Translations of

MATH EMATI CAL

MONO GRAPHS
Volume 226

Hubert

V. M. Manuilov
E. V. Troitsky

American Mathematical Society

Translations of
MATH EMATI CAL
MONO GRAPHS
Volume 226

Hilbert
V. M. Manuiov
E. V. Troitsky
Translated by the authors

American Mathematical Society

Providence, Rhode Island
 EDITORIAL COMMITTEE
AMS Subcommittee
Robert D. MacPherson Grigorii A. Margulis James D. Stasheff (Chair)
ASL Subcommittee Steffen Lempp (Chair)
IMS Subcommittee Mark I. Freidlin (Chair)
B. M. E. B.
FI4JIbBEPTOBbI MOIIYJIM
MOCKBA, 2001
This work was originally published in Russian by Faktorial Press under the title "C
©2001. The present translation was created under license for
the American Mathematical Society and is published by permission.
Translated from the Russian by V. M. Manuilov and E. V. Troitsky.

2000 Mathematics Subject Classification. Primary 46L08; Secondary 46Lxx.

For additional information and updates on this book, visit

www.ams.org/bookpages/mmono-226

Library of Congress Cataloging-in-Publication Data

Manuilov, V. M. (Vladimir Markovich). 1961-
[C* Gil'bertovy moduli. English]
Hilbert / V.M. Manuilov, E.V. Troitsky ; translated by V.M. Manuilov and E.V.
Troitsky.
p. cm. (Translations of mathematical monographs, ISSN 0065-9282 v. 226)
Includes bibliographical references and indexes.
ISBN 0-8218-3810-5 (acid-free paper)
1. 2. Hilbert algebras. I. Troitskii, E. V. (Evgenii Vadimovich) II. Title.
III. Series.
QA326.M3413 2005
512'.556—dc22 2005042811

Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of ally material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society.
201 Charles Street. Providence. Rhode Island 02904-2294. USA. Requests can also be made by
e-mail to reprint-permission@ams .org.
© 2005 by tile American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.
® The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10987654321 100908070605
 Contents
Preface vii

Chapter 1. Basic Definitions 1

1.1. C-algebras 1

1.2. Pre-Hilbert modules 3

1.3. Hubert 4
1.4. The standard Hubert module HA 8
1.5. Hubert and strong Morita equivalence 11

Chapter 2. Operators on Hubert Modules 15

2.1. Bounded and adjointable operators 15
2.2. Compact operators in Hubert modules 18
2.3. Complementable submodules and projections in Hubert C-modules 22
2.4. Full Hilbert 24
2.5. Dual modules. Self-duality 27
2.6. Banach-compact operators 32
2.7. operators and index. Mishchenko's approach 33
2.8. Representations of groups on Hilbert modules 42
2.9. Equivariant Fredholm operators 52

Chapter 3. Hubert Modules over 55

3.1. 55
3.2. Inner product on dual modules 58
3.3. Hilbert and dual Banach spaces 61
3.4. Properties of Hilbert Wt-modules 62
3.5. Topological characterization of self-dual HUbert 65
3.6. F'redholm operators over 66
3.7. Dupré — Fillmore theorem for Hilbert modules over finite 69

Chapter 4. Reflexive Hilbert 75

4.1. Inner product on bidual modules 75
4.2. Ideals and bidual modules 79
4.3. Reflexivity of Hilbert modules over V 82
4.4. Reflexivity of modules over C(X) 84
4.5. Hilbert modules related to conditional expectations of finite index 86

Chapter 5. Multipliers of A-Compact Operators. Structure Results 99

5.1. Extension of a Hilbert by the enveloping 99
5.2. Multipliers and centralizers 101
5.3. Multipliers of A-compact operators 107
5.4. Quasi-multipliers of A-compact operators 110
V
vi CONTENTS

5.5. Strict topology 114

5.6. Multipliers and Hubert modules. The commutative case 118
5.7. Inner products on Hubert 126

Chapter 6. Diagonalization of Operators over 133

6.1. Problem of diagonalizing operators in Hubert 133
6.2. Quadratic forms on related to selfadjoint operators 136
6.3. Diagonalizing operators in the 138
6.4. Continuity of "cigenvalues" 143
6.5. Case of infinite 145
6.6. Case of of real rank zero 146
6.7. Case of continuous fields of trace 148
6.8. Schrödinger operator as an operator acting on a Hubert 154
6.9. Example: A continuous field of rotation algebras 157

Chapter 7. Homotopy of Groups of Invertible Operators 159

7.1. Technical lemmas 159
7.2. Proof of the Cuntz—Higson theorem 164
7.3. The case A C 166
7.4. Some other cases 169
7.5. Dixmier-Douady Theorem for 12(A) 173
7.6. Some generalizations 175
7.7. Neubauer type homotopy 176

Chapter 8. Hilbert Modules and KK-Theory 181

8.1. Tensor products 181
8.2. Main definitions 182
8.3. Cuntz's approach 184
8.4. Generalized Kasparov bimodules 188
8.5. Classifying spaces for some K- and KK-groups 189

Bibliography 193

Notation Index 199

Index 201
 Preface
Hubert provide a natural generalization of Hubert spaces arising
when the field of scalars C is replaced by an arbitrary This generaliza-
tion, in the case of commutative appeared in the paper [591 of 1. Ka-
plansky; however the noncommutative case seemed too complicated at that time.
The general theory of Hilbert (i.e., for an arbitrary serving
as 'scalars') appeared 25 years ago in the pioneering papers of W. Paschke [100] and
M. Rieffel [108]. This theory has proved to be a very convenient tool in the theory of
operator algebras, allowing one to obtain information about by study-
ing Hilbert over them. In particular, a series of results about some
classes of (like and monotone complete
was obtained in this way [38]. An important notion of Morita equivalence for
was also formulated in terms of Hubert [109, 18]. This
notion also has applications in the theory of group representations. It turned out
to be possible to extract information on group actions from Hubert
arising from these actions 1102, 1101. Some results about conditional expectations
of finite index [5, 133] and about completely positive maps of [2] were
also obtained using Hilbert
The theory of Hilbert may also be considered as a noncommutative
generalization of the theory of vector bundles [33, 69]. This was the reason Hilbert
modules became a tool in topological applications — namely in index theory of
elliptic operators, in K- and KK-theory [92, 90, 62, 63, 64, 65, 128] and in
noncommutative geometry as a whole [24, 29].
Among other applications, one should emphasize the theory of quantum groups
[135, 136], unbounded operators as a tool for Kasparov's KK-theory [3, 4] (also
Section 8.4) and some physical applications [72, 80].
Alongside these applications, the theory of Hilbert itself has been
developed as well. A number of results about the structure of Hubert modules and
about operators on them have been obtained [74, 42, 88, 77, 81, 127]. Besides
these results, an axiomatic approach in the theory of Hubert modules based on the
theory of operator spaces and tensor products was developed [12, 11].
A detailed bibliography of the theory of Hubert can be found
in [43].
Some results presented here were only announced in the literature or the proofs
were discussed rather briefly. We have tried to fill such gaps. We could not discuss
all the aspects of the theory of Hilbert here, but we tried to explain
in detail the basic notions and theorems of this theory, a number of important
examples, and also some results related to the authors' interest.

Vt'
viii PREFACE

A major part of this book formed the content of the lecture course presented
by the authors at the Department of Mechanics and Mathematics of Moscow State
University in 1997.
We are grateful to A. S. Mishchenko for introducing the theory of Hubert
modules to us. Together with Yu. P. Solovyov, lie has acquainted us with the circle
of problems related to its applications in topology.
While working in this field and in the process of writing the present text, a
significant influence oii us was niade by our friend and co-author M.
We have discussed a number of problems of the Hubert C*_mnodule theory
with L. Brown, A. A. Irmatov, G. G. Kasparov, R. Nest, G. K. Pederseii and
'N. Paschke. Some applications were considered also with J. Cuntz, A. Ya. Helein-
skii, J. Kaminker, V. Nistor, J. Rosenberg, K. Thomsen, B. L. Tsygan and others.
Our research was partially supported by a series of subsequent Russian Foim-
dation of Basic Research grants.
 CHAPTER 1

Basic Definitions

1.1.
For basic information about we refer to the books [31, 104, 123,
58]. We present here only some results on C*_algebras, which will be necessary for
our purpose.
Recall that an involutive Banach algebra A is called a if the equality
IIa*aIl 11a112

holds for each element a E A. Any can be realized as a norm-closed

subalgebra of the algebra of bounded operators 13(H) on a Hilbert space H. We
do not assume existence of the unit element in By A+ we denote
the obtained from the A by unitalization. As a linear space
with involution, coincides with A C and multiplication in is given by the
formula (a, z)(b, w) = (ab + zb + aw, zw), a, b e A, z, w E C. Any pair (a, z) E A+
defines an operator A —p A by b '—+ ab + zb and the norm of (a, z) is the norm of
this operator.
The spectrum of an element a of a unital is the set Sp(a) of complex
numbers z such that a — z• 1 is not invertible. If a C*_algebra A has no unit, then
the spectrum of an element a E A is defined as its spectrum in the
A+ D A. The spectrum is a compact subset of C. An element a E A is called
positive (we write a > 0) if it is Hermitian, i.e., if it satisfies the condition a* = a,
and if one of the following equivalent [31, 1.6.1] conditions holds:
(i) Sp(a) C I0,oo);
(ii) a = b*b
for some b E A;
(iii) a = for some Hermitian h E A.
h2

The set of all positive elements (A) forms a closed convex cone in A and P+ (A) fl
= 0. Among the Hermitian elements h defined in (iii) there exists a
unique positive element, which is called the positive square root of a (we write
h a1!2).
A linear functional : A C is called positive if > 0 for any positive
element a E A positive linear functional is called a state if = 1. We
have hail = sup where a 0 and the supremurn is taken over all states.
A of an algebra A into the 13(H) of all bounded
operators on a Hilbert space H is called a representation. A vector E H is
called cyclic for the representation ir : A —÷ 13(H) if the set of all vectors of the
form a E A, is dense in H. The vector H is called separating for the
representation ir : A —+ 13(H) if the equality = 0 implies a 0.
To each positive linear functional w on a A we can associate a
unique (up to the unitary equivalence) representation of the algebra A on some
2 1. BASIC DEFINITIONS

Hubert space and a vector E such that w(a) = for all a e A

and the vector is cyclic. The construction of such a representation is called the
GNS-construction.
An approximate unit of a A is an increasing net ea E A, E A.
such that Ileall 1 and urn la — aeall = 0 for any a E A. Each has an
approximate unit ea such that ea 0 and e0 e0 for [311.

DEFINITION 1.1.1. A possessing a countable approximate unit is

called a-unital.
DEFINITION 1.1.2. An element h E A is called strictly positive if for any positive
nonzero linear functional (or, equally, for any state) one has cp(h) > 0.
REMARK 1.1.3. Existence of a strictly positive element is equivalent to existence
of a countable approximate unit. One can assume that 0. Then h :=
is strictly positive. Conversely, := hut is a countable approximate unit. Any
separable is a-unital. The details can be found in [1041.
We will often use the following statements.
LEMMA 1.1.4 ([104, Lemma 1.4.4]). Let x, y and a be elements of a
A such that a 0 and
x*x<_aa, yy*<a/3

Then the sequence u,1 = x[(1/n) + is norm-convergent in A to an element

u such that hull

PROOF. Put dnrn := [(1/n) + a]'/2 — [(1/rn) + a]'/2. Then

— Urn
2_— XanmY 2_ *j — XanrnY

_Y
< nma nmya°12d nmY — 2

= II II
112.

Using, for example, the Dini theorem, we can see that the sequence of functions
[(1/n) + t E Sp(a),
is uniformly convergent to on the spectrum of a, hence,
0. Therefore, by the Cauchy criterion, is norm-convergent to an element u e A.
Then, reasoning as above, we obtain
hlunhl = hlx[(1/n) + a]"2yhl + a]_h/2a13/2h1
Hence huh o
PROPOSITION 1.1.5 ([104, Prop. 1.4.5]). Let x and a be elements of a
algebra A such that a 0 and a. For any 0 < there exists an element
u e A such that huh S and x =

PROOF. Put u,-, := x[(1/n) + By Lemma 1.1.4, is norm-

convergent to an element u E A such that
(1I1_2a1) = —all.
huh II
 1.2. PRE-HILBERT MODULES 3

Then
lix — 112 = iix(1 — [(1/n) +
Iia"2(i — [(1/n) + a]"2a"2)ii2 0

as n —p oo, by the Dini theorem applied to the corresponding functions on the

spectrum. Thus, x 0
1.2. Pre-Hilbert modules
Let M be a module over a A. An action of an element a E A on
M is denoted by x a, where x E M.
DEFINITION 1.2.1. A pre-Hilbert A-module is a (right) A-module M equipped
with a sesquilinear form : M x M —' A with the following properties:

(i) (x,x)OforanyxEM;
(ii) (x, x) = 0 implies that x = 0;
(iii) (y,x) = (x,y)* for any x,y EM;
(iv)
The map is called an A-valued inner product.
Here are some examples.
EXAMPLE 1.2.2. Let J C A be a right ideal. Then J can be equipped with the
structure of a pre-Hilbert A-module with the inner product of elements x, y E J
defined by (x, y) := x'y.
EXAMPLE 1.2.3. Let be a countable set of right ideals of a A
and let M be the linear space of all sequences (x2), E J2, satisfying the condition
lxi <oc. Then M becomes a right A-module if the action of A is defined by
(xi) a for (xe) E M, a E A, and becomes a pre-Hilbert A-module if the
inner product of elements (x2), (yz) E M is defined by ((xe), := xyj.
Let /C be a right A-module equipped with a sesquilinear map [.,.]:K x K —+ A
satisfying all properties of Definition 1.2.1 except (ii). Put
N := {x E K;: [x,x] 0}.
For each positive linear functional on the A the map (x, y) i—* y])
is a (degenerate) inner product on K, hence the set = {x E K: x]) = 0}
is a linear subspace in K. By taking the intersection of all such subspaces we see
that N = is also a linear subspace in K. It follows from properties (iii) and
(iv) of Definition 1.2.1 that N. A C N. Therefore N is a submodule in K;. The
quotient module M = K/N is equipped with the obvious structure of a pre-Hilbert
A-module with the inner product (x + N,y + N) := [x,y].
Let M be a pre-Hilbert A-module, x E M. Put IIXIIM := ii(x,x)ii"2. We
usually skip the subscript M when it does not lead to confusion of norms.
PROPOSITION 1.2.4 ([100]). The function is a norm on M and satisfies
the following properties:
(i) lix . aiiM < laD for any x E M, a E A;
(ii) (x,y)(y,x) (x,x) for any x,y E M;
(iii) ii(x,y)D 1IY1IM for any x,y E M.
4 1. BASIC DEFINITIONS

PROOF. For any positive linear functional on A, the function x x))1/2

definesa seminorm on M. For each x E M,
IIX11M =
iI(x,x)Iih/2 =
where the supremum is taken over all states p on A. Therefore I'M is a seminorm
and, by property (ii) of Definition 1.2.1, is a norm on M. Statement (i)
follows from the equality
lix = (x a, x a) ii = iia* (x, x)aii II = iiaii2

To prove (ii) we take x, y E M and a positive linear functional p on A. Applying

the Cauchy—Bunyakovskii inequality for the (degenerate) inner product (•,•)) on
M we obtain
p((x,y)(y,x)) = (y,x)))

(y,x),y. (y,x)))"2

= p((x,x))"2 .

1/2 1/2 1/2

Thus, for any positive linear functional we have (y, x) (x, y)) < . p( (x, x)).
so statement (ii) is proved. This evidently implies statement (iii). D
We call inequality (ii) (and also its consequence, inequality (iii)) of Proposition
1.2.4 the Cauchy-Bunyakovskii inequality for Hubert
REMARK 1.2.5. For any module, or more precisely, for any
sesquilinear form the polarization equality

4(y,x) =

is obviously satisfied for all x. y e M.

1.3. Hubert
DEFINITION 1.3.1. An A-module M that is at the same time a Banach space
with a norm satisfying the inequality x E M, a E A, is called
a Banach A-module.
DEFINITION 1.3.2. A pre-Hilbert A-module M is called a Hubert if
it is complete with respect to the norm
If M is a pre-Hilbert A-module, then the action of the A and the
A-valued inner product on M can be extended to the completion M, which thus
becomes a Hilbert Consider some examples.
EXAMPLE 1.3.3. If J C A is a right ideal, then the pre-Hilbert module J is
complete with respect to the norm = In particular, the C*_algebra A
itself is a free Hubert A-module with one generator.
EXAMPLE 1.3.4. If is a finite set of Hilbert A-modules, then one can
define the direct sum The inner product on is given by the formula
(x,y) := (xj,yj), where x = (xj),y = e EBM1. We denote the direct sum of
n copies of a Hilbert module M by or L,1(M).
 1.3. HILBERT 5

Here and further we denote by the direct sum of orthogonal submodules and
by the direct sum of Banach subspaces without orthogonality.
ExAMPLE 1.3.5. If {M2}, i E N, is a countable set of Hubert A-modules, then
one can define their direct sum On the A-module of all sequences
x = (xi) x2 E M?, such that the series
: (x2, x?) is norm-convergent in the
A, we define the inner product by

(x,y) >2 (Xj,yj) for x.y E

Let us check that this series converges. Since the series (xi, and (ye. yj)
are convergent, for any > 0 there exists a number N such that for all n > 0 we
have
N+n N+n

LN L=N
Theii
N+n N+n N+n
.

L=N
This proves that the inner product is well defined.
Let us verify completeness of the module Let = e be a
Cauchy sequence. Then for any E > 0 there exists a number N such that

(1.1) - -
for all n, m N. Since all summands in (1.1) are positive, the inequality

— — <6

holds for each number i separately. But then the sequences E M2 are Cauchy
sequences, hence they converge to the limits x1 = lim E M1. Let us verify that
the series (xi, x2) is norm-convergent in A. Let us fix 6 > 0. There exists a
number n> N such that the estimate (1.1) holds. Let us choose a number K such
that

Then, for any k > 0, we have

K+k
— J2))
+ + — + (X(n)
i=K
— —
= — — < 6.
6 1. BASIC DEFINITIONS

Therefore
K+k K+k K+k
<2e+ + _xcm))V
L=K
K+k 1/2
12
/
< 2e + 2 — —

1/2

Now, by solving the quadratic inequality, we obtain that

K+k
(1.2) <(1+ <8e.
L=K
Passing to the limit m —' oo in the inequality (1.2), we obtain that
K+k
<8e.

This proves that the series is norm-convergent.

The direct sum of a countable number of copies of a Hilbert module M is
denoted by 12(M) or HM. The Hilbert 12(A) (another notation is
HA) is called the standard Hilbert over A. If the is unital,
then the Hilbert module HA possesses the standard basis }, i e N, where =
(0,. .0, 1,0,. . 0,...) with the unit being the i-th entry.
. , . ,

EXAMPLE 1.3.6. Let B C A be a of a A. Let A

and B be unital with the common unit. Assume that there exists a linear map
E: A —+ B that is a projection (i.e., E2 = E) of norm 1. Such a map is called
a conditional expectation from A to B. Conditional expectation is a positive map,
i.e., E(a*a) 0 for all a E A, and it satisfies the equality
E(biab2) = biE(a)b2 for a e A,b1,b2 E B
(see [123]). A conditional expectation is called faithful if, for any positive element
aE the equality E(a) = 0 implies a = 0. When the conditional expectation
is faithful, one can introduce the structure of a pre-Hilbert B-module on the
algebra A by
(x,y) = E(x*y), x,y e A.
We will give a condition for this module to be a Hilbert (i.e., to be
complete) in Section 4.5.
Let .N c M be a closed submodule of a Hilbert M. We define the
orthogonal complement Al'-'- by the formula
= {y eM: (x,y) = 0 for all XE A(}.
Al'-'-

Then is a closed submodule of the Hilbert M too. However, the

equality M = .N' N-'- does not always hold, as the following example shows.
EXAMPLE 1.3.7. Let A = C[0, 1] be the of all continuous functions
on the segment [0, 1]. Consider, in the Hilbert A-module M = A, the submodule
Al' = C0(0, 1) of functions that vanish at the end points of the segment. Then,
obviously, Js[.L = 0.
 1.3. HILBERT C-MODULES 7

If M is a Hubert A-module, then we denote by M A the closure in M of the

linear span of all the elements of the form x a, x E M, a E A.
LEMMA 1.3.8. M . A = M.
PROOF. Let E A be an approximate unit. Then, for any x E M,
= II(x—xea,x—x.ea)II
< (1 + II(x,x) — (x,x)eaII 0,

hence the elements of the form x ea are dense in M. D

We will often usc the following statement.

LEMMA 1.3.9. For any x E M
x=
PROOF. Let (x, x) = a. Then
IIx(x,x)((x,x) + e)_1 — x112

= II(x((x,x)((x,x) +e)' — 1),x((x,x)((x,x) +e)' — i))II

= IIa(a2(a+EY2 — 2a(a+e)1 + 1)11
=
since the following inequalities hold under the condition t 0:

/1 \
=e et+2t2
3
(t+e)2 \2 j 2

and

t2(t+ey'—tI= <E.
t+e
0
The following statement is an analog of the polar decomposition for Hilbert
We will see below that, similarly to the case of C*_algebras, the exact
polar decomposition exists only in the case of Hilbert over

PROPOsITION 1.3.10 ([711). Let M be a Hubert A-module, x E M, and 0 <

a < 1/2. Then there exists an element z E M such that x
PROOF. For n E N put
I if A 1/n,

1 if A> 1/n.
Then, by the spectral theorem,
(gn((X,X)) — gm((X,X)))II =
=sup{IA(gn(A) —gm(A))I : A E Sp((x,x))}.
8 1. BASIC DEFINITIONS

Therefore the sequence x ( (x, x)) is a Cauchy sequence, so it has a limit z E M.

Then
— xli = 11x

— 1)11
=

This completes the proof. 0

1.4. The standard Hubert module HA

DEFINITIoN 1.4.1. A Hubert M is called finitely generated if there
exists a finite set c M such that M equals the linear span (over C and A)
of this set. A Hilbert M is called countably generated if there exists a
countable set {x, } c M such that M equals the norm-closure of the linear span
(over C and A) of this set.
THEOREM 1.4.2 (Kasparov stabilization theorem, [63]). Let A be a
a countably generated Hilbert A-module. Then M HA HA.

PROOF. We start by proving the theorem for the case where A is unital. It is
convenient here to use the procedure of almost orthogonalization [36]. An element
x of the Hilbert C*_module is called nonsingular if the element (x, x) E A is
invertible. The set {x1} e Jv is called orthonormal if (x,, x3) = It is called a
basis of the module 1V if finite sums of the form E A, are dense in
LEMMA 1.4.3 ([36]). LetJ'f be a Hubert A-module that contains the orthonorrnal
elements e1 x E J'.f, > 0. If an element y E Al satisfies (y,y) = 1 and
y±{x,ei then there exists an element EAl such that
(i) the elements e1, . . . are orthonormal,
(ii) efl+lESpanA(el en,x,y),
(iii) dist(x,SpanA(el, . . . < e.
PROOF. Let
=x — x" = x' +Ey.

Then
(x", x") = (x', x') + E2 e2 > 0.
1/2
Therefore the element x" is nonsingular. Put = x". (x", x") — Then
E SpanA(x',y).L{el,.. .
Therefore the elements e1, . . . , are orthonormal. Since we have taken x' E
SpanA(x,el,. . . and E SpanA(x',y), we obtain (ii). Finally, put

w= x")'12 + x) E SpanA(el,. . . ,

Then the equality lw — xM = lix" — = E proves (iii). 0

 1.4. THE STANDARD HILBERT MODULE HA 9

We return now to the proof of Theorem 1.4.2. Let {y,,} be the sequence of
all generators of module M. By } we denote the standard basis of the module
HA. Let c U be a sequence, in which one meets each element
and each element Yn infinitely many times. Then the set } is generating for the
module M HA. We will prove the theorem by induction. Let us assume that the
orthonormal elements E M HA and the number in(n) n are already
,

constructed in such a way that

(i) C SpanA(xl,... .

(ii) . 1 k n.
Since each element is equal to e3 or one can find a number m' > Tn(n) such
that em' J-{xi }. Since em' ..L{ei,. . . , it follows from (i) that
. .

By Lemma 1.4.3, there exists an element

(1.3) E . . ,ën,xn÷i,em')
such that the elements . , are orthonorinal and
— — 1
SpanA(el,.. Cfl.44))

It follows from (1.3) and from condition (i) that

{e1,. ..,efl+1} C SpanA(xl,. ..
By setting m(n + 1) Tn', we complete the induction step. Thus, an orthonormal
sequence satisfying properties (i) and (ii) has been constructed. But property (ii)
means that this sequence generates the whole module M HA, so M HA HA.
Thus, Theorem 1.4.2 is proved for unital Let A be a nonunital
and let A+ be its unitalization. Defining tile action of on the Hilbert
A-module M by the formula x (a, A)
M with the structure of a Hilbert At-module. Consider the Ak-module
"A+ and denote by "A+ A the closure of the linear span of all the elements of the
form x a, x HA+, a A, in HA+. It is easy to see that HA+A = HA. The
isomorphism M HA+ HA+ implies the isornorphism
M HA = MA HA+A = (M HA+)A = HA.
0
DEFINITION 1.4.4. A Ililbert A-module M is called a finitely generated projec-
tive A-module if there exists a Hilbert A-module .N such that M for
some n.
The following two theorems of Dupré and Fillmore show that finite-dimensional
projective submodules lie in Hilbert C*_modules in the simplest way.
THEOREM 1.4.5 (Dupré — Fillmore, [36]). Let A be a unital
a finite-dimensional projective A-submodule in the standard Hilbert A-module
HA. Then
(i) the nonsingular elements of the module M1 are dense in M';
(ii)
(iii)
 __

11) 1. BASIC DEFINITIONS

PROOF. We begin the proof of the theorem with the case where M L71(A).
Let 91'... be an orthonormal basis in M. Fix > 0. For each m. put

— gj(gj,em).

Thcii E M± and

Since (x, Cm) 0 for each x e HA, we conclude that 1. Therefore

there exists a number m0 such that for any m m0, the element is nonsingular.
Then one can define
, / , ,
Cm —

with = 1. Let x M1. Then

, , —1/2 , , , —1/2
x) = em) x) = (em, (eflL, . 0.

Choose a number rn 'rn0 such that x)II < and set

= x +E
It is easy to see that
(1.4) lix' — xii =
Let us check that the element x' is noiisingular. Put
= x— v= +E1).
Then u±v (since u + v. Therefore
(1.5) (x', x') (u, u) + (v, v) = (u, u) + x) + e' x) + El),
and tile right-hand side of equality (1.5) is invertible since x) 1 < e. Therefore
(x', x') is invertible too. Together with estimate (1.4), this proves statement (i).
Let be a sequence in which each element Cm is repeated infinitely many
times. Put x = Il — gj(9j,Xi). Themi (taking E = 1) one can find an el-
etnent 9n+1 M1 such that (9n+i Yn÷i) = 1, dist(x,
, 1. Therefore
dist(xl,SpanA(gj gn÷i)) 1. At the next step we replace the module M by
SpanA(gl gn+i), x1 by £2, and E = 1 by E = 1/2. Going on with this proce-
dure, we obtain an orthonormal basis k N, extending tile basis . , . .

of submodule M, and the remaining part {gk : k > n} is a basis of the module
M'. This proves statements (ii) and (iii).
We pass now to the case of an arbitrary finitely generated projective module
M. Let M By Theorem 1.4.2, HA HA, hence

Therefore, if is the orthogonal complement to tile submodule M in the

then HA and =JVWHA. But = M1 is
obviously the orthogonal complement to the submodule M in the module HA. D
THEOREM 1.4.6 ([36]). Let A be a unital a finitely
generated projective Hubert submodule in an arbitrary Hubert A-module Then
 1.5. HILBERT C-BIMODULES AND STRONG MORITA EQUIVALENCE 11

PROOF. As in the previous theorem, the proof can be reduced to the case where
M is a free module, M = If {g1,. .. is the standard basis of M, then
put x' = x — for x E H. Then x' EM and x — x' E M', hence
J'/=MEDM'. 0
For a Hubert A-module M denote by (M, M) C A the closure of the linear
span of all (x, x), x E M. The set (M, M) is obviously a closed two-sided involutive
ideal in the A.

DEFINITION 1.4.7. A Hubert A-module is called full if (M, M) = A.

One can always consider any Hilbert module as a full Hilbert module over the
(M, M).
The standard Hilbert is obviously full.

1.5. Hubert and strong Morita equivalence

In this section we briefly discuss the case where a module has two
structures over two Such were studied by Rieffel in
[109] and we follow these papers here.
Let A and B be two Let M be a right pre-Hilbert
over A with the inner product (., and a left pre-Hilbert over B with
the inner product The latter means that the sesquilinear form (., .)B is
conjugate linear in the first variable.
DEFINITION 1.5.1. The module M is called a pre-Hilbert A-B-bimodule if the
following conditions hold:
(i) (x,y)Bz=x(y,z)A for any x,y,zEM;
(ii) (bx,bx)A 11b112(x,x)A and (xa,xa)B

LEMMA 1.5.2. One has II(x,x)AIl = II(x,x)BIl for any x EM.

PROOF. Denote (x,x)A = a, (x,x)B = b. Then bx = xa for any x E M. Since
a3 = (xa,xa)A =
IIbIIIIaII2, hence hail c Similarly one obtains hibli hall. 0
Thus we see that lxii = hI(x,x)AII"2 II(x,x)B11112 defines a norm on M.

DEFINITION 1.5.3. A pre-Hilbert A-B-bimodule M is called an A-B-equiva-

lence bimodule if it is complete with respect to the norm and if it is full both as a
right and as a left Hilbert
LEMMA 1.5.4. One has (bx,y)A = (x,b*y)A and (xa,y)B = (x,ya*)B for any
x,y E M, a E A, bE B.
PROOF. It suffices to prove the first statement since the second is similar. Since
the bimodule M is full as a B-module, we can assume without loss of generality
that b = (Z,t)B for some z,t E M. Then
(bx,y)A = (Z(t,X)A,Y)A
= (x,t)A(Z,y)A.
12 1. BASIC DEFINITIONS

On the other hand,

(x, = (x, (t, z)By)A = (x, t(z, Y)A)A
(X,t)A(Z,y)A.
0
DEFINITION 1.5.5. Two A and B are called strongly Morita equiv-
alent if there exists an A-B-equivalence bimodule.
LEMMA 1.5.6. Strong Morita equivalence is an equivalence relation.
PROOF. Reflexivity of this equivalence relation is clear, symmetry follows from
considering the conjugate module X, which consists of the same elements, but the
bimodule structure is given by = xf
and = b*x, x E M, a E A, b e B. So
we have to check transitivity. Let M be an A-B-equivalence bimodule and let
be a B-C-equivalence bimodule, where C is one more Then ®B M
is a left C-module and a right A-module. For finite sums n2 0 m2, ®m
in OB M define an A-valued inner product by
n2 0 > n 0 m3)A = m3)A

and similarly define a C-valued inner product. It is easy to check that these inner
products satisfy all the necessary properties, except, possibly, nondegeneracy. Let
£ C JV®BM be the (maybe empty) set of all finite sums 1 = E .A1®BM
with (I, l)A = 0. Then the completion of ®B M/C with respect to the norm
obtained from any of the two inner products gives us an A-C-equivalence bimodule
(for example, fullness follows from Lemma 1.3.8). 0
EXAMPLE 1.5.7. Consider the standard right Hilbert A-module HA and endow
it with the left module structure over the C*_algebra A 0 K, where IC denotes the
of compact operators on a separable Hilbert space. Elements of A ® IC
can be written as infinite matrices of the form (a23), i,j e N, with entries in A. The
left action of such a matrix (a23) on a sequence (x2) E HA results in the sequence
Define the A 0 IC-valued inner product on HA by
((xi), = (Xjy) E A® IC
(it is easy to check that the matrix (xjy) defines an element from A 0 IC because
both sequences (xi) and vanish at infinity). All properties of an A-A 0 K-
equivalence bimodule can be easily checked and thus we see that A is strongly
Morita equivalent to A ® IC. Similarly, taking instead of HA, we obtain the
strong Morita equivalence between A amid the C*_algebra (A) of n x n-matrices
with entries from A.
This example and Lemma 1.5.6 imply the following statement.
COROLLARY 1.5.8. Let A and B be If the A® IC and
B ® IC are isomorphic. then A and B arc strongly Morita equivalent.
A criterion for the strong Morita equivalence was obtained in [15, 18]. Since
the method of proof is far removed from the Hilbert technique, we skip
the proof.
 1.5. HILBERT C-BIMODULES AND STRONG MORITA EQUIVALENCE 13

THEOREM 1.5.9 ([15, 18]). Let A and B be strongly Morita equivalent

algebras. If they both have countable approximate identities, then the C* -algebras
A 0 K and B 0 K are isomorphic.
 CHAPTER 2

Operators on Hubert Modules

2.1. Bounded and adjointable operators
Let M, be Hubert over a A. A bounded C-linear A-
homomorphism from M to is called an operator from M to Let HomA (M , .i\f)
denote the set of all operators from M to .N. If .Af = M, then EndA(M) =
H0mA (M, M) is obviously a Banach algebra. However, we shall soon see that
there is no natural involution on this algebra. Let T E HomA(M..N). We say that
T is adjointable if there exists an operator T* E HornA(ftf, M) such that
(Tx,y) = (x,T*y)
for all x E M, y E .iV.
LEMMA 2.1.1. Let M be a Hubert A-module and let T : M —' M and T*
M —+ M be maps such that

(x.Ty) = (T*r,y)

for all x, y E M. Then T is a bounded C-linear A-homomorphism (and T* is as

well).

PROOF. For any x,y,z EM, wE C and a E A one has

(z,T(x+y)) = = (T*z,x) + (T*z,y)
= (z,Tx) + (z,Ty) = (z,Tx + Ty),
(z,Twx) = (T*z,x)w (z,Tx)w = (z,wTx),
(z,T(xa)) = (T*z,xa) = (T*z,x)a = (z,Tx)a = (z,(Tx)a).

Since z is an arbitrary element, it follows that

T(x + y) = Tx + Ty, T(wx) = wTx, T(xa) = (Tx)a,
and linearity properties hold.
To prove the continuity of T we should verify that its graph is closed. Let
x. T(Xa) y in M, and let z E M be an arbitrary element. Then

o = (T*(y — — (T4(y —

= (y — — (T*(y —
—+ (y — Tx,y) — (T*(y — Tx),x) = (y — Tx,y — Tx). D

We show now that there exist nonadjointable operators.

15
16 2. OPERATORS ON HILBERT MODULES

EXAMPLE 2.1.2. Let A be a unital As above, the standard basis

of the Hubert module HA consists of the elements e2 = (0, .. , 0, 1,0,...), where
.

1 is the i-th entry. To each operator T E EIIdA(HA) one can associate an infinite
matrix with respect to this basis.
IIt&iII. = (e1,Te1).
Then the adjoint operator has the matrix
Let A = C([0, 1]) and let the functions çoj E A, i = 1,2,..., be defined by the
formula
0

on and [xi,
Let an operator T E EndA(HA) have the matrix

0
P2
0 0...
0 0 0...
(actually it is an operator from the module HA to A, i.e., an A-functional). It is
easy to verify that T is bounded. But the operator T* is not well defined since it
should have the matrix
0 0
0 0
0 0

and the image of the basis element e1 should be an element of HA having the first
column as its coordinates and it has to he an element of HA, which is impossible
since the series is not norm-convergent in the C*_algebra A.
Denote by (M , N) the set of all adjoint able operators from M to N. The
algebra = M) is an involutive Banach algebra. Moreover, it
is a C*_algebra; this follows from the estimate
IIT*TII > sup {(T*Tx,x)} sup {(Tx,Tx)} = 11Th2,
xEB1(M) xEB1(M)
where B1(M) denotes the imit ball of the module M.
We will use the following statement frequently without special reference.
PROPOSITION 2.1.3. For an operator T : M —+ M, the following conditions
are equivalent:
(i) T is a positive element of the End*(M);
(ii) for any x E M the inequality (Tx, x) 0 is fulfilled, i.e., is positive in
tile A.

PROOF. The first condition is equivalent to the equality T = S*S for some
S E End*(M). Therefore
(Tx, x) = (Sx. Sx) 0 for any x E M.
Now let (Tx,x) 0 for all x E M. Then
(Tx, x) = (Tx, x E M.
 2.1. BOUNDED AND ADJOINTABLE OPERATORS 17

The map (x,y) (Tx,y) defines a sesquilinear form on M. Therefore, by the

polarization equality 1.2.5, (Tx,y) = (x,Ty) for all x,y E M. By Lemma 2.1.1,
this means that T E End* (M) and T = T*. So T is a selfadjoint element of the
C*_algebra End*(M), so (see [31, 1.6.5]) it can be represented as the difference
T — T_ of two elements of End*(M), where > 0, T_ ? 0 and =
= 0. Then (T_y,y) for any y E M. In particular,
= ( (T+T_x,T_x) = 0.
Oii the other hand, T_ > 0 and > 0, hence x) 0 (because the statement
in this direction is already proved). So the only possibility left is
x e M. By the polarization equality, this implies y) = 0 for all x, y e M,
U

THEOREM 2.1.4 ([100]). Let M and Al be Hubert A-modules and let T: M

Al be a linear map. Then the following conditions are equivalent:
(i) the operator T is bounded and A-linear, i.e., T(x a) = Tx a for all
x E M, a E A;
(ii) there exists a constant K 0 such that the inequality (TX, Tx) K(x, x)
holds in A for all x e M.
PROOF. To obtain the second statement from the first, let us assume that
T(x a) = Tx. a and that 11Th 1. If the A is not unital, then
we consider M and Al as modules over the A+ obtained from A by
unitalizing. For x E M and n e N put
—1/2
1
,

Then = = (x,x)((x,x) + 1. Therefore 1, hence

1. Then the inequality Tx,.,) 1 holds in A (or in for all
n E N. But
(2.1) (Tx,Tx) = = (x,x) +!.
Passing to the limit as n —+ ocin the inequality (2.1), we obtain (Tx,Tx) (x,x).
To derive the first statement from the second, assume that the inequality
(Tx, Tx) < (x, x) holds for all x E M. It clearly follows that T is bounded,
11Th <1. Let x e M, y e Al. Define the map r : by the formula
r(a) = (y,T(x. a)).
Then
2
r(a)*r(a) = (T(x.a)y)(y,T(x.a)) (T(x.a),T(x.a))
11y112 (x a, x a) = 11y112 a*(x, x)a
11xI12 a*a.

To complete the proof we use the following statement.

LEMMA 2.1.5 (156, 100]). Let A be a unital
a linear map such that for some constant K 0, the inequality r(a)*r(a) Ka*a
holds for alla e A. Then r(a) = r(1)a for alla E A.
18 2. OPERATORS ON HILBERT MODULES

Thus r(a) = r(1)a, i.e.,

(y, T(x a)) = (y, Tx)a = (y, Tx a)
for all y E dv, x E M. This implies the first statement of the theorem. 0
COROLLARY 2.1.6. Let M,A( be Hubert A-modules, T E EndA(M,dv). Then
11Th = inf{K'/'2: (Tx,Tx) K(x,x) Vx e M}.
EXAMPLE 2.1.7. Let M = a decomposition into an orthogonal direct
be
sum of Hubert modules. We define the operator P : M —p M to be the projection
onto the submodule .IV along the submodule £. Then P is bounded, = 1, arid
= P, hence P E

2.2. Compact operators in Hubert modules

Let M,JV be Hilbert A-modules, x E dv, y E M. Define the operator
M by its action on an element z E M by the formula
(2.2) := x(y, z).
Operators of the form (2.2) are called elementary operators. They clearly satisfy
the equalities
(.\ ' x,y/\*_fl
— y,x,
' /
(ii) 9x(y.u),v 9x.v(u.y) for u e M, v E dv;
(iii) = OTx,y for T E HomA(dv,r);
(iv) = for S E
We denote the closed linear span of the set of all elementary operators by K(M, dv).
The elements of IC(M, are called compact operators. In the case = M the
equalities (i)--(iv) mean that the algebra K(M) = ftC(M, M) is a closed two-sided
ideal in the Ends (M). Compact operators acting on Hilbert modules
are not compact operators in the usual sense, when one considers them as operators
from one Banach space to another. However, they are a natural generalization of
compact operators on a Hilbert space.
PROPOSITION 2.2.1. Let HA be the standard Hubert module over a unital
C HA be the free submodule generated by the first ii
elements of the standard basis. An operator K E EndA(HA) is compact if and
only if the norms of restrictions of K onto the orthogonal complements (A)' of
the submodules vanish as n 00.

PROOF. Denote by the projection in HA onto the submodule Then,

for any one has

II 1 (y, z)* (x, x) (y, z)

z)112 = 11x112 z)112

2 2 2
IIXII . . IIzhI

Since IIpnyII tends to zero, the same is true for the norm of the restriction of
the operator to the submodule hence, for the norm of any compact
operator. Let us assume now that for some operator K, one has —, 0.
 2.2. COMPACT OPERATORS IN HILBERT MODULES 19

Then, since Kem(em,z) = 0 for any for lizil 1 and

one has

(2.3) sup —*0

as n 00. If z e then Kz = z). This means that (2.3) still

holds if the supremum is taken over the unit ball of the whole module HA. Therefore
the operator K is the norm limit of the operators = U

Note that, in the case of modules over nonunital the statement of

2.2.1 is not valid.

Let denote the of compact operators on a separable Hilbert space

H. Since the algebra IC is nuclear [70], there is a unique on the
algebraic tensor product of IC by any A and we denote its completion
with respect to this seminorm by /C 0 A. Denote by 0 A the
algebra of all n x n-matrices with entries from A, where is the algebra of complex
n X n-matrices.

PROPOSITION 2.2.2. There exist natural isometric isomorphisms:

(i) /C(A) A;
(ii)
(iii)

PROOF. If a C*_algebra is unital, then statement (i) is clear. In the general

case consider the map (9a,b : a, b E A) —' A defined by the formula

Let us verify that this map is well defined. If = >1j /.Lj9c3,d3, then
= for any x E A. Therefore = The
map is multiplicative and involutive,

co(Oa,b 9c,d) = 'P(Oab,dc) 'P(0b,a) cO(Oa,b)*.

Surjectivity of follows from the possibility of the factorization a = for

any a E A (see 1.1.5). If (un), E A, is an approximate unit of the A,
then

Therefore Ikil for k = )'jOa,,b,. This means that the map can
be extended by continuity up to a map defined on the whole algebra /C(A). The
estimate

= sup
IxII1 i=1 i=1
20 2. OPERATORS ON HILBERT MODULES

shows that the map is an isometry, so statement (i) is proved. Statement (ii) can
be proved in a similar way using the map
/ ...
p—'
t ... 1S*

Finally, since there exists an isometric map from the linear space to
the linear space Mn(A) and since these spaces are dense in IC(HA)
and /C ® A, respectively, statement (iii) follows. 0
LEMMA 2.2.3. For any x e M there exists z E M and k = E ftC(M) such
that x = kz.
PROOF. Put
u := v z := limxfr+ (x,x)'13)'.
Since s2 (e + s) —1 is uniformly convergent to s on bounded sets, in order to prove
that u is well defined, note that for t = (x, x), one has
(x (e+(x,x)'13)' — x + (x,x)'13)',x (E + (x,x)'13)' — x + (x, x)h/3)_1)
= [(E + t'/3)' — + t113Y1]tI(E + t'/3)' — + t"3)11
= [(e + t1/3)' — +
The same arguments show that x = kz. 0
Note that we have also proved that M(M, M) = M.
THEOREM 2.2.4. A Hubert A-module M is countably generated if and only if
the C8-algebra ?C(M) is o-unital.
PROOF. Let IC(M) be cr-unital and let be a countable approximate unit for
it. Then

(2.4) foranyxeM.
Indeed, by Lemma 2.2.3, x = kz holds for some k E /C(M), z E M. Since __+ k
with respect to the norm, one has = kz = x.
By definition, any compact operator can be approximated by a linear combi-
nation of elementary ones. Hence, for each there exist elements and in
M such that
8(n)
1
< —, n=1,2

Let us show that the countable set x', i = 1, s(n), n = 1,2,..., generates
. . . ,

the module M. Consider an arbitrary element x E M and a number e > 0. By

(2.4) one can find a sufficiently large n such that

E 1 E
lix — <— and — —.
2 n 2
 2.2. COMPACT OPERATORS IN HILBERT MODULES 21

Then
s(n) 8(n)
—
ç Ix —
+ Fn(x)

+ =
Now let the module M be countably generated. It can be viewed as a module
over the algebra A+ obtained by unitalizing A (if it was not unital) with respect
to the action x eM, a C A, e C. If it were countably
generated over A, then it should be countably generated over A+ as well. Since in
the definition of elementary compact operators only the A-valued inner product is
involved, one has KA(M) = 'CA+ (M). Thus we can restrict ourselves to the case
where A is unital.
So let M be a countably generated Hilbert module over a unital algebra A.
By the Kasparov stabilization theorem, M HA HA. Let ,. M —' HA be
the corresponding inclusion and let ir : HA —p M be the corresponding selfadjoint
projection. Let denote the standard basis of HA. Recall that for a
the property of being o-unital is equivalent to that of having a strictly positive
element. Consider
00
:=

or, in matrix form,

11
Then is a strictly positive element in Indeed, on one hand, by Proposi-
tion 2.2.1, we have E AC(HA). On the other hand, if p: K(HA) —' C is a state
such that = 0, then p(Oe,,,e,,) = 0 for any n, since all 9e,,,e,, 0. Then, for
any x = (x1, x2,.. .) E HA, one has
*)
p(Oe,,,x0x,e,,) = p

=p ( p
/
P =0,

where the last inequality follows from

= (en. (ejxj,ejxj)(en,z),z)
= (z,en)xxj(en,z)
Thus, for any x,y,z EM,
= Z) = Ox,e,,Oe,,,y(Z)
and
Ip(9x,e,,9en,y)I p"2(Oe,,,yOy,e,,) = 0,
22 2. OPERATORS ON HILBERT MODULES

due to the vanishing of the second factor. So p vanishes on a dense subset, hence,
everywhere on 1C(HA). We have shown that is a strictly positive element of
?C(HA). Then a countable approximate unit in AC(HA) and
is a countable approximate unit in IC(M). Indeed, if k E K(M), ink k, then
tkir E and
Ilk — = IIir(tkir — = Iltkir — —i 0 (n —+ oo).
0
Complementable submodules and projections
2.3.
in Hubert
Recall that a closed submodule A( in a Hilbert M is called or-
thogonally complementable if M Ii jV'. As we have already seen, a closed
submodule in a Hilbert is not necessarily orthogonally complementable.
DEFINITION 2.3.1. A closed submodule iV in a Hilbert M is called
(topologically) complementable if there exists a closed submodule £ in M such that
jV+C=M,A1nC= {0}.
The following example shows that there exist topologically complementable
submodules that are not orthogonally complementable.
EXAMPLE 2.3.2. Let J C A be a closed ideal such that the equality Ja = 0,
a E A, implies that a = 0. Put M A
Jv':= {(b,b) : bE J}.
Then
= {(c, —c) : CE J}.
Therefore A1 EBA1-'- = M. However, the submodule £ = {(a,O) : a E A} C
M is a topological complement to in M.
Recall that we denote the nonorthogonal direct sum of Hilbert
by A decomposition into a direct sum M = allows us to define the
projection P onto A1 along £. The operator P is A-linear and, by the closed graph
theorem, is bounded, hence P E EndA(M). However, as is clear from Example
2.3.2, the projection P can be nonadjointable. But if M = Iv' C, then the
corresponding projection is selfadjoint, P E Since it is more convenient
to work with orthogonal decompositions, we would like to describe situations where
such a decomposition exists.
THEOREM 2.3.3 ([90]). Let M, iv' be Hubert A-modules and T E
an operator with closed image. Then
(i) Ker T is an orthogonally complementable submodule in M,
(ii) ImT is an orthogonally complementable submodule in A1.
PROOF. (i) Let ImT = and let T0 M be an operator such that its
action coincides with the action of T. By the open mapping theorem, the image of
the unit ball T0(B1(M)) contains some ball of radius > 0 in Therefore for
each y E M such that T0x = y and lxii IIyii. One
has
= < .
 2.3. COMPLEMENTABLE SUBMODULES AND PROJECTIONS 23

and hence,
1/2 1/2
IIyU2 = II (Tox, y) II II (x, II lxii <
We obtain that for any y E

Let us show that the spectrum of the operator does not contain the origin.
Suppose the opposite, i.e., that 0 e Let f be a continuous function on
R such that
1(0) = 1 = f(t) =0 if ItI 2 152
Using functional calculus in tile E
the S = f(T0T1fl. Then IISII = 1 and < We
can choose an element x E M such that lxii = 1, llSaii > Then the inequality

l!Sxii

contradicts the assumption (with y = Sx). So 0 Therefore the

operator is invertible and, in particular, surjective. For any z E M one can
find an element w .M3 such that Toz = Then z — E KerT and
z E KerT+ImTJ.
Since the module is obviously orthogonal to KerT, it is a complement of
Ker T. This completes the proof of (i).
(ii) Since M = Ker T Tm the submodule Tm is closed. Note that
Tm = Tm T*, so one can apply the proof of (i) to the case of the operator T*
instead of T and it gives the orthogonal decomposition = Ker T* Tm T. 0
COROLLARY 2.3.4. lIP e is an idempotent, then its image ImP is
an orthogonally complementable submodule in M.
COROLLARY 2.3.5. Let be Hubert A-modules and let F : M Al be
a topologically injective (i.e., IIFxil 2 ö lixil for some 5 > 0 and for all x e Al)
adjointable A-homomorphism. Then F(M) F(M)1 = Al.
COROLLARY 2.3.6. Let M be a Hilbert A-module and let J : M M be a
selfadjoint topologically injective A-homomorphism. Then J is an isomorphism.
LEMMA 2.3.7 Let M be a finitely genemuted Hubert submodule in a
Hubert module Al over a unital C* -algebra. Then M is an orthogonal direct sum-
mand in Al.
PROOF. Let x1 E M be a finite set of generators. Define the operator
F: by the formula = where e i = 1,... ,n, is
the standard basis. It is easy to see that F is adjointable with the adjoint F*
acting by the formula F*(x) = ((x1,x),. .. where x E Al.
,

By Theorem 2.3.3 the module Tm F = M is an orthogonal direct summand. 0

LEMMA 2.3.8. Let A be a unital HA = M a
projection and Al a projective module. Then HA = M if and only p is
adjointable.
24 2. OPERATORS ON HILBERT MODULES

PROOF. If exists, then (1 — =1 — exists as well. Therefore, by

Theorem 2.3.3, Ker(1 — p) M is the image of a selfadjoint projection.
To prove the converse, let us verify first that HA = jV' + M'. By the
Kasparov stabilization theorem, one can assume, without loss of generality, that
.Af = spanA(el,.. . , = spanA(en+1, efl+2,...). Let gi be the image of
under the projection onto M':

Since the projection induces an isomorphism of A-modules Al Mi-, the elements

gi,. . , gn are free generators and (g,,
. > OA (i.e., the spectrum of this positive
operator is separated from 0 and hence it is invertible). So, if

then
i=1

On the other hand,

1 = (ek,ek) = (fk,fk) + (gk,gk), 1— > 0,

i.e., the spectrum of the element 1 — is separated from the origin, hence this
element is invertible in A,

1
ek= k
E + M' (k= 1,...,n).
1 fk

ThusAl1 = HA.
has the form y = m+n, one has (x, y) = (x, m)+(x, n) = 0; in particular, (x, x) = 0,
thus x = 0. Therefore HA = Consider the map q
= {
which is a bounded projection since HA = Let x+y E x1 +yl E
Then
(p(x+y),x1 +yi) = (x,x1 +yi) = (x,xi),
(x+y,q(xi +yi)) = (x+y,xi) =
Therefore exists and equals q. 0

2.4. Full Hubert

Let M be a Hilbert A-module. Recall that it is full if the closure (M, M) C A
of the linear span of all (x, x), x E M, coincides with A. One can always consider
any Hilbert module as a full Hilbert module over the (M, M). The
following statement shows that full modules are easier to handle.
THEOREM 2.4.1 ([71, 36]). Let A be a cy-unital a full
Hubert A-module. Then:
(i) There exists a Hubert A-module Al such that 12(M) HA Ef3A(. If A is
unital, then there exist a number n and a Hubert A-module Al' such that

(ii) If the module M is countably generated, then 12(M) HA.

