Algebraic Methods in Operator Theory

Algebraic Methods
in Operator Theory
RaUl E. Curto
Palle E. T. Jf2jrgensen
Editors
Springer Science+Business Media, LLC

RaulCurto Palle E. T. J~rgensen
Department of Mathematics MLH Department of Mathematics MLH
University of Iowa University of Iowa
Iowa City, IA 52242 Iowa City, IA 52242
Library of Congress Cataloging-in-PubUcation Data
Algebraic methods in operator theory I Raul E. Curto, Palle E. T.

J~rgensen, editors.
p. cm.
Includes bibliographical references.
ISBN 978-1-4612-6683-9 ISBN 978-1-4612-0255-4 (eBook)
DOI 10.1007/978-1-4612-0255-4
1. Operator theory. 2. Banach algebras. 1. Curto, Raul E.,
1954- . II. J~rgensen, Palle E. T., 1947-
QA329.A44 1994 94-5179
512'.55--dc20 CIP
Printed on acid-free paper

~Springer Science+Business Media New York 1994
Origina11y published by Birkhauser Boston in 1994
Softcover reprint of the hardcover 1st edition 1994
Copyright is not claimed for works of U.S. Government employees.
AII rights reserved. No part ofthis publication may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, electronic, mechanica1, photocopying, recording,
or otherwise, without prior permission of the copyright owner.
Permission to photocopy for internal or personal use of specific clients is granted by

Springer Science+Business Media, LLC for libraries and other users registered with the
Copyright Clearance
Center (CCC), provided that the base fee of $6.00 percopy, plus $0.20 per page is paid directly
to CCC, 222 Rosewood Drive, Danvers, MA 01923, U.S.A. Special requests should be
addressed directly to Springer Science+Business Media, LLC.
ISBN 978-1-4612-6683-9
Camera-ready copy by the Authors.

Contents
Preface . viii
I. Single Operators and Applications
On the commutant lifting theorem and Hankel operators

Radu Gadidov . . . . . . . . . . . . . . . . . . 3
Operator semigroups, invariant sets and invariant subspaces
John Froelich and Michael Marsalli . . . . . . . . . . 10
The local De Branges-Rovnyak construction and complete
Nevanlinna-Pick kernels
Scott McCullough . . . . . . . . . . . . . . . . . . 15
Local spectral theory for multipliers and convolution operators
Vivien G. Miller and Michael M. Neumann . . . . . 25
Operator-valued Poisson kernels and standard models in
several variables
F.-H. Vasilescu . . . . . . . . . . . . . . . . . . . . . . 37
II. Nonselfadjoint Algebras
Elementary operators and subalgebras

Keith J. Coates. . . . . . . . . . . . . . . . . . . . . 49
Questions on bimodules of nest algebras
Xingde Dai . . . . . . . . . . . . . . . . . . 52
The universal factorization property for commutative
subspace lattices
John Daughtry . . . . . . . . . . . . . . . . . . . . . . 59
Compression limit algebras
Alan Hopenwasser and Cecelia Laurie 66
Inverse semigroups, groupoids and a problem of J. Renault
Alan L. T. Paterson . . . . . . . . . . . . . . 79
On Banach algebras generated by two idempotents
Ysette Weiss . . . . . . . . . . . . . . . . . . . . . . . 90
III. C· Algebras
Berezin-Toeplitz quantization
L.A. Coburn . . . . . . . . 101
vi Contents
Normal elements of a simple C·-algebra

Georye A. Elliott . . . . . . . . . . 109
A Gelfand-Naimark theorem for C* -algebras
I chiro Fujimoto . . . . . . . . . . . . . 124
The generalized Weyl-von Neumann theorem and C* -algebra
extensions
Huaxin Lin. . . . . . . . . . . . . . . . . . . . . . . 134
The cb-norm of a derivation
Martin Mathieu. . . . . . 144
Hopf C* -algebras and their representations
Xiu-Chi Quan . . . . . . . . . . . . 153
Quasi-diagonalizing unit aries and the generalized
Weyl-von Neumann Theorem
Shuang Zhang . . . . . . . . . . . . . 163
IV. von Neumann Algebras and Subfactors
On the structure of finite depth subfactors

Dietmar Bisch . . . . . . . . . . . . 175
Universally bounded operators on von Neumann algebras
of type III
Erik Christensen . . . . . . . . . . . . . . . . . . 195
Non occurrence of star graphs as principal graphs
Said A. Rida . . . . . . . . . . . . . . . . 205
Depth 2 subfactors and Hopf algebra crossed products
W ojciech Szymanski. . . . . . . . . . . . . . . 218
V. Representations of Groups and Algebras on Hilbert Space
Generalized characters of U 00
Robert P. Boyer . . . . . . 225
Module structures on Hochschild and cyclic cohomology
of crossed products
Ronghui Ji . . . . . . . . . . . . . . . . . . 236
q-Relations and stability of C*-isomorphism classes
P.E. T. J{t1rgensen, L.M. Schmitt, and R.F. Werner 261
A test for injectivity for asymptotic morphisms
Terry A. Loring . . . . . . . . . . . . . 272
On the "quantum disk" and a "non-commutative circle"
Gabriel Nagy and Alexandru Nica . . . . . . . . . . 276
Contents vii
Landstad duality for coactions on C· -algebras

John C. Quigg . . . . . . . . . . . . . . 291
Quantization of Poisson SU(2)
Albert Jeu-Liang Sheu . . . 296
VI. Geometry and Topology
"Vector bundles" over quantum Heisenberg manifolds

Beatriz Abadie . . . . . . . . . . . . . . . . . 307
Deformations of topological spaces predicted by E-theory
Marius Diidiirlat and Terry A. Loring . . . . . . . . . 316
Analyticity, uniforni averaging and K-Theory
Paul S. Muhly, Chaoxin Qiu, and Jingbo Xia . 328
Boundary value problems for functions analytic
on multiply connected domains on spaces
with a general weight
D. Kurtz, Yu. Latushkin, and 1. Spitkovsky . . 350
Preface
The theory of operators stands at the intersection of the frontiers of modern

analysis and its classical counterparts; of algebra and quantum mechanics;
of spectral theory and partial differential equations; of the modern global
approach to topology and geometry; of representation theory and harmonic
analysis; and of dynamical systems and mathematical physics. The present
collection of papers represents contributions to a conference, and they have
been carefully selected with a view to bridging different but related areas of
mathematics which have only recently displayed an unexpected network of
interconnections, as well as new and exciting cross-fertilizations. Our unify-
ing theme is the algebraic view and approach to the study of operators and
their applications. The complementarity between the diversity of topics on
the one hand and the unity of ideas on the other has been stressed. Some
of the longer contributions represent material from lectures (in expanded
form and with proofs for the most part). However, the shorter papers,
as well as the longer ones, are an integral part of the picture; they have
all been carefully refereed and revised with a view to a unity of purpose,
timeliness, readability, and broad appeal.
Raul Curto and Paile E. T. J¢rgensen

Department of Mathematics
University of Iowa
Iowa City, Iowa
December, 1993
Section I
SINGLE OPERATORS AND APPLICATIONS
ON THE COMMUTANT LIFTING
THEOREM AND HANKEL OPERATORS
RADU GADIDOV
In this note we give another proof of the commutant lifting theorem (see
[6] and [7]), based on the Adamjan-Arov-Krein techniques introduced in [1]. We
then apply the construction given in Theorem 1 below to obtain a generalization
of a result in [2] (see also [4]).
Let us begin with some notation. Throughout this note all Hilbert spaces
are understood to be complex and separable. If?i,?i' are Hilbert spaces,
the algebra of all bounded linear operators from ?i into ?i' will be denoted
by 8(?i, ?i') and by 8(?i) if?i = ?i'. If T is a contraction on the Hilbert space
?i, we denote as usual the defect operator of T by DT (:= (1 - T*T)i), the
defect space of T by 1)T (:= DT?i ), the minimal isometric dilation of T by
U E 8(Je), and the orthogonal projection from Je onto?i by P. Furthermore, if
1) is a Hilbert space, the spaces L2(1) and H2(1) are the usual Lebesgue and
Hardy spaces of 1)-valued functions on the unit circle T relative to normalized
Lebesgue measure m on T, S E 8(H2(1)) is the multiplication by z on H2(1),
H:'(1):=L2(1)9H2(1) and P_ the orthogonal projection from L2('D) onto
H:'(1). If 1)' is another Hilbert space and S' E 8(H2(1)~» is the multiplica-
tion by:Z on H:'(1)'), a Hankel operator is an operator H in 8(H2(1) ,H:'(1)'»
such that HS = S'* H; moreover, if 9: T -+ 8(1),1)') is a bounded measurable
function (weakly or strongly, which amounts to the same since both 1) and 1)'
are separable) and / is in L2(1), the function %-+9(z)/(z) on T will be de-
noted by 9 f. For any d in 1) the constant function on T whose value is d will
also be denoted by d, and the function z -+:zd on T will be denoted by :Zd.
Recall that the usual form of the commutant lifting theorem is as follows
(cf. [7, Theorem 2.2.3]).
Theorem 1. Let T and T' be contractions in 8(?i) and 8(?i'), respectively,
having minimal isometric dilations U and U' in 8(Je) and 8(Je'). If A is a
contraction in 8(?i, ?i') such that AT = T' A, then there exists a contraction A
in 8(Je, Je') such that AU = u' A and P' A = AP.
We give another proof of this fundamental theorem below. When applied
to Hankel operators, the construction given in the proof of Theorem 1 yields
the following generalization of a result in [2], where the spaces 1) and 1)' are
finite dimensional (see also [4]).
Theorem 2. Let 1), 1)' be Hilbert spaces such that dim 1) ::; dim 1)'. If H
is a Hankel operator in 8(H2(1) ,H:.(1)'» with IIHII < 1, then there exists a
4 R. Gadidov
bounded measurable function e : T -+ 8(V, V') such that e(z) is an isometry

a.e. on T and Hh = ~eh for all h in H2(V).
Proof of Theorem 1. Let B in 8(/C, 1t') be defined by B = AP. Then BU =
T' B, so it suffices to produce a contraction A in 8(/C, /C') such that AU = u' A
and P' A = B. Since B is a contraction, the operator X on 1t' ffi /C with matrix
is clearly a positive operator in 8(1t'ffi/C). Using X to define a new inner product

on 1t'ffi/C and the usual process of factorization and completion, we obtain a
Hilbert space it such that 1t', /C cit, it = 1t'V/C and for all h', h~ in 1t' and
k, kl in /C,
< h' + k,h~ + k~ >it =< h',h~ >1t' + < Bk,h~ >1('
+ < B*h',k 1 >IC + < k,kl >IC .
Let us define T: 1t'VU/c-+T'*1t'V/C by
(1) T(h' + Uk) = T'* h' + k , for all h' in 1t' and k in /C.
Since
(2) IIT'*h' + kll~ = IIT'*h'II~, + 2Re < Bk,T'*h' >1t' +lIkllk

= 1IT'*h'II~, + 2Re < BUk,h' >1(' +IIUklik
~ 11k'1I~' + 2Re < BUk, h' >1(' +IIUklik
= Ilk' + Ukll~,
it follows that T is well defined and is a contraction, so it can be extended to a
contraction (denoted again by T) acting on the whole spate it. Let U* in 8(i*)
be the minimal isometric dilation of T* and set
-,
/C = Vu-*n-u, , U-, = U-*Ii', A=...r.·
/I. = r,i'11C .
n~O
We ass~rt that U' in 8(i') is a minimal isometric dilation of T', AU

- = U' A-
and P'A = AP. Since i ' = V u,n1t' and T'* = 111t' = U:I1t' = Urn' it follows
n~O
that U' is a minimal isometric dilation of T'. To show that P' A = B, suppose
h' E 1t' and k E /C. Then
<Ak,h' >i'=< k,h' >i. =< k,h' >it=< Bk,h' >1(',

On the commutant lifting theorem and Hankel operators 5
Finally, to show that AU

- = if'A,- let h' be in 1t', kin /C and n a positive integer.
Then
<AUk, if'nh' >i;. =< Uk, if;:h' >i;. =< Uk,f*nh' >it
=< fUk,f*(n-l)h' >it=< k,f*(n-l)h' >it
=< k, if:(n-l)h' >i;. =<1k, if!n-l)h' >i;.
_z _, _z _
=< U.Ak, U;:h >i;. =< U'Ak, u'nh' >i;•.
Also, for n = 0, we have
< 1Uk,h' >i;. =< Uk,h' >i;. =< Uk,h' >it

=< BUk, h' >1('=< T' Bk, h' >1{.'
=< Bk,1"*h' >1{.'=< k,1"*h' >it
=< k, 1"* h' >i;. =< Ak, 1"*h' >i;'
- -
=< Ak,fh' >i;.=<Ak, if:h' >i;.
- , >i;. =< U'Ak,
--
=< U*Ak, h
-
-,-, h >i;.
Since the subspaces {if'n1t'}n:o span it' we conclude that if'A = AU. By
the uniqueness (up to isomorphism) of the-minimal isometric dilation of T' (cf.
[7, Theorem 1.4.1]), there exists a unitary_operator q,' in 8(/C' , it') such that
q,11{.' = 11{.' and q,'U' = if'q,', so A := q,'* A is a contraction in 8(/C, /C'), such
that AU = u' A, P' A = AP, and the proof is complete.
Remark. The construction made above d~n't just luckily fit. With a
•
little effort one can show that the contraction A in Theorem 1 is unique if
and only if ii. = 1t'VU/C or it = T'*1t'V/c. Also by noting that the operator
0: ii.-vAe1t'e(/Ce1t) defined by O(h' + k) = DAPke(h' + APk)e(1 - P)k,
for all h' in 1t' and k in /C, is unitary, one can see that
O((1t'VU/C).L) = 1lA (:= (VA e VT) e {DATh e DThj he 1t}),
O((1"*1t'V/c).L) = 1lA (:= (VA eVT') e {DAhe DT.Ah j he 1t}),
so one regains the main result in [3].
Proof of Theorem 2. Note that since HS = S'* H, we may put ourselves in the
framework of Theorem 1 by taking A = H, T = S and T' = S'*. Since S is an
isometry, it follows that 1t(:=H2(V» = /C, and since IIHII < 1,
x= [;. ~l
6 R. Gadidov
is invertible, so any vector h in if. can be uniquely expressed as h = h' + h for

some h' in 1t'(:=H:,(V'» and h in 1t (=H2(V». Moreover, it follows easily
from (2) that T defined in (1) is an isometry and
(3)
(1t'VT1t)..l = { - HDEld + DEld; deV},
(T'*1t'V1t)..l = { DH~ (zd') - H* DH~ (zd'); d' e V'} ,
so
dim«1t'VT1t)..l) = dimV:::; dirnV' = dim«T'*1t'V1t)..l).
Hence T can be extended to an isometry (denoted again by T) acting on the

whole Hilbert space it. Then the minimal isometric dilation of T*, denoted as
above by U.e8(K.*), is a unitary operator, and we show that A
(:= P:,j1t)
is an isometry. Once we show this the proof can be completed as follows.
Since U' is a minimal isometric dilation of 8', there exists a unitary operator
t/J' e 8(K.' ,L2 (V'» such that t/JIH2(V') = ~H2(V') and 5' := t/J'U't/J'* is the multi-
plication by z on L2(V'). Thus if := t/J' A is an isometry in 8(H2(V) ,L2(V'»
such that if 8 = 5' if and P~ if = H, and hence there exists a bounded measur-
able function 8 : T -+ 8(V, V') such that for all h in H2 (V), if h = 8h. Since
if is an isometry, 8(z) is ~ isometry a.e ..
Now we show that A is an isometry. Note first that this is equivalent to
if. c ie'
and since K.* = it V K.', what we really need to show is that K.* = K.'.
Let now k. in K.* be orthogonal to K.'. There exist h and {h n } n':1 in it such
that k* = h+ f
n=i
fJ!n-i)(U. - T*)h n and IIk*1I 2 = IIhll 2 + f
n=i
IIDT.i~nIl2 (cf.
[7, Theorem 2.2.1]) and since for any positive integer n, fJ!n-i)(U* - T*)h n is
orthogonal to 1t', it follows that h is orthogonal to 1t'. It is trivial to check that
(4) ite1t' = { - Hh + h; he 1t},
so there exists h in 1t such that
(5) h = -Hh+h.
For all h' in 1t',
0=< k*, U*h' >K:. =< k., T·h' + (U* - T*)h' >K:.
=< h,T h >it + < DT.hi,DT.h >it
- -. I - ,
=< Th + DT.hbh' >it'

so there exists hi in 1t such that
(6)
is invertible, so any vector h in ii can be uniquely expressed as h = h' + h for

some h' in 1-£'(:=H:(V'» and h in 1-£ (=H2(V». N~oreover, it follows easily
from (2) that T defined in (1) is an isometry and
(1-£'VT1-l)J.. = { - H Di/d + Di/d; dE V} ,

(3)
(T'*1-£'V1-£)J.. = {Di~ ("Ed') - H· Di~ (zd') ; d' E V'} ,
so
dime (1-£'VT1i) J..) = dim V ~ dim V' = dime (T'·1-£'V1t) J..) .
Hence T can be extended to an isometry (denoted again by T) acting on the
whole Hilbert space ii. Then the minimal isometric dilation of T·, denoted as
above by U.EB(f.), is a unitary operator, and we show that i
(:= PZ,j'H)
is an isometry. Once we show this the proof can be completed as follows.
Since U' is a minimal isometric dilation of 8', there exists a unitary operator
4>' E B(f' ,L2 (V'» such that 4>iH2(7)') = ~H2(7)') and S' := 4>'U' 4>'* is the multi-
plication by z on L 2(V'). Thus ii := 4>';' is an isometry in B(H2(V) ,L2(V'»
such that ii 8 = s'ii and P!.. ii = H, and hence there exists a bounded measur-
able function 9 : T -+ B(V, V') such that for all h in H2(V), iih = 9h. Since
ii is an isometry, 9(z) is ~ isometry a.e ..
Now we show that;' is an isometry. Note first that this is equivalent to
ii c f' and since f.= ii Vf', what we really need to show is that f. = f'.
Let now k. in f. be orthogonal to f'. There exist h and {h n } n':l in ii such
that k. = h + Emn-l)(U. - T*)hn and IIk.1I
n=l
2 = IIhll 2 + E IIDf.hnl 2
n=l
(cf.
[7, Theorem 2.2.1]) and since for any positive integern, mn-l)(U. - T*)hn is
orthogonal to 1-£', it follows that h is orthogonal to 1-£'. It is trivial to check that
(4) iie1-£' = { - Hh + h; h E 1-£ },

so there exists h in 1-£ such that
(5) h =-Hh+h.
For all h' in 1-£',
0=< k., U.h' >i.. =< k., T*h' + (U. - T·)h' >i..
- -. I - ,
=<h,T h >it+<Df.hl,Df.h >it
-- -,
=< Th + Df.hl,h >it'
so there exists hi in 1-£ such that
(6)
8 R. Gadidov
By (3) and (5) there exist h' in 1{,', h2 in 1{, and d in 'D such that
Since D'jo is the orthogonal projection onto (TH.)J., 'D'jo C (T'*1{,'V1{,)J., 80

there exists d' in 'D' such that
(8)
By the way we extended T to all of ii, T«'H.'VTH)J.) c (T'*1{,'V1t)J., 80 there

exists d~ in 'D' such that
(9) T( -HDii2d + Dii2d) = Dii~ (zdD - H* Dii~ (z~) .
Putting together (1) and (6)-(9) we get

--
Th+D'johl =T«h
- -,
+ Th2) + (-HDii2 d+Dii2 d»+D'johl
-
= T(h' + T~) + T(-HDii2d + Dii2d) + (Dii~(zd') - H* Dii~(zd'»

= (T'*h' +~) + (Dii~(zdD - H* Dii~(z~» + (Dii~(%d') - H* Dii~(%d'»
= (T'*h' +~) + (Dii~ (z(d' + d~» - H* Dii~ (z(d' + d~))) = -Hhl + hi.
Thus
(10)
hi = h2 - H* Dii~ (z(d' +~»,
-Hhl = T'*h' + Dii~(%(d' +~».
Therefore
Hhl = Hh2 - HH* Dii~(%(d' +~» = -T'*h' - Dii~(%(d' + dD>,
from which it follows that zed' + dD = -H~ - T'*h'. From this relation and
(10) we get
(11) hi = ~+H*Dii~(Hh2+T'*h')
= ~+H*HDii2~+H*Dii~T'h'
= Dii2~ + H* Dii~T'*h' = Dii2(~ + H*T'h').
By (7) we get
h = -Hh + HDii2d
h2 =T*h-T*Dii2d,
80
~ +H*T'*h' =T*h-T*Dii2d-T*H*Hh+T*H*HDii2d
= T* Dih - T* d =T* Dih.
Taking into account (11) we obtain h1 = D;lr DJ,h. From the above relation
and (6) we deduce
11Th + Dt.h1112 = IIhll2+ II Dt. h111 2= 11- Hh1 +h111~

= IIDHh1112 = IIDii1rDJ,hIl2,
80
IIhll2+ IIDt.h1112 = IIDii1rD~h1l2 .

By induction (using k.l. U:1t' for any n~O) one obtains that for any nonneg-
ative integer n,
IIhll2+ II E U!"'-1)(U. - r)h",11 2

n
"'.1 n
= IIhll2+ E IIDt. h",1I 2

"'=1
= II Dii1rn D~h1l2 .
Since T·n !: 0 , we get k. = 0 and the proof is complete.
Remark. This paper constitutes part of the author's Ph.D. thesis written at
•
Texas Ak M University under the direction of Professor Carl Pearcy.
References
[1]. V.M. Adamjan, D.Z. Arov, and M.G. KreYn, Infinite Hankel matrices and
generalized problems of Caratheodory-Fejer and F. Ries%, Funkcional. Anal. i
Priloren. 2 (1968), 1-19. [Russian]
[2]. V.M. Adamjan, D.Z. Arov, and M.G. Krein, Infinite Hankel block matrices
and some continuation problems, Izv. Akad. Nauk. Armjan. SSR. Ser. Mat. 6
(1971), 87-112. [Russian]
[3]. T. Ando, Z. CeaUljellCU, and C. Foi&lj, On intertwining dilations, II, Acta
Sci. Math. (Szeged) 39 (1977), 3-14.
[4]. R. Gadidov, Two problems of operatorial eztrapolation, An. Univ. Craiova
Mat. Fiz.-Chim. 12 (1984), 18-26.
[5]. J.B. Garnett, Bounded analytic functions, Academic Press, New York, 1981.
[6]. D. Sarason, Generalized interpolation in Hoo, 'frans. Amer. Math. Soc.
127 (1967), 179-203.
[7]. B. Sz.-Nagy, and C. Foi&lj, Harmonic analysis of operators on Hilbert space,
North Holland, Amsterdam, 1970.
Texas A& M University
College Station, Texas, 77843
OPERATOR SEMIGROUPS, INVARIANT SETS AND
INVARIANT SUBSPACES
JOHN FROELICH AND MICHAEL MARSALLI
1. INTRODUCTION
Let B(H) be the algebra of bounded operators on a complex, separable

Hilbert space H. Let (B(H)h = {t E B(H) : lit II ~ I}. We will say I:
is an operator semigroup if I: is a unital, absoultely convex subsemigroup
of (B(H)h. In particular, if A is a unital subalgebra of B(H), then Al =
{a E A : lIall ~ I} is an operator semigroup. If I: is an operator semigroup,
then I: has a natural action on HI, the unit ball of H, namely O"(s, x) =
sx, s E I:, x E H. Consequently, (I:, HI, 0") is a flow when HI is endowed
with the weak topology [1]. We apply ideas from the theory of dynamical
systems to study operator semigroups, guided by a close analogy between
the invariant sets of an operator semigroup I: and the invariant subspaces of
an operator algebra A. In particular, we will exhibit invariant set versions
of the transitive algebra problem and Lomonosov's Theorem. There are
also interesting connections between certain invariant sets of Al and the
hyperinvariant subspaces of A. Finally, we study norm precompact orbits
for various operator algebras, including self-adjoint algebras, triangular
algebras, CSL algebras, and the standard function algebras acting on an
L2 space. We summarize some of our results below. Full details will appear
in [3].
2. METRIC SPACES OF INVARIANT SETS
Let I(I:) denote the collection of norm closed convex subsets of HI

which are invariant under I:. Then I(I:) is a complete lattice under inclu-
sion. By convexity, each element of I(I:) is also weakly closed. For x E H,
the orbit of x is the set I:x = {sx : s E I:}. We lay the foundation for the
study of I(I:) by first considering it as a metric space. (Recall that the
collection of closed subsets of a metric space is also a metric space via the
Hausdorff metric [5].) Let C be the collection of norm closed convex sub-
sets of HI. Because the weak topology is metrizable on HI, the weak and
norm topologies induce Hausdorff metrics dw and du on C. Let Cc be the
norm compact elements of C, and let Ic(I:) be the norm compact elements
of I(I:). We collect some basic facts about these metric spaces.
Theorem l.
a) (C, dw ) is a compact metric space.
Operator semigroups 11
b) (I(E),d",) is a closed subspace of (e,d",).

c) (e, du ) is a complete metric space.
d) (I(E), du ), (ee, du ), and (Ie(E), du ) are closed subspaces of (e, du ).
3. TRANSITIVE OPERATOR SEMIGROUPS
In order to study ergodic properties of operator semigroups, we make

the following definitions. An operator semigroup E is n-transitive if for
every n-tuple of vectors XI, ... ,Xn and every a E (B(H)h there exists
sEE such that aXi = SXi for 1 ~ i ~ n. We say E is orthogonally
n-transitive if for every pair of n-tuples of orthonormal vectors eI, ... , en
and II, ... , fn there exists sEE such that sei = fi for 1 ~ i ~ n.
Finally, we say E is completely transitive if E is n-transitive for every n.
Clearly, when H is infinite dimensional, (B(H)h is completely transitive
and orthogonally n-transitive for every n. In fact, these ergodic properties
characterize (B(H)h as an operator semigroup, as the following result
shows.
Theorem 2. Let E be a weakly closed operator semigroup on an infinite

dimensional space H. The following are equivalent.
a) E is orthogonally n-transitive for all n.
b) E is completely transitive.
c) E = (B(H)h.
Note that E is I-transitive if and only if J(E) = {rHl : 0 ~ r ~ I},

and these are precisely the invariant sets for (B(H)h. As an analogue of
the transitive algebra problem, we have the following question.
'fransitive Semigroup Problem. Is (B(H)h the only weakly closed

1-transitive operator semigroup?
The answer is no. In fact, for each n there is a non-trivial weakly

closed n-transitive operator semigroup En. We sketch the construction.
Let Fn = {a E (B(H)h : rank a ~ n}. Let Gn be the closure of the
absolutely convex hull of Fn in the Hilbert-Schmidt norm. Then G n is a
weakly closed semigroup which is n-transitive. Finally, let En be the weak
closure of the absolutely convex hull of 1 and Gn • In contrast, if A is a
weakly closed operator algebra, and Al = {a E A : \lall ~ I} is I-transitive,
then A = B(H). Nevertheless, we have the following result, which is an
analogue of the transitive algebra problem for commutative algebras.
Theorem 3. If E is a commutative operator semigroup, and dimH ~ 2,

then E is not 1-transitive.
12 J. Froelich and M. Marsalli
We now construct another example of a I-transitive operator semigroup

which seems to be fundamental. Let C1(H) be the trace class operators
on H with trace class norm II . Ih· Let Tl = {t E C1(H) : IItlh ::; I}.
Then the weak closure of the absolutely convex hull of Tl and I is a non-
trivial I-transitive operator semigroup. We have the following analogue of
Lomonosov's Theorem.
Theorem 4. If L. is a weakly closed, i-transitive operator semigroup which

contains a compact operator k with Ilkll = 1, then TI C L..
In particular, when H is finite dimensional, TI is contained in every 1-

transitive operator semigroup on H. The previous theorem suggests that
TI is "minimal."
Question 1. If L. is a i-transitive weakly closed operator semigroup, 's

Tl C L. ?
4. MINIMAL INVARIANT SETS
The existence and properties of "minimal" invariant sets is a principle
theme in dynamical systems. For operator semigroups the only invariant
set which does not properly contain another invariant set is {O}. In order to
obtain a more fruitful notion of minimality, we will say that a set J E I(L.)
is a minimal invariant set for L. if the only invariant subsets of J are of the
form rJ with 0 ::; r::; 1. If M is an invariant subspace for L. and p is the
orthogonal projection of H onto M, let L.M = {sp : s E L.}. Then L.M is
an operator semigroup on M. We can characterize the minimal invariant
sets of L. as follows.
Theorem 5. A set J is a minimal invariant set for L. if and only if there

is a subspace M invariant for L. and a real number r with 0 ::; r ::; 1 such
that J = {x EM: IIxll ::; r} and L.M is i-transitive on M.
Because commutative operator semigroups cannot be I-transitive ex-

cept on a one dimensional space, we obtain the following corollary.
Corollary 6. Suppose L. is commutative. Then J is a minimal invariant

set if and only if J is generated by an eigenvector.
A somewhat surprising consequence of the existence of minimal in-

variant sets for the ball of an algebra is the existence of hyperinvariant
subspaces.
Theorem 7. Let A be a weakly closed unital subalgebra of B(H). If

A i= {AI : A E C} and Al has a nonzero minimal invariant set, then A
has a hyperinvariant subspace.
Operator semigroups 13
The previous corollary shows that there exist commutative operator

semigroups which have no minimal invariant sets. Consequently, we in-
troduce another type of minimality. A subset J of HI is projective if
sup{lIxll : x E J} = 1. A projective set J E I(~) is a minimal projec-
tive invariant set if J does not properly contain any projective invariant
sets. Unfortunately, a minimal projective invariant set might not be a
minimal invariant set. Furthermore, there exist commutative semigroups
with no minimal projective invariant sets. (Consider inner functions acting
on the Hardy space H2.) Nevertheless, the existence of minimal projec-
tive invariant sets can be guaranteed for a fairly large class of operator
semigroups.
Theorem 8. Every norm compact projective invariant set contains a min-
imal projective invariant set.
Corollary 9. If ~ is strongly compact, then ~ has a minimal projective
invariant set.
Once again, we can obtain hyperinvariant subspaces from a ntinimal-
ity condition; in this case, the existence of sufficiently "large" minimal
projective invariant sets.
Theorem 10. Let A be a weakly closed unital subalgebra of B(H). If J is
a minimal projective invariant set for Al such that J contains two linearly
independent unit vectors, then A has a hyperinvariant subspace.
5. NORM PRECOMPACT ORBITS
In view of the previous results, we turn our attention to the con-
sideration of norm compact invariant sets. For a unital algebra A, we
will consider the set of vectors with norm precompact orbits. So let
Mp(A) = {x E H : Alx is norm precompact}. The set Mp(A) is in fact an
invariant and hyperinvariant subspace for A [4]. We can characterize the
space Mp(A) for a wide variety of well known operator algebras.
Theorem 11. If A is a self-adjoint algebra, a triangular algebra or a
CSL algebra, then Mp(A) is the span of the finite dimensional invariant
subspaces of A.
Let I' be a positive finite Borel measure on the complex plane with
compact support K. Let A be a unital sub algebra of Loo(p.). We consider
A as an algebra of multiplication operators on L2(p.). Then Mp(A) = H if
and only if the natural injection A - L2(p.) is a compact operator. So we
seek necessary and sufficient conditions on the measure I' for this injection
to be compact. We summarize our results for the usual function algebras
as follows. (See [2] for notation and terminology.)
14 J. Froelich and M. Marsalli
Algebra Conditions
LOO(Il) Il
discrete
C(K) Il
discrete
P(K) Il
restricted to 8e I< discrete
R(K) Il
restricted to 8I< discrete
A(K) Il
restricted to 8I< discrete
POO(Il), Il cnr 1l(8L) = 0, where L is the
spectrum of z in POO(Il)
Question 2. For A a unital subalgebra of LOO(Il), when is the natural

injection A -+ L 2 (1l) a compact operator?
In closing, we give an application of these ideas to the theory of subnor-

mal operators (cf. [2]). Let W (S) be the weak closure of the polynomials
in S, where S is a subnormal operator on H. Recall that W(S) is an-
tisymmetric if W(S) n W(S*) consists only of scalar multiples of 1. If,
in addition, Mp(W(S)) = H, then the lattice of invariant subspaces is
enormous.
Theorem 12. If S is a subnormal operator on H, W(S) is antisymmetric

and Mp(W(S)) = H, then Lat W(S) contains a sublattice isomorphic to
the lattice of subspaces of H.
REFERENCES
1. J. Berglund, H. Junghenn and P. Milnes, Analysis on Semigroups,

Wiley-Interscience, New York, 1989.
2. J. Conway, Subnormal Operators, Pitman Res. Notes Math. Ser.,
no. 56, Pitman, Boston, 1981.
3. J. Froelich and M. Marsalli, Operator semigroups, flows and their
invariant sets, J. Functional Analysis, 115 (1993), 454-479.
4. M. Marsalli, A classification of operator algebras, J. Operator Theory,

24 (1990), 155-163.
5. S. Nadler, Jr., Hyperspaces of Sets, Marcel Dekker, New York, 1978.
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF HOUSTON, UNI-

VERSITY PARK, HOUSTON, TEXAS 77204
DEPARTMENT OF MATHEMATICS, ILLINOIS STATE UNIVERSITY, NOR-
MAL, ILLINOIS 61790
THE LOCAL DE BRANGES-ROVNYAK CONSTRUCTION
AND COMPLETE NEVANLINNA-PICK KERNELS
Scott McCullough
o. Introduction.
The purpose of this paper is to characterize finite di-
mensional complete Nevanlinna-Pick interpolation kernels. This re-
sult complements a recent result of Agler [2] and the characterization
of finite dimensional Caratheodory interpolation kernels in [8]. Peter
Quiggin has also shown that the condition is sufficient for scalar val-
ued interpolation [12]. Further, an abstract finite dimensional version
ofthe de Branges-Rovnyak construction is presented. This construc-
tion generalizes versions of the de Branges-Rovnyak construction [7]
used by a number of authors, including Agler [5] and Muller [9] for
weighted shifts (diagonal kernels) satisfying various polynomial iden-
tities.
Let k denote the Szego kernel,
1
k(z,A) = --_.
1- zA
A finite dimensional version of the classical Nevanlinna-Pick Theo-
rem is the following. Let N distinct points AI, ... , AN in the unit disc,
an integer M < N, and v x v matrices Wl, ••• ,WM be given. If the
Pick matrix
is positive semidefinite, then there exists v x v matrices wM+I, ••• , WN

such that the matrix
is positive semidefinite.
Following Agler [3], it is natural to replace the Szego
kernel, or more precisely the matrix (l-~j .>.), with other kernels.
Accordingly, fix k = (kij )f.j=l a positive definite matrix, with ki,j ::F 0
16 S. McCullough
for all i,j. Normalize k by kl,j = ki,l = 1. Formally the finite

version of the Nevanlinna-Pick problem has not changed. Given
v x v matrices WI. ••• , WM such that
(0.1)
we are to find v x v matrix WM+I. •.• , WN such that
(0.2)
However, unlike for the Szego kernel, this may not always be possible.
k is said to be a complete Nevanlinna-Pick kernel [3], and NP kernel
for short, if for every M < N, every v and every choice of v x
v matrices WI. ••• , WM satisfying (0.1) there exists v x v matrices
WM+I. ••• , WN satisfying (0.2).
Let b = (~) and b' = (l-b i ,j). It is a result of Agler's
I"
that if -b' is a positive semidefinite matrix, then k is an NP kernel.

For example, if k is the Szego kernel and AI, ... , AN are distinct points
from the unit disc, then b = (1 - AjAi) and -b' = (AjAi), which is
obviously positive definite. Agler shows that the Dirichlet kernel
is an NP kernel and it turns out that the reproducing kernel for
"Dirichlet like" spaces are also NP kernels. In [12] Quiggin gives
a number of interesting examples of NP kernels which do not arise
from considering spaces of analytic functions. Our main result is
0.3 Theorem. If k is an NP kernel, then -b' is positive semidefinite.
Recently Agler has established a version of (0.3) for
reproducing kernel Hilbert spaces over domains in en (see [2]).
As in the case of the classical Nevanlinna-Pick prob-
lem, (0.3) has a natural operator theoretic formulation [14]. In the
classical case, where k is the Szego kernel, the adjoint of the unilat-
eral shift and the Sz.-Nagy-Foias and de Branges-Rovnyak Dilation
Theorems playa central role. Here the relevant operator (matrix) S
is defined as follows. The matrix k induces a Hilbert space structure
on CN via the inner product < x, y >= y*kx. Let H denote this
Hilbert space. Define S on H by S*ej = Ajej, where {e}, ... , eN}
is the standard orthonormal basis for CN and AI, ... , AN are given
distinct points in C. The corresponding dilation theorem is the fol-
lowing. Let T denote an N x N matrix with eigenvalues AI, ... , AN
in the unit disc D and corresponding eigenvectors hI, ,.. , h n • We can
define b(T*, T) by
The local de Branges-Rovnyak construction 17
It follows from a theorem of Agler [3] that if b(T*, T) ~ 0, then T

extends to S*(J1.) , J.L copies of S* for some J.L. We give an explicit de
Branges-Rovnyak type construct of such an extension.
Define, for M $ N, Let T denote the compression of S*
to LM, where LM denote the orthogonal complement of the span of
{eM+b ... ,eN} in H. It is not hard to see that T is then an M X M
matrix with spectrum Ab ... , AM. Our main application of the de
Branges-Rovnyak construction is to give a proof that, under the as-
sumption that -b' is positive semidefinite, T extends to (S*IHM )(M),
M copies, where HM is the span of {et. ... ,eM} in H.
It is worthwhile to interpret our discussion for the Szego
kernel k = (l-~; ,,/ In this case S* is unitarily equivalent to the
adjoint of the unilateral shift on the classical Hardy space H2 re-
stricted to the span of the vectors {kxj(A) = l-~;"}' As noted
before, in this case b = (1 - AjAi). If T is an M x M matrix with
spectrum Al! ... , AM, then
b(T*, T) = 1- T*T.
Thus, the condition that b(T*, T) is positive is that T is a contrac-
tion; and the conclusion is that T can be lifted, via the de Branges-
Rovnyak construction, to the adjoint of the unilateral shift. In this
case, the compressions of S* are particularly simple. If T is the
compression of S* to the orthogonal complement of the span of
{eM+1, ... ,eN}, then T is unitarily equivalent to the restriction of
S* to the span of {el' ... , eM}.
1. The local de Branges-Rovnyak construction.

Throughout the remainder of this paper we fix distinct
points A}, ... , AN in e and a positive definite matrix k = (k ij ){;j=l
satisfying k1,j = ki,l = 1 and k ij i= 0 for all i,j.
We use el, ... , eN to denote the standard basis for eN
and let H denote the Hilbert space eN with inner product deter-
mined by < ej, ei >= k ij • Let S denote the matrix from the intro-
duction, i.e., S*ej = 'Xjej.
An operator T on a Hilbert space H is said to extend
(or lift) to an operator J on a Hilbert space K if there exists an
isometry V : H -+ K such that VT = JV. The following is a local
version of the de Branges-Rovnyak construction. Versions of the de
Branges-Rovnyak construction have been employed by a number of
authors, including Agler [5] and Muller [9] who used it to establish
versions of (1.1) for certain weighted shifts.
18 S. McCullough
1.1 Theorem. Let T be a matrix acting on a Hilbert space 1l of

dimension J.l < 00. Suppose there exist subspaces £}, ... , £N of 1l
such that
(1) for x E £j, Tx = AjX; and
(2) if x E 1l, then there exists x j E £j such that x = LX j.
Let {hj,m} be a basis for the subspace £j of1l. If the matrix b(T*, T)
defined by
( *) < b(T* , T)h j,m, h i,n >:= « hj,m,k. hi,n

.
1,1
» '
is positive semidefinite, then T extends to S*(/J).
Local de Branges-Rovnyak construction. Define Jon 1l®H by Jx®

y = x ® S*y. J is simply S*(/J), where J.l is the dimension 1l. Since
(*) is positive semidefinite, there exists vectors x j,m E 1l such that
Define an operator V : 1l -+ 1l ® H by Vhj,m = Xj,m ® ej. To see

that V is an isometry, compute
= L aj,mai,n < hj,m, hi,n >= II L aj,m hj,mll 2•
It is straightforward to verify the relation VT = JV. 0

Recall from the introduction the matrices b = (~)
',I
and b' = bi,j - 1 (also recall we assume kl,j = ki,I = bI,j = bi,I
so that b' is obtained from b via the Cholesky algorithm, b~,j =
-bi,Ibj,I + bi,j.) In the remainder of this section, we indicate how to
apply the de Branges-Rovnyak construction to show that if -b' is
positive semidefinite, then k is an NP kernel. This also allows us to
gather, in a natural way, some facts needed in the next section.
Recall, for T an N x N matrix with spectrum AI, •.. , AN
and corresponding eigenvectors hI, ... , hN we have defined b(T*, T)
by (1.1)(*). The following lemma gives an alternate definition.
1.2 Lemma. Let T be a matrix acting on a Hilbert space 1t of

dimension Il < 00. Suppose there exist subspaces £}, ... , £N of 1t
such that
(1) for x E £j, Tx = AjX and
(2) if x E 1t, then there exists x j E £j such that x = 2: x j.
Let {hj,m} be a basis for the subspace £j of1t. If {fIH and {91H
are functions analytic in D such that
where b = .,
(kl. ), then
b(T*, T) = L 91(T)* fl(T).

I
Proof. We first observe that if f is a function analytic in D, then
f(T)hj,m = f(Aj)hj,m.
Thus, we may compute,
< L91(T)*fl(T)hj,m,hi,n >= LL91(Ai)f,(Aj)< hj,m,hi,n >

I iJ I
=bi,j < hj,m, hi,n >

= < b(T*, T)hj,m, hi,n >. 0
Recall that ej is a basis for eN such that < ej, ei > =

ki,j and S is the operator whose adjoint S* satisfies S*ej = Ajej.
The following is a simple, but important observation.
1.3 Proposition. b(S, S*), computed with respect to the basis ej,
is the rank one positive with every entry 1.
The following theorem explains the significance of the
condition -b' positive semidefinite in terms of the dilation theory of
S*.
1.4 Theorem. Suppose -b' is positive semidefinite and T is a ma-
trix (i.e., an operator on a finite dimensional Hilbert space). If T*
extends to S(/J), then b(T*, T) is defined and positive semidefinite
and T extends to S*(N/J).
Proof. What we mean by b(T*, T) is defined, is that T satisfies the
hypothesis of Theorem 1.1. To see this, let 1t denote the space that
20 s. McCullough
T acts on and let K denote the space that J := S*(J.I.) acts on. The
hypothesis is that there exists an isometry V : 1t -+ K such that
VT* = J*V. This can be reformulated as
where M is the invariant subspace for J* given by the range of V.

Let P denote the orthogonal projection of K onto M and Q = 1- P
the orthogonal projection onto Mol., the orthogonal complement of
M. If x is an eigenvector for J with corresponding eigenvalue A,
then, since Mol. is invariant for J, we have
TPx =PJPx
=PJPx + PJQx
=PJ(P + Q)x
=PJx = APX.
Thus, since J satisfies the hypothesis of (1.1) so does T.

If -b' = (1 - bi,j) is positive semidefinite, then there
exists vectors hs such that 00* - L: hsh: = b, where 0 is the vector
with each entry 1. Consequently, there exist analytic functions, also
denoted hs, such that hs(Aj) is hs(j), the j-th component of hs. We
are now in a position to compute, using Lemma 1.2,
b(T*, T) =I - L hs(T)* hs(T)

= 1- P L ha(J)* Pha(J)P.
The inequality ha(J)* Pha(J) $ hs(J)*hs(J) produces
b(T*, T) ~ I - P L hs(J)* hs( J)P

= Pb(S, S*)(IL) P.
Since b(S, S*) is positive semidefinite by (1.3), it follows that b(T*, T)
is positive semidefinite. An application of (1.1) shows that T extends
to S*(NJ.I.). 0
1.5 Theorem. If -b' is positive semidefinite, then k is an NP kernel.
Proof. Recall that we let H· denote the Hilbert space eN with the
inner product determined by < ej, ei >= kij where e}, ... , eN is a
basis for eN and that S is the matrix (operator on H) whose adjoint
S* satisfies S*ej = Ajej.
We argue by induction. Thus, we may assume N =

M + 1.
Let LM denote the orthogonal complement ofthe span of eN
in H and P the orthogonal projection onto LM. Compute
TPej =PS*Pej
=PS*(P + Q)ej
=PS*ej
=Ajej,
=
where Q 1- P, and where we have used PS*Qej =
0, since the
range of Q is invariant for S*. Thus, for j = 1, ... , M, Pej is an
eigenvector for T with corresponding eigenvector Aj.
T extends to (S*)(M), by (1.1). This means there exists
an isometry V : LM --+ K such that VT = JV, where J = S*M =
I ® S* and K = LM ® H. Since T Pej = AjPej, JV Pej = AjPej.
Thus, VPej = Xj ® ej for some vector Xj ELM. Therefore,
kij - kiNkNj = < Pej, Pei >
(1.6) = < VPej, VPei >
Therefore, if we are given v x v matrices WI. ••• , WM such that
(1.7)
is positive semidefinite, then, from (1.6) and (1. 7),
(1.8) (( J - w·w":)(k·
1 3 1,3. - k·I, NkN ,1.)))M
1
is positive semidefinite. From these two inequalities Quiggin [12]

shows that it follows from the Parrott Lemma [10] in a non-trivial
way that there exists w N such that
(( J _ w.w":)k
1 1
.. )N
1,1 1
is positive semidefinite. More generally, this is a consequence of an

abstract interpolation theorem of Agler [3]. 0
2. NP kernels.
In this section we establish (0.3) from the introduction.
Recall that k is an NP kernel if, whenever vxv matrices
WI. ••. , WM satisfying
22 S. McCullough
are given there exists v Xv matrices W M +1, ••. , W N such that
where A 2: 0 means the matrix A is positive semidefinite. Recall also

that b = (iJ) and b~j = bij - 1.
To prove (0.3), we must show that if k is a NP kernel,
then -b' is positive semidefinite. We argue by induction. Accord-
ingly, we assume that M = N - 1, that (ki,j)f is an NP kernel, and
that e' := (-b~,;)r is positive semidefinite. We must show (-b~,;)f
is positive semidefinite.
Since e' is positive, there exist a positive integer p and
vectors Xj E CP such that < Xj,Xi >= c'. In particular, Xl =
O. With e = (bi,j)r, we have 00· - 2: xsx: = e, where 0 is the
vector with each entry 1. Let Wj denote the p X P matrix with first
column given by the vector xj and 0 in all other entries. The matrix
(I - WiWJ) is the matrix with 1 - eLi in the (1,1) entry 1 down
the remainder of the diagonal, and 0 elsewhere. Compression of the
matrix
((J - wiwj)ki,i)7
to the (1, 1) entries obtains the matrix ((1- e~,i)ki,j) = (bi,jki,i)j i.e.,
the matrix with all entries 1. Consequently,
is a direct sum of the rank one positive with all entries 1 and the
matrix (ki,j!), where J is the identity on 0'-1 j and is thus positive
semidefinite. Since k is an NP kernel, there exists a p X P matrix W N
such that
((I - wiw;)ki,j)~
is positive semidefinite. Computing as above, we find that the matrix
1 ... 1 (1-<X1'X»k 1'N)

1 ... 1 (1- < X2,X »k2 ,N
··· ..
.
... (l-<x,x»kN,N
must be positive semidefinite, where x· is the first row of the matrix

W N.We must have
(1- < x,xi »ki,N = (1- < X1,X »k1,N

for all j = 1,2, ... , M. Since, Xl = 0 and kl,N = 1, this means

(1- < X, Xj > )kj,N = 1.
Consequently
1
1- -k- =< X,Xj >,
j,N
for j = 1,2, ... , M. Further, we must have (1- < X, X > )kN,N ~ 1.
Therefore
-b~,N ~< X,X >.
If we let XN = x, then
-b' = « Xj,Xi »i" + y,
where y is the matrix with (N, N) entry -b~,N- < x, X >~ 0 and
all other entries O. It follows that -b' is positive semidefinite. 0
3. A Remark.
In our definition of NP kernel the order of the points
>.t, ... , >'N was fixed. Provisionally say k is a strong NP kernel if for
any index set I C {I, 2, ... , N} and for any v X v matrices {Wi: i E I}
such that
( J - wiwj)ki,j) i,jET
is positive semidefinite, there exists matrices {Wi : i ~ I} such that
(J - wiw;)ki,j)~j=l
is positive semidefinite. It is immediate that a strong NP kernel is an
NP kernel. Moreover, if k is a strong NP kernel, then -b' is positive
semidefinite. The proof in section one that -b' positive semidefinite
implies k is an NP kernel also shows that in fact k is a strong NP
kernel. Hence, NP kernel and strong NP kernel are the same.
It is also worthwhile to point out that if it is assumed
that there exists a positive definite matrix P such that
(3.1)
where * denotes Schur product, then the inequality (1.7) implies the
inequality (1.8). The relation (3.1) is (1.6) and this is a consequence
of the assumption -b' ~ 0, which in turn is equivalent to the exis-
tence of a positive matrix Q such that
(3.2)
Of course, (3.2) is just (3.1) with N replaced by 1. This approach is
very close to that of Quiggin [12].
24 s. McCullough
REFERENCES
1. M.B. Abrahamse, The Pick interpolation theorem lor finitely connected do-
mains, Mich. Math. 26 (1979), 195-203.
2. J. Agler, A talk at the Southeastern Analysis Meeting (SEAM) in Knoxville
Tenne ....ee, March 1992.
3. ___ , Some interpolation Theorems 0/ Nevanlinna-Pick type" J. Operator
Theory (to appear).
4. ___ , Operator Theory and Nevanlinna-Pick interpolation, preprint.
5. ___ , Hypercontractions and Subnormality, Journal of Operator Theory
13 (1985), 203-217.
6. J .A. Ball, A lifting theorem lor operator models 0/ finite rank on multiply-
connected domains, Journal of Operator Theory 1 (1979), 3-25.
7. L. de Branges and J. Rovnyak, Appendix on square summable power series,
Canonical models in quantum scattering theory, in Perturbation Theory and
its Applications in Quantum Mechanics, Wiley, New York, 1966, pp. 347-392.
8. S. McCullough, Caratheodory interpolation kernels, Integral Equations and
Operator Theory 15 (1992).
9. V. Muller, Models/or operators using weighted shifts, Journal of Operator
Theory 20 (1988), 3-20.
10. S. Parrott, On a quotient norm and the Sz.-Nagy-Foias lifting theorem,
Journal of Functional Analysis 30 (1978), 311-328.
11. V.I. Paulsen, Completely bounded maps and dilations, Wiley, New York,
1986.
12. P.P. Quiggin, For which reproducing kernel Hilbert spaces is Pick's Theorem
true?, preprint.
13. J. Rovnyak, Some Hilbert spaces 0/ analytic functions, Yale Dissertation
(1963).
14. D. Sarason, Generalized interpolation in Hoo, American Math society Trans-
actions 27 (1967), 180-203.
University of Florida
201 Walker Hall
PO Box 118000
Gainesville FL 32611-8000
LOCAL SPECTRAL THEORY FOR MULTIPLIERS
AND CONVOLUTION OPERATORS
VIVIEN G. MILLER AND MICHAEL M. NEUMANN
ABSTRACT. This note centers around the class D(G) of decomposable measures
on a locally compact abelian group G. This class is a large subalgebra of the mea-
sure algebra M(G), has excellent spectral properties, and is related to a number
of concepts from commutative harmonic analysis. The discussion of D(G) in
Section 1 is to illustrate this point. The main features of this class are collected
in Theorem 1.1, which improves recent results from [19] and includes some new
properties related to the involution of M(G). We shall present a different ap-
proach, which avoids previous tools like the hull-kernel topology [19], [23] or the
spectral theory of several commuting operators [2], [10]. Theorem 1.1 is an im-
mediate consequence of the spectral theory for multipliers on Banach algebras in
Section 3. The emphasis is here on multipliers with the decomposition property
(6) from [3], which characterizes the quotients of decomposable operators. We
show that mUltipliers with property (6) behave very nicely and coincide with the
strongly decomposable multipliers under fairly mild conditions on the underlying
Banach algebra. Our results on multipliers require some new results on general
local spectral theory in Section 2, which should be of independent interest. In
particular, some basic results on decomposable operators from [8] and [28] will be
extended to the more flexible case of quotients and restrictions of decomposable
operators in the spirit of [3].
1. Decomposable Measures. Given a complex Banach space X and the Ba-