 2.4. FULL HILBERT CS-MODULES 25

PROOF. Consider the set

s= {cEA: 1cM EM}.

To prove the theorem, two following lemmas are required.
LEMMA 2.4.2 ([16]). For any a E A, a 0 and any c > 0, there exists cES
such that 11(1 — c)aII
PROOF. Since the module M is full, one can find a finite set of elements m E M
such that
— k
(2.5) <E/2.

Put

= yj(E2 (yj,yj))"2, i = 1,... ,k;

b=>2(yj,yj).
Then
IIcII = + b)"2b(E2 + <1
hence c E S. Let 1(t) := E4t2(E2 + t)2. Applying this function to the element b
we obtain the estimate
IIf(b)II = k4b2(E2+bY2II = 1k2(E2+bY1b2E2fr2+bY1II
= 11(1 — c)b2(1 — <E2/4.
Therefore 11(1 — c)bII < E/2 and, together with the estimate (2.5), this proves the
lemma. 0
LEMMA 2.4.3 ([161). There exists a sequence (xe), E M, such that the
sequence of partial sums of the series (xi, x2) is a countable approximate unit
of the algebra A. If A is unital, then there exist a finite number k and elements
E M such that k = 1.
PROOF. Consider first the case of a unital Then, by Lemma 2.4.2,
one can find an element c E S such that Ill — dl < 1/2. Therefore the element c is
invertible and c = (y3, for some k and some Yj E M. Put = c_h/2; .

then (x3,x3) = 1.
In the case of a nonunital let h E A be a strictly positive element.
By induction we shall construct a sequence (c3) in S such that

i; <

By Lemma 2.4.2, we can find an element c1 E S such that 11(1 — ci)hll < Under
the assumption that the elements c1, . . , Ck are already found, by Lemma 2.4.2 we
.
26 2. OPERATORS ON HILBERT MODULES

can find an element d e S such that

k 1/2
1
(2.6) h

and put
/ k / k

1/2

d 5, we have ck+1 E S. Since Ildil < 1 and

Ck+1 < 1— c3, we have c3 < 1. Finally, inequality (2.6) implies that
k+1 k 1/2 k 1/2

(i d)(1 <

thus completing the induction step. So we obtain that

as k —* oc. Since the strictly positive element h generates the whole A

[104], the lemma is proved. 0
Let us continue the proof of the theorem. By Lemma 2.4.3, we can choose a
sequence (x2) in the module M such that the partial sums of > (xi, form an
approximate unit in A. Define the map T: A 12(M) by the formula
(2.7) T(a) = (xia,. ..,xka,...), a E A.
Since the series (Ta,Ta) = = a*a is convergent in A, one has
T(a) E 12(M). Moreover, the adjoint operator is well defined, T* 12(M) —+ A,
T* (y2) = (Xj,Yj) E A for (yr) E 12(M). Since T* T acts identically on A, the
.

operator T is an isometry and

12(M) = Im T Ker T* A
where = Ker T*. This completes the proof of statement (i) of the theorem for
the case of a nonunital In the case of a unital the previous
reasoning can be applied literally if we replace the module 12(M) by and replace
the infinite sequence in (2.7) by a finite one.
We pass to the proof of statement (ii). Renumber tile sequences in the module
12(M) with the help of a bijection N —+ N x N. Then elements of the niodule
12(M) can be viewed as sequences (rn23), M, and, for each i N, the set of
sequences (mj3),j E N, is isomorphic to the module 12(M). Such a renumbering
defines an isomorphism 12(M) 12(12(M)). Taking into account the isomorphism
12(M) A .N, we conclude that
12(M) 12(12(M)) 12(A = 12(A) 12(AI).
 2.5. DUAL MODULES. SELF-DUALITY 27

Note that the Hubert module is countably generated, hence

12(JVI) 12(A) 12(jV) 12(A)

by the Kasparov stabilization theorem. U

2.5. Dual modules. Self-duality

For a Hubert A-module M let us denote by M' the set of all bounded A-linear
maps from M to A. The structure of a vector space over the field C is introduced
by the formula f)(x) := Af(x), where E C, f E M'. x E M. This definition
may seem artificial, however it is convenient since it allows us to define a linear
inclusion of the module M into M' (there is also the alternate approach: to define
as the set of all anti-linear maps from M into A). The formula
(f.a)(x)=a*f(x),
where a e A, introduces the structure of a right A-module on M'. This module is
complete with respect to the norm hf II = sup{jjf(x)II lixhl :1}. Such modules
are called dual (Banach) modules. The elements of the module M' are called
functionals on the Hubert module M. Note that there is an obvious isometric
inclusion M c M', which is defined by the formula x (x,.) = Sometimes we
shall write (f, x) instead of f(x).
DEFINITION 2.5.1. A Hilbert module M is called self-dual if M = M'.
The condition of self-duality is very strong. Below we shall see that there are
quite a few self-dual modules: any Hilbert module over a A is self-dual
if A is finite dimensional. Self-dual Hilbert behave quite like Hubert
spaces. In the same way as in the case of Hilbert spaces, the following statements
can be easily checked.
PRoPosITIoN 2.5.2 ([100]). Let M be a self-dual Hubert A-module, Al an ai-
bitrary Hilbert A-module and T : M a bounded operator. T E
Then there exists an operator T* Al : M such that the equality (x, T*y)
(Tx, y) holds for all x E M, y E Al.
COROLLARY 2.5.3. Let M be a self-dual Hilbert A-module. Then EndA(M) =

PROPOSITION 2.5.4. Let M be a self-dual Hilbert A-module and let M C Al.

Then Al= MeM-'-.
PROOF. Since M is self-dual, i : M —+ Al is an isometric adjointable inclusion.
Therefore M = iM has an orthogonal complement by Corollary 2.3.5. U

If A is a unital C*_algebra, then the Hubert module is obviously self-

dual. For an arbitrary module this is not true; moreover, the Banach module M'
may not admit the structure of a Hubert at all. A description of the dual
module for the standard Hubert module HA is given by the following statement.
PROPOSITION 2.5.5. Consider the set of sequences f = A, i E N,
such that the norms of partial sums are uniformly bounded. If A is a
28 2. OPERATORS ON HILBERT MODULES

unital then this set coincides with the action off on elements of
the module HA is defined by the formula

(2.8) f(x) =

where x= (x1) E HA, and the norm off is given by the equality

(2.9) 11112 = sup

f and let e be the standard basis in HA. Put 12 (f(ej))*.

We shall show that the sequence (f2) determines a functional f in a unique way. Let
us assume that there exists a functional g f, g(ej) = 1(e1). Choose x E HA such
that 111(x) — g(x)iI = C 0. Denote by x under the projection
onto the submodule LN(A) C HA, x(N) = = (x1,... ,XN,0,...). Let us
find a number N such that
1/2
'ç-' C
2(111 II +

Since = g(x(N)), we have IIf(x — — g(x — x(N))M = C. But, on the

other hand, we have
f(x — x(N)) — g(x — x(N)) (11111 + — x(N)

<(11111 + C/2.

The obtained contradiction shows that f = g. The Cauchy—Bunyakovskii inequality

2
N N N
(2.10)

shows that

(2.11)
N

Note that if we take x2 = ft/ then equality is obtained in (2.10).

Put 1(N) = (fi, f(N) E LN(A)' LN(A). It is clear that
. . . .

(2.12) IIfII

But = hence (2.11) and (2.12) imply (2.9). Since for any
e > 0 one can find a number N such that the estimate
N-f-n N-f-n N+n N+n
111112
< If 112

holds for all n > 0, the convergence of the series (2.8) follows. Li
 2.5. DUAL MODULES. SELF-DUALITY 29

Note that for the functional f = from Example 2.1.2, the partial sums
are uniformly bounded, however the corresponding series is not conver-
gent.
Let us describe an interesting example of a dual module.
EXAMPLE 2.5.6 ([44]). Let A = 13(H) be the algebra of all bounded operators
on a separable Hilbert space H. Consider pairwise orthogonal projections E A,
i E N, such that the series converges to 1 E A, and each projection
pj is equivalent to 1. We can consider H = as an orthogonal sum of Hilbert
spaces isomorphic to H, : H —+ H2 being isometries, so that

p, = 1 =

As was shown above (see Proposition 2.5.5),

l2(A)' = {{ai}
a A with respect to the inner product
:=

The maps

: 12(A)' —' A, : {a,} —+

define an isometric isomorphism of A and 12(A)'.

Let be a positive linear functional on A. If M is a Hilbert A-module and
if = {x E M : c.p((x,x)) = O} is a linear subspace of M, then is a
pre-Hilbert space with the inner product (., given by the formula
x,yEM.
We denote the norm defined by this scalar-valued inner product by and the
Hubert space obtained by completion of with respect to this norm by
Let f E M'. By Theorem 2.1.4 and Corollary 2.1.6, we have
f(x)*f(x) If
for all x E M. Therefore, if x E then
= 0 =
Hence, the map
(2.13) x+ '—p
defines a linear functional on Since
1/2 (x)*f(x)) 1/2 1/2
< < III II ço((x, x) )112
1/2
= lU If lix + II

the functional (2.13) is bounded. Then there exists a unique vector E such
that ff11 and + = ço(f(x)) for all x E M. For x E M,
we shall denote by the functional (x,.) E M'. Note that = y+ for all
y E M.
30 2. OPERATORS ON HILBERT MODULES

Let be another positive linear functional on A such that ço. Then

c and the natural map x + '—* x + N,1, can be extended to the map

It is easy to see that = It turns out that the same holds for all
elements of lvi'.
PROPOSITION 2.5.7. Let M be a Hubert A-module and let and be positive
linear functionals on A with Then = f,p for any functional
f E M'.
PROOF. Let f M'. Since the quotient space M/N,,O is dense in one can
choose a sequence of elements in such that + — 0.
Then
= urn + = + Ne,).
To prove the statement, it is sufficient to show that x)) —i for all
x E M. But
— f(x))12
< INI'II x)(x, yn) — (yn, — f(x)(x, yn) + f(x)f(x)t)
< — (yn,x)f(x)* — + f(x)f(x)*)
for each n E N. Since
p((yn, x)f(x)*) = p((yn, x (f(x))*)) ço(f(x (f(x))*)) =
.

it is sufficient to show that x)(x, — f(x)(x, —* 0. Note that

(P((Yn,X)(X,Yn) — = (X,yn)) — f(x.
= (yn + — (x,yn) +
and the sequence {x . (x, + is bounded with respect to the norm I,.
Indeed,

(X,yn) + = (x. (X,yn)))

= co((yn,x)(x,x)(x,yn))
2

. w(11x112 . (y,,,, yn))

Ilyn +
and the sequence is bounded. Since + — 0, the statement
is proved. 0
It turns out that the property of a Hubert over a unital
being self-dual is very close to the property of being finitely generated and projec-
tive. We complete this section with a complete description [131] of that interplay
in the commutative case.
DEFINITION 2.5.8. A A is called module-infinite (MI) if each count-
ably generated Hubert A-module is projective and finitely generated if and only if
it is self-dual.
 2.5. DUAL MODULES. SELF-DUALITY 31

Recall that a projective finitely generated module over a unital is

always self-dual.
DEFINITIoN 2.5.9. A commutative unital C* algebra A = C(Y) is said to be
DI (divisibly infinite) if for any infinite sequence u2 of elements of norm 1 IItLiII
C > 0 in A there exist a subsequence i(k) and elements 0 bk e A of norm 1 such
that
(i) the supports of bk in Y are pairwise disjoint;
(ii) for each k there exist points such that bk(yk) = 1, 0
U?(k)(yk)I 5, 5, and the sequences {yk} and have a common
accumulation point. In particular, is a discontinuous function for any integer
s>1.
THEOREM 2.5.10. If a commutative unital A is DI, then it is MI.
PROOF. We have to prove that any countably generated self-duai Hilbert A-
module M is finitely generated and projective. By the Kasparov stabilization
theorem 1.4.2, one has M EBI2(A) 12(A), where 12(A) is the standard Hilbert
module (see p. 6). Denote by p : 12(A) —+ M C 12(A) the corresponding orthogonal
projection and let : 12(A) E3 N be the orthogonal projection onto the first
j standard summands of 12(A). Then either 1)11(1 — p3)pII 0 as j cc or 2)
this is not the case.
1) Let us show that in this case M is finitely generated and projective. One
can argue as in 1921: for a sufficiently large j, the operator

J(kXI\fpJ(x) ifxEM,
x ifxEM-12(A),
is close to the identity, hence it is an isoniorphism. It maps M isomorphically onto
a direct summand of E3.
2) In this case (changing the standard decomposition, if necessary) we can find
a sequence j(k) such that for each k there exists an element zk E M of norm 1
such that its j(k)-th entry (with respect to the standard decomposition) has
norm greater than some fixed C > 0. Let us choose functions bk as in Definition
2.5.9 (for simplicity we assume that we do not need to pass to a subsequence for
the second time). Then the formula /9(x) = >k defines a functional on 12(A).
It is evident that this functional is not adjointable on 12(A) since bk 74 0 as k cc.
Let us show that there is no adjointable functional on 12(A) with =
so that /31M is a nonadjointable functional on M and M is not a self-dual module.
Indeed, suppose that there exists an element a = (a1, a2,...) E M C 12(A) such
that := = /3(x) for any x E M. Then
(2.14) (a,a)

(the last element is a bounded measurable discontinuous function; see Definition

2.5.9 (ii)), since for any continuous positive function fk equal to I on supp
and equal to 0 on Ui>k supp b3 (the existence follows from normality of Y), one
has /3fk = + (/9 — But /3fk is an element of 12(A) (or, more precisely,
an adjointable functional /3fk(x) = Let bfk be the corresponding
element. Then bfk = afk + (b — a)fk is the decomposition corresponding to
12(A) = M e12(A). Indeed, a E M and ((b — a)fk,x) = 0 for any XE Al, since
32 2. OPERATORS ON HILBERT MODULES

a(x) = fl(x) for those x. The Pythagorean theorem for Hubert. shows
that (a, >Jj<k Taking all the fk's, one obtains (2.14).
On the other hind,
(a,a)(y2) 2 (a,zj)(zj,a)(yj) = = 262
m,k
But (2.14) implies that (a, a) does not belong to A. This
contradiction proves the theorem. D

Now we describe some consequences of this theoreni.

THEOREM 2.5.11. A commutative separable unital A is MI if and
only if its Gclfand spectrum Y has no isolated poiats.
PROOF. If V has an isolated point, then a separable Hilbert space arises as one
of self-dual modules, hence, MI does not hold.
Now, suppose that Y has no isolated points, in particular it is infinite. Since V
is a compact Hausdorif separable space, the topology is generated by some metric
p. For any given sequence of elements with noriii 2 C we can find a sequence of
different points such that > 2C/3.
I

Since Y is compact, one can pass to a convergent subsequence yk = Yi(k). For

a convergent sequence, by induction, we can choose Ek such that the corresponding
Uk of yk are pairwise disjoint. Then we can choose Yk
inside these neighborhoods so close to yk that IUi(k)I 2 C/2. Finally one can
choose functions bk such that bk (Yk) = 1 and supp bk C Uk \ It remains to take
6=C/2. D

COROLLARY 2.5.12. Suppose that X is a compact connected separable Hausdorff

C-space. Then is MI if and only if X/G has at least two separated points.
PROOF. The Gelfand spectrum V of is a quotient space of X/G with
respect to the equivalence of nonseparated points. Since X is connected, Y is
connected too. So, by Theorem 2.5.11, GG(X) is MI if and only if Y is not reduced
to one point. D

2.6. Banach-compact operators

DEFINITION 2.6.1. Let be Hilbert A-modules and let M' he the dual
module for M. Consider the closure BK(M , of the linear span of all operators
= y . 1(x)
.rE M, y E f M', in the Banach space HomA(M,J'f). We call the elements
of the set Banach-compact operators.
hi the case JV' = M the set BK(M , N) is equipped with the natural structure
of a Banach algebra. If T e EfldA(M) is a (not necessarily adjointable) operator,
then the equalities
Tx = y f(Tx) = 9y,foT(X), TOv,i(x) Ty. f(x) = OTy.f(X)
show that BK(M) is a two-sided ideal iii the Banach algebra EndA(M).
hi the case of the standard Hilbert module over a unital we give
one more (geometric) description of compact and Banach-compact operators. Let
 2.7. C-FREDHOLM OPERATORS AND INDEX. MISHCHENKO'S APPROACH 33

S C HA be a bounded set. We call it A-pre-compact if for each E > 0 there exists

a free finitely generated A-module N N C HA, such that dist(S,N) <e.
PRoPosITIoN 2.6.2. Let T e EndA(HA) (resp. T E Then the
following conditions are equivalent:
(i) T E 13K(HA) (resp., T E
(ii) the image T(Bl(HA)) of the unit ball Bl(HA) is A-pre-compact.

PROOF. If statement (i) holds, it is sufficient to prove that one can find an
approximating module N for a finite set of elements in HA and this
can be easily done by the Dupré - Fillmore method, as in the proof of Theorem
1.4.5. So suppose that (ii) is fulfilled. Then for any E > 0, one can find elements
b1, . , bk E HA, generating the module N C HA, such that (b2, b3)
. . öjj and
diSt(T(Bl(HA)),.Af) <E, where B1(HA) is the unit ball of HA. Denote by PAr the
projection onto N and consider the operator PET. It can be decomposed as
(2.15) = bi(fi,x) + ...
where E Since x E Bl(HA), one can find an element b N such that
IITx — bII <E. Therefore
IITx—PgTxII=IITx—b+b—PgTxII
(.216 ) = IITx - b[I + IIPAr(b - Tx)II <E + E = 2E,

hence IT — PNTII <2e and T lies in the norm closure of the set of operators of the
form (2.15). If T is adjointable, then PArT is adjointable as well, hence E HA
and T is compact. 0
2.7. C*.Fredholm operators and index. Mishchenko's approach
The material of this section is taken mainly from [92]. Recall first the definition
of K-groups.
DEFINITION 2.7.1 ([60, § 11.1]). Let lvi be an Abelian monoid. Consider the
Cartesian product M x M and its quotient monoid with respect to the equivalence
relation
(m,n) (in',n') (m,n) + (pep) = (m',n') + (q,q).
This quotient inonoid is a group, which is denoted by S(M) and is called the sym-
metrization of M. Consider now the additive category 1'(A) of projective modules
over a unital A and denote by [M] the isomorphism class of an object
M from P(A). The set of these classes has the structure of an Abelian
monoid with respect to the operation [MI + [N] = [M eN]. In this case the group
is denoted by K(A) or K0(A) and is called the K-group of A or the
Grothendieck group of the category P(A). If A has no unit, then the natural map
—* C induces a map of K-groups and K0(A) is defined by

K0(A) Ker(Ko(Aj —+ Ko(C)).

The groups := K0(A ® C0(RT')) for integer n > 0 turn out to be 2-periodic
in 11 and the definition caii be extended to n E Z by periodicity. For unital
algebras one could use classes of stable homotopy of projections in (i.e., allowing
addition of direct suinmands) instead of classes of isomorphic projective modules.
34 2. OPERATORS ON HILBERT MODULES

More precisely, projections p: A'1 —' A'1 and q : are equivalent if there
are m' and n' such that n + n' = in + m' = s and the projections
(p 0
0
A8=A'1EDA'1 —i =A8,
(q 0
0
A8 = Atm A"1 —i Atm Atm = A8
can be connected by a norm continuous homotopy in the set of all projections from
End(A8) = End*(A8) (A is unital!). One can also consider the equivalence classes
of projections in the algebraic sense. The details can be found in [10, 60, 94, 134].
Recall the following well-known statement.

LEMMA 2.7.2 (cf. [115, Theorem 4.151). The set of epimorphisms is open in
the space of bounded linear maps of a Banach space E to a Banach space C.
LEMMA 2.7.3 ([92, 1.41). Let A( be a finitely generated Hubert A-module over
a unital A and let a1,... ,a5 be generators of J'.f. Then there exists a
number E > 0 such that if some elements aç,.. a'8 E .N satisfy. ,

Ilak—akII <E
then these elements also generate
PRooF. The map
f L5(A) —'.N, (0,...,O,i,0,...,0)
an epimorphism. Hence, by Lemma 2.7.2, there exists E > 0 such that if
is <
SE, then g is an epimorphism. Let

g is an epimorphism and the elements aç,.. . , a'8 generate .M. 0

In this section we always assume that the algebra A is unital. Recall the
definition of a Fredholm operator suggested by Mishchenko in [92].
DEFINITION 2.7.4. A (bounded A-linear) operator F : HA HA is called
A-Fredholm if

(i) it is adjointable;
(ii) there exists a decomposition of the domain, HA = and the
range, HA = (where M1, M2, , are closed A-modules and
ft/1,.M2 have a finite number of generators), such that F has the matrix
form F with respect to these decompositions and F1
= ( F2 )
M2 is an isomorphism.
 2.7. C-FREDHOLM OPERATORS AND INDEX. MISHCHENKO'S APPROACH 35

THEOREM 2.7.5 ([92]). Let HA where M and N are closed A-

modules arid N has a finite number of generators a1,. . . , a8. Then N is a projective
A-module of finite type.
PRooF. By Lemma 2.7.3, there exists some r > 0 such that the estimate
— akII < E, a'k E N, k= 1, . . . , 5,

implies that generate N. Let P HA N be the projection along the

summand M in the module HA. Then P is bounded and A-linear. Therefore there
exists some 6 > 0 such that lxii < 6 implies IlPxlI < &. Choose a number n0 such
that
llak—akll <6, k=1,...,s,
where is the projection of ak onto along Write with respect to the
decomposition HA = as the sum
— I II I AP II
ak = ak + ak, ak E ./V, ak E
Then ak — = P(ak — ak). Therefore llak — <e and generate N. Let
V be the module generated by Then HA coincides with the sum M + V.
Indeed, if x E HA, then
X = + = (XM — +
Consider the operator Q projecting onto along Then
Q(ak)=ãk, Q(N)=V,
= P(V) = N.
Since are close to the composition of operators

is an isomorphism. Therefore Q : N V and P : V

8
N are isomorphisms. In
particular, if Aka'k = 0 holds, then Akak = 0. Therefore M n N = 0, i.e.,

HA = It is clear that V is a closed A-submodule in HA and

= (M fl
Indeed, (M fl L.,LO) n V = 0; on the other hand, if x E then x = x' + x",
x' E V, x" E M. Since V C one has x" e i.e., x" e M fl Thus N is
isomorphic to a direct summand in the free finitely generated A-module Li

THEOREM 2.7.6 ([129]). fn the decomposition mentioned in the definition of

a Fredholm operator (see 2.7.4), one always can assume that the modules M0 and
M1 are orthogonally complementable. More precisely, there exist decompositions
for F,

such that Vo = HA, V1 = HA or (which is the same by Lemma 2.3.8)

such that projections V0eW0 V0 and V1 Vi are adjointable.
36 2. OPERATORS ON HILBERT MODULES

PROOF. Let W0 = .N3, V0 Wi-. The orthogonal complement exists by

Lemma 2.3.7. The restriction Fl is an isomorphism. Indeed, if E We-, then
let = + 4, 4 E M0, 4 E W0, = 1. Let us assume that llFxnII 0;
then 11F4 + F411 —' 0. Since F4 E M1, F4 E = HA, this
means that 1F411 —* 0 and 11F41l —i 0. The operator F1 is an isomorphism,
hence 11411 —.0. If a1,.. .,a8 are generators for W0 = Al0, then 0 = =
(4,a3) + (4,a3),
= n—-.oo, j= 1,...,s.
Since 4 4 --. 0 (as n —' oo) and
with the condition lIxflhI 1 shows that
= 4 + 4 --. 0. This contradiction
is an isomorphism. Let V1 = F(Vo).
Since W0 = Al0, one can assume that W1 = Indeed, any y E HA is of the
form y = m1 ± n1 = F(mo) + n1, where m1 E M1, E Alj, m0 E M0. Further,
m0 = v0 + n0, where v0 E V0, nçj E W0 = Al0, and
y = F(vo + no) + ni = F(vo) + (F(no) + ni) E V1 +Al1.
Thus HA = V1 + W1.
LetyE V1flW1 = V1nAl1, i.e., n1 =y—F(vo),ni EAl1,vo E V0. Let us
decompose Va into two summands, Va = m0 + n0, where m0 E M0, n0 E Then
n1 = F(mo) + F(no),
F(mo) n1 — F(no), F(mo) E M1, n1 — F(no) E Al1,
whence we obtain that F(mo) = 0, n1 —F(no) = 0. Since F: M0 M1, m0 = 0.
Further, V0 E Vo = Al(j1-, hence
0=(vo,no)=(mo+no,no)=(no.no), n0=0.
Thus v0 = m0±n0 = 0, y = F(vo) = 0. Therefore V1 flWj = 0 and HA =
V1 the orthogonal complement Vp-, e
= HA and the proof is complete. U

REMARK 2.7.7. If we do not require the operator F to be adjointable, then there

still exists a decomposition F : Alo'EEAlO —' where = spanA(el, . , en), . -

but M1 does not necessarily have an orthogonal complement. This result was
obtained in [54].
DEFINITION 2.7.8. Let the conditions of Definition 2.7.4 hold. By Theo-
rem 2.7.5, Al1 and are projective A-modules and we can define the index of
Fby
index F = [Al1] — [Al2] E K(A).

In [134] this index is called Mindex. We suggest this notation be interpreted as

emphasizing Mishchenko's role in introducing the index of operators.
THEOREM 2.7.9. The index is well defined.
PROOF. It is necessary to check that the index does not depend on decompo-
sitions of the range and the domain in Definition 2.7.4. Let be the projection
onto Lm along Let F be an A-Fredholm operator and let HA = M1eAl1 and
 2.7. C-FREDHOLM OPERATORS AND INDEX. MISHCHENKO'S APPROACH 37

HA = be decompositions of the domain and of the range, respectively.

Then
F_(F1 0 "\
0 F4)'
where F1 : —÷ M2 is an isomorphism. According to the proofs of Theorem 2.7.5

and Theorem 2.7.6, one can assume that

c= M1 = P1
where Pi is a projective finitely generated A-module. Let another decomposition
of the domain and the range be given:

Then there exists m n such that

Lm =
where is a projective finitely generated A-module. This is exactly the result of
the proof of Theorem 2.7.5.
Let us show that there exists in> n such that if
= F(Lm) +M, and : HA —' HA
is the projection on along = then
=
where P2 is a projective finitely generated A-module, and

where is a projective finitely generated A-module. Indeed, HA = If

a1,. .. , ak are generators of the module ..N, then
— I II
a3—a3+a3, I TI I! •_i
For m —+ oo we have la/Il —' 0 since a7 = where x is arbitrary,
x onto and —p 0 as m oo. Then for m sufficiently large,
we have
= fl
(the proof repeats the proof of Theorem 2.7.5). Similarly
= fl 1S4.

Since m n, we have Lm N1 where is a finitely generated projective

A-module. From the equalities

F C M1
that F : Therefore we have the following equalities in K(A):
[JV1] + = + [Pfl = [LmI, = [P2],
+ [22] = VI;] + = [Pfl =
Thus [.N1] — [..V2] = — and we have proved that the index is well defined. 0
1This projection is well defined since C Mi form n and hence FIL.L is an isomorphism,
whence it follows that HA is a closed A-module, L'm fl = 0, L'm + = HA. Therefore
HA = is a direct sum of closed A-modules and Q'm is bounded and A-linear.
38 2. OPERATORS ON HILBERT MODULES

LEMMA 2.7.10. Let an operator F : HA —4 HA be A-Fredholm. Then there

exists a number e> 0 such that any adjointable operator D satisfying the condition
IF — DII <r is an A-Fredholm operator and indexD = index F.

PROOF. By the definition of Fredhoimness

F= (
F1 : M2. Then IIF1II < IIFII; moreover, if D : HA —' HA is an arbitrary
operator, then
D_(Di D2'\
D4)'
arid there exists a constant C such that lID1 II <GilD II (cf. [92]). Therefore, if D is
an arbitrary operator satisfying the estimate hF — Dii <e, then hF1 — Dill < C•e.
Since F1 is an isomorphism, we can find 5> 0 such that if liP'1 — D1 <5 and D1
is an operator, then D1 is also an isomorphism. By taking e = 5/C we obtain that
for the operator

D4)'
the element D1 is an isomorphism. Then
(D1 0

where

( ? ): —'

(1
= ):
are isomorphisms. Using U1 and U2 we obtain a new decomposition of the domain
and the range into direct sums:
M=U1(M1),

With respect to the new decomposition the matrix of D is equal to U2DU1. Thus
D is A-Fredholm with the index
— = [.N1] — = index F.
0
LEMMA 2.7.11. Let F and D be A-Fredholm operators,

Then DF : HA HA is an A-.Fredholm operator and index DF = index D +

index F.
PROOF. For F and D, consider decompositions from the definition,
HA = HA,

HA
 2.7. OPERATORS AND INDEX. MISHCHENKO'S APPROACH 39

where
F_(F1 0 D_(D1 0
0 F4)' 0
F1 and D1 are isomorphisms, and AIj , Al2, Al11, are projective finitely generated
A-modules. As earlier, without loss of generality we can assume that

M2 is an isomorphism, one can change the decompo-

sition into direct sums by setting
M2=L;t,
Thus a number n can be chosen as large as necessary. Choose n in such a way that

where, as before, : HA —4 HA is the projection onto along Then

ri ,&tl
=
M2 = = P'EpN1'. With respect to the new decomposition, HA =
the matrix of the operator F has the form

0 F4)'
and F1 is an isomorphism. Then
(F1 F2)(1 (Fi 0)
_Fç1F2
Denote by U the matrix ( and put = U(V1).
)
We obtained a new decomposition, HA = M1 and the matrix F with respect
to the decomposition
HA = = HA
has the previous diagonal form. Consider the projection
T: HA = -*
Since HA the restriction M'2 : is an isomorphism.
Consider the matrix D with respect to the decomposition
HA = —p-' = HA.
This matrix has the form D ) , where D1 is an isomorphism. Put
= (
.( 1 o\(D1 0 0
D4
Therefore one can change the decomposition of the range,

in such a way that the matrix of the operator D with respect to the new decompo-
sition
= =
40 2. OPERATORS ON HILBERT MODULES

has diagonal form. Let us change the decomposition of the range once again:
= D(M2), =
The matrix D with the respect to the latter decomposition

HA = = HA
has diagonal form. Then the composition DF with the respect to the decomposition
HA = M1E V1 -. = HA

has the form DF (DF)4 ),

and (DF)1 is an isomorphism. Taking
= (
into account the fact that End* HA is a we conclude that DF is an
A-Fredholm operator and
indexF = — [V2], indexD = [V2] —

indexDF = —

It follows that index DF = index D + index F. 0

LEMMA 2.7.12. Let K : HA —i HA be a compact operator. Then 1 + K is an
A-Fredholm operator and index(1 + K) = 0.
PROOF. It is clear that 1 + K is adjointable. Choose a number n such that
the inequality < 1 is fulfilled. With respect to the decomposition HA =
we have the following matrix presentation:
K2'\ K2
K4)' ( + 1
K3 1+K4
By the estimate < 1, the operator 1 + K1 is invertible, hence, as before,
there exist invertible operators U1 and U2 such that

U2(1 + K)U1 =
(1+K1 0
0 (1 + K4) — K3(1 +

Then, with respect to the new decomposition, HA =

M1 M2 = = the operator
(1 + K) has the diagonal form and is thus an A-Fredholm operator with
index(1 + K) = — = 0.

LEMMA 2.7.13. Consider an A-Fredholm operator F: HA HA and let K e

'CA. Then the operator F + K is A-Fredholm and index(F + K) = index F.

PROOF. Consider decompositions of the module HA into direct sums such that
the matrix F has the diagonal form:
HA = = HA.
Without loss of generality we can assume that
= =
 2.7. C-FREDHOLM OPERATORS AND INDEX. MISHCHENKO'S APPROACH 41

where is a finitely generated closed A-module. Choose a number ri. sufficiently

large to provide that IKIL.1. II < IIFi'IL'. Consider the new decomposition of the
module HA:
= = L,1, M2 = =

LetF= ( ) andK= ( bematricesofFandKwithrespect

)
to the decomposition HA = —÷ = HA. Then
F+K_(F'1+K1 K2
K3 F4+K4)'
and the operator F1 + is invertible. By repeating the construction of Lemma
2.7.10 (about operators close to a Fredholm operator), we obtain
index(F+ K) = [v'] — = [La] — IF(2i)
= [ft/i] + [P1] — [Pi] — = index F.
0
THEOREM 2.7.14. Let
F: HA . HA, D: HA —' HA, D' : HA —4 HA
be adjointable operators such that
FD = IdHA +K1, D'F = IdHA +K2, K1, K2 AC(HA).
Then F is an A-Fredholm operator.
PROOF. Consider a decomposition of HA, for which the operator FD = 1HA +
K1 has the diagonal form (Lemma 2.7.12)
HA = M1 M2 = HA,
and assume that the decomposition of HA satisfies the conditions of Theorem 2.7.6.
Consider the projection
P:HA —*j\/2.
Viewed as an operator on HA, it is a compact operator. The image of the operator
(1 — P)(1 + K1) = (1 P)FD is exactly equal to M2. It is easy to see that, up to
an isomorphism,
(1-P)FD=(1--P)(1+K1)=1+(—P(1+K1))+K1 =1+K1,
D'(1-P)F=D'F-D'PF=1+K2,
where K1 )CA, K2 By Lemma 2.7.13, one can assume without loss of
generality that F : HA —* M2 is an epimorphism. Otherwise, we will pass to the
operator (1 — P)F. Consider now the decomposition for 1 + K2

HA
M2
HA M2 is an F M2 maps M1 isomorphically into
M2 and KerF C M2 Let us show that F(M1)flF(V1)
0. For this purpose, decompose F into a composition
-* M2,
42 2. OPERATORS ON HILBERT MODULES

where F is an isomorphism. Therefore

=
Since the A-module is finitely generated, F(X(1) is finitely generated too. We
have obtained a decomposition
HA = = HA,
where M1 is an isomorphism. 0
LEMMA 2.7.15. If adjointable operators D, D' and F are such that FD and
D'F are A-Fredholm operators, then F is an A-Fredholm operator.
PROOF. By the definition of Fredholmness of FD and D'F, we can find ad-
jointable operators T and T' such that
(FD)T =1+ K,
T'(D'F) = 1+ K'.
By Theorem 2.7.14, the operator F is A-Fredholm. For T, for example, one can take
an operator with the matrix ( ) , where FD has the form (
)
in the sense of Definition 2.7.4. 0
REMARK 2.7.16. In the case where A is a the properties of A-
Fredholm operators are more similar to the properties of usual Fredhoim operators.
This problem will be addressed in Proposition 3.6.8.
REMARK 2.7.17. For applications to elliptic operators it is important to develop
the theory for modules of the form 12(P). This can be done similarly (see [121]).
2.8. Representations of groups on Hubert modules
In this section we assume that C always denotes a compact group. First of all,
we prove an equivariant version of the Kasparov stabilization theorem. We follow
here the original proof [631. For closely related problems see also [87].
DEFINITION 2.8.1. For a B, put

®c B),

where {V2} is a countable set of finite-dimensional spaces, in which all irreducible

unitary representations C are realized (up to isomorphism) and each representation
is repeated an infinite number of times; the B-Hilbert completion of the infinite
algebraic sum is implemented with respect to the norm given by the following B-
inner product on summands:
(x1 ®bi,x2 ®b2) := (x1,x2)v, . Xi, x2 E
DEFINITION 2.8.2. A B with a continuous action of a group C is
called a -algebra.
DEFINITION 2.8.3. A Hilbert E over a B equipped
with a unitary action of C is called a C-B-module if this group action agrees with
the Hilbert module structure, i.e.,
g(xb) = g(x)g(b), (g(x),g(y)) = g((x,y)), x, y E E, bE B, g E C.
 2.8. REPRESENTATIONS OF GROUPS ON HILBERT MODULES 43

THEOREM 2.8.4 ([63]). Let B be a a countably gen-

erated Hilbert C-B-module. Then there exists an equivariant B-linear isomorphism

which respects the inner product.

PROOF. Denote by C+ the module C viewed as a and, for a
module .A/, denote by (J'f)B the same module viewed as a B-module. Let the
action of C on B+ be extended from that on B by the formula g(1) = 1. Sup-
pose that we have proved the theorem for unital so that there is an
equivariant B-linear isomorphism

C ((C (fl2+)B = 'NB,

so we can restrict ourselves to the unital case.
Let {xk} be a countable system of generators for C and let {ek} be an or-
thonormal basis of fl2 such that each e.k is of the form e, = Vk ® 12, where
vk E V8(k). In other words, if {wk} is a union of some orthonormal bases of all
then ek = wk ® 12. Let be a system of elements in C 7t2, in which each
element of the form Xk ek is repeated an infinite number of times. We can
0 or 0
assume that = 0 and put W1 = 0 V1 ® B. Suppose that, by induction, we
have already constructed subspaces W1, satisfying the following conditions:
. . . ,

(i) W, is a C-finite-dimensional C-invariant subspace in C

K(i)
(ii) each has a basis (;, z1 . ) = (fi, .. , fr,) such that
. .
, .

(iii) there exists m = m(n) such that

wi+. + Wn C Cm := Ce B)

consequently (W1
and + ) B C Cm,
(iv)the distance between and (W1 + ... + does not exceed 1/n.
Note that (1) and (ii) imply that the modules W2B are pairwise orthogonal and
C-invariant and (W1 + ... + is also C-invariant. The last module is free,
so, by Lemma 2.3.7, it has an orthogonal complement, which is C-invariant due to
unitarity of the action.
Let us pass to the construction of Put
p
, . II — /
Yn+i Yn+i Yn+i Yn+i
j=1

Then

— =0
44 2. OPERATORS ON HILBERT MODULES

for any w e (W1 + + Since by the definition of the sequence y3, the
element Yn+1 lies either in C or in some ® B, we have E Cm' for some
m' > m. Consider the orthogonal complement Sn,m' to (Wi + + in
Cm'. It is an invariant module and e Sn.m'. By the Mostow theorem on
periodic vectors [93], the elements with C-finite-dimensional orbits arc dense in
Snm'. Hence one can find a vector z E such that liz — < 2fl+2 and
the linear span 1? of the orbit Gz of z is an invariant finite-dimensional subspace of
Sn,m'. Since z is a totalizing vector, R is an irreducible G-module. Therefore there
is rn" > in' such that there exists an equivariant isomorphism r : R Vm". Let
{ h1,... hk} be an orthonormal basis of Vm" and
, := r—' (hi), i = 1,. , k. Then
. .

for the corresponding irreducible matrix representation T : C U(k), we have

9EC.

Since R c Cm' and rn" > m', R is orthogonal to Vm". More precisely, each element
of R is orthogonal to Vm" ® B in C ® flB. Hence (ri, /i3) = 0 for any i and j. Let
k k 1/2

z := a : + 1,

Then = + {(2n + and the matrix L := is

positive and invertible in Mk(C) C Mk(B). Let D = := E Mk(B) and
:= Set to be the C-linear span of vectors i = 1, . , k, or, . .

equivalently, to the span of since D has complex coefficients. Then Wm+ 1 C Cm"
and

= = = Sjj.
p,q=1

Since all Ii, and are orthogonal to W1 +... + is orthogonal to it as

well. Further, put F L'/2, so that r := ij'Fjj. Then

g = (gri +
= (2n + 2)a)

= + 0 1B). (2n + 2)a)

31 s=1 s=1

= >>Tf(g) (rs + (h5 0 (2n + 2)a)

=
j=1 s=1 3j s=1

= = E
j=1 s=1 t=1 t=1 j=1 s1
 ___________________

2.8. REPRESENTATIONS OF GROUPS ON HILBERT MODULES 45

Thus is C-invariant. Let us estimate the distance by taking z' =

so that
=
p(z, p(z, z') = - (2n
®

1 1

— (2n+2) k
1/2 (2n+2)
+1

Therefore
p(yn+i, (W1 +••• + z) + p(z,
1
<
2n+2 n+1 n
Thus, by induction, the C-subspaces W2 with properties (i)—(iv) are constructed for
all i. From the explicit expression for we obtain that W,., is isomorphic to some
i.e., is irreducible. Further, the B-Hilbert completion M (i.e., the closure in EE?flB)
of the algebraic orthogonal sum of modules B gives the whole E Indeed,
by property (iv), the algebraic sum is dense in So, M It remains
to prove that M is isomorphic to flB, i.e., that each irreducible representation is
repeated infinitely many times among the modules WEB. Suppose the opposite,
then M flB flB, or

0
Let us now prove the theorem on decomposition of representations [911. Let
M be a Hilbert B-module with a strongly continuous unitary representation C,

and suppose that the group acts trivially on B. Now let {V8} be the complete
collection of pairwise nonequivalent unitary representations of C of dimensions d8
and let be their matrix elements, which are continuous functions on C. For an
invariant normalized Haar measure dg on C, we define the operator

(2.17) :M —'M,
Since a product of continuous complex-valued functions by a continuous module-
valued function is integrable for a fixed x E M and since the group is compact,
the integral is well defined and equals some element of M. We obtain a bounded
operator. Indeed,

d8 f IIT9(x)II dg sup lxii.

46 2. OPERATORS ON HILBERT MODULES

Therefore
sup
9EG
It is well known [7, I, §7.1, Theorem 5] that
0 s s',
(2.18)
fG = { d. zm Jn, —

We need the following Peter—Weil theorem (see, e.g., [7, I, §7.2, Theorem 1]).

THEOREM 2.8.5. The functions form a complete orthonormal sys-

tem in L2(G).
LEMMA 2.8.6. The operators Pq have the following properties:
(i) P;q is adjointable and
(2.19) =
(ii) the following equality holds:
(') ')fl\ D8 — Os
I' . / pq p'q' — qp'

(iii) the following equalities hold:

(2.21) =

(2.22) =

PROOF. First of all, note that for unitary operators in M, the mapping F '—i F*
iscontinuous in the strong operator topology. In other words, for unitary operators,
the strong continuity implies the *-strong one. Indeed.
lI(F'5 — F*)xIj = II(F'' — F1)xII = IIF'(F'' — F')FzII = IIFz — F'zII —* 0.
Therefore one can take T instead of T9 in (2.17), and then take it outside the
integral. More precisely, the first equality in the chain
(P;q)* = d5
f D;q(g)T;(x) dg = f d(g1)

= d5
f dg =

has to be verified at first at the level of integral sums and then one has to take
the limit, which is possible due to the *-strong continuity. Remaining equalities
in the chain above follow from the invariance of Haar measure and the relations
T9* = = Tg1. Statement (i) is proved.
It follows from (2.17) that

=
ff dgdg'.

Since TgT9' = we obtain, by taking := gg', from

D;q(g) = = =
 2.8. REPRESENTATIONS OF GROUPS ON HILBERT MODULES 47

and from relations (2.18) that

= dg' J
JG
= =

To prove statement (iii) let us note that

dh
TgPfm(X) =
L D;m(h)Tgh(X) dh = f
d5 d8

= d5 D:m(h)Th(x) dh
f
= D:m(h)Th(X) dh
d5 f
i=1 Gj1
= =

The second equality of this statement can be proved similarly. 0

LEMMA 2.8.7. The operators P := are selfadjoint pairwise orthogonal
projections.
PROoF. If we rewrite the statement of the lemma as
(2.23) (p)* = p8 pSp8' =

then the proof follows immediately from (2.19) and (2.20). 0

LEMMA 2.8.8. Put

P8

The operators P8 have the following properties:

(2.24) (ps)* = PS,

(2.25) =

(2.26) T9P8 = P8T9.

In other words, P5 are selfadjoint invariant pair-wise orthogonal projections in M.
PROOF. By the definition of formulas (2.24) and (2.25) immediately follow
from (2.23). To verify the third relation consider the character of the representation
vS,

x8(g) :=>D,,(g),
48 2. OPERATORS ON HILBERT MODULES

which, being a trace, satisfies the relation = One also has

P8 =dsfxS(g)Tgdg,

T9P8 = dsTgj X8(g')Tg' dg' =

= d8
f dg'T9 = P8T

LEMMA 2.8.9. Define

(2.27) M8 P8(M), M' :=

where the sum is supposed to be completed either as a Hubert sum. or as a closure
in M of the algebraic sum (which is the same). Then
(2.28) M' = M.
PROOF. Assume that a C-linear functional f on M vanishes on M and that
x E M is an arbitrary vector. Then, for any set of indices, we have E M',
so that
0=
f
Therefore, by the Peter—Weil Theorem 2.8.5, f(T9 (x)) = 0
dg.

holds almost everywhere

and, by continuity, it vanishes everywhere. In particular, = f(x) = 0.
Hence, by the Hahn—Banach theorem, M = M. 0
THEOREM 2.8.10 ([91]). Let M be a Hubert B-module equipped with a strongly
continuous unitary representation of C and let the group act trivially on B. Now
let {V8} be the complete collection of pairwise nonequivalent unitary representations
of C and let
M8 := Hornc,c(V8,M) C Homc(V8,M) ®M
be a Hilbert B-module with the inner B-valued product defined by the formula

:= orthonormal basis for V8.

i,j=1
Then, for the Hubert sum, we have an equivariant B-linear isomorphism

vEV8,çoEM8,

and
r(V8®M8) =M5,
where M8 is defined by (2.27).
PROOF. To begin with, let us note that the r8 are algebraically injective. In-
deed, let

o=r,
 2.8. REPRESENTATIONS OF GROUPS ON HILBERT MODULES 49

Since, by the Schur lemma, is either an isomorphism or 0, the above equality can
be true only if either = 0 or = 0. But then ® p = 0.
By Lemma 2.8.9, it is sufficient to prove only that r5 maps ®M5 bijectively
onto M8.
Note that, by setting := P8(M) we obtain, by relation (2.20),
that the operators induce isoinorphisms

Thus M8 = M is a sum of isomorphic modules.

Let { . . . , } be the orthonormal basis of with respect to which the
matrix elements were defined. Define the homomorphism
(2.29) : [Mfl —* M5, ® x) =
where we have put in square brackets to emphasize that there is no action of
G on it. By the properties of the operators the map V is an isomorphism.
Since, by (2.21),

®x) = T9Pf1(x) =

and
V ®
=
V map
V : —, W5(x)(v) := V(v 0 x).
Then
o Ox) = V(v ® x).
OWS)(v
Since we have an isomorphism on the right-hand side and since is algebraically
injective, r5 is an isomorphism (see Lemma 2.8.12), whence V is an isomorphism.
In particular, the image of coincides with M8 and these images are orthogonal to
each otl1er. Hence r is topologically injective and its image coincides with M. 0
REMARK 2.8.11. Let the G-A-module M belong to the class P(A) of projective
finitely generated modules. Then obviously M8 = Honxc(V3,M) e P(A). Let us
show that only a finite number of summands do not vanish in the sum Denote
generators of M by a1,.. . , a8. Using the Mostow lemma [93], choose C-periodic
vectors b1,. .
. , so close to a1,.. . , a8 that they generate M as an A-module (see
Lemma 2.7.3). By decomposing the linear span of the orbit Gb3, which is a finite-
dimensional C-C-module, into irreducible modules, let us find a new system of
generators c1,.. . now belonging to irreducible G-C-modules. Then it is evident
,

that the number of nonzero summands does not exceed N.

LEMMA 2.8.12. Let F: L M, T : N L be continuous maps of Banach
spaces, S = FT an isomorphism and Ker F = 0. Then F is an isomorphism.
PROOF. Since S is an isomorphism and F is bounded, T is topologically mice-
tive and its image T(N) is closed in L. Suppose it does not coincide with L. Choose
a vector 0 x E L\T(N). Then 0 F(x) FT(N). Indeed, let F(x) = FT(y) for
some yEN. Since z = Ty E T(N). z—x 0, while F(z—x) = FT(y).—F(z) = 0.
50 2. OPERATORS ON HILBERT MODULES

We get a contradiction to the condition KerF = 0. Hence, T is a topologically

injective epirnorphism, i.e., isomorphism, as well as F = ST'. U

Let us recall some facts about integrating operator-valued functions [62]. Let
X be a compact space, A a and p: C(X) —p A an involutive homomor-
phism of unital algebras. Let F : X A be a continuous map and let, for each
x e X, the element F(x) commute with the image of In this case the integral

EA

can be defined as follows. Let X = be an open covering and let

= 1 be a subordinate partition of unity. Choose points e and form the integral
sum

If the limit of such an integral sums exists, then it is called an integral.

If X is a Lie group G, it is natural here to use a Haar measure C(X)
= dg and to define

fQ(g)dg :=

for a norm continuous map Q G —p 8(H), where the A is viewed as

a subalgebra in the algebra 8(H) of bounded operators on a Hubert space H. If
Q:G C 8(H), then, since

>0,
fG
we obtain that
e and JQ(g)dg

(the positive cone P+ (A) is convex and closed). Hence we have proved the following
lemma.
LEMMA 2.8.13. Let Q: C be a continuous function. Then, for the
inteqral in the sense of [62], the following inequality holds:

0.
THEOREM 2.8.14 ([128]). Let GL = GL(A) be the full general linear group,
i.e., the group of invertible operators in End 12(A), and suppose that for the group
C, a representation g (g C. T9 E GL) is given, and that the map

Cx 12(A), (gu) T9u

is continuous. Then there exists an A-valued inner product on l2(A) equivalent to

the initial one (i.e.. generating an equivalent norm) and such that the representation
g '—* is unitary with respect to this new product.
 2.8. REPRESENTATIONS OF GROUPS ON HILBERT MODULES 51

PROOF. Let (, )' be the initial inner product. For any u, v E 12(A) there exists
a continuous map C A, x Define the new product by the
formula
(u,v) =
JG
I
where the integral can be considered in the sense of any of the two definitions in [62],
since the map is norm continuous. It is easy to see that this new product gives a
sesquilinear map 12(A) x 12(A) —+ A satisfying properties (i)—(iv) of Definition 1.2.1
and that, by Lemma 2.8.13, (u, u) 0. Let us show that this map is continuous.
Fix an arbitrary u E 12(A). Then x i—p C 12(A) is a continuous map
defined on a compact space, thus the set I
x E G} is bounded. Therefore, by
the uniform boundedness principle ['7, v. 2],
(2.30) lim = 0

uniformly in x e C. If u is fixed, then

-= const

and, by equality (2.30), one has

(u, v) 11=1 f(Tx (u), dxli

. vol C sup IITx(v)II —40 (v —40).

ccEG

We have obtained the continuity at the point 0, hence on the whole space 12(A) x
12(A). For = (ai(x),a2(x),...) e 12(A), the equality (u,u) = 0 takes the form

=
JC
0.

Let A be viewed as a subalgebra of the algebra of bounded operators on a Hubert

space L with an inner product (, )j.,. For each p e L we have

0=

= j dx
= f dx

(cf. [62]). Therefore = 0 almost everywhere, hence, by continuity, = 0

for all x and = 0. In particular, u = 0.
Since each operator is an automorphism, we obtain

f = (u,v).

Now we show the equivalence of the two norms, which, in particular, implies
the continuity of the representation. There is a number N > 0 such that N
52 2. OPERATORS ON HILBERT MODULES

for any x E G. Hence, by [62], we have

11u112 = II(u.u)IIA = if
<N2(j]uII')2.
\XEG I
On the other hand, applying Theorem 2.1.4 and Lemma 2.8.13, we obtain that

(u. u)'
= f (T9-iT9u. T9-iT9u)' dg j 11T9-t 112(Tgu, T9u)' dg

f = N2f(Tqu,Tqu)' dg = N2(u,u).

Then (IIuII#)2 = j(u, u)'IIA < II

(u, u)IIA = N211u112. D

REMARK 2.8.15. Since 12(P) is a direct summand in 12(A), the previous theo-
rem remains valid for 12(P) and for any other countably generated module M.
DEFINITION 2.8.16. Denote by GL* C GL the general linear group formed by
adjointable invertible operators.
REMARK 2.8.17. Before averaging we have had operators, which, in general, are
nonadjointable. but after averaging we have obtained unitary operators out of them.
In relation to this remark the following question arises: is it true that if a given
operator represents an element of compact group, then it is adjointable? A negative
answer to this question is contained in Example 2.3.2, since a decomposition into
a direct (topological) sum defines a representation of the group Z/2Z.
COROLLARY 2.8.18 ([1291). Let M = be a topological decomposition
into a direct sum of closed Hubert C* -modules (not necessarily an orthogonal sum).
Then there exists a new inner product on the module M equivalent to the initial
one, with respect to which the given decomposition is orthogonal.
PROOF. Define the operator J : M M by the formula
x,
Jx=<I —x, ifxEM1,
ifxeM2.
One can consider the operator J as a representation of the group Z/2Z on the
module M and, by Theorem 2.8.14, the inner product (x,y) + (Jx,Jy)
is equivalent to the initial one. Orthogonality of M1 and M2 with respect to this
inner product is evident.. 0
In Theorem 5.7.13 we shall show how the averaging Theorem 2.8.14 can be
generalized from the case of a compact group to the case of an amenable group,
but only for Hubert We-modules.