nach algebra C(X) of all bounded linear operators on X, an operator T E C(X)
is called decomposable if, for every open covering {U, V} of the complex plane
C, there exists a pair of T-invariant closed linear subspaces Y and Z of X such
=
that u(TIY) ~ U, u(TIZ) ~ V and Y + Z X, where u denotes the spectrum.
This definition of operator decomposability is equivalent to the slightly more
complicated original version introduced by Foialj in 1963 and covers numerous
interesting classes of operators. For instance, all spectral operators in the sense
of Dunford, all generalized spectral operators in the sense of Colojoara and Foialj,
and all operators with a totally disconnected spectrum are decomposable; see [8]
and [28] for a thorough treatment of spectral decompositions in operator theory.
In the present paper, we wish to discuss the notion of decomposability in the
context of multipliers on certain Banach algebras, which includes the case of
convolution operators on group algebras as an important special case.
For a locally compact abelian group G and a regular Borel measure JL E M( G),
let T" : LI(G) -+ LI(G) denote the corresponding convolution operator on the
group algebra LI(G) given by T,,(J) := JL * f for all f E LI(G). For a compact
group, it has been known for quite some time and was rediscovered by a number
26 v. G. Miller and M. M. Neumann
of authors that T,. is compact if and only if I' is absolutely continuous with
respect to Haar measure [1], [15]. However, a characterization of the slightly
more general class of those measures which induce Riesz convolution operators
has been found only recently in [19]. The difficulties which arise in the general
spectral theory of convolution operators are, of course, at the core of harmonic
analysis and closely related to the fact that the spectrum 0'(1') = O'(T,,) is, in
general, much larger than the closure of the range ii(r) of the Fourier-Stieltjes
transform ii on the dual group rj see for example [14], [30].
It seems natural to investigate the role of decomposabilty in this context and
to apply the theory of Banach algebras to gain more insight into convolution
operators. This kind of approach was initiated by Colojoara.-FoiCUj [8] and has
been pursued further, for instance, in [2], [9], [10], [18], [19], [20], [23], [24]. In this
note, we shall give a fairly self-contained proof of the following result. Indeed,
since the group algebra A = Ll (G) is a regular and symmetric semi-simple
commutative Banach *-algebra with a bounded approximate identity [26] and
since its multiplier algebra can be identified with M(G) via convolution [16],
Theorem 1.1 follows immediately from the general theory to be developed in
Section 3.
1.1 Theorem. Let D(G) consist of all measures I' E M(G) for which T,. is
decomposable on Ll (G). Then D( G) is a full and symmetric closed *-subalgebra
of the measure algebra M( G) and contains the greatest regular subalgebra of
M(G). Each I' E D(G) has a natural spectrum in the sense that 0'(1') = O'(T,.) =
ii(r)-. Moreover, the spectrum 6(D(G)) of D(G) contains the dual group r
and is a quotient of the spectrum 6(M(G)).
Thus the algebra D( G) of decomposable measures is large enough to contain,

for instance, all measures whose continuous parts are absolutely continuous and
also certain singular measures [12], [19], [23], [24]. On the other hand, as a sym-
metric subalgebra, D(G) avoids the Wiener-Pitt phenomenon [14] and hence is
considerably smaller than M(G) whenever G is non-discrete. More precisely, in
this case, G supports symmetric probability measures with independent convo-
lution powers [14], and it is easily seen that such measures cannot have natural
spectrum and hence are not decomposable [12]. It has also been observed in
[19] that D(G) is stable under absolute continuity and hence an L-subalgebra of
M(G). Moreover, any measure on G which is supported by a compact indepen-
dent set is decomposable if and only if it is discrete [12]. Thus, non-zero continu-
ous measures on compact Kronecker sets provide examples of non-decomposable
measures with natural spectrum [12]. Since compact Kronecker sets are Helson
sets [26], these examples do not belong to the class Mo(G) of all measures I'
in M( G) for which p. on r vanishes at infinity. The case of Mo( G) has been
completely settled in [2], [12], [19], where it was shown, among other things,
that Moo(G):= RadLl(G) := {I' E M(G): p. == 0 on 6(M(G))\r} is the great-
est regular sub algebra of Mo(G) and that D(G) n Mo(G) = Moo(G). Also, for
compact G, a measure I' E Mo(G) is decomposable if and only" if I' has natural
spectrum [19], but this equivalence ceases to be true for non-compact groups.
For instance, if 11 E Mo(R) is a symmetric probability measure with independent
Local Spectral theory for multipliers 27
powers and if 5 E M(Z) denotes the Dirac measure at 1, then it can be seen that
the product measure JL := v X 5 E Mo(R x Z) has natural spectrum, but JL is
not decomposable, since Lemma 4.7 of [19] shows that JL ~ Moo(R x Z). We are
indebted to Colin Graham (Thunder Bay) for providing us with this and some
other examples.
1.2 Open Problems. It would be interesting to obtain an internal measure-

theoretic characterization of the decomposable measures and their precise re-
lation to the greatest regular sub algebra Reg M(G) of M(G) and to the class
Dec M( G) of all JL E M( G) for which convolution on M( G) is decomposable.
From [19] we know that Reg M( G) ~ Dec M( G) ~ D( G), but it is not clear if
these inclusions may be strict. Also, Dec M( G) has been shown [23] to consist
precisely of all JL E M(G) for which jJ, is hull-kernel continuous on ~(M(G)), but
the measure-theoretic role of the hull-kernel topology in this context is not yet
fully understood. Similar questions arise for more general types of multipliers
[12], [19]. Finally, it may be of interest to relate D(G) and its spectrum to the
concept of analytic structure in ~(M(G)).
2. Quotients and Restrictions of Decomposable Operators. In this

section, we shall be concerned with two natural extensions of the notion of
operator decomposability. Given an arbitrary operator T E £(X) on a complex
Banach space X and a closed subset F of C, let XT(F) consist of all x E X for
which there is an analytic function f : C\F -+ X such that (T - >.)f(>.) = x
for all >. E C\F. Then T is said to have the decomposition property (5) if
X = XT(U-) + XT(V-) holds for every open covering {U, V} of C. According
to [3], this means precisely that T is similar to a quotient of a decomposable
operator, i.e. that there exist a decomposable operator R E £(Z) on some
Banach space Z and a continuous linear surjection Q E £(Z, X) such that
TQ = QR. It has also been shown in [3] that property (5) is closely related to
a classical property from [5]: the operator T is said to have Bishop's property
({3) if, for every open subset U of C and for every sequence of analytic functions
fn : U -+ X for which (T - >')fn(>') converges locally uniformly to zero on U, it
follows that fn(>') -+ 0 as n -+ 00, locally uniformly on U. Obviously, ({3) implies
that T has the single valued extension property (SVEP), which means that, for
every open U ~ C, the only analytic solution of the equation (T - >.)f(>.) = 0
for all >. E U is the constant f == 0, cf. [8].
It follows easily from Proposition 1.3.8 of [8] that an operator T E £(X) is
decomposable if and only if it has both properties ((3) and (5). More signifi-
cantly, Albrecht and Eschmeier [3] have recently shown that ({3) characterizes,
up to similarity, the restrictions of decomposable operators onto closed invariant
subspaces, that an operator T E £(X) satisfies ((3) if and only if its adjoint on
the dual space of X satisfies (5), and that the corresponding statement remains
valid if both properties are interchanged; see also [11]. These results have been
very useful in recent progress on the invariant subspace problem for operators on
Banach spaces; see for instance [13]. In the following, we shall extend some basic
results on decomposable operators from [8], [27], [28] to the case of operators
with ((3) or (5). These extensions will then be used to investigate the spectral
28 V. G. Miller and M. M. Neumann
properties of multipliers.
Note that, by Proposition 1.1 of [21], an operator T E leX) has SVEP if and
only if XT(F) = XT(F) holds for all closed F ~ C, where XT(F) := {z EX:
qT(Z) ~ F} and qT(Z) denotes the local spectrum ofT at Z in the sense of [8], i.e.
the complement of the set of all ~ E C for which there exists an analytic function
I: U - t X on some open neighborhood U of ~ such that (T - 1')/(1') = Z for
all I' E U. The spaces XT(F) seem to be more appropriate for certain general
questions oflocal spectral theory than the classical counterparts XT(F) from [8].
The following elementary properties will be useful: XT(0) = {OJ, XT( q(T» = X,
XT(F) = XT(q(T) n F), and XT(F) = X implies that T - ~ is surjective for
all ~ E C\F and hence that q(T) ~ F whenever T has SVEP. The last result
follows from the fact that surjective operators with SVEP have to be bijective
[29].
Let V(z,r) and V(z,r) denote the closed, resp. open disc in C with radius
r 2: 0, centered at z E C. We shall need the following extension of Corollary
2.3 in [17], concerning the local spectral radius of T E leX) at Z E X given by
PT(z) := limsuPn_oo IITnz I1 1 /n.
2.1 Proposition. For arbitrary T E .c(X), we have XT(V(O, r» = {z EX:
PT(Z) $ r} for all r 2: O. In particular, ifT has SVEP , then for each DOD-zero
Z E X we have the local spectral radius formula max{I~I: ~ E qT(Z)} =
PT(z).
Proof. Since XT(F) = XT(F) for all closed F ~ C whenever T has SVEP,
the last assertion follows immediately from the first. Now, let Z E X such
that PT(z) $ r. Then the infinite series 1(>') := _>.-I Z - ~-2Tz - ~-3T2z -
>. -4 T3 Z - ••• converges locally uniformly on C\ V (0, r) and hence defines an
X-valued analytic function on C\ V(O, r). Since (T - ~)/(~) = Z for all ~ E
C\ V(O, r), we conclude that:!: E XT(V(O, r». Conversely, let:!: E XT(V(O, r» be
arbitrarily given, and consider an analytic function I : C\ V(O, r) - t X such that
(T - ~)/(~) = Z for all >. E C\ V(O, r). Then, for ~ E C with 1>'1 > max{r,IITII},
we obtain I(~) = (T - ~)-IZ = _~-IZ - ~-2Tz - ~-3T2Z - >.-4T3z _ ... and
therefore 1(>') - t 0 as I~I - t 00. Hence the definition g(O) := 0 and g(p.) :=
1(1/1') for 0 < 11'1 < l/r yields an analytic function g from V(O,l/r) into
X with the property that g(p.) = -I' Z - p.2Tz - p.3T2z - p.4T3 z - ... for
11'1 < 1/ niax{r,IITII}. Since g is analytic on the disc V(O,l/r), it follows exactly
as in the scalar-valued setting from Cauchy's integral formula that the preceding
identity holds even for all I' E V(O,l/r). This shows, in particular, that the
radius of convergence of the power series representing g(p.) is 2: l/r. By the
standard formula for the radius of convergence of a vector-valued power series,
we conclude that PT(Z) $ r, which completes the proof.
Now, given a pair of operators T E leX) and S E leY) on complex Banach

spaces X and Y, respectively, let C(S, T) : .c(X, Y) - t l(X, Y) denote the
corresponding commutator defined by C(S,T)(A) := SA - AT for all A E
.c(X, Y). Further, for closed F ~ C and any 0 $ r < 00, let Fr := {z E C :
dist(z, F) $ r} = F+ V(O,r).
2.2 Proposition. Let T E £(X), S E £(Y), and A E £(X, Y) be arbitrarily

given, and let T := PC(s,T)(A). Then A XT(F) ~ !Ds(Fr) holds for all closed
F~C.
Proof. We use a basic idea from the proof of Theorem 2.3.3 in [8], see also [27].
Given Z E XT(F) and an analyticfunction f : C\F -+ X such that (T-.\)f(.\) =
Z for all.\ E C\F, we consider an arbitrary closed disc E:= V(z,s) in C\Fr
and choose a real number t > T + s such that the disc D := V(z, t) is contained
in C\F. Then K := sup{lIf(.\)1I : .\ E D} < 00, and, for each .\ E E, we obtain
from Cauchy's integral formula
for all n ~ O. Therefore
limsup sup
n_oo >'EE
Ilc(s,T)n(A/(n)~.\)lll/n
n.
:5 T/(t - s) < 1,
which implies that
converges uniformly on E and hence locally uniformly on C\Fr • We conclude

that the function 9 : C\Fr -+ Y is analytic. Finally. following the telescoping
argument from the proof of Theorem 2.3.3 of [8], we obtain (S - .\)g(.\) = Az
for all .\ E C\Fr and therefore Az E !Ds(Fr).
Under suitable assumptions on S and T, there is a converse to Proposition
2.2. This is included in the following result. which generalizes Theorem 4.1 of
[27]. Recall that an operator T E £(X) is said to have Dunford's property (C)
if XT(F) is closed for each closed F ~ C. It is well-known and easily seen that
this property follows from property (fJ). and it has been observed in [21] that
property (C) implies SVEP. Note that T is decomposable if and only if T has
both (C) and (6).
2.3 Theorem. Assume that T E £(X) has property (6) and that S E £(Y)
has Dunford's property (C). Then, for all A E £(X, Y) and T ~ 0, we have the
equivalences:
PC(s,T)(A):5 T <==? O'C(s,T)(A) ~ V(O,T)

<==? A XT( F) ~ !D s( Fr) for all closed F ~ C.
Proof. Since it has been shown in Theorem 2.4 of [21] that C(S, T) has SVEP,
the first equivalence is clear from Proposition 2.1. In view of Proposition 2.2, it
remains to be seen that the last statement implies the first, so assume that A
XT(F) ~ !lls(Fr) holds for all closed F ~ C. Since T has property (5), we know
from [3] that there exists a decomposable operator R E .l( Z) on some Banach
space Z and a continuous surjection Q E .l(Z,X) such that TQ = QR. Then
clearly (AQ)ZR(F) = (AQ)3R(F) ~ AXT(F) ~ !lls(Fr) = Ys(Fr) for all closed
F ~ C. Now, although Theorem 4.1 of [27] is stated only for the case of decom-
posable operators, it is easily seen that its proof goes through verbatim if the
operator on the range space has property (C). Hence PC(S,R) (AQ) ~ r. In view
of the open mapping theorem for the surjection Q, there exists some constant
M > 0 such that MIIBII ~ IIBQII for all B E .l(X, Y). Since C(S, R)n{AQ) =
C(S, T)R(A)Q and therefore MIIC(S, T)R(A)II ~ IIC(S, R)n(AQ)1I for all n E N,
we conclude that PC(s,T)(A) ~ PC(s,R)(AQ) ~ r.
2.4 Proposition. Let T E .l(X), S E .l(Y), and A E .l(X, Y) be arbitrarily

given, and let r := PC(s,T)(A). Then we have:
(a) IfS has property (C) and A has dense range, then u(S) ~ u(T)+V(O, r).
(b) 1fT has property (6) and A is injective, then u(T) ~ u(S) + V(O, r).
Proof. Under the assumptions of part (a), Proposition 2.2 implies that Y =
(AX)- = (AXT(U(T)))- ~ !lls(u(T) + V(O,r)). Since S has SVEP, this yields
u(S) ~ u(T) + V(O, r) by one of the properties of spectral subspaces mentioned
above. To prove assertion (b), let F := u(S) + V(O, r), and consider an arbitrary
~ E C\F. Then V(~, r) and u(S) are disjoint. Again from Proposition 2.2 and
basic properties of spectral subspaces, we obtain the inclusions Aker(T - ~) ~
AXT({A}) ~ !lls(V(~,r)) = !lls(V(~,r) n u(S)) = !lls(0) = {OJ. Since A is
injective, we conclude that ker(T - ~) = {OJ so that T - ~ is one-to-one. To
show the surjectivity of T - ~, we choose open neighborhoods U of F and V
of ~, respectively, such that U U V = C, F n V- = 0, and ~ f/. U-. Since T
has property (5), we obtain X = XT(U-) + XT(V-). Also, from Proposition
2.2 and F n V- = 0, we conclude that AXT(V-) ~ !lls(V- + V(O,r)) =
!lls«V- + V(O, r)) n u(S)) = !lls(0) = {OJ. The injectivity of A implies that
XT(V-) = {OJ and therefore X = XT(U-), Since ~ f/. U- by our choice of U,
the last identity shows that T - ~ is surjective. Hence C\F ~ C\u(T), which
completes the proof.
Part (a) of the preceding result implies, in particular, that quasi-similar sub-
decomposable operators on Banach spaces have the same spectrum. Since hy-
ponormal operators are known to be subscalar [25], this subsumes, for instance,
a classical result due to Clary [7]. Later we shall use the following special case
of part (b), which improves a recent result from [20]: ifT has property (6) and
A is injective such that SA = AT, then u(T) ~ u(S). We next show that the
properties (6) and (fj) are preserved under certain limits.
2.5 Proposition. Let S E .l(Y), and assume that the operators Tk E .l(X)
have property (6) and Ak E .l(X, Y) are surjective for all kEN. If rk :=
PC(S,Tk)(Ak) -+ 0 as k -+ 00, then also S has property (6).
Proof. Given an arbitrary open covering {U, V} of C, we choose closed F, G ~ C

with F ~ U and G ~ V so that C is covered by the interiors of F and G.
Since rk --+ 0 as k --+ 00, there exists a kEN such that Fr. n O'(S) c U
and Gr. n O'(S) ~ V. From (5) for Tk and Proposition 2.2, we conclude that
Y = AkX = AkXT. (F) + AkXT• (G) ~ !Ds(Fr.) +!Ds(Gr.) = !Ds(Fr. n O'(S)) +
!Ds(Gr• n O'(S)) ~ !Ds(u-) + !Ds(V-), which shows that S has (5).
2.6 Corollary. Let T E leX), and assume that the operators Sk E leY)
have property ((3) and Ak E leX, Y) are bounded below for all kEN. If
PC(s.,T)(A k ) --+ 0 as k --+ 00, then also T has property ((3).
Indeed: we know from [3] or [11] that the properties ((3) and (5) are dual to
each other, hence the assertion follows from Proposition 2.5 by a straightforward
duality argument. The most important special case arises for X = Y and Ak =
I, the identity operator on X. Recall from [28] that the spectral distance of two
operators S, T E leX) is defined to be pes, T) := ma.x{pC(s,T) (I), PC(T,s)(I)}.
Note that peS, T) coincides with the spectral radius of S - T whenever Sand T
commute. We obtain the following generalization of a classical limit result due
to Apostol; cf. Theorem IV.6.10 of [28].
2.7 Theorem. Let T k , T E leX) such that rk := p(Tk , T) --+ 0 as k --+ 00. If
Tk has property (5) for each kEN, so does T. The same conclusion holds with
respect to the properties ((3), (C), SVEP, and decomposability.
Proof. The case of SVEP follows easily by means of Proposition 2.2, but has
already been settled in Theorem 3.2 of [27]. In view of the preceding results, it
remains to consider the case of Dunford's property (C). In this case, we know
that T and Tk have SVEP for all kEN so that the XT(F)-spaces coincide with
the XT(F)-spaces. We conclude that T has (C), since a double application of
Proposition 2.2 shows that, for all closed F ~ C,
00 00
k=l k=l
Recall from [8J that two operators S, T E leX) are called quasi-nilpotent
equivalent if p( S, T) = O. Obviously, Theorem 2.7 implies that each of the
properties (5), ((3), (C), SVEP, and decomposability is preserved under quasi-
nilpotent equivalence. This improves classical results from [8J. We close this
section with another kind of permanence.
2.8 Theorem. Let T E leX), and consider an analytic function f : U --+ C

on an open connected neighborhood U of O'(T). IfT has property (5), so does
f(T). The same conclusion holds with respect to the properties ((3), (C), SVEP,
and decomposability.
Proof. The case of SVEP, Dunford's property (C), and decomposability is clear
from Theorem 1.1.6 and 2.1.10 of [8]. If T has ((3), then the construction of [3]
shows that there exists a decomposable extension S E leY) ofT on some Banach
space Y so that O'(S) is contained in the given open connected neighborhood
U of O'(T). Hence f(5) E £(Y) is well-defined and decomposable by [8]. Thus

f(T) = f(5)IX has (f3). Finally, the case of property (6) follows by an obvious
duality argument.
3. Spectral Properties of Multipliers. Given an arbitrary complex Banach

algebra A, let Mr(A) consist of all T E £(A) which commute with all right
multiplication operators on A in the sense that T(ab) = (Ta)b for all a,b EA.
Similarly, let MI(A) denote the left mulipliers on A, and let M(A) := Mr(A) n
MI(A) be the set of all two-sided multipliers on A. IT A is without order, then
this definition of M(A) coincides with the standard notion from [16]. Finally,
let A2 denote the linear span of all products ab for a, b E A. We shall see that
the condition A2 = A implies remarkable permanence properties for multipliers
on A with property (6). By the Cohen factorization theorem [6], this condition
is fulfilled if A has a left or right bounded approximate identity. To put the
following result into perspective, let us note that it is still an open problem
whether the sum and product of two commuting decomposable operators are
again decomposable; see also [2], [4], [10], [18], [20] for related results.
3.1 Theorem. Assume that the Banach algebra A satisfies A2 =

A. If 5 E
=
Ml(A) and T E Mr(A), then 5T T5. Moreover, if 5 and T have (6), so do
5 +T and 5T.
Proof. First observe that 5T(ab) = 5«Ta)b) = (Ta)(5b) = T(a5b) = T5(ab)

for all a, b E A. Hence the condition A2 = A implies that 5 and T commute.
Now assume that 5 and T have (6). To show'that 5 + T has (6), we consider
a pair a, b E A. Given an arbitrary open covering {U, V} of C, we choose
closed sets F, G ~ C with F ~ U and G ~ V such that the interiors of F
and G cover C. Then we find an r > 0 such that Fr n 0'( 5 + T) ~ U and
G r n 0'(5 +T) ~ V and cover O'(T) by finitely many open discs with radius r and
centers .\l""'.\m E C, respectively. Since T has (6), there exist a decomposable
operator R E £(Z) on some Banach space Z and a surjection Q E £(Z,A) such
that RQ = QT. Let W ~ C be open such that W- n O'(T) = 0 and C = W U
V(.\t,r)U·· ·UV(.\m,r). Since QZn(W-) = Q3n(W-) ~ ~T(W-) ~ ~T(W-n
O'(T» = ~T(0) = {OJ and since R is decomposable and Q is surjective, we obtain
A = ~T(V(.\lJr» + ... + ~T(V(.\m,r» and therefore a = a1 + ... + am with
suitable ak E ~T(V(.\k,r» for k = 1, ... ,m. Then clearly ak E ~T_>..(V(O,r»
and consequently, by Proposition 2.1, PT->..(ak) ~ r for k = 1, ... ,m. Let
Lk E £(A) be given by Lk('II.):= ak'll. for all'll. E A and k = 1, ... ,m. From
the assumptions 5 E Mi(A) and T E Mr(A), we obtain PC(s+T,s+>..)(Lk) =
PC(T,>..)(Lk) ~ PT->..(ak) ~ r. Hence, by Proposition 2.2, Lk~S+>..(F) ~
2ls+T(Fr) = ~S+T(Fr n 0'(5 + T» ~ ~S+T(U-) and similarly Lk~S+>..(G) ~
2lS+T(V-) for k = 1, ... , m. Since 5 + .\k has property (6), we also know that
A = ~s+>..(F) + ~s+>..(G) and therefore Lk(b) E ~S+T(U-) + !2lS+T(V-) for
k = 1, .. . ,m. This implies that ab = L1(b) + ... + Lm(b) E ~S+T(U-) +
2lS+T(V-), Because of A2 = A, we conclude that A = ~S+T(U-)+~S+T(V-),
which shows that 5 + T has (6). By Theorem 2.8, it follows that f(.\,/-') :=
exp(.\5 + /-,T) has (6) for all .\,/-, E C. But 5T = T5 implies that 5T =
8)..8IJ f().,p.)(0,0). Thus ST has (0) by Theorem 2.7.