2.9. Equivariant Fredholm operators

In this section we transfer the main definitions and statements related to A-
Fredhoim operators to the G-equivariant case, where C is a compact group. We will
present only a list of results of [1281. The proofs (very similar to the nonequivariant
case) and details can be found, e.g., in [121].
 2.9. EQUIVARIANT FREDHOLM OPERATORS 53

By the averaging Theorem 2.8.14 we can restrict ourselves to a unitary action

of G on i2(P). Suppose the action of G on A is trivial and apply the Kasparov
stabilization Theorem 2.8.4 to 12(P):
(2.31) 12(P) NA,
where NA = V2) and is a countable collection of finite-dimensional
vector spaces of all irreducible representations of C (up to their equivalence), and
each representation is repeated countably many times. The isomorphism (2.31) is
a C-isomorphism of Hilbert A-modules and the sum on the left-hand side of (2.31)
is orthogonal. Put
m
Rm
i=m+1 i=1
For any bounded C-A-operator F : 12(P1) 12(P2) denote by S(F) : NA NA
the following (stabilized) operator:

( )
Note that this stabilization is defined up to a C-A-isomorphism.
THEOREM 2.9.1 ([1281). Let NA where M and H are closed C-
A-modules, and let H have a finite system of generators , a8. Then H is a
. . .

projective finitely generated C-A-module.

DEFINITION 2.9.2. A bounded C-A-operator
F: 12(1'l) -* 12(P2)
is called C-A-Fredholm if
1) the operator F is adjointable;
2) there exist decompositions of the domain NA = M1EBA(1 and the range
NA = of the operator S(F) (where M1, M2, Hi, are closed C-A-
modules and H1, are finitely generated), such that the matrix of S(F) with
respect to these decompositions has the form S(F) where F1
= ( F2
M2 is an isomorphism of C-A-modules. By the previous theorem, and
are projective C-A-modules and we can define the index of F by
index F = V/il — 1H21 E KG(A).
REMARK 2.9.3. Since the action of C is unitary, an analog of 2.7.6 evidently
holds.
THEOREM 2.9.4 (11281). The index is well defined.
LEMMA 2.9.5 ([128]). Let an operator F : 12(A) i2(A) be C-A-Fredholm.
Then there exists a number E > 0 such that any bounded adjointabie C-A-operator
D satisfying the inequality hF — DII <E is a C-A-Fredholm operator and index D =
index F.
LEMMA 2.9.6 ([128]). Let F and D be C-A-Fredholm operators:
F : 12(21) 12(72), D : 12(22) 12(23).
Then DF : 12(71) is a C-A-Fredholm operator and
index DF = index D + index F.
54 2. OPERATORS ON HILBERT MODULES

LEMMA 2.9.7 ([128]). Let K : 12(21) —, 12(22) be a compact G-operator. Then

1 + K is a G-A-Fredholm operator and index(1 + K) = 0.
LEMMA 2.9.8 ([128]). Suppose F 12(21) —* 12(72) is a C-A-Fredholm operator
and an operator K E is G-equivariant. Then the operator F + K
is G-A-F'redholm and index(F + K) = index F.
LEMMA 2.9.9 ([128]). Suppose F : 12(21) is a bounded adjointable
G-A-operator, D E End* NA and K E K(NA) are G-A-operators, and for D there
exists a decomposition of NA, as in the definition of a G-A-Fredholm operator, and
D = S(F) + K. Then F is a G-A-Fredholm operator.
THEOREM 2.9.10 ([128]). Suppose
F : 12(21) 12(22), D : 12(72) _' 12(21), D' 12(22) 12(71)
are bounded adjointable G-A-operators and
S(FD) = 1flA +K1, S(D'F) = 1flA +K2, K1, K2 e JC(NA).
Then F is a G-A-Fredholm operator.
LEMMA 2.9.11 ([128]). If, for bounded adjointable G-A-operators D, D' and F,
the operators FD and D'F are G-A-Fredholm operators, then F is a G-A-Fredholm
operator too.
 CHAPTER 3

Hubert Modules over

3.1.
Detailed information al)out can be found in the books [123, 13,
32, 116, 58]. We also recommend the original papers of Murray and von Neumann
[96], which are still current. We list here only the basic definitions and some
necessary facts.
Topologies on 8(H). Besides the norm topology on the algebra 8(H) of
bounded linear operators on a Hilbert space H, a number of other locally convex
topologies is useful to consider. These topologies are defined by various sets of
seminorms on 8(H). Let a E 8(H), E H.
1. The a-weak topology is defined by the seminorms
00 00
2 2
p(a) = , <0° <00.
i=1 i=1 i=1
2. The a-strong topology is defined by the serninorms

p(a) = <00

3. The topology is defined by the seminornis

p(a) = + < 00

4. The weak topology is defined by the seminorms p(a) =

5. The strong topology is defined by the seminorms p(a)
6. The strong* topology is defined by the seminorms p(a) = +
When restricted to bounded subsets in 8(H), the a-weak topology coincides
with the weak topology, the a-strong topology coincides with the strong topology
and the topology coincides with the strong* topology. We will use the
notation w-lim and s-lim for weak and strong convergence, respectively.
A commutant of a subset R C 8(H) is the set R' {a E 8(H) : ar
ra for any r E R}. A bicommutant of a set R is the set = (R!)!.
THEOREM 3.1.1 (von Neumann bicommutant theorem). Let A C 8(H) be an
involutive subalgebra. Then the following conditions are equivalent:
(i) the algebra A contains the identity operator and is closed with respect to
the a-weak topology;
(ii) the algebra A contains the identity operator and is closed with respect to
the a-strong topology;
(iii) the algebra A coincides with its bicommutant, A!! = A.

55
56 3. HILBERT MODULES OVER W-ALGEBRAS

In particular, it follows from here that if A C B C 8(H) are two subsets, then
A!! c B!!.
DEFINITIoN 3.1.2. An invohitive suhalgebra in 8(H) is called a von Neumann
algebra if it satisfies the conditions of Theorem 3.1.1.
Any von Neumann algebra is a Ct-algel)ra.
Unlike general C*..algebra.s. there exist polar decompositions in von Neumann
algebras: any element a E A can he written iii a unique way as a = uh, where u is
a partial isometry, h is a positive element of the algebra A and Ker u = Ker h.
Universal enveloping von Neumann algebra. Let w he a positive linear
functional on a Ct-algebra A and let H,.,) he a cyclic representation of A on a
Hilbert space obtained from the GNS-construction. Put
(ir.H)=
wE .4

where denotes the set of all positive linear on the Ct-algebra A.

The representation (ir, H) is called universal. The universal representation of a
Ct_algebra contains any representation of this algebra as a subrepreseutation.
Let us recall that the a(Et, E)-topology on the dual space (of functionals) Et
for a Banach space E is defined by the system of seminorms = 1(x) I, f E E*,
x E E. The E is defined by the system of serninorins
q1(x) =
THEOREM 3.1.3. The second dual space Att of a Ct -algebra A equipped with
the o.(A**. At)-t.opology is hom.eomorphic to the hicommutantA!! of the algebra
A C 8(H) with respect to the universal representation equipped with the a-weak
topology.
The von Neumann algebra where the bicommut.ant is taken with respect
to the universal representation. is called a universal enveloping von Neumann al-
gebra for the A and is denoted by A**. Any homomorphism of Ct-
algebras A B canonically extends to a homomorphism of the second dual spaces
Att Btt. If A C 8(H0) and A —p8(H) is the universal representation, then
H0 C H and A!! c A**.
Wt-algebras. The notion of Wt-algehra allows us to talk about von Neumann
algebras without alkv relation to the concrete Hubert space where they act.
DEFINITION 3.1.4. A Ct-algebra A, which, as a Banach space, is dual to some
Banach space F, A = Ft, is called a Wt-alqebra
A Banach space F is called pre-dual for A.
DEFINITION 3.1.5. A linear on a von Neumann algebra A is called
normal if for any increasing net aA E A, A E A, with the least upper bound a e A,
the value p(a) is tIme least upper bound of the set
THEOREM 3.1.6. Let A be a Wt-algebra. Then there exists a unique (up to an
isomorphism,) pre-dual spare for A. which coincides with the space of all normal
linear f'unctionals on A.
Denote by the pre-dual space of a Wt-algebra A. Let P denote the set of
normal positive functionals on A, P C The pre-dual space is the linear
span of the set P.
 3.1. W-ALGEBRAS 57

DEFINITION 3.1.7. A W*_algebra is called a-finite if it contains no more than

a countable set of iriutually orthogonal projections.
The condition of a-finiteness is equivalent to the existence of a faithful normal
positive linear functional on A.
Traces and classification of Let A be a and
P+(A) the cone of positive elements in A.
A trace on A is a function r: [0, +oo] such that
..

(i) ifs, t then r(s + t) = T(s) + r(t);

(ii) ifs E A A 0, then r(As) = Ar(s);
(iii) if s c and if n E A is unitary, then r(n*su) = r(s).
Being defined on tile set P+(A), a trace can be easily extended to a linear functional
on the whole algebra A.
A t.race T is called finite if r(s) < +00 for any s E
A trace r is called semifinite if for any s E P+(A), the number r(s) is the exact
upper bound of the values r(t) for all t E satisfying t s and r(t) < +00.
A trace r is called faithful if the conditions s E (A) and r(s) = 0 imply that
$ =0.
A trace r is called normal if the functional on A obtained by extending r is
nornial.
A W*_algebra A is called finite (resp. semifinite) if for any nonzero element
t E P+(A) there exists a finite (resp. semifinite) trace r on A such that r(t) 0. A
l,V*_algebra A is called properly infinite (resp. purely infinite) if there is no nonzero
finite (resp. semifinite) trace on A.
THEOREM 3.1.8. Any (semi)flnite possesses a faithful normal (semi)
finite trace.
A A is called an algebra of type I if it is isomorphic to a von
Neumann algebra B with commutative commutant
Let p be a projection in a Wi-algebra A. Denote by Ap C A the set of elements
of the form pap, a A. It is easy to check that Ap is also a
A W*_algebra A is called continuous if there is no projection iii the center 2(A)
of A, for which the algebra Ap is of type I. A semifinite continuous A
is called an algebra of type II. A finite type II is called an algebra of
type Iii, arid a properly infinite type II Wa-algebra is called an algebra of type
A purely infinite is called an algebra of type III.
THEOREM 3.1.9. For a A there exist mutually orthogonal central
projections E 2(A) such that their sum equals 1 and the algebras
Apr. Ap111, and Ap111 are algebras of type I, Il', and III, respectively.
THEOREM 3.1.10. For a A the following conditions are equivalent:
(i) A is finite;
(ii) there exists a linear map T from A into its center 2(A) such that:
(a) T(x*x) = T(xx*) 0 for any x E A,
(b) T(ax) = aT(x) for any a E 2(A), x A,
(c) T(1) = 1,
(d) T(x*x) 0 for any nonzero element x E A.
58 3. HILBERT MODULES OVER W-ALGEBRAS

In this case this map (a center-valued trace) T is unique and a-weakly

continuous.

A W*_algebra, whose center is isomorphic to C, is called a factor. Like the

direct integral of Hubert spaces, one can define the direct integral of von Neumann
algebras (see, e.g., [123]).

THEOREM 3.1.11 (central decomposition). Let A be a that admits

a faithful representation in a separable Hubert space. Then there exists a Borel
space r with a measure d-y and factors E r, such that

AJ I,

Let A be a von Neumann algebra. Two projections e, f E A are called equiva-

lent, p q, if there exists a partial isometry v E A such that e f=
v E A such that e = and f, we write e -< f.
It can be easily checked that the relation is an order relation on the set of all
projections in a von Neumann algebra. In the case where e f' e f, we write
e-<f.
A projection e E A is called finite if the conditions e f and f < e imply
e = f. Otherwise a projection e is called infinite. If there are no finite projections
I E A with f S e, then the projection e is called purely infinite. If the projection
ze 0 is infinite for any central projection z, then the projection e is called properly
infinite. If a von Neumanii algebra eAe is commutative, then the projection e is
called Abelian. For a projection e E A the least projection z E Z(A) from the
center Z(A) such that e z is called the central support for a projection e and is
denoted by z(e).

3.2. Inner product on dual modules

Hubert over will be called Hilbert Some
aspects of the theory of Hilbert become simpler in the W*_case.

THEOREM 3.2.1 ([100]). Let M be a Hilbert A-module. An A-valued inner

product (.,.) admits an extension to the Banach module M', making it a self-dual
Hilberi A-module. In particular. the extended inner product satisfies the equality
f EM'.
PROOF. Let f, g E M'. Our aim is to define the inner product of these\func-
tionals (f, g). For this purpose define the map F : P —'. C by =
where E P is a normal positive functional on A. Let us show that the F
admits an extension to the set of all normal functionals on A. To do this we
require the following two technical lemmas.

LEMMA 3.2.2. If E C and E P satisfy = 0,

then = 0.
 3.2. INNER PRODUCT ON DUAL MODULES 59

PROOF. Consider a normal positive functional = Then 2

i = 1,... ,n. If x,y eM, then, by assumption,

therefore = 0. Note that, by Proposition 2.5.7,

=
=
= = 0.

This means that the map F can be extended to To avoid extra notation
we denote this extension still by F.
LEMMA 3.2.3. The map F is bounded.
PROOF. One can write an arbitrary normal functional i/ E as = —
where P, i = 1,...,4, and Then

I II II

as required. 0
Let us continue the proof of the theorem. Since A is isomorphic to the space
of linear functionals on the pre-dual space [123], there exists a unique element
(f,g) e A such that = for any e in particular, =
C P. Thesesquilinearityofthedefinedmap (.,.) :

follows from the linearity of the map f '—i from M' to e P. Let us
for
show that (.,.) satisfies properties (i)—(iv) of Definition 1.2.1.
(i) Since f)) = 0 for any e P, the inequality (f,f) 2 0
follows.
(ii) Let (f,f) = 0 for f eM'. Then = 0 for all e P, hence = 0
for all x e M. Therefore f = 0.
(iii) Since

=
=
=
=
for any e P, we conclude that (f,g) =
(g,f)*•
(iv) Let a e A, e P. Let us define the functional on the algebra A by
pa(b) = b e A. Then e and çoa = where E C, E P.
Let us put i/ = + then is a positive functional and > . .
,
60 3. HILBERT MODULES OVER W-ALGEBRAS

It follows from Proposition 2.5.7 that

But

= = w(a*f(x)) = a)(x))

= ((f. x+ = ((f +
for any x E M. Since the subspace is dense in H,1,, the equality

= = ((f.

= ((f = a,g))
holds for any p E P. hence a*(f,g) = (f a,g). Passing to adjoints, we also obtain
(f,g)a = (f,g a). .

The obtained inner product on M' is an extension of the inner product from
M. Indeed, if x,y EM, E P, then
= = =
Hence = (x,y). Further, if f EM', then
ço((f,x)) = =
therefore (f, = f(x). Let us show that M' is complete with respect to the norm
I'M' defined by the inner product constructed above. There also exists the norm
on M' defined as the norm of linear maps from M to A, with respect to which
the space M' is complete. Let us prove that II•I'M' = 1.11. Since
2
1(x)* f(x) = (x,f)(f,x) —
(x,x),
we obtain that 11111 If But since 11111 for each w E

Ill = = : E P, IIwII = 1} 111112,

and =
So, we have proved that M' is a Hubert A-module. It remains to verify that
it is self-dual. Let F E (M')'. The restriction of F to M C M' is an element of
the module M', hence one can find a functional I E M' such that F(s) = f(x) for
all x E M. Let us define the functional F0 E (M')' by
Fo(g) = F(g) — (f,g), gE M'.
It is clear that for all x E M. We should verify that Fo(g) = 0 for all
= 0
g E M'. Let E P. Choose a sequence {yn + in converging to
Since F0 is bounded, we can find a number K such that Fo(h)*Fo(h) K(h, h) for
allhE M'. For alln= 1,2,... one has
= —
— i,,)) Kço((g — —
 3.3. HILBERT W-MODULES AND DUAL BANACH SPACES 61

But since
— yn,g — — — +
— + +(yn + +
= — +
we have — — 0. Therefore
(3.1) = 0.

Since the equality (3.1) is valid for any normal functional E P, we obtain that
Fo(g) = 0. D
3.3. Hubert and dual Banach spaces
PROPOSITION 3.3.1. Let M,H be Hubert over a A
and let T : M —9 H be a bounded operator, T E HornA(M,AI). Then there exists
a unique extension of the operator T to an operator T : M'
PROOF. Define the operator T# : by (T#y)(x) := (y,Tx), x E M,
y E f'I. Since II(T#y)(x)II lxii the operator T# is bounded, IIT#yII
iTilliyii. ForanyaEAwehave
(T#(y . a))(x) = (y. a,Tx) = a* (y, Tx) ((T#y) a)(x): .

therefore the map T# is A-linear. Define the map T : M' by (Tf)(y) =

(f, where y E H, f E M'. Since T (T#)#, the map T is also a bounded
A-linear map. The equality = (Tx5(y) shows that the operator T is an
extension of the operator T.
Let us show uniqueness of this extension. Let S M' H' be a bounded
A-linear map that coincides with T on the submodule M = x E M} c M'.
Then their difference V = T — S vanishes on M. Since the module M' is self-dual,
the operator V has an adjoint operator : H' M'. If g E H', x E M. then
(V*g)(x) = = = 0,
i.e., V= 0, hence S T. U

COROLLARY 3.3.2. Let M be a Hubert A-module. Then the map T '—* T

defines a monomorphism = EndA(M').
Let us show that self-dual Hilbert are dual Banach spaces, as well
as the of operators acting on them.
PROPOSITION 3.3.3 (1100]). Let M be a self-dual Hilbert Then
M is a dual Banach space.
PROOF. Let us introduce the notation M° for the Hubert module M viewed
as a Banach space with multiplication by scalars given by the formula A x := Ax,
x E M°. Consider an algebraic tensor product 0 over C, where is a
pre-dual space of normal functionals on A. Equip the space 0 M° with the
maximal cross-norm and define for x E M the linear functional on 0 M° by
the formula
=
62 3. lULBERT MODULES OVER W-ALGEBRAS

where . . E A,, .
M°. It is easy to see that this functional is
well defined. Since

® lii) lxii Ilwiil

it follows from the definition of the maximal cross-norm [661 that <
Let us show that in fact the equality = lxii holds. Let be a sequence
of functionals of norm 1 in A, such that —' iixii2. For each element
of the form ® x E A, ® M° we have iixii lxii and ®
iixii2, hence lxii < Therefore the map x '—' defines an isometric inclusion
M C (A. ® M°)'. To prove the statement it is sufficient to check that the set
= : x E M} is closed in (A, ® Mo)* with respect to the weak* topology,
since it would mean that M is isometric to the dual space of some quotient space of
A, ®M°. Let be a net in M that converges to some element F E (A, ®M°)'
with respect to the weak* topology. For y E M, let us define the linear functional
on A. by = ® y), where E A,. The functional is bounded,
there exists a unique element f(y) E A satisfying the properties
liFil iiyii, hence
iiFii and = f is obviously
linear. Let us show that it is A-linear as well. Let y E M, a, b E A, E A,. Define
the normal functional E A, by = ço(a'b). Then it follows from the equalities
cp(f(y a)') = F(p® (y .a)) = . a)) = .

= = =
= = p(a'f(y)') = cp((f(y)a)'),
which hold for any E A, that f(y . a) = f(y)a. Since the module M is self-dual,
we can find an clement x0 E M such that 1(y) = (x0, y). Thus F = hence
is closed in (A, ®M°)'. 0
Consider the weak* topology on the dual Banach space M. A net } in
M obviously converges to an element x E M with respect to this topology if
for every A, and for every y EM.
Some modification of the previous argument allows us to also obtain the fol-
lowing statement.
PROPOSITION 3.3.4 (11001). Let M be a self-dual Hubert W'-module. Then
the C'-algebra is a W'-algebra.

3.4. Properties of Hubert W'-modules

The elements of self-dual Hilbert W'-modules admit the following convenient
representation (an analog of the polar decomposition).
PROPOSITION 3.4.1 ([100]). Let M be a self-dual Hubert Any
element x EM can be written as x = z• (x,x)'12, where z EM is such that (z,z)
is the projection onto the image of (x, x)V2. Such a decomposition is unique: if
x = z' a, where a 0, and if (z', z') is the projection onto the image of a, then
= z and a = (x,x)hh/2.
 3.4. PROPERTIES OF HILBERT W-MODULES 63

PROOF. For x E M and n E N put

= ((x,x) + 1/n)'!2, x
Since = (x,x)((x.x) + < 1. Let y eM be an accumulation
point for the sequence } in the weak* topology (which exists due to the com-
pactness of the unit ball). Since ha,, — (x,x)"2h1 0 and x,
x= (x,x)'!2. Let p be the projection onto the image of Then
p(x, x) 1/2 = (x, x)
1/2
p = (x, x)
1/2

hence x = y.p(x,x)"2 and

(x,x) (x,x)'!2p(y,y)p(x,x)'!2.

Therefore
(x,x)'!2(p = 0.

Since 1, we have r — p(y,y)p 2 0. Therefore

0,

whence it follows that p(p — p(y, y)p)1/2 = 0, hence p p(y, y)p. Put z y p.
Thenz.(x,x)'!2=y.p(x,x)'!2=x,(z,z)=p(y,y)p=pandz.p=z.
In order to prove the uniqueness of the decomposition suppose that x = z' a,

where a 2 0, and that (z', z') is the projection onto the image of a. Then (x, x) =
a(z',z')a a2, hence a (x,x)'!2 and (z',z') =p. Since (z'—z'.p,z'—z'.p) = 0,
we obtain z' = z'p. Also one has
(z,x) = (x,x) 1/2 = (z,z , )(x,x) 1/2
i.e., = 0. whence we obtain that (p—(z,z'))p=p—(z,z'.p)
p — (z, z') = 0. Now it can be easily seen that (z — z', z — z') = 0, hence z' = Z. 0
Let be a set of projections in a A. For each of them the
set M0 = paA C A has the natural structure of a singly generated projective
Hilbert A-module. Similarly to the definition of the standard Hilbert module we
can define the module as the set of sequences (ma), ma C Mc, C A,
such that the series is norm-convergent in A. The dual filbert module
(sc, Ma)' is called the ultra weak direct sum of the modules Mc,. For self-dual
Hilbert W*_modules we have the following structural result.
THEOREM 3.4.2 ([1001). Let M be a self-dual Hubert over A. Then
there exists a set {pc, } of projections in A such that the module M is isomorphic
to the ultra weak direct sum of the modules pc,A.
PROPOSITION 3.4.3. Let Al C HA be a Hilbert submodule over a W* -algebra
A. If Jsf-'- = 0, then the dual module 15/' coincides with
PROOF. Let j Al HA be an inclusion of modules. The restriction of
functionals f fig, f E defines the map j' dual to j. If :

fE is such that = 0, then f±Isf and, by assumption, one has f = 0.

Therefore the map j' is injective. Consider the composition of maps

i=j'ooj '—+
64 3. HILBERT MODULES OVER

If n E JV, then i(n) = j'(j(n)) = = Therefore the map i coincides with

the inclusion map Al '—p /'f'• The dual map (after identification of the first and
second dual modules)
(i' o i)' = i' o i" : .IV' = .Af" —, —+.N'•'
is an isomorphism, hence the map i' should be surjective; then the map j' is sur-
jective as well. 0
PROPOSITION 3.4.4. Let A be a and 1Z C HA an A-submodule
without orthogonal complement, i.e., 1?)- = 0 in HA. Then R! =
PROOF. It is sufficient to show that if the orthogonal complement to a sub-
module 1?. in HA is equal to zero, then the orthogonal complement to 7?. in
is equal to zero too. Let us assume the contrary. Suppose that there exists a
functional f such that f(r) (f, r) 0 for some r E 7?.. Since the series
*
is norm-convergent in A, there is a number n such that f (n) (z) 0 for
. . .

(Ii,... .
. .). But, as f(Th) E 11A, we get a contradiction. 0
Even Hilbert over commutative behave differently
from Hubert spaces. The following example shows what kind of pathology can
occur.
Let A = be the of bounded sequences and let co C be
its subalgebra of sequences vanishing at infinity. Consider the Hubert
M = l2(co) over A.
PROPOSITION 3.4.5 ([85]). There is a submodule .iV C M such that it coincides
with its bi-orthogonal module, but is not complementable.
PROOF. Put
//= {(f1,f2,f3,...),(f1,2f2,3f3,...)},
where fk E c0 for all k eN and the series k2IfkI2 is convergent in c0. It
is easy to see that the submodule Al C M is closed and coincides with its bi-
orthogonal complement, ftf-'--'- = Al. Let us show that Al is noncoinpiementable.
Assume the contrary: let M = .)VEB/C for some submodule K C M. Then there
exists an A-linear (nonorthogonal) bounded projection P : M —' Al. Since P
commutes with the action of A, it can be written as a sequence (Pk)kcN, where
Pk is a projection acting on the k-th coordinate. More precisely, if Pk E A is the
projection onto the k-th coordinate, then Pk = PkP. Since P is bounded, the norms
of the projections Pk should be uniformly bounded. But any projection Pk projects
onto where N : 12 —4 is an unbounded operator, N(aj,a2,...) =
(a1,2a2,. . .). For each k there is a decomposition Mk = 12 =ft4 DKk, where
JCk = pkJC, j'4 = Pk implies
that the complements JCk should consist of the pairs {(x, y) : x, y l2} such that
> for some > 0 independent of k.
Consider an arbitrary element 12,.. .), (gi' 92,...) M, where C0,
gi for any i. If =M then this element can he written a.s the sum

For a fixed n, we use a superscript to denote the number of an element in the

sequence A = = For sufficiently large n the elements of
the submodule K satisfy — > — But the left-hand side of this
 3.5. CHARACTERIZATION OF SELF-DUAL W-MODULES 65

inequality vanishes as k oo (when ii is fixed) and the right-hand side does not
vanish, since e c0 (it is sufficient to take the constant sequences = $ 0).
This contradiction shows that the submodule .N is not complementable. D

3.5. Topological characterization of self-dual Hubert

Let M be a Hubert over a A and let P E be the
set of normal states on A. Let us define (see [42]) two topologies on M as follows.
Denote by r1 the topology given by the system of seminorms E P,
and denote by r2 the topology given by the system of serninorms y E M,
E P. In the case where A = C and M is a Hilbert space, the topology r1 is the
norm topology and the topology r2 coincides with the weak topology, so in general
these two topologies do not coincide.
THEOREM 3.5.1 ([42]). Let M be a Hubert Then the following
conditions are equivalent:
(i) the module M is self-dual;
(ii) the unit ball B1 (M) is complete with respect to the topology r1;
(iii) the unit ball Bi(M) is complete with respect to the topology r2.
PROOF. Let us prove the implication (i) (ii). Assume, for this purpose, that
the unit ball B1(M) is not complete with respect to the topology r1. Let us denote
by L the linear span of the completion of B1(M) with respect to the topology
Ti. For the extensions of seminorms from M to L we use the same notation. By
assuniption, there exists an element r E L \ M and a bounded net E A,
such that for any p E P and for any e > 0 there exists some (1 E A, for which
— — <r for all /3 E A. /3 o. For arbitrary XE M we have
— = —

— y_y,yfl —

for all /3,y Therefore there exists the limit (with respect to the As)-
topology)
E A
in the A for each x E M. The inequality

: E A}

shows the continuity of the map R: M —+ A, x '—p R(x). It is clear that the map
R is an A-module map, so R is a functional on M. By assumption, the module
M is self-dual, so there exists an element z E M such t.hat R(x) = (z,x). Then
lime (yo, x) = (z, x) (where the limit is taken in the o(A, Therefore
the net {ya} converges to z E M in the topology r1 and r = z, giving a contradiction
to our assumption.
Let us now prove the implication (ii) (i). The extension of the inner product
to the dual module M' we again denote by (.,.). It is easy to see that the ideal
(M,M) E A is norm dense in the ideal (M',M') E A, hence (M,M) E A =
(M', M') E A. Let us first consider the case where a W*_algehra A is u-unital.
Then there exists an exact normal state E (see [13, Prop. 2.5.6]). Let
be the cyclic representation associated with The vector E H is
66 3. HILBERT MODULES OVER W-ALGEBRAS

simultaneously cyclic and separating. The linear space M with the inner product
(•,•) = becomes a pre-Hilbert space and the map i!'(f(.)) : M —+ C, where
I E M', becomes a linear functional on M. Then one can find an clement
in the completion of the space M with respect to the norm .))1/2 such that
(fe,, x) = ip(f(x)) for all x M. This means that there exists a sequence (x2),
x1 E M, i N, such that

where the image of the element x under the canonical inclusion M C M' is denoted
by Since the vector is cyclic and separating, there exists the limit (in the
on A) lim1 — — f) = 0 (see [13, Lemmas 2.5.38, 2.5.391).
Therefore f E M and the module M is self-dual. Let us pass to the general case.
If a A is not a-unital, then one can choose a directed set of projections
E A, E A, such that for each E A, the algebra paApa is a o-unital
and limo Pa = 1, where the limit is taken in the o(A, As
was proved earlier, the functional on the Hilbert paM is an
element of the module paM for all o E A. But then there exists the limit (in the r1
topology) of the net { }, which belongs to M and is equal to f, so self-duality
of the module M is proved.
It remains to show the equivalence of conditions (ii) and (iii). If B1(M) is
complete with respect to the topology r1, then M is self-dual. Therefore it is a
dual Banach space with respect to the topology r2 (see Prop. 3.3.3), hence B1(M)
is complete with respect to the topology Let us assume now that B1(M) is
complete with respect to the topology r2 and {Xa}, E A, is a bounded Cauchy
r1-net. ForallyEM, have
(3.2)
As before, let L denote the linear span of the ri-completion of B1(M). There
exists the limit lime = t L in L (with respect to the topology Ti). It follows
from inequality (3.2) that {XQ} is a Cauchy net with respect to the topology T2 as
well. But since the topology r2 is weaker than the topology r1, one has L ç M.
Therefore the limits lirn2 x1 with respect to the topologies Ti and r2 coincide and
are equal to t E M, implying the completeness of M with respect to the topology
TI. 0
3.6. Fredhohm operators over
Throughout this section, let A denote an arbitrary -algebra. We show here
that the properties of Fredholm operators in the We-case are more similar to those
of the usual Fredholrn operators (over C) than in the general which may
be more pathological (see Example 3.6.3 below). We present here slightly modified
results of [48].
LEMMA 3.6.1. Let M be a self-dual Hubert over a A.
For each closed submodule .iV c M the biorthogonal set .AP' c M is a Hubert
A-submodule and is a direct summand of M, as well as its orthogonal complement
 3.6. FREDHOLM OPERATORS OVER W-ALGEBRAS 67

PROOF. It is clear, by the definition of the orthogonal complement. that Al'-'- ç

Jvl is an A-submodule. Consider the inclusion i : Al M aiid its adjoint map
M= Al'. Since dual W*_modules are self-dual, admits an
and, consequently, its kernel is an image of a selfadjoint projection. i.e., it has an
orthogonal complement in M. But
i*(rn)0 i*(m)(n)=OVnEAl inEJt['.
0
LEMMA 3.6.2. Let be a bounded A-module morphism of a self-dual module
M. Then the kernel of the map is a direct suminand in M and satisfies
the equality =
PROOF. By Proposition 3.3.4, the algebra EndA(M) is a
algebra. Therefore there exists a polar decomposition d = US. where S is positive
and U is a partial isometry and Ker = Ker U. There also exists a selfadjomt
projection p e onto Ker = Ker U. We are clone. 0
EXAMPLE 3.6.3. Note that the kernel of a bounded A-linear operator oil a
Hilbert A-module over an arbitrary C*_algebra A need not be a direct sutninand.
For example, consider the C*_algebra A = C( [0, 1]) of all continuous functions on
the segment [0,11 as a Hilbert A-module over itself equipped with the standard
inner product (a, b)A = a*b. Define the mapping by the formula Oq(f) = f
for the fixed function
I—2x+1 ifx1/2,
ifx1/2
and for every f E A. Then equals the Hilbert A-submodule and the (left)
ideal {f E A : f(x) = 0 for x E [0, 1/21) is not a direct sunmiand of A, but coincides,
nevertheless, with its bi-orthogonal complement in A.
COROLLARY 3.6.4. Let :M Al be a bounded A-linear mapping of self-
dual modules. Then the kernel of is a direct suinmand of M and has the
property
PROOF. Consider the self-dual Hilbert A-module £ = M Al. The mapping
can be identified with a bounded A-linear mapping on £ that acts on M as
and on Al as the zero operator. Since the kernel of is a direct suinmand of £.
which contains Al by Lemma 3.6.2, its orthogonal complement is a direct sunimaiid
ofM. 0
EXAMPLE 3.6.5. Let A be the set of all bounded operators B(H) on a separable
Hilbert space H with the basis :i E N}. Denote by k the operator =
for a sequence : i E N} e e0(R). Then the mapping : A —' A. : a k a
is a bounded A-linear mapping on the right projective Hilbert A-module A. But its
image is not a direct summatid of this A-module and is not even a Hilbert module.
since all direct sirnimands of A are of the form pA for some projection p of A. and
k. 'A should equal p. The image of is a subset of the set of all compact operators
on H. Note that the mapping is not injective.
The following statement tinder some restrictions can he proved in the
as well [71,
 ____

68 3. FIILBERT MODULES OVER W-ALGEBRAS

PROPOSITION 3.6.6. Let M be a self-dual A-Hubert module and let {ftf, (., .)}
be an arbitrary Hubert A-module. Suppose there exists an injective bounded module
mapping a : M .,V such that ci(M)-'--'- = Af. Then the operator o(a*a)_h/2 is a
bounded module isomorphism between M and A!. In particular, these modules are
isomorphic as Hubert A-modules.
PROOF. The mapping a possesses an adjoint bounded A-module mapping a*:
M due to the self-duality of M. Since a positive element of the
algebra EndA(M) of all hounded (adjointable) module mappings on the Hubert
A-module M, its square root (a*a)l/2 is well defined by the series

(a*a)h/2 = II(a*a)IIh/2 (idM (idM -

where {Ak} are the coefficients of the Taylor series of the function f(x) = —x
at x = 0. Moreover, since
((a*a)l?'2(x) (a*a)h/2(x)) (cm(x), cm(x)),
due to the injectivity of a the mapping (cm*a)h/2 has trivial kernel. Note that the
range of (a*a)l/2 is r1-dense in M. Indeed, for every A-linear bounded functional
r(.) = (.. y) on the self-dual Hilbert A-module M that maps the range of (a*a)l/2
into the origin one has
0= y) = (x
for every x e M. Hence y = 0, since (a*a)h/2 is injective and x E M was chosen
arbitrarily.
Now consider the mapping a(a*cm)_l/2 defined on M. Since, by the assump-
tions on a, the range of (a*a)h/2 is r1-dense and the kernel is trivial, one concludes
that its inverse unbounded module operator (a*ci)_u/2 is ri-densely defined. One
gets
(a(a*a)_h/2(x), = (x. y)
for every x, y from the (Ti-dense) domain of (a*a)_h/2. Consequently the operator
can be extended to a bounded isometric module operator on M by r1-
continuity. Its range is r1 -closed (i.e., is a self-dual direct summand of .iV), hence
it equals by assumption. D
COROLLARY 3.6.7. Let M be a self-dual Hubert A-module and let {.N, (., .)} be
an arbitrary Hubert A-module. Every injective module mapping from M into I'..! is
a Hubert A-module isomorphism between M and a direct summand of 1%!.
PROPOSITION 3.6.8. Let M and I..! be countably generated Hubert A-modules
and let F M .A! be an A-Fredholm operator. Then Ker F and (Im F)' are
projective finitely generated A-submodules and index F = [Ker F] — [(Tm F)'] in
K0(A).
PROOF. Let M = M0 iV = .MJ be the decompositions from the
definition of A-Fredholm operator,
(Fo O'\ -
F
F0 : M0 !VO, F1 : —, .iV1, where M1 and are projective finitely generated
modules. Let x = x0 + x1, x0 E M0, x1 E M1 and F(x) = 0, so 0 = Fo(xo) +
 ____

3.7. DUPRE - FILLMORE THEOREM 69

Fi(xi) E Thus Fo(xo) = 0. Fi(xi) = 0, so x0 = 0 and x E M1. Thus

KerF = KerF1 C M1. By Lemma 3.6.2, KerF is a projective finitely generated
A-module and has an orthogonal complement. So, by Corollary 3.6.7,
(FO 0 o\
F= 0 Ff 0 M0 KerF F(M'1)) (ImF)'
(

\ 0 0 0)
and index F [KerF] — [(ImF)']. U

The following example shows that the case of general Hilbert and
injective mappings between theni is more complicated.
EXAMPLE 3.6.9. Consider the A C([0, 1]) of all continuous func-
tions on the segment [0,1] as a self-dual Hilbert A-module over itself equipped with
the standard A-valued inner product (a, b)A = a*b. The mapping : f(x)
x 1(x) (x E [0, 1]) is an injective bounded module mapping. Its range has trivial
orthogonal complement, but it is not norm closed and, consequently, not a direct
summand of A. Nevertheless, the bi-orthogonal complement to the range of 0 with
respect to A equals A.
LEMMA 3.6.10. Let P and Q be self-dual Hubert A-submodules of M. Then
P fl Q is a self-dual Hubert A-module and a direct summand of M. Moreover,
P + Q c M is a self-dual Hubert A-submodule. If P is projective and finitely
generated, then the intersection P fl Q is projective and finitely generated too. If
both P and Q are projective and finitely generated, then the sum P + Q is projective
and finitely generated too.
PROOF. Let p: M = P 1" P' be the canonical orthogonal projection,
which exists by Proposition 2.5.4. Let PQ = p : Q P'. Since Q is a self-
dual Hilbert A-module, PQ admits an adjoint operator and KerpQ c Q is a direct
summand by Lemma 3.6.2. Consequently it is a self-dual Hilbert A-submodule of
Q c M. But KerpQ P fl Q. To obtain the second assertion, one has to again
apply the fact that every self-dual Hilbert A-submodule is a direct summand by
Proposition 2.5.4.
If P is projective and finitely generated, then any direct suminand of P is
projective and finitely generated, whence the remaining assertion follows. 0

3.7. Dupré — Fillmore theorem for Hubert modules

over finite
Let A be a finite Denote by r the normal faithful finite trace on
the algebra A normalized by r (1) = 1.
Let M c Hk be a Hilbert A-submodule. By M1- we denote its orthogonal
complement in the module It is known from Theorem 1.4.5 above that
if M is a finitely generated projective Hilbert A-submodule in Hk, then it admits
an orthogonal complement, = M M'. In the case where M is a submodule
in the module HA instead of Hk, the orthogonal complement to M in the module
HA is isomorphic to HA. However, in the general case, nothing is known about the
relation between the modules M' and Hk. The following theorem describes this
relation in the case of finite
70 3. HILBERT MODULES OVER W-ALOEBRAS

THEOREM 3.7.1 ([81]). Let A be a that admits a decomposition

into a direct integral of finite factors. If M is a finitely generated projective A-
submodule in Hk, then the modules M-1- and Hk are isomorphic.
PROOF. The idea of the proof is contained in [36]. Let . . , gn be generators
of the module M. Without loss of generality we can assume that the operators
(g,. g) E A are projections, (fit. flu) = Let {Cm} be the standard basis of the
module HA C "k Fix > 0 and define the elements E M-'- by

= em —

Theii
1

The inecluality
r((gj,gj))

shows that the series T((gj, em)) is convergent, hence there exists a
number m0 such that for all m> nz0, one has
I * \
(Yuiem)) <
E

> 1—
To proceed wit.h the proof, the following lemma is required.
LEMMA 3.7.2. If SE Hk, lxii = 1 and if r((x,x)) > 1— then there exists
a projection p e A such that r(p) > 1 — E and the operator p(x, x)p is invertible in
the W* -algebra pAp.
PROOF. Let us denote by dP(A) a projection-valued measure defined by the
positive operator a (x,x) E A, a = f1 AdP(A). Put

A
Jo, A<A0,
i, A>A0,
where Ao e [0; 1]. Then f(a) = p is a projection. Denote dr (P(A)) by dp(A). It is
a C-valued measure on [0; 1] and, by [96], one has

r(a)=f Adp(A).

Then

(3.3) 1— <r(a)=j Ad/L(A)

Since P(1) = 1, we obtain

(3.4) ([0; A0)) + ([Ao; 1]) = 1.

From (3.3) and (3.4) we get the inequality

p([Ao;1]) >1
— 2(1
 3.7. DUPRE - FILLMORE THEOREM 71

Choosing in an appropriate way a number A0 $ 0, we get ([Ao; 11) > 1 — e. It

follows from the definition of the function f(A) that

r(p) = r (f(A)) / f(A) dp(A)

=
= fdi(A) -E.
Now consider an operator pap E pAp. From the spectral theorem we obtain that

A I (A) dP(A).

Therefore the spectrum of the element pap of the pAp lies within the
segment [Ao; 1], hence it is separated from zero, so this operator is invertible in
pAp. 0
Now we can proceed with the proof of Theorem 3.7.1. Let
—1/2
= pbp = b E pAp
—
an em—em
' . en

= = P.
Let us now take an element y E M1 such that y $ 0, 1. Since the series
converges, for any e >0 we have
/ T((em,y)*2
,, * ,,
b (em,y))

11b112T((em,y)*(em,y)) <c
for sufficiently large numbers m.
Denote an operator (em,y)* (em,Y)
n
E A by c. Then r(c) < -i-; Ilcil 1 and
c 0. If dQ(A) is a projection-valued measure and if we denote a measure dr(Q(A))
by dv(A), then Adzi(A) < f. If A1 E[0; 1], then
2
A1.J dv(A)f 2

hence f dv(A) < Taking A1 we obtain that v([Ai; 1]) <E. Put
1, A<A1,
A
o, AA1.
Then the operator q = g(c) is a projection such that r(q) = A1))> 1 — E and

(3.5) IqcqII A1 =

Denote by p V q (resp. p A q) the least upper (resp. greatest lower) bound of the
projections p and q. Put p' = p A q. Since (see [123])
r(p)+r(q) = r(pVq)+r(pAq),
we have
r(p') ='r(p)+r(q) —r(pVq) > (1 —E)+(1 —1 = 1 —2E,
72 3. HILBERT MODULES OVER W-ALGEBRAS

due to the inequality r(p V q) T(1) = 1. We get from the estimate (3.5) that

(3.6) jp'cp'II

Put = •p'. Then = p'. Put also y' = E M'. The element y'
can be decomposed into the sum of two mutually orthogonal summands: y' = u+v,
where u = y v= + i). u,v E M.L. Then

(y',y') = (u,u) + +EP') +Ep')

and
*

p'(y', y')p' = p'(u, 'u)p' + y)p' + EP') y)p' + EP')

(3.7) *
= p'(u, u)p' + y)p' + EP') y)p' + El.)').
Since
*
(etm y)p' y)*pl(en y)p'
y)p' = p'cp',
it follows from (3.6) that y)p'jj < Therefore the operator y)p' +
Ep' is invertible in the W*_algchra p'Ap'. Invertibility of the operator p'(y', y')p'
follows from (3.7). Let us consider a trace norm on the module M1 (and on Hk)
defined by the formula
1/2

The inequality

— —
= = r(E2p') <E2
implies the estimate Thus we have proved that the set of the elements
of the module M1, for which there exists a projection p' such that r(p') > and
the operator p'(x. .r)p' is invertible in the algebra p'Ap', is dense in the module
with respect. to the trace norm. In particular the following statement holds.
LEMMA 3.7.3. There exists an element x E M' such that
> and lxii 1.
Let us come back to the proof of Theorem 3.7.1. Let } be a sequence that
contains each an infinite number of times. Put y = Yi — gk(gk. y1): Then
for e1 = 1 one can find an element y' E such that Ily'M 1 and
(3.8) — y'llr <E',
and a projection pi with T(pi) > such that the operator pi(y'. y')pi is invertible in
—1/2
the algebra p1Ap1. Then, taking h1 = y'b', where b' = E p1Ap1,
we obtain from (3.8) the inequality
< = distT(y,y') < ei,
where B1 denotes the unit ball of the Hubert module in the initial norm. Therefore
distT(yl.Bj(SpanA(M,hl))) <Er.
 3.7. DUFRE - FILLMORE THEOREM 73

Further, taking 62 = wecan find an element h2 E (SpanA(M, h1))-'- such that

the operator (h2, h2) =P2 is a projection, r(p2) > and
<e2.
Repeating this procedure and taking Ek = we obtain a set of mutually orthogonal
elements E M-t- such that the operator (hi, = pj is a projection, T(pj) >
and for which one has
(3.9) distT(yk,Bl(SpanA(M,hl,h2 hk))) <6k.
Such elements h1 generate a A-module .AJ C M' and it follows from (3.9) that
<
Therefore the closure of B1 (M .N) with respect to the trace norm contains the
unit ball of the whole module Hk and the closure of B1 (At) with respect to the
trace norm contains B1 (M
Thus the constructed basis { } of the module Jv' is not convenient to work
with, since the elements (hi, are not equal to one. Therefore we should pass to
another basis. Denote by T the standard center-valued trace on A [123].
LEMMA 3.7.4 ([48]). For any number C there exists n such that T p2)
C.
PROOF. Suppose that there is a normal state f on the center Z of the algebra
A such that for some number C the inequality (f o T) <C holds. Then
one can find a central projection z E Z such that

(3.10)

Consider the zAz. Multiplication by z turns Hilbert modules over

A into Hilbert modules over zAz and respects the orthogonality of submodules.
Therefore I'.1 z c M-'-z and the set B1 (J\fz) is dense in B1 (M1z) with respect to
the trace norm defined by the faithful trace on zAz induced by the trace
The inequality (3.10) means that for any 6 > 0, one can find (substituting, if
necessary, z by a smaller central projection) a number k such that the inequality
T <6 holds. The module J'fz
is a zAz-module generated by the elements h1z, .. . , and the module lZk is
the orthogonal complement to 4 in Nz. Since Bj(JVz) is dense in the
set B1 should be dense in Bi((Mz ® 4)-'-) with respect to the trace norm
Let x = E Estimate its trace norm:

= =
i>k i>k

= <Tz(>PiZ.
i>k i>k

1x112) (>PiZ).
i>k i>k
Since <6, we obtain <6. Therefore < e for all
x e Bj(lZk). But since the set B1(R.k) is dense in Bi((Mz ® 4)-i-), one has the
74 3. HILBERT MODULES OVER W-ALGEBRAS

estimate < e for all y E Bi((Mz On the other hand, if we apply

Lemma 3.7.3 to the module (Mz £k)' instead of to the module M', then we
can find in an element y satisfying the inequality > This
contradiction completes the proof of the lemma. 0
To complete the proof of Theorem 3.7.1, let us choose a projection q in the
algebra A with properties
(3.11) T(q) = min(T(pi + p2); 1)— T(pi)
and qJ..p1. It follows from (3.11) that T(q) T(p2), hence there exists another
projection q' equivalent to the projection q and such that q' Equivalence of
projections q and q' implies the existence of a unitary element u E A such that
qu = uq'. Put r = h2q!u* e Then r is orthogonal to h1 and
(r,r) = = uq!p2qlu* = = q.
Put = h1 + r. Then
= (h1, h1) + (r, r) = Pi + q.
/41)))
Note that = min(T(pi + p2); 1). Passing to the next element, 113, we
(2) (2) (2) (2)
can obtain an element h1(2) such that (h1 , h1 ) is a projection and T((h1
. . .
,
)) =
min(T(pi +p2 + P3); 1). By Lemma 3.7.4, repeating this procedure and increasing
the value hr)), we construct an element such that hr) = 1.
The orthogonal complement to h?° in N is generated by the elements i > 1,
where qj are some projections. Applying the construction described above to these
generators, we can construct, by induction, a set of elements such that they
satisfy the equality (hr, hr) = 1 and generate the module N. Therefore } is
a basis in N and the module H is isomorphic to HA.
It remains to prove that Al' = M'. Since the module H C M' is closed in the
initial norm, the restriction f E (M1)' belongs to H'. Note that the
module M-'- is self-dual, (M-'-)' = M', because of the self-duality of the modules
and M. Suppose that f = 0. Since the submodule H is dense in with
respect to the trace norm, one has f = 0 on due to the continuity of the map
f: —* A with respect to this norm because of the inequality

(3.12) T((f(y))*f(y)) 11f112.r((y,y)),

where y E M'. This proves injectivity of the map M' Now let 0 E H'.
This functional can be extended up to a map from to A. If C H is a
sequence converging to the element y E with respect to the trace norm, then
put 0(a) = It is well defined due to the estimate (3.12) with 0 instead
of f. Thus, the A-modules and H' coincide and the theorem is proved since
the modules H' and are isomorphic. 0
 CHAPTER 4

Reflexive Hubert
4.1. Inner product on bidual modules
For a Hubert C5-module M over a C5-algebra A we define the bidual Banach
right A-module M" as the set of all bounded A-homomorphisms from the dual
module M' into A. It turns out that the A-valued inner product on M can always
be extended to the bidual module, as opposed to the dual module, which admits
an extension of the inner product only in the case of W5-algebras.
Let x E M, f E M'. Put
±(f)
The map x '—p ± is an isometric map from the A-module M into the A-module

1±0 = fe 11111 1} 11111 IIxIl lxii

lxii
l._ilx(x) Ii 1
II (x, x) II = lixil
=
For a functional F E we define the functional F e M' by the formula
F(x) :=
Identifying M and M = x e M} C M', we obtain that F is the restriction
of F onto M c M'. Note that for all x G M. It is clear that the map
F '—i F is an A-module map from M" to M' and IFil iF. We shall check
soon that this map is an isometry.
Let us define the inner product (•,•) M" x —+ A by the equality

(4.1) (F,G):=F(G), F,GeM".

It can be directly checked that (F . a,G) = a5(F,G) for a E A. Besides, for
x,y E M, one has
= = = = (y,x)5 = (x,y).
Therefore the inner product defined by formula (4.1) is an extension of the inner
product on M. To check the properties of an inner product we require the following
construction.
Consider the right A-module A x M. In addition to the natural direct sum
inner product (., defined by the formula ((A, x), (b, = a5b + (x, y), where
a,b E A, x,y E M, let us consider another inner product on Ax M. Take I E M',
f 0, and a number t> 11111 and put
(4.2) ((a, x), (b, Y))f,t := t2 a5b + as 1(r) + f(x)5b + (x, y).
75
76 4. REFLEXIVE HILBERT C-MODULES

Properties (iii) and (iv) of Definition 1.2.1 obviously hold. Let us check properties
(i) and (ii). The first one is valid due to the inequality
((a, x), (a, = t2 a*a + a*f(x) + f(x)*a + (x, x)
(4.3) + a*f(x) + f(x)*a +

(4.4) > t2 + a*f(x) + f(x)*a +

= (ta+ (ta+ >0.

Suppose that ((a, x), (a, X))ft = 0. Then equality should be reached at each step
in (4.3)—(4.4). Subtracting line (4.3) from line (4.4), we obtain
(IIfli_2 — t_2)f(x)*f(x) = 0,
hence 1(x) = 0. Therefore t2a*a + (x, x) = 0 and we can conclude that a = 0 and
x = 0, so we have checked the validity of property (ii). Thus, the module A x M
with the inner product defined by formula (4.2) is a Hilbert A-module. We denote
the Hubert module A x M equipped with this inner product by (A x M)f,t and let
the norm on this module corresponding to this inner product be denoted by
Note that lxii. For x,y E M, a E A. we have
(f a + ii = Iia*f(y) + (x, y) ii 1 ((a, x), (0, Y))1,t ii
(a, x) 11(0, y) = 1 (a, x)
Therefore
(4.5) II(f a + £)iI Ii(a,
PROPOSITION 4.1.1 ([1011). Let 1sf c M' be a submodule containing the module
M. Then the norm of any functional i,b E H' satisfies the equality =
PROOF. Without loss of generality we assume that = 1. Define the func-
tional f E M' by the formula 1(x) := x E M. Then if ii ( 1. It is necessary
to prove the inverse inequality, if ii 1. Take g E H such that < 1 and put
ForaE AandxEM wehave
iica+f(x)ii =
(the last inequality follows from (4.5)), hence the map

is a bounded modular map, 1. Therefore

(4.6) x)*fc(a, x) ((a, x), (a, x))91
for all a E A, x E M. From the estimate (4.6) we get
A*c*ca + a*c*f(x) + f (x)*ca + f(x)*f(x) a*a + a*g(x) + g(x)*a + (x, x).
Taking a = —2g(x), we obtain
4g(x)*c*cg(x) + f(x)*f(x) <(x, x) + 2(g(x)*c*f(x) + f(x)*cg(x)).
But since
G(x)*c*f(x) + f(x)*cg(x) g(x)*c*cg(x) + f(x)*f(x),
 4.1. INNER PRODUCT ON BIDUAL MODULES 77

so that
2g(x)*c*cg(x) (x,x) + f(x)*f(x) (1 + If ll2)(x,x),
we have c*ll + 111112)1/2; thus

c*)II = lIcc*II = 11c112

+ Ill 112)1/2.
The last inequality is valid for all g E with IlgI < 1 and, since = 1, the
inequality 1 + IIfII2)h,/2 should be valid too, whence it follows that 11111 1.

Thus 11111
= = 1.
Note that in the case where = M', Proposition 4.1.1 means that the map
F F is an isometric inclusion M" C M'.
PROPOSITION 4.1.2 ([101]). For aliF EM" one has (F,F) 0 and II(F,F)II =
IIFIl2.
PROOF. Let F E M", F 0. Put c = F(F), D = IIFII. Let us show first that
D2 E Sp(c). For t> D consider the inner product (•, on the module A x M.
Since IF, a for all (a,x) E A x M, the map
Ax M—'A,
is bounded and its norm does not exceed D (we mean here the norm defined by the
inner product (•, Therefore

(4.7) (ca+F(x))*(ca+F(x))
Inequality (4.7) holds for all t > D and, taking the limit t —* D, we obtain
(ca + F(x))*(ca + F(x)) D2(D2a*a + F(x)*a +
a = —D2F(x), we get
— 1)*(D_2c — 1)?(x) D2(_D_2F(x)*F(x) + (x, x)),
hence
1)*(D_2c_ 1) + D2(x,x).
Suppose that D2 Sp(c). Then one can find a number 0 such that
— 1)*(D_2c —

for all x E M. But then

- - D2
F(x)*F(x) < —(x,x),
- 1+o
whence
D2=
2D22
- 1+8
The obtained contradiction shows that D2 E Sp(c). But since
Ilcil = IIF'II = hE'll2 = D2,
we have hlchl = D2, hence II(F,F)hI = hIF1I2 and II(F,F)ll E Sp((F,F)). For an
arbitrary element a E A we have
IIacaII = = hI(F'a,F'a)II E Sp((F'a,F.a)) = Sp(aca).
78 4. REFLEXIVE 1-IILBERT

The following lemma concludes the proof.

LEMMA 4.1.3 (1101]). Let an element c E A be such that the inclusion Ilacall E
Sp(aca) holds for any a E A, a 0. Then c> 0. 0

Proposition 4.1.2 shows that the inner product defined on M" satisfies con-
ditions (i) and (ii) of Definition 1.2.1. It remains to check condition (iii). Notice
that
(F+G,F+G).>0, (F+iG,F+iG)0,
hence these expressions are selfadjoint. Then
((F,G)+(G,F))* = (F,G)+(G,F), _i((F,G)_(G,F))* =i((F,G)—(G,F)).
Therefore (F,G) = (G,F)*. The module M" is a Hilbert A-module, since the
operator norm on M" coincides (by Proposition 4.1.2) with the norm defined by
the inner product. Thus, we have proved the following theorem.
THEOREM 4.1.4 ([1011). The map (.,.) x —f A defined by (F, G) =
:

F(G), F, G E M", is an A-valued inner product on M". The norm defined by this
inner product coincides with the operator norm on M". The map F '—p F is an
isometric inclusion M" C M'.
Let us pass now to dual modules of higher order. Let '1" E (M")'. Define the
functional M' by the formula
xcM.
Further, let f E M'. Define E (M")' by the formula
:= (F(f))*, F e M".
The maps ;— f i— are A-module inorphisms. Consider their
composition
(4.8) M' F
E M we have
F41(x) = =
= 1(x)
the composition (4.8) is the identity map, whence the map M' —* M'" is an
isometric inclusion and the map M" M' is an epimorphism. Let us show that
the last map is also monomorphic. For that purpose, apply Proposition 4.1.1 for
the case = M". Let E I'!' = M". Then the functional E M' is the
restriction on M of the functional = Suppose that = 0. Then,
by Proposition 4.1.1, we have = 0, hence the map
is monomorphic. Thus this map is an isometric isomorphism.
COROLLARY 4.1.5. For a Hubert M one has (M")' = M' and
(Jv1tI)II _A4".

So, the series of dual modules M, M',... stabilizes on the third step and the
inclusions
M c M" = M" c M' = M"
 4.2. IDEALS AND BIDUAL MODULES 79

are isometric. Thus M and M11 are Hubert unlike the module M',
which is. generally speaking, only a Banach module. Let us illustrate by examples
that all possible variants can be realized:

(i) Let A be a unital a free A-module with

n generators. Then the module M is self-dual, hence M = M" = M'.
(ii) Let A be a By Theorem 3.2.1, for any Hilbert A-module M
its dual module M' is a self-dual Hilbert module, hence M = M'.
(iii) {44] Let A = C0(O, 1] be the (nonunital) of functions on a
segment [0, 1] vanishing at zero, M = A. Then M' = C[0, 1], M" =
C0(0, 1] and M = M'.
(iv) [44] Consider the module C0(0, 1) of functions on the segment [0, 1] van-
ishing at the end points, over the A = C0(0. 1]. In this case
one has M' = CEO, 1], M" = Co(0, 1], that is, M M'.
DEFINITION 4.1.6. A Hubert M is called reflexive if M" = M.
Later we shall discuss other examples of reflexive Hilbert C*_modules.

4.2. Ideals and bidual modules

Let J C A be a right ideal in a A. The ideal J is equipped with the
natural structure of a pre-Hilbert A-module with the inner product (x, y) = x*y for
x, y E J. In this section we describe the bidual module J". Note that the inclusion
map J c A is bounded and A-linear, hence it defines an element of the dual module
J'. Denote this element by e J' and put
J:= :F J"}.
Obviously .J is a linear subspace of A. Since for a E A and for F e one has
F(ft)*a = (a*F(fj))* = ((Frn

a right ideal in A. Since the equality = x holds for any x C J, we get

J ç J.
PROPOsITION 4.2.1 ([101]). If J C A is a right ideal in a A, then
(i) the ideal J is closed;
(ii) the map i : F '—* F(f1)*, where F C J", is an isomorphism of Hilbert
A-modules J and J".
PROOF. The map i is obviously a homomorphism of A-modules. This map is
also bounded, since = 1. Since the equality
(f1 . x)(y) = xfL(y) = =
holds for any x, y C J, one has x= Therefore
= iIPIIj' = : x C J, lxii 1}
= x
x C J, 1} =
C .J". Hence Iii(F)ii = i an isometry from J"
onto J. Applying condition (ii) of Theorem 2.1.4 to the map i, we obtain that
80 4. REFLEXIVE HILBERT C-MODULES

(i(F), i(F)) < (F, F). Changing the map i by its inverse, we get (i(F), i(F)) =
(F,F), i.e., (F,F) = (F(fL)*,F(f,)*) = It is easy to derive from here
that (F, C) = for F, C E J". Since the module J" is complete with
respect to the norm J is closed. 0
LEMMA 4.2.2 ([1011). In a A consider the set of right ideals K,
Jc K C A, with the following property: any bounded A-module homomoTphism
from the ideal J to A admits a unique extension to an A-module homomorphism
from K to A. There is a unique maximal element in this set of ideals and it
coincides with J.
PROOF. Let K, J c K C A, bea right ideal in A such that any element f E J'
can be uniquely extended to a map f E K'. For a E K define F E J" by the formula
F(f) = f(a)*. Due to uniqueness, the map fL has to be the inclusion map K C A,
hence = fL(a) = a. Therefore K C J. On the other hand, Corollary 4.1.5
shows that any element of J' admits a unique extension to an element in J'. 0
Note that J J C K are two right ideals in a A, then
the inclusion J c K holds. Indeed, for any F E J" the map g from K'
to A belongs to the module K".
Consider now the commutative case. Let X be a locally compact Hausdorif
space and let S C X be its subspace. Denote the of continuous functions
on X vanishing at infinity by Co (X), the of bounded continuous func-
tions on S by C(S) and the ideal of functions vanishing on S by C0(X, S) C C0(X).
If W C X is an open subset, then it is easy to see that there exists a unique maxi-
mal open set W, WC W C X, such that any function in C(W) extends uniquely
to a function in C(W). There is an obvious inclusion W c W, where W denotes
the closure of W. If X is a metric space, then one can show that W = W for any
open subset W. This is not true in general. For example, if X is a Stone space and
if W C X is a dense open subset, then one can show that W = X [34].
Let E C X be a closed subspace and let J = C0(X, E) be an ideal in A =
C0(X). Put U = X \ E and denote by Y the space of all bounded functions oii
X vanishing on E such that their restrictions onto the subset U are continuous on
U. The set Y is equipped with the structure of an A-module (with the pointwise
multiplication), and products of functions in Y by functions in J belong to J. For
g E Y define f9 J' by the formula f9(x) = gx, x E J. The following proposition
describes the dual module J'.
PRoposiTioN 4.2.3 ([101]). The map g is an isometric isomorphism
between the Banach A-modules J' and Y.
PROOF. The map g '—p f9 is obviously an isometric A-homomorphism. We
have to show surjectivity of this map, i.e., to show that for any f E J' there is
an element g E Y such that f = f9. Without loss of generality we can assume
that If III' < 1. Then f(x)f(x) < x E J, i.e., If(x)(t)I Ix(t)I holds
at each point t E X. For any point t U choose a function Xj e J that satisfies
= 1. Define the function h : U —' C by h(t) = f(xt)(t). Note that the
estimate Ih(t) < 1 holds for all t E U. For each y E J at any point t E U one has
(xty — y)(t) = 0, hence f(xty — y)(t) = 0. Therefore
f(y)(t) = f(xty)(t) = f(xt)(t)y(t) = h(t)y(t).
 4.2. IDEALS AND BIDUAL MODULES 81

Since f maps J into A, the function hy should be continuous on the set U for any
function y E J. Now one can take g = h. Then f = 0
After describing the dual module .1' we can describe the bidual module 5 = J".
Let Xu be the characteristic function of the set U. Theii the map is the inclusion
map J C A and J = : F E Y'}.
PROPOSITION 4.2.4 ([101]). The sets J and C0(X, X \ U) coincide.

PROOF. Let D C X be the intersection of the zero sets of all functions

F E Y'. By the closedness of the ideal J one has J = Go(X. D). Since J c J, One
has D c E. Put V= X \ D. It is easy to see that V c U. Let. us show that in fact
the equality V = U holds, i.e.. that any function in C(U) extends uniquely to a
function in C(V) and that if W 3 U is an open set such that any function in C(U)
uniquely extends to a function in C(W). then 14' c V.
Note that for all F E Y', g E V. t U. the equality F(g)(t) =
holds. Indeed, take a function x E J with x(t) = 1. Note that gx E A. Since
gx = XUgx, one has
F(g)(t) = F(g)(t)x(t) = F(gx)(t) =
= =
Let h E C(U); let us show that the function h can be extended to a bounded
continuous function on V (uniqueness of such an extension follows from V c U).
Let g E V be a function such that its restriction coincides with h. Let 14'
be a precompact set that lies in V together with its closure. Choose a functional
F E V' such that = land IIF(xu)II = 1 (by Proposition 4.2.1, IIF(xu)M =
IlFik.,). Define the function h on W by h = F(g)(t), t E W. The function h is
continuous and hi = lihil. For each t E W and for any net with EU
and limo = t, one has, for sufficiently large a,
= = =
h(t) t h(t) Since
any point t E V is contained in some neighborhood of the form W, one can use this
argument to define the function h on the whole V.
Now let W 3 U be an open set with the property that any continuous function
in C(U) extends uniquely to a function in C(W). Let us show that then W c V.
Take a point t E W and a function a E A such that a(t) = 1 and aix\w = 0.
For a function g e Y, let denote the unique bounded continuous extension of the
function to W. Define F V —' A by

F(g)(s) = if s E W,
(0
It is clear that F E V'. Since F(xu)(t) = = 1, it is easy to see that
t D, hence t e V, whence 14' ç V and the proof is finished. 0
COROLLARY 4.2.5 (1101]). Let E be a closed subset of a metric locally compact
Hausdorff space X. Then the Hilbert Co(X)-modnle C0(X, E) is reflexive.
82 4. REFLEXIVE HILBERT C-MODULES

4.3. Reflexivity of Hubert modules over

In this section we describe the reflexivity results from [124]. Let AC he the
of compact operators on a separable Hubert space H, and let AC+ be
the of operators of the form a = A + K, where A E C, K E AC.
THEOREM 4.3.1 ([124]). Any countably generated Hubert is reflex-
ive.

PROOF. By the stabilization Theorem 1.4.2, any countably generated Hubert

module is a direct summand in the standard module 12(J(+). Therefore it is suffi-
cient to prove reflexivity of the module Proposition 2.5.5 gives a description
of the dual module:

= = (fi):

LEMMA 4.3.2. 1ff = e 12(ACj' and jfK e AC, then f.K = (f2K) e
PROOF. Since the operator K can be approximated by finite-dimensional op-
erators, it is sufficient to prove the lemma in the case where K is finite dimen-
sional. Notice that the operator = is a positive operator,
the kernel of which contains Ker K and the image of which is contained in Im K*.
Since dim Irn K* aiid codim Ker K are finite, the norm convergence of the series
follows from its weak convergence. 0
Let FE 12(ACj". Put F1 = E where is the standard basis of
E Since M" C M' for any Hilbert module M. the sequence F =
(Fi) is an element of the module 12(AC+)'. Let us prove that the series
is convergent in the to the element F(F) = (F, F).
Let K E AC be a finite-dimensional operator in H. By Lemma 4.3.2,

(F,F)K = (F,F. K) =

hence
K*(F,F)K =

and
=

for any e H, where (•,•) is a scalar-valued inner product on H. Let 'i e B1(H),
where B1 (H) is the unit ball of H. Choose an element E H and a finite-
dimensional operator K' such that mj = Then

Therefore

(4.9) ij) II (F, F) 11 IIiiII2

for any q.
 4.3. REFLEXIVITY OF HILBERT MODULES OVER 83

LEMMA 4.3.3. Let f = E and E for all i E N. Then

F is continuous and the algebra K is closed, one can assume

that all are finite dimensional operators. Denote by c H the image of the
operator k, dim < oo. Let H = H
a of two closed infinite-dimensional subspaces. Assume first that each of
the subspaces lies in one of the siibspaces H1 or H2. Let F(f) = K + A, K E
A C. Choose a compact operator k with an image L C H2 such that dim L =
Replace by zeros those terms in the sequence (k1, k2,. .), for which C H1, and .

denote the obtaimied sequence by .).

Then (k1k, k2k,...)
.
.
= 14k....)
because of the equivalence of the conditions = 0 awl Tm I link. Thus

i.e.,

=
Since dim L= oc, one has 14,...) = K' + A for some K' E IC. Interchanging
the subspaces H1 and H2, one can construct a sequence 14',...) such that
= K"+A for some K" E IC. Thus
(1. 1. \—I \ ui..'I
— '1' 2

hence
F(k1, k2, .
. .)
= K' + K" + 2A.
Therefore A = 0. In the case of arbitrary subspaces V1 one can find finite-diumen-
sional operators n1 such that
k1 + = ni, + n7, (Iii 12,.. .) E 12(IC+), (nil, rn2,.. .), (ni, 112....) E l2(1C1).
and the image of each of the operators in one of the subspaces H1.
H2. 0
Put F, K + A, where K E IC, A,, A E C. Then
F*F = + +
LEMMA 4.3.4. = A.
PROOF. At first let us show that the series lAd2 is convergent. Suppose
thecontrary. PutM = A112 > M+1.
Choose > 0 to satisfy the estimate

Choose, further, a vector e H with = 1 to satisfy the inequalities

<E. i= 1 N.
Then the inequalities

M+
84 4. REFLEXIVE HILBERT C -MODULES

contradict (4.9). So,

0C 2
< oo. But this means that ()q,,t2,...) E
+
).
Tlieim

But since F(K1,K2,...) + E we conclude that = 0

LEMMA 4.3.5 ([124]). Let X be a compact Hausdorff space and let f,
g be real-valued functions on X, n e N. Assnme that the functions f are
continuous, that the functions + and are nonnegative, that the function 9
is bounded, that the series g
is convergent the function
the series + is convergent to the function g.
the the weak topology. Take
E X and put
= ieii2.
= = A.
All hypotheses of Lemma 4.3.5 are satisfied, so the series uniformly
converges on X to the function Therefore the series con-
verges to (F, F) iii the thus F E 0
4.4. Reflexivity of modules over C(X)
In this section we present results of the papers [91, 125].
DEFINITION 4.4.1 ([125]). A compact Hausdorif space X is called an L-space
for any sequence Ii' 12'... of continuous functions on X converging pointwise to
if,

some bounded function f, the set of the continuity points of the function f is dense
in X.

Examples of L-spaces are compact subsets of a finite-dimensional Euclidean

space [75]. Infinite Stone spaces are not L-spaces.
Definition of L-spaces allows us to give a description of the bidual Hilbert
modules over of functions on such spaces.
THEOREM 4.4.2 ([91, 125]). Let A = C(X), where X is an L-space. Then any
countably generated Hubert A-module is reflexive.
PROOF. By the stabilization Theorem 1.4.2, any countably generated Hilbert
module is a direct summand in the standard module HA. Therefore it is sufficient
to prove reflexivity of the module HA. Proposition 2.5.5 gives a description of the
dual module,

H'4 = {i = : e
 4.4. REFLEXIVITY OF MODULES OVER C(X) 85

Since the sequence of partial sums is moiiotone and bounded at each

point t E X, the corresponding series is convergent pointwise to a bounded function.
By assumption, the set of the continuity points for the limit fuiiction
is dense in X. Let us fix a continuity point t0 E X and let A(t) be a continuous
function on X equal to 1 at to. For F E we have

(4.10) F(f A) = F . A) + F ((i - A).

where is the standard basis in HA C The element A belongs to

the Inodule HA, hence

where by F, = F2(t) we denote Let w(f) be the least upper bound of

oscillation of the function Then, obviously,
2
N
w(f) = lim sup f2(t)12 = un f—
N=OCtEX

Let us choose an arbitrary E > 0 and a function A(t) such that the oscillation w(f)
on the support Supp A(t) is less than s2. Then let us find N such that the inequality
2

(4.11) <2E2

holds for all N' > N. It follows from the inequalities (4.10), (4.11) that at the
point to (where A(to) = 1) we have

IF(f)(to) - (t )f(t)I IIFII .

for all N' > N. Thus, the series converges to a continuous function
F(f)(t) at all the continuity points to of the sum Moreover, we shall
see soon that this series is uniformly convergent, hence it converges to a continuous
function, which should coincide with the function F(f)(t) due to the property L of
X. The uniform convergence of this series would follow from that of the series

(4.12)

because of the Cauchy-Bunyakovskii inequality (Proposition 1.2.4). So we have to

prove the uniform convergence of (4.12) on X.
Since C the series (4.12) converges at each point and it.s sum coincides
with the continuous function (F, F)(t) = F(F)(t) at each point of continuity. De-
note by E C X the set of the continuity points for the series (4.12). Now let t0 be
some point of discontinuity for the series (4.12). Without loss of generality (mul-
tiplying by some continuous function, if necessary) one can assume that there is a
sequence of points E E converging to the point to and satisfying the conditions
 4. REFLEXIVE HILBERT C-MODULES

(F, = 1 and 1F1(to)12 < 1. Choose functions E C(X) to satisfy

the conditions
(i) h,(t0) =
h = (hi) E H,1, i.e., the series
(ii) uniformly converges.
Define the function A(t) on X by the following conditions:
(i) 0 < < 1;
(ii) outside the point to the function A(t) is continuous.
Consider the element
f(t) =
+ — h2(t)).
It can be easily checked that the functions f. (t) are continuous on the whole X.

Moreover, since we have f = (fe) E Then

F(f)(t) = (F,f)(t) = + -
The series

(4.13)

is uniformly convergent since h E HA. The series

uniformlyconverges on the set E C X to the function (F, F — h)(t), which is

continuous on the whole X. Then, for t E E, one has
(F, f)(t) = (F, h)(t) + A(t)((F, F)(t) — (F, h)(t))
and, for each
(4.14)
(F, f)(t) (F, h)(t) are continuous and, since the series (4.13) is
uniformly convergent,
(F,h)(to) <1.
=
But this contradicts the continuity of the function (F, f)(t), because, due to our
choice of the function the limits of the right-hand side and the left-hand side
of the equality (4.14) are different for even and for odd n. 0
4.5. Hubert modules related to conditional expectations of finite index
In this section we describe results of the paper [47]. Let E: A —p B ç A be a
faithful conditional expectation on a A, i.e., a projection of norm one
onto a B such that the condition E(x*x) = U, x e A, implies x = 0.
A conditional expectation E : A B c A is called a conditional expectation of
algebraically finite index if there exists a set of elements {u1, .. , u,, } C A such that
.

x= for any x E A. Then the element Ind(E) of the

center of the A is called the index of E. Ind(E) does not depend on a
choice of the elements {u1,. c A, it is positive and Ind(E) 1A [133, 5]. It
. . ,

is shown in [5] that the algebraic finiteness of the index is equivalent to the property
 4.5. CONDITIONAL EXPECTATIONS 87

of A to be a projective finitely generated module over the

B.
It is also interesting to consider another class of conditional expectations E
A — B c A, for which there exists a number K 1 such that the map
(K - E — idA) is a positive operator on the A (cf. [45]). Such con-
ditional expectations are called conditional expectations of finite index and they
have the following property:
PRoPoSITION 4.5.1. Let A be a and let E : A —i B c A be a
conditional expectation, for which the set of fixed points coincides with B. Then
there exists a finite number K 1 such that the map (K . E — IdA) is positive if
E is faithful and the (right) pre-Hilbert B-module {A, E((., is complete with
1/2 *
respect to the norm IIE((., )A)IIB (where (a, b)A = a b for a, b E A).

PROOF. If the algebra A is complete with respect to the norm IIE((.,

then there exists a number K such that the inequality KIIE(x*x)II holds
for each x E A. For a E A, E >0, put x =a(E+E(a*a))_h/2. Note that
(E + .
. + E(a*a))_V2
whence the inequality K . 'A + E(a*a))_V2 . ama. (E + E(a*a))_h/2 follows.
Multiplying both sides of this by (E+E(a*a))h/2 , we conclude that K. (E+E(a*a))
a*a for all a E A, E > 0. The inverse statement easily follows from the inequality
IIE(x)II which is valid for any x E A. D

Notice that, unlike algebraic finiteness of the index, in the case of conditional
expectation of a finite index the Hubert {A, E((., can be infinitely
generated.
Put
K(E) = inf{K such that (K. E — idA) is positive on A}.
We call K(E) tile characteristic number of the conditional expectation E.
Let X be a compact Hausdorif space with an action of a group G. Denote by
CG(X) the of G-invariant continuous functions on X and by
G : gx x} the stabilizer of a point x E X.

DEFINITION 4.5.2. A continuous action of a group C on X is called uniformly

continuous if for each point x E X and for each neighborhood U,, of x there is a
neighborhood of the point x such that gE

Note that the continuous action of a compact group satisfies this definition.
DEFINITION 4.5.3. Let a group C act uniformly continuously on a compact
Hausdorif space X in such a way that the cardinality of each orbit #Gx does not
exceed some number k E N. Define the conditional expectation : C(X)
EG(C(X)) c C(X) by the formula

EG(f)(x) f(gax), x E X.
#(C/C
LEMMA 4.5.4 ([47]). The conditional expectation is well defined.
 4. REFLEXIVE HILBERT C-MODVLES

THEOREM 4.5.5 Let a group C act uniformly continuously on a locally

compact Hausdorff space X so that k := max(#(Cx) : x E X} < +oc. Then the
characteristic number of the conditional expectation E0 satisfies the equality
K(EG) k max#(Gx).
vEX
PROOF. Let. .r E X be an arbitrary point and let = #G.r. Then
KEc(f)x = f(qr) f(x).
'
where f is an arbitrary nonnegative function in C(X). ASSume that K(EG) < k
and choose a point .r such that k.r > K(E(;). Then OflV caii choose a sufhciently
small neighborhood of thc point .r such that fl = 0 (I j) for the
set = 1, 92 } = G/GJ.. Let f 1 a colit iiiuous liulinegative function with
support iiiside LI1. Then

K(Ec:)EG(f)x = > <f(x).

Contradiction with the definition of K (Es) completes the proof. 0

THEOREM 4.5.6 (combination of Proposition 4.5.1, Leninia 4.5.4, and Theorem
4.5.5). Suppose a discrete gn.up C acts on a locally compact Hausdorff space X in
such a way that k := max{#(Gx) : x E X} < +00. Then E0 is well defined.
If X is a noriTial space. then K(EG) = k. Hence, by Proposition 4.5.1, C(X)
a Hubert over GG(X).
THEOREM 4.5.7. Let X be a compact Hausdorff space and let C be a group
acting uniformly continuously on X. If all orbits of the action of C have the same
finite number of points, then the conditional erpectation
E(f)(x) > f(g1.r)
= #(Gx)
is well defined on C(X), and the Hubert {C(X). E((.. .))} is finitely
generated and projective.
PROOF. The idea of the proof is contained in [133J and requires two technical
lemmas.
LEMMA 4.5.8. Let X he a compact Hausdorff space with a uniformly continuous
action of a group C and let all orbits contain an equal finite number of points. Then,
for any point x E X and any element g E C1. one can find an open neighborhood
U1 of a point x, on which g acts identically.
PaOOF. Let us denote the cardinality of orbits #(Gx) by n. Let x1 E X
be the orbit of the point a: and let E C be elements such that = x1. Choose
E and assume that each neighborhood U1 of the point x contains some point
y E U1 such that g0y y. Fix neighborhoods U1, of the points satisfying the
condition U.r, fl Us.., = 0 for i j. Then we can find a neighborhood c U1 of
the point x such that c Ut,. Since the group C acts uniformly continuously,
one can find a neighborhood W1 c V1 of the point x such that g(W1) ç V1 for
each g e C1. If y W1 and goy y, then the orbit Cy of this point contains at
 4.5. CONDITIONAL EXPECTATIONS 89

least n+ 1 different points {h7y E : i= 1 E The obtained

contradiction proves the lemma. 0
4.5.9. Under the hypotheses of Lemma 4.5.8. for any point x E X one
can find a neighborhood of this point such that the action of the subgroup G1 on
V1 is trivial.
PROOF. We should show that a neighborhood from Lemma 4.5.8 can be
choseii for all g E simultaneously. For each g E put
U1(g)={yEX:gy=y}.
Suppose the contrary, i.e., that the set U1(g) does not coiitain a neighbor-
hood of .r. It means that any neighborhood U1 of x contains some point z such
that for some e C1 we have g:z :. Consider a neighborhood of a: and
neighborhoods {Uh.r} for fixed representatives {h1 e} e G of cosets in C/C1
such that their intersections are empty pairwise and c U,,,.. As in the
proof of Lemma 4.5.8. we find a neighborhood W,. c Va. of the point x such that
g(W.E) ç V,. for each g e C1. Put. U.,. = U9EG, G(W1) c Va. It is a G1-invariant
open neighborhood of the point x E X. The assumption g:z z for some z E U,.,
E means that the orbit of the point z consists of at least ii + 1 points. 0
Let C Va be a neighborhood of the point x such that the action of C,. on
V,. is trivial. Then one can find a function f,. E C(X) such that supp f1 C V1 and
= 1. For each g E G one has either (gV,.) fl V,. = 0 or gV,. 1/,.. Therefore
— JQ(f)2 if g.r = x.
10
where denotes the action of C on functions. = f(g1.r). Consider a
finite covering {U,. , } of the space X by sets of the above form. Put

v= > 1, = v"2(f,.)112 e C(X).

Note that if we take one element in each coset C/Ga., theim, by Lemma 4.5.4, the
map
Ec(f)(x)
= #(Gx)
is well defined for all a: E X. f C(X) and it is a conditional expectation omm C(X).
Moreover, one has

= f,

E C(X); hence the set {ui, . . . , is a basis of the Hilbert

CG(X)_module {C(X), EG((., .))}. Therefore this Hilbert module is finitely gener-
ated and projective. 0
Theorem 4.5.7 generalizes results of [133) and shows that if all orbits consist
of an equal finite number of points, then the corresponding conditional expectation
is of algebraically finite iiidex. Being finitely generated and projective, the Hilbert
module A = {C(X), E((., .))} is self-dual. In the case of a finite index (when
90 4. REFLEXIVE HILBERT C-MODULES

< cc and the pre-Hilbert module A is complete) we cannot expect this

module to be self-dual. However, sometimes this module is reflexive, i.e., A" = A,
where A' is the dual Banach Cc (X )-module of bounded Cc (X)-homomorphisms
from A into CG(X).
THEOREM 4.5.10 ([47]). Let the group G act uniformly continuously on a com-
pact Hausdorff space X. Suppose that all orbits consist of not more than ii points,
and that the number of points for which the length of their orbit is less than ii is
finite. Then the Hubert {C(X), .))} is reflexive.
PROOF. Describe first the dual Banach A'. Let x1,. . . ,

be the points with cardinality of orbits smaller than n. One can choose open
neighborhoods U1,.. . , Urn of these points in such a way that each neighborhood
U2 is invariant with respect to the action of the subgroup and if, for some
h E C, one has hx2 = x3, then hU2 U3. Denote by V the C-invariant compact
set X \ (U1 U . . U Urn). Let F E A' be a functional on the module
A. Consider its restriction on the Hubert {C(Y), .))}. For a
function g C(Y) we take its extension E C(X) and define by the formula
F]y(g) = This definition does not depend on the choice of an extension
If V' V is another compact G-invariant subspace not containing the points
then (Fly')ly = F]y. Since the orbit of each point of the set V has
the same cardinality, by Theorem 4.5.7, the C'(Y)-rnodule {C(V), Ec((., is
finitely generated and projective, hence self-dual. Denote by C(X \ {xi,. })
. . ,

the set of continuous functions on the noncompact space X \ {xi,. ,Xm}. The . .

restriction onto this space defines the map

(4.15) A' —b C(X \ {x1,.. . ,Xrn}).

It is easy to verify that the map (4.15) is injective.

Let us study local properties of functionals from A' in a neighborhood of the
points x1 Xrn. Let Xo be one of these points. It has a neighborhood such
that if gx0 = x0, then = The group contains a normal subgroup
C0 of the elements that do not move points from the neighborhood Choose a
representative 9j in each coset G/GXO. Then outside the point x0 the action of the
functional F E A' can be written as

(4.16) F(f)(x) = > .

and this action can be continuously extended to the point x0. Consider the orbit
= x0,x',.. . of the point x0. Then one can write the sum (4.16) in the
form
k—i f
F(f)(x) = .
(
jrO
Passing to the limit (which exists by assumption), we obtain

i F*(gjx)
F(f)(xo) .
= j0 ( i:g,x0=xJ
 4.5. CONDITIONAL EXPECTATIONS 91

hence there exists (for f 1 E C(X)) the limit

F(g2x)

for any x E X \ {x1,. . . , Recall that the function F(x) is defined only out-
side the point x0. If we would like the action F on the Hubert
{C(Y), .))} to be of the form (4.16) on the whole X, it would be necessary
to define the function F(x) at the point x0 by
(4.17) F(xo) = lim F(g2x).
n x—'xo

To complete the proof we need the following lemma.

LEMMA 4.5.11. The module A' is isomorphic to the module of all bounded func-
tions F(x) on X that are continuous on X \ {x1,. . , Xm} and satisfy the condition
.

(4.17).
PROOF. We have to show that the image of the monomorphism (4.15) consists
of bounded functions. Suppose the contrary. Then there exists a point such that
> n.IIFII, where IIF1I is the norm ofF in the dual Banach A'.
Moreover, one can choose a neighborhood of the point such that =0
for those elements of the group G, for which ±. Consider a function f E C(X)
such that f(±) = 1 and supp f C Then it follows from the equality (4.16) that
= . f(±),
and the inequality
= = > IIF1I

gives a contradiction. 0
Having the above description of the dual module A', we can describe the bidual
module A". Since one has the canonical inclusion A" C A', and since the inner
product on A can be naturally extended to an inner product on A" (see Theorem
4.1.4) making it a Hilbert it is sufficient to verify to which functions in
A' one can extend the inner product. Consider a function F from
A' fl A". Adding to it (if necessary) a continuous function from A, we can assume
that F(xo) = 0. Then the C°(X)-valued inner product of F by itself is an element
of CG(X) of the form
(4.18) (F, F)(x) = E(IF(x)12) =

for all x E X. But since (F, F)(xo) = 0, it follows from the assumption F E A"
that
lim F(g2x) = 0
x—.xo
for each summand of the equality (4.18). Therefore we obtain from (4.17) that the
function F is continuous at the point x0, hence the module A is reflexive. 0
Later in this section we will prove a final version of Theorem 4.5.10 due to
V. Seregin.
92 4. REFLEXIVE HILBERT C-MODULES

CONSTRUCTION 4.5.12. Consider a uniformly continuous action of G on a Haus-

dorif compact space X with bounded cardinality of orbits. Let us develop the
construction used in the proof of Theorem 4.5.10. For any point x0 E X let
be a neighborhood of x that is invariant under the action of the stabilizer of
the given point and is disjoint from its own images under the action of the group
elements not belonging to Such a neighborhood exists due to uniform con-
tinuity: indeed, take a neighborhood with all these properties except that of
invariance; then choose, by the definition of uniform continuity, a neighborhood
Wxo with CxoWxo C and put :=
In this case one can take representatives g3 (j = #Gx°) for the cosets
. . . ,

for the subgroup C and the neighborhoods that appear in this construction
are of the form = = g3(x°), where is the orbit of x0. Let
X —* [0, 1], E H, be a continuous function such that

= Io,
11, x=x.
Since the cardinality of orbits is bounded, E is well defined and one can assume
that is Suppose = so C Then
:= is invariant, equals 1 on the orbit x0 and is supported in U, U1o. All
these properties, except the invariance, are obvious. If g C and y U1o. then
y gjz, where z E and gg, = gkh for some and h E Gxo, so

pu(gy) = = = pxo(z) = Pj(Y)

Similarly, we can choose a neighborhood with the same properties as
and contained in together with its closure. Then, taking a function in-
stead of such that it equals 1 on and has support in we obtain the
corresponding function
THEOREM 4.5.13 ([118]). Let X be a compact Hausdorff space and let G be
a discrete group acting uniformly continuously on X. If the cardinalities of all
orbits are uniformly bounded by some number N, then the Hubert Cc(X)-inodule
H {C(X), is reflexive.

PROOF. Denote by B(X) the space of all bounded functions on X and by H

the space of all maps from H to B(X).
Choose and Px for any x as in Construction 4.5.12 and define the maps
H' B(X) and 71: B(X) H as follows:
(4.19) =
and

ij(f)(h)(x) =
(4.20)
# Xc
where F E H' and #Gx denotes the number of points in the orbit of x, as before.
Let us verify that is well defined, i.e., that the function f = is bounded:
f(x)I <NIIFII.
 4.5. CONDITIONAL EXPECTATIONS 93

Let us prove that = F. Take an element h E C(X) and a point x E X.

Then

F(h)(x) = F (h — + (x)

= F (h - (x) (x)

F(hi)(x) +
EGx

= F(hi)(x) +
x,EGx

= F(hi)(x) +

h — Here we have taken into account that E

CG(X). Note that the function h1 is identically 0 on the orbit of x. Then F(h1)
also is identically 0 on the orbit of x. Indeed, choose (as in Construction 4.5.12) the
neighborhoods so small that IhiI <e holds on their union. Then <e
for the corresponding invariant function PU and for arbitrary e > 0 we have

IIF(hi)(x) II = = IIF(hipu)(x) <

Since x is an arbitrary point, this implies o = Id and = H'.
Let us prove now that the map ij is injective. Suppose that f 0 E B(X), so
we have f(x°) 0 for some point x0 E X. Choose a continuous function h E H
satisfying the following conditions:

h(x°)=1,
Then = 0, hence 0 and = 0. Therefore

(4.21) H' C B(X), ,j: H' C H.

LEMMA 4.5.14. The image of e consists of functions f E B(X) such that for
any point xo E X there exists a neighborhood of xo such that
#Gxo
(4.22) f(xo) = lim f(xt).
#Gx

PROOF. Let f satisfy condition (4.22). To prove that f E it suffices to

show that, in this case, ij(f) E H'. Then (4.21) immediately implies that = f.
The property ij(f) E H' means that zi(f)(h) is a continuous invariant function
for any function h E H and that the correspondence h '—+ ij(f) (h) is bounded and
linear over CC(X). Invariance follows at once from (4.20) and continuity follows
from the calculation below, where we denote the points of the orbit of x (resp.
x0) by (resp. by x?) and we suppose that x E (see Construction 4.5.12), so
Gx C U1 and one can take := — (j = 1, #Gx°). . . .
94 4. REFLEXIVE HILBERT C-MODULES

Note that 0 as the neighborhood decreases (as well as Then

Iii(f)(h)(x)-ii(f)(h)(x°)I =
-
=
j ( > - >
i
J
<sup

+
( > -
2
/
11111
+
-
Hence, by (4.22) (with instead of x°),
ij(f)(h)(x) — ii(f)(h)(x°)I = 0.

Additivity of 77(f) follows immediately from (4.20). Now let r E C'(X) be an

invariant function. Then
77(f)(hr)(x) = > =
x,EGx x,EGx
= 77(f)(h)(x)r(x)
and ri(f) is a
For any h E H one has

77(f)(h)(x) = > > . 11111 IIhII,

x,EGx
so 77(f) is a continuous functional. Hence, 11(f) E H'.
Conversely, let us prove that condition (4.22) is satisfied for any function f
from the image of For f, there exists a functional F E H' such that f =
Namely, F = 17(f). Consider an arbitrary point x0 and choose for it a neighborhood
Vso c (see 4.5.12). Take the function (see 4.5.12) and evaluate F at this
function. Then, for x E Vxo, the function defined by the formula

F(4)(x) = = >
z

is continuous because it lies in the image of the functional F. This is equivalent to

the condition
f(x1)
= f(z)
x, zEGx°flV,,o

Thus, (4.22) holds for f and Lemma 4.5.14 is proved. 0

 4.5. CONDITIONAL EXPECTATIONS 95

Let us continue the proof of the theorem and let us give a functional description
of H". Define the maps ii: H" .8(X) and B(X) —* H, where H is the space
of all maps from H" to B (X), by the formulas
v(F)(x) =
and
=

x E X, h = h= Note that
(4.23) IIhII I((h)(coz))(x)I lihil.

Then

F(h)(x) = F (h — +> (x)

x,EGX

=F (h - h(x1)fx) (x) + F ( (x)

x,EGx
= F(hi)(x) +
x, EGx

= F(hi)(x) +
x, EGx

= F(hi)(x) +>
h — >xEGx and invariance of the functions F(f1,) was used.
Further, for any point y E Gx, one has
hi(y) = h(y) —

x,EGx
= h(y) — h(y) = 0,
so the function F(h1) equals 0 on this orbit.
Indeed, choose, as in Construction 4.5.12, the neighborhoods so small that
Ihil < r holds on their union. Then < e for the corresponding invariant
function and

= = <Elif II.
x,EGx
So we obtain
IIF(hi)(x)II = = IIF(hipu)(x)II <rilFil
for arbitrary e > 0, hence
(4.24) pv=Id.
96 4. REFLEXIVE HILBERT C-MODULES

Let us prove now that p. is injective. Suppose that f 0 E B(X), so, for some
point x0 E X, we have f(x°) 0. Then

0,
= =

hence 1u(f) 0 and Kerp = 0.

Suppose that f Im ii. Then there is a functional F E H" such that f =
Since p is injective, by (4.24), one has F = p(f). For an arbitrary point x choose
and (see Construction 4.5.12). The function is continuous and one
has

=
= y, EGY

=
=
in the neighborhood of x. As was already shown, condition (4.22) is satisfied
for 1. Thus I is bounded due to the estimate

If(x)I = <NIIFII.
Hence fc Moreover, this implies the continuity of ii: H' —' v(H') C B(X),
when the last space is equipped with the uniform norm. On the other hand, taking
into account (4.23), one has

IIF1I = sup IIF(h)II = sup II,.t(f)(h)(x)II = sup

IIhIl=1 x,IIhII=1 x,IIhII=l #Gx
sup N . 11111 .
llhfl=l
Hence, H' is isomorphic to the image of v(H') equipped with the uniform norm.
Let us prove that each f E Tm z' is continuous. Let x E X. One can assume
that f = 0 on Cx. Indeed, it is clear from the definition of ii that f is invariant and
vp(1) = 1 for the identity element, since otherwise we can change f by a constant.
It follows from above that one can define F := p(f) E H" and F = E H'.
Then the function

is continuous and invariant.

Let xa be a directed set of points of X that converges to x0. Then
urn = 0,

hence
un
 4.5. CONDITIONAL EXPECTATIONS 97

In particular, lima f(xa) = 0. Therefore the function f is continuous at an arl)itrary

point x. Hence f c C(X) and H = H". 0
Let us complete this section with some statements, which are not directly re-
lated to reflexivity. Details can be found in [131].
THEOREM 4.5.15. Let C be a discrete group acting uniformly continuously on
a compact separable Hausdorff connected topological space X. Suppose that either
the averaging over orbits defines the structure of a self-dual Hubert GG(X)-module
on C(X) and GG(X) is MI (cf. 2.5.8) or C(X) is a finitely generated projective
module over C1G(X). Then all the orbits have the same finite cardinality.
DEFINITION 4.5.16. An action of a group C on a metric space X is called
(Lyapunov) stable if for any E > 0 and any x c X there exists S > 0 such that
p(gx, gy) for each g E C provided p(x, y) <5

THEOREM 4.5.17. Let a discrete group G act on a compact metric space X. If

the action is stable, then EG is well defined.
 CHAPTER 5

Multipliers of A-Compact Operators.

Structure Results
5.1. Extension of a Hubert Ct-module by the enveloping Wt-algebra
CONSTRUCTION 5.1.1. Let A be a Ct-algebra, Att the enveloping Wt-algebra
of A and M a Hubert A-module. Consider the algebraic tensor product M ® A**
(over C). One can equip this tensor product with the structure of a right
module by the formula (x 0 a) . b := x ® ab, x E M, a, b e A**. Define the inner
product
[.,.]:M ® Att x M ® Att Att
by the formula
n m

i=1 j=1 i,j

where x2,y3 e M, e Att. The sesquilinearity and the properties [z,w] =
[w, z]* and [z, w a] = [z, w]a
clearly hold. To verify that this inner product is
positive, we use the following well-known statement.
LEMMA 5.1.2 ([123. Lemma IV.3.21). Let B be a Ct -algebra, e B. i,j =
1,...,n. A matrix[cjjl e is positive > Oforanyb1 b,1 E
B.
Since for any al E A, one has

the matrix E is positive. Therefore the clement

is positive for any b1 e Att, hence [z, z) 0 for any z E M ® Att. Set
.N={ZEM®Att :[z,z]=0}.
Then is an Att-submodule in M 0 A*t and the quotient module M 0 At*/Ar
is a pre-Hilbert Att-module. Denote by M# the Hilbert Att-module obtained by
completion of M 0 Att/A( with respect to the norm given by the inner product
[., .]. We call this module the extension of the module M by the algebra Att. The
Att always contains the unit element and for any x E M, a E A, we
have (x. a) ® 1 — x 0 a E .A/. Therefore the A-module map x x ® 1 + .ftf,
is well defined. Since [xO 1 1-i-N] = (x,y), this map is an
isometric inclusion.
Denote by HomA(M, Att) the set of all bounded A-linear maps from M to
At*. Introduce on this set the structure of a vector space over C by the formula
99
100 5. MULTIPLIERS

(Ao)(x) := where A E C. x E M. E HomA(M.A**). and also with the

structure of a right A**_module by the formula b)(x) b E A**. For a
functional I E (M#)' we can define the map fR E HomA(M, A**) as the restriction
of f onto M. namely. fR(.r) := f(x ® 1 + 1V). Obviously 11111.

THEOREM 5.1.3 ([100]). For any A and for any Hubert A-module
M. the map f fR is an isometry of (M#)' onto Ho1nA(M, A**).
PROOF. Let. a matrix e satisfy the condition >0
for any e A. Let us show that it is sufficient to prove positivity of the
matrix For this purpose, it is sufficient to show that
(5.1) >0

for any b1.. . ,.E A**. Without loss of generality one can assume that the
elements are in the unit ball Bi(A**) of the W*_algebra A**. Since the unit ball
B1 (A) of the C*_algehra A is dense in B1 (A**) with respect to the strong* topology.
oiie can find iiets E A, A E A, converging to the elements E A** with respect
to the strong* topology. Then the net converges to with
respect to the weak topology (see [123, whence the inequality (5.1) follows.
We have to show that any map e HO1I1A(M, A**) with < 1 can be ex-
tended up to a unique functional f E (M#)' with If II 1. Consider the functional
fo : M A** A** given by the formula

fo ®ai) =

Obviously fo is an map. For E A, one has

= . aj)*Ø(xj . a3)

;r7 ai)
= *

i=1 i=l i.j

Therefore for any E A**, the following inequality holds:

i.e.,
fo(z)*fo(z) <
for any z E M ® A**. Therefore the functional f : A** is well defined by
the formula
xi ® + :=

and satisfies the inequality f(g)*f(y) y

f e (M#)'. It follows from the equality f(x ® 1 + = 0(x) that f iS all
extension of 0. 0
 5.2. MULTIPLIERS AND CENTRALIZERS 101

COROLLARY 5.1.4. Let A be a a Hubert A-module.

Then an A-valued inner product on M can be extended up to an inner
product on the set HomA(M, A**) making this set a Hubert
COROLLARY 5.1.5. Let A be a a self-dual Hubert
A-module. Then the Hilberi M# is self-dual too.
As one more corollary we shall present the following characterization of self-dual
Hubert modules.
THEOREM 5.1.6 ([42, 40]). Let A be a Then the following state-
ments are equivalent:
(i) the Hilbert A-module HA is self-dual;
(ii) the A is finite-dimensional.
PROOF. Note that each condition of the theorem implies the existence of a
uiiit in the C*_algebra A. Indeed, if A is finite—dimensional, then 1 E A. If the
module HA is self-dual, then the bounded A-module map f HA A defined by
the formula 1(a) = a1. where a = E A, i e N, has to be an element of the
module HA itself. This means in particular that End.4(A) = and so the
identit.y mapping A —* A is identified with the unit element of A.
In what follows, let us use the description of the dual module H'4 as the set of all
sequences b = (bk), E A, such that the partial sunis = of the series
are uniformly bounded. Self-duality means that for any increasing sequence
(efl) of positive elements of the C*_algebra A, its boumidedness is equivalent to its
norm-convergence. But, for finite-dimensional C*_algebras, the monotone bounded
sequences are convergent, and this proves the implication (ii) (i). To prove the
opposite implication we pass to the Hubert module = over the enveloping
147*_algebra A**. This module is self-dual by Corollary 5.1.5. Since any monotone
bounded sequence in a I'V *_algcbra is convergent, any positive linear functional on
A** has to be normal, i.e., = (A**)*. Suppose that the W*_algebra A** is
infinite—dimensional. Then it contains an infinite collection of mutually orthogonal
projections Pk E A** such that Pk = 1. Hence there exists an inclusion of the
commutative of bounded sequences into A**. Let p be a
positive linear functional on the algebra Let us extend it to a positive linear
functional on the larger algebra A**. According to time assumption, is normal.
Therefore its restriction = to the algebra is normal too. Hence we have
obtained a false conclusion = This contradiction shows that the
algebra A** is finite-dimensional, hence the A is finite-dimensional as
well and this proves the implication (i) (ii). 0
An interesting general theory of extensions of Hubert. C*_moclules can be foimd
in [6].

5.2. Multipliers and centralizers

While writing this section we extensively used results of [104, 134]. Let 8(H)
be the algebra of all bounded operators on a Hubert, space H and let A be a
algebra.
DEFINITION 5.2.1. A two-sided closed ideal J C A is called essential if JflJ'
0 for any nonzero ideal J' C A.
102 5. MULTIPLIERS

REMARK 5.2.2. Au ideal J C A is essential if and only if

J± :={aeAIaJ=O}=O.
DEFINITION 5.2.3. A representation p : A 13(H) is called nondegenerate if
for any ii E H there exists an element a e A such that p(a)h 0.
REMARK 5.2.4. For an arbitrary representation p, we can consider its restric-
tion onto the orthogonal complement H' to the invariant subspace
H p(A)h = 0}, which is invariant as well. The new representation p' : A —+ 13(H')
is then nondegenerate. Thus, roughly speaking, we lose nothing working only with
nondegenerate representations.
LEMMA 5.2.5. A representation is nondegenerate if and only if p(A)(H) is
dense in H.
PROOF. Let a representation p be nondegenerate and let h±p(A)(H), i.e., for
any f E H and any aE A,
0= (h,p(a)f) (p(a*)h,f)
holds. Then p(b)h = 0 for any b e A, hence h 0.
Conversely, let p(A)(H) = H and let h e H be an arbitrary nonzero vector.
Without loss of generality one can assume that lihil = 1. Since p(A)(H) is dense in
H, one can find g e H and a E A such that Ih—p(a)gII <1/2. Then > 1/2,
1/4> (h — p(a)g,h — p(a)g) = 1— (g,p(a*)h) — (p(a*)h,g) + 1/4,
(g,p(a*)h) + (p(a*)h,g) > 1, p(a*)h
u
DEFINITION 5.2.6. Let p: A 13(H) be a faithful nondegenerate representa-
tion, so we can assume A c 13(H). An operator x e 8(H) is called a (two-sided)
multiplier of A if
axEA
for each a e A. Denote by M(A) the set of all multipliers. It is easy to see that
M(A) is au involutive unital algebra.
REMARK 5.2.7. It seems that the definition of multiplier depends on the choice
of a (nondegenerate faithful) representation. In Theorem 5.2.11 we shall see that
this is not the case.
PROPOSITION 5.2.8. The set M(A) is a unital
C
A an essential ideal in M(A). If A has no unit, then C M(A).
PROOF. Three statements are nontrivial: 1) that M(A) is norm-closed, 2) that
the ideal A is essential, and 3) that there exists an inclusion into the bidual space.
1) Let x with respect to the norm, Xe. E M(A), x E 13(H). Then
x0a xa and ax for any a. Since A is closed, xa E A and ax E A for any
a, i.e., x e M(A) by definition.
2) Let J be an ideal in M(A), JflA = 0, and let x E J be an arbitrary element.
Then xa E A (since x is a multiplier) and xa E J (since a E A C M(A) and J is
an ideal) for any a E A. Therefore xa E J fl A = 0, xa = 0 for any a E A. Then
x = 0 by Lemma 5.2.5.
 5.2. MULTIPLIERS AND CENTRALIZERS 103

3) Since A!! A** (cf. remark after Theorem 3.1.3), it is sufficient to

prove that M(A) C For this purpose, note that for any x E M(A) and for any
weakly convergent net aA E A, we have
x w-limaA = w-lim(xaA) E
.XEA AEA

since xaA E A. where [AIW is the weak closure of A in 8(H) and where w-Iim
denotes the limit with respect to the weak topology. One has = A!! due to
the nondegeneracy of the representation, hence XA!! = C = A!!. Since
1EA",onehasxEA1'. D

The next definition historically precedes Definition 5.2.6.