3.2 Corollary. If the Banach algebra A satisfies A2 = A, then the set Mo(A)
of all multipliers T E M(A) with property (0) is a closed full commutative
subalgebra of C(A).
Proof. By Theorem 2.7 and 3.1, Mo(A) is a closed sub algebra ofthe commutative
Banach algebra M(A). Moreover, if T E Mo(A) is bijective, then its inverse is
obviously a multiplier and has property (0), since Theorem 2.8 may be applied
to the analytic function given by f().) := 1/), for all), "I- O. This shows that the
subalgebra Mo(A) is full.
To relate the class Mo(A) to the more restricted class D(A) of all decompos-
able multipliers T E M(A), we assume the Banach algebra A to be semi-prime in
the sense that every two-sided ideal J of A with J2 = {OJ is trivial [6]. This con-
dition will ensure that all multipliers on A have SVEP. Note that all semi-simple
algebras are semi-prime and that a commutative algebra is semi-prime if and
only if it contains no non-zero nilpotent element. Hence the following theorem
covers also examples like the radical weighted convolution algebras L1(R+,w)
on the half-line, since, by the Titchmarsh convolution theorem, these algebras
are integral domains. In the special case of a semi-simple commutative Banach
algebra with a bounded approximate identity, the identity Mo(A) = D(A) has
recently been established in [20] with a completely different approach. A de-
composable operator T E C(X) on a Banach space X is strongly decomposable
if TIXT(F) is decomposable for each closed F ~ C.
3.3 Theorem. Assume that the Banach algebra A is semi-prime. Then every
multiplier T E M(A) has SVEP. If, in addition, A has a bounded left approxi-
mate identity, then, for every T E Mo(A) and every closed left ideal B in A, the
restriction S := TIB E C(B) is decomposable on B and satisfies Bs(F) Es(F)=
with Es(F) := {u E B : au = 0 for all a E A with O'T(a) n F = 0} for all closed
F ~ C. In particular, every T E Mo(A) is strongly decomposable on A, and
D(A) is a closed full commutative subalgebra of C(A).
Proof. Let T E M(A) and :z: E A such that T2:z: = O. Since (T:z:)2 = :z:T2:z: = 0
and (T:z:)u(T:z:)= =
T(:z:uT:z:) =
:z:uT2:z: 0 for all u E A, we conclude that the
two-sided ideal J generated by T:z: satisfies J2 = {OJ and therefore J = {OJ
because A is semi-prime. Hence T2:z: = 0 implies T:z: = 0, which shows that
ker(T - ).)2 = ker(T - ).) for all ). E C. Thus T has finite ascent and hence SVEP
by Proposition 1.8 of [17]. Now, assume that A has a bounded left approximate
identity, and consider a multiplier T E Mo(A) and a closed left ideal B in A. By
the module version of the Cohen factorization theorem [6], it follows easily that
TB ~ B. To prove the decomposability of S := TIB, by Theorem IV.4.28 of
[28], it suffices to show that Es is a spectral capacity for S. Clearly, Es(0) = {OJ
and Es(C) = B. We also have B = Es(UJ) + ... + Es(U;;;) for every finite
open covering {UI, ... , Urn} of C. Indeed, given an arbitrary u E B, we first
use the Cohen factorization theorem to obtain a E A and v E B such that u =
avo Then, as in the proof of Theorem 3.1, property (0) yields a representation
a = a1 + ... + am for suitable ak E AT( Uk) for k = 1, ... , m. This implies that
u = al1 = a1"+·· +am l1 with CTT(akl1) ~ CTT(ak) ~ Uk and hence ak" E ES(Uk)
for k = 1, ... , m. We next claim that, for every collection {Fa: a E J} of closed
subsets of C with intersection F, we have Es(F) = naeJEs(Fa). Here the
inclusion ~ is obvious, and to see the converse let u E na€JEs(Fa) and a E A
with CTT(a) n F = 0 be arbitrarily given. By compactness, there exist finitely
many ab ... , am E J such that CTT(a) n Fa, n··· n Fa m = 0. Again by property
(6) for T, we obtain ao, ab ... , am E A such that a = ao + a1 + ... + am,
CTT(ao) n CTT(a) = 0, and CTT(ak) n Fa~ = 0 for k = 1, ... ,m. Because of
u E Es(Fa~), we conclude that aku = 0 for k = 1, ... , m and consequently
au = aou = 0, since CTT(au) = CTT(aoU) ~ CTT(a) n CTT(ao) = 0. Thus, as desired,
u E Es(F). We finally show that, for each closed F ~ C, the 5-invariant closed
subspace Es(F) satisfies CT(5IE s (F)) ~ F. Given any A E C\F and u E Es(F),
we use again the module version of the Cohen factorization theorem to write
u = al1 with a E A, 11 E Es(F). Then, by (6) for T, we obtain 6, c E A such that
Aft CTT(C), CTT(6) n F = 0, and a = 6 + c. Now c = (T - A)d for some d E A and
therefore u = al1 = (6 + C)l1 = Cl1 = (T - A)(d)l1 = (5 - A)(dl1) with dl1 E Es(F).
This shows that 5IEs(F) is surjective and hence bijective by SVEP, see [29].
We now conclude from Chapter IV of [28] that 5 is decomposable and that
Bs(F) = Es(F) for all closed F ~ C. In particular, taking B := A, we see
that T is decomposable. Also, since B := AT(F) = ET(F) is a closed ideal
in A for each closed F ~ C, we conclude from the above that T is strongly
decomposable. The last assertion follows from Corollary 3.2 and the Cohen
factorization theorem.
We finally specialize to the case that A is a semi-simple commutative Banach

algebra with spectrum .!l(A). Then M(A) is also a semi-simple commutative
Banach algebra, A can be regarded as an ideal in M(A), and toeA) can be
canonically embedded in the spectrum to(M(A)) of the multiplier algebra, see
[16] for details. A Banach algebra A endowed with an algebra involution * is
called a Banach *-algebra. If A is semi-simple, then such an involution on A is
automatically continuous [6]. Also, if A is a semi-simple commutative Banach *-
algebra, then so is M(A) with respect to the involution given by T*(u) := T(u*)*
for all T E M(A) and u E A. Finally, recall that A is said to be symmetric if
CT(U) ~R for all '1£ E A with u = '1£*. The case of group algebras shows that, in
general, symmetry does not carryover from A to M(A).
3.4 Theorem. Assume that A is a semi-simple commutative Banach algebra.

Then Reg M(A) ~ M.s(A) , where Reg M(A) denotes the greatest regular Ba-
nach subalgebra of M(A), and each T E M.s(A) has a natural spectrum in the
sense of CT(T) = T(to(A))-. If, in addition, A is a symmetric Banach *-algebra
with A2 = A, then M.s(A) is a fuII and symmetric closed *-subalgebra of M(A)
whose spectrum is a quotient of to(M(A)).
Proof. The existence of Reg M(A) is clear from [19] or [23], and Theorem 2.3
of [18] shows that RegM(A) ~ D(A) ~ M.s(A), see also [19]. Now, given
T E M.s(A), let 5 denote the operator of multiplication by Tlto(A) on the
Banach algebra Co(L\(A». Since the Gelfand transformation G from A into

Co(L\(A» is injective and satisfies SG = GT, we conclude from Proposition 2.4
that T(L\(M(A))) = u(T) ~ u(S) ~ T(L\(A»-, which shows that T has natural
spectrum. If A is a Banach *-algebra and T E M(A), then AT. (F*) = {u* : u E
AT(F)} for a.ll closed F ~ C. Thus T* E M6(A) whenever T E M6(A). Hence,
if A is a Banach *-algebra with A2 = A, then, by Corollary 3.2, we conclude that
M6(A) is a closed full *-subalgebra of M(A). Now assume, in addition, that A
is symmetric. Given T E M(A) and cp E L\(A), let u E A so that cp(u) ¥- O.
By Theorem 35.3 of (6), (T*ncp)cp(u) = cp(T*(u» = cp(T(u*)*) = cp(T(u*»* =
(T(cp)cp(u*»* = T(cp)*cp(u) and therefore (T*)~= (F)* on L\(A). Hence, if
T = T*, then T is real-valued on L\(A) and thus u(T) ~ R whenever T has
natural spectrum. Consequently, the symmetry of A implies that B := M6(A)
is symmetric. By the Stone-Weierstrass theorem, Ii is dense in C(L\(B», which
implies that the Choquet boundary and hence the Shilov boundary of B is the
whole of L\(B). Hence, by a result of Shilov, each cp E L\(B) extends to some
.,p E L\(M(A», cf. Theorem 13.3 of (22). Thus L\(B) ~ L\(M(A»/ "', where the
equivalence relation", on L\(M(A» is, of course, given by .,p1 '" .,p2 if and only
if.,p1 == .,p2 on L\(B).
In particular, if A is a semi-simple commutative Banach *-algebra which is

both regular and symmetric and has a bounded approximate identity, then the
preceding results show that D(A) is a full and symmetric closed *-subalgebra
of M(A) whose spectrum is a quotient of L\(M(G}). Moreover, by regularity,
D(A) contains A as an ideal which implies that L\(A) is canonica.lly embedded in
L\(D(A». This establishes Theorem 1.1. We finally note that similar results can
be obtained for Reg M(A) and also for the algebra Dec M(A) of all T E M(A)
which induce decomposable multiplication operators on M(A). It remains open
for which algebras A the inclusions Reg M(A) ~ Dec M(A) ~ D(A) from (19)
are actua.lly identities. The authors are indebted to Garth Dales (Leeds) for
valuable discussions on the topics of this note.
REFERENCES
1. C. A. Akemann, Some mapping properties 0/ the group algebras 0/ a compact group, Pacific
J. Math. 22 (1967), 1-8.
2. E. Albrecht, Decomposable systems 0/ operators in harmonic analysis, Toeplitz Centennial,
Birkhauser-Verlag, Basel, 1982, pp. 19-35.
3. E. Albrecht and J. Eschmeier, Analytic functional models and local spectral theory, (sub-
mitted).
4. C. Apostol, Decomposable multiplication operators, Rev. Roumaine Math. Pures Appl. 17
(1972), 323-333.
5. E. Bishop, A duality theorem/or an arbitrary operator, Pacific J. Math. 9 (1959), 379-397.
6. F. F. Bonsall and J. Duncan, Complete Normed Algebras, Springer-Verlag, New York,
1973.
7. W. S. Clary, Equality 0/ spectra 0/ quasi-similar hyponormal operators, Proc. Amer. Math.
Soc. 63 (1975), 88-90.
8. I. Colojoarl and C. Foia§, Theory 0/ Generalized Spectral Operators, Gordon and Breach,
New York, 1968.

9. J. Eschmeier, Operator decomposability and weakly continuous representations o/locally
compact abelian groups, J. Operator Theory 7 (1982), 201-208.
10. _ _ _ , Spectral decompositions and decomposable multipliers, Manuscripta math. 51
(1985), 201-224.
11. _ _ _ , A nalytische Dualitat und Tensorprodukte in der mehrdimensionalen Spektralthe-
orie, Habilitationsschrift, vol. 2.42, Schriftenreihe des Math. Inst. Universitat Miinster,
Miinster, 1987.
12. J. Eschmeier, K. B. Laursen and M. M. Neumann, Multipliers with natural local spectra
on commutatille Banach algebras, (submitted).
13. J. Eschmeier and B. Prunaru, Inllariant subspaces lor operators with Bishop's property
(f3) and thick spectrum, J. Functional Anal. 94 (1990), 196-222.
14. C. C. Graham and O. C. McGehee, Essays in Commutatille Harmonic Analysis, Springer-
Verlag, New York, 1979.
15. J. W. Kitchen, The almost periodic measures on a compact abelian group, Monatsh. Math.
72 (1968), 217-219.
16. R. Larsen, An Introduction to the Theory 0/ Multipliers, Springer-Verlag, New York, 1971.
17. K. B. Laursen, Operators with finite ascent, Pacific J. Math. 152 (1992), 323-336.
18. K. B. Laursen and M. M. Neumann, Decomposable operators and automatic continuity,
J. Operator Theory 15 (1986), 33-51.
19. _ _ _ , Decomposable multipliers and applications to harmonic analysis, Studia Math.
101 (1992), 193-214.
20. _ _ _ , Local spectral properties 0/ multipliers on Banach algebras, Arch. Math. 58 (1992),
368-375.
21. _ _ _ , Asymptotic intertwining and spectral inclusions on Banach spaces, to appear in
Czech. Math. J. (1993/94).
22. M. A. Naimark, Normed Algebras, Wolters-Nordhoff, Groningen, 1972.
23. M. M. Neumann, Commutatille Banach algebras and decomposable operators, Monatsh.
Math. 113 (1992), 227-243.
24. _ _ _ , Banach algebras, decomposable conllolution operators, and a spectral mapping
property, Function spaces, Marcel Dekker, New York, 1992.
25. M. Putinar, Hyponormal operators are subscalar, J. Operator Theory 12 (1984), 385-395.
26. W. Rudin, Fourier Analysis on Groups, Interscience Publishers, New York, 1962.
27. F.-H. Vasilescu, Some properties 0/ the commutator 0/ two operators, J. Math. Anal. Appl.
23 (1968), 440-446.
28. _ _ _ , Analytic Functional Calculus and Spectral Decompositions, Editura Academiei
and D. Reidel Publishing Company, Bucure§ti and Dordrecht, 1982.
29. P. Vrbova, On local spectral properties 0/ operators in Banach spaces, Czech. Math. J. 23
(1973), 483-492.
30. M. Zafran, On the spectra 0/ multipliers, Pacific J. Math. 47 (1973), 609-626.
MISSISSIPPI STATE UNIVERSITY, P.O. DRAWER MA, MISSISSIPPI STATE, MS 39762, U.S.A.
E-mail: neumann@math.msstate.edu
Operator-Valued Poisson Kernels and
Standard Models in Several Variables
F.-H. Vasilescu
1. INTRODUCTION
Let 'Ji be a complex Hilbert space, and let £('Ji) be the algebra of all
bounded linear oeprators acting on 'Ji. For each T E £('Ji) we denote by
a(T) the spectrum of T. Let lD c C be the open unit disc. If T E £('Ji) is
such that a(T) C lD, then for each analytic polynomial I (more generally,
for each I holomorphic in lD and continuous on Ji)) we have (see [Vas])
(1.1) I(T) = r
laD
l(w)(1 - wT*)-l(I - T*T)(I - wT)-lda(w) ,
where a is the normalized Lebesgue measure on alD. If, moreover, IITII $ 1,

then
(1.2) P(T, w) = (I - wT*)-l(1 - T*T) (I - wT)-l , wElD
is positive for all wElD.
Definition. If T E £(1t), IITII $ 1 and a(T) C lD, then P(T, w), given
by (1.2), is called the opemtor-valued Poisson kernel associated with T.
Consequences of (1.1). 1° If I is holomorphic in IDl, continuous on

Ji), and Rei ~ 0, then ReI(T) ~ o.
2° II/(T)II $ 1111100, i.e., the von Neumann inequality, which can be
easily derived from 1° (see IRiNa]).
Natuml Problem: To extend (1.1) and (1.2) to several variables.
2. A POLYDISC TYPE POISSON KERNEL

Let T = (T1 , ••. ,Tn) E £(1t)n be a commuting multiopemtor (briefly a
c.m.). If the (Taylor) joint spectrum a(T) is in lDn , then
C(T, w) = (I - wIT1)-1 ... (I - wnTn)-l, wE lDn ,
makes sense, and we may set
(2.1) P(T,w)=C(T,w)*~TC(T,w), WElDn ,

38 F.-H. Vasilescu
where
AT = L (-l)!o!T*oTo ,
oEZ+
aSe
and e = (1, ... ,1) E Z+ (if a = (at. ... ,an) then lal = ai + .. .+an, and
TO = 1fl ... T::").
Definition. Let T E £(1i)n be a c.m. such that u(T) C Jl)n and

~T ~ o. Then peT, w), given by (2.1), is called the opemtar-valued Poisson
kernel associated with T.
I do not intend to insist too much on this case. It will be separately
treated in a forthcoming work (with R. Curto).
3. A BALL TYPE POISSON KERNEL

Let B = Bn C en be the open unit ball. Let also S = 8B. 1fT E£(1i)n
is a c.m. such that u(T) C B, then
G(T, w) = (1 - WITI - ... - wnTn)-n, wE h

makes sense, and we may define
(3.1) peT, w) = G(T, w)* ~r;)C(T, w) , wE h,

where
Definition([Vas]). Let T E £(1i)n be a c.m. such that u(T) C B

and ~S;) ~ o. Then peT, w), given by (3.1), is called the opemtor-valued
Poisson kernel associated with T.
3.1. THEOREM([Vas)). Let T E £(1l)n be a c.m. such that u(T) C

B. Then for each analytic polynomial f (more generally, for each f holo-
morphic in B and continuous on fJ) one has
(3.2) f(T) = Is J(w)P(T,w)du(w) ,
where u is the normalized Lebesgue measure on S.
Consequences. If, in addition, ~r;) ~ 0, we have:
1° If f is holomorphic in B and continuous on h, and if Ref ~ 0, then

Ref(T} ~ O.
Operator-valued Poisson kernels 39
2° IIf(T)II:::; IIflloo, which is a version of von Neumann's inequality.
Definition([Vas)). Let T E C(Ji)n be such that aCT) C Band tl¥:) ;:::

O. For each f E Ll(a) we define
(3.3) P[f](T) = Is f(w)P(T, w)dn(w) ,
which is called the Poisson tmnsform of f at T.
Lemma([VasJ). With T as above, if f E LOO(a), we have the estimate
(3.4) IIP[f](T)1I :::; 211fll00 .

Natural Problem. Under what conditions does there exist
(3.5) lim P[f](rT) ,

r-+l
r<l
where T E £(Ji)n is a c.m. with a(T) C [17

An answer to this question is given by the following.
3.2. THEOREM([VasJ). Let T E C(Ji)n be a c.m. such that
and tl¥:) ~ O. Then (3.5) exists in the strong operator topology of £(Ji)
for each f E C(S).
As a matter of fact, Theorem 3.2 is an improved version of the corre-
sponding result in [Vas], due to some subsequent assertions obtained in
[MiiVa].
For n = 1, a direct proof of Theorem 3.2 can be given. If n ;::: 1, only an
indirect proof of Theorem 3.2 is so far known. This is based on a structure
theorem for those c.m. T E £(Ji)n such that tl¥) ~ 0 and tl¥:) ;::: 0 (see
Theorem 5.1).
REMARK. If tl¥) ;::: 0, then a(T) C [J [MiiVa].
4. POSITIVITY CONDITIONS
The fact that T E C(Ji) is a contraction is expressed by the relation
I - T*T ~ O. There are versions of this condition valid for one or several
variables.
40 F.-H. Vasilescu
Definition([Ag1]). An operator T E £(ll) is said to be an m-hyper-

contraction if
(4.1) t(-1)k(~)T*kTk~0 , 1~r~m.

k=O
J. Agler proved structure theorems for m-hypercontractions, using C*-

algebra techniques (see [Ag1]).
If T E £(ll)n is a c.m., then the condition (of polydisc type)
(4.2) L (_1)laIT*aro ~ 0 , 0 ~ I-' ~ e

aEZ"
as,.
can be used to obtain the existence of a regular dilation for T 1 , ••• ,Tn (see
[SzFo]).
If T E £{ll)n is a c.m., the condition (of ball type)
(4.3)
can be used to obtain a von-Neumann type inequality (see [Dru]).

Condition (4.3) can be further extended:
(4.4)
and a~m) ~ 0{1 ~ m ~ n) has been used by A. Athavale to obtain exten-

sions of Agler's results (see [Ath1], [Ath 2]).
We mention that all these conditions can be also expressed in terms of
hereditary polynomials, introduced by J. Agler.
Let us also note that if T E £(1l)n is a c.m., we may define the operator
MT : £(1l) --+ £(1l) by
n
(4.5) MT{X) = LT; XTj ,
;=1
and we have
(4.6) a~m) = (1 - MT)m{J) , m ~ 0 an integer.

5. STANDARD MODELS
Let n ~ 1, m ~ 1 be fixed integers, and let 8m : Z+ - Z+ be given by
(m+ lal-1)! n
(5.1) 8m (a) = a Z+ .
a.I ( m _ 1) .I ' E
Let also JC = f2(Z+, 1£). We define on JC the operators S1m ) by

(S;m)f)(a) = [8 ~m(a)
m a+eJ
.) rightF/ 2f(a+ej),j=1, ... ,n ,(5.2)
where {ej = (0, ... ,1, ... , On is the canonical basis of lRn.
Definition([MiiVa]). The c.m. s(m) = (Sf m), ... , S~m» is called the
backwards multishift of type (n, m).
Definition.([Ath1]). A c.m. WE £(1£)n is said to be a spherical isom-

etry if
(5.3)
5.1. THEOREM([MiiVaJ). Let T E £(1£)n be a c.m., and let m ~ 1

be an integer. The following conditions are equivalent.
(a) 6.T(1) >

-
(m) > O.
0 and 6.T -
(b) T is unitarily equivalent to the restriction of s(m) ED W to an invari-

ant subspace, where s(m) is the backwards multishift of type (n,m) and
W is a spherical isometry.
Theorem 5.1 is an extension of a result from [Vas]. Indeed, if n = m,

the operator U : JC _ H2(S, 1£) given by
(5.4) (U f)(z) = L [8n (a)]1/2 f(a)zCi

CiEZ+
is unitary (where H2(S,1£) = H2(S) ® 1£, with H2(S) as in [RudJ). Note

that
(5.5) u S;n) = Mj U , j = 1, ... ,n ,
where (Mjh)(z) = zjh(z) , hE H2(S,1£). This shows that for n = m the

conditions below are equivalent:
42 F.-H. Vasilescu
(a') 6,(1)
T -
> 0 and 6,(n)
T -
> o·,
(b') T is unitarily equivalent to the restriction of M* EB W to an invari-

ant subspace; which is essentially proved in [Vasj (modulo some redundan-
cies).
Let us also mention that the case n = m = 1 is implicitly proved by
[SzFoj.
5.2. THEOREM([MiiVa]). Let T E £(1i)n be a c.m. and let m 2:: 1

be an integer. The following conditions are equivalent.
(a) ~) 2:: 0, 6,~m) 2:: 0, and
(5.6) s - lim MNI) = 0

k-+oo
(" s" stands for the strong operator topology.)
(b) T is unitarily equivalent to the restriction of s(m) to an invariant

subspace.
For n = m = 1, condition (5.6) means precisely
s - lim T*kTk =0
k-+oo
which is in turn equivalent to
s- lim Tk=O
k-oo
i.e., T is in the class Co. in the sense of [SzFoj.

Theorems 5.1 and 5.2 can be obtained via some technical lemmas. Here
are some of them:
1° ~¥) 2:: 0 if and only if IIMTII ~ 1.
2° If ~(1)
T
>
-
0 and 6,(m)
T
>
-
0, then
3° The backwards multishift s(m) has the following properties:
(a) ~~(~) 2:: O.
(b) A (m)
US(m) 2:: 0.
°
REMARK. If T e £(11.)n is a c.m. such that 6.¥) 2: and 6.¥:) 2: 0,
then T has a spherical dilation (see [Vas]). In other words, there is a Hilbert
space M c 11. and a c.m. N e £(M)n consisting of normal operators such
that
on M,
and Tj = P'H.NjI11.(j = 1, ... ,n), where P'H. : M -+ 11. is the orthogonal
projection.
If m i- n, we do not have any spherical dilation. Thus, for n = m there
are chances to reconstruct the Sz.-Nagy-Foias theory for contractions (see
also [MiiVaJ).
Nevertheless, if T e £(11.)n is a c.m. such that 6.¥) 2: 0, 6.~m) 2: 0, then
sCm) $ W (W given by Theorem 5.1) may be regarded as a "dilation" of
T. For instance, we have the estimate
(5.7)
for all analytic polynomials f, which is a version of von Neumann's inequal-

ity (for m = 1 and n arbitrary, (5.7) can be found in [Dru]).
6. THE POISSON TRANSFORM REVISITED

As mentioned in Section 3, Theorem 5.1 can be used to prove Theorem
3.2.
6.1 PROPOSITION([Vas]). We have
(6.1) lim P[J](rM*)h = C[ih)

r-+l
r<l
for all f e LOO(O') and h e H2(S, 11.), where C[g) is the Cauchy transform
of 9 ([Rud]) and }(z) = f(z).
6.2 LEMMA([Vas)). If W is a spherical isometry, then
(6.2) lim P[f](rW)

r-+l
r<l
exists in the uniform topology of £(11.) for each f e C(S).

Sketch of proof of Theorem 3.2. Since
T~M· $ WI invariant subspace ,

44 F.-H. Vasilescu
then P(rT,w) may be replaced by
P(rM*,w) $P(rW,w)
The assertion follows from Proposition 6.1 and Lemma 6.2.

Proposition 6.1 also implies the next result.
6.3 THEOREM((Vas]). Let T E £('H)n be a c.m. such that a¥) ~ 0,

a(n) > 0 and
T - ,
s - lim M~(I) = 0 .
k-+oo
Then there exists a contractive unital algebra homomorphism
CPT : HOO(S) -+ £('H)
such that 'PT(f) = J(T) for each analytic polynomial J.

In addition, if {Jkh E HOO(S) is a bounded sequence that convergens
cr-almost everywhere to zero, then {CPT(fk) h is strongly convergent to zero.
The proof is based on the fact that T may be replaced by M*, and we
have ((Vas]).
(6.3) lim J(rM*)h = C[ih]

r-+l
r<l
REMARK Theorem 6.3 is a several variables version of the HOO_

functional calculus for some contractions ([SzFo)).
7. INVARIANCE UNDER AUTOMORPHISMS

The concept of Poisson transform may be extended, as a consequence of
Theorem 3.2, in the following sense.
Definition([CuVa]). Let T E £('H)n be a c.m. such that a¥) ~ 0 and

a':;) ~ O. Then for each J E C(S) we set
(7.1) P[J](T) = s -lim P[J](rT)
r-+l
r<l
The lefthand side of (7.1) is called the Poisson trons/orm 0/ J at T.

As in the scalar case (see [Rud]), we have a certain invariance property
under automorphisms of B (Le., biholomorphic mappings of B into itself),
even for (7.1).
7.1 THEOREM([CuVa]). Let T E £(1i)n be a c.m. such that ~¥) ~

o and ~¥:) ~ o. Then for all f E 0(8) and t{; E Aut(B) one has
(7.2) P[J 0 t{;)(T) = P[J)(t{;(T)) .
(Of course, t{;(T) is defined via the holomorphic functional calculus.)
Related results:
1° ~(m)
T
>
-
0 iff ~(m) > O·
1/J(T) - ,
2° s - lim M~(I) = 0 iff s - lim M.~(T)(I) =

k-oo k-oo .,
o.
8. OPEN PROBLEMS
There is a long list of open problems in this context. We select some of
them.
Problem 1. Is the class of c.m. T E £(1i)n with ~¥) ~ 0, ~) ~ 0 an

appropriate extension to several variables of the class of contractions?
Problem 2. State and prove a commutant lifting theorem for the above
class.
Problem 3. Prove a Wold type decomposition for spherical isometries.
Problem 4. State and prove some classical problems in the unit ball
related to operator theory (e.g., Schur's problem, etc.).
REFERENCES
[Agl] Agler J., Hypercontractions and subnormality, J. Operator The-
ory 13(1985), 203-217.
[Ath 1] Athavale A., On the intertwining of joint isometries, J. Operator
Theory 23(1990), 339--350.
[Ath 2) Athavale A., Model theory on the unit ball in on, J. Operator
Theory (to appear).
[CuVa] Curto, R. E. and Vasilescu F.-H., Automorphism invariance of
the operator-valued Poisson transform, Acta Scient. Math. (to
appear).
46 F.-H. Vasilescu
[Dru] Drury S.W., A generalization of von Neumann's inequality to

complex ball, Proc. Amer. Math. Soc. 68(1978),300-304.
[MiiVa] Standard models for some commuting multioperators, Proc.
Amer. Math. Soc. (to appear).
[RiNa] Riesz F. and Sz.-Nagy B., L~ons d'analyse fonctionelle, Akad.
Kiad6, Budapest, (1952).
[Rud] Function theory in the unit ball of en, Springer-Verlag, New
York!
Berlin, (1980).
[SzFo] Sz.-Nagy B. and Foias C., Analyse harmonique des operateurs de
l'espace de Hilbert, Masson, Paris, and Akad. Kiad6, Budapest,
(1967).
[Vas] An operator-valued Poisson kernel, J. Functional Analysis 108
(1992).
Institute of Mathematics
Romanian Academy of Sciences
P.O. Box 1-764
Ro-70700
Bucharest, Romania
Section II
NONSELFADJOINT ALGEBRAS
Elementary operators and subalgebras
Keith J. Coates
Texas A&M University
College Station, Texas 77843
September 26, 1993
Abstract
In this note, we construct an elementary operator on B(H) of length
two which leaves invariant a nontrivial triangular subalgebra of B(H} but
which cannot be written as a finite sum of elementary operators of length
one that each leave the triangular subalgebra invariant.
Let A be a Banach algebra and let A, B E A. A multiplication (operator)

~A,B : A -+ A is defined by the equation MA,B(X) = AXB, X E A. Let
A = (A 1 ,A2 , ••• ,An), B = (BI,B 2 , ••• ,Bn) be n-tuples of elements of A. An
= =
elementary operator R R(A, B) is an operator of the form R I:?=l MA;,B;.
Clearly, R can have different representations I: MC;,D;. We define the length I
of R to be the smallest number of multiplication terms required for any repre-
sentation of R.
Elementary operators have been studied in great detail in recent years. The
spectral theory has received much attention [2, 3, 4]. Most of the results concern
the relationship between the spectrum of R(A, B) and the joint spectra of A
and B.
Certain structural properties of R also relate to structural properties of the
Ai, Bi. One example of the work in this direction is the so-called range inclusion
problem [1, 5, 6, 7]: If I is an ideal in A, and if Ran(R(A, B» c I, what can
be said about the relationship between I and the Ai, Bi? In [1], Apostol and
Fialkow showed that for A = B(H), H a Hilbert space, if {BI,B 2 , ••• ,Bn}
is linearly independent modulo K(H), the ideal of compact operators, then
Ran(R(A, B» c I if and only if Ai E I, i = 1,2, ... , n.
In this note, we examine a somewhat different problem. If C is a subalgebra
of A, define Ec(A) to be the algebra of elementary operators on A which leave C
invariant. The question we ask is the following: If R is an elementary operator in
=
Ec(A), must there exist a representation R I:?=l MA;,B; such that MA;,B; E
Ec(A), i = 1,2, ... ,n?
50 KJ. Coates
We give an example where the answer to the above question is negative.