DEFINITION 5.2.9. A pair (L, R) of maps
L:A A, R: A A, R(a)b = aL(b) for each a, b E A,
is called a double centralizer of A. Let us denote the set of all double centralizers
of A by DC(A).
PROPOSITION 5.2.10. Let (L,R) e DC(A). Then
(i) L(ab) = L(a)b and R(ab) = aR(b);
(ii) L and R are linear;
(iii) L and R are bounded and IILIt = IIRII.
The set DC(A) with the operations
(Li,Ri)+(L2,R2):zr(Lj+L2,Ri+R2), z(L,R)=(zL,zR), zEC,
(L1,Ri)(L2,R2) := (L1L2,R2R1),
(L,R)* :=
L*(a) (L(a*))*, R*(a) := (R(a*))*, a A,
is a normed involutive algebra with respect to the norm
(L, R) II := = II RD.
PROOF. 1) Let a,b E A, z E C and let ea (ck E A) be an approximate unit of
A. Then
eaL(ab) = R(ea)ab = eaL(a)b, L(ab) = L(a)b,
CQL(za + zb) = + zb) = + R(ea)zb +
= + = + L(b))),
L(za + zb) = z(L(a) + L(b)).
2) Thus L is a linear operator on the Banach space A and, to prove its continu-
ity, it is sufficient to check that its graph is closed. Let a and —p b.
Then
IIv(L(a) — b)II IIvL(a) — + — vbII
= IIR(v)(a — an)jj + IIvL(an) — vblI
IIR(v)II . ha — + hivhl hhL(an) — bhI 0

for any v E A, hence vL(a) = vb, whence b = L(a), since v was taken arbitrarily.
Thus the graph of L is closed, so L is continuous. Properties of R can be verified
in the same way.
104 5. MULTIPLIERS

3) Let us compare ILl! and IIRII:

IL!!2 = IIL(a)112 = IIL(a)*L(a)II = sup IIR(L(a)*)aII

IIaII=1 IIulI=1

< sup IIRII IIL(a)*II Mall < sup IIRII IILII 11a112 = IIRII IILII.
IIaII=1 IIaII=1

whence IILII IIRII. A similar calculation gives the opposite inequality.

The remaining statements are obvious, one has only to verify that
R2(R1(a))b = R1(a)L2(b) = aL1(L2(b)).
0
THEOREM 5.2.11. The map
p : M(A) DC(A), x (Li, = = ax.
is an isometric *-isomorphism between M(A) and DC(A). Therefore DC(A) is a
and M(A) does not depend on our choice of a nondegenerate represen-
tation.
PROOF. First of all one has aL1(b) = axb = so the pair (Lx, does
belong to DC (A). The linearity of the map p is clear. Also,
= (xy)a = x(ya) = =
= a(xy) = (ax)y = =
= (LXL,J.RYRX) =
thus p. is a honiornorphisin of algebras. It is involutive,
= (Lxa*)* = (xa*)* = =
(a) = (Rxa*)* = *
= (a).
Since IlxalI IIxlI hall, one has 11L111 lIxIt. Conversely, since the representation
is nondegenerate, x with respect to the strong topology, where is
an approximate unit of A. Indeed, for any a E A, the net a is norm-
convergent, whence xa with respect to the norm. From the nondegeneracy
we obtain the strong convergence x on the dense set AH, and, due to
the houndedness, lIxM, the strong convergence takes place everywhere
on H. Let > 0 be taken arbitrarily. Choose 11. e H such that IIhM = 1 and
I
Ia'!! !IxhhI +e/2. Since .r = xh and one can find o such that
hlxh — <E/2. Then lIxlI +&. Therefore
= IlxII —
and lxii, since is arbitrary. Hence p is an isornetry.
It remains to verify that Imp = DC(A). Consider an arbitrary element
(L. R) C DC(A). Then the sets are bounded and by the weak
L(eQ) and
compactness of the unit ball 8(H), they have points of accumulation .rL E and
XR e A!!, respectively, with respect to the weak topology. Passing, if necessary, to
sub-nets, we can suppose without loss of generality that
= XR =
 5.2. MULTIPLIERS AND CENTRALIZERS 105

Then ZR = XL. Indeed, for any a,b E A we have

axLb = a = a = aL(b) = R(a)b,
axRb = a = = R(a)b,
whence XL = ZR (cf. the proof of the second part of Proposition 5.2.8). Let us
denote x XL = ZR. Then
L(a) = = xa = Lr(a).
R(a) = w-liinR(ae0) = ax =
in particular, x E M(A). Tile above equalities show that p.(x) = (L, R). 0
EXAMPLE 5.2.12. 1) The equality M(A) = A holds if and only if A is unital.
2) For a commutative algebra A = c0(X) one has
M(c0(X)) = Cb(X) c(Iix).
where Cb(X) is the C*_algebra of all bounded functions with the sup norm and I3X
is the Stone-Cech compactification of X.
3) For the algebra = of compact operators one has = 8(H).
The proof of 1) and 2) can be found, for in [1341 and 3) will be proved
below in a more general situation (see Theorem 5.3.1).
DEFINITION 5.2.13. Let p: A 8(H) be a faithful non(legenerate representa-
tion, so we can assume A c 8(H). An operator x E 8(H) is called a left multiplier
of A if
Xe E A
for ally a E A. Denote by LM(A) the set of all left multipliers. It is clear that they
form a unital algebra. Similarly one can define right multipliers, RM(.4).
An operator x E 8(H) is called a quasi-multiplier of A if
axb E A
for any a, b E A. Denote by QM(A) the set of all quasi-multipliers. It is clear that
they form an involutive linear space.
DEFINITION 5.2.14. A linear map A A A is called a left centralizer if
A(ab) = A(a)b, for each a. b E A.
Similarly one defines a right centralizer. Denote the spaces of left and right Ccii-
tralizers by LC(A) and RC(A) respectively.
DEFINITION 5.2.15. A linear map q: A x A —* A is called a quasi-centralizer if
A0 E LC(A), where A0 : b '—' q(a, b), for any a, b E A,
and
PbERC(A), foranva, bE A.
In other words,
q(ca,bd) = cq(a.b)d, for any a, b. c, d E A.
LEMMA 5.2.16. If p E RC(A), then E LC(A).
106 5. MULTIPLIERS

PROOF. Recall that p' is defined by p*(a) (p(a*))*. Then

p*(ab) = (p((ab)*))* = (p(b*a*))* = (b*p(a*))* = p*(a)b

D
LEMMA 5.2.17 ([104, Lemma 3.12.21). Each right centralizer, each left central-
izer and each quasi-centralizer is bounded.
PROOF. Let p E RC(A). Suppose that it is unbounded. i.e., there exists a
sequence E A such that < 1/n and > n. Then the element
a := is well defined. By Proposition 1.1.5, for each there exists an
element e A such that IIa"611 and = Then the equality
IIp(xn)II = Ilunp(a"3)II
gives a contradiction, hence p is bounded. In a similar way one can prove that any
left centralizer is bounded too. Thus, q E QC(A) is continuous separately in each
variable as a map A x A —i A. By the uniform boundedness principle, such an
operator is jointly continuous (see [35]). D

PROPOSITION 5.2.18 ([104, Prop. 3.12.3]). Let A 13(H) be a nondegenerate

faithful representation. Then there exists a one-to-one isometric linear correspon-
dence between left, right and quasi-multipliers and left, right and quasi-centralizers,
respectively. In the first two cases this correspondence is a homomorphism of alge-
bras. in the third one it is a homomorphism of involutive spaces.
PROOF. The correspondence for the left and right multipliers, together with
its properties, were already described in Theorem 5.2.11. Let q E QC(A) and
let x E A!! be an accumulation point (with respect to the weak topology) of the
bounded directed net {q(ea, eQ)}, where {eQ} is an approximate unit for A. Passing,
if necessary, to a sub-net, we can. as before, assume that x = w-limQ q(eQ, ea). Then
A q(a,b) = = axy

for any a. b e A, so x e QM(A) C A". The required properties can be verified

exactly as the similar properties in 5.2.11. 0
PROPOSITION 5.2.19. Let A be a (closed two-sided *-) ideal of a B.
Then there exists a unique homomorphism B M(A) such that its restriction
to A is the identity map.
PROOF. Put := (L,,, Rb), i.e.. Lb(a) = ba, Rb(a) = ab, where we identify
M(A) = DC(A). Since A C B is an ideal, ba E A and ab E A, so E DC(A).
Thus, obviously : A '—p M(A).
Now assume that in addition to 'y, there exists a homomorphism ö: B —' M(A)
that coincides with 'y on A. Then
ö(b)a = ö(b)ö(a) = ö(ba) = ba, 'y(b)a = 'y(b)y(a) = y(ba) = ba

for any b e B and a E A, i.e., -r(b) and ö(b) coincide as multipliers of A, hence
0
COROLLARY 5.2.20. Let p: A 13(H) be a faithful representation of A and let
A C B be an ideal. Then there exists a representation of B that extends p.
 5.3. MULTIPLIERS OF A-COMPACT OPERATORS 107

PROPOSITION 5.2.21. Let A and B be a

be a morphism
a morphism M(A)/A —, M(B)/B, which makes the diagram

A M(A) M(A)/A

H1
B M(B) M(B)/B
commutative. If is an isomorphism, then and are isomorphisms too.
PROOF. Let (L,R) e DC(A). Define L,R: B B by
1(b) := ço(L(a)), := p(R(a)). b e B, b =
Let us show that these maps are well defined. Let ec, be an approximate unit of
the A and let b = Then
cp(L(a) — L(a')) = lim — e0L(a')) = lim ço(R(e0)) ço(a — a') 0.

A similar equality holds for right multipliers. Since for b1 = and b2 =

one has
= p(a2) = =
= = 131L(b2),

(1, E DC(A). Define DC(A) —i DC(B) by ço"(L, R) = (1, ñ). Then it is a

:

*-morphism of that extends This map induces a map for quotients.

Indeed, if (x — y) e A, x, y E M(A), then — y) = — y) e B. We have
obtained the desired commutative diagram.
If now is an isomorphism, then = Hence,
is an isomorphism and the inverse map is defined by

(L,R) o o Row).
By the five-lemma, is also an isomorphism. 0
REMARK 5.2.22. The homomorphism coincides with the canonical extension
of to Au restricted onto M(A).

5.3. Multipliers of A-compact operators

THEOREM 5.3.1 ([63]). Let M be a Hubert A-module. Define the map
DC(K(M)), T '—' (T1,T2),

0Tx,y, 0x,Ty.
Then q5 is an isomorphism.
PROOF. First of all, note that
OTx,yZ = Tx(y,z) = Ox,T*yZ = x(T*y,z) = x(y,Tz) = Osy oT(z),
so that T1 and T2 can be defined in an equivalent way (and for all compact operators
simultaneously) by the formulas
T1(k):=Tok, T2(k):=koT,
108 5. MULTIPLIERS

From these equalities we obtain at once that T1 and T'2 are well defined as maps
IC(M) —+ IC(M) (since AC(M) C is a two-sided ideal) and are bounded

by 11Th. Since

T2(k1)k2 = k1Tk2 = k1T1(k2), k1, k2 E

(T1,T2) E DC(AC(M)). Since

(TS)1(k) = TSk = T1(S1k), (TS)2(k) = kTS = S2(T2k),

is a homomorphism of algebras. It also respects the involution:

= = = O.r.yT*, = (T*)2,

= = = = (T*)i.
The map is algebraically injective. Indeed, let T1 = 0 and T2 = 0. Then for
any x E M, 0 = = T.r(Tx,Tx) holds, whence (Tx, Tx)3 0 and
Tx = 0. Hence T = 0.
To prove the surjectivity of 0. we construct an inverse continuous map 0. Let
(T1, 7'2) be an clement of DC(IC(M)) and let x E M. Consider the limits
(5.2) T(x) := limT,1(x), := + 1/ri]',
(5.3) T*(x) = := [T2(O.t.x)]*(x)[(x,x) + 1/n]'.

Let us prove that these limits exist. By Theorem 5.2.11, (T1, T'2) = (LF, Rp), where
F E M(IC(M)). Then
(T1(k))*Ti(k) = (Fk)*Fk = k*F*Fk < IIF!12k*k = lIT1 l12k*k,
T2(k)(T2(k))* = kF(kF)* = kFF*k* llFlI2kk* =

where the inequalities are the operator inequalities in ) Then

(T,1(x) — T71(.r) —

= {[(x,x) + 1/rij' — [(x,x) +

{[(.r, x) + 1/n.j_i — [(.r, x) + 1/rnj_1}
< {[(x,.r) + 1/ni_i — [(x.x) + 1/fliil}((Tl(ozX))*TI(OXX)(x),x)
x {[(x,x) + 1/n]' — [(x,x) + 1/m]'}
{[(x,x) + 1/n]' — [(x,x) +
x {[(x. x) + 1/nj' — [(x, x) + 1/m]1}
= l1T1I12{[(x,x) + 1/nj' — [(x,x) +
x {[(x,x) + 1/n]' — [(x,x) + 1/rn]1}
= ItT1 ll2(x, x)3{[(x, x) + 1/n]' — [(x. x) + }2

Thus, the Cauchy criterion of convergence for (5.2) is the same as for the limit from
Lemma 1.3.9. Therefore the convergence is proved. The convergence of (5.3) can
he proved similarly. We obtained the maps T and T* defined everywhere on M.
 5.3. MULTIPLIERS OF A-COMPACT OPERATORS 109

Also, by Lenuna 1.3.9,

(x,T*y) = [(y,y) + 1/ri]')
= lim + [(y,y) + 1/n]')
= lirn + 1/nr',y. [(y,y) + 1/n]')
= Em + 1/n]1,09.9(y) [(y,y) + = (Tx,y).
Hence, by Lemma 2.1.1, T, T* E It remains to verify that =
(T1,T2). Put
/ +
iV'
—)
and note that, by Lemma 1.3.9,
(5.4) = x(y, z) lim x(x, (y, z) = lim
Therefore
= =
= = T(x)(y,z) =
A similar argument for 7'2 completes the proof. 0
It is easy to obtain the following extension of this theorem.
THEOREM 5.3.2 ([73, Theorem 1.5]). There exists an isometric isomorphism
of Banach algebras
0: EndA(M) —+ LM(K(M))
that extends the homomorphism 0 from Theorem 5.3.1.
PROOF. As usual, unitalizing, if necessary, we can assume that the
A is already unital. As before, define 0 by the formula
0(T)(k) = Tk, k e JC(M),
so that it extends 0 from Theorem 5.3.1. Then the calculations presented in the
proof of 5.3.1 for T1 show that 0 is an algebraically injective homomorphism of
algebras and 1. To prove that it is surjective, define the continuous inverse
map for 0 similarly to 5.3.1 by
:= Em :=

S E x E M.
By the same argument as in Theorem 5.3.1, the above limit exists. It defines an
A-linear map. The boundedness of the operator : M —+ M and the continuity
of can be verified as follows. For any x e M, < 1, we have (see [100, 3.11]
and Proposition 3.4.1)x = n(x,x)'/2, whereu E (M#)', and one has u(x,x)0 EM
for any /3 > 0. Put
/2 EM, <a <
y := u(x, x)3 EM, z:

Then
(x,y) = (x,x)3G, (z,z) =
110 5. MULTIPLIERS

and
z(z,z)(y,v) = =
Foranyn=1,2,...,onehas
= [(x,x) + 1/n]1 [(x,x) +
= [(x,x) + 1/n]1 [(x,x) + 1/n]'
= [(x,x) + 1/n]' [(x,x) + 1/n]'
+ 1/n]' (y,x)*(z, z)(y,x) [(x,x) +
= IIS(Oz,z)112[(x,x) + 1/n]1 [(x,x) +
whence, as n —i oo, we obtain that
1/4 < a < 1/2.
Passing here to the limit as a 1/2, we get the estimate

11S112(x,x).

Thus, IISII and, by 2.1.4, is an A-homomorphism. Hence E

EIIdA(M) and 1.
Let us show that = IdLM. For this purpose it is sufficient to verify that
= ib(S) 0 °x,y Using (5.4) and the notation from it, we obtain
=
= lim = =

Since the norm of and does not exceed 1, is an isometry. 0

5.4. Quasi-multipliers of A-compact operators

In this section we present a modified proof of Theorem 1.6 of [73]. Note that this
theorem and similar statements about left and double multipliers can he deduced
from some general results about multipliers (see [103]).
THEOREM 5.4.1. Let M be a Hubert A-module. Then the map 0 from Theo-
rem 5.3.2 can be extended up to an isometric involutive isomor'phism

EndA(M,M') QM(K(M)).
PROOF. The formula
:
Ox',y.(T(x)(y')), X, 7,1, X', E M, T E EndA(M,M'),
obviously is bilinear, thus, it defines a bilinear map on a dense subset in K(M) x
JC(M) with values in K(M).
Let us estimate the norm of Let x = u(x, x)'12 be the polar decomposition
of x in (M#)'. Let := n(x, x)e, where 0 < e < 1/2. Recall that the structure
of a right module on M' is defined by = a E A, M' and
 5.4. QUASI-MULTIPLIERS OF A-COMPACT OPERATORS 111

y E M. By 2.1.4, we obtain that

II0x',y(T(x)(y')) (z) 12 = (z, y) (T(x)(y'))(x', x') (y, z) II
I I *
= ll(z,y)(x,x)2 I
)](x ,x )] (x,x)z
II (x', 112 (x, x) 4_E(y, z) 112
[T
= (x', x') )1* (x' x') 1/211 (x, x) (y, z) 112

112
. 1
(x', x') 1/2 (x' x') 1/2,1 .
(x, x) —E (y, z) (z, y) (x, x)

II (x', xi') 1/2(1/ 1/> 1/2112 . II

(x, x) —E y)112 II lIz 112.

Passing to the limit E —i 0, we obtain IIWE 1 and

II0x',y(T(x)(y')) (z) II 11Th . II (x', xi') 1/2(1/ 1/) 1/211 II (x, x) 1/211 hizIl.

Therefore Thus, cb(T) AC(M)xK(M)

:

ftC(M). Since
cb(T) 0(T) (Ok'x',y', Ox.ky)
)
— —

we have 0(T) E QC (K(M)). Moreover, the previous calculation shows that is

continuous: 11011 < 1.
Let us show the algebraic injectivity of i.e., that Ker = 0. Let T 0. This
means that there exists a vector x E M such that T(x) is a nonzero functional,
T(x)(y') 0 for some (nonzero) y' E M. Then
(T(x)(yI))*T(x)(ij) IIT(x)h12 (y', (y', 1/)
Therefore
(1/ . (T(x) (1/) )* (T(x) (1/)) . (T(x) (1/) )* (T(x) (y')))
= (1/,
}3
IIT(x)1h2{ (T(x)(yI))*T(x)(yI) 0

and
.
Ox,y'.(T(x)(y'))
. (T(x)(yI))*(T(x)(yI))I)
= (Oy'.(T(x)(y')) (T(x)(y')).y'(T(x)(y')) (T(x)(y')) [1/ . (T(i) (ye) (T(x) (1/))],

= (y'. (1/. 1/ .
1/. (T(x)(yI))*(T(x)(1/)) . (1/.
(T(x) (T(x) 1/ . (T(x) (y') ) * (T(x) 0.

Thus, 0(T) 0 and the algebraic injectivity of 4' is proved.

To prove that the map 4 is surjective and isometric, it is sufficient to define, as
in the previous theorem, a map
i,1': QC(K(M)) —+ End(M,M'), = S, <1.
For any S E QC(K(M)) and any k E K(M), the map S(k..) : K(M) K(M)
is a left centralizer, or, in terms of multipliers, for any S and
112 5. MULTIPLIERS

any k e the element kS E (K(M))!! is a left multiplier. Then the map

1/' : LM(K(M)) End(M) from the previous theorem is applicable to it. To
distinguish between different maps, we denote the maps constructed in the previous
theorem by and u". To define II', put. for each x, y E M.
:=

(5.6) :=
+
We have to verify the following properties:
1: existence of the limit in (5.6);
2: the linearity (over A and C) of this expression in x and y;
3: the estimate lxii iiyii;
1

4: the identity = S.
Let (y,y)1/'2, u E (M#)'. Put
3' i 1 1

Then y = . and y = hence z1 E M and z2 E M, since

1 — 3a < 1/4 < 1/2 an(l < 1/2. For n = 1,2,..., we have (similarly to the proof
of Theorem 5.3.2)
-1 -1
(r), y> [(u, y) = .Z20Z2,Z2 S) (x), y) y) +
+
S)(x), y) y) +

= y) y)

= Z2) (Zi, y> y)

+
= [y.y +
One can deduce three corollaries from this equality. At first, for any y with 1

one has
y), Tr,(X, y))
-1 -1
y) y)3a(Z2 Z2)(y, [(n, y) +
+

1i211S11211x112(y, y)
+!]
-2
iiSii2iixii2(y,

iiSii2iixii2(y, y)2 y) +!] (iiSII iixM)2

 5.4. QUASI-MULTIPLIERS OF A-COMPACT OPERATORS 113

and statement 3 is proved. Secondly, by fixing some a, we have

— Tm(X,y),Tn(x,y) — Tm(X,y))
+

+
-
since 4a> 1. So statement 1 is proved. Also,
-2
y). y)) y)
+
IISII2IIxII2(y, 11S11211x112(y, y) —f

and this proves 2 by 2.1.4, as the linearity in x is clear.

To prove 4 it is sufficient to verify, for elementary compact operators, that

SE QC(K(M)), X, y, X', y' E M.

Let

x = y' = u' (y',y')"2, v := u. (x,x)"6 E .M,

:= it'S (y',y')1"6 EM, u, u' E (JVf#)'

Then for w zE M, we have

(5.7) V. (x,z) =

while = = (x,x)2/3. Therefore, if T E LM(ftC(M)),

then for w' := u• (x,x)h/12

=
+
= [(xix)
+ —

-1
= (x,x)213 +
-1
= [(Xix)
+
= (x,x)+F1 =

Similarly to (5.7), we obtain that

= WI' := v'(y',y')'13 E M.
114 5. MULTIPLIERS

Then, after putting := v'(y', E M, we have

y') [(y', y') + 1/n]'
= y') [(y', y') +
= y') [(y', y') + 1/n]1
= lirn(w" (w", y') [(y', y') +

= w") (w", y') [(y', y') + 1/n1'

= [(y',y') + 1/n]'
= [(y',y') +
= S) (x), (y', y') = S) (x),v'),
whence
= liii [(y',y')+1/n]
= °x',y. ,,, S)(x),v') = 6x',y• ((8,,, S8,,,,)(v),v'>

In the last expression we used the presentation of quasi-multipliers in the form of

quasi-centralizers. On the other hand, y' = v'. (v',v'), x = v• (v,v) and
S(Ox',y', =
= Ox',v'9S(O,,, ,,, ,8,,,,)v.y 6x'.y (S(8,,,,,, 8,,,, )v.v').
Statement 4 is proved, and this completes the proof of the theorem. 0
5.5. Strict topology
DEFINITION 5.5.1. Let A 8(H) he a nondegenerate faithful representation
of a A. The strict topology on 13(H) is the topology satisfying one of
the following (obviously, equivalent) conditions:
(i) it is the weakest topology, for which the maps

ra : 13(H) 13(H), ra : x i—+ xa,

Ia : 13(H) —÷ 13(H), Ia : x ax, x E 13(H), a E A,
are continuous;
(ii) it is the topology generated by the system of seminorms

IIxaII, IIaxII.

Usually this topology is denoted by /3 analogously with the Stone-tech compactifi-

cation (cf. 5.2.12). For example, we denote by the space of the maximal ideals
for the closure of the algebra C(X) in 13(H) with respect to the strict topology,
and we denote by /3 -lim the limit with respect to the strict topology.
PRoPosITIoN 5.5.2. The set M(A) is strictly closed,
[Al0 c [M(A)]0 = M(A).
 5.5. STRICT TOPOLOGY 115

PROOF. Let the net {XQ}aEA c M(A) be strictly convergent to x E 13(H).

Then for any a E A there exist the norm-limits
L(a) := R(a) :=
defining the maps L, R: A A. Since

aL(b) = a = b = R(a)b,
the pair (L, R) is an element of DC(A), i.e., a double centralizer. Identifying double
centralizers with multipliers, by Theorem 5.2.11 we obtain that y E M(A). Then
13
y. Indeed,
ya — xaa = L(a) — xaa —p 0, ay — axa = R(a) — axa 0

with respect to the norm topology. So, M(A) is fl-closed. 0

LEMMA 5.5.3. The net is an approximate unit for A if and only if
M(A) 1 = fl-lima
PROOF. The result follows immediately from the definition. 0
PROPOSITION 5.5.4. (i) The conjugation in M(A) is fl-continuous.
(ii) Multiplication in M(A) by a fixed element is fl-continuous.
(iii) Multiplication in M(A) is jointly fl-continuous on bounded sets.

PROOF. Let x E M(A) and let x. Then

xa, axa ax for any a E A,

whence, after conjugation,
foranybEA (b=a*),
i.e., —f-'

Now let y e M(A) be a fixed element and xe,, —i x. Then 13

I(xay)a — (xy)aII — x(ya)Il —p

— a(xy)Il — axII . 0

for any a E A. This means that xy.

13 (3
Now let XQ x, for any a E A and
IIXaD —' y, for any
'y E r. Then, for any a E A and for any > 0, there exists a pair (ao, 'yo) such that
for any pair (cl,-y) > e A x r (i.e., for a > and > 'ye) one has
II(xaYy)a — (xy)aIl — + — xyall
— yalJ + — x(ya)II <E,
— a(xy)II — + — axyll
— axII + — (ax)yII
This means that (xy) = 0
THEOREM 5.5.5. The algebra of multipliers M(A) coincides with the fl-closure
ofA inB(H),
M(A) = [A]13.
116 5. MULTIPLIERS

PROOF. By Proposition 5.5.2, it is sufficient to prove that M(A) C Let

be an approxiniate unit for A. By Lemma 5.5.3, 1 E M(A).
Let x E M(A) be an arbitrary element. Consider the net Since x is a
multiplier, the elements of this net belong to A and, by Proposition 5.5.4, the
net is 8-convergent. Hence, y = e But y = x. Indeed, for any
a E A, one has
ay = a = = ax,
yb = = = xb,
so x is the same multiplier as y. D
DEFINITION 5.5.6. Let A '—* 8(H) be a nondegenerate faithful representation
of a A. The left strict topology on 13(H) is the topology satisfying one
of the following equivalent conditions:
(i) it is the weakest topology for which the maps
ra xEB(H), aEA,
are continuous;
(ii) it is the topology generated by the system of seminorms
:= lixall.
For the left strict topology, one can prove the following analog of Theorem 5.5.5.
THEOREM 5.5.7. The algebra of left multipliers LM(A) coincides with the clo-
sure of A in A!! with respect to the left strict topology.
PROOF. This statement can be obtained by the same method as in 5.5.5, it is
sufficient to use only the "left half" of the argument. D
For Hubert modules, it is natural to consider the following two topologies on
the space of bounded homomorphislns.
DEFINITION 5.5.8. Let M be a Hilbert A-module. The strong module topology
on End(M) is the topology generated by the system of seminorins
{sh}hEM sh(x) := IIx(h)II, x E End(M),
and the *-strong module topology on End*(M) is the topology generated by the
system of seminorms
{sh; sh(x) := IIx(h)II, := x E End*(M).
PRoPosITIoN 5.5.9. The strong topology is not weaker than the *-strong module
topology on bounded sets in IC(M) (hence, by Theorems 5.5.5 and 5.3.1, everywhere
on End*(M)).
The left strong topology is not weaker than the strong module topology on IC(M)
(hence, by Theorems 5.5.7 and 5.3.2, everywhere on End(M)).
The corresponding topologies coincide on bounded sets in
PROOF. We will verify the equivalence of the appropriate seminorms. We have
sh(x) = IIx(h)II = IIxk(g)II IIxkII = .

for some k e 1C(M), g e M (see Lemma 2.2.3) and

= jx*(h)II = IIx*k(g)II . ugh = IIk*xhI = (x)
 5.5. STRICT TOPOLOGY 117

for some k e K(M), g e M. Conversely, let k e K(M) be an arbitrary element

and let 0 with respect to the strong module topology. Assume that the net
Xa is bounded: < c. Then for any E > 0 there exist vectors h1,.. . and ,

91,... from M such that

and there is such that

= 1,... ,fl,
<
for any > Then, for these a,

= IIXakII .

c• + s+n <2e.

Similarly, if 0 with respect to the *-strong module topology, then we can

additionally require that
= 1,. ..,n,
<
for > Then

= IIxaII + >

c. E+ =
+ MgiII

PROPOSITION 5.5.10. Let A and B be

a morphism such that B C Then is strictly continuous.

PROOF. Let x in M(A) and let b be an arbitrary element of B. Then

b= for some a E A. The nets and axa are norm-convergent in A.
The map being a homomorphism of does not increase the norm.
Therefore = and = are Cauchy nets, so they are
norm-convergent. Since b was arbitrary, this means that -L y E M(B).
Thus, for any b' e B, b' = one has

yb' = = = = i1(xa') =
D

REMARK 5.5.11. In particular, the map from Proposition 5.2.21 is the ex-
tension by continuity of the morphism thereby, is unique.

A similar theory can be developed for quasi-multipliers, if one uses the following
definition.
118 5. MULTIPLIERS

DEFINITION 5.5.12. Let A 13(H) be a nondegenerate faithful representation

of a A. The quasi-strict topology on 8(H) is the topology satisfying one
of the following equivalent conditions:
(i) it is the weakest topology for which the maps
xEB(H),a,bEA,
are continuous;
(ii) it is the topology generated by the system of seminorms

{Vab}a.bEA, l/ab(X) := IaxbII.

5.6. Multipliers and Hubert modules. The commutative case

The following results describing the modules 12(C0(X, A)) and spaces of oper-
ators on them in terms of spaces of maps are obtained by combination and small
modification of [41, 1].
DEFINITION 5.6.1. Denote by C0(X,M) (resp. by A0(X)) the space of contin-
uous maps X M (rcsp. X A) vanishing at infinity.
Note that Ao(X) is a with respect to the sup-norm and C0(X, M)
is a Hilbert Ao(X)-module with the inner product given by (f,g) =
where f,g E C0(X,M), XE X.
DEFINITION 5.6.2. Let X be a locally compact Hausdorif topological space
and let M be a Hilbert A-module. Let us call the pair (X, M) compatible if the
following conditions are satisfied:
(i) the
j : M ®c c0(X) —, c0(X,M), j(m® f)(x) := f(x)m, m EM, I E c0(X),
is an isometric isomorphism;
(ii) let E C0(X,M) be such that p(x0) = 0 for some x0 E X and let
F E EndA0(x)(Co(X,M)) be an arbitrary operator; then (Fp)(xo) = 0.
REMARK 5.6.3. Here and further, by a tensor product we always mean the
projective tensor product. In the case of C*_algebras it is also called spatial or
minimal. However in (i) (e.g. if M = A) we use the tensor product with a commu-
tative algebra, which is nuclear, so all on this tensor product coincide.
For more details see [94, §6.3].
REMARK 5.6.4. Generalizing [41], we shall prove in the following two lemmas
that the pair (X, 12(A)) is compatible (in [41] the case of compact X is considered).
Note that a weaker compatibility condition for an arbitrary pair (namely, if in (ii) we
replace End by End*) follows from [1]. Thus, the second part (the one concerning
adjointable operators) of Theorem 5.6.8 [41] below can be deduced from [1] using
identification of (M) with multipliers for
LEMMA 5.6.5. The map
j : 12(A0(X)) C0(X, 12(A)), j(f)(x) := (fi(x), 12(X),...),
I = (11,12,...) E 12(A0(X)),
is an isometry.
 5.6. COMMUTATIVE CASE 119

PROOF. It is clear that j is an isometric inclusion. Let us show that it is

surjective. Let p E C0(X, 12(A)) be an arbitrary element. Then, since
— < —

we have A0(X). It is necessary to verify that the series

is norm-convergent. Take an arbitrary E > 0 and let K C X be a compact set such
that the inequality <E holds for any yE Y := X\K. For each point x E K
choose a number n(x) such that

<
i—n(x)
Since the map

is continuous, for each x E K one can find an open neighborhood in K such

that
<E
7=n(x)
for each z Due to the compactness of K one can choose a finite subcovering
Take ii := m > n,

sup
xEX

<max , sup
}

maxIsup
I xEY 3=1

<max *e} = E.

By the Cauchy criterion the series is convergent. U

LEMMA 5.6.6. Let E c0(X,12(A)) and let FE EndA0(x)(Go(X,12(A))) be an

arbitrary operator. Suppose that co(xo) = 0 for some x0 E X. Then (Fço)(xo) = 0.
PROOF. By 2.1.4, one has (Fço,Fço) 11F112(co,co), so
(Fp(xo),Fp(xo)) = (Fp,Fp)(xo) < = = 0.
U

DEFINITION 5.6.7. Denote by 8(X, EndA(M)) (resp. B*(X, the

algebra of bounded continuous maps from X to End,, (M) = LM(K(M)) equipped
with the left strict topology (resp. the algebra of bounded continuous maps from
X to = M(IC(M)) equipped with the strict topology). By Proposi-
tion 5.5.9, one can consider the strong module (resp., *-strong module) topol-
ogy instead of the left strict (resp., strict) topology. We equip these algebras,
B(X,End,,(M)) and with the sup-norm, making the first one a
Banach algebra and the second one a
120 5. MULTIPLIERS

THEOREM 5.6.8. Let (X,M) be a compatible pair. Then

(i) the Banach algebras EndA0(X)(M ® co(X)) and 8(X, EfldA(M))) are
naturally isomorphic;
(ii) the 0 co(X)) and are nat-
urally isomorphic.
PROOF. By condition (i) of Definition 5.6.2, we can identify M ®c co(X) =
C'0(X, M) and define the map
J : 13(X,EndA(M))) EndA0(x)(Co(X,M)),
:= D(x)(ço(x)).
XE X, E c0(X.M).
First of all, let us show that .J(D)y e c0(X, M). For this purpose we should verify
that it is continuous and vanishes at infinity. Since II D(x) is bounded by some
constant C, we can choose a compact set K C X such that < r/C outside
K. We obtain that
II(J(D)p)(x)II = <C = outside K.
Vanishing at infinity is verified. To verify continuity, we choose an arbitrary x E X
and e > 0. Let us find an open neighborhood V1 of the point x in X such that the
estimate II
<nC holds for any y E V1. By the definition of continuity
with respect to the strong module topology, for (a fixed element) E M, there
exists an open neighborhood V2 of the point a' in X such that the inequality
IID(x)(p(x)) — <E
holds for any y E V2. Then, for any y E U := V1 fl V2, one has
— (J(D)p)(y)IIM = IID(x)(p(x)) —
— D(y)(p(X))IIM + IID(y)(p(x)) —
<n + C• = 2n,

so the continuity is proved.

The linearity over C arid A() (X) of the operatOr .1(D) is clear. Since
= sup
sEX
sup IID(x)IlEndM sup <C
sEX sEX
the operator J(D) is bounded. Thereby the map J is well defined. It is clear that
it is C-linear and
(J(DC)ço)(x) = = D(x)[C(x)(cc(x))I
= = (J(D)J(C)cp)(x),
so that J is a homomorphism of algebras. Let us check the (algebraic) injectivity
of J. Suppose that the relation = 0, i.e., = 0, holds for any
co(X, M) and any a' X. Note that passing to the one-point compactification
X+ of the space X, we can define a continuous function x5° : [0, 1] on the
normal space X+ that equals 1 at a'0 and vanishes at oc. Then, for any m M, the
function := nrfo(x) tends to zero at infinity. Since 0 = =
D(xo)rn and since m and a'0 are arbitrary, we have D(xo) = 0 for any x0, i.e.,
D =0.
 5.6. COMMUTATIVE CASE 121

Note that. the estimate above, which proves the continuity of J(D), also gives
the inequality lull < 1. Thus, the first part of the theorem will be proved if we can
manage to define an inverse linear map
S EndA0(x) C0(X,M) 8(X,EndA(M)), IISII 1.

Put
(5.8) :=
where is a (nonuniquely defined) map e C10 (X, M), satisfying the condition
= in. Let us verify the independence of the definition from this nonunique
choice of Let = 0. Then, by property (ii), one has = 0 for any
operator T and we have proved that (5.8) is well defined. Once more, the linearity
of S(T) (x) is clear, so we should verify its boundedness. We have
ll(S(T)(x))(in)ll =
llTllErid(c0(x..M)) II IlC0(X,J4) llTllErid(c0(x,.k.l))
This estimat.e shows that S(T) (x) is bounded, that the bound does not depend on
x, and that 1. To complete the proof of the first part of the theorem, the
continuity of S(T) (x) in x with respect to the strong module topology should be
verified. Let x E X be an arbitrary point, E > 0 an arbitrary number and m E M
an arbitrary (fixed) element. One can find an open neighborhood U of the point x
such that the inequality
—

holds for any y E U and for some map of the form Then one has

llS(T) (x)rn — S(T) (y)inll = ) (x) — (y) <

for any y e U. The first statement is proved.
Now let J' = Then, for any D E one has
= =
= (cp(x).
= ) (x)) = ) (v')) (x),
so that Im,J' C c0(X,M).
To complete the proof of the second part, we have to verify, firstly, that
C

i.e., that the operator S(T)(x) is adjointable for each x E X, and, secondly, that
S(T)(x) is continuous in the *-strong module topology.
The first assertion follows from the calculation
(S(T)(x)m, m') = (x)) =
= = (m,S(T*)(x)ml),
rn, m' EM.
Moreover, it implies that S is involutive. Now let x E X be an arbitrary point,
E> 0 an arbitrary number and m E M an arbitrary (fixed) element. One can find
an open neighborhood U of the point x such that
— <E
122 5. MULTIPLIERS

holds for any y E U and for some map of the form Then one has
IS(T)*(x)m — S(T)*(y)mII = — <E
foranyyEU. 0
Denote by 13(X, GL*(A)) C GL*(A)) the set of functions with bounded
(pointwise) inverse.
COROLLARY 5.6.9. The homomorphism J defined above gives an isometric iso-
morphism of the groups
GL(A0(X)) 8.(X,GL(A)), GL*(Ao(X)) 8(X,GL*(A)).
PROOF. Since J is a homomorphism of algebras, the statement immediately
follows from its unitality. 0
Note that one has to be cautious when identifying different classes of operators
in the standard Hilbert module over a commutative with continuous
sets of operators of the same class on a Hilbert space. For example, although an
operator of finite rank on the Hilbert module Hc(x) defines a continuous set of
operators of finite rank on a compact space X, the inverse statement is not true,
as can be seen from the following example obtained by D. Kucerovsky [67]. It is
interesting that this example is of topological origin.
Denote by the standard tautological vector bundle over the complex projec-
tive space CP(n). For information about vector bundles and their characteristic
classes we refer the reader, for example, to the books [49, 60]. Let F(ert) be the
Hilbert C(CP(n))-rnodule of sections of the bundle
LEMMA 5.6.10 ([67]). Let K be a compact operator with an algebraically n-
generated image on the Hilbert C(CP(n))-module r(en). Then there exists a point
x0 e CP(n) such that K(xo) = 0, where K(x) E C(CP(n),K.) is the ftC-valued
function defined by the operator K.
PROOF. The operator K has the form sk(rk,-), where 8k, rk are contin-
uous sections of the bundle Let E = . be the vector bundle equal to
the direct sum of ii copies of the bundle Then si Sn is a section of the
bundle E. Let us calculate the higher order Chern class of the bundle E. We have
Cn(E)
because 0. The nonvanishing of cn(E) means that any section of the
bundle E vanishes at some point. In particular, the section s1 ED.. . ED vanishes at
some point x0 E CP(n), hence all the sections i= 1, . . . , n, vanish at the point
x0 too. 0
EXAMPLE 5.6.11 ([67]). Let

X =JJCP(n)
be the disjoint union of complex projective spaces and let X+ be the one-point corn-
pactification of X. Define a Hilbert fl as a direct sum of spaces
of sections of the bundles fl = The rncdule fl is countably gen-
erated, so it can be realized as an orthogonally complemented submodule of the
 5.6. COMMUTATIVE CASE 123

standard Hubert module Hc(x+). Define the compact operator K on the module
fl by the formula

where s = E fl. The operator K defines a continuous family of rank one

operators over the space X+. However, since the set K(x) does not vanish at any
point of X, by Lemma 5.6.10 the operator K is not an operator of a finite rank
on the module fl. At the same time, since the set K(x) is continuous, K E K(fl).
Extending the operator K by zero, one can obtain a compact operator on the
module Hc(x+) possessing the same property.
Now we consider nonadjointable operators in the commutative case [54]. Let
A = C(X), where X is a compact Hausdorif space. We will use the shortcut ICA
for the space IC(HA) of A-compact operators on the Hilbert module HA = 12(A).
We also will use the characterization of compact operators from Proposition 2.2.1.
As before, let be the module generated by the first n vectors of the standard
basis and let be the orthogonal projection onto that module (for A = C we
use the notation pg). The space of all operators F: 12(A) —' 12(A) satisfying the
conditions of Definition 2.7.4, except, possibly, the condition to be adjointable, will
be denoted by TA.
THEOREM 5.6.12. An operator K E End HA is compact if and only if, for the
map S introduced in the proof of Theorem 5.6.8, S(K) maps X to K and S(K)
X /C is norm continuous.
PROOF. Let us prove that 8(K) is norm continuous if K E ftCA. Choose an
arbitrary point x E X and a number c > 0. There exists a number n E N
such that IIK(1 — Pn)II < e/4. Note that S(K) is strongly continuous. Hence,
= is norm continuous. Thus, we can choose a neighborhood
x such that —S(Kp71)(x)II <e/2 for each y E Then, for these
y, the following inequality holds:
IS(K)(y) — S(K)(x)II IIS(K(1 —
+ pn))(X)II + —
E
++
E
= E.

Now let us show that S(K)(x) E ftC for each x E X. For an arbitrary e > 0
choose a number n such that IIK(1 — Pn)II <r. Then, for any x E X, one has
IIS(K)(x)(1 — = IIS(K(1 — pn))(X)II IK(1 — <
Hence, S(K)(x) is a compact operator.
Now let a map D : X ftC be norm continuous. Let us prove that J(D) E /CA,
i.e., IIJ(D)(1 — pTh.)fl —' 0 as ii oc. Choose c > 0 and for each x e X find a
number such thit
<E/2 for
and an open neighborhood x such that the inequality
ID(y) — D(x)II <e/2
124 5. MULTIPLIERS

holds for any y E Let . , be a finite sub-covering of and m :=

.. , }. Then, for each y E X, we can choose x2 such that
IIJ(D)(1 — pm)II sup IID(y)(1 —
yEX
— + II(D(y) — D(x1))(1 —
yEX
< + e/2) = E.
yEX
D
THEOREM 5.6.13 ([54]). For each F E .FA, there exists an invertible operator
G E GL such that the map S(GF) : X is norm continuous.
PROOF. One can a.ssume that F has the following form:
F1
F 12(A) 12(A) =
= ( F ) :

where F1 : iVI1 M2 and N1, N2 = are finitely generated projective modules.

The proof of this fact is similar to the case of adjointable operators (Section 2.7).
Since N1 is a finitely generated projective module, N1 Nj'- = 12(A) (by Lemma
1.4.3). We can consider N1 as a locally trivial vector bundle over X and Nj'- as a
locally trivial bundle over X with a Hubert space as a fiber. Indeed, let P : 12(A)
N1 be an orthogonal projection. By Lemma 5.6.14 (see below), the map S(P)
X Proj (12) is norm continuous. Hence, for an arbitrary point 10 E X there exists
an open neighborhood such that the inequality <1 holds
for each x E Denote the fiber of the bundle N1 by For x e the
restriction of the map 1 — S(P)(x) to is an isomorphisnl onto and
the map (1 — S(P)) : X Proj(l2) is norm continuous.
By Kuiper's theorem [68], the bundle Nj' is trivial. Hence, there exists an
isomorphism of A-modules T: Nj'- such that the map S(T(1 — P)) : X
Endc 12 is norm continuous.
The operator F has the matrix form
F=
F10
- - i,
-
where F1 : N1 M2 is an lsomorphism
I
\. 4J
3

with respect to the decomposition l2(A) = Nj'- l2(A) = Then

Fi
iJF= ( F ),where
0
: ;

—F3F1 1
,,
Hence, F has a diagonal form,

F ): Nj' N1 I%/12
= (
where
AV12 .— çì .— i—hr \ —
(/1 —
r
Define the operator G E GL by the formula

C:= ( ? )
 5.6. COMMUTATIVE CASE 125

Then

Since F4 = GFP is a compact operator, by Theorem 5.6.12 the operator S(F4) is

norm continuous. Hence, the map S(GF) = S(T(1—P))+S(F4) is norm continuous
too. 0
LEMMA 5.6.14 ([33, Lemma 5, p. 253]). Let H be a Hubert space, n > 0 an
integer and the space of rank n projections in H. The strong topology and the
norm topology coincide on
PROOF. Take Po E and e> 0. Let (fl,... be an orthonormal basis of
po(H). If a projection p E is sufficiently close to P0 with respect to the strong
topology, the norms — are arbitrarily small and i = 1, n, is a basis
. . . ,

of pH. Hence, there exists a neighborhood (with respect to the strong topology) V
of P0 in such that the inequalities
i= 1,...,n,
hold for the orthogonalization . .. , of the basis . Let H be
an arbitrary element with iieii 1. Then, for p E V. one has

— = —

EXAMPLE 5.6.15 ([54]). The following example shows that there exists a strong-
ly continuous map F: X 1c such that J(F)
Let us identify 12 and L2[0, 1]. Let be the set of all functions f E L2[0, 1]
such that 1(x) = 0 for x t. Let us denote by the orthogonal projection
L2 —, il'. Define the following family of isometries [33]:

Ut: L2[0, 11, (Utf)(x) := t E (0,1], f E

Let R E 1c be a Fredhoim operator with index R = 1. Suppose also that RR* = 1
and R* R = 1 — p', where p' is a one-dimensional projection. Define a strongly
continuous map F: [0, 1] 1c by the formula
— + UI1RUtPt fort> 0.
F(t)
fort=0.
Since F*(t) = (1 — + one has
F*(t)F(t) = 1-Pa + = 1+ K1,
F(t)F*(t) = 1
P(t) + P(t) = 1,
—

where K1 is some rank one operator. Hence, F(t) E 1c. We have

indexF(0) = 0 1 indexF(1)
and, by Theorem 5.6.13, we see that J(F) A = C[0, 1].
126 5. MULTIPLIERS

Thus, an arbitrary strongly continuous map F : X '1c does not define an

operator in Nevertheless, one can define a new topology on cIC (independent
of X!) such that the continuity of F: X with respect to this new topology
implies J(F) E cIC(X) and vice versa.
DEFINITION 5.6.16 ([54]). An F-topology is the topology on generated by
the following pre-base:
Ue,ai a,,,D = {R E Fc I II(R — < i = 1,..., n};
= {R E FcI there exists an operator T GL(l2),
such that T(V) C V and IITR — DII <E},
where a1,.. . are some vectors from 12 and V is a finite-dimensional subspace
of 12.
We complete this section with the following two theorems. The proofs can be
found in [54].
THEOREM 5.6.17 ([54]). The set of all F-continuous maps f : X —* coin-
cides with
THEOREM 5.6.18 ([54]). The spaces (tIC, norm) and (cIc, F-topology) are
weakly homotopy equivalent.

5.7. Inner products on Hilbert

We have already encountered the case where several inner products on the
same Hilbert module were used. In this section we will study the relations between
different inner products more thoroughly.
Let M be a module over a A, on which two sesquilinear maps (•)
and are defined in such a way that M is a pre-Hilbert module with respect
to each of these maps.
DEFINITION 5.7.1. Two inner products, (., and (, are called equivalent
if the norms defined by these inner products are equivalent.
Note that if the inner products are equivalent to each other and if {M,, (.,
is a Hilbert C*_module, then {M,, is a Hilbert C*_module too.
Let us consider first the case of different inner products on self-dual Hilbert
modules.
PROPOSITION 5.7.2 ([42]). Let M be a self-dual Hilbert over a
A with the inner product (, If is another inner product
equivalent to the given one, then there exists a unique invertible positive operator
S such that (x,y)1 = (Sx,Sy)2 for all x,y EM.
PROOF. For x E M consider the functional on M defined by the formula
y i—' (x, Since the module M is self-dual, there exists an element Bx E M
such that
(x.y)2 = (Bx,y)1
for all y M. The map x '—i Bx is an A-homomorphism. Denote by the
norm defined by the inner product (., i = 1,2. By our assumption, there exist
constants k,l > 0 such that
I
 5.7. INNER PRODUCTS ON HILBERT C-MODULES 127

for any x E M. Then

12
II = II(x, Bx)211 IIBxII2 lixiL IIBxII1

Therefore IBxIJ1 11x01, i.e., the map B is bounded. The equality (Bx,x)1 =
(x, x)2 0 implies that the operator B is positive with respect to the initial inner
product. The inequality
= k2 (x, x)2 = k2 (Bx, x) < k2 IIBxII 1 lxiii

shows that the estimate iiBxiI1 IixiI1 holds, from which we obtain, by [92] (cf.
the proof of Theorem 2.3.3), invertibility of the operator B. To complete the proof
it remains to put S = B—h1'2. 0
PROPOSITION 5.7.3 Let M be a Hubert module with an inner product
(•, Let be another inner product equivalent to the initial one. Then the
map M given by the formula x (x, )2' x E M, defines an invertible
positive element in QM(K(M)) C Conversely, any invertible positive
element in C (which can be identified with an element T of
the set EndA(M,M)) defines an inner product (x,y) T(x)(y). x,y E M.
PROOF. Let EndA(M, M') —p QM(AC(M)) be the isometric isomorphism
defined in Theorem 5.4.1, = Ox,t.T(z)(y), x,y,z,t e M. The map
x (x, •)2 is bounded, hence it defines a map T M —+ M' and the element
E QM(1C(M)) is defined by the formula

(note that the elementary operators of the form are considered with respect to
the initial inner product Then for s E M one has
= (x (x. (y,y)2,s)1,s)1
.

= (x 0.

Since linear combinations of elementary operators are dense in the algebra

the operator is positive. Let us show that it is invertible. Let us pass to
the Hubert module M# over the enveloping A**. Both inner prod-
ucts can be extended to the module M# and to the self-dual module (M#)'. By
Proposition 5.7.2, there exists an invertible operator S E ((M#)') such
that these extensions of the inner product are related by (x, y)1 = (Sx, Sy)2
for any x, y E (M#)'. But the image of the operator under the inclusion
QM(K(M)) c ((M#)') obviously coincides with the product S*S. Since
the operator S is invertible, the spectrum of the operator ç)(T) is separated from
zero, hence 4(T) is invertible. In the opposite direction the statement can be proved
similarly. 0
COROLLARY 5.7.4 ([44]). Let M be a Hilbert with an inner product
(., The following conditions are equivalent:
(i) any other inner product equivalent to the initial one is defined by an
invertible operator S E EndA(M) and is given by the formula (x, y)2 =
(Sx,Sy)1, x,y EM;
128 5. MULTIPLIERS

(ii) each positive invertible quasi-multiplier T E QM(ftC(M)) can be de-

composed into a product T = S*S for some invertible left multiplier
Se
THEOREM 5.7.5 ([44], see also [17]). Let M be a countably generated Hubert
with an inner product (., Then for any inner product (•, equiva-
lent to the initial one, there exists an invertible operator S e EndA(M) such that
(x,y)2 = (Sx,Sy)1.
PROOF. By assumption, the K(M) is a-unital, hence it contains
a strictly positive element H E K(M). It is sufficient to show that each positive
invertible quasi-multiplier admits a decomposition T = S* S with some left mul-
tiplier S. Put K = (HTH)1/2 E ,kC(M), = K (H2 + H e ,kC(M). Then
IVnII and the sequence converges to K with respect to the norm.
Then, for any K' E H . ftC(M), the sequence is norm-convergent to KK'.
Since H is dense in K(M), we conclude that the sequence (Va) is conver-
gent to some element S e LM(K(M)) with respect to the left strict topology and
SF! = K. Therefore HS*SH K*K = HTH and, finally, = T. D

As we can see from the following example of a nontrivial inner product, the re-
quirement for Hilbert to be countably generated is essential in Theorem
5.7.5.

EXAMPLE 5.7.6 ([17, 44]). Let H be a nonseparable Hilbert space. Consider

the space

equipped with the weak topology and the standard Hilbert C(X)-module Hc(x).
Let us show that there exist inner products on Hc(x) equivalent to the standard
one, which cannot be written as (S., 5.) for any operator S E Endc(x)(Hc(x)).
For this purpose it is sufficient to find a quasi-multiplier T E QM(AC ® C(X))
that cannot be represented as T = S*S, S E LM(K ® C(X)). Let us use the
identification of LM(ftC ® C(X)) (resp. QM(K ® C(X))) with the set of bounded
maps from X to 5(H) continuous in the strong (resp. weak) topology (it is discussed
in detail in Section 5.6). Define the new inner product on the module Hc(x) by
the formula
(5.9) (y, = (y, x(x))(x),
where y,z E Hc(x), x E X. it is easy to see that (y,y)0 (y,y).
This inner product defines a positive invertible quasi-multiplier T. Suppose that
T = S*S for some S E ® C(X)). Let us show that one can choose a
separable infinite-dimensional Hilbert space HT C H such that T(X)HT C HT
and T1(X)HT C HT for all x e X. Let {ei,... ek, .} be a basis of some
, . .

separable subspace H0 C H. Since X is compact, the sets T(X)ek and T'(X)ek

are compact subsets in H for each number k. Therefore they generate a separable
Hilbert subspace H1 C H such that T(X)H0 C T'(X)Ho C Hi. Further,
by induction we find separable subspaces H such that C
C Finally, set HT := Ha), i.e., the closure of the union of
all
 5.7. INNER PRODUCTS ON HILBERT C-MODULES 129

Let us denote by Xo C X the subset

3/4 r
1/4, ris linear
The restriction of the operator S onto the subspace X0 has the form
S1
0 53)' 1/ 23/
82 * IA * * *
s1s2=r,

with respect to the decomposition H = HT Since the subspace is

invariant under the action of the operator s1 e B(HT) is invertible and the
operator is unitary. Since the map u is continuous in the strong
topology on the group of unitary elements, we conclude that is continuous on
X0. Therefore the map r = is also strongly continuous as a map from X0
to Thus, the assumption that T = S*S implies that any weakly
continuous bounded (by 1/4) linear map r : HT —4 turns out to be strongly
continuous. But, since the strong and the weak topologies do not coincide on the
ball of radius 1/4 in B(H#, HT), the obtained contradiction shows that the inner
product (5.9) is not related to the standard inner product on the module Hc(x) by
any invertible bounded operator S e EndA(Hc(x)).
If one considers various equivalent inner products on a Hilbert
the problem of whether an operator is adjointable depends on the concrete inner
product. By (resp. we denote the of adjointable
operators (resp., compact operators) with respect to the inner product i=
1, 2. The adjoint operator for the operator T with respect to this inner product is
denoted by T(j).
PRoPosITIoN 5.7.7 Let M be a Hubert A-mod'ule over a A
with the inner product (., Let S e EndA(M) be an invertible operator defining
the inner product = (S., S.)1. Then the operator S is adjointable with respect
to the first inner product if and only if it is adjointable with respect to the second
one.
*(1) *(2)
If S is then the sets EndA and EndA coincide and the same is
true for IC(1)(M) and ic(2)(M).
PROOF. Let the operator S be adjointable with respect to the inner product
Then one has
(Sx,y)2 = (S2x,Sy)1 = =
for any x, y e M, and the operator S with respect
to the second inner product. The converse statement can be proved similarly.
Let us assume now that S e Let B e Then
(Bx, y)2 = (SBx, Sy)1 = (Sx,
= (x,
for any x,y E M, hence BE The
statement about compact operators can be proved in the same way. 0
However, if the inner product is defined with the help of a nonadjointable
operator S, then operators adjointable with respect to one of the equivalent inner
130 5. MULTIPLIERS

products need not be adjointable with respect to another. Thus a problem of

whether a functional on M can be represented as an inner product by elements
from M arises. More precisely, we define the set F C M' by
F= U I3EB; yEM
where B is the set of all inner products (•, equivalent to the initial one. The
functional f E M' is called representable if f e F. We study the set F for the
standard Hilbert HA. Denote the extension of the initial inner product
from the module HA to = HA.. and to its adjoint module still by (., •).
It is easy to see that C
PROPOSITION 5.7.8. If I E is representable, then there exists an element z
in the module HA, for which the operator inequality
(5.10) a(z,z) </3(f,f) < (f,z) <8(z,z)
holds with some positive constants fi, -y, 8.
PROOF. By Theorem 5.7.5, any inner product equivalent to the given one has
the form (x, = (Sx, Sy), where S E EndA(HA) is an invertible bounded op-
erator. If f is representable, then f = S*Sz for some S and some z E HA, the
operator z) A is positive, and we have
(f, z) = (S*Sz, z) = (Sz, Sz) = (z,
Since S is invertible, there exist positive numbeis a and b such that
a(z,z) < (z,z)13 <b(z,z),
hence
(5.11) a(z,z) < (f,z) <b(z,z).
Let us now estimate (f, f) = S*Sz). Since a2 < (S*S)2 b2, we have
(5.12) a2(z,z)
< (f,f) < b2(z,z).
Combining (5.11) and (5.12), we obtain the estimate (5.10). U

A functional f is called nonsingular if there exists an element z in HA

such that the spectrum of the element (f, z) E A is separated from zero (then one
can assume that the operator inequality (f, z) c> 0 holds for some number c).
The following example shows that there exist singular functionals with the property
(f,f)=1.
EXAMPLE 5.7.9. Let A = L°° ([0; 1]). Let us define f as a sequence of
functions f = (fk(t)),
for E
[l/2k; 1/2k—1]
fk(t) = t

10 for remaining t.

The property 1 is obvious. Let us show that the spectrum of the oper-

ator (f, z) is not separated from zero for any z = (zk) E HA. Since the series
4zk is norm-convergent, for any e > 0 there exists a number n such that
II 4zkII < e. But then the estimation fk(t)zk(t)I < e holds for t <
Hence f is singular. Condition (5.10) for f is false, thus f is not representable.
 5.7. INNER PRODUCTS ON HILBERT C-MODULES 131

PROPOSITION 5.7.10. Let M be a Hubert and let f E M' be a

nonsingular functional. Then it is representable.
PROOF. Nonsingularity and the Cauchy-Bunyakovskii inequality for Hubert
modules give us the estimate 0 < c (1' z) hf hl(z, z)"2, from which it follows
that the module SPanA z is isomorphic to A. Let us show that the decomposition
M = SpanA Ker f into a (nonorthogonal) direct sum holds. If x e M. then put
a = (f,zr' . (f,x); y = x — za. Then x = za + y and y E Kerf. Uniqueness of
this decomposition is clear. Set M1 = SpanA z and M2 = Ker f and let us choose,
using Corollary 2.8.18, a new inner product in such a way that the submodules M1
and M2 become orthogonal. Then z± Ker f. Put z' = (z, . (f, z). Then

(f,x) = i.e., f is representable. 0

Recall that a A is of topological stable rank one [111] (notation
tsr(A) = 1) if invertible elements of A are dense in A.

PROPOSITION 5.7.11. Let a A be of topological stable rank one. Then

the representable functionals are dense in with respect to the initial norm.
PROOF. It is sufficient to verify that nonsingular functionals arc dense in
If f = e then invertible element gi in A such that
one can find an
hg' — fill <e. Putting g = (91,12,13,...) E and taking z = e1 = (1,0,0,...) E
we obtain that (g, z) is invertible and — I <E. lb
0
The situation with representability of functionals in general is more compli-
cated. Consider the following example.
EXAMPLE 5.7.12. Let A be the of hounded operators on an infinite-
dimensional Hilbert space. As is shown in 2.5.6, there exists an isomorphism of
Hubert modules S A Let a E A, f = S(a). Then the condition (f, x) = 0
can be written as

(5.13) (S(a),x) = (S(a),S(S'z)) = (a,S'x) = a* = 0. .

If a E A is invertible, then it follows from (5.10) that S1x = 0, i.e., Kerf = 0.

But if f is representable, (f,.) = (z, with z E HA, then the kernel of f cannot
vanish. Therefore the functional f = S( 1 A) is not representable, moreover, it has
an open neighborhood consisting of nonrepresentable functionals.
To complete this section we show how the averaging Theorem 2.8.14 can be
generalized from compact groups to amenable ones in the case of Hilbert
modules.
THEOREM 5.7.13 ([82]). Let M be a Hilbert module over a A, and
let p: C —p EndA M
be a uniformly bounded representation of an amenable group
C, C, g E G. Then there exists an inner product (., equivalent to the
initial one and such that the representation p is unitary with respect to it.
PROOF. For any normal linear functional e where denotes the pre-
dual Banach space for A, let define the function on the group C the
formula
=
x,y E M, g E C. By our assumption, this function is bounded. Put
=
132 5. MULTIPLIERS

where it'I is an invariant mean on the group G. By fixing x and y, we obtain a

linear bounded map
C; '—i
This map is an element of A. Define the new inner product on the module
M by (x, y)3 = E A. Let us verify that it is well defined. Its sesquilinearity is
obvious. If E is a state, then 0 for any g G, hence =
0. Suppose that 0 for some x E M. Then = 0. But, since
(x,x) = (p(g')(p(g)x),p(g')(p(g)x)) C2(p(g)x,p(g)x)
1
for any g E G, the estimate holds, where e denotes the neutral
element of C, and

Hence, > and = 0, i.e., Ø((x,x)) = 0 for an arbitrary state

4. But then (x.x) = 0, hence x = 0. Therefore is an inner product. The
property (p(g)x, = (x, y),3 obviously holds, hence the representation p is
unitary. Equivalence of (.,.) and (•, follows immediately from the estimate
1
(p(g)x,p(g)x) C2(x,x),
which is valid for any g E G. 0
 CHAPTER 6

Diagonalization of Operators over

6.1. Problem of diagonalizing operators in Hubert
It is well known that in the commutative case, i.e., for A = C(X) a commutative
of continuous functions on a compact topological space X, compact
selfadjoint operators acting on the Hilbert HA (which are the same as
norm continuous families of classical compact selfadjoint operators K(x), x E X [11)
can be diagonalized pointwise and that diagonal elements determine continuous
functions A(x) E A by the theorem of the continuity of spectra. We consider here
a noncoinmutative analog of this well-known fact. We prove in particular that a
compact positive operator acting on a Hilbert over a finite W*_algebra
A can be reduced to diagonal form with diagonal entries from the algebra A.
If A c A is a weakly dense in a finite then operators
acting on the Hilbert C HA can be diagonalized in a larger A-
module. When we pass to noncommutative "eigenvalues" are not
defined uniquely (only up to the unitary equivalence), but in some cases one can
choose them from the initial A instead of the larger A. We
describe some cases, when such possibility can be realized. We also show how these
results can be applied to describe spectra of operators commuting with an action
of some
The following theorem due to Kadison was the first result about diagonalization
of operators in the noncommutative case, which initiated a series of further results
in this direction [50, 95, 137, 46, 79, 81, 831.
THEOREM 6.1.1 LetA be a a-finite a selfadjoint
(nxn)-matrix with the entries from A. Then there is a unitary (nxn)-matrix U
with the entries from A such that a matrix U* KU is diagonal with "eigenvalues"
from A.
Consider a simple example showing what kind of difficulties can occur while
diagonalizing matrices over Let A = C[—1, 1]. Put
F(t)=IT ift>0,
iftO.
Define the operator K: A A —÷ A A by
K_(1+t 1(t)
—

For t < 0, the eigenvectors of an operator K are (1,0) and (0, 1), and for t > 0,
they are and — One can glue these eigenvectors together in such
a way that their coordinates are continuous functions on [—1, 1], hence one can
diagonalize the operator K over the A.
133
134 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

Let A be a A compact operator K acting on a Hubert

M is called strictly positive if the element (Kx, x) E A is positive for any x E M
and if the spectral projection of the operator K corresponding to the origin is equal
to zero. This condition means that the operator K is a strictly positive element of
the
By Theorem 3.2.1 due to Paschke [100], in the case where A is a
the inner product can be extended in a natural way to the dual module M'.
For modules (A) and HA we use common notation MA.
DEFINITION 6.1.2. Let A be a A selfadjoint operator K in a
Hilbert A-module MA is called diagonalizable if there exists a set {x2} of elements
in M'A and a set of elements E A such that
(i) the set is orthonormal, i.e., (x2,x3) =
(ii) the module M'A coincides with the A-module H' dual to the module H
generated by the set {x2},
(iii) =
(iv) for any unitary elements u2, E A one has the operator inequality
(6.1)

Elements x2 are called "eigenvectors" and the operators "eigenvalues" of the

operator K. We should keep in mind that the "eigenvectors" and the "eigenvalues"
are not defined uniquely. Condition (6.1) means that our notion of diagonalizability
includes the natural ordering of "eigenvalues". For example, if A = L°° (X) is a
commutative and K is an (already diagonal) operator of the form
(1(x) 0
0 g(x)
on the module A A, then, after diagonalizing this operator, we get the operator
(max(f(x),g(x)) 0
0 min(f(x),g(x))
Before we proceed, we give a few additional examples explaining Definition
6.1.2.
EXAMPLE 6.1.3. Consider the A of all complex (2x2)-matrices,
the module M = A A with the usual A-valued inner product and a selfadjoint
operator K on M acting by the formula K = + where
((1 o\ (0 o\\ 1(0 o\ (2 0
o))' 2

If we take x, y E A e A as "eigenvectors" of the operator K, then the corresponding

"eigenvalues" would be
(1 (4 0'\
9)' 4)'
but they are not comparable as elements of the positive cone of A. By choosing
the "eigenvectors" in another way,
((0 o\ (i o\\ 1(1 o\ (0 o\\
 6.1. DIAGONALIZATION PROBLEM 135

we are in the situation of Definition 6.1.2. In this case the "eigenvalues" are
(4 o\ (1 0
9)'
These "eigenvalues" are ordered and the "eigenvectors" x1, £2 satisfy the condition
(xi, x3) = Throwing out requirement (1) of Definition 6.1.2, one can choose
K-invariant submodules in M and generating "eigenvectors"
((0 o\ (1 o))'
i\\ ((i i\ (00 1

Then the corresponding "eigenvalues" are

(4 o\ (1 0
9)'
They are ordered in a positive cone of algebra A. But the "eigenvectors" generating
the K-invariant submodules in M do not satisfy the condition (xi, x3) =
The following example shows the necessity (even in the commutative case) of
using dual Hilbert modules for diagonalizing compact operators.
EXAMPLE 6.1.4. Let A = L°°([0; 1]) and let bk be a monotone sequence of
positive numbers converging to zero. Put

Ak
(1 ii. 1 1

— 0, for other t,

and let fk(t) = bk . ak(t). Let K be a compact operator given by the matrix

fi 12
12 0 •.. 0

K=

in the standard basis of the module HA. This operator can be easily diagonalized
pointwise. The eigenvector corresponding to the maximal eigenvalue has the form
x= where

X1(t) = ai(t) + and = çan(t), n> 1.

k> 1

Then (x, x) = >k ak(t) = 1. This series is not norm-convergent in A, so x E

\ HA, hence the operator K cannot be diagonalized in the module HA.
This example leads to the following statement.
PRoPosITIoN 6.1.5 ([46]). Let A be a finite with infinite center.
Consider a Hubert A-module M such that HA c M c Hk. If each strictly positive
compact operator acting on M can be diagonalized in M, then the module M is
self-dual, i.e., M = Hk.
136 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

6.2. Quadratic forms on related to selfadjoint operators

If A is a C*_algebra with a faithful finite trace r and if D is a selfadjoint
operator acting on a Hubert A-module M, then a quadratic form on M can be
defined by the formula Q(x) = T ((Dx, x)) for x E M. Below we will see that such
C*_mnodule quadratic forms have properties similar to those of the usual Hubert
space quadratic forms. By B1(M) we denote the unit ball of the module M.
PRoPosITIoN 6.2.1. Let D be a positive operator on M such that Ker D = 0.
If the quadratic form Q(x) reaches its supremum over B1(M) at some element x,
then the operator (x, x) is a projection.
PROOF. Denote (x, .r) by h E A. Then we have h 0 and h* =
Suppose that the spectrum of the operator Ii contains sonic number c different from
zero and one. Define the function 1(t) on the segment [0; 1] j Sp /i by the formula
( _L,
f(t)—1 ——
e<t<1,
where 0 < & < c. Put a = f(h), x' = xa. Then (x',x') = aha = ha2. This
operator is equal to the value of the function t applied to the operator ii.
.

Since f2(t) < 1 for 0 < t < 1. we obtain the estimate IIha2II < 1 and the element
x' lies in B1(M). By assumption, (Dx. x) is a positive operator. Denote it by k2,
where k > 0. Then
Q(x') — Q(x) = r((Dxa,xa) — (Dx,x)) = r(ak2a — k2)

= r(a2k2 — k2) = r(ka2k — k2)

= r(k(a2 — 1)k).
By definition, the operator a2 — 1 is also positive. Denote it by b2, where b 0.
Then
r(k(a2 — 1)k) = r(kb2k) = r(b(Dx,x)b) =
hence
(6.2) Q(x') — Q(x) = T((Dxb,xb)).

But (xb, xb) = b(x, x)b = bhb and the operator bhb corresponds to the function
t(f2(t) — 1). This function differs from zero at t = e, so the operator bhb differs
from zero, hence xb differs from zero as well. Note that the operator D'/2 is well
defined and that Ker D1/2 = 0. Therefore
r((Dxb,xb)) = T((D11'2xb,D11'2xb)) >0,
so we obtain from (6.2) the inequality Q(x') — Q(x) > 0, which contradicts the
assumption that Q(x) is a supremum of the quadratic form Q over B1(M). Thus,
we have proved that Sp /i does not contain values other than zero and one, so h is
a projection. 0
PROPOSITION 6.2.2. Let x E B1 (M) be a vector at which the quadratic form
Q (D.,.) reaches its suprcmum over B1, and let £ C M denote the submodule
generated by the vector x. Then M = £ and the submodules £ and £' are
D-invariant, i.e.. DL C £, C
 6.2. QUADRATIC FORMS ON RELATED TO SELFADJOINT OPERATORS 137

PROOF. By the previous proposition, the operator (x, x) is a projection. There-

fore £ is a projective module and, by the Dupré - Fillmore Theorem 1.4.6, M =
£ £L. Choose an element y E such that = 1. Put Xt = xcost + ysint.
Since Q reaches its maximum at the point x0 = x, one should have = 0 at
t = 0. It is easy to see that

= T((DX,y) + (Dx,y)*).

If (Dx,y) 0, then put a = (Dx,y). Obviously, Jail = 1. Further, put

z = ya* and = xcost + zsint. Then

0 = + (Dx.z)*)
T((Dx,y)a* +a(Dx.y)*) = r(2aa* .

= 211 (Dx, y) Ii

r is faithful, it follows that a = 0. Therefore the vector Dx is

orthogonal to any y E hence Dx E £. Since the operator D is selfadjoint, we
conclude that (Dx,y) = 0 for any y E Therefore c £.L.

PRoPOsITION 6.2.3. Let x E Bl(HA) be a vector, at which the quadratic form

its supremizm over the set Bl(HA). Then (x, x) = 1.
Q reaches

PROOF. If the operator (x, x) is less than one, then there exists a vector y E
such that < 1, y 0 and yq = y, where q = 1 — (x,x) is a projection. Then
(x + y,x + y) = (x,x) + (y,y) 1, so x + y E Bl(HA). But by the previous
proposition, one has Q(x + y) Q(x) + Q(y) and since y 0 and KerD 0,
the inequality Q(y) > 0 should be valid, hence Q(x + y) > Q(x) and the obtained
contradiction proves the proposition. U

PROPOsITION 6.2.4. Let the operator D on the module M be positive and let
its "eigenvectors" form a basis in M. If these "eigenvalues" satisfy the inequality
Sp Sp then the supremum of the quadratic form Q over B1 (M) is reached
at the first "eigenvector" x1 and is equal to

PROOF. Any vector x E B1(M) can be written as x = where e A.

Then

Q(x) = r((Dx,x)) =r
+

Let Sp d SpA2. Then

Q(x)
i> 1
138 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

since the inequality 1 follows from IIX II

1. Further, we have
Q(x) + d(1 — = — +d
= — d)a1) + d = T((Al — d

Thus, rc7ti) is the supremum of the form Q(x) over B1(M) and it is reached at the
vectorx1. 0
6.3. Diagonalizing operators in the
THEOREM 6.3.1 ([81]). Let A be a that admits a decomposition
into a direct integral of finite factors. Then a compact strictly positive operator
K in the module HA can be diagonalized in Hk and its "eigenvalues" are defined
uniquely up to the unitary equivalence.
PROOF. Note that the is semifinite and its center is
isomorphic to the center Z of the algebra A. Therefore (see [123]) both these
algebras can be decomposed into a direct integral of factors over a Bore! space r
with a measure d'y such that L°°([') = Z. The operator K then also admits the
decomposition
K=J
r
Put T = T ® tr, where T is the standard Z-valued finite trace on A and tr is the
standard trace in the Hubert space 12. This is a semifinite center-valued trace on
the (Hi). Let us show first that if we separate the spectrum of
K from zero, then the resulting operator would lie in the idea! of operators with a
finite trace in the algebra (Hi). By XE we denote, as usual, the characteristic
function of a set E C R.
LEMMA 6.3.2. For each e> 0, the inequality
<00
holds almost everywhere on F.
PROOF. Denote the spectral projection by P. Then the operator
inequality
(6.3) KIimp E
holds on the A-submodule ImP C by the spectral theorem. The compactness
of the operator K implies that the module HA can be decomposed into the direct
sum HA 1?. in such a way that < C for sufficiently large n.
Passing to dual modules, we obtain the estimate
(6.4) IIKIwII <C,
where = 7Z'. Denote by Q the projection onto 7?.' in Hk. Then the
projection on ImP fl 1?.' equals PA Q and the projection onto (KerP fl
is equal to P V Q (here we use standard notation for projections onto intersections
and sums of subspaces; see, e.g., [123]). It follows from Theorem V.1.37 of [123]
that
(6.5)
 6.3. DIAGONALIZING OPERATORS IN THE W-CASE 139

Since P V Q 1, where 1 denotes the unit operator in (Hk), we get the

inequality T(P V Q — Q) T(l — Q). But 1 — Q is the projection onto
so its trace is equal to n. Therefore it follows from (6.5) that — P A Q) n.
Comparing (6.3) with (6.4), we see that ImP fl RI = 0, hence P A Q = 0 and,
finally, we can conclude that T(P) ( n. 0
Now we need one simple fact about measurable functions.
LEMMA 6.3.3. Let r be a Borel space and let : r x R —' R be a function
such that
(i) for each A E R the function A) is measurable on
(ii) the function i/.'('y; A) is right-semicontinuous and monotone nonincreasing
in the second argument for almost all 'y.
For each real number put
(6.6) = inf{A <
Then the function is measurable.
PROOF. We should show that the set V = {-y ç /3} is measurable
:

for each /3 E R. But the definition of the function cQ('y) and (ii) imply that
V = {'y : inf{A : A) ( /3} = {-y : The measurability follows
from (i). 0
Recall that the operator K is decomposable over r. Let
Pi(7;A) = P2(y;A) =
be the spectral projections of the operator corresponding to the sets (A; +oo)
and [A; +oo) respectively. Put
P1(A) = X(A;+oo)(tfl, P2(A) =
and A) = i'(P1(A)). Then this function satisfies the conditions of Lemma 6.3.3,
hence the function
(6.7) A(7) = inf{A : q5(y; A) 1}

is measurable.
Now let us define two new projections in Hk:

(gQ\ Jr = jr
P2
= j1'

and let us check that they are well defined.

d'1
= j 1'
P2(y; A(y)) d'y

LEMMA 6.3.4. The operator-valued functions and A(-y)) are

measurable.
PROOF. We can assume that the A is faithfully represented in the
direct integral of Hilbert spaces H = H(y) d-y equipped with the scalar product
(. .) We have to show that the function

(6.9)
140 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

is measurable for all = E H. By Theorem XIII.85 in [107] the function

=
satisfies the conditions of Lemma 6.3.3. The measurability of (6.9) follows from the
measurability of the set U = {'y : ( a} for each a. But the definition of
the function (6.6) implies that

This set is measurable because of the measurability of the function A('y) — (-y).
The case of the projection P2 can be dealt with in the same way. 0
COROLLARY 6.3.5. The projections P1 and P2 given by formula (6.8) are well
defined and
1, T(P2) 1, P1 ç P2.

These two projections define the decomposition of the module Hk into the
direct sum of three submodules,
(6.10)
where H+ = ImP1, H0 = Im(P2 — P1), H_ KerP2. The operator K commutes
with these projections, since the operators K('y) commute with the projections
P1(y;A(y)) and for almost all -y. Therefore the operator K can be
written, with respect to the decomposition (6.10), as
fK+ 0 ü
K=( 0
0KJ
K0 U

and the operator K0 is an operator of multiplication by the scalar .X(-y) for almost
all -y. Therefore any submodule of the module H0 is invariant for the operator K.
We can conclude from Corollary 6.3.5 that there exists a projection P such that
Pi P P2 and T(P) 1. Then the operator K is diagonal with respect to the
decomposition = ImP W Ker P,
K_(K1 0
0 K'
Note that the module Tm P is isomorphic to A, since the projections onto these
two submodules in Hk both have the same center-valued traces equal to 1, hence
these projections are equivalent [123]. Let x1 E Hk be a generator of the module
Tm P, (x1, x1) = 1. If this generator is fixed, then one can represent the operator
K1 Tm P —p Tm P as the operator of multiplication by some element A1 E A,
i.e., K1x1a = x1)qa for a E A, x1a E ImP. By Theorem 3.7.1, the module KerP
is isomorphic to the module Hk and the operator K' is obviously compact in the
module Ker P, so we can apply Lemma 6.3.2 to K'. One also has the operator
inequality K1 = K'.
Proceeding by induction, we can find elements E Hk and operators EA
such that (xi, x3) = Kx2 = and Denote by .N' the Hilbert
A-module generated by all these elements It is clear that .iV HA. Note that
the operator KIN is not necessarily compact. We need to show that N' = Hk.
LEMMA 6.3.6. The norms of the operators tend to zero as i —' oo.
 6.3. DIAGONALIZING OPERATORS IN THE W-CASE 141

PROOF. Since the sequence is monotone nonincreasing, it converges to

some number b 0. Suppose that b 0. The operators X, can be decomposed into
a direct integral over r. From the construction of the operators we can conclude
that there exist numbers (y) such that
(6.11)
If we decompose x2 coordinatewise into the direct integral, x2 = jj! x2('y)
then, for almost all the vectors xj(7) are orthonormal in Hk and K('y)x2('y) =
Define the function b('y) as the limit of the norms in the algebras
Then
(6.12) = Xj('y)) >
where the inner product is taken also in the A('y)-modules Let x be an
element of the module Then it can be written as

x(7) = for some = f A.

If (x, x) = 1 for almost all then

(6.13) = 1.

From (6.11) and (6.12) we can conclude that the operator inequality
is valid for all i. Therefore, since b(y) is a scalar, due to (6.13) one has
=

Further, we obtain that

II(Kx,x)II = ess ess supb('y) = b,

and, since b> 0 by assumption, we have

(6.14) II(Kx,x>II > b

for all x E satisfying (x, x) = 1. Consider now the projection —i

If the spectrum of the operator K is separated from zero, then P,., defines an
inclusion of the module A( into the module which is impossible in the case of
finite Therefore for any e> 0, we can find a vector x E such that
(x,x) = 1 and <E. Put x' = x — x'. Then Ilx'II IIx"II
Estimate the norm of (Kx, x):
II (Kx, x) (Kx', x') + 2 (Kx', x") —4— (Kx", x") 1
11.1(11 Ix'jJ2 —i— 2 IIx'II IIx"II —1— II(Kx", x")II
+ II(Kx",x")IL
IIKIIE

Since due to the compactness of the operator K, one should have the
estimate (Kx",x") & for sufficiently large n. Therefore II(Kx, x)M < &', where
= IIKIIe2 + 2IIKIIE + &. If we choose E sufficiently small, then this estimate
contradicts (6.14), hence our assumption b> 0 is false. 0
142 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

We have shown that the norm of the restriction of the operator K onto the
orthogonal complement to the vectors x1,. .. tends to zero. Hence x E
,

implies that IKxII = 0. But since the spectral projection of K corresponding

to the origin is zero, Ker K = 0 and hence = 0. Therefore it follows from
Proposition 3.4.3 that .A(' = Hk. 0
The "eigenvalues" of the operator K are obviously not defined uniquely and
the same is true for the "eigenvectors" x=
where u, E A is a unitary element, then the elements = u are also the
"eigenvalues" of the operator K. Nevertheless the following statement shows that
if the "eigenvalues" are ordered, then they are unique up to the unitary equivalence.
PROPOSITION 6.3.7. Let A1 and x2 be the "eigenvalues" and "eigenvectors"
constructed in the proof of Theorem 6.3.1 and let be "eigenvalues" of the operator
K corresponding to another "basis" {y2 } of the module (i.e., the orthogonal
complement to the submodule generated by {yj } is zero). If the operator inequality
V holds for all unitaries v1 E A and for all numbers i, then A1
and coincide up to the unitary equivalence.
PROOF. It is easy to verify that, by assumption, the inequality
in the factors A('y) for almost all 'y E r. Therefore each
projection onto a submodule generated by each vector y1 in the module Hk is a
spectral projection for the operator K. Denote the projection onto SpanA (yi) by Q.
Then obviously P1 Q P2, where P1 and P2 were defined by formulas (6.8). We
can decompose Q into the sum Q = P1 R and the projection P onto SpanA(x1)
can be_ written as the sum P = P1 S, where R and S are some projections.
Since T(P) = T(Q) = 1, R and S are equivalent and ImR ImS. This module
isomorphism commutes with the action of the operator K, since its restriction onto
these modules is scalar and coincides with multiplication by d1 ('y) for almost all
'y E r. Therefore there exists a unitary element u1 E A such that it induces this
isomorphism between the modules ImP and Im Q and A1 = Proceeding by
induction, we obtain the unitary equivalence of the two sets of "eigenvalues". 0
To finish this section let us discuss the theorem about diagonalizing operators
in the case when the requirement of the strict positiveness of the operator K is
discarded. If K is an arbitrary selfadjoint compact operator in HA or Hk, then the
module Hk can be decomposed into the direct sum Hk = H_ Ker K H+ in such
a way that the restriction of K onto (resp. H_) is positive (resp. negative). We
can find two sets of the "eigenvectors" separately in H+ and in H_, but we should
have to dismiss the requirement for the "eigenvectors" to be normalized, i.e., the
inner products of such vectors by themselves are now equal not to unit but to some
projections. Let us consider the following example.
EXAMPLE 6.3.8. Let p E A be a nontrivial projection, p 0, p 1. Consider
the module M = By the Kasparov stabilization theorem (Theorem 1.4.2),
we have M HA. Let {e1} be a basis in the module HA, e = (O,p) E HA epA.
Define the operator K on M by the formulas Ke1 = A2, Ke = 0, where {A1} is a
sequence of positive numbers converging to zero. Then the operator K cannot be
diagonalized in the sense of Definition 6.1.2, since it is impossible to simultaneously
satisfy conditions (i) and (iv) of this definition.
 6.4. CONTINUITY OF "EIGENVALUES" 143

Note also that in [46] it is shown that any selfadjoint operator acting on a
self-dual Hilbert module over a can be diagonalized, but without the
requirement for "eigenvectors" to satisfy the condition (x, x) = 1, and "eigenvalues"
are defined not uniquely (even as classes of the unitary equivalence).

6.4. Continuity of "eigenvalues"

In this section we show that "eigenvalues" of compact operators on Hilbert
over enjoy the property of continuity. Throughout this
section A denotes a that admits a decomposition into a direct integral
of finite factors.
LEMMA 6.4.1. Let K1 = K2 = be strictly positive
operators on with a finite spectrum and let IlK1 — K2I1 <E. Then
(i) there exists a unitary operator U in (A) mapping the "eigenvectors"
of K2 to the "eigenvectors" of K1 such that IIU*KIU — K211 < E,
(ii) the "eigenvalues" of the operators Kr (r = 1,2) can be chosen to
—
satisfy the inequality < e.

PROOF. Since the algebra A can be decomposed into a direct integral of finite
factors, it is sufficient to prove the lemma for the case where A is a type factor
(the proof for type factors is obvious). Denote by EK (A) the spectral projection
of the operator K corresponding to the set (—oc, A). The faithful finite trace r
on A can be extended to the (infinite) trace f = tr ®r on the algebra EndA (Hi)
and then restricted to a finite trace on the smaller algebra EndA(Lfl(A)), where
f(1) = n. Put
= inf A, 0 < < n.

Using the continuous minimax principle [96], we obtain

(6.15) = iflf (Ke,e)

P is the set of projections in EndA(Lfl(A)) and where denotes the

scalar product in the Hilbert space where the algebra EndA (A)) is represented.
It follows from (6.15) that JJK1 — K211 <E implies the estimate

(6.16) — <E.

Let be the projections onto the "eigenvectors" of the operators K,. corre-
sponding to the greatest "eigenvalues" f(Qcr)) = 1. Out of the two partitions
of unity and given by decompositions of K1 and K2, one
can construct a finer partition of unity. By [1231, there exist two sets of mutually
orthogonal projections E EndA(Lfl(A)) such that
(i)

m for some i or j.
144 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

Then (after possible renumbering) the operators Kr can be written in the form
(r)
Kr >2 where E R. This enables us to define a
unitary operator U —' such that
(6.17)

U maps the A-modules generated by the

"eigenvectors" into the modules generated by i.e., = U,
=
for some unitary elements u2 E A. Put

n(a) = cl}.
m>ri

Then and it follows from (6.16) that 1(Q) — < e. But,

changing a, we obtain that
(6.18) —

a= 1 (then i = 1), we obtain

— (r) (r)
K —
— I — am
mn( 1)

It follows from (6.17) and (6.18) that

(6.19) —
= (au) — =
mn(1)
(r)
Choosing to satisfy the conditions A1(1)x1
. . . __(1)
= _(1)
x1
(1)
A1
(2)
and A1 x1(2) = (2)
x1
(2)
A1
we get the estimate

(6.20) — <E.

In the same way we can obtain the estimates (6.19), (6.20) for other i.

THEOREM 6.4.2. Let K,. : HA —p HA, r= 1,2, be compact strictly positive

operators and let IlK1 — K211 Then
(i) there exists a unitary operator U in Hk such that it maps the "eigenvec-
tors" of K2 into the "eigenvectors" of K1 and IIU*KIU — K2jj < e;
(ii) the "eigenvalues" of the operators K,. (r = 1,2) can be chosen to
satisfy the inequality — <

PROOF. Let (A) E Hk denote the Hilbert submodule generated by the

first n "eigenvectors" of the operator Kr, By Theorem 3.7.1,
the orthogonal complement to this submodule is isomorphic to and the norm
of the restriction of the compact operator K,. onto the orthogonal complement to
in Hk tends to zero. Therefore it is sufficient to consider only the case
of operators acting on the module (A) and there one can approximate these
operators by operators with finite spectrum. 0
 6.5. CASE OF INFINITE W-ALGEBRAS 145

6.5. Case of infinite

Let us consider now the case of Hubert over infinite
Results obtained jointly with M. Frank [461 show that diagonalizing operators in
self-dual Hubert modules over infinite is significantly different both
from the case of finite and from the classical case of Hilbert spaces.
This explains why the case of infinite is of lesser interest in applica-
tions.
PROPOSITION 6.5.1 ([461). Let A be a with an infinite number of
pairwise orthogonal nontrivial projections {pj i E N} equivalent to 'A. the sum
:

of which is equal to 1A (the infinite sum is understood in the sense of the

convergence). Then the Hilbert A-module Hk equipped with the standard A-valued
inner product is isomorphic to the Hilbert A-module {A, (., where (a, b)A =
a*b.

PROOF. Suppose that the equivalence of projections {pj i N} and 1A is

realized by partial isometries i E N} E A, 1A =
: = Then the
map
S: Hk —+ A,

with the inverse map

a Hilbert A-module isomorphism of the modules Hk and A. 0

COROLLARY 6.5.2. Let A be an infinite Then each bounded oper-
ator T on the module can be diagonalized and the formula
=
holds for all for some AT E A and for partial isometries EA
described in the proof of Proposition 6.5.1.
PROOF. A of one of the types or III possesses a set of
partial isometries with properties described in the proof of Proposition 6.5.1. It
follows from this proposition that the operator T acting on the module is
unitarily equivalent to the operator STS' acting on the module A. But each
bounded operator in A is an operator of multiplication (from the left) by some
element of the algebra A. 0
COROLLARY 6.5.3. Let A be a not containing type ii < 00,
and Iii factors in its decomposition into the direct integral. Let M be a self-dual
Hubert A-module possessing a countably generated A-pre-dual Hubert A-module.
Then each bounded operator T on M can be diagonalized and the formula
(6.21) T(x) = UAT(U,X)
holds for all x E M, for some AT E A and for an "eigenvector" u E M, common
for all T.
PROOF. Since the module M possesses a countably generated pre-dual Hilbert
A-module, M is a direct summand of the Hilbert A-module Hk by the Kasparov
stabilization theorem [631 (see Theorem 1.4.2). Therefore it is sufficient to prove
the statement only for the self-dual Hilbert A-module "kU For convenience we
146 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

denote the projection onto the module M in the module Hk by P. Consider the
central decomposition of the algebra A into a direct integral. Each fiber in it is a
factor of one of the types or III by assumption. It is easy to see that the
modules Hk and A are isomorphic: we can either apply Corollary 6.5.2 fiberwise or
directly find an appropriate set of partial isometrics } E A. Then any bounded
operator T in the module Hk can be diagonalized. To get formula (6.21), it remains
0
REMARK 6.5.4. Let A = 13(H) be a Then there are selfadjoint
operators T E A that cannot be diagonalized in a stronger sense. More precisely,
it is impossible to write such operators as the sum > where E C = Z(A)
and P1 = = E A (compare with the Weyl theorem). Therefore, in the case
of infinite W*_algebras, Corollaries 6.5.2 and 6.5.3 are the best results one could
hope for.
6.6. Case of of real rank zero
It is known that in the commutative case, i.e., when A = C(X), the com-
pact selfadjoint operators cannot be diagonalized in the module HA. However
this becomes possible if we pass to a larger module over the larger
L°° (X) = A D A, which is obtained from the algebra A by completion (on bounded
sets) with respect to the weak topology in the natural representation of the alge-
bra A in the Hilbert space H L2(X) with a new (integral) norm assigned to a
finite faithful trace (measure) on A. Unlike the the
are continuous, i.e., they lie in the initial C*_algebra A. This leads to the following
definition.
Let A be a separable with a finite faithful trace r. Let (a,
b a nondegenerate inner product on A and let = (., be
the norm defined by this trace. Completing A with respect to this norm we obtain
the Hilbert space L2(A) and the faithful representation of the algebra A on this
space. Let A = L°° (A) be the corresponding trace finite containing
A as a weakly closed subalgebra. Let K be a compact strictly positive operator
in the module HA. This operator admits a natural extension to the larger module
HA over the larger algebra A, where it is still compact and strictly positive, and
where, by Theorem 6.3.1, it can be diagonalized in this module.
DEFINITION 6.6.1. Let A be a that admits a decomposition into a
direct integral of finite factors and let a separable C-algebra A he a weakly dense
subalgebra in A. Let K be a compact strictly positive operator on HA. We call
A an algebra admitting weak diagonalization if, while diagonalizing K in Hk, its
diagonal entries can be chosen within the A E A.
Note that if a admits weak diagonalization, then each selfadjoint
operator of a finite rank K e 0 A can also be diagonalized over the
A in such a way that its diagonal entries can be chosen within the C*_algebra A.
Denote by 2 = C(Z) the center of the A and let T: A C(Z)
be the standard faithful 2-valued trace on A with T(1) = 1. Assume that a
A E A possesses the following property:
(*) for any two projections p, q c A one can find projections Rq, <p,
rq q in the A such that they are equivalent in A and
= = min{T(p)(z),T(q)(z)}, z E Z.
 6.6. CASE OF C-ALGEBRAS OF REAL RANK ZERO 147

Below we show that the class of admitting weak diagonalization con-

tains, besides commutative algebras, also weakly dense separable
of real rank zero with property (*) in finite
Recall that real rank zero (RR(A) = 0) means [19] that each selfadjoint oper-
ator in A can be approximated by operators with finite spectrum, i.e., of the form
where p2 e A are selfadjoint mutually orthogonal projections and E R.
By [19], in this case one also has = 0.
THEOREM 6.6.2. Let A be a weakly dense separable with property
(*) and with RR(A) = 0 in a Wt-algebra A that admits a decomposition into a
direct integral of finite factors. If K is a compact strictly positive operator in the
A-module HA, then the "eigenvalues" of the natural extension of the operator
K to the A-module can be chosen in such a way that A1 e A.

PROOF. By results of S. Zhang [137, 19], an operator K can be approximated

by operators E EndA(Lfl(A)) with a finite spectrum. By [137, Corollary 3.5],
there exist unitary operators E such that the operators
o

0 A(n)

are diagonal and E A have a finite spectrum. Let us show that, using property
(*) and choosing such in an appropriate way, one can satisfy condition (6.1)
for the "eigenvalues" Let Aa = Ab = where r1 e A
are projections. Suppose that a < b, but for some rn and n one has 13m >
Diagonalizing projections [137], we can find projections s, e A such that they are
equivalent to rj, and sj = and Put

= >2 akqk >2

(in) (m)

and note that the operators

(Ai 0\ f)1 0

I I
and

are unitarily equivalent. Repeating this procedure each time when > ak, one
can satisfy (6.1) for and In the same one can order all "eigenvalues" of
within A. By property (*), if — K,1_1 < then one can find unitaries
in A such that
(6.22) — A?Ll) <

E A. Taking (if necessary) a subsiibsequence }, we can

assume that = in (6.22). Then the sequence
(1) (1) (2) (3) * (3)
= A2 ,
= Uj,2, = *
Ui,2U7,3,
148 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

is a Cauchy sequence in A. Denote its limit by E A. By Theorem 6.4.2, for any

one can find a unitary operator mapping first ii. "eigenvectors" of K into
the of Put = e EndA(Hk). Then
(6.23)
and — Ku —+ 0. Passing to the limit in (6.23) we obtain = Therefore
are the "eigenvalues" of the operator K. 0
Note that condition (*) is necessary for a to be weakly diagonaliz-
able. Indeed, if the operator K is the direct sum of two projections,

°),
then its can be ordered only if the "common part" of projections 1 —p
and q is a projection from the algebra A.

6.7. Case of continuous fields of trace

Let X be a compact Hausdorif space and let {A(x), x E X} be a family of unital
equipped with faithful finite traces = 1. By A(x) we
denote the set of functions a = a(x) defined on X and such that a(x) e A(x) for
any x E X.
DEFINIT!ON 6.7.1. Let A C A(x) be a subset with the following proper-
ties:
(i) A is a *-subalgebra in HxEX A(x);
(ii) the set {a(x),a E A} is dense in the C*_algebra A(x) for any x E X;
(iii) the function x IIa(x)fl is continuous for each a E A:
(iv) let a e flxEX A(x): if for each x E X and for all e > 0 there exists an
element a' E A such that — a'(x) < s in some neighborhood of
the point x. then a E A;
(v) the function x is continuous for each a A.
Then the triple (A(x), X, A) is called a continuous field of trace
Note that the first four properties of this definition make the standard definition
of a continuous field of [31]. It is known that A is a For
the sake of brevity we denote by IIa(x)ll the norm of an element a = a(x) A, i.e.,
IIa(x)Ii. If V C X is a closed subset, then by AIv we denote the restriction of
a continuous field of onto V.
It is clear that r : A rX(a(x)) is a trace on A with values in C(X) C A.
This trace defines an inner product on A(x) by
= =
and a C(X)-valued inner product on A,
(6.24) E C(X)
with the norm
Jb(x)117 = sup(b(x), = sup

Let H(x) = L2(A(x)) be the completion of A(x) with respect to the norm
Then the algebra A(x) is faithfully represented in the Hilbert space H(x) and we
 6.7. CASE OF CONTINUOUS FIELDS OF TRACE C-ALGEBRAS 149

can pass to the appropriate B(x) = L°°(A(x)), i.e., the weak closure
of A(x) (the notation L°° is used by analogy with the commutative case).
Now let dx be a o-finite Borel measure on X. Note that the function (6.24) is
continuous. Therefore one can put

= (b,b)r.

Denote the completion of the algebra A with respect to this norm by H = L2 (A).
The algebra A is faithfully represented on the Hubert space H and we denote the
corresponding We-algebra by A = L°° (A) A. It is clear that
ED

A= / B(x)dx.
Jx
Since the L°°(X)-valued trace = dx defined by tile family of traces on
A is faithful and finite, A is a finite
We are interested here oniy in the case, when X is either a segment or a
circle. Then, if X is divided by points {xk} into segments Dk [1k: xk+1] and
if ak(x) E AIDk is a continuous field for each k. we call a set a(x) = {ak(x)} a
piecewise continuous field on X. Such piecewise continuous fields clearly belong to
the A. The distance from such a field to the A is given by
the formula
dist(a(x), A) = sup{IIak(xk) — ak+1(xk)IJ}.
k
For convenience of notation we write K1 (x) K2 (x) for two piecewise continuous
fields Ki(x) and K2(x) if there exists a piecewise continuous unitary field u(z)
such that u*(x)Ki(x)n(x) = K2(x). Recall that tsr(A) = 1 means that invertible
elements are dense in the algebra A.
Further we are going to use the following result on almost commuting operators.
PRoPosITIoN 6.7.2 ([841). Let A be a with the properties
(i) RR(A) = 0 and tsr(A) = 1;
(ii) the unitary group of the pAp is connected for every projection
p E A.
Let h A be selfadjoint, let u E A be unitary and let the estimate
(6.25) IIu*hu — hil
hold. Then there exist a constant C depending only on thu and a pat/i u(t) con-
necting u with 1 such that for sufficiently small S, one has
IIu*(t)hu(t) — hit

for all t.
THEOREM 6.7.3 ([83]). Let (A(x), X, A) be a continuous field of trace
algebras over a segment or a circle. Let (B(x), X, B) be the corresponding continu-
ous field of W* -algebras and let A be the corresponding W* -algebra. Suppose that
for each point x E X, the A(x) has the following properties:
(i) RR(A(x)) = 0 and tsr(A(x)) = 1;
(ii) the unitary group of the C* -algebra pA(x)p is path-connected for any
projection p E A(x);
(iii) the trace on A(x) is finite and faithful for any x E X;
150 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

(iv) the homonzorphisrn Ko(A(x)) —p Ko(B(x)) induced by the inclusion

A(x) C B(x) L°°(A(x)) is injective and Ko(A(x)) is a sublattice in
K0 (B(x)).
Then the C* -algebra A A has the weak diagonalization property.
PROOF. We denote the algebra of (nxn)-matrices by Let us begin the
proof with the following lemmas.
LEMMA 6.7.4. Let Ki(x), K2(x) be two continuous fields of selfadjoint opera-
tors with a finite spectrum in the algebra such that Kk is unitarily equivalent
to k = 1,2. If IIK1(x) — K2(x)II <E, then for any x0 e X there exists
a closed neighborhood W of the point x0 and continuous unitary fields u;(X) in W
such that

(6.26) — <e.
PROOF. By assumption, we can write the operators Kk(x) in the form
Ki(x) = K2(x) =
where (x), qj (x) are spectral projections and the eigenvalues are ordered. Since
Ko(A(xo)) c Ko(B(xo)) is a sublattice by assumption, we can find projections
ri(xo) A(xo) such that =1 E A(xo) and each of the projections
Pm (x0), q3 (x0) and also each of the projections from ® 1 is unitarily equivalent
to a sum of certain projections rj(xo). By renumbering the eigenvalues of Kk(xo)
(counting them with multiplicities), we can write
Ki(xo) = K2(xo) =

Since in the ® B(xo) = ® L°°(A(xo)) one has

IIK1(xo) — K2(xo)II <E
(cf. we obtain
(6.27) — <E
for all 1. Decompose tile set of the projections {rj (x0) } into s subsets,

in such a way that the sum of all projections in each subset is unitarily equivalent
to a one-dimensional projection in 0 1. Then each subset corresponds to the
"eigenvalime" A(k) (xo):

= +... + ai,ri,(xo);
= +... + 81,r,(xo).
Choose unitaries E A(xo) such that = for li—i +1
I <li. Then the estimate (6.26) follows from (6.27). 0
LEMMA 6.7.5. Let D = [xk; xk+11 be a segment in X, and let A(x), E AID
be two continuous fields of selfadjoint operators with a finite spectrum on D such
that — ,i'(x)II < e. Let j.t E A(Xk) be a selfadjoint element with a finite
spectrum and let u A(xk) be unitary and satisfy
11/1 — u*1zF(xk)uII <5 and — <E.
 6.7. CASE OF CONTINUOUS FIELDS OF TRACE C-ALGEBRAS 151

Then for sufficiently small S and E there exists a piecewise continuous unitary field
u(x) such that
(i) U(Xk) = u, U(Xk+1) = 1;
(ii) the estimate
(6.28) IIu*(x)p/(x)u(x) — +S
holds on D, where C is a constant depending only on the norm of A(x).
PROOF. By assumption, we have — <2E. Therefore
IIP'(xk) — u*/f(xk)uII < + 5.
By Proposition 6.7.2, there exist a constant C depending only on the norm of A(xk)
(or of /i(xk) since these elements are close to each other) and a path uj, 0 t 1,
connecting u with the unit in the unitary group of the algebra A(xk) such that
— A(xk)II +S
for any t [0, Continuity implies that there exists a point e (Xk; xk+1)
such that there exists a unitary continuous field Ut(X) over the segment [xk;
such that it extends and the estimate
II?4(x)/1'(x)ut(x) — .X(x)II +S
still holds for all x e [xk; Define the unitary field u(x) by the formula

IUt(X) for t= k
, where x E [xk;x'k + 11'
1 for xE
It is a unitary continuous field and the estimate (6.28) holds on [xk; On the
other hand, if x then u(x) = 1, so this estimate is also valid there. 0
LEMMA 6.7.6. Let K(x) be a selfadjoint operator on the module K(x) E
®A. Then for any e > U there exists a piecewise continuous field K' (x) of opera-
tors with a discrete spectrum such that it locally lies in l!K'(x) — K(x)II <E
and the operator K'(x) is diagonalizable on ® A with piecewise continuous
"eigenvalues" with A) < 2E.
PROOF. Fix > 0 and choose a point xo E X. Since RR(A(xo)) = 0, we
can find an operator K'(xo) E ® A(xo) with a finite spectrum, K'(xo) =
ampm(xo) (where pm(xo) are spectral projections of K'(xo)) such that one
has IIK'(xo) — K(xo)II < There exists a neighborhood of the point x0 such
that the projections prn(xo) can be continuously extended in this neighborhood up
to projections pm(X). Put K'(x) = >m ampm(x). Then one can find a smaller
neighborhood of the point x0, where
(6.29) IIK'(x) — K(x)II <E.
So, for each point in X there exist a neighborhood W and a continuous field on W
such that the inequality (6.29) is valid. By taking a finite covering of X by such
neighborhoods, one can obtain a partition of X by segments and a piecewise
continuous field (continuous on each segment) K'(x) = e
AIDk, such that dist(K'(x), 0 A) < 2e. Let us diagonalize the operator field
K'(x) on each segment Dk, K'(x) = diag{A1(x)}. Then the fields A1(x) =
e AIDk, are piecewise continuous and, since — < 2E,
152 6. DIAGONALIZATION OF OPERATORS OVER

using Lemmas 6.7.4 and 6.7.5, we can replace the diagonal entries on each
interval by unitarily equivalent ones in such a way that the estimate
= sup Ack)(Xk)M <

is valid. U

LEMMA 6.7.7. Let Ki(x),K2(x) E be piecewise continuous fields of op-

erators with a finite spectrum and let IIK1(x) — K2(x)II <E, Ki(x)
K2(x) Suppose that their diagonal elements A2(x) and ,u2(x) are
piecewise continuous and A) < 8. Then there exist piecewise continuous
unitary fields u7(x), piecewise continuous fields and a piecewise continuous
field of operators such that
(i) — K2(x)II <
(ii) — <
(iii) — + cs,.
(iv) <S.
PROOF. Let x0 E X be a point of continuity for K2(x) and let D x0 be an
interval in X. By Lemma 6.7.4, we can diagonalize the operator K2(xo) in such a
way that its "eigenvalues" (x0) satisfy the inequality
— <e.
Let us extend these "eigenvalues" arbitrarily up to continuous fields (x). Then
the inequality
— Ai(x)II <E
isstill valid in some neighborhood of the point x0. By assumption, there exist uni-
tary elements w2 E A(xo) such that = Take unitary extensions
w2(x) of the elements w1(x0) = and choose a neighborhood of the point x0
in such a way that the estimate
— <
holds in this neighborhood. Thus we obtain a partition of X into segments Dk =
•

[xk;xk+1J, piecewise continuous fields (x), (x) E AIDk

and piecewise continuous unitary fields = E AIDk, such
that the estimates
(6.30) — <e
and
(6.31) *
— Iij(X) <6/2
hold on Dk. There also exists a piecewise continuous field of operators K2 (x)
unitarily equivalent to the operator (x) } and such that

It follows from (6.31) that at the point xk one has

(6.32) — < 6,
 6.7. CASE OF CONTINUOUS FIELDS OF TRACE C-ALGEBRAS 153

and (6.32) implies existence of unitary elements v1.k(xk) E A(xk) such that
— <&

By Lemma 6.7.5, we can find such extensions of the unitary elements Vj,k(Xk) to
some neighborhood of the point xk, for which, by (6.30), one has the estimate
— + 8.
Put
Pck)(X) =
Then
(6.33) — <8
and
—
Ai(x)M + 8.