The example is constructed in B(12}, which we regard as bounded, infinite di-
mensional matrices relative to the standard basis of 12 : el = {1, 0, 0, .. . }t, e2 =
(0,1,0, ... }t, .... The subalgebra we consider is Too, the algebra of upper trian-
gular matrices in B(/ 2 ). Specifically, we give an example of an operator R in
ETco (B(/2» oflength 2, but for which there does not exist a finite representation
R = L:?=l MA.,B. such that each MA.,B. E ETco {B(/ 2 )}. The full work will be
published at a later date, but we present here the actual example, together with
an indication of the proof.
Let 0' = (0'2,0'3, •.. ,) E 100 , with the O'~s all distinct and nonzero. Let
Ei,j be the matrix unit in B(12} with 1 in the (i,j) position and O's elsewhere.
Construct the following operators in B(/2):
00 00
2: 2: 2: Ei,i-l
00 00
Al = 2: Ei,i, Bl = O'iEi,i-lo A2 = O'iEi,i, B2 =

i=2 i=2 i=2 i=2
Note that Al and A2 are diagonal matrices and that B} and B2 are forward
unilateral shift operators.
Define R, an elementary operator on B{h), by R = MAt,B t - M A2 ,B.' It
is not difficult to see that R lies in ET",,(B(12» but that the individual multi-
plications MAt,B" M A2 ,B. do not: Let Ei,j E Too. Then we must have i :5 j.
If i < j, then R(Ei,j} = MAt,B t (Ei,j) - MA.,B.(Ei,j} = O'jEi,j-l - O'iEi,j-l =
(aj - ai}Ej,j-lo which is a nonzero element of Too. If i = j, then we have
R(Ei,i) = MAt,B t (Ei,i) - M A2 ,B2(Ei,i) = O'iEi,i-l - O'iEi,i-l = 0. The individ-
°
ual multiplications each map Ei,i to the same nonzero element outside of Too
(i > i-I), so R(Ei,i) = E Too.
Theorem 1 R E ET",,(B(l2» but R admits no representation E?=l Me.,D.
such that each Me.,D. E ET",,(B(l2».
Idea of proof. We have seen above that R E ET"" (B(/2» but that MAt,B"
M A2 ,B. ¢ ET",,(B(12}}' Suppose we have a representation R = L::::l Me.,D.
with each Me.,D. E ET",,(B{l2»' Let A~,BLA~,B~,Ci,Di be the upper left
m x m submatrices of A 1 ,B1 ,A2,B2,Ci,Dj, respectively. Then standard com-
putations show that R' =
M A,l' B'1 - M A,2' B'2 =L..J - Me'i' D'i E Er.m (Mm),
"'~l
where Mm is the algebra of m x m matrices and Tm is the subalgebra of upper
triangular m x m matrices. Similarly, it is also not difficult to see that each
Me~,D~ E ETm(Mm} as well.
By'the construction of the Ci, Di, we have that for every 2 :5 j :5 m, there
are iI, i2 with 1 :5 i}, i2 :5 n such that Cit contains the matrix unit Ei,i and
Di2 contains the matrix unit Ei,i-l' Since MC;,D: E ETm{Mm} for each i, we
must have that il '" i 2 •
. = M A,)' B'1 -MA'2' B',
L....J - Me~, 'D~
Also, since "'?-l 2
the construction of AI, B 1, A2, B2
give that for every it '" h, there is an i such that Ci contains the matrix unit
Ej.,j. and Di contains the matrix unit Eh,h-l'
Elementary operators and subalgebras 51
A combinatorial argument can now be used to show that the above two
conditions force the number of terms n in the sum to be larger than log2m.
Since this must be true for every m, we arrive at a contradiction, proving the
theorem.
This paper provides one example of an elementary operator that leaves in-
variant a subalgebra of B(H) but which does not lie in the linear span of length
one operators which leave the subalgebra invariant. The full proof of this ex-
ample will appear in an upcoming paper which will explore in greater detail the
kinds of algebras for which this situation can occur.
References
[1] C. Apostol and L. A. Fialkow, Structural properties of elementary operators,
Can. J. Math. 38 (1986), 1485-1524.
[2] R. E. Curto, Spectral theory of elementary operators, preprint.
[31 L. A. Fialkow, Spectral properties of elementary operators, Acta. Sci. Math.
46 (1983), 269-282.
[4] __, Spectral properties of elementary operators, II, Trans. Amer. Math.
Soc. 290 (1985),415-429.
[5] __, The range inclusion problem for elementary operators, Michigan
Math. J. 34 (1987),451-459.
[6] L. A. Fialkow and R. Loebl. Elementary mappings into ideals of operators,
Illinois J. Math. 28, 555-578.
[7] C.K. Fong and A. R. Sourour, On the operator identity L: AkX Bk == 0, Can.
J. Math. 31 (1979),845-857.
Questions on Bimodules of Nest Algebras
Xingde DaP
the University of North Carolina at Charlotte, Charlotte, NC 28223
Abstract
We will describe some recent results on bimodules of nest alge-
bras and we will propose some related questions.
In this notes we will describe some recent results on bimodules of nest

algebras. Also, we will propose some related unanswered questions.
A nest in a separably acting factor M is a totally ordered family of (self-
adjoint) projections containing {O, I} which is closed in strong operator
topology. Let (3 be a nest in M. The set {I - N : N E (3} is also a nest in
M which is called the dual nest of (3. The nest algebra alg(3 related to (3 is
alg(3 = {T EM: T P = PT P for each projection P in the nest (3}.
If M = 8(1i), an equivalent definition for a nest in 1i is a totally

ordered family of closed subspaces containing {O,1i} which contains the
join and meet of any of its subfamily and the corresponding nest algebra
will be the set of all bounded operators in 8(1i) which leave each element
in the nest invariant.
By a bimodule of alg(3, we mean a linear subspace S of M which is
closed under left and right multiplication by elements of alg(3. We will call
S a norm-principal bimodule of alg(3 if it is norm-closed and, if there is
an element GEM such that S is the norm closure of the linear $pan of
the set {AG B : A, B E alg (3}. The similar terminology is used for ideals of
nest algebras.
Power [17] proved that every w* -closed (hence strong-closed) ideal of a
nest algebra is w*-principal. Davidson [8] generalized the Power's results
into bimodules. If one considers the norm-topology, are all norm-closed
bimodules of nest algebras also norm-principal? In [12] Hopenwasser stud-
ied the Jacobson radical R/If of the algebra. He proved for many nests,
the Jacobson radical is principal. He raised the following question: is the
ISupported in part by funds from the Foundation of the UNCC and State of North
Carolina.
Questions on bimodules of nest algebras 53
Jacobson radical of a nest algebra, for a continuous nest, a norm-principal

ideal of the algebra? The answer to this question is negative. It turns
out, in order that the radical RJ{ to be norm-principal, the nest must be
countable. Orr [15] provides a complete characterization for this. We [4]
have obtained necessary and sufficient conditions for 8(1-£), the algebra of
operators acting on a Hilbert space 1-£, to be a norm-closed singly generated
bimodule of a nest algebra. As a consequence of the Ringrose's criterion
on the radical, the Jacobson radical of a nest algebra is the norm-closed
linear span of spaces of the form NB(1-£)N1. where N E N. Based on this
fact and the results on B(1-£), one can prove the condition on radical [4].
Recently we obtained a necessary and sufficient condition for a separa-
bly acting factor to be a norm-principal bimodule of a nest subalgebra [7].
Let 1 be a nest in a factor M. We will use the following notations:
E>.b) = V{N E 1: N is finite}

Epb) = V{N1.: N E 1 and N1. is finite}
Let 13 be a nest in a type I I or type I I I factor M. We define a nest 13

to be of order type I if it is an infinite set, and if one of the nests 13 or 131.,
we write it 1 satisfies the following conditions A or B:
1. Condition A :
• Both of projections E>.b) and (Epb))1. are left limit points of

1;
• The projection E>.b) is infinite and the projection Epb) is fi-
nite.
2. Condition B :
• 0 is a limit point of 1. Each non-zero Q E 1 is infinite. There is

projection P E 1 such that p1. is non-zero and finite.
We define a nest 13 to be of order type I I if it is not of order type I. Since

there is not non-zero finite projection in type I I I factor and, there is no
infinite projection in type I II factor, so all nests in type I II or type I I I
factors are of order type I I.
In [7] we have proven that the property that a nest 13 is of order type
I I is a necessary and sufficient condition for a (type I I or type I I I) factor
M to be a norm-principal bimodule of algf3. As a corollary, we proved that
any separably acting type I II or type I I I factor is always a norm-closed
singly generated bimodule of an arbitrary nest sub algebra in the factor.
54 X.Dai
Let (3 be an arbitrary nest in a separably acting type I II or type I I I

factor M. Using the results in [7], we proved [6] that, in order to exist an
operator Gin 'R.{3, the Jacobson radical of alg(3, such that the linear span
of ( alg (3) . G· (alg(3) is dense in 'R.{3 under norm topology of M, a necessary
and sufficient condition is that the nest is countable. (That is the nest (3
has at most countable infinite elements.) This is quite different from the
results in 8(1l) case [15].
The type 1100 factor case is much complicated. The count ability con-
dition is necessary but not sufficient. Let (3 be a nest in a separably acting
type IIoo factor M. An element P E (3\{0, I} is said to be a singular point
of (3 if it satisfies the following conditions 1 or 2.
1. There is a strictly increasing sequence {Qn} ~ (3, liffin_oo Qn = P

and P - Qn is infinite for each n E N. Also, there is a projection
Q E (3 such that Q > P and Q - P is finite.
2. There is a strictly decreasing sequence {Qn} ~ (3,liffin_oo Qn = P
and Qn - P is infinite for each n E N. Also, there is a projection
Q E (3 such that Q < P and P - Q is finite.
We proved [6] that a necessary and sufficient condition for the Jacobson
radical 'R.{3 of alg(3 to be a norm-closed singly generated ideal of alg(3 is
that the nest (3 is countable, and however, it does not contain a singular
point. In all these cases (type Ill, IIoo and type I II) single generation is
equivalent to countable generation.
Some questions remain open. For example,
Question 1 Characterize the conditions for the Jacobson radical to be a

norm-principle ideal of the nest algebra in a von Neumann algebra.
Question 2 Let M be an arbitrary von Neumann algebra and let (3 be

a nest in M. Characterize the conditions for M to be a norm-principle
bimodule of a nest subalgebra alg(3.
Using direct integral theory it may now be possible to do this for any von
Neumann algebra. However, the approach might be tedious. Is there a
different approach to the questions? We don't know the answer to this.
Erdos and Power [10] have successfully classify all bimodules of nest
algebras which are closed in the strong operator topology. They proved that
each strong closed bimodule is singly generated under the strong topology.
In a conversation, Arveson suggested to the author the following question.
Question 3 Classify the norm-closed bimodules of nest algebras.

This question is open even in B(1i) case. Recently Orr classified the
maximal ideals and the automorphism invariant ideals of continuous nest
algebras.
Question 4 Extend above results on bimodules and Jacobson radicals into

CSL-algebras.
The results [15] (and [4], an alternative proof of the results in [15]) on
the Jacobson radicals of nest algebras is based on the Ringrose's structural
theorem for the radical. In general, the corresponding results for CSL alge-
bras are still unknown. Hopenwasser proposed a structure for the radical
of CSL algebras analogous to Ringrose's theorem. Hopenwasser-Larson
further studied this conjecture. Recently Davidson-Orr verified this con-
jecture for width-two CSL algebras. Let C be a CSL in B(1i). If C contains
an order type I nest, then, as a consequence of a result in [4], B(1i) is not
a norm-principal bimodule of algC. We don't know whether this condition
is also sufficient.
Let C be a CSL in 1i. Assume that C has a subnest N which is of order

type I. Notice that algC ~ algN. By our result B(1i) is not a n.p.b. of
algN. So it is not a norm-principal bimodule of algC. Is the converse true?
In particular,
Question 5 If the CSL is a tensor product of two nests. Characterize the

condition that B(1i) is a n.p.b. of algC.
Using Larson-Pitts' factorization results ([13] [16]) and David Larson's

technique (cf. [13] Theorem 4.9) one can prove a generalization of the
'weak' factorization of a nest algebra. Namely, for an arbitrary nest {3 in
a separably acting factor M, for each positive invertible operator P E M,
there exist operators A and B in alg{3, which are invertible in M but not
necessarily invertible in alg{3, such that P factors in the form of AA* and
B* B. This 'weak' factorization property of nest algebra is one of the tools
in [12], [4] and [7].
Question 6 Is the weak factorization property valid for all nest algebras
in an arbitrary von Neumann Algebra?
We conjecture that the answer is 'yes'. A similar question is concerned

with the 'weak' factorization on CSL algebras.
In an important paper [13] David Larson introduced his strong radical
Rj}. Some unknown aspects on this ideal(for example, the construction of
56 X.Dai
a concrete idempotent in the strong radical when N is uncountable nest)

is crucial for one to construct the similarity transform which can change
multiplicity of continuous nest. Let N be a nest in a Hilbert space 1i. The
set R'N is the collection of operators A in algN for which, given £ > 0,
there exists a countable £-partition. The set R'N is a norm-closed ideal of
algN which contains the Jacobson radical R.N. The inclusion is proper if
and only if the nest N is infinite.
Let N be a nest in 8(1i) and let x be a separating unit vector for
the abelian von Neumann algebra generated by N. Then the mapping
p : N ~ [0,1] by N 1--+ IINxll is an injection. We write N by N>. where
>. = IINxll and N = {N>.: A E A} where A = p(N) which contains and °
1. Erdos introduced the following diagonal function ~ : algN X [0, 1] ~ IR
as follows. For A E algN,
~(A,t) = inf{IIN[A1,A2]AN[At,A2]1I: >'1,A2 E A,A1 < t < A2}, t E (0,1);
~(A,O) = inf{IIN>.AN>.1I : A E A, 0< A};
~(A, 1) = inf{II(I1{ - N>.)A(hf - N>.)II : A E A.,A < I}.
The functions ~ are due to Erdos.

In [3] we characterized conditions for an operator A E algN to be in
°
R'N. We proved that a necessary and sufficient condition for A E R'N is
that ~(A, t) = almost everywhere (Lebesgue measure) ( in [0, 1]). The
same result was independently obtained by Orr.
Question 7 Characterize conditions for Larson's ideal R'N to be a norm-

principal ideal of algN.
An example. Let 1i = £2([0,1]) and let C be the Cantor set in [0,1].

Denote (3 = {X[O,c)£2([0,1]) : c E C}. The set (3 is an uncountable pure
atomic nest in 1i. Let G = [0, 1]\C. Then G is a disjoint union of countable
many open intervals {En = (an,bn) : n EN}. Let {en,k : k E Z} be an
orthogonal basis for the infinite dimensional atom XEn£2([0, 1]) of (3. Refine
the nest (3 into 'Y as follows. Insert the subspace X[O'''n)£2([0, 1]) + [en,l :
I ~ k] into the nest for all n E N, k E Z. The nest 'Y is uncountable and
pure atomic of multiplicity free. Define an operator R mapping e.. ,k into
e..,k-1, n E N, k E Z. The operator R is in RC;. We have RR* = I. It is
easy to verify that R* . RC; ~ a1n. So, we have
Therefore, R is a generator for RC;. []

We proved [3] that a necessary condition is that the nest must be pure
atomic. The above example provides an uncountable pure atomic nest N
for which R'N is a norm-principal ideal of algN. This suggests to us that
the pure atomic condition might be sufficient also.
References
[1] Andersen, N. T., Compact perturbations of reflexive algebras, J.
Funct. Anal. 38(1980), 366-400.
[2] Andersen, N. T., Similarity of continuous nests, Bull. London Math.
Soc. 15(1983), 131-132.
[3] Dai, X., Norm-closed bimodules of nest algebras, Ph.D thesis, Texas
A&M University, 1990
[4] Dai, X., Norm-principal bimodules of nest algebras, Journal of Func-

tional Analysis, Vol. 90, No2, (May 1990), 369-390.
[5] Dai, X., On the generators of bimodules of nest algebras, Jour. Op-
erator Theory, to appear.
[6] Dai, X., Jacobson radicals of nest subalgebras in factors, Proc. AMS,
to appear.
[7] Dai, X., Norm-principal bimodules of nest algebras II, preprint.
[8] Davidson, K. R., Similarity and compact perturbations of nest alge-

bras, J. Reine Agnew. Math. 348(1984), 72-87.
[9] Davidson, K. R., Nest Algebras, Pitman Research Notes in Mathe-

matics Series 191(1988).
[10] Erdos, J. A. and S. C. Power, Weakly closed ideals of nest algebras,
J. Operator Theory 7(1982), 219-235.
[11] Gilfeather, F. and Larson, D., Nest-subalgebras of von Neumann al-
gebras, Advances in Mathematics, Vol. 46, No.2, November 1982,
171-199.
[12] Hopenwasser, A., Hypercausal ideals in nest algebras, J. London

Math. Soc., 34 (2) (1986) 129-138.
58 X.Dai
[13] Larson, D. R., Nest algebras and similarity transformations, Ann.

Math. 121 (1985), 409-427.
[14] Larson, D. R., A solution to a problem of J. R. Ringrose, Bull. (New
Series) A.M.S. 7 (1982), 243-246.
[15] Orr, J. L., On generators of the radical of a nest algebra, Proc. Lon-
don Math. Soc.
[16] Pitts, D. R., Factorization problems for nests: Factorization methods

and characterizations of the universal factorization property, Journal
of Functional Analysis, 79 (1988) 57-90.
[17] Power, S. C., Analysis in nest algebras, Pitman Research Notes in

Mathematics Series 192(1988), 189-234.
[18] Ringrose, J. R., On some algebras of operators, Proc. London Math.

Soc. (3) 15 (1965), 61-93.
[.19] Ringrose, J. R., On some algebras of operators II, Proc. London
Math. Soc. (3) 16 (1966), 385-402.
The Universal Factorization Property for
Commutative Subspace Lattices
JOHN DAUGHTRY
Abstract.
We present a necessary and sufficient condition for a commutative
subspace lattice (CSL) to have the property that every sublattice has the
universal factorization property (UFP). Then we derive a sufficient condi-
tion for a CSL to have the UFP which applies to all known examples of
CSL's with the UFP. (The preceding results are, in fact, obtained for CSL-
subalgebras offactor von Neumann algebras.) Finally, we give an example
of a CSL which has the UFP, even though it is not a direct sum of a com-
plemented lattice and a CSL with the property that every linearly ordered
sublattice is countable.
H denotes a separable Hilbert space with complex scalars. B(H) is

the space of all bounded, linear operators on H. The word "projection"
always refers to a selfadjoint idempotent operator. For any A in B(H),
rp(A) denotes the projection of H on the closure of the range of A. R(A)
is the range of A, and I is the identity operator. V always denotes a von
Neumann subalgebra of B(H).
A lattice C is complete if the least upper bound and greatest lower
bound of each subset of C is in C. A complete lattice of projections in
B(H) (with lattice operations corresponding to intersection and closed lin-
ear span of subspaces) which contains 0 and I is a subspace lattice. A
subspace lattice C which satisfies PQ = QP for all P and Q in C is a
commutative subspace lattice (CSL). C is complemented if PEe implies
60 J. Daughtry
I - P E C. A subspace lattice which is linearly ordered is a nest. An

interval in a CSL C is a projection of the form P - Q where P and Q are
elements of C.
Let 'P be any set of projections in B(H), and let T be an invertible
element of B(H). r'P denotes {rp(TP) : P E 'P}. Algy 'P denotes {A E
V : A leaves invariant the ranges of the elements of 'P}. In this and other
=
notations, the subscript is usually omitted when V B(H). For C a CSL,
algC is called a CSL algebra. The study of such algebras was pioneered by
W. Arveson in [2].
A CSL C has the Universal Factorization Property relative to
V(U F Pv) if for every invertible T in V such that rC is commutative, T =
U A with U unitary and A and A-1 in algy C. The characterization of all
CSL's with the U F Pv is a well-known open problem, even for V B(H).=
A CSL C has the Hereditary Universal Factorization Property relative
to V (HU F Pv) if every sub-CSL of C has the U F Pv.
It is obvious that factoring T in the form U A as in the definition of
U FPy is equivalent to factoring T*T in the form A* A with A and A- 1 in
algyC. Related factorization problems are often stated in these terms (as
factorization problems for positive definite operators), instead.
Another equivalent formulation of the condition that T = U A with
U unitary and A and A -1 in algy C is that algy( rC) is unitarily equivalent
to algy C.
The problem of characterizing the nests which have the U F PB(H) was
solved by David Larson in [6], following the pioneering work of Arveson [3].
(For nests, the condition that rC is commutative is superfluous.) Specif-
ically, a nest has the U F P if and only if it is countable. It is clear from
this solution that every subnest of a nest with the U F P also has the U F P.
The U F Py for the case that V is a factor and C is a nest was solved by
David Pitts in [7; p. 85]: the necessary and sufficient condition is that a
certain subnest (which depends upon V) of C, called "the reduced nest" of
C, is countable.
The situation for CSLs is more complicated. We will demonstrate
below that C has the H U F P relative to B( H) if and only if every sublattice
of C which is linearly ordered is countable. More generally, we show that
C has the H U F Pv for a given factor V if and only if every sublattice of C
which is a nest has the property that its "reduced nest" is countable. Even
though this condition is general enough to apply to some uncountable CSLs,
it certainly does not include all complemented CSLs, and it is well-known
that all complemented CSLs have the U F P. More surprisingly, there exist
CSLs with the U F P which are not direct sums of complemented lattices
and CSLs such that every linearly ordered sublattice is countable. We will
conclude the paper with an example of such a CSL.
Theorem 2 gives a sufficient condition for a CSL to have the U F Py .
The condition applies to every example known to the author.
The universal factorization property 61
Now, for the convenience of the reader, we introduce some of the

required results from [4] specialized to the setting of this paper. Fix a von
Neumann algebra V. Recall that if W- is a linear operator on V satisfying
i. for all X E V such that X ~ 0, w-(X) ~ 0,
ii. w-(X*) = w-(X)* for all X E V,
lll. w- is idempotent, and
IV. w-(X)w-(Y) = w-(X(w-(Y)) for all X and Y in V, then W- is a
conditional expectation from V onto R(w-).
Remarks:
1. ii is a consequence of i because the scalars are complex.
2. [5; Chapter 8] and [8; Chapter II] are good references for the proper-
ties of conditional expectation operators. In particular, they demon-
strate that if V is a von Neumann subalgebra of B(H) which is either
commutative or has a commutative commutant, then there exists a
conditional expectation operator of B(H) onto V.
3. [4] does not actually require properties i and ii for establishing Lemma
1 (below).
Let V be a von Neumann algebra with W- a conditional expectation

with domain V. W- is faithful if the only element X in the kernel of W- sat-
isfying X ~ 0 is X = O. W- is normal if it is continuous with respect to the
strong operator topology. It is well-known that a faithful, normal expec-
tation of V onto the commutant of a CSL C is an algebra homomorphism
when restricted to alg C [1; Thm. 3.1.1].
Our main tool is the following
Lemma 1 (Theorem 1 of [4]): Let T E V be invertible and let C C V be
a commutative set of projections. Assume that rC is commutative, and M
is a subset of C such that M' =C', (rM), =
(rC)', and U is a unitary
operator such that U PU* = rp(T P) for all P EM. Let w- be any condi-
tional expectation operator from V onto C'. If w-(U*T) and w-(T-1U) are
one-to-one, then R(U P) = R(TP) for all P E C.
Following [7; p. 82], we fix V, a von Neumann subalgebra of B(H),

and we fix a complete linearly ordered set N in V. For each PEN, let
A.t(P) = sup{Q EN: Q ~ P and Q - P is a finite projection in V},
AX(P) = inf{Q EN: Q ~ P and P - Q is a finite projection in V}
and
N r = {A.t(P) : PEN} u Px(P) : PEN}.
If N contains 0 and the identity operator then N r is also a nest, called
the reduced nest of N.
In addition to Lemma 1, we need a fundamental result of David Pitts:
Lemma 2 (Theorem 6.4 of [7]): Let V be a factor and let N be a nest in

V. Then N has the U FPv if and only if N r is countable.
62 J. Daughtry
Theorem 1: Let V be a factor, and let C be a CSL in V. Then C has the

HU F Pv if and only if every sub-CSL of C which is a nest has the property
that its reduced nest is countable.
Proof: If N is a nest which is a sublattice of C and Nr is not countable,

then N does not have the U F Pv by Lemma 2. Thus C does not have the
HUFPy .
Now suppose that every sub lattice of C which is a nest has a countable
reduced nest. By the first Lemma of [4] (cr. [2, p. 482]), there exists a nest
N in C such that N' = C' and ( TN)' = (T C)' for all T invertible in V such
that TC is commutative. (The somewhat surprising order of the quantifiers
is correct: there is a nest which satisfies the condition simultaneously for
all such T!) Let T represent an arbitrary invertible element of V such that
TC is commutative. By Lemma 2, T = U A with A and A-1 in al!JV N. We
wish to apply Lemma 1 with M = N. By Lemma 6.2 of [7], there exists
a faithful, normal conditional expectation W of V onto N'. Recall that
N' = C' and W is multiplicative on algy N. Thus w(U*T) and w(T- 1U)
are inverse to each other, hence one-to-one. Thus by Lemma 1, A and A- 1
actually belong to algC. This provides the desired factorization for T. 0
Recall that an invertible operator T has a polar decomposition of the

form UITI where U is unitary and ITI = (T*T) t. The following observation,
which was pointed out to the author by Baruch SoleI, seems to be widely
known. We omit the proof, which involves standard von Neumann algebra
arguments. Observe that the proposition provides a measure of how far
Theorem 1 misses characterizing the CSLs with the UFP.
Proposition 1: If C is a complemented commutative subspace lattice on