Proceeding by induction and passing from Dk_1 to Dk, we substitute the fields
(x) by (x) on each Dk and we obtain new piecewise continuous fields (x) =
(x) } such that there exist piecewise continuous fields of unitary elements (x)
for which
— < 8.
By setting = we get statement (i) and the estimate
<8 follows from (6.33). 0
Now let K(x) be a strictly positive compact operator on the module HA. Take
a sequence > 0 vanishing as m oc. By Lemma 6.7.6, we can find a sequence
of piecewise continuous fields of diagonalizable operators Km (x) e ® A of
finite rank with "eigenvalues" Ai.m (x) E A such that
(i) IKm(X) — K(x)II
(ii) Km(X) diag{Aj,m(x)}, where dist(Ai,m(x),A) <Cm.
Since
IIK1(x) — K2(x)jI K(x)II + IK(x) —
IIK1(x) — <Cl +62,
one can find, by Lemma 6.7.7, a piecewise continuous operator
such that
— K2(x)II < — < e2, <
and
— <C</2(e1 + e2) + e2 + 62.
Then, since
— Km(X)M < — Km_i(X)Jj + IIKm_i(X) — K(x)II
+ IIK(x) — Km(X)II <2Cm_i + Cm,
we can find, by induction, a sequence such that
(6.34) — Km(X)II <Em, — <Em,
(6.35) <Cm,
(6.36) — +Em) +Cm +Cm.
154 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

Since, by (6.34), we have

— K(x)II < — Km(X)II + IIKm(X) — K(x)II
the sequence of operators K(x)
i the sequences are Cauchy sequences if the
numbers Em tend to zero sufficiently fast, namely if the series >m 1 + Em is
convergent. Therefore the limits
= urn
exist and, since, by (6.35), tends to zero, we conclude that E
A. Let us show that are the "eigenvalues" of the operator K(x). By The-
orem 6.4.2, it follows from the estimate — K(x)II < that there exist
unitary operators Urn E ® A such that they map the "eigenvectors" E Hk
of K(x) into the "eigenvectors" of and IIUnKnUm — Ku < 2Em. Put
Km = E ®A; Km K. Then
(6.37) KmXi
Passing to the limit in (6.37), we obtain Kx2 = x1A2.
6.8. Schrödinger operator as an operator acting on a Hubert
Let D be an operator (possibly unbounded) acting on a Hilbert space H.
Suppose that D commutes with operators U1,... , Then we can consider D
as an operator acting on a Hilbert over the A generated
by Uj,. .. , Let us assume that the operator D is positive and the operator
(1 + D) —1 is A-compact. These assumptions are usually satisfied in the case where
the operator D is elliptic in some sense. If the operators U1, . , .are normal and
.

commute with each other, then the algebra A is the C*.algebra of continuous func-
tions on the space X = Sp U1 x ... x Sp so the operator D defines the family of
operators x E X, such that the operators (1 are compact (in the usual
sense). Then we can diagonalize pointwise the family of operators and obtain
a family of eigenvalues E C(X). The spectrum of the operator D is deter-
mined by the spectra of the operators of multiplication by the functions In
the case, where the operators U1, . , U,2 generate a noncornrnutative
. .

the situation is similar. As an example, we consider in this section an operator

commuting with two unitary operators that generate a A9 of irrational
rotation, namely a perturbed Schrödinger operator with irrational magnetic flow
\2
(6.38) (i—±2irOy) ————+W(x,y)
j
with a double-periodic perturbation W(x,y) = W(x + 1,y) = W(x,y + 1). This
operator has been studied in a number of papers (see [76, 991). Applying to tile
operator (6.38) the Fourier transform in the variable x (x —+ and the change of
variables t = — + 9y, s = we obtain the operator

(6.39) D=A+W
with

(6.40)
 6.8. SCHRODINGER OPERATOR 155

and
W—

where (resp. T8) denotes the unit translation in variable t (resp. s), Tt s) =
+ 1, s), and Wkj denote the Fourier series coefficients of the function W(x, y).
We suppose that the function W(x, y) is sufficiently smooth to satisfy the condition
>.k,1 IWktI < 00.
Let A9 be the C0-algebra generated by two noncommuting unitaries U and V
such that UV = [14, 22] and let C A9 be its "infinitely smooth"
subalgebra of elements of the form >k1 aki V1, where the coefficients aki are of
rapid decay. The Schwartz space S(R) of rapidly decreasing functions on R can be
endowed with a structure of a projective right with one generator [22].
We denote this module by M°°. The action of on M°° is given by formulas
= + 9), =
where E M°°. The module M°° is generated by a projection p E M°°
with r(p) = 9 and, since M°° C A9, M°° inherits the norm from A9. Its
closure M = M°° A9 with respect to this norm is a Hilbert Ae-module. Note
that the eigenfunctions of the operator A (6.40) form an orthonormal basis (t) }
in S(R) C L2(R) and the functions from S(R2) = can be represented
as the series with m2(s) E M°°. Define the A9-valued inner product
on 8(R2) by the formula

where n3(s) E M°°. By S(R;M) (resp. L2(R; M)) we denote the Schwartz space
of functions (resp. the space of square-integrable functions) taking values in the
Banach space M. The inclusion
S(Rx R) "-
allows us to consider S(R2) as a dense subspace in the Hilbert A9-module
N= : (m1, rn) is convergent in A9}.

One can see that, since the A9 is simple and (M, M) should be its ideal,
the module M is full, i.e., (M,M) = A9. By Theorem 2.4.1, one has HA0.

THEOREM 6.8.1 ([81]). The operator D defined by (6.39) is a selfadjoint un-

bounded operator in N with a dense domain.
PROOF. The proof consists of the following five steps.
1. Denote by N1 C N the subspace of all sequences (m2) such that the series
is norm-convergent to an element of A9. If E =
then = — and the series — is convergent
in A9. Therefore A is an unbounded operator in N with the dense domain JVi.
2. Here we show that the action of the operator Ckj = can be extended
from M°° to M. Since this operator commutes with the action of the algebra
156 6. DIAGONALIZATION OF OPERATORS OVER

on the M°°, we have Ckj e MOC. The image of the generator

p of M°° can be written as CkI(p) = paki E M°° for some aki E and we have
Ckl(p) = Ckl(p2) = Ck,(p)p = PUkIP.
Obviously the map m paklpm = Ckl (m) can be continuously extended from
M°° to M. Besides that, since Ckl is a unitary operator, we have IICklII = 1 and
= 1.
3. Consider now the operator Bkl = It is obviously continuous in
S(R2). Let be the matrix coefficients of the operator with respect to
the basis
=

Since this operator is unitary, we get = Let = e N.

Then Estimate its norm:

(Bkl(e), = QfliCki(mfl))

= ,

i,n,j

= =

= =

< lIakLII2 = =

Hence IIBkIII 1 and it is a continuous operator on 1Sf.

4. We have
IIWII < IWklI IIBkiII IwklI.

By our assumption, the last sum is finite, hence W is continuous on .iV.

5. It remains to show that D commutes with the action of the A9 on
.N. This is clear for the operators and Bkl. Since the series W = >k,1 WkiBkl is
convergent, W also commutes with the action of A9. D is selfadjoint if the function
W(x,y) is real-valued. 0
Now let A be a type factor containing A9 as a weakly dense subalgebra (cf.
[141). This inclusion induces the inclusion of HA0 into HA and operators acting in
HA9 can be extended to operators acting in HA. Note that if I1WII <c, then the
operator D+c is invertible and its inverse = (1+1(W+c))1&1
is compact due to the compactness of the operator So, by Theorem 6.3.1,
it is diagonalizable in hence the same is true for the operator D. Slightly
changing the proof of that theorem (namely, taking 9 instead of 1 in (6.7)) we
 6.9. EXAMPLE: A CONTINUOUS FIELD OF ROTATION ALGEBRAS 157

can obtain the set of "eigenvectors" {x2} for D with (xi, = p. In that case
the corresponding "eigenvalues" can be viewed as elements of where
A.
Since the A9 is a of real rank zero [20], the "eigenval-
ues" )t, can be chosen from the smaller algebra by Theorem 6.6.2. If
0, then the spectrum of the operator D lies in the set U1(20(i — 1); 20i).
Therefore the spectral projections = P(29(j_l);29j)(D) lie in and the
standard methods of perturbation theory [80] can be applied. However the problem
of analyticity of "eigenvalues", which takes place in the commutative case for the
integer values of 0 [99, 107], is far from being understood.
6.9. Example: A continuous field of' rotation algebras
We give an example here of a without the weak diagonalization
property. Let X = [a, b] be a segment containing at least one integer point, for
example 1, and let S be a circle. Let the group Z of integers act on the
C(X x S) of continuous functions on the cylinder by the formula
(6.41) (o(n)f)(x, t) = f(x, t + nx),
where f(x, t) e C(X x S), x E X, t E 5, n E Z. By Ax we denote the cross-product
C(X x
As is shown in [112, 37], the algebra Ax is a continuous field of rotation algebras
= A(x) over X. Note that the continuous field (A(x),X,Ax) is a continuous
field of trace and that for irrational x one has RR(A(x)) 0 and
tsr(A(x)) = 1 [20, 111]. Moreover, the algebra A(x) has the weak diagonalization
property for each x e X. On the other hand, the unitary group of the
A(x) is not path-connected, but we believe that the obstruction to diagonalizability
is not related to this fact. Note that A1 C(T2) is the commutative
of continuous functions on a torus, so it is not a real rank zero algebra, unlike the
irrational case, and the homomorphism K0(A(1)) —, Ko(B(1)) is not injective.
There exists a nontrivial projection in M2 ® A1, which is the Bott generator in the
group K°(T2). This projection can be extended ([111, 78]) up to a continuous
field of projections p(x) in some neighborhood of the point 1 E X in such a way
that
(6.42) Tx(P(X)) 1+x
for the standard trace on We can diagonalize this field of projections in the
direct integral of type factors: p(x) where >
Then, for x = 1, we should have A1(1) = 1, = 0. For arbitrary x the
"eigcnvalues" (x), i = 1, 2, should be projections. If these "eigenvalues" are
continuous fields in Ax, then we should have Ai(x) = A1(l) = 1, )¼2(x) = =0
for all x, hence
= 1.
But since the trace is invariant with respect to the unitary equivalence, this contra-
dicts (6.42). Therefore the Ax does not have the weak diagonalization
property.
Note that on a dense set of irrational points in X, the A(x) satisfy
all the conditions of Theorem 6.7.3 except condition (ii). Slightly changing the
proof, for any given continuous field of operators K(x) and for any E > 0 we can
158 6. DIAGONALIZATION OF OPERATORS OVER C-ALGEBRAS

find a subset X such that the measure of the set X \ does not exceed e
and the operator field K(x) is diagonalizable on with "continuous" "eigenvalues"
(i.e., lying in the Axj.
 CHAPTER 7

Homotopy Triviality of Groups

of Invertible Operators
As before, denote by EndAl2(A) (resp. by End the Banach algebra of all
bounded A-homomorphisms of the Hilbert A-module 12(A) (resp. the
(cf. Section 2.1) of adjointable operators). Let GL(A) and GL*(A) denote the
corresponding groups of invertible operators. The question about contractibility of
general linear groups is very important for K-theory in order to construct classi-
fying spaces in terms of Fredholm operators. A series of papers is devoted to this
problem: [88, 61, 127, 89]. These results were used to construct the classifying
spaces for A) in [126] (cf. Section 8.5). In [30] Cuntz and Higson proved
contractibility of GL* (A) for any C*_algebra A with a strictly positive element (or,
equivalently, with a countable approximate unit, i.e.,
In this chapter, which is based on [130], we give another simple proof of the
theorem of Cuntz and Higson based on a generalized construction of a homotopy
from [97]. We also show that a similar argument can be applied to prove con-
tractibility of GL(A) in some special cases, in particular, for A a subalgebra of the
of compact operators in a separable Hilbert space and for A = G0(M),
where M is a finite-dimensional manifold.
The classical paper of Dixmier and Douady [33] gives a proof of the contract ibil-
ity of the unitary group of a Hubert space with respect to the strong topology. We
prove several versions of the generalization of this theorem to the case of GL(A)
and GL*(A) for a ti-unital A. There strict topology plays the role of
strong topology.
We complete this chapter by a detailed exposition of the modified Neubauer
homotopy.

7.1. Technical lemmas

It is known that the set of invertible operators in a Banach space is open with
respect to the norm topology, while the set of bounded A-homomorphisms is closed
in the set of all endomorphisms. Thus, GL is an open set in a Banach space. The
same argument is valid for CLt. According to Milnor's theorem [86], such sets
have the homotopy type of CW-complexes and, by Whitehead's theorem, strong
and weak homotopy triviality are the same for them. Thus, we have proved the
following statement.
LEMMA 7.1.1. To prove contractibility of GL (resp.. CLt) it is sufficient to
verify that any continuous map f: S —* GL of a sphere S of an arbitrary dimension
is homotopic to the trivial (single point) map Id E CL. A similar statement holds
for CLt.
159
160 7. 1-IOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Let us make one more reduction. In order to simultaneously consider the case
of GL and of GL* we introduce the following common notation: g := GL (resp.,
GL*), 6(M) := EIIdA(M) (resp.,
LEMMA 7.1.2 (a version of the Atiyah theorem about small balls). Let f: S —i
g be a continuous map of a sphere of arbitrary (but finite) dimension. Then f is
homotopic to a map f' : S —+ such that f'(S) is a finite polyhedron.
PROOF. Let> 0 be such that the E-neighborhood of the compact set f(S) lies
in a fine simplicial subdivision of the sphere S such that diani(f(o)) <
Choose
E/2 for any simplex a- of this subdivision. This is possible, since S is compact. Let
f' be a piecewise linear map extending the restriction of f onto the 0-dimensional
skeleton. Thus diam(f'(a-)) < diam(f(a)) <E/2 for any s. For any point s E S
there exists a vertex S such that If(s) — = Ill(s) — f(st)Il < r/2 and
Ilf'(s) — < r/2, hence the segment [f(s),f'(s)] c G for any point s S.
Therefore the linear homotopy It(S) tf'(s) + (1 — t)f(s) lies in Passing to a
subdivision of f'(S), we obtain the structure of a simplicial complex. 0
REMARK 7.1.3. Note that this argument is not valid for other topologies to be
considered. For example, with respect to the strong topology on the set of linear
operators in a Hilbert space, the sequence Id91 is convergent to Id, where has
the matrix diag(1, . 1,0,0,...) (unit up to the n-th place). Hence, the general
. . ,

linear group is not an open set with respect to this topology.

One more step taken from the original work of Kuiper [68] is universal.
LEMMA 7.1.4. Let V C be defined by
V={gEgIgIHl =Idff',g(Hi)=Hl}.
where
12(A)—H'EBH1,
Then V is contractible in G to 1 c.
PROOF. Let us decompose H' as

so that 12(A) = H1 H2 H3 s.... The matrix of g with respect to this decom-

position has the form
m(1, 1) = u = m(i,i) = 1 e i > 1, m(i,j) = 0, i
g=diag(u,1,1,1,...)=diag(u,u'u,l,u'u,l,...).
We would like to define a hornotopy gj G, t [0, -ir], in such a way that

For this purpose, put

mj(1, 1) = u
for t [0, ir/2],

( mt(2i,2i+1)
mt(2i+1,2i+1)
(cost —sint'\(u 0'\( cost 0
cost 0 1
 7.1. TECHNICAL LEMMAS 161

for i> 1, and

mt(r,s) = 0 for the remaining r, s.
For t e [ir/2, ir] put

( rnt(2i — 1,2i— 1) mt(2i — 1,2i)

1) mt(2i,2i)

(cost —sint'\(u' 0'\( cost sint'\(u 0'\

cost 0 1)'
and
mt(r,s) = 0 for remaining r, s.

0
LEMMA 7.1.5. Let W C be defined by
W={g EGIgIH' =IdH'},
where
12(A) H' ® H1, H' H1 12(A).
Then W is contractible inside G to
V = {g e = IdH', g(H1) = H1}.
PROOF. With respect to the decomposition 12(A) = H' H1 we define the
hornotopy by
., (1 /3(s)(1-t)
7(8)
I

Let the operator ( ) be the inverse for ( ). Then

eRo 1)'
(1

(1 (1
(1 (1 (1—t).0
1

Hence, the homotopy lies in G. 0

162 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Let 1CA, LM(KA) EndAl2(A), M(KA) and

QM(/CA) EndA(l2(A),l2(A)')
denote the of A-compact operators on 12(A). the algebra of the left miii-
tipliers, the C*_algebra of multipliers and the space of quasi-multipliers, respectively
(see [17, 73, 44, 63, 92, 1001 and Sections 5.3 and 5.4).
Let ci be a strictly positive element (see, e.g., Section 1.1) in a u-unital
algebra A and let := (ci) be a countable approximate unit, where the functions
i = 1,2,..., are defined by the following graph:

1/2 forz3andw2=ci2 1/2 . Then

71
j=i+2,i+3,...,
32 13 ,..., —

If A is not unital, the notion of the "standard basis" {e1} for the module 12(A)
makes no sense. Nevertheless, the elements ej'y are well defined for any -y E A,
namely,

Denote by Q, the orthogonal projections onto these one-dimensional submodules

LEMMA 7.1.6. The injection i: A 12(A) defined by

X >ek(i)wix, k(1) < k(2) < k(3) <

is adjointable and preserves the inner product. In particular, its image Tm i is

defined by the selfadjoint projection of the form
(7.2) p=
PROOF. First of all,

(ix, iy) ek(i)wiY) ek(i)wiy)

=
= >X*wjwjy = = (x,y).

Consider the operator t : 12(A) —+ A of the form

t(z) := = >WjZk(j).
 7.1. TECHNICAL LEMMAS 163

This series satisfies the Cauchy criterion: if tl1e number in is so large that

i=m+1
then
r r 1/2 r 1/2
< <1

The same argument for s = 1 implies that IIt(z)II lizil. Also one has (ix,z) =
(x, tz), i.e., t =
Consider arbitrary elements x. y E A. Then
(i*ix)*y = (i*ix y) = (ix,iy) = (x,y) =
x and = Id. Hence.
ii*ii* = ii*
i.e., p is a projection. Since = Id, i = Imp (see
also [71, Sect. 31). U

We need the following stronger version of this lemma.

LEMMA 7.1.7. The injection J: 12(A) 12(A) given by

(a1, a2,.. . ) (vjj, = E

{k(1, 1); k(1, 2), k(2, 1); k(1, 3), k(2, 2), k(3, 1);... } = {1, 2,... },
isadjointable and preserves the inner product. In particular, its image is defined
by the selfadjoint projection of the form
PROOF. Letx=(ai,a2,...)E12(A),y=(bi,b2,...)E12(A). Then

(Jx, Jy) = =

= = = (x,y).

In particular, J is bounded. Consider the operator T: 12(A) 12(A) of the form

T(z) := (t1,t2,. .
. ), :=

For this series the Cauchy criterion holds. Indeed, let a number N = N(z) be so
large that 11(1 — pN)zII < 5 and let m be so large that s(k(m,j)) > N (j is fixed).
Then (by Proposition 1.2.4)

. 11(1 PN)ZII <18.

164 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

By 1.2.4, the inequality

r r r r
= <

holds for any r, where qj is the orthogonal projection onto Hence

(T(z),T(z)) < (z. z).
So, T is bounded and it clearly is the adjoint for J.
The proof of the second statement literally repeats the argument from the
previous lemma. D
Consider an operator F E GL. Then, with respect to the standard decomposi-
tion of 12(A) into the direct sum of E1 A, the operator F has a matrix Fj with
entries from LM(A). If F E GL*, then E M(A), since =
LEMMA 7.1.8. For any F E GL*, any E > 0 and any there exists a number
iii(k) such that for any in ni(k) and any A with 1, one has
<C.
PRooF. Consider the bounded operator F*. Since F*eky e 12(A), there exists
a number m(k) such that
— Pm(k))1 ek','ll < C, < (in > m(k)).
Hence.
Fe?flp)lI = IIQmF8ekYlI . < (in > rn(k)).
0
7.2. Proof of the Cuntz—Higson theorem
LEMMA 7.2.1. Let Fr E GL8. r = 1 N. be arbitrary operators and let e > 0
be any number. Then we can choose increasing nonintersecting sequences i(k) and
j (k) of natural numbers such that
(7.3) 1(1 — < .2k' s= k, k + 1 r= 1 N,
4

(7.4) < .2k' s = 1,..., k — 1, r= 1 N.

PROOF. Put i(1) := 1 and choose j(1) > i(1) in such a way that
1 r
—pj(I))Frei(l)alIj < 21.21' r = 14
N.
Find i(2) > j(1) such that (by Lemma 7.1.8)
1
< r= 1,...,N.
Then choose j(2) > i(2) such that

11(1 — < 22 .2k' k = 1,2, r = 1...., N,

4
and i(3) > j(2) such that
1 r
II
< 2s .
s = 1,2, r= 1, . . . , N.
2
 7.2. PROOF OF THE CUNTZ-HIGSON THEOREM 165

Let us continue this process by induction. Suppose that i(1),.. i(k — 1) and . ,

,j(k — 2) are already found and satisfy conditions (7.3) and (7.4). Find
j(k — 1) > i(k — 1) such that
11(1 < rn = 1,...,k— 1, r= 1 N,

and after that find i(k) > j(k — 1) such that

1 e
2k' s = 1,... ,k
<_ — 1, r = 1,..., N.
By induction, we obtain the required statement. 0
Let us now define embeddings J and J' similarly to the constructions in
Lemma 7.1.7. To define J, we choose the vectors among that sat-
isfy g = g(s,j) > s +j, g > s; then = and Let us
similarly define for J', but taking instead of Ci(k) and denoting by h the
corresponding g. From conditions (7.3) and (7.4) we obtain that
II = II t)Ws, ej(h(n.m))ah(n.m)Wn) 1
II

II 1

h 1 N.

Denote by P and P' the corresponding orthogonal projections. Then PP' = P'P =
0. Let x = (ai, a2,...) and y = (b1, b2,..) be arbitrary norm one vectors from
12(A). Then, for any r 1,... ,N, by (7.5), (7.6) one has

II
(FrJX, J'y) = II

( +
t,s,n,m \h(n,m)g(8,t) h(n.m)<g(s,t)

2h(n,zn). 29(t,s)
t,8 ,n,m

since h(n, m) > n + m, g(t, s) > t + s by the construction. Therefore one has
(7.7) IIP'F'rPII <E, r = 1,. .. , N.
As was shown in Lemma 7.1.2 and Lemma 7.1.5, it suffices to know how to
construct a homotopy of any piecewise linear map with the image in a finite poly-
hedron in GL8 with vertices F1,. . , FN into a map taking values in a compact set
.

{D(x)} C GL8 such that

PD(x)=D(x)P=P VxES.
For this purpose we can apply a homotopy of Neubauer type (see Section 7.7). By
(7.7), we have to verify only that there exists an operator H0 : P'(12(A)) P(12(A))
such that operators H0P' and P are adjointable. Let us assume H0 =
166 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Then H0P' = = JJl* where Jl* is an isomorphism P'(12(A)) 12(A)

and J: 12(A) P(12(A)).
We have proved the following statement.
THEOREM 7.2.2 ([30]). Let A be a ci-unital Then GL*(A) is con-
tractible with respect to the norm topology.

7.3. The case A C AC

Let a A be a subalgebra of the algebra AC of compact operators on
a separable Hilbert space H. Under these restrictions we can prove the following
statement.
LEMMA 7.3.1. Let a, b E A, (fit 12,...) e Then

Ilafibli 0 (i co).

PROOF. Since a* E AC, for any E > 0 we can find a number N = N(e) and a
basis h1, h2,... in H such that

<
HN=spanc(hl,...,hN),
PN and are the corresponding projections on HN and Since the partial sums
of the series form a nondecreasing uniformly bounded sequence of positive
operators in 13(H) (see [42]), is strongly convergent to the zero operator.
Hence,
111h11 = (f.*h, f'h) = (f1fh, h) 0

for any h E H. Thus, is strongly convergent to 0. Let i0 be so large that the

estimate
IIIPNII
<
holds for i > io. Then
= +
< 2 . SUP hill
D

Note that similar properties for matrix elements themselves (which belong to
LM(AC) 8(H)) do not hold even for operators from GL*. Moreover, the following
example shows that all matrix elements can have norm equal to 1.
EXAMPLE 7.3.2 (A. V. Buchina, [105]). Consider the standard Hilbert AC-
module

We are going to construct an invertible operator F : L2(AC) —i with

matrix elements (with respect to the standard decomposition of 12 (AC)) satisfying
the inequality 1 for any i and j. These elements belong to LM (AC) = 8(H),
i.e., to the algebra of all bounded operators.
 7.3. THE CASE A C K 167

Let } be a basis of H and let denote the rank one projection onto the i-th
vector. Define F by the matrix
Pi P2 P3 P4 P5 P6 P7 P8
P2 P1 P4 P3 P6 P5 P8 P7
P3 P4 P1 P2 P7 P8 Ps P6
P4 P3 P2 P1 P8 P7 P6 Ps
Ps P6 P7 P8 P1 P2 P3 P4
P6 P5 P8 P7 P2 P1 P4 P3
P7 P8 P5 P6 P3 P4 P1 P2
Ps P7 P6 P5 P4 P3 P2 P1

where the first line is the ordered sequence of projections, the second line is obtained
from the first line by permutations in pairs P2i—1 and P2i for i = 1,2,..., the third
line is obtained from the first line by permutations of adjacent pairs, the fourth line
is obtained by permutations of pairs in the third line, the fifth line is obtained by
permutations of 4-tuples in the first line, and so on. Let us introduce the notation
and note that, for ii i2, one has for any j. Let us
show that this operator F satisfies all necessary conditions.
1. Let us prove that the image of (k1, k2, k3,..) E 12(IC) is compact for
the action of each line of the operator F, i.e., let us verify that the inequality
II Pa(j)kj < oo holds for each i.
1

For any 0 we can find N E N such that the estimate

2 /
n+p (n+p \ (n+p
=

n+p a+p
= < <E

holds for any n> N and any p e N and, by the Cauchy criterion, it follows from the
norm-convergence of k k3 that k3 is also norm-convergent. Thus, the
series is norm-convergent and its entries are compact operators, hence
is compact as well.
2. Let us verify that the image of a vector (k1,k2,k3,...) E 12(IC) under the
action of F belongs to 12 (IC), i.e., let us prove the norm-convergence of the series
*

We have
*

=
168 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Since II < oo, the inequalities

oo> =

Ilk.jxII2 = 2 = 2

hold for any x E H of norm 1. Hence, for any s > 0 and for any x with II.rlI
= 1,
one can find a number N(x) e N such that I(kji)a,U)12 < r for any
m> N(x) and any p E N. Thus,
/n+p \ n+p
=
j i7Z j
n+p n+p
= = <E.
i=fl j i=fl j
Consider the operators B = and = acting on
H. The first of them is compact and, consequently. the remaining ones are compact
as well, being the sums of norm-convergent series with compact entries. For a fixed
vector x E H of norm 1, the following statements hold:
(i) the inequality

(B7,x,x) = (Bx,x)

holds;
(ii) .x) = (Bx, x).
Put E,, := B — It is easy to see that
(i) —' 0 as ii 00;
(ii) =
(iii) E,L 0;
(iv) lIE1, II = II B — < 111311.

It follows from (i) and (iii) that 0 with respect to the strong topol-
ogy. By (iv), we obtain 0 with respect to the strong topology, due to the
joint strong continuity of multiplication in both variables on bounded sets. We
have proved the strong convergence of the increasing sequence of positive compact
operators B. Choose a finite-dimensional projection p
such that IIB(1 — < r and then choose n such that Il(B — < E for any
ni> n. Then, since the sequence increases,
lB — Bmll lI(B — + lp(B —

+ llp(B — Bm)pII + 11(1 — p)(B — Bm)(1 —

+ 11(1 —p)B(1 —p)II <4E.

311(B
Thus, the sequence of operators converges to B, i.e., lB — 0 as n 00. .'

Since IIBII <00, we have <00 and we are done.

3. By the construction of the operator F, it is clear that = 1 for any i
and j.
 _____________

7.4. SOME OTHER CASES 169

4. Note that F*F = F2 = FF* = Id. Therefore F is invertible.

By 1.—4. the operator F constructed above has all the necessary properties.
THEOREM 7.3.3. The group GL(A) for A C K: is contractible with respect to
the norm topology.
PROOF. Since Lemma 7.3.1 is the analog of Lemma 7.1.8, the proof can be
obtained by literally repeating the arguments from Section 7.2.
7.4. Some other cases
DEFINITIoN 7.4.1. Let us say that a A has property (K) if for any
functional f 12(A) A, any > 0 and any a E A, one can find a vector x E 12(A)
such that
(x,x) = a*a.
DEFINITION 7.4.2. A A has property (E) if for any functional
I = (fi ft,...) E (A) and any s > 0, one can find another functional
g (gi, . . . E and a number k E Z such that
i=k+1,k+2
and YILk : Lk —. A is epimorphism, where
EXAMPLE 7.4.3. Let A be the algebra of continuous functions on a smooth
n-dimensional manifold Al. Then A has property (E) (with k = n + 1).
To establish the relation between properties (E) and (K) in Theorem 7.4.5
below, we require the following lemma.
LEMMA 7.4.4. Let M be a Hilbert module, x E M, (x,x) > a > 0, hall 1.
Then one can find an element y=xb, Ilbil 1, such that (y,y) =a2.
PROOF. Put
/ 1\_h/'2
(x,x),
fl_00\ + —)
lim ('y b a.
flJ
This limit (with respect to the norm topology) exists, since
/ 1\_h/2 / 1\_h/2 / 1\h/2 f 1\_h/2
a2 (v+—) m
n m n
—1/2 —1/2 2
1 1

]
and
— (1 1\ 1 1

= =
for any nonnegative z. The estimate hlbhl 1 follows from
/ \1 /
\ nJ " nI
The condition (y, y) = a2 is now clear. 0
THEOREM 7.4.5. Property (E) implies property (K).
170 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

PROOF. Without loss of generality we can assume that hail = 1. Consider an

arbitrary functional f = (1's...) E and take some E > 0. Let g and k be
given by property (E) for E/2. Put f' := f Since 12(A), there exists, by
(E), a functional g' : A such that
iif'—g'ii<E/2, i=k'+1,k'+2,...,
and is an epimorphism. Then the functional

onLk,
on

satisfies the following conditions: ii! — hil < h is an epimorphism on Lk and

fl Lk' separately. Without loss of generality we can assume that h i} = 1. Let
x E Lk and y E fl Lk' satisfy h(x) h(y) = a. Then h(x — y) = 0 and, by 2.1.4,
a*a= (h(x),h(x)) ( (x,x), a*a= (h(y),h(y)) < (y,y).
By Lemma 7.4.4, one can find an element b such that libhl 1 and z = (x — y)b
satisfies (z,z) = a2. Thus h(z) = h((x — y)b) = 0 and, since = 1, one has
iif(z)iI<E.
REMARK 7.4.6. Let i and i' be adjointable embeddings preserving the inner
product and let the corresponding projections q = and q' = satisfy
(7.8) h!qq'iI <E, IIq'qII <E.
Note that qq' = where i is an isometric embedding and is an epimorphism
of norm 1. Therefore the inequalities (7.8) are equivalent to iii*ihII < E, < E.
Then the map 1= (i,i') : 12(A) e12(A) 12(A) is also an adjointable embedding
with the adjoint I*(x) = (j*(x) iF*(x)). Indeed, the operator 1* given by this
formula is continuous and
(I(x,y),z) = (i(x) +i'(y),z) = (xi*(z)) + (y iI*(z)) = ((x,y),I*(z)).
One also has
I*I(x y) = (i*(ix + i'y), + i'y)) = (x, y) + (j*jly iI*ix)
so that
(7.9) lId <2E
and 1*1 is invertible. Therefore I is an embedding. Note that, for invertibility, it
is sufficient to have E < 1/2.
THEOREM 7.4.7. Let a A have property (K). Then the group GL(A)
is norm-contractible.
PROOF. As above, one has to prove a statement similar to Lemma 7.1.8. In
the present situation we argue as follows. Let F1 be the first line (i.e., a functional)
of the matrix of F with respect to the standard decomposition of 12(A). Note
that any vector from 12(A), up to any 6 > 0 given earlier, belongs to for a
sufficiently large n = n(E). Hence, applying property (K), one can assume from
the very beginning that x E Applying property (K) infinitely many times with
 7.4. SOME OTHER CASES 171

constants decreasing as a geometrical progression, we can find a sequence of vectors

x. E 12(A) satisfying the following conditions:
(7.10)
E 12(A) = M1 ,
= ((0,... . . },

(7.11) (x2,x2) = is an approximate unit for A,

(7.12) <

Note that fork> k(i), one hasw2 = IId(i,k)II <1. Therefore, similarly
to the argument above, the map
J1 : 12(A) 12(A), (a1, a2,...) '—i > k(i,j))a3,

where
k(1, 1); k(1,2), k(2, 1); k(1, 3), k(2, 2), k(3, 1);...
is some increasing sequence, is an adjointable embedding preserving the inner prod-
uct. If we denote H1 := ImJ1, then, by (7.12),
IIF1IH1II <

Let C1 be the orthogonal complement to the image of the first copy of A under
J1. Let m(2) > m(1) := 1 be so large that 11(1 — Pm(2))F!J(1)II < e/2, where
Yi := Denote by F2 the restriction of the m(2)-th line of the
matrix F onto G1 12(A) and find, by the same algorithm, a new embedding J2
such that its image equals H2 and
11F21H211 <
LetC2 c H2 be the orthogonal complement to the image of the first copy of A
under J2. Let m(3) > m(2) be so large that
E 1/2
11(1 Pm(3))F'YiII < .2i'
.
i 1,2, Y2 J2(O!2 ,0,...),

11(1 Pm(3))YiII < j= 1,2.

Proceeding by induction, we obtain sequences m(j) and such that

E
(7.13) II(1Pm(j))FYiII<

(7.14) 11(1 Pm(j))YiIl <

<
(7.15) = 1,...,i.
Using again, we can construct an embedding J of the module 12(A) on a
submodule H of the linear span of y2 and an embedding J' of the module 12(A)
on the submodule H' := Em(j). Since these modules are e-orthogonal (i.e.,
< E for any h E H, h' E H', IhII 1, IIh'II 1), there exist projections
p and p' onto them such that pp' = p'p = 0. More precisely, note first that the
172 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

embedding J is adjointable. Indeed, the image of each vector (aj, a2,...) under J1
is a sum of the form
E Mk(2.3).

For constructing the higher J3 's, the corresponding lie again in direct sums of
modules Mr and, for these sets of modules are not intersecting. We can apply
Lemma 7.1.7. The operator J will be defined by the formula
(7.16) J: (al,a2,...) i—p = y3/13a3.

Hence, there exist the orthogonal projections q and q' onto H and H'. respectively.
Note that this argument can be refined as follows. In particular, we have shown
that for any J8 and any m there exists no more than one r such that QmJ3Qr 0.
Therefore, throwing out, if necessary, a finite number of canonical summands in
12(A) and restricting J8 to the remaining module, we can assume that
(7.17) Qm(j)Js = 0,
(7.18) Qm(j)Yi 0, j = 1,... ,j.
Also, IIqq'II IIq'qII <e. Indeed, consider a vector of the form

x=>>v:las=>ys/28a8,
It is necessary to show that Iq'xII <E. This estimate follows from (7.14), (7.18):

jq'xII = >>

>1 11(1 _Pm(s))YsIlsasIl + >i: =

Since the projections q and q' are selfadjoint, we obtain the second estimate as well.
Then, by Remark 7.4.6, HeH' is the image of an adjointable embedding and,
by Theorem 2.3.3, the decomposition 12(A) = (Hi- fl holds. Denote
by p and p' the projections onto H and H', respectively, which correspond to this
decomposition, so that pp' p'p = 0. Thus,
(7.19) IIp—qII <&, IIp'—q'II < 3E, < 1+3e <2, < < 2.

Indeed, let x lixil = 1. Then x = for some y and, by (7.9),

IIIyII Therefore

— q)xII = — q)(iily + iIiI*Iy)II

= II(r — q)(q + = II — 2E(1 + e) <3g.

Besides, IIp'FpII < 7IIFIle. In fact,

lIp'FpII = — q')Fp + q'FpII <3EIIFII +

 7.5. DIXMJER-DOUADY THEOREM FOR I2(A) 173

and, by (7.19), it is sufficient to prove that for x E H, II.rII 1. the estimate

I
q'Fxii <2E holds. Any such vector x can be presented as
= <1.

Then

iq'FxiI =

— + +=
Note that one can obtain a similar statement not only for a single operator
F (actually for two operators; F and Id), but also for a finite set (vertices of a
siinplicial complex) . .
, F(N). For this purpose, one has to argue for F =
with a constant e and to obtain projections P1 and After that one has to apply
the algorithm to and to obtain projections and P2 such that
P1P2 = P2P1 = P2, P1 P2 = P2P1 = P2, P2 P1 = P1 P2 =0,
<e,
and so on. This completes the proof, since now one can apply the Neubaucr homo-
topy. U

Let us complete this section with a discussion about the following example,
which gives no hope that the class (K) is sufficiently large.
EXAMPLE 7.4.8. Under the conditions and notation of Example 2.5.6 the func-
tional

does not have property (K).

REMARK 7.4.9. Characterization of topological spaces X such that C(X) has
property (K) was obtained in [106].

7.5. Dixmier-Douady Theorem for 12(A)

Let us consider 12(A) as the completion of the algebraic tensor product H® A
L2([0, 1]) ® A with respect to the A-inner product (f ® -y, g 0 8) = (f, g) We
assume here that the inner product in L2([0, 1]) is linear in the second entry.
LEMMA 7.5.1 ([33, p. 250]). There exist, for each t E [0, a closed linear
subspace H and, for each t E (0, 1], a linear isometry Ut : H such that
(i) the map t where is the orthogonal projection onto is strongly
continuous;
(ii) the maps t UtPt and t are strongly continuous;
(iii) H1 = H, H0 = 0, U1 = 1.
174 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Recall that in [331 the subspaces are defined in the following way:
:={IEL2([O,1I)If(x)=O for xt}.
LEMMA 7.5.2. If F(t) is bounded and if F with respect to the strong
topology on B(H), as t —+ 0, then ØIdA —4 F®IdA with respect to the left strict
topology.

PROOF. It is sufficient to prove that

II(Ft ® —F® IdA)Ox,yII 0 (t 0),
where
=
= x(y, z), x ® xi E C, E A, lxii = = 1,

and {h2} is an orthogonal basis of H. Then, for z = h2z. ® one has

ii(Ft ®IdA = — <E

if t is so close to 0 that
ii(Ft — < E

LEMMA 7.5.3. Let a set {C(t)} be uniformly bounded (by a constant C) and
let G(t) —p G and S(t) S (t 0) with respect to the left strict topology. Then
G(t)S(t) CS (t 0) with respect to the left strict topology.
PROOF. Let k 'CA be an arbitrary operator. Then 5k E 'CA and
— SkiI 0, ii(C(t) — G)(Sk)iI 0 (t 0).

Hence
iiG(t)S(t)k — GSkii ii(G(t) — C)Sk + G(t)(S(t) — S)kii
Ii(G(t) — C)Skii + CII(S(t) — S)kii 0 (t —' 0).
0
THEOREM 7.5.4. The unitary group U of operators in 12(A) is contractible with
respect to the left strict topology.
PROOF. For any U E U and t E (0, 1] we define
t) := (1d12(A) —Pt ® IdA) + ® Ida) U (U1 ® IdA) ® IdA)
and
U.
The operator 1), t e (0, 11, defines the identity mapping ®A and coincides
with the restriction of the unitary map (U1 ® IdA) Olito H1 ® A.
Therefore

Thus, since U' = Ut', all operators are adjointable.

 7.6. SOME GENERALIZATIONS 175

It is clear from Lemma 7.5.3 that 1 is continuous in t e (0, 11 and, similarly,

in (U, t). Indeed, let (U', t') U x (0, 11 tend to (U, t) e U x (0, 1]. Then, for any
k e K:n, one has
t)k — t')kII
=II(1d12.A)—Pt®IdA)k+(Ur'®IdA)U(Ut®IdA)(Pt®IdA)k
— (1d12(A) —Pt' ® IdA)k — ® Ida) U' (Ui' ® IdA) (Pt' 0 IdA)kI[
—Pt, ® IdA)k[I
+ 0 Ida) — 0 ® (Pt® IdA)kII
+ ® Ida) [U — U'] ® IdA) (Pt ® IdA)kII
+ ® 1d4 U' [(Ui ® IdA) — (Ui' ® IdA)] (Pt ® IdA)k[)
+ ®IdA)U'(Ut' 0 IdA) [(Pt 0 IdA) — (Pt' ®IdA)]kII
®IdA)kII
+ ® Ida) — 0 Ida)] k1 + U']
II[(Ut ® — (Ui' o IdA)] II[(Pt ® Ida) — (Pa' 0 IdA)]k[t 0

by Lemma 7.5.2. Here k1, k2 and k3 are fixed operators from KA. Now let (U', t') E
U x (0,1] tend to (U,0) E U >< [0,1]. Then Pt' —+ 0 with respect to the strong
topology and ®IdA 0 with respect to the left strict topology by Lemma 7.5.2.
Therefore, for any k e )CA,
0, —+0.

7.6. Some generalizations

Note that in the proof of Theorem 7.5.4 we used only the boundedness of the
set of invertible operators {U} but not its unitarity. Thus, we have actually proved
the following statement.
THEOREM 7.6.1. Every bounded set of invertible operators in a Hubert space
H is contractible with respect to the strong topology in the set of invertibles.
Every bounded set of invertible operators from GL (resp. GL*) is contractible
in GL (resp. GL*) with respect to the left strict topology.
LEMMA 7.6.2. Let S be a compact set and let
f:S—+B(H), si—+F8,

be continuous in the strong topology. Then is bounded.

PROOF. Since S is compact, the set is bounded by some C(x) for any
x E H. Therefore, by the uniform boundedness principle [35, 11.3.21], there exists
a constant C such that
C, Vs e S, .r e B1(H).
Therefore <C. 0
176 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

LEMMA 7.6.3. Let S be a compact set and let

f:S EndAl2(A) = s F3,

be continuous in the left strict topology. Then {11F311} is bounded.

PROOF. Let x E 12(A) be au arbitrary element. Choose k E and z such that

x = kz (see Lemma 2.2.3). Then s F3x is continuous: we apply the definition of
the left strict topology to the inequality

IIF3x — FtxI( = IIFskz — FtkzII < IIF3k — FtkII IIzII.

Tile proof can be completed similarly to the proof of Lemma 7.6.2. 0

Now, from Theorem 7.6.1 by Lemma 7.6.2 and Lemma 7.6.3, we obtain the
following statement.

THEOREM 7.6.4. The group G(H) of invertible operators in a filbert space H

is weakly contractible (i.e., the homotopy groups ir2(G(H)) = 0) with respect to the
strong topology.
The group GL (resp. GL*) is weakly contractible with respect to the left strict
topology.

REMARK 7.6.5. We suppose that the results of this section in the part con-
cerning Hilbert spaces were known earlier but we have not found them published
anywhere.

7.7. Neubauer type homotopy

In this section we describe how to modify the homotopy from [97] for our
purposes. Although we work with completely different objects, the construction in
[971 is so universal that the proofs can be transferred almost without modifications.

LEMMA 7.7.1. Let M be a Hubert A-module, X a topological space, T : X —+

Q = G(M) a continuous map and P and P' projections from e = E(M) such that
PP'=P'P=O,
H0p'Ee, P'T(x)P=O VxEX.
Then there is a homotopy T D in Q such that

PD(x) = D(x)P = P Vx E X.
PROOF. Put Q := Id —P, Q' := Id —P',

P(x) := Q(x) := Q' — P(x).

Then P(x) is a projection on T(x)PM and there is a decomposition into projec-

tions. Id = Q(x) + P(x) + P', and the projections Q(x), P(x) and P' are such that
 7.7. NEUBAUER TYPE 1-IOMOTOPY 177

the composition of any two of them vanishes for any x. Indeed,

Q'T(x)P = (Id —P')T(x)P = T(x)P,

PT(x)'Q'T(x)P = P, T(x)PM C P(x)M C T(x)PM,
P(x)P(x) = T(x)PT(x)' (Id —P')T(x)PT(x)1 (Id —P')
= = T(x)PT(x)'(Id—P') 2(x),
Q(x) + 2(x) + P' = Q' - 2(x) + P(x) + F' = Id,
P(x)P' = T(x)PT(x)'(Id—P')P' = 0,
P'P(x) = P'T(x)PT(x)'(Id—P') = 0.
So 2(x) + P' is a projection, whence Q(x) = Id —(2(x) + P') is a projection too.
Define
H = -H0P' +
Then, since P'P = PP' = 0, one has P'H0 = = 0 and

H2
HP'H = (H0P' + +
= = -F,

Q'HT(x)'P(x) = = 0,

P(x)T(x)HP' = T(x)PT(x)'Q'T(x)HP'
= T(x)PT(x)'(1 — P')T(x)(—HoP')
= T(x)PT(x)' (1 — P')T(x)P(—HoP')
= T(x)PT(x)'T(x)P(—HoP') = T(x)P(—H0P') = T(x)HP'.

Assume that
G(x) := HT(x)'P(x) + T(x)HP'.
Then

G(x)2 = (HT(x)'7(x) + T(x)HP')(HT(x)'P(x) + T(x)HP')

= + T(x)(—P)T(x)'P(x)
+ HT(x)'7(x)T(x)HP' + T(x)HP'T(x)HP'
=

HT(x) T(x) HP'T(x)HP'

= 0 — — F' + T(x)(—H0P')T(x)(—HoP')
=O—2(x)—P'+O=—(P'+P(x)),
G(x)Q(x) = 0,
178 7. HOMOTOPY TRIVIALITY OF GROUPS OF INVERTIBLE OPERATORS

Q(x)G(x) = (Q' — P(x))(HT(x)'P(x) + T(x)HP')

= Q'HT(x) + (Id —P')T(x) (—PH0P')
— P(x)HT(x)'P(x) — P(x)T(x)HP'
= 0 + T(x)HP' — (T(x)PT(x)'Q')(—HoP'
+ '(T(x)PT(x) -1 — T(x)HP'

=
(T(x)PT(x) - 'Q') = 0.
Hence, for
U(s,x) := Q(x) + (1 — s)(P(x) + P') + sG(x),
we obtain that
U(s, x)' = Q(x)
+ + (1— s)2
[(1 — s)(P(x) + P') - sG(x)].
Therefore U(s, x)T(x) defines a homotopy in
U(0, x)T(x) = Id oT(x) 'S-' U(1, x) o T(x).
Thus, since P(x)T(x)P = T(x)P, one has
U(1, x)T(x)P = Q(x)P(x)T(x)P + G(x)P(x)T(x)P
=0+ HT(x)'P(x)T(x)P = = HP.
Since H(P + P') = (P + P')H = H, for
V(s) := QQ' + (1 — s)(P + P') — sH,
we have
V(s)' = +
s2 + (1 — s)2
[(1 — s)(P + P') + sH].

Besides, V(0) = QQ' + P + P' = Id. Therefore the following homotopy is defined:
R(x) := V(1)U(1,x)T(x) U(1,x)T(x) in C(X,g(M))
and
R(x)P = V(1) U(1, x) T(x)P
=V(1)HP=QQ'HP-H2P=O+(P+P')P=P.
Put
R(s, x) := R(x) — sPR(x)Q.
For some e e M, let the equality R(s,x)e = 0 hold. Then
o = R(s, x)e = R(x)(P + Q)e — sPR(x)Qe = Pe + R(x)Qe — sPR(x)Qe,
0 = QR(s,x)e = QR(x)Qe.
Let f = PR(x)Qe, whence f = Pf. Then
PR(x)(Qe — P1) f — P1 = 0, QR(x)Pf = 0.
Therefore R(x)(Qe — Pf) = 0, Qe = Pf = f=0 and PR(s,x)e = Pe = 0, e = 0.
One also has
R(x)M = M, R(x)P = P, QR(x)QM QR(x)(1-P)M = QR(x)M = QM.
 7.7. NEUBAUER TYPE HOMOTOPY 179

Therefore, with respect to the decomposition M the operator R(s, x)

has the matrix
QR(x)QM =
QR(x)Q)'
Hence, R(s, x) is an epimorphism and R(s, x) belongs to as an epimorphism
without kernel. It remains to put D(x) R(1,x). D
LEMMA 7.7.2. Let M be a Hubert A-module, X a compact set, T: X —+
a continuous map with 0 < e < mm IIT(xY'1L1 and P and F' projections from
= E(M) such that
IIP'T(x)PII <E Vx E X.
Then there exists a homotopy S(s, x) in such that
S(0,x)=T(x), P'S(1,x)P=0 VxEX.
PROOF. Put S(s,x) := T(x) — sP'T(x)P. Since
IIS(s,x) — T(x)II
one has S(s,x) E 0
 CHAPTER 8

Hubert Modules and KK-Theory

In this chapter we formulate some definitions and theorems related to the
Kasparov bivariant KK-functor. The details and the discussion of different ap-
plications can be found, e.g., in the original paper of Kasparov [64] and also in
[65, 119, 28, 55, 120, 3, 132, 24, 51, 114, 8, 9, 98]. A more detailed bibli-
ography is contained in [391. Besides the original approach of Kasparov, we also
discuss the Cuntz approach [28, 29], which is very useful in several situations.
We briefly present some generalizations of Kapsarov A, B-birnodules (triples) and
discuss some applications of this approach. In our presentation, we mainly follow
L65, 120, 55, 3]. As before, we deal with complex although sometimes
(cf. [113]) the real approach is more useful [117].
Let us emphasize that starting at some point we shall be forced to restrict
ourselves to the case of separable
An application of the Kuiper type Theorem 7.2.2 to the construction of clas-
sifying spaces for K(X; A) by means of A-Fredholm operators is discussed at the
end of the chapter. We consider the simplest case. (For more general situation
see [126, 1211.) Relation to KK(C,C(X) 0 A) is indicated (details can be found
in [30, 89, 134]).