Hand T is any invertible element of B(H) such that TC is commutative,
then T = UITI with U unitary and ITI in alg C n (alg C)-1. (Thus the
factors belong to any von Neumann algebra which contains T.)
We are now prepared to prove the main result:
Theorem 2: Let V be a factor. Let C C V be a CSL with E E C such

that CE is complemented (as a subspace lattice on the range of E). If every
linearly ordered complete sublattice C of {P E C : P ~ E} is such that the
reduced set c,. is countable, then C has the U F Py.
Proof: By the first Lemma of [4] (cf. [2, p. 482]), there exists a nest
N in C such that N' = C' and for all T invertible in V such that TC is
commutative (TN)' = (TC)'. Having chosen such a nest N, let C denote
{PV E : PEN}. Let T E V be invertible and such that TC is commutative.
Let N be the nest obtained by adding 0 to C. (In the case E = 0, the desired
result is already known.) N r is countable by hypothesis, so T = VB with
V a unitary element of V and B and B-1 in algy N, by Lemma 2. In
particular, V*T maps the range of E onto itself. Let Uo denote the unitary
factor in the polar decomposition of V*T E as an operator on the range of
E: V*TE = UoIV*TEI. CE is a complemented lattice on the range of E.
The commutativity of {rp(V*TPE) : P E C} = {V*rp(TPE)V : P E C}
and Proposition 1 imply that Uo maps the range of PE onto the range of
V*T P E and W*T EI maps the range of PE onto itself for all P in C.
Define U = VUoE + V(I - E). Then it is easily verified that U is
unitary. For all P E C, R[U P E] = R[T P E] (by the preceding paragraph),
so U*T and T- 1U leave invariant the range of each element of CE. Recall
that V*T = B with B and B- 1 in algy N. Thus R(TP) = R(VP) for
PEN. Consider PEN, and note that P ~ E. Thus rp[U P] = rp(U PE)+
rp[U P(I - E)] (because U is unitary) = rp[VUoEP] + rp[V(I - E)P] =
rp(TE)+rp(VP)-rp(VE) = rp(VP) = rp(TP). That is, P = rp(U*TP).
Thus, U*T and its inverse, T- 1U, map the range of Ponto itself for all P
in N uCE.
The commutant of Cis [CUCEl' and (TC)' (TN)' = =
[T[CUCE)]'.
Let 1/J denote a conditional expectation of B(H) onto (CEY, and define
WE(X) = E1/J(EX E)E. By Lemma 6.2 of [7; p. 83], there exists a faithful,
normal conditional expectation X of V onto N' and X is multiplicative on
VnalgN. Let E.L = I-E, and WE.l.(X) = E.LX(E.LXE.L)E.L. Define
W : V --+ C' by W = WE + WE.l.. Then W is a conditional expectation of V
onto C' (because C' = [C U CE]').
Consider wE(U*T) = E1/J(EUoV*TE)E = E 1/J(EW*TEIE)E. The
last expression equals EIV*TEIE because EW*TEIE commutes with CE.
From the definition of UO, W*TEI =
UoV*TE, so WE(U*T) =
EU*TE.
=
Thus W(U*T) EU*TE + WE.l.(U*T). Similarly, WE(T- 1U) E =
1/J(ET- 1VUoE)E = E1/J(EW*TEI-1UQ'UoE)E =
E1/J(EW*TEI- 1E) =
EW*TEI- 1E by the fact that EW*TEIE commutes with CE. That is,
WE(T-1U) = T-1VUoE = ET-1U E. Thus W(T-1U) = ET-1U E +
WE.l.(T- 1U).
Then WE.l.(U*T) WE.l.(T- 1U) = E.LX(E.LU*TE.L)E.L X
(E.LT- UE.L)E.L = E.L X(E.LU*TE.LT- 1UE.L)E.L (by the multiplicativ-
1
ity of X on algN) = El. by the fact that U*T maps the range of E into
itself. Likewise, WE.l.(T- 1U)WE.l.(U*T) = E.L. It follows from the pre-
ceding calculations that W(U*T) and W(T- 1U) are inverse to each other,
hence each has trivial kernel.
Now by Lemma 1 with M = C U CE, it follows that the range of UP
is the same as the range of T P for all P E C. In other words, U*T and
T- 1U belong to algy C. Then letting A =
U*T, we have T =U A with A
and A-1 in algv C. c
The hypotheses of the preceding theorem apply to some lattices which

are not direct sums of complemented lattices with CSLs having the property
that every linearly ordered sublattice is countable. That is, there does not
exist E E C with I - E E C such that CE is complemented (as a CSL on
64 1. Daughtry
R(E» and every linearly ordered sublattice of £(1 - E) is countable. We

now give an example of such a lattice.
Example: Let I' be counting measure on Q, the rational numbers. Z
denotes the integers, and 9\ denotes the real numbers. For r E 9\, [[r]] is
the largest integer n such that n ::5 r. Let H = L2(1') = 12(Q). Let E be
the collection ofsubsets S of Q such that q E S implies ITq]] E S. 'P denotes
the lattice of projections onto subspaces of the form {f : f(q) = 0 for all
q E S} where SEE.
Proposition 2: If E E 'P, E =f:. 0 and I - E E 'P, then 'P E contains an

uncountable nest.
Proof: For such an element E there must exist an integer n such that the
set of all functions supported on Q n [n, n + 1) is contained in R(E). For
each real number rEin, n + 1), the functions supported on Q n [r, n + 1)
belong to 'P E. Such an uncountable linearly ordered subset is contained in
an uncountable nest. c
It is clear from the preceding proposition that 'P is not a direct sum
of a complemented lattice and a CSL with the property that every linearly
ordered sublattice is countable. Consequently, the following proposition
shows the usefulness of Theorem 2.
Proposition 3: 'P has the Universal Factorization Property.
Proof: Let E denote the projection on the functions f such that f(n) = 0
for all integers n. Then £E is complemented. Every subset C of {P E £ :
P ~ E} is determined by the set of integers at which elements of C must
vanish. Thus if C is linearly ordered, it must be countable. The conclusion
that 'P has the UFP now follows from Theorem 2. c
Theorem 2 and the preceding example demonstrate that count ability
and complementation can be combined in somewhat subtle ways to yield
the universal factorization property for a CSL. Consequently, the most ap-
propriate open problem to state in conclusion is whether there is a third
property of a CSL which implies the UFP (a property which does not ap-
pear to be some combination of countability and complementation).
References
1. Arveson, W. B., Analyticity in operator algebras, Amer. J. Math. 89
(1967), 635-647.
2. Arveson, W. B., Operator algebras and invariant subspaces, Ann. of
Math. (e) 100 (1974), 433-532.
3. Arveson, W. B., Interpolation problems in nest algebras, J. Func.
Anal. 20 (1975), 208-233.
4. Daughtry, J., and Johns, R., Arveson nests and operator factorization
along commutative subspace lattices, Proc. A.M.S. v.107, 4 (1989),
943- 947.
5. Davidson, K. R., Nest Algebras, Pitman Research Notes in Mathe-
matics v. 191, Essex, U.K. (1988).
6. Larson, D. R., Nest algebras and similarity transformations, Ann. of
Math. 121 (1985), 409-427.
7. Pitts, David R., Factorization problems for nests: factorization meth-
ods and characterizations of the universal factorization property, J.
Funct. Anal. 79 (1988), 57-90.
8. Stratila, S., Modular Theory in Operator Algebras, Abacus Press,
Kent, U.K., 1981.
9. Takesaki, M., Theory of Operator Algebras I, Springer-Verlag, NY,
1979.
Subject Clauification: Primary 47C05, 47A15, 47A68
John Daughtry
East Carolina University
Greenville, NC 27858
madaught@ecuvm1.bitnet
COMPRESSION LIMIT ALGEBRAS
ALAN HOPENWASSER
CECELIA LAURIE
I. INTRODUCTION
While there is now a large and rapidly growing literature on the study
of direct limits of subalgebras of finite dimensional C·-algebras, the focus
in almost every paper on the subject has been on systems with embeddings
which have *-extensions to the generated C* -algebra. One notable excep-
tion is the paper by Power [PI]. The purpose of this note is to extend
the study of systems which are not *-extendible and to produce examples
which illustrate some new phenomena.
When all the embeddings in a direct system are *-extendible, then the
limit algebra is, in a natural way, a subalgebra of an AF C* -algebra. On
the other hand, if the embeddings are not *-extendible, then it is no longer
a priori obvious that the limit algebra is an operator algebra. (Initially, the
direct limit must be taken within the category of Banach algebras.) Even
if the limit algebra is an operator algebra, its image under some represen-
tations may generate a C*-algebra which is not approximately finite (as
happens in the situation studied by Power). For systemS with the type of
embeddings which we consider in this paper, the limit algebra will always
be an operator algebra, a fact which can easily be seen with the aid of
the abstract characterization of operator algebras by Blecher, Ruan, and
Sinclair [BRS]. We also produce a representation of the limit algebra which
is "natural" in the sense that the C* -envelope of the image algebra (as
defined by Hamana [HJ) is isomorphic to the C*-algebra generated by the
image algebra.
II. COMPRESSION EMBEDDINGS

The key observation behind the choice of embeddings under investiga-
tion is that if A is a CSL-algebra and if p is an interval from the lattice of
invariant projections for A, then the mapping x 1-+ pxp is an algebra ho-
momorphism. Since we are interested in direct limits of finite dimensional
algebras, we shall assume that every CSL-algebra is finite dimensional.
This implies that, with respect to a suitable choice of matrix units, A is
The authors would like to thank Vern Paulsen for helpful suggestions on the content
of this paper.
Compression limit algebras 67
a subalgebra of some full matrix algebra Mn which contains the algebra

Dn of diagonal matrices. Such algebras go under a variety of names: poset
algebras, incidence algebras, or digraph algebras; henceforth, we use the
term digraph algebra.
The invariant projections of a digraph algebra are all projections in the
diagonal Dn; in particular, they all commute with one another. If e and
f are two invariant projections such that e ~ f, then p = f - e is called
an interval from the lattice. (The term semi-invariant projection is also
sometimes used.)
Since we want our embeddings to be unital, a compression will be the
mapping which takes x to the restriction of pxp to the range of p, where p is
an interval from the lattice. The image algebra under the compression will
be viewed as a subalgebra of a full matrix algebra, usually of rank smaller
than the rank of the original containing matrix algebra.
Definition. Let A ~ Mn be a digraph algebra. A compression embedding
is a direct sum of compression mappings on A subject to the proviso that
at least one summand is the identity mapping.
The purpose of the assumption that at least one summand is the iden-
tity mapping is to ensure that the embedding is completely isometric.
(Compressions are completely contractive, but not completely isometric
except in the trivial case of the identity mapping.)
Definition. A compression limit algebra is the limit of a direct system
<PI A
A 1--+ if>2 A if>3 A
2--+ 3--+···-+
in which each embedding is a unital compression embedding.

A compression embedding of Ak into Ak+1 is a unital map of Ak onto
its range; we are assuming then that the unit of Ak+1 is equal to the unit of
the subalgebra ¢k(A k ). (This avoids unwanted degeneracies.) The special
case in which each summand is compression to the identity is just the
familiar case of a standard embedding. In this special case, compression
embeddings are • -extendible; in general they will not be.
In this note, we shall focus on a special class of compression algebras:
direct limits of full upper triangular matrix algebras,
The non-' -extendible embeddings studied by Power [PI) require non-trivial

cohomology for the digraph algebras in the system; hence nest algebras
can never appear as finite dimensional approximants. The systems studied
here have one other new feature: the only constraint on nk and nk+ 1 is
that nk < nk+l; no divisibility is required.
68 A. Hopenwasser and C. Laurie
For each k, one of the summands of tPle is the identity compression; let
qle denote the support interval for this summand from the nest of invariant
projections for T n "'+l' Also, let ""Ie denote the compression of Tn "'+l onto
qle. We may, in a natural way, view ""Ie as an algebra homomorphism from
Tn"'+l onto Tn",. It is clear that ""Ie 0 tPle = id, for all k. Thus compression
embedding systems are structured in the sense introduced by Larson [L]; in
particular, compression limit algebras are all structured Banach algebras.
Compression embeddings satisfy one additional property: they map
matrix units to sums of matrix units. This implies that they are regular
embeddings, in the sense used by Power [P2, section 4.9].
Remark. Compression embeddings can be placed into a broader context.
If C = [c;j 1is an n x n matrix, the mapping tPc: Mn - Mn given by
tPc[ ~j] = [c;j~j 1is known as a Schur mapping. It is easy to check that
tPc is an algebra homomorphism if, and only if, C satisfies the cocycle
condition: c;le = c;jCjle. Furthermore, tPc is unital if, and only if, all
c;i = 1. By taking a direct sum of algebra homomorphisms each of which is
a compression composed with a Schur mapping, we can obtain embeddings
more general than compression embeddings. (Compression embeddings
arise by insisting that appropriate Cij = 1.) We defer the study of these
more general systems to another time.
III. REPRESENTATIONS
Let
be a direct system with compression embeddings. Since each tPle is a unital

complete isometry, we may identify each Tn", with a unital subalgebra of
A; the subalgebras so obtained are nested and the closure of their union is
A. Each subalgebra carries a matricial norm and the sequence of norms is
compatible with the nesting; consequently, A has a matricial norm. It is
easy to see that this matricial norm satisfies the axioms of Blecher, Ruan
and Sinclair [BRS], so A is an (abstract) operator algebra, i.e., there is a
Hilbert space 11. and a completely isometric unital algebra homomorphism p
mapping A into 8(11.). This argument remains valid for systems of digraph
algebras and for systems with the more general Schur embeddings, provided
that the embeddings are complete isometries.
Our primary interest, however, is to obtain an explicit representation
which is, in an appropriate sense, canonical. To that end, let 11. denote a
Hilbert space with basis {en}. This index set will depend on the specific
direct system, but will always be either Z or {n: n ~ b} for some integer
b ~ 0 or {n: n ~ a} for some integer a ~ O. For each k, let Pie denote
the projection on the closed linear span of { en: n ~ k } and let N denote
the nest consisting of the projections Pie together with 0 and I. Alg (N)
denotes the nest algebra associated with N. The representation which we
will construct will map A to a weakly dense subalgebra of Alg (N).
For each k, we may write tPk = 1/1! $ ... $ 1/1:, where each 1/1j is a
compression map. At least one of these compression maps must be the
identity; in order to keep track of a selection of identity compressions,
arrange the indexing so that 1/1S = id for all k. Thus, we will always have
that the first index a is non-positive and the last index b is non-negative.
Now suppose that tP: Tn" -+ Tn"+l is a compression embedding and
that 1/1 is a single compression to an interval from the invariant projection
lattice of Tn,,+l. It is easy to check that 1/1 0 tP is a sum of compressions to
intervals from the lattice of Tn". Note that the intervals associated with
1/1 0 tP may include some which were not associated with tP itself.
A consequence of the observation above is that a composition of two
compression embeddings is again a compression embedding. Since we need
to consider multiple compositions, let tPjk denote the composition tPk 0 .. ·0
Eal 77i , where each "Ii is a simple

b
tPj. Then tPjk can be written in the form
II
compression. Furthermore, since 1/1'[J = id for each n, we may arrange the
indexing so that '70 = id. Thus, a ~ 0 and b ;::: O. The indices a and b
depend, of course, on j and k, but that dependence is suppressed in the
notation.
Observe also that the string of 77'S in the summation for tPjk appears
as a substring in the expression for tPj,k+1. The reason, of course, is that
1/1~+1 = id. Furthermore, this substring includes the '70 = id term. Thus,
as we increase k, we change the summation for tPjk by adding terms at one
or both ends of the sum. By increasing k indefinitely, we obtain a mapping
Pj = E al 77i • The index set for this sum will either be Z or a set of the
form { i: i ;::: a} or of the form {i: i ~ b} for some a :5 0 or b;::: o.
An index set consisting of integers greater than or equal to some a
will occur precisely when all but finitely many of the embeddings tP/c have
tP~ = id as the first term; the other singly infinite index set arises when the
distinguished identity summand occurs as the last term in all but finitely
many embeddings. In either case, we can change the presentation by delet-
ing finitely many embeddings to arrange that the index set for the 77'S is
either the non-negative integers or the non-positive integers, as appropri-
ate. In either of these cases, we may now view Pj as a representation on the
Hilbert space 1l with respect to the appropriate basis. The doubly infinite
case still presents an ambiguity: we need to "anchor" the representation Pj
with respect to the basis {en: n E Z}.
This "anchoring" may be done as follows. Let the index set for the
matrices in Tnl be {O, 1, ... , b }, so that f~ is the matrix unit which has
a 1 in the upper left corner and zeros elsewhere. Identify 1/1J with the
mapping 0 $ ... $ 0 $ 1/16 $ 0 $ ... $ 0, a mapping of Tnl into T n2 . (In
other words, replace all terms in tPl except for the distinguished copy of the
identity by a 0 mapping of appropriate dimension.) The image of f~ under
1/16 is now a matrix unit in Tn2 ; arrange the index set for Tn2 so that this
matrix unit is f&'. By iterating this procedure, we can index all the Tn" SO
that the successive images of the matrix unit fJo under the distingushed
identity compressions will be f&. In the doubly infinite case, the indexing
of the Tn" 's will begin with negative integers and end with positive integers
each of arbitrarily large magnitude for sufficiently large k.
It is now clear how to define the representation Pk: the image of the
distinguished matrix unit fifo in the component TJo should be the one di-
mensional projection onto the span of the basis vector eo. For each k, Pk
is a completely isometric representation of Tn" into Alg (N) acting on the
Hilbert space 1-£. Furthermore, the following diagram commutes:
Alg (N) ~ Alg (N)

This system of representations induces a completely isometric representa-
tion P from the direct limit A into Alg (N).
Remark. Observe that the image p(A) is weakly (or a-weakly) dense in
Alg (N). To see this it is enough to note that, given a ~ 0 and b ~ 0, the
image contains matrices with arbitrarily specified entries in the locations
ij for a ~ i ~ j ~ b. To accomplish this, choose k sufficiently large, select
an appropriate matrix in Tn", and inspect the image of that matrix under
the representation p.
In the special case in which each ,pk is a direct sum of identity map-
pings, i.e., ,pk is a standard embedding, the representation which we have
constructed is just the representation of the standard limit algebra intro-
duced by Roger Smith. For a description of this representation, see toP].
Depending on the locations of the distinguished copy of the identity, the
representation will act on a Hilbert space with a basis indexed by the in-
tegers, by the non-negative integers, or by the non-positive integers. The
choice here is arbitrary, but since the representations in this special case are
all *-extendible, the C* -algebras generated by the images are all isomor-
phic. (Indeed, they are all the UHF algebra associated with the appropriate
supernatural number.)
In the general case, it is possible to have two completely isometric
representations, p and a, of A with C*(p(A)) not isomorphic to C*(a(A)).
It is desirable, then, to find a representation for which the C* -algebra
generated by the image is the C* -envelope in the sense of Hamana. We
shall show below that p, as constructed above, has just this property. It is
in this sense that p is canonical.
First, we provide a brief review of the idea of a C* -envelope. The ap-
propriate setting for this is actually the category of unital operator spaces
with unital complete order injections, but we only need to deal with oper-
ator algebras so we will restrict the discussion to that domain.
If A is an operator algebra, then a C*-extension of A is a C*-algebra

B together with a unital complete order injection P of A into B such that
C*(p(A» = B. A C*-extension B is a C*-envelope of A provided that,
given any operator system, C, and any unital completely positive map
T: B -+ C, T is a complete order injection whenever TOp is. Hamana [II]
proves the existence and uniqueness (up to a suitable notion of equivalence)
of C* -envelopes; further, he shows that the C·-envelope of A is a minimal
C* -extension in the family of all C·-extensions of A.
After proving the existence of C·-envelopes, Hamana then uses this
to prove the existence of a SHov boundary for A. The SHov boundary is
a generalization to operator spaces of the usual notion of SHov boundary
from function spaces; it was first developed by Arveson in [AI). Here is
the appropriate definition. Let B be a C* -algebra and let A be a unital
subalgebra such that B = C*(A). An ideal J in B is called a boundary ideal
for A if the canonical quotient map B -+ B / J is completely isometric on
A. A boundary ideal which contains every other boundary ideal is called
the Silov boundary for A.
Hamana shows that if B is a C·-extension for A, then the C* -envelope
of A is isomorphic to B / J, where J is the SHov boundary for A. In par-
ticular, B is the C*-envelope for A if, and only if, the SHov boundary is O.
This last fact is the one which we shall use to show that C*(p(A)) is the
C* -envelope of the image p(A) for the representation p defined above.
The first step needed to accomplish this goal is to compute the C*-
algebra generated by each Pk(Tn,.). The following simple lemma is helpful:
Lemma. Let Tn be the n x n upper triangular matrices acting on C n and
let p and q be distinct intervals from the nest of invariant projections for
Tn. Then
Proof. Here, we view pap and qaq as being restricted to the ranges pen and
qCn of p and q respectively. Since p and q are distinct, there is an element a
of Tn such that one of pap and qaq, say pap, is non-zero while the other is o.
So we have an element of the form bEBO in C·( {papEBqaq: a E Tn }). From
this and the fact that the C*-algebra generated by Tn is M n , it follows
that 8(pCn) EB 0 ~ C*({pap $ qaq: a E Tn}. This in tum implies that
O$8(qCn) ~ C*({pap$qaq: a E Tn} and hence that 8(pCn) $8(qCn) ~
C· ( {pap $ qaq: a E Tn }. The reverse containment is evident.
By using the obvious extension of this lemma to multiple direct sums
(including countable direct sums), we can describe the C·-algebra,
C·(Pk(Tn ,.»,both as a subalgebra of 8(1t) and as an abstract (finite-
dimensional) C·-algebra. Indeed, write Pk = E(D "7i and, for each i, let qi
be the interval in the nest for Tn" to which "Ii is a compression. Then
C*(Pk(Tn ,,)) = {(bj ) E L(D8(qjC n,,): bi = bj whenever qi = qj}.
If r1, ... , r m is the list of distinct intervals to which the 1Ji are compressions,
then
m
C*(Pk(Tn,.)) ~ LeB(rjCn,.).
j=l
We shall need both the isomorphism class of C*(Pk(Tn,,)) and its actual
expression as an operator algebra acting on 'H. Also, since C*(p(A)) is the
closure of the union of the C*(Pk(Tn,,)), we have proven the following:
Proposition. Let A be the direct limit of a system of full upper triangular
matrix algebras with compression embeddings and let p be the representation
of A defined above. Then C*(p(A)) is an AF C*-algebra.
Our main result is the following theorem.
Theorem. Let A be the direct limit of a system of full upper triangular
matrix algebras with compression embeddings and let p be the representation
of A defined above. Then C*(p(A)) is the C*-envelope of p(A).
Proof. In order to prove that C*(p(A)) is the CO-envelope of p(A), it is
sufficient to show that the SHov boundary of p(A) is O. Since the SHov
boundary is the largest boundary ideal, we need merely show that if J
is a non-zero ideal then J is not a boundary ideal. This requires proving
that the canonical quotient map C*(p(A)) -+ C*(p(A))j J is not completely
isometric when restricted to p(A). We shall, in fact, prove that p(A)nJ =I- 0;
the quotient map is not even isometric.
For convenience, let Bk = C*(Pk(Tn,.)) and B = C*(p(A)). Each Bk is
isomorphic to a direct sum of full matrix algebras, one for each compression
map which appears in the expression for Pk. Since the identity must appear,
one summand must be Mn,.. This summand is the largest rank summand
and it appears one time only. M n ,,-l appears at most two times, reflecting
the fact that the nest for Tn" has exactly two interval projections of rank
nk -1. More generally, M n ,. _ j appears at most j + 1 times in the expression
for B k .
Thus, the isomorphism class of Bk is Mn " $ Mtl $ ... $ Mt ., where
tt, ... ts are integers less than nk. In the Bratteli diagram for B, the kth
level has a node for each summand in the isomorphism class of Bk. Next,
we need the following observation: given k and a summand Md in the
isomorphism class of Bk, there is an integer j > k such that Md partially
embeds into the summand Mn; in the isomorphism class of Bj .
The observation is verified by inspecting the chain of finite dimensional
CO-algebras acting on 'H. Now, Pk(Tn",) is an infinite sum whose terms
are selected from finitely many compressions of Tn". Consequently, there
exist integers a :5 0 and b ;::: 0 such that if p is the projection onto the
linear span of {ei: a :5 i :5 b}, then p is reducing for C*(Pk(Tn,,)) and
every compression, and in particular the one corresponding to Md, appears
in the restriction of C*(Pk(Tn,,)) to p. Then, by the way in which the
representations are constructed, there is an integer j > k so that the 1]0 = id

term in Pj acts on a subspace of 1£ which includes the range of p. (This was
the reason for "anchoring" the Pj so that the 170 terms act on the vector
eo and that the support space for the 1JO terms "grows away" from eo.)
This shows that in the abstract Bratteli diagram, each node eventually
partially embeds into a node corresponding to the identity summand (the
Mn; node).
Now let J be a non-zero closed two-sided ideal in B. By a result in [EI,
J is the closure of the union of the J n B k • In particular, there is a positive
integer k such that J n Bk =I- O. Since Bk is isomorphic to a finite direct
sum of full matrix algebras, J n Bk is isomorphic to a direct sum of some
of those algebras, with O's as the remaining summands. Let Md be one
of the non-zero summands appearing in J n B k • Let j > k be an integer
such that the Md term partially embeds into Mn ;. It now follows that
Mn; is one of the non-zero summands for J n B j . By utilizing the support
projection for the Mn; term (a reducing projection for C·(pj(Tn ;))), we
see that the subalgebra of C·(pj(Tn ;)) isomorphic to Mn; is contained in
J. This subalgebra consists of all sequences (bi ) in C·(pj(Tn ;)) for which
bi = 0 whenever the corresponding interval qi is not the identity and the
remaining bi are all equal.
The proof is now completed by observing that there is an element of
this subalgebra which lies in pj(Tn ;). Indeed, let v be the matrix unit in
Tn; which has a 1 in the extreme upper right corner and zeros elsewhere.
Then the compression of v to any interval q other than the identity is zero.
Thus, pj(v) E J n p(A), and the proof is complete.
Remark. We have shown that the C·-envelope of a direct limit of full up-
per triangular matrix algebras with compression embeddings is an AF C·-
algebra. If, on the other hand, A is the direct limit of full upper triangular
matrix algebras with •-extendible embeddings, then A can be represented
as a generating subalgebra of a UHF algebra. Since UHF algebras are
simple, it is immediate that the generated C·-algebra is the C·-envelope.
More is true. It is easy to show that if A is a generating subalgebra of an
AF C·-algebra B and if A contains a Stratila-Voiculescu masa, then B is
the C·-envelope of A. Thus, if A is the direct limit of a system of digraph
algebras with •-extendible embeddings, then the C·-envelope of A is an AF
algebra.
The simple proof of this last fact does not apply to compression limits.
Examples in the next section will show that the image of the limit algebra
under the representation P need not contain a masa in the generated C·-
algebra. (The natural diagonal in p(A) need not be a masa in C·(p(A)).)
The following problem is suggested.
Problem. Is the C·-envelope of a direct limit of digraph algebras with com-
pression embeddings an AF C·-algebra?
Remark. In [PI), Power studies a direct limit of digraph algebras with

embeddings which are neither *-extendible nor compression embeddings.
For systems of tri-diagonal algebras with certain natural non-*-extendible
embeddings, Power shows that the limit algebra is completely isometrically
isomorphic to a generating subalgebra of an appropriate Bunce-Deddens
algebra. Since this latter algebra is simple, it is the C* -envelope of the
limit algebra for the tri-diagonal system. In particular, this provides an
example of a system of digraph algebras whose limit algebra has a C*-
envelope which is not AF.
Remark. Since p(A) is weakly dense in Alg (N), C*(p(A)) is an irreducible