8.1. Tensor products

We need the following notions of tensor products of Hilbert
DEFINITION 8.1.1. Let be and let M2 be Hilbert B2-modules,
i= 2. Let B1 ®
1, be the product (cf. Remark 5.6.3). The exterior
tensor product of modules M1 ® M2 is the completion of the algebraic tensor
product over C (which is a B1 ® B2-module) with respect to the norm defined by
the B1 0 B2-valued inner product
(e1 ®e2, Ii 012) = (ci, e2)®(fi, 12), e2,f2 E
DEFINITION 8.1.2. Let A and B be two M1 and M2 Hilbert
modules over A and B, respectively, and p: A a *..homomorphism.
Consider the degenerate B-valued inner product
(8.1) (ei ® fi ® 12) = (e2, p((e1, 11)) ft E
Let N be the submodule of isotropic vectors in the algebraic tensor product (M1 ®alg
M2). Then the completion of the quotient space (M1 ®ajg M2 ) /N with respect to
the product (8.1) equipped with the obvious Hubert structure is called
the interior tensor product M1 0,, M2. The action of B is defined by the formula
(ci ®e2).b=ei®e2 .b; also, ei®p(a)e2 =e1
181
182 8. HILI3ERT MODULES AND KK-THEORY

8.2. Main definitions

REMARK 8.2.1. To explain the motivation for the conditions in the definition
of KK-groups. note that this definition is close to the definition of Atiyah's functor
Eli or, more precisely, to the idea of representing elements of K-homology by
elliptic operators. An elliptic pseudodifferential operator acting in Sobolev spaces
on a manifold can be distinguished among all bounded operators in some abstract
Hilbert space by two features: it is Fredholm and it almost commutes with the
operators of multiplication by functions. Evidently, to state this condition, the
Hilbert space should be equipped with an appropriate representation of the algebra
of functions on the manifold. Passing to a noncommutative algebra A instead of
the algebra of functions and to a Hilbert B-module instead of a Hilbert space,
we obtain the prototype for the notion of KK-groups and a motivation for the
following definition.
DEFINITION 8.2.2. Let A and B be An A, B-bimodule is a pair
(M, ir), where M is a Z/2Z-graded Hilbert B-module equipped with an action
of the A given by a *-homomorphism ir : A End* (M), where the
operator ir(a) is homogeneous of degree 0 for any a E A: ir(A) C End*(M)(O).
Denote by E(A, B) the set of all triples (Kasparov bimodules) (M, ir, F), where
(M,ir) is an A,B-bimodule and the operator F e End*(M) is homogeneous of
degree 0, such that the conditions
(i) ir(a)(F2 — 1) E K(M),
(ii) the commutator [ir(a), F] E
hold for each a e A. A triple (M, ir, F) is called degenerate if the conditions
(i) ir(a)(F2 — 1) = 0,
=0
(ii) [ir(a), F]
hold for each a e A. The set of all degenerate triples is denoted by D(A, B).
A homotopy in E(A, B) is an element of E(A, B[0, 1]), where, as before, B[0, 1]
denotes the of all continuous B-valued functions on the unit interval.
Define addition on E(A, B) by setting
(M, ir, F) + (M', ir', F') = (M
M', ir ir', F F').
Define KK(A, B) to be the set of homotopy classes of elements of E(A, B).
This set turns out to be an Abelian group. It is easy to verify that any de-
generate triple is homotopic to the triple (0,0,0) and that the inverse element for
(M, ir, F) equals (—M, ir, —F), where —M denotes M equipped with the inverse
grading.
REMARK 8.2.3. A natural question arises: why do we consider operators from
End* (M), and not from End(M)? The answer has two sides: on one hand, it is
convenient to deal with adjointable operators, on the other hand, any pseudodif-
ferential operator where the symbol is a homomorphism of bundles of projective
modules over a "nice" is adjointable. The details can be found in [121].
A "nice" A means here that LM(A) = M(A).
In the sequel we assume to be separable and Hilbert to
be countably generated. In some cases the first assumption, with a lot of difficulties,
can be replaced by a weaker one, but this is not important for us here since only
separable algebras are used in applications: the (reduced) group of
 8.2. MAIN DEFINITIONS 183

finitely presented groups (fundamental groups of manifolds) and commutative

algebras of continuous functions on some "nice" spaces.
For a *-homomorphism f: A1 A2, define the map
f*:E(A2B)_,E(A1B)
and, for a *-homomorphism g: B1 —+ B2, define the map
(M,ir,f)i—4(M®9B2,ir®l,F®l),
where the tensor product is understood in the sense of Definition 8.1.2 and where
ir ® 1 is defined by the formulas
ir®1 :A_4End*(M®gB2), (ir®1)(a)(e®b)=ir(a)e®b.
By definition, this immediately implies the following statement.
THEOREM 8.2.4 ([641). The groups KK(A, B) define a homotopy invariant bi-
functor from the category of separable to the category of Abelian groups,
which is covariant in B and contravariant in A. One has KK(C, B) = K0(B).
To define the higher KK-groups, one can either use representations of Clifford
algebras or one can set

We require the following definition in order to introduce the basic technical tool
of KK-theory, namely, the Kasparov product.
DEFINITION 8.2.5. Define by 1AEKK(A, A) the class of the triple (A,
A, = 0,

ZA : lC(A) C End*(A), iA(a)b = ab, a,b E A.

Define also the map
: KK(A, B) -# KK(A ® D, B® D),
TD

TD(class(M, ir, F)) = class(M ® D, ir® lD, F® 1).

THEOREM 8.2.6 ([64]). There exists a well-defined pairing (Kasparov product)
KK(A,D) x KK(D,B) -+ KK(A,B),
denoted by (x, y) x ®D y, which enjoys the following properties:
(i) the pairing is covariant in B and contravariant in A;
(ii) if f: D E is a 'i'-homomorphism, then
f*(y), x E KK(A, D), y E KK(E, B);
®E Y = X ®D
(iii) the product is associative:
(xØDy)®Ez "X®D(y®EZ)
Vx E KK(A, D), y e KK(D, E), z E KK(E, B);
(iv) TE(x ®B y) = TE(X) ®B®E TE(y) Vx E KK(A, B), Vy KK(B, D);
(V)x®B1B=1A®X=x VxeKK(A,B).
Due to these properties, the product can be generalized to a more general
setting.
DEFINITION 8.2.7. For x KK(A1, B1 ® D) and y E KK(D 0 A2, B2) put
X Y TA2(X) ®Bj®DØA2 TBI(Y) E KK(A1 ® A2, B1 ® B2).
184 8. HILBERT MODULES AND KK-THEORY

The original proof of this theorem and even the construction of the product
itself is very complicated. That is why we will pass now to the presentation of the
main aspects of the Cuntz approach to KK-theory [28]. This approach allows us
to consider elements of KK(A, B) as generalized homomorphisms from A to B and
to define the product as their composition.
8.3. Cuntz's approach
First of all, let us note that for each class in KK(A, B), one can choose a
representing triple (M, ir, F) such that F2 = 1. Indeed, for the opposite B-module
(—M), define the map
ir':A—9MED(—M), ir'(a)=ir(a)EDO,
and define F' E End* (M ED (—M)) by the formula
= + U(1 —

—
F

U U 1 is the canonical identification. Then it is easy

to check that the class of (M ED (—M), ir', F') coincides with the class of (M, ir, F)
and that F'2 = 1.
Moreover, for a class of (M, ir, F) with F2 = 1 one can assume that F =
F*. Indeed, put F' = F(FF*)l/2. Since FF* = (F*F)_l, one has =
(F*F)_h/2 and the operator F' is invertible. Since
= F(FF*)F* =1,
FFFF*

the operator F' is unitary. Then, for any polynomial f, we have Ff(F*F) =
f(FF*)F. Using approximation of the function x '—' by polynomials on the
spectrum of F*F, we obtain the equality
F' = = = F'-' =
Under this assumption, = F(FF*)t/2, t 10, 1], defines an operator homotopy
between the triples (M, ir, F) and (M, ir, F'). Namely, the bimodule in an operator
hoxnotopy is fixed and a family of operators is norm continuous. This is a stronger
equivalence relation on E(A, B) than homotopy, but, nevertheless, one can define
KK using operator hornotopy.
So, we assume that for (M, ir, F) E E(A, B), the equality F = F* =
holds. In this case one can identify and using F. Thus, we arrive to
a description of a representative of an element of KK(A, B) as a trivially graded
Hilbert B-module equipped with two representations, and 1r, of the
A:
R.+(a)
= lr(a)IM(o), ir(a) = Fir(a)FIM(1).
The Kasparov stabilization Theorem 1.4.2 allows us to restrict ourselves only to
the module HB = 12(B).
DEFINITIoN 8.3.1. Let A and B be A quasi-homomorphism from
A to B is a couple of *-homomorphisms ir± : A End* (HB) such that
7r(a) E B® IC I((H8)
holds for any a E A, where IC is the algebra of compact operators on a separable
Hubert space.
 8.3. CUNTZ'S APPROACH 185

So, KK(A, B) is the set of homotopy classes of quasi-homomorphisms from A

to B [271.

For further constructions we need the qA defined by using free

products of Such products can be defined in two ways. We shall
indicate one of them and present in detail another one. The first approach is the
following. Let A and B be and let A *alg B be their algebraic free
product. It is a *-algebra equipped with two canonical *-homomorphisms

For any D and for any pair of *-homomorphisms : A D,

B D, there exists a unique *-homomorphism i,1') : A *alg B —' D such that

Let
IxII= sup XEA*aigB.
D,
The corresponding completion of the algebraic free product is a This
algebra is denoted by A * B and is called the product of A and B.
This free product can be defined alternatively in the following way (see [55,
Appendix B] and [21]). For each n = 1,2,..., consider the following algebraic
tensor products:
A®cB®cA®c B®cA®c n tensor factors,neven,
ntensorfactors,nodd,
ntensorfactors,neven,
ntensorfactors,nodd.

Consider the following direct sum of vector spaces:

Co := Ba).

For the purpose of obtaining multiplication in C0, define the bilinear maps
X Bm ' An+m_i, n even, m even,
X Bm An+m, n odd, m even,
X Am ' Bn+m_i, n even, m even,
X Am Bn+m, n odd, m even,
X Bm 4 An+m_i, n even, m odd,
X Bm An+m, n odd, m odd,
X Am —I Bn+m_i, n even, m odd,
X Am Bn+m, n odd, m odd,
X Am An+m_i,) n even, m any,
X Am An+m, ri odd, rn any,
X Bm 4 Bn+m_i, n even, m any,
X Bm Bn+m, ) n odd, m any,

by setting, for the first row,

(ai®b2®as®•..®bn).(bi®a2®b3®..•®am)
186 8. MODULES AND KK-THEORY

The remaining maps are defined in a similar way. Define the bilinear map

by the formula

(x1,yi,x2,y2,...) := +yj
i,j=1

where x2, E A2, Ys, B2. Then one can check that this bilinear map equips
C0 with the structure of an algebra. Define the involution * on elementary tensors
as the operation that transposes tensor factors in the inverse order and takes their
adjoints in A or B. For example, (a 0 := 0 a*. Then, for an even n, one
obtains maps * : * : and for an odd n, one obtains maps
*: —÷ *: This turns C0 into an involutive algebra with respect
to the map
* : C0 —' C0,
I \* I * * * * * * * * \
.=1Sx1,yl,y2,x2,x3,y3,y4,x4,...),
where x2 E A2, y2 E
Let CpA : A —+ D and 'PB : B D be two *-homomorphisms. Define the map
D on for even n on elementary tensors by the formula

:= 'PA(al)'PB(b1)'PA(a2)'PB(b2) . . .

In the three remaining cases the map can be defined in a similar way. Then one
can define the following *-homomorphism:
Xn EAR,

In particular, *-representations of C0 on a Hilbert space H can be obtained in this

way. Consider all representations ir: C0 8(H) (not necessarily of the mentioned
form) and put
lixil := sup IIir(x)Il, x E C0.

Each representation ir generates *-representations of A and B:

lroiA:A—).B(H),
lrozB:B-4t3(H),
Then one concludes that the norm fi is finite on C0 and that IIiA(a)II = hail and
IIiB(b)hl = hlbU. Put N := {x E C0 hixhl = O}. Then N is a two-sided *-ideal. Let C
I

be a completion of the quotient algebra C0/N with respect to the norm . One
can easily verify that C is a and, since and are isometrics, the
maps
a—'ZA(a)+N, 3B:B—'C, b'—'iB(b)+N,
are isometric embeddings. Put A * B := C. The universal property, i.e., exis-
tence of the homomorphism (cf. the first approach to the definition), is now
completely evident.
 8.3. CUNTZ'S APPROACH 187

CONSTRUCTION 8.3.2. For a A, let QA be the product A*A and let

denote the two canonical embeddings A QA, namely i :=
for A = B. Denote by qA the ideal of QA generated by all the elements of the form
{q(a) := — i(a)Ia E A}. It is evident that

qA = = (IdA,IdA) QA : A}
and we have the short exact sequence

where any of the two embeddings, and i, splits More precisely, this follows
from the following statement.
LEMMA 8.3.3. Elements of the form
qa1 qa2 ... and qa2 ... qan, a2 E A,
generate a dense involutive subalgebra C in qA and qA = Ker
PROOF. Since (IdA, IdA)(qa) = 0, one has qA C Ker It remains to show
that C is dense in Ker
Denote by C0 the linear span of elements of the same form as for C and also
of elements of the form i+(a). Then C0 is a dense involutive subalgebra in QA.
Indeed,
q(ab) + q(a)q(b) = — F(ab) + — — i(b))
= i(ab) — —

= — —

+ = — + —

Hence,
= q(ab) + q(a)q(b) —
Since = — q, we obtain, by induction, that C0 = A *alg A C QA.
Suppose that x e Ker C QA is an arbitrary element. Then x = xi,,
e C0. Hence, there exists the limit
y=lim{x,, C,

— o = — = 0,
so y e Ker and
0= — y) = lim o o = lim

x=y.
0
Thus, any quasi-homomorphism ir± from A to B defines a *-homomorphism
= (irk, 7r) : QA — B satisfying the property ir(qA) C B®K. Let qA —, B®/C
be the restriction of ir. Conversely, using the Kasparov stabilization Theorem
1.4.2, one can show that each *-homomorphism : qA —, B ® /C defines (up to a
small deformation) a quasi-homomorphism from A to B. We obtain the following
definition of the KK bifunctor due to Cuntz.
188 8. HILBERT MODULES AND KK-THEORY

THEOREM 8.3.4 ([27]). The set of homotopy classes of *-homomorphisms qA

B 0 K; coincides with KK(A, B).
The details of the proof can also be found in [55, §5.2].
For an arbitrary A denote by PA : qA A the restriction of
(IdA,O) QA — A onto qA. Put q2A := q(qA).
Using the Pedersen derivation lifting theorem [104], one can prove the following
result (see also [55, § 5.1]).
THEOREM 8.3.5 ([28]). For any separable A there exists a *-homo-
morphism : qA M2(q2A) such that the operator is homotopic to the
embedding of the algebra q2A into M2(q2A) in the class of *-homomorphisms
and (PqA 0 ° c°A is homotopic to the embedding j' of the algebra qA into
M2(qA) (for embeddings into two-by-two matrices the formula x '—i ) is
(
used).
COROLLARY 8.3.6. The map PB : qB B induces an isomorphism
KK(A, qB) KK(A, B).
PROOF. The inverse map for can be defined as follows. Let : qA B®/C
represent an element of KK(A, B). It induces a *-homomorphism q2A
q(B ® K;). Moreover, there exists a canonical *-homomorphism ic q(B 0 K;)
qB ®/C. Then the desired element of KK(A, B) is represented by the *-homomor-
phism
qA M2(q2A) M2(q(B ® K;)) M2(qB 0/C) qB ®K;.
0
Since IC 0 /C K;, we obtain the following corollary.
COROLLARY 8.3.7 ([281). The set of homotopy classes of *-homomorphisms
qA OK; qB 0/C coincides with KK(A, B).
After obtaining this identification, the Kasparov product can be defined to be
the composition of *-homomorphisms.

8.4. Generalized Kasparov bimodules

In this section we shall briefly present some generalizations of the Kasparov
A, B-bimodules (triples) and discuss some applications of this approach. We begin
with a brief sketch of the construction of [3] (where the details can be found) based
on unbounded operators in Hilbert and its application to an algebraic
description of the Kasparov product.
DEFINITION 8.4.1. Let 6 be a Z/2Z-graded Hubert over a Z/2Z-
graded B and let T : Dom T 6 be a densely defined operator on 6.
The operator T is regular if
(i) T is closed;
(ii) the adjoint operator T* is densely defined on 6;
(iii) 1 + T*T has a dense image in 6.
 8.5. CLASSIFYING SPACES 189

Similarly to the case of bounded adjointable operators, T is B-linear in this

case too. If E = B, then regular operators on B are unbounded multipliers of B (see
[23]). For any regular operator T, the operator F := T(1 + T*T)_h/2 is bounded
and adjointable and one has
T = FH1, T* _ F*L_l, F* = T*(1 +TT*)_l/2
where H := (1 + T*T)_h/2 and L := (1 + TT*)_h/2.
DEFINITIoN 8.4.2. An unbounded Kasparov A, B-bimodule is a pair (E, D),
where e is an A, B-bimodule (cf. Definition 8.2.2) and D is a regular operator of
degree 1 satisfying
(i) D=D*;
(ii) ir(a)(1 + D2)' E K(E) for any a E A;
(iii) the set of all a e A, such that the commutator [D, a] =
D a bounded adjointable
operator, is dense in A.
Denote by (A, B) the set of all unbounded Kasparov A, B-bimodules.
With the help of a special approximate unit in the following proposition
is proved in [3].
PrtoPosrrloN 8.4.3. Suppose that (i', D) E EU(A, B) and F D(1 + D2)'/2.
Then (i', F) is a Kasparov A, B-bimodule.
The induced map /3: EU(A, B) —, KK(A, B) is surjective.
Suppose that is separable and B1 is o-unitaL One can verify that for
(Es, D2) e B2), the operator D0 := + admits a closure D,
which is a regular operator on := Hence, a coupling
x —'

is defined. It is proved in [3] that the diagram

x

i3xf3j,

K(A1,B1) x K(A2,B2)
commutes. This gives an algebraic description for the Kasparov product.
The approach based on unbounded modules turned out to be a useful tool and
was used in a number of papers (see, e.g., [51, 25, 26]).
Among more recent results let us indicate the notion of a Hubert module with
boundary introduced by M. Hilsum, which allowed him to make progress in some
topological problems [52, 53]. Unfortunately, these results (and other applications
of the Hilbert C*_module theory to elliptic theory) need an extended presentation
of pseudodifferential calculus and other topics which are far from the main subject
of our book.

8.5. Classifying spaces for some K- and KK-groups

Let A be a unital Let, as above, P(A) be the category of finitely
generated projective A-modules. Recall the definition of the group K(X; A) for a
190 8. HILBERT MODULES AND KK-THEORY

compact Hausdorif space X. The details and some generalizations can be found
e.g. in [121].
DEFINITION 8.5.1. Consider the set of locally trivial bundles E —+ X with the
fiber P P(A) and with the structure group of invertible continuous A-homomor-
phisms P P. We denote the symmetrization of the semigroup of the isomorphism
classes of such bundles (with respect to the direct sum operation) by K(X; A).
By HA-bundles we mean locally trivial bundles with the fiber HA and with the
structure group GL* (A).
LEMMA 8.5.2. Each HA-bundle over X is trivializable, i.e., isomorphic to X x
HA.
PROOF. This follows immediately from the contractibility of the structure
group (Theorem 7.2.2). 0
CONSTRUCTION 8.5.3. Denote by the set of all adjointable A-Fredholm
operators. Let
FE F : x '— F(x),
be a continuous family of Fredholm operators, x E X. Then (cf. Theorem 2.7.6)
we can choose an appropriate decomposition of HA such that = =
For each x E X there exists a neighborhood U, such that for any y
the map is an isomorphism onto its image. Since X is compact, one can
. Putn := min{n(xi),.. ,n(x8)},.Mj :=
.

and M0 := Then F(x)IM0 is an adjointable isomorphism onto its image for

each x. Moreover, by Corollary 2.3.5, the module F(x)(Mo) has an orthogonal
complement. This orthogonal complement (see the proof of Corollary 2.3.5) is
defined by a continuous family of projections. Hence, it is locally trivial. Two
subbundles arise in X x HA: KF := X x ..M3 and CF := (x, (F(x)Mo)-'-). Define
the map
index: —i K(X,A),
where [E] denotes the class of the bundle E in K(X; A). Evidently, the result does
not depend on the choice of n.
LEMMA 8.5.4. The map index is invariant under small perturbations of F with
respect to the uniform topology. Hence, it defines a map
index: —i K(X,A).
PROOF. Indeed, since X is compact, for any F there exists some e > 0 such
that for any C: X satisfying the inequality IIG—Fil <E, the submodule
still satisfies all restrictions and the projection onto (x, is close to
the projection onto CF. 0
THEOREM 8.5.5 ([126, 89, 30]). The map index defines an isomorphism
index: K(X,A).
PROOF. It is evident that this map is an epimorphism, since, for any bun-
dles E and F, there exists a family of operators making the following diagram
 8.5. CLASSIFYING SPACES 191

commutative:
(10)
00

XXHA >XXHA
Here we have used Lemma 8.5.2. To prove injectivity, one has to construct a
homotopy connecting an arbitrary map F: X with X —p IdHA, where is
the set of operators of index zero. For F, one has [NFl — [CFI = 0, i.e., after taking
sums of these bundles and some trivial bundle X x one obtains an isomorphism
J:Npe(X
x x e La)) on the right-hand
side. Hence, taking ii + k instead of n and writing J fiberwise as J(x), define, for
t E [0, 1], the following family of maps:

E
F
for h E
Since the fiberwise perturbation is compact, the map takes values in F* and the
index is constant, hence, zero. For t = 1 one obtains a map to GL* (A), which is
homotopic (in the class of maps to GL* (A)) to a map into a single point, according
to Theorem 7.2.2. El

REMARK 8.5.6. Using Theorem 7.2.2 it is proved in [89, 30] that

K(C(X) 0 A) [X, F*I K(C, C(X) ® A)
(the last identification in fact is by definition) Thus,
K(X; A) K(C(X) ® A) K(C, C(X) 0 A).
 Bibliography
1. C. Akemann, C. Pedersen, and J. Tomiyama, Multipliers of J. Funct. Anal. 13
(1973), 277—301.
2. C. Anantharaman-Delaroche and J. F. Havet, On approximate factorizations of completely
positive maps, J. Funct. Anal. 90 (1990), 411—428.
3. S. Baaj and P. JuIg, Théorie bivariante de Kasparov et opérateurs non bornés dans les
-modules hilbertiens, C. R. Acad. Sci. Paris Sér. I 296 (1983), 875—878.
4. S. Baaj and C. Skandalis, C -algkbres de Hopf et théorie de Kasparov dquivariantc, K-
Theory 2 (1989), 683—721.
5. M. Baillet, Y. Denizeau, and J.-F. Havet, Indice d 'une esperance conditionelle, Compositio
Math. 66 (1988), 199—236.
6. D. and B. Extensions of Hilberi C -modules, Houston J. Math. 30 (2004),
537—558.
7. A. 0. Barut and R. Theory of group representations and applications, PWN-Polish
Scientific Publishers, Warszawa, 1977.
8. P. Baum and A. Connes, Chern character for discrete groups, A Fête of Topology, Academic
Press, Boston, MA, 1988, pp. 163—232.
9. P. Baum, A. Connes, and N. Higson, Classifying space for proper actions and K-theory of
group Cs-algebras, Cs-Algebras: 1943-1993, A Fifty Year Celebration, Contemporary Math.,
vol. 167, Amer. Math. Soc., Providence, RI, 1994, pp. 241—291.
10. B. Blackadar, K-theory for operator algebras, Springer-Verlag, New York, 1986.
11. D. P. Blecher, A generalization of Hubert modules, J. Funct. Anal. 138 (1996), 365—421.
12. , A new approach to Hilbert C-modules, Math. Ann. 307 (1997), 253—290.

13. 0. Bratteli and D. W. Robinson, Operator algebras and quantum statistical mechanics. I,
Springer-Verlag, New York — Berlin — Heidelberg, 1981.
14. B. Brenken, Representations and automorphisms of the irrational rotation algebra, Pacific
J. Math. 111 (1984), 257—282.
15. L. G. Brown, Stable isomorphism of hereditary subalgebras of Pacific J. Math.
71 (1977), 335—348.
16. , Stable isomorphism of hereditary subalgebro.s of Pacific J. Math. 71
(1977), 335—348.
17. , Close hereditary C -subalgebras and the structure of quasi-multipliers, Preprint
MSRI 1121 1—85, 1985.
18. L. G. Brown, P. Green, and M. A. Rieffel, Stable isomorphism and strong Morita equivalence
of C-algebras, Pacific J. Math. 71 (1977), 349—363.
19. L. G. Brown and C. K. Pedersen, C-algebras of real rank zero, J. Funct. Anal. 99 (1991),
13 1—149.
20. M.-D. Choi and G. A. Elliot, Density of self-adjoint elements with finite spectrum in an
irrational rotation C-algebra, Math. Scand. 67 (1990), 73—86.
21. E. Christensen, E. Effros, and A. Sinclair, Completely bounded multilinear maps and
algebraic cohomology, Invent. Math. 90 (1987), 279—296.
22. A. Connes, et géométrie différentielle, C. R. Acad. Sci. Paris Sér. I 290 (1980),
599—604.
23. , An analogue of the Thom isomorphism for crossed products of a C -algebra by an

action of R, Adv. in Math. 39 (1981), 31—55.

24. , Noncommutative geometry, Academic Press, San Diego, CA, 1994.

193
194 BIBLIOGRAPHY

25. A. Connes and N. Higson, Deformations, morphismes asymptotiques et K-théorie bivariante,

C. R. Acad. Sci. Paris Sér. I 311 (1990), 101—106.
26. A. Connes and H. Moscovici, Cyclic cohomology, the Novikov conjecture and hyperbolic
groups, Topology 29 (1990). no. 3, 345—388.
27. J. Cuntz, Generalized homomorphisms between C5-algebras and KK-theory, Dynamics and
Processes, Lecture Notes in Math., vol. 1031, Springer-Verlag, 1983, pp. 31—45.
28. , A new look at KK-theory, K-Theory 1 (1987), 31—51.

29. , A survey on some aspects of non-commutative geometry, Jahresber. Deutsch. Math.-

Verein. 95 (1993), 60—84.

30. J. Cuntz and N. Higson, Kuiper's theorem for Hubert modules, Operator Algebras and Math-
ematical Physics, Contemporary Math., vol. 62, Amer. Math. Soc., Providence, RI, 1987,
pp. 429—435.
31. J. Dixmier, Les C5-algèbres et leurs representations, Gauthier-Villars, Paris, 1967.
32. , Les algèbres d'operateurs dans l'espace Hilbertien, Gauthier-Villars, Paris, 1969.

33. J. Dixmier and A. Douady, Champs continus d'espaces Hilbertiens, Bull. Soc. Math. France
91 (1963), 227—284.
34. R. C. Douglas and C. Pearcy, On the spectral theorem for normal operators, Proc. Cambridge
Philos. Soc. 68 (1970), 393—400.
35. N. Dunford and J. T. Schwartz, Linear operators. I. General theory, Interscience, New York,
1958.
36. M. J. Dupré and P. A. Fillmore, Thviality theorems for Hilbert modules, Topics in Modern
Operator Theory, 5th International Conference on Operator Theory (Timisoara and Her-
culane (Romania), June 2—12, 1980), Birkhäuser Verlag, Basel — Boston Stuttgart, 1981,
—

pp. 71—79.
37. 0. A. Elliott, T. Natsume, and R. Nest, The Heisenberg group and K-theory, K-Theory T
(1993), 409—428.
38. 0. A. Elliott, K. Saitô, and J. D. M. Wright, Embedding AW* -algebras as double commutants
in type I algebras, J. London Math. Soc. 28 (1983), 376—384.
39. S. C. Ferry, A. Ranicki, and J. Rosenberg (eds.), Novikov conjectures, index theorems and
rigidity, Vols. 1, 2, London Math. Soc. Lecture Note Ser., vols. 226, 227, Cambridge Univ.
Press, 1995.
40. 0. C. Filippov, On C-algebras A over which the H%lbert module l2(A) is self-dual, Vestnik
Moskov. Univ. Ser. I Mat. Mekh. (1987), no. 4, 74—76; English transi., Moscow Univ. Math.
Bull. 42 (1987), no. 4, 87—90.
41. M. Frank, A set of maps from K to isomorphic to Enda(k)(l2(a(k))). Applica-
tions, Ann. Global Anal. Geom. 3 (1985), no. 2, 155-171.
42. , Self-duality and C -reflexivity of Hilbert C* -modules, Z. Anal. Anwendungen 9

(1990), 165—176.
43. , Hilbert C -modules and related subjects a guided reference overview, Leipzig
University. ZHS-NTZ preprint No 13, 1996, funct—anbabbage.sissa.it, preprint #9605003.
44. , Geometrical aspects of Hilbert -modules, Positivity 3 (1999), 215—243.
45. M. Frank and E. Kirchberg, On conditional expectations of finite index, J. Operator Theory
40 (1998), no. 1, 87--ill.
46. M. Frank and V. M. Manuilov. Diagonalizing "compact" operators on Hilbert W -modules,
Z. Anal. Anwendungen 14 (1995), 33—41.
47. M. Frank, V. M. Manuilov, and E. V. Troitsky, On conditional expectations arising from
group actions, Z. Anal. Anwendungen 16 (1997), 831—850.
48. M. Frank and E. V. Troitsky, Lefschetz numbers and geometry of operators in W5-modules,
Funktsional Anal. i Prilozhen. 30 (1996), no. 4, 45—57; English transl., Functional Anal.
AppI. 30(1996), 257—266.
49. P. Criffiths and J. Harris, Principles of algebraic geometry. Vol. 1, Wiley-Interscience, New
York, 1978.
50. K. Grove and G. K. Pedersen, Diagonalizing matrices over C(X), J. Funct. Anal. 59 (1984),
64—89.
51. M. Hilsum, Signature operator on Lipschitz manifolds and unbounded Kasparov bimodules,
Operator Algebras and their Connections with Topology and Ergodic Theory (H. Araki, C. C.
Moore, Stratila, and D. Voiculescu, eds.), Lecture Notes in Math., vol. 1132, Springer—
Verlag, 1985, pp. 254- 288.
 ______

BIBLIOGRAPHY 195

52. , Index classes of Hubert modules with boundary, Preprint 2001-03 Institut de
mathématiques do Jussieu, 2001.
53. , Hilbert modules of foliated manifolds with boundary, Foliations: Geometry and
Dynamics (Walczak, Pawel et al., eds.), Proceedings of the Euroworkshop (Warsaw, Poland,
May 29-June 9, 2000), World Scientific, Singapore, 2002, pp. 315-332.
54. A. Irmatov, On a new topology in the space of Fredholm operators, Ann. Global Anal. Ceom.
7 (1989), no. 2, 93-106.
55. K. K. Jensen and K. Thomsen, Elements of KK-theorij, Mathematics: Theory and Applica-
tions, Birkhkuser, Boston, MA, 1991.
56. B. E. Johnson. Centralisers and operators reduced by maximal ideals, J. London Math. Soc.
43 (1968), 231—233.
57. R. V. Kadison. Diagonalizing matrices, Amer. J. Math. 106 (1984), 1451-1468.
58. R. V. Kadison and J. R. Ringrose, Fundamentals of the theory of opcator algebras. \ToIs.
1—2, Graduate Studies in Math., Amer. Math. Soc., Providence, RI, 1997.
59. I. Kaplansky, Modules over operator algebras, Amer. J. Math. 75 (1953), 839-858.
60. M. Karoubi, K-theory. An introduction, Grundlehren Math. Wiss., vol. 226, Springer-Verlag,
Berlin — Heidelberg — New York, 1978.
61. V. A. Kasimov, Homotopy properties of the general linear group of the Hilbert module 12(A).
Mat. Sbornik 119 (1982), 376—386; English transl., Math. USSR-Sb. 47(1984), 365—376.
62. G. G. Kasparov, Topological invariants of elliptic operators, I: K-homology. Izv. Akad. Nauk
SSSR Ser. Mat. 39 (1975), 796—838; English transl., Math. USSR-Izv. 9 (1975), 751—792.
63. , Hilbert theorems of Stinespring and Voiculescu, J. Operator Theory
4 (1980), 133—150.
64. , The operator K-functor and extensions of Izv. Akad. Nauk SSSR Ser.
Mat. 44 (1980), 571—636; English transl., Math. USSR-Izv. 16 (1981), 513—572.
65. , Novikov's conjecture on higher signatures: the Operator K-theory approach, Repre-

sentation Theory of Groups and Algebras (J. Adams, R. Herb, S. Kudla, J.-S. Li, R. Lipsman.
and J. Rosenberg, eds.), Contemporary Math., vol. 145, Amer. Math. Soc., Providence. RI,
1993, pp. 79—99.
66. A. Ya. Khelemskij, Banach and locally convex algebras, Clarendon Press, Oxford, 1993.
67. D. Kucerovsky, Finite rank operators and fanctional calculus on Hilbert modules over abelian
Canad. Math. Bull. 40 (1997), no. 2, 193—197.
68. N. Kuiper, The homotopy type of the unitary group of the Hilberl space, Topology 3 (1965),
19—30.
69. A. Kumjian, On equivari ant sheaf cohomology and elementary -bundles, J. Operator The-
ory 20 (1988), 207—240.
70. E. C. Lance, On nuclear Ce-algebras, J. Funct. Anal. 12 (1973), 157-176.
71. , Hilbert C5-modules - a toolkit for operator aiqebraists, London Math. Soc. Lecture

Note Ser., vol. 210, Cambridge Univ. Press, England, 1995.

72. N. P. Landsman, Rieffel induction as generalized quantum Marsden- Weinstein reduction,
J. Ceom. Phys. 15 (1995), 285—319.
73. H. Lin, Bounded module maps and pure completely positive maps, J. Operator Theory 26
(1991), 121—138.
74. , Injective Hilbert Pacific J. Math. 154 (1992), 131—164.
75. N. N. Lusin, Theory of functions of real variable, Uchpedgiz, Moscow, 1948. (Russian)
76. A. S. Lyskova, On Schrödinger operator in magnetic field. Uspekhi Mat. Nauk 36 (1981).
189—190; English transl., Russian Math. Surveys 36 (1981), no. 2, 181—182.
77. B. Magajna, Hilbert -modules in which all closed submodules are complemented Proc.
Amer. Math. Soc. 125 (1997), 849—852.
78. V. M. Manuilov, On Ko-group of continuous family of algebras A0, Uspekhi Mat. Nauk 44
(1989), no. 3, 163—164; English transl., Russian Math. Surveys 44 (1989), no. 3, 206-207.
79. , Diagonalization of compact operators in Hilbert modules over We-algebras of finite

type, Uspekhi Mat. Nauk 49 (1994), no. 2, 159—160; English transl., Russian Math. Surveys
49 (1994), 166—167.
80. , On eigenvalues of perturbed Schrödinger operators in magnetic field with irrational
magnetic flow, Funktsional Anal. i Prilozhen. 28 (1994), 57—60; English transl., Functional
Anal. AppI. 28 (1994), no. 2, 120-122.
 ______

BIBLIOGRAPHY

81. , Diagonalization of compact operators in Hubert modules over finite

Ann. Global Anal. Geom. 13 (1995), 207—226.

82. , Adjointability of operators on hubert -modules, Acta Math. Univ. Comenian. 65
(1996), 161—169.
83. , Diagonalization of operators over continuous fields of C* -algebras, Mat. Sb. 188
(1997), no. 6, 99—118; English transl., Sb. Math. 188 (1997), no. 6, 893—911.
84. , On almost commuting operators, Funktsional Anal. i Prilozhen. 31 (1997), no. 3,
80—82; English transl., Functional Anal. Appl. 31 (1997), no. 3, 212—213.
85. , An example of non-completed Hubert -module, Vestnik Moskov. Univ. Ser. I
Mat. Mekh. (2000), no. 5, 59—60; English transl., Moscow Univ. Math. Bull. 55 (2000), no.5,
38—39.
86. J. Milnor, On spaces having the hornotopy type of a CW-cornplex, Trans. Amer. Math. Soc.
90 (1959), 272—280.
87. J. Mingo and W. Phillips, Equivariant triviality theorems for Hilbert C -modules, Proc.
Amer. Math. Soc. 91 (1984), 225—230.
88. J. A. Mingo, On the contractibility of the general linear group of Hilbert space over a
algebra, J. Integral Equations Operator Theory 5 (1982), 888—891.
89. , K-theory and multipliers of stable Trans. Amer. Math. Soc. 299 (1987),
397—411.
90. A. S. Mishchenko, Banach algebras, pseudodifferential operators and their applications to K-
theory, Uspekhi Mat. Nauk 34 (1979), no. 6, 67—79; English transl., Russian Math. Surveys
34 (1979), no. 6, 77-91.
91. , Representations of compact groups on Hilbert modules over Trudy Mat.
Inst. Steklov. 166 (1984), 161—176; English transl., Proc. Steklov Inst. Math. 166 (1986),
179—195.
92. A. S. Mishchenko and A. T. Fomenko, The index of elliptic operators over Izv.
Akad. Nauk SSSR Ser. Mat. 43 (1979), 831—859; English transl., Math. USSR-Izv. 15 (1980)
87—1 12.
93. G. D. Mostow, Cohornology of topological groups and solvmanifolds, Ann. of Math. 73 (1961),
20—48.
94. G. J. Murphy, C* -algebras and operator theory, Academic Press, San Diego, 1990.
95. , Diagonality in C-algebras, Math. Z. 199 (1990), 279—284.
96. F. J. Murray and J. von Neumann, On rings of operators, Ann. of Math. 37(1936), 116—229.
97. G. Neubauer, Der Hornotopietyp der Automorphisrnegruppe in der Räumen 1p und co, Math.
Ann. 174 (1967), 33—40.
98. V. Nistor. A bivariant Chern-Connes character, Ann. of Math. 138 (1993), 555—590.
99. S. P. Novikov, Two-dimensional Schrödinger operators in periodic fields, Modern Problems
of Math., vol. 23, VINITI, Moscow, 1983, pp. 3—32.
100. W. L. Paschke, Inner product modules over B5-algebras, Trans. Amer. Math. Soc. 182 (1973),
443—468.
101. , The double B-dual of an inner product module over a C-algebra, Canad. J. Math.
26 (1974), 1272—1280.
102. , integrable group actions on von Neumann algebras, Math. Scand. 40 (1977), 234—

248.
103. A. A. Pavlov, Algebras of multipliers and spaces of quasirnultipliers, Vestnik Moskov. Univ.
Ser. I Mat. Mekh. (1998), no. 6, 14—18; English transl., Moscow Univ. Math. Bull. 53 (1998),
13—16.
104. G. K. Pedersen, C5-algebras and their automorphisni groups, Academic Press, London —
New York — San Francisco, 1979.
105. P. Popov and A. Buchina, Quasi-orthogonalization of functionals on l2(A), Acta. AppI.
Math. 68 (2001), 123—135.
106. P. 5. Popov, A topological criterion for almost orthocoinplernentation of all functionals on
12(C(X)). Mat. Zametki 65 (1999). no. 4, 636—640; English transi., Math. Notes 65 (1999),
no. 3-4, 532—536.
107. M. Reed and B. Simon, Methods of modern mathematical physics. V. 4. Analysis of opera-
tors, Academic Press, New York — London, 1978.
 BIBLIOGRAPHY 197

108. M. A. Rieffel, Induced representations of Adv. in Math. 13 (1973). 176—257.

109. , Morzta equivalence for C -algebras and I1/ -algebras, J. Pure Appi. Algebra 5

(1974). 51—96.
110. , Strong Morita equivalence of certain transformation group -algebras, Math. Ann.
222 (1976), 7—22.
111. , C-algehras associated with irrational rotations, Pacific J. Math. 93 (1981), 415—

429.
112. , Continuous fields of C* -algebras coming from group cocycles and actions. Math.

Ann. 283 (1989), 631—643.

113. J. Rosenberg, positive scalar curvature, and the Novikov conjecture. PubI.
Math. I.H.E.S. 58 (1983), 197-212.
114. , K and KK: Topology and operator algebras, Proc. Sympos. Pure Math. 51-1

(1990), 445—480.
115. W. Rudin, Punctional analysis, McGraw-Hill, New York. 1973.
116. S. Sakai, and Springer-Verlag, Berlin — New York, 1971.
117. H. Schröder, K-theory for real and applications. Pitman Res. Notes Math. Ser..
Longman Sci. Tech., Harlow, 1993.
118. V. V. Seregin, Reflexivity of modules obtained by the actions of a group. Vestnik
Moskov. Univ. Ser. I Mat. Mekh. (2003). no. 1, 40—45; English transi., Moscow Univ. Math.
Bull. 58 (2003). no. 1, 44—48.
119. G. Skandalis, Some remarks on Kasparov's theory. J. Funct. Anal. 56 (1984), 337-347.
120. , Kasparov 's bivariant K-theory and applications. Expositiones Math. 9 (1991), 193—

250.
121. Yu. P. Solovyov and E. V. Troitsky, C -algebras and elliptic operators in differential topology,
Amer. Iviath. Soc., Providence, RI, 2001.
122. V. S. Sunder and K. Thomsen, Unitary orbits of selfadjoints in some C' -algebras, Houston
J. Math. 18 (1992), 127—137.
123. M. Takesaki, Theory of operator algebras, 1, Springer-Verlag, New York — Heidelberg --
Berlin, 1979.
124. V. A. Trofimov, Reflexivity of Hilbert modules over the algebra of compact operators with
adjoint identity, Vestnik Moskov. Univ. Ser. I Mat. Mekh. (1986), no. 5, 60—64; English
transi., Moscow Univ. Math. Bull. 41 (1986), no.5, 51-55.
125. , Reflexive and self-dual Hilbert modules over some C -algebras, Uspekhi Mat. Nauk

42 (1987), no. 2, 247-248; English transl., Russian Math. Surveys 42 (1987), 303—304.
126. E. V. Troitsky, The classifying space of a K-functor related to a Vestnik Moskov.
Univ. Ser. I Mat. Mekh. (1985), no. 1, 96—98; English transl., Moscow Univ. Math. Bull. 40
(1985), 111—115.
127. , Contractibility of the full general linear group of the C*.Hilbert module l2(A), Funk-

tsional Anal. i Prilozhen. 20 (1986), no. 4, 50—64; English transl., Functional Anal. Appl. 20
(1986), 301—307.
128. , The eqivariant index of elliptic operators over Izv. Akad. Nauk SSSR
Ser. Mat. 50 (1986), no. 4, 849—865; English transl., Math. USSR-Izv. 29 (1987). 207—224.
129. , Orthogonal complements and endomorphisms of Hilbert modules and
complexes, Novikov Conjectures, Index Theorems and Rigidity, Vol. 2 (S. C. Ferry, A. Ran-
icki, and J. Rosenberg, eds.), London Math. Soc. Lecture Note Ser.. vol. 227, Cambridge
Univ. Press, 1995, pp. 309-331.
130. , Functionals on 12(A), Kuiper and Dixmier-Douady type theorems for
modules, 'frudy Mat. Inst. Steklov. 225 (1999), 362—380: English transl., Proc. Steklov Inst.
Math. 225 (1999), 344—362.
131. , Discrete group actions and corresponding modules, Proc. Amer. Math. Soc. 131
(2003), 3411—3422.
132. A. Valette, Les fibres en thdorie de Ko.sparov, Mem. Acad. Roy. BeIg. Cl. Sci. XLV (1988),
no. 6.
133. Y. Watatani, Index for C*.subalgebras, Mem. Amer. Math. Soc., vol. 424, Amer. Math. Soc..
Providence, RI, 1990.
134. N. E. Wegge-Olsen, K-theory and Oxford Univ. Press, Oxford. 1993.
198 BIBLIOGRAPHY

135. S. L. Woronowicz, Unbounded elements affiliated with C* -algebras and non-compact quantum
groups, Comm. Math. Phys. 136 (1991), 399—432.
136. S. L. Woronowicz and K. Napiórkowski, Operator theory in the C* -algebra framework, Rep.
Math. Phys. 31 (1992), 353—371.
137. S. Zhang, Diagonalizing projections in multiplier algebras and in matrices over a C -algebra,
Pacific J. Math. 145 (1990), 181—200.
 Notation Index

a112 1 4
a> 0 24 LC(A) 105
a0 1 LM(A) 105
I M' 27
A" 56

M
1 19
/3-urn 114 M(A) 102
Co(X) 80 19
DC(A) 103 6
GL 50 43
GL 52 29
E:A—.B 6 pVq 71
EndA(M) 15 pAq 71
16
P(A)

29
6 RC(A) 105
flB 42 RM(A) 105
HornA(M,H) 15 RR(A) 147
16 s-tim 55
index F 36 Sp(A) 1

/C 19 tsr(A) 149
IC(M) 18 65
IC(M,Jsf) 18 65
ICA 123 18
8K(M,Jtf) 32 w-lim 55
K(E) 87 27
K(A) 33 La 3
Ko(A) 33 114
33 Z(A) 57
KK(A, B) 182 5
12(A) 6 e 5
12(M) 6

199
 Index

A-orthobasis, 8 equivalent projections, 58

A-valued product, 3 essential ideal, 101
equivalent, 126 F-topology, 126
A, B-bimodule, 182 factor, 58
A-B-equivalence bimodule, 11 functional, 27
algebra nonsingular, 130
1
normal, 56
divisibly infinite (DI), 31 positive, 1
module-infinite (MI), 30 representable, 130
u-unitai, 2 G-B-module, 42
universal enveloping, 56
42
von Neumann, 56
GNS-construction, 2
56
group,
continuous, 57
finite, 57 Grothendieck, 33
of type 1, 57 linear,
of type II, 57 full general GL, 50
of type lii, 57 general GL*, 52
of type 1100, 57 inner product,
of type III, 57 equivalent, 126
properly infinite, 57 Kasparov bimodule, 182
semifinite, 57 unbounded, 189
u-finite, 57 KK-groups,
approximate unit, 2 Cuntz definition, 188
bicommutant, 55 Kasparov definition, 182
Cauchy-Bunyakovsky inequality, 4 lemma
centralizer of Dixmier-Douady, 173
double, 103 module
left, 105 dual, 27
quasi-, 105 Hilbert, 4
commutant, 55 countably generated, 8
compatible pair, 118 4
conditional expectation, 6
full, 11
faithful, 6
reflexive, 79
of finite index, 87
algebraically, 86 self-dual, 27
characteristic number of, 87 standard, 6
element of pre-Hilbert, 3
hermitian, 1 projective finitely generated, 9
positive, 1 multiplier, 102
strictly positive, 2 loft, 105
element of module quasi-, 105
nonsingular, 8 right, 105
i-orthogonal submodules, 171 operator homotopy, 184

201
202 INDEX

operator on module, 15 topology,

A-Fredholm, 34 left strict, 116
index of, 36 quasi-strict, 118
adjoint, 15 strict, 114
Banach-compact, 32 strong, 55
compact, 18 strong*, 55
elementary, 18 strong module, 116
orthogonal complement, 6 strong* module, 116
polarization equality, 4 weak ,55
positive square root, 1 ci(E, E)-, 56
pre-Hilbert A-B-bimodule, 11 o(E, Em)-, 56
property (E), 169 0-strong 55
property (K), 169 tl_strong* 55
representation of an algebra, 1 or-weak, 55
nondegenerate, 102 trace, 57
universal, 56 faithful, 57
set finite, 57
A-precompact, 33 normal, 57
space semifinite, 57
pre-dual, 56 triple,
spectrum of an element, 1 degenerated, 182
state, 1 ultra weak direct sum, 63
strong Morita equivalence, 12 unitalization, 1
submodule, vector
complemented, 22 cyclic, 1
symmetrization, 33 separating, 1
tensor product
exterior, 181
interior, 181
theorem
of Cuntz-Higson, 166
of Dixmier-Douady for modules, 174
of Dupre-Fillmore, 10
of von Neumann about bicommutant, 55
Titles in This Series
226 V. M. Manuilov and E. V. Troitsky, Hubert C'S-modules, 2005
225 S. M. Natanzon, Moduli of Riemann surfaces, real algebraic curves, and their
superanaloges, 2004
224 Ichiro Shigekawa, Stochastic analysis, 2004
223 Masatoshi Noumi, Painlevé equations through symmetry, 2004
222 G. G. Magaril-I1'yaev and V. M. Tikhomirov, Convex analysis: Theory and
applications, 2003
221 Katsuei Kenmotsu, Surfaces with constant mean curvature, 2003
220 1. M. Gelfand, S. G. Gindikin, and M. I. Graev, Selected topics in integral
geometry, 2003
219 S. V. Kerov, Asymptotic representation theory of the symmetric group and its
applications to analysis, 2003
218 Kenji Ueno, Algebraic geometry 3: Further study of schemes, 2003
217 Masaki Kashiwara, D-modules and microlocal calculus, 2003
216 C. V. Badalyan, Quasipower series and quasianalytic classes of functions, 2002
215 Tatsuo Kimura, Introduction to prehomogeneous vector spaces, 2003
214 L. Grinblat, Algebras of sets and combinatorics, 2002
213 V. N. Sachkov and V. E. Tarakanov, Combinatorics of nonnegative matrices, 2002
212 A. V. Mel'nikov, S. N. Volkov, and M. L. Nechaev, Mathematics of financial
obligations, 2002
211 Takeo Ohsawa, Analysis of several complex variables, 2002
210 Toshitake Kohno, Conformal field theory and topology, 2002
209 Yasumasa Nishiura, Far-from-equilibrium dynamics, 2002
208 Yukio Matsumoto, An introduction to Morse theory, 2002
207 Ken'ichi Ohshika, Discrete groups, 2002
206 Yuji Shimizu and Kenji Ueno, Advances in moduli theory, 2002
205 Seiki Nishikawa, Variational problems in geometry, 2001
204 A. M. Vinogradov, Cohomological analysis of partial differential equations and
Secondary Calculus, 2001
203 Te Sun Han and Kingo Kobayashi, Mathematics of information and coding, 2002
202 V. P. Maslov and G. A. Omel'yanov, Geometric asymptotics for nonlinear PDE. I,
2001
201 Shigeyuki Morita, Geometry of differential forms, 2001
200 V. V. Prasolov and V. M. Tikhomirov, Geometry, 2001
199 Shigeyuki Morita, Geometry of characteristic classes, 2001
198 V. A. Smirnov, Simplicial and operad methods in algebraic topology, 2001
197 Kenji Ueno, Algebraic geometry 2: Sheaves and cohomology, 2001
196 Yu. N. Lin'kov, Asymptotic statistical methods for stochastic processes, 2001
195 Minoru Wakimoto, Infinite-dimensional Lie algebras, 2001
194 Valery B. Nevzorov, Records: Mathematical theory, 2001
193 Toshlo Nishino, Function theory in several complex variables, 2001
192 Yu. P. Solovyov and E. V. Troitsky, C-algebras and elliptic operators in differential
topology, 2001
191 Shun-ichi Amari and Hiroshi Nagaoka, Methods of information geometry, 2000
190 Alexander N. Starkov, Dynamical systems on homogeneous spaces. 2000
189 Mitsuru Ikawa, Hyperbolic partial differential equations and wave phenomena, 2000
 TITLES IN THIS SERIES

188 V. V. Buldygin and Yu. V. Kozachenko, Metric characterization of random variables

and random processes, 2000
187 A. V. Fursikov, Optimal control of distributed systems. Theory and applications, 2000
186 Kazuya Kato, Nobushige Kurokawa, and Takeshi Saito, Number theory 1:
Fermat's dream, 2000
185 Ken.ji Ueno, Algebraic Geometry 1: Ftom algebraic varieties to schemes, 1999
184 A. V. Mel'nikov, Financial markets, 1999
183 Hajime Sato, Algebraic topology: an intuitive approach, 1999
182 I. S. Krasil'shchik and A. M. Vinogradov, Editors, Symmetries and conservation
laws for differential equations of mathematical physics, 1999
181 Ya. G. Berkovich and E. M. Zhmud', Characters of finite groups. Part 2, 1999
180 A. A. Milyutin and N. P. Osmolovskii, Calculus of variations and optimal control,
1998
179 V. E. VoskresenskiT, Algebraic groups and their birational invariants, 1998
178 Mitsuo Morimoto, Analytic functionals on the sphere, 1998
177 Satoru Igari, Real analysis—with an introduction to wavelet theory, 1998
176 L. M. Lerman and Ya. L. Umanskiy, Four-dimensional integrable Hamiltonian
systems with simple singular points (topological aspects), 1998
175 S. K. Godunov, Modern aspects of linear algebra, 1998
174 Ya-Zhe Chen and Lan-Cheng Wu, Second order elliptic equations and elliptic
systems, 1998
173 Yu. A. Davydov, M. A. Lifshits, and N. V. Smorodina, Local properties of
distributions of stochastic functionals, 1998
172 Ya. G. Berkovich and E. M. Zhmud', Characters of finite groups. Part 1, 1998
171 E. M. Landis, Second order equations of elliptic and parabolic type, 1998
170 Viktor Prasolov and Yuri Solovyev, Elliptic functions and elliptic integrals, 1997
169 S. K. Godunov, Ordinary differential equations with constant coefficient, 1997
168 Junjiro Noguchi, Introduction to complex analysis, 1998
167 Masaya Yamaguti, Masayoshi Hata, and Jun Kigami, Mathematics of fractals, 1997
166 Kenji Ueno, An introduction to algebraic geometry, 1997
165 V. V. Ishkhanov, 13. B. Lur'e, and D. K. Faddeev, The embedding problem in
Calois theory, 1997
164 E. I. Gordon, Nonstandard methods in commutative harmonic analysis, 1997
163 A. Ya. Dorogovtsev, D. S. Silvestrov, A. V. Skorokhod, and M. I. Yadrenko,
Probability theory: Collection of problems, 1997
162 M. V. J3oldin, G. I. Simonova, and Yu. N. Tyurin, Sign-based methods in linear
statistical models, 1997
161 Michael Blank, Discreteness and continuity in problems of chaotic dynamics, 1997
160 V. G. OsmolovsklT, Linear and nonlinear perturbations of the operator div, 1997
159 5. Ya. Khavinson, Best approximation by linear superpositions (approximate
nomography), 1997

For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/.
;i\IS the