C* -algebra. If it happens that p(A) contains a non-zero compact operator,
then Arveson's boundary theorem [A2) immediately implies that the Silov
boundary for p(A) is o. The boundary theorem is a deep theorem, so the
argument above could be considered more elementary.
In most of the examples in the next section, p(A) contains non-zero
compact operators. Here is how to determine in general if p( A) contains
compact operators. Each representation Pk of Tn" can be written as an infi-
nite direct sum LEI) 7Ji, where the TJi'S are compressions. Let Zk be the num-
ber of times that 7Ji = id in the expression for Pk. Then Zk E { 1, 2, ... , 00 }
and the sequence Zk is decreasing. Thus there are two possibilities: all
Zk = 00, or there is a finite integer y such that Zk = Y for all large k.
In the first case p(A) will contain no non-zero compact operators. In the
second case, p(A) contains finite rank operators and C*(p(A)) contains all
compact operators.
For the first assertion, for any a E Tn", lIall = IIpk(a)1I = IIPk(a)lIess. It
follows from the density of UPk(Tn ,,) that IIp(a)1I = IIp(a)lIess for all a E A.
Thus, p(A) contains no non-zero compact operators. As for the second
assertion, choose k so that Zk is finite. Let v be the matrix unit in Tn"
which has a 1 in the extreme upper right hand comer and zeros elsewhere.
Then Pk(V) has finite rank and lies in p(A). Since p(A) is weakly dense in
Alg (N), it has no non-trivial reducing subspaces. Consequently, C*(p(A))
is irreducible and contains a non-zero compact operator, which implies that
it contains all compact operators.
IV. SOME EXAMPLES
A. Fix an integer i between 1 and n and consider the system:
tPl T.n+kl --+

T.n --+ tf>2 T.n+kl+~ --+
tP3
... -+
Ai·
Each embedding tPn is given by

where lk is the k x k identity matrix. For example,
J
<lii
[a
a 1-+ <lii 1-+
<lii
<lii 1 1-+
<liJ
aii
The representation p of the limit algebra Ai will act on a Hilbert space

'H with othornormal basis {ej} indexed by N U {O}. Let Pk denote the
projection on the linear span of { eo, ... , ek } and N the nest consisting of
the Pk'S. Also, let lCN be the algebra of compact operators which leave N
invariant.
It is easy to check that for each j, the image of any matrix under Pj
is the sum of a finite rank matrix and a scalar multiple of the identity I.
Consequently, p(Ai) ~ lCN + Cl. The containment will, in fact, be proper.
For each k, let fk = Pk - Pk-i. Let
The image of each Pk is clearly contained in Bi , whence p(Ai) ~ Bi . To

see the reverse containment, let b E B i . There is a compact operator c in
lCN and a scalar a such that b = c + al. Observe that
(Strictly speaking, this is abuse of notation, since, for example, f .. bli is not
a scalar but an infinite matrix with a single non-zero entry, viz., bii in the
ith position on the diagonal. In cases like this we identify the two and the
meaning should be clear.)
Now, fix an integer k > i such that bkk is close to a and PkCPk is close
to c. Then a = Pkbpk + apt is in p(Ai) and is close to b. Thus, p(Ai) is
dense in, and hence equal to B ...
Fact. The family of algebras Ai are pairwise non-isomorphic.
Indeed, the identity and the Pk are the only projections in lCN+Cl (or
in B('H), for that matter) which are invariant under P(Ai). But Pk E p(Ai)
if, and only if, k < i. Thus, keeping in mind that the indexing starts with
0, there are exactly i + 1 invariant projections in Ai. This establishes the
claim. Note also that the C* -envelope for each Ai is lC + Cl. The "linking
condition" that the ith diagonal element is equal to the limit of the diagonal
elements disappears in the passage to the generated C* -algebra. This is an
example in which the diagonal of p(A) is not a masa in C*(p(A)).
B. This time consider the system
with the embeddings tPj defined by
where P = n + kl + ... + kj_l, Le., Q pp is the last entry on the diagonal of

the matrix Q.
The Hilbert space 'H and the nest N are as before. This time, however,
p(A) = ICN + Cl; the arguments are much the same as in the previous
example. The limit algebra A now has infinitely many invariant projections,
and so is not isomorphic to any of the ~ from the previous example. Also,
the C* -envelope of A is evidently once again IC + C1.
C. For this example, it is most convenient if the indices for each matrix
algebra T m are drawn from the set { -m + 1, ... , -1,0 }. The basis for
'H is indexed by the non-positive integers and the Pk will be the obvious
infinite rank projections. Fix an integer i such that -n + 1 :5 i :5 O. The
embeddings in the system
are given by
Arguments analogous to the one in example A show that
Once again, Bi ~ B j if, and only if i = j, and the C·-envelope of each Bi

is IC + Cl. This is another example in which the diagonal of p(A) is not a
masa in the C* -envelope.
The limit algebras in examples A and C are all non-isomorphic. All
that needs to be checked is that Bi '1- Ai for each i, since when i =I j, Bi
and Aj have a different number of invariant projections. Observe that if
P is a projection in P(Ai) n p(~). which is invariant under p(~) and is
not the identity, then for any sequence Xj in Ai, the set {Xjp } is linearly
dependent. On the other hand, if p =I 0 is an invariant projection in
P(Bi) n P(Bi)*. then there exist operators Xl,X2 •... E p(Bi) such that the
set {Xjp } is linearly independent.
D. In this example, 1£ is a Hilbert space with basis indexed by N U {O},

N = {pic} and ICN are as before, and V denotes the algebra of diagonal
matrices. Also, d will denote the conditional expectation from Tn onto Dn;
in other words, if a is an upper triangular matrix, d(a) is the diagonal part
of a. Note that d is a direct sum of rank one compressions.
Now consider the stationary system
,... 4>1 ,... tf>2 ,... "'3
.L2 --+ .L4 --+ .L8 --+ ... -+
A
in which tPlc: T 2" -+ T2 ,,+1 is given by tPlc(a) = a EB d(a).

The representations Pic: T2" -+ICN+V are given by plc(a) = aEBd(a)EB
d(a) $ ... = a $ I:E9 d(a). The image p(A) of the limit algebra under the
canonical representation is thus a subalgebra of ICN + 'D. Once again, the
image will be a proper subalgebra.
Definition. We say that an element b E V is periodic with period p if
bm +p = bm , for all m. Here, bm denotes the matrix entry bmm . We say b is
dyadic periodic is the period p is a power of 2. Let'DAP(2OO) denote the
closure in the norm topology of the dyadic periodic elements of 'D.
Let ICJv = {a E ICN: d( a) = O}. Clearly, Pic (T2" ) ~ ICJv + VAP(2OO),
for each k. Thus, p(A) ~ ICJv + 'D AP(2OO). Furthermore, an argument
analogous to the one used in example A shows that we have equality:
p(A) = ICJv + 'DAP(2OO). The C· -envelope of A is therefore IC + VAP(2OO).
E. The notation is the same as in the previous example, but the embedding
is changed to
a 1--+ a $ dlh(a) EB dlh(a)
where dlh(a) is the diagonal last half of a. (When a is a 21c x 21c matrix,
dlh(a) is a 21c- 1 x 21c - 1 matrix.)
If a is a matrix in T2", then the first 21c- 1 diagonal terms in Pic (a) bear
no relation to the remaining diagonal terms in plc(a). This difference in
behaviour compared with the last example results in a slightly different
image for the limit algebra under the canonical representation: p(A) =
ICN + 'D AP(2OO). The C·-envelope is unchanged.
F. Consider the stationary system
4>1 ,... <1>2 ,... "'3
T2 --+ .L4 --+ .L8 --+ ... -+
A
in which tPlc: T21c -+ T2 ,,+1 is given by tPlc(a) = a$lh(a)$lh(a). Here lh(a)

denotes compression to the last half of a, resulting in a 21c - 1 x 21c - 1 matrix
when a is 21c x 21c. Let S(2OO) denote the norm closure of all the dyadic
periodic matrices. (A matrix on 1£ is periodic with period p if it has the
form a $ a $ .. " where a is a p x p matrix.) The intersection of S(2OO) and
Alg (N) will be denoted by SN(2OO). With this notation, the image of the
limit algebra A under the canonical representation p is ICN + SN(2OO) and
the C*-envelope is IC + S(2OO).
G. We conclude by mentioning an old example, the system
'T' 4>1 'T' 4>2 'T'

.L2 --+ .L4 --+ .L8 --+ ... -
4>3 A
with the standard embeddings of multiplicity 2: 4>k(a) = a 6) a. The

canonical representation is the Smith representation described in [OP);
peA) = SN(2°O) and the C·-envelope is the UHF algebra 8(200).
REFERENCES
[AI] W. Arveson, Subalgebras of C"-algebras, Acta Math. 123 (1969),141-224.
[A2] _ _ , Subalgebras of C"-algebras II, Acta Math. 128 (1972), 271-308.
[BRS] D. Blecher, Z. Ruan, and A. Sinclair, A chamcterization of opemtor algebras, J.
Functional Analysis 89 (1990), 188-201.
[B] O. Bratteli, Inductive limits of finite dimensional C" -algebras, Trans. Amer.
Math. Soc. 111 (1972), 195-234.
[H] M. Hamana, Injective Envelopes of Opemtor Systems, PubI. RIMS, Kyoto Univ.
15 (1979), 773-785.
[L] D. Larson, Structured Limit Algebras, This journal.
toP) J. Orr and J. Peters, Some representations 0/ TAF algebras, Preprint.
[PI] S. C. Power, Non-sel/-adjoint opemtor algebras and inverse systems 0/ simplicial
complexes, J. Reine Angewandte Mathematik 421 (1991), 43-61.
[P2] S. C. Power, Limit algebras; an introduction to subalgebras 0/ CO -algebras, Pit-
man Research Notes, Longman, 1992.
University of Alabama
Tuscaloosa, AL 35487
INVERSE SEMIGROUPS, GROUPOIDS AND A
PROBLEM OF J. RENAULT
ALAN L. T. PATERSON
September 1992
Abstract
In [8], J. Renault showed that a topological groupoid G relates
to inverse semigroups through its ample semigrou.p Ga. In the case
when G is r-discrete, the semigroup G a is "large" and determines the
topology of G. The main theme of this paper is the following natural
problem which was raised by Renault : given an inverse semigroup
S (assumed, for convenience, unital) does there exist a (Hausdorff)
r-discrete groupoid G with S "determining" G as an inverse sub-
semigroup of G a ? What kind of uniqueness can we expect? Renault
shows that there exists such a groupoid in the case where S is the
Cu.ntz inverse semigrou.p On, the groupoid G in that case being the
Cu.ntz groupoid On.
We show that the answer to both questions is negative in general.
The natural class of groupoids associated with an inverse semigroup
S is that of S -grou.poids. Such groupoids are not always Hausdorff
but there always exists a faithful S-groupoid whose representation
theory is essentially the same as that of S. We describe briefly a
construction which produces this and many other S-groupoids.
1 Introduction
In [8], J. Renault showed that there is a natural way of relat-

ing groupoids to inverse semigroups : each groupoid G is asso-
ciated with its inverse semigroup of G-sets. Of particular signif-
icance for the topological theory of groupoids are the r-discrete
groupoids with their related ample inverse semigroups of com-
pact open G-sets. This suggests that the categories of inverse
semigroups and r-discrete groupoids are closely related, and
raises the possibility of using one category to study the other.
Inverse semigroups are conceptually simpler than groupoids, but
the latter are better developed for the purposes of representation
theory and operator algebras. The main objective of this study
is that of understanding the C* -algebras associated with inverse
80 A. L. T. Paterson
semi groups in terms of groupoid C* -algebras. To achieve this,

we need an algebraic framework for producing groupoids from
inverse semigroups. The present paper begins the development
of such a framework.
We take as our starting point a problem which was raised
by J. Renault in his book ([8, p.142]). Let S be a unital in-
verse semigroup. Does there exist an r-discrete groupoid G with
S "determining" G as an inverse subsemigroup of the ample
semigroup Ga of G? What kind of uniqueness can we expect?
Renault shows that there exists such a groupoid in the case
where S is the Cuntz inverse semigroup On, the groupoid G in
that case being the Cuntz groupoid On.
In §2, we briefly discuss the relevant facts needed from the
theories of inverse semigroups and groupoids. We then introduce
for a given inverse semi group S a class of topological groupoids
called S-groupoids and use On and On as illustrations. The
above problem is then formulated in terms of the existence and
uniqueness of a faithful, Hausdorff S-groupoid.
Two examples relevant to the above problem are discussed.
In the first, we exhibit an inverse semigroup which does not ad-
mit a faithful Hausdorff S-groupoid, and in the second, show
that even if a faithful S-groupoid exists, it need not be unique.
(The author is grateful to J. Renault for pointing out that Ex-
ample 1 has also been constructed in earlier work by A. Nica.)
In §3 we sketch a general construction which produces a large
class of S-groupoids. In particular, it produces for every inverse
semi group S a faithful S-groupoid Gu '
lt can be shown that Gu is the universal S-groupoid, and
this is the groupoid whose representation theory gives that of
the inverse semigroup S. The details will be given elsewhere.
The author is grateful to the referee for his helpful comments,
and in particular, for pointing out an error in the author's earlier
treatment of the topology of the Cuntz groupoid.
2 Inverse semigroups and S-groupoids
A semigroup S is called an inverse semigroup ([4, p. 129]) if

for each S E S, there is a unique element s* E S such that
ss*s = s, s*ss* = s*. (1)
Inverse semigroups 81
It can be shown that the map s-+s· on an inverse semi group

S is an involution. For functional analytic purposes, an in-
verse semi group is just a *-semigroup of partial isometries on
a Hilbert space ([1, 11, 2]). We will assume throughout that
the inverse semi groups under consideration have units. Inverse
semigroups can also be characterized as the *-subsemigroups of
the semigroup I(X) of bijective maps T: A-+B where A and B
are subsets of a set X. Such maps are called partial one-to-one
maps. The product of two such partial one-to-one maps T1, T2
is the composition of the two, defined wherever it makes sense.
(So, for example, x is in the domain of TIT2 if and only if T2 (x)
is in the domain of Td The involution on I(X) is the inverse
map Tl-+Tl1.
Let S be an inverse semigroup. The set E of idempotents in
S is a semilattice with e A f = ef in E. It is obvious from (1)
that ss·, s· sEE for all s E S. Less obviously, s·es E E for all
sE S and e E E. T then 5.1]Howie.
We now recall the definition of a groupoid. A groupoid is a
small category with inverses. Spelling out the details, a groupoid
([8, p.6]) is a set G together with a'subset Q2 c Gx G, a product
map (u,v)-+uv from G2 to G, and an inverse map u-+u- 1 such
that
(i) if {u, v), (v, w) E G2, then (uv, w), (u, vw) E G2 and
(uv)w = u(vw) (2)
(ii) (u, u- 1 ) G2 for all u

E E G, and if (u, v) (resp. (w, u))
belongs to G2, then
u-1(uv) = v, (resp. (wu)u- 1 = w). (3)
The elements of the set G2 are called the composable pairs.

The unit space GO of G is the set {uu- 1 : u E G}. Important
maps r, s from G onto GO are defined :
(4)
The elements r(u) and s(u) are called the mnge and source of
u.
A subset A of the groupoid G is called a G-set ([8, p. 10])

if the restrictions of r and s to A are both one-to-one. This
property is equivalent to AA-I U A-I A c GO. The family of
G-sets is an inverse semigroup with product (A, B)~AB and
involution A~A-I.
If G is in addition a topological space with G2 given the
relative topology from G x G, then G is called a topological
groupoid if the product and inverse maps are continuous. The
maps r and s are obviously continuous.
In the sequel, topological groupoids are not assumed to be
Hausdorff.. ( The topological groupoids in [8] are assumed to be
Hausdorff.) The reason is that, as we shall see, the topological
groupoids associated with inverse semigroups are not Hausdorff
in general. Non-Hausdorff groupoids do occur in other contexts.
For example, the holonomy groupoid of a foliation is not usually
Hausdorff, and the Hausdorff condition is not required for the
topological groupoids of [10].
The topological groupoids that we will be considering will be
r-discrete.
A topological groupoid G is called r-discrete ([8, p.19]) if
every u E G has an open neighborhood V such that r Iv and
s Iv are homeomorphisms onto open neighborhoods of r(u) and
s(u) respectively. It immediately follows that if Gis r-discrete,
then GO is open in G.
Let Ga be the set of compact Hausdorff open G-sets in G. It
can be shown that Ga is an inverse subsemigroup of the semi-
group of G-sets.
We now introduce the groupoids naturally associated with
an inverse semi group S. A pair (G,,¢), where G is an r-discrete
groupoid, is called an S-groupoid if '¢ : S~Ga is a homomor-
phism such that :
(i) '¢(1) = GO;
(ii) the family of sets of the form
where, for e, el,···, en E E,

Inverse semi groups 83
is a basis for the topology of GO. (Here, AC = G \ A for

A C G.)
(iii) USES'IjJ(S) = G.
It follows (cf. [8, p.20]) that the family of sets of the form
Ue,el. ... ,en 'IjJ (s) is a basis for the topology of G.
An S-groupoid (G, 'IjJ) is called faithful if 'IjJ is injective.
A good example of an S-groupoid introduced by J. Renault is
the Cuntz groupoid On where S is the Cuntz semigroup On. For
the convenience of the reader and in order to relate S-groupoids
to the work of Renault, we now briefly discuss On and On.
For convenience, let 1 < n < 00. The semi group On is
generated by the unit 1, the zero 0 and elements Pi, qi (1 ~ i ~
n) where qjPi = bij 1. Apart from 1, 0, the elements of On can
be written uniquely in the form
Paqj3 = Pal' .. Pa r qj34 ... qj31
with a = al'" a r , 13 = 131 ... j3s and 1 ::; ai,j3j < n. The
involution is given by :
(Paqj3)* = Pj3Qa.
The Cuntz groupoid On ([8, pp. 138-140)) is defined by Re-

nault as a reduction of a semidirect product groupoid Gn x 1> Z
with G n a certain Glimm groupoid. The following convenient
realization of On as a groupoid is given in [8, p.140j.
Let Y = {1,"', n}N, the set of sequences in 1,2,"" n with
the product topology. Then On is the set of triples (u, v, z)
where u, v E Y and Ui = Vi-z for all but finitely many i. Multi-
plication and inversion in On are given by :
(u, v, z)(v, w, z') = (u, w, z + z'), (u, v, Z)-1 = (v, u, -z).
The unit space can be identified with Y.

In this realization of On, the topology is determined by the
basis described below. Let a = a1 ... a r+z and 13 = 131 ... j3r be
finite sequences with ai,j3i E {I"" ,n}. Note that if.e denotes
length for a finite sequence, then
£(a) - £(13) = z.
For W E Y, elements aw,/3w E Yare defined by juxtaposition.

Every element (u, v, s) E On can be written in the form:
(aw, /3w, z)
with 0'., /3 as above. Given finite sequences 0'., /3 as above and an
open subset W of Y, we define Sa./3.w C On by :
Sa./3.w = {(aw, /3w, z) : w E W}.
The family of all such sets Sa./3.w is a basis for a topology on

On. This is the groupoid topology for On and On is an r-discrete
(Hausdorff) groupoid.
Renault ([8, p.141]) defines an isomorphism 'l/Jn : On~O~ by
setting: 'l/Jn(1) = Y, 'l/Jn(O) = 0 and
'l/Jn(paq/3) = {(a,,/3,,£(a) - £(/3)) :, E V}.
It is easily shown that (On' 'l/Jn) is a faithful On -groupoid. (Re-

nault's result [8, Proposition 2.3, p.141] gives detailed informa-
tion about the structure of the compact open subsets of On and
O~.)
This example led Renault to raise the question stated in
§l. This can be reformulated as follows: Given an inverse
semigroup S, does there exist a faithful (Hausdorff) S-groupoid
(G, 'I/J), and if so, is it unique up to S -isomorphism?
In this connection, we discuss two examples. The first shows
that a faithful Hausdorff S-groupoid need not be unique, while
the second shows that a faithful Hausdorff S-groupoid need not
exist.
Example 1 The author is grateful to J. Renault for point-

ing out that this example appears in a preprint of A. Nica enti-
tled "G* -algebras generated by isometries and Wiener-Hopf op-
erators". The groupoid En appears (explicitly or implicitly) in
the work of a number of people, and is called the Guntz- Toeplitz
groupoid since its G* -algebra is the Cuntz-Toeplitz algebra ([5]).
We have seen that On is a faithful Hausdorff On-groupoid.
This On-groupoid is not unique in this respect. To get a differ-
ent one, let X = Y U Z where Z is the set of finite sequences
Inverse semi groups 85
Q from 1,2"", n. A sub-basis for X is given by sets of the

form B Q and their complements, where Q E Z and B Q consists
of all sequences, finite or infinite, in X beginning with Q. In
particular, for Q E Z, the singleton
{ Q} = BQ n B~l:1 n ... n B~n
is both open and closed in X and Y c X has the relative
topology. Further, if y E Y and Q r = Yl ... Yr, then Qr-+Y.
It is readily checked that X is a compact Hausdorff space
and that we can define an On-groupoid En in the same way as
On with X in place of Y. The groupoids On and En are not On-
isomorphic. For if they were, then their units spaces would be
homeomorphic, which is impossible since X has isolated points
while Y does not.
In groupoid terminology ([8, pp.8, 10]), Y is a closed subset
of X, invariant under the natural action of On on X, and On is
the reduction of En by Y.
Example 2 Let S be the inverse semigroup whose elements

are the identity 1, the zero 0, idempotents en (n ~ 1) and z. The
product on S is abelian and is determined by the expressions :
zen = en, enem = 0 (n =I m), Z2 = 1.
( The semi group S is an abelian Clifford semi group and {I, z}

is the maximal subgroup of S with 1 as identity.) We will show
that there does not exist a faithful Hausdorff S-groupoid.
Suppose that (H, 't/J) is a faithful S-groupoid. Let
w = (U 't/J(en)) U '¢(O).
00
n=l
Then W is open in HO. Since zen = en and zO = 0, it follows

that W C't/J(z). Since z =I 1, we have 't/J(z) =I't/J(l). Now 't/J(1) =
HO, and since 't/J(z) is a G-set, it follows that T = HO \ W =I 0.
We now claim that T contains exactly one point. For suppose
not, and let Q,f3 E T, with Q =I 13· Then Q,f3 E 't/J(1), Q,f3 E
't/J(en)C for all nand Q,f3 E 't/J(O)c. Let I, It" .. ,In E E and
U = U/,/l'''',/n be as in the definition of S-groupoid. Then either
both Q,f3 E U or both Q,f3 E UC • Since the set of such U's
is a basis for the Hausdorff space HO, we can't separate a, (3

in HO and a contradiction follows. So T contains exactly one
element a. Since 't/J(z) is a G-set and Z2 = 1, there exists a
unique element U E 't/J(z) such that s(u) = a = r(u). Note that
'¢(z) = {u} U W and 't/J(I) = {a} U W. Since z =I- 1 and H is
faithful, we have u =I- a.
We claim that u and a cannot be separated in H. Indeed, let
Vi, V2 be neighborhoods of u and a in H. We can suppose that
Vi = U1""(z)
'f'
where U1 = Ue, '1'
e' ... e"
'n
Now using the fact that S
is abelian, we have a = s(u) E s(Vi) = U1 • We will have shown
that u, a cannot be separated if we have
s(Vi) n V2 n W =I- 0. (5)
(For if W E s(Vd n V2 n W, then, since 't/J( z) is a G-set containing
W, we would have W E U1't/J(z) and so W E Vi n V2.) To prove
(5), we observe that a E (s(Vi) n V2) \ Wand a E 't/J(en)C for all
n. Then the open set s(V1) n Va contains a set of the form
't/J(I) n 't/J(el)C n ... n 't/J(er)C = HO n 't/J(ed c ••• 't/J(err
for some r. Since er+1 =I- 0, we have 't/J(Cr+d =I- 't/J(O), and there
exists W E 't/J(er+d \ 't/J(O). Now for 1 ~ i ~ r, eier+1 = 0 so that
,¢(ei),¢(er +1) = '¢(o). So W E '¢(eiY' So W E (HO n '¢(ed c •• . n
't/J(erY) n't/J(er +1»
c s(Vi) n V2 n W, and u and a cannot be
separated.
So there does not exist a faithful, Hausdorff S-groupoid for
this S.
QUESTION Give conditions on an inverse semigroup S

which will ensure the existence of a faithful Hausdorff S-groupoid.
3 Constructing S-groupoids
Let S be a unital inverse semigroup. This section is devoted

to a brief description of a construction which will produce a
universal S-groupoid and many others. The former is always
faithful and its Hilbert space representation theory can be shown
to coincide in the appropriate sense with that of S. A detailed
description of the construction will be given elsewhere.
Let 4> be an inverse semi group homomorphism from S onto
an arbitrary inverse semi group T. Let E, F be the idempotent
semilattices of S and T. We can construct an S-groupoid from
4> as follows.
The map 4> lifts in the obvious way to a homomorphism from
i 1 (E) onto i 1 (F). The enveloping C* -algebra of i 1 (F) can be
identified with C(Y) where Y is the set of non-zero semicharac-
ters on F with the pointwise topology. Hence we have a homo-
morphism denoted by 4>' from i 1 (E) into C (Y). Let E' be the
subset of il (E) consisting of elements of the form
e' = e(l - el) ... (1 - en)

with e, ei E E. Using the fact that the elements of E commute,
E' is a semilattice in the product inherited from i 1(E). Clearly,
E is a subsemilattice of E'. Restricting 4>' to E' gives a ho-
momorphism into the projections in C(Y). (These determine a
"localization" on Y in the language of Kumjian([6]).) There is a
natural right action y-+y.s (y E D s = {y E Y : y( 4>( ss*)) = I})
of S on Y by partial one-to-one maps.
The subset T' = {f't : f' E F', t E T} is an inverse semi group
in il(T) and we have a pair (a, ¢) of homomorphisms from E', S
into T' where a = ¢ lEI. This is a covariant pair for the inverse
semigroup semi direct-product E' XES where
E' XES = {(e' , s) : e' :5 ss*, S E S}
where (e', s) is identified with (e', t) if there exists e E E U {O}

with e' :5 e and es = et. The product and involution E' XE S
are given by :
(e',s)(f',t) = (e'sf's*,st), (e',s)* = (s*e's,s*).
The pair (a, ¢) therefore gives a homomorphism X : E' XES -+T'.

The S-groupoid G associated with ¢ is then given as follows. Let
Y Xs S = {(y, s) : y E D s}.
Let R be the equivalence relation on Y x s S where
(y, s) R( z, t) if and only if y = z and there exists e' E E' such
that y ~ e' and x((e', s)) = x((e', t)). Let G = (Y XS S)/ R.
The equivalence class of (y, s) in G is denoted by (y, s). Then

G is an r-discrete groupoid in a natural way. The product and
inverse for G are given by :
""""-"""7"-1
(y, s)(y.s, t) = (y, st), (y, s) = (y.s, s*).
A basis for the topology on G is given by sets of the form Du,s,

where s E S, U is an open subset of D s , and
Du,s = {(y,s) : y E U}.
Let ¢ : S -+Ga be defined by :
¢(s) = {(y, s) : y E Ds}.

Then (G, ¢) is an S-groupoid. The case where S = T and
¢ is the identity map gives the universal S-groupoid (in the
appropriate sense).
REFERENCES
[1] B. A. Barnes, Representations of the iI-algebra of an in-

verse semigroup, 'frans. Amer. Math. Soc. 218(1976), 361-
396.
[2] J. Duncan and A. L. T. Paterson, C*-algebras of inverse
semigroups, Proc. Edinburgh Math. Soc. 28(1985), 41-58.
[3] J. Duncan and A. L. T. Paterson, C* -algebras of Clifford
semigroups, Proc. Roy. Soc. Edinburgh, AlII (1989), 129-
145.
[4] J. M. Howie, An introduction to semigroup theory, Aca-
demic Press, London, 1976.
[5] P. E. T. Jorgensen, L. M. Schmitt and R. F. Werner, Q-
canonical commutation relations and stability of the Cuntz
algebra, Pacific J. Math. (to appear).
[6] A. Kumj ian , On localizations and simple C* -algebras, Pa-
cific J. Math., 112(1984), 141-192.
[7] C. C. Moore and C. Schochet, Global analysis on foliated
spaces, Mathematical Sciences Research Institute Publica-
tions, No.9, Springer-Verlag, New York, 1988.
[8] J. Renault, A groupoid approach to C*-algebras, Lecture
Notes in Mathematics, No. 793, Springer-Verlag, Berlin,
1980.
[9] J. Renault, C* -algebras of groupoids and foliations, Proc.
Sympos. in Pure Math., 38(1982), Part 1, 339-350.
[10] J. Renault, Structure des ideaux des produits croises de
C* -algebras de groupoides, Preprint, 1988.
[11] J. R. Wordingham, The left regular *-representation of an
inverse semigroup, Proc. Amer. Math. Soc. 86(1982), 55-58.
DEPARTMENT OF MATHEMATICS
UNIVERSITY OF MISSISSIPPI
UNIVERSITY, MS 38677
On Banach Algebras Generated by
Two Idempotents
Y sette Weiss
A matricial analogue of the Gel 'fand map is constructed for the el-
ements of a unital Banach algebra generated by two idempotents. For
commutative unital algebras, the Gel'fand map realizes an imbedding into
the algebra of all continuous functions on a well-defined set, namely the
space of maximal ideals. Analogously, a semisimple algebra can be shown
to be isomorphic to a subalgebra of all continuous 2 x 2 matrix-valued
functions on a compact set with possible poles of order two at two points,
and continuous trace. Conditions for the semisimplicity of the algebra are
given, as well as a formula for the computation of the spectral radii of its
elements.
Introduction
The desire to describe the unital Banach algebras that are generated
by two idempotents is motivated by the Gohberg-Krupnik-Sarason symbol
calculus for algebras of Toeplitz, Hankel, Cauchy, and Carleman operators
on Lp- spaces (for a summary of this work, see [1]-[3]). The Calkin and
Allan-Douglas images of these algebras can be described with the aid of
extensions of various theorems on two projections. In these applications,
symbol maps are families of continuous representations and so the con-
struction of the symbol for the elements of the closure of a Banach algebra
generated by two projectors is not explicit but 'pointwise' as limit in the
representations in the induced factor norms. In this paper, we construct in
terms of the Gel'fand map a matrix symbol with continuous determinant
for all elements of the algebra. This symbol is continuous for the elements
of a dense subalgebra and continuous with possible poles of order two on
the closure of the algebra. For a C·-algebra generated by two orthogonal
projections, by using such properties as the positivity of certain elements,
this construction can be made into an imbedding of this algebra into the
algebra of all continuous 2 x 2 matrix-valued functions.
In order to construct the matricial analogue of the Gel 'fand map for
the unital Banach algebra generated by two idempotents p and q, we first
consider the free algebra U with identity generated by two elements p and
q with p2 = P and q2 = q. Using properties of U, we then construct a
set of representations {c/>T }TE17 of the algebra A = (p, q) of polynomials
in p and q. This algebra has the properties that it is the holomorphic
On Banach algebras 91
image of U and that it is dense in A = alg (p, q). The representations are
shown to be bounded and so extendable to the whole algebra A. The set of
representations is then parametrized in terms of G, the spectrum of (p_q)2,
which belongs to the center of A.
The spectrum G can then be endowed with the weak topology relative
to the space of maximal ideals of the subalgebra of A generated by (p -
q)2. Then for every a E A, the matrix-valued function ¢T(a)d:;JaCT) is
continuous at all of the points corresponding to irreducible representations,
and may have poles at two points, corresponding to the reducible (that is,
one dimensional) representations. The map A 1-+ A(T), T E G, is our
analogue of the Gel 'fand map.
The structure of an algebra generated by two idempotents
For the reader's convenience, we repeat some details found in [5].
Theorem 1. Let A be a unital algebra generated by two idempotents over

the field K of characteristic other than two. If dimK A = 00, then A is
isomorphic to the algebra of matrices of the form
(1.1 )
where Zi E K[t] and v 2 = 1 - t. Under this isomorphism,
p 1-+ ( 1 +v
1 and q 1-+ ( I+V
-1
-t )
1- v .
Corollary 1.1. For A as in Theorem 1, the center of A is isomorphic to

K[t].
Corollary 1.2. The algebra A is a PI-algebra of order 3 with
[la, W,e] = 0
for all a, b, e E A.
Proof of Theorem 1. Let U be the free algebra with identity e generated

by p and q, p2 = p, q2 = q, over the field K. It is not difficult to see that
t = (p - q)2 belongs to the center of U and generates, with the identity e,
a subalgebra Z isomorphic to the algebra of polynomials in t, K[t].
92 Y. Weiss
We show that U is a free module of rank 4 over Z. We do this by

finding elements el, e2, ea, and e4 of U such that every element a E U has
a unique representation of the form
4
a = LZiE:i, Zi E Z. (1.2)
i=1
Let
e2 =p-q
(1.3)
ea = p+q-e
e4 = [P, q) = pq - qp.
Then the identities
e22 -
- t, e~=I-t (1.4)
hold, and
(1.5)
It follows then that (1.2) is valid.

The representation is next shown to be unique. In order to do this,
we define maps 7ri : U 1--+ Z by
(1.6)
where a = L Ziei, Zi E Z. Then it can be by using the relations given in

(1.4) that
7ri(a) = t(I - t)Zi(t), Zi(t) E K[t). (1.7)
Now suppose that a = O. Then by (1.6) and (1.7), we have zi(t)t(I-t) = O.
But since char K =I 2, the algebra K[t) admits no zero devisors, which
implies Zi(t) = 0 for i = 1, ... ,4. Thus U is uniquely represented by means
of (1.2) and (1.3) as a free module of rank 4 over Z.
In order to represent an element a E U as a 2 x 2 matrix, we endow
U with the structure of a free right module of rank two over the unital
subalgebra W generated by ea. Then the left regular representation of U
can be written in terms of 2 x 2 matrices over W.
Observe that by (1.4)

4
a = LZiei
i=l
= el(Zl + Z3 e 3) + e2(Z2 + Z4 e 3)
= elw1 + e2W2,
where Wl = Zl + Z3e3 and W2 = Z2 + Z4e3 are in W. Set W1 = Zl - Z3e3

and W2 = Z2 - Z4e3. Finally define La : U -4 U by
In other words,
(1.8)
is our representation of U.
We note from (1.8) and (1.5) that the generators p and q of the free
algebra U correspond to the matrices
Now assume A is an algebra fulfilling the conditions of the theorem;

that is, it is a unital algebra generated by two idempotents over a field K
of characteristic other than two and dimK A = 00. Let "( : U -4 A be
an epimorphism which sends p and q to the generating idempotents of the
algebra A. If a = L Ziei is a nonzero element of the kernel of ,,(, then from
the above it follows that at least one of the elements Zit( e - t) E ker "y
is nonzero. Since the quotient algebra of Kltl over nontrivial ideals is
finite dimensional over K and the free algebra is of dimension four over
KIt], the algebra A must be be finite dimensional over Kltl as well. So
since it is assumed that A is infinite dimensional over K, the map "( is an
isomorphism, and the theorem is proved. •
The matricial Gel'fand map
From here on we assume that K = C is the field of complex numbers

and that A = alg (p, q) is a semisimple unital Banach algebra generated by
two idempotents p and q. Bya we denote the spectrum of t = (p - q)2 in
the algebra Z = alg (t) generated by t.
Theorem 2. If the points 0 and 1 are not in the a or are isolated points
of a, then the algebra A is isomorphic to a subalgebra of the algebra of
94 y. Weiss
continuous 2 X 2 matrix-valued functions on a. The algebra A is isomorphic

to a subalgebra of continuous 2 x 2 matrix-valued functions on a with poles
of order two at 0 or 1 in case either of these points is a limit point of a.
In any case, the trace and determinant of the matrix-valued function are
continuous on a and give rise to continuous mappings from the algebra A
into the algebra of all continuous functions.
Proof. Since the algebra A = alg (p, q) is aPI-algebra with identity of

order 3, it follows from Corollary 1.2 and the continuity of the identity
element that the irreducible representations of A are either one or two
dimensional. We will show that the two-dimensional irreducible represen-
tations correspond to points of a \ {O, I}.
First consider the situation where 0 is an isolated point of a (the case
where 1 is such a point is handled in an identical fashion). Let RadA
be the intersection of the kernels of all of the irreducible representations s
of the algebra A, and let I be the intersection of all such representations
where s(t) = O. In this case the identity element in the factor algebra
Z / Rad Z may be represented as the sum of two projections corresponding
to the components {OJ and a \ {OJ. These are defined in terms of integrals
around the two disjoint pieces of a; that is, one about the point {OJ and
the other about a \ {OJ. From this it is seen that A/RadA is the direct
sum of the commutative finite dimensional algebra A/I and the Banach
subalgebra I, which has only two-dimensional irreducible representations.
We next construct the set of two dimensional representations of A and
put them in one-to-one correspondence with the points of a\ {O, I}. Denote
by A the possibly nonclosed subalgera of A generated by Z = alg ((p_q)2),
p, and q. Then A is a Z-module with generating elements ei (see equation
(1.3)) and is dense in A. The relations given in (1.6) can be used to define
bounded operators Pi : A 1--+ Z -+ C(a), where the first part of the map is
given by 7ri from (1.6), and the second part is the usual Gel'fand map. For
this reason, the maps Pi may be continuously extended to Pi : A 1--+ C(a).
Let
(2.0)
and
(2.1)
where TEa \ {O, 1} and u = ±J1="T. Then M: and M; are equivalent

irreducible representations, which we henceforth denote by Mr. By varying
TEa \ {O, I}, nonequivalent representations are found, depending contin-
uously on the parameter. In this way we get all classes of nonequivalent
two-dimensional irreducible representations of the algebra A. Note that
we obtain the representation (2.1) by using the left regular representation

of A. It is not difficult to see that maximal left ideals are generated by
(v"f="T - e3) = (v"f="T + l)e - p - q, (2.2)
and that elements of the form
ael + (3e2 = ae + (3(p - q) (2.3)
are a basis in all factor spaces formed with these ideals. The matrix-valued
map (2.1), the elements of which were constructed with the help of the
Gel 'fand transform of the elements of commutative algebra, is now seen to
give an imbedding of A into the algebra of continuous 2 x 2 matrix-valued
functions, possibly with poles of order two at 0 or 1 (in the next section we
will give an example of an algebra with an imbedding constructed as above
which actually has such poles). Obviously, if 0 or 1 do not belong to or are
isolated points in a, then the matrix elements of the two dimensional rep-
resentations are continuous functions, and the map (2.1) defines for every
a E A a continuous matrix valued function on a \ {O, I}. The definition of
the one-dimensional representations depend on how many there are. There
may be two one-dimensional representations corresponding to the point
o and the eigenvalues {O,O} and {I, I} of the generating elements{p, q},
as well as two one-dimensional representations corresponding to the point
1 and the eigenvalues {l,O} and{O, I}. By using the equalities (1.6) and
(1.7), we get
C; E C, i = 1, ... ,4, (2.4)
and hence the continuity of the map (2.1). Recall that the trace is defined
by
(2.5)
By definition, this is continuous as a map from A to the continuous func-
tions on a \ {O, I}. We now show that the image of trT a is continuous on
all of a for any a E A. Indeed, if s is an eigenvalue of MT(a), then se - a
is not invertible in
(3.1)
where trTa and detTa are continuous functions defined by (2.6) and (2.7).
The proof of this result follows from the local principle of Allan-Douglas
(see [2], p.31), and the form of the functions involved. In fact, let a E A,
then a point sEC belongs to the spectrum of a if and only if there is a
point TEa with
det(MT(a) - s * e) = O.
96 Y. Weiss
The proof of the last equality may be found in [5]. Using this equality the
modulo maximal point of the spectrum of the element a can be computed
by formula (3.1).We also have the following corollaries.
Corollary 4. The spectrum of the element (p - q)2, which is in the center

of A = alg (p, q), is a polynomially convex subset of the complex plane, and
coincides with the spectrum of A.
Corollary 5. If (p - q)2 is not quasi nilpotent, and if 0 and 1 are not iso-
lated points of the spectrum of (p - q)2, then A = alg (p, q) is semisimple.
Corollary 6. Let a = Et=l Ziei belong to the submodule A of A. If a is

invertible in A, then a is invertible in A and
where ~ = z~ - tz~ - (e - t)z§ - t(e - t)zl E Z.

As we saw, the matrix-valued map (2.1) gives an imbedding of A into
the algebra of cocontinuous 2 x 2 matrix-valued functions, possibly with
poles of order two at 0 or 1. We now give an example of an algebra where
this map has such poles. Let A=alg(p,q) the C* -algebra generated by
(~ ~) 1- r Vr( lr - r)) .
p= q = ( Vr(l- r)
Then it is not hard to see that the element
a = (-JT
VT=T VT=T)
JT
is in A and its symbol
is continuous with a pole at O.

For an operator theoretic point of view to some of the results in this
paper, and for corresponding applications, the reader is referred to [4].
References
[1] A. Boettcher, S. Roch, B. Silbermann, and I.M.

Spitkovskil, A Gohbery-Krupnik-Sarason symbol calculus for algebras
oJ Toeplitz, Hankel, Cauchy, and Carleman opemtors, 1990, Birkhauser,

Basel-Boston, Mass., Series: Oper. Theory: Adv.Appl., 48, 189-234.
[2] A. Boettcher and B. Silbermann, Analysis of Toeplitz Opemtors,
Akademie-Verlag, Berlin and Springer-Verlag, Heidelberg, 1989.
[3] N.Ya. Krupnik, Banach Algebms with a Symbol and Singular Integml
Opemtors, Birkhauser, Basel-Boston, Mass., 1987.
[4] S. Roch and B. Silbermann, Algebms genemted by idempotents and
the symbol calculus for singular integml opemtors, Integral Equations
Operator Theory 11 (1988), 385- 419.
[5] Y. Weiss, Algebms that are genemted by two idempotents, Seminar
Analysis of the Karl-Weierstrass-Inst. (Berlin,1987/
1988), 1988, Akademie-Verlag, Berlin, 139-145. (in Russian)
Section III
C* ALGEBRAS
Berezin-Toeplitz Quantization *
L. A. Coburn
A version of Toeplitz operators which "interpolates" between classical

Toeplitz operators on the circle and pseudo-differential operators on R n
was introduced in a series of papers by F. A. Berezin [Bl. B2 , B3]. Subse-
quently, C. A. Berger and I undertook a detailed analytic study of Berezin's
operators in order to find an analog of the classical symbol calculus of
pseudo-differential operators. In a series of papers [BC l , BC2 , BC3], we
dissected the largest class of bounded "symbols" for which Berezin-Toeplitz
operators on en have a "good" symbol calculus and commute modulo the
compact operators. Then joined by K. H. Zhu and D. Bekolle, we set-
tled the corresponding problem for Berezin-Toeplitz operators on bounded
symmetric domains n in en [BCZl. BCZ2 , BBCZ]. Roughly speaking, to
admit a good Berezin-Toeplitz symbol calculus, the symbols must satisfy
a "vanishing mean oscillation" condition. On en, exp(i vlIzl) is good while
exp(ilzl) is not good.
More recently, Berger and I have [BC 4 ] obtained sharp estimates for
the norms of Berezin-Toeplitz operators on en at roughly the level of the
Calderon-Vaillancourt Theorem for pseudo-differential operators in Rn. I
will say a little more about this later,
Since pseudodifferential operators provide the celebrated Weyl quan-
tization of R2n, it is natural to think of Berezin-Toeplitz operators in this
connection, as well. The appropriate asymptotic analysis was first carried
out for Berezin-Toeplitz operators on the unit disc lI)) (in the complex plane
C) by S. Klimek and A. Lesniewski [KL]. Earlier, A. Sheu [S] used some of
the [BC3] results to quantize the 2-sphere.
Using [KL] methodology, I was recently able to construct a reasonable
Berezin-Toeplitz quantization of en [C]. In turn, D. Borthwick, Lesniewski
and H. Upmeir [BLU] and Borthwick, Klimek, Lesniewski and M. Rinaldi
[BKLR] have very recently extended the results of [KL] and [C] to provide
a Berezin-Toeplitz quantization for "generic" bounded symmetric domains
in en and a "super" quantization for lI)) and en ("super" = Z2 graded =
bosons and fermions).
Ideally, a "quantization" should be a rule which assigns to some "rea-
sonably large" algebra of smooth function f, g on "phase space" n in en
operators AI> Ag (not necessarily bounded) on some Hilbert space so that
* Research supported by grants of the NSF.

102 L. A. Coburn
the map f --+ Af is linear, positive, 1-1 and
(*)
where [A, B] = AB - BA, L .} is the Poisson bracket in the appropriate

metric on n, h is "Planck's constant" (a very small positive scalar) and "i"
is i. Unfortunately, various "no go" theorems show that is is very hard to
do - probably impossible. As a result, weakened version of (*) have been
proposed.
A natural "first order quantum deformation" notion of [B2]' [R], [KL]
replaces (*) by
(**)
where, now, h is a (variable) positive real number and the operators A}h)
act on Hilbert spaces H(h)' [B2]' [RJ, [KL] also require norm continuity of
the map h --+ A}h), or at least that
(***)
Most of the usual versions of quantization require, in addition to (**),

(***), that
with f Xh 9 again in the appropriate algebra of smooth functions on n.

The requirement (0 is not suited to a quantization where the A}h) are
Toeplitz operators. In particular, for n = II)) and H(l) = H2(1I)), dA/,rr) the
usual Bergman space with P the Bergman projection from L2(1I)), dA/,rr)
onto H(l) and A?) = Tf where
for k in H(l), we have

(tt)
so that (t) fails! The point of [KLJ, [e], [BL U] is that, if we require only (**)
and (***) then the Toeplitz operators provide a reasonably general quanti-
zation scheme.
I will give a detailed description of the Berezin-Toeplitz quantization
of C n , as carried out by [e]. This quantization can be constructed for "sym-
bols" f,g in TP(C n) + c~(cn) where TP is the algebra of trigonometric
polynomials on C n and c~(cn) is the usual algebra of compactly sup-
ported COO-functions. Note that T P(C n ) consists of linear combinations
of characters
Xa(z) = eiIMz.a
Berezin-Toeplitz quantization 103
for a, z in en and z . a = z1iit + ... + zniin. This "Berezin-Toeplitz"

quantization is of interest because the C"-algebra generated by {Tt) :
I E TP + C~} can be precisely identified and is non-trivial, i. e. =F
/C, the compact operators. By comparison, the Weyl pseudo-differential
quantization gives /C as a quantization of C~(en) but its extensions to
larger symbol classes do not admit a very precise description.
For n = en, we consider the family of Gaussian probability measures
Here, Izl2 == z . z. The space of entire dJLr-square integrable functions is

denoted by
Hr = H2(dJLr) == H 2(e n ,dJLr).
For 9 in L2(dJLr), the Berezin-Toeplitz operator TJr) is defined by
for k in a linear subspace of H2(dJLr) (say, polynomials). We think of r as

h- 1 •
Theorem A. For 9 bounded and uniformly continuous on en,
Theorem B. For I,g in TP+c~n+6, r> 0,
IIT(r)T(r)
f 9
- T(r)
fg + !T(r) II < C(I ,g)-2
r ~(8jf)(8jg) (r) - r
holds for C(J,g) independent ofr.

For
n
{J,g} = i~)(8;J)(8;g) - (8;J)(8;g)]
;=1
we now have
Corollary. IIT;r),TJr)]- ~TI;~g}ll(r) ~ 2C(J,g)r- 2 •

The proofs of these results are not particularly instructive. It is more
interesting to describe the C* -algebra generated by
104 L. A. Coburn
Direct calculation shows that
T(r)T(r) = exp{b . a/4r}T(r) .

Xu Xb Xa+b
It follows that
is unitary with
w(r)w:(r)
a b
= exp{iImb. a/4r}W(r)
a+b·
Thus, the C* -algebra generated by {Tt) : a E Cn } is just the canonical

commutation relation algebra CCRr(C n ) with respect to the symplectic
form
a(a,b)=Ima·b/2r.
It is easy to check that the CCRr(C n ) for fixed n are all *-isomorphic via
spatial dilations. It is also easy to check (!) that the C* -algebra generated
by
is just Kr , the full algebra of compact operators on Hr. It follows that

C*{TY) : 1 E TP + Cgo} is
and these algebras are *-isomorphic for all r > O. Note that CCRr(C n ) is
so simple that we have the closed ideal lattice
CCRr(C n ) + Kr
I
Kr
I
(0)
The sup norm closure ofTP+Cgo is just the algebra AP+Co generated
by the almost-periodic functions and the continuous functions which vanish
at infinity. Thus, in roughly the sense of A. Connes, CCRr + Kr is non-
commutative AP + Co.
I provide next a very brief discussion of the [BC4 ] norm estimates.
Clearly, IITY)II(r) ~ 11/1100. Fixing r = 1/2 for convenience with T, =
r?/2), IIT,II(1/2) = IIT,II and
pt)(a) = (4rrt)-n r
len
l(w)e-la-wI2/4tdvw,
there are constants C(t) so that
C(t)IITfll ~ IIj<t)lloo t> 1/4

C(t)llj<t)lloo ~ IITfl1 t < 1/4.
These estimates are "sharp." An interesting collateral result is that, lor I
in cm(c n ) with
Tf is bounded iff I is bounded. Buried in this is the following calculus

challenge problem: Suppose I is in cm(JRn) n V'(JRn), then
For I positive (or real and semi-bounded), it is not hard to check that
j<t) is bounded if and only if j<s) is bounded for arbitrary s, t > O. It
follows for such I that Tf is bounded iff j(1/2) is bounded. This is an
earlier result of [BC3 ].
That j(1/2) should playa special role is not surprising since, for r = ~,
the normalized reproducing kernel functions are just
ka(w) = exp(w· a/2 -lal 2 /4)

and
j(1/2)(a) = (Tfka,k a) == j(a)
is the Berezin translorm of Tf. In general, the Berezin transform of an
operator A is given by
A(a) = (Aka, ka)
and A -+ A is 1-1. Clearly, IIAlloo ~ IIAII by the Cauchy-Schwarz inequal-
ity.
The special role of j(1/4) is a longer story. It is easy to provide ex-
amples of I for which j<t) is not bounded while j<t+E) is bounded. The
family
f>.(w) = exp(AlwI 2 )
provides the necessary examples.
The representation of the CC R algebra on the Segal-Bargmann space
H 1/ 2 = H2(C n , dJ.Ll/2) makes available some useful analytic machinery. As
an example, we have
Proposition. For A in CCR 1/ 2 (C n), the Berezin transform A is almost

periodic and the formula
tr(A) = f Ad{3 (ttt)

len B
106 L. Ao Coburn
defines a faithful finite trace on CCR1/2(Cn ) where d/3 is normalized Haar

measure on the Bohr compactifica.tion C B of cn.
Proof. We first note that, for 1 almost-periodic on C n , 1 extends to be

continuous on CB and
1 c~
.
Id/3= Lim (2T)2n
T-+oo
1 jT ... jT I(Zl, ... ,zn)dxldYl ... dYn
-T-T
for Zj = Xj + iYj. Next, we check that (ttt) defines a trace on {Tf : 1

trigonometric polynomial}. Using the discussion in [BC3 ], it is easy to
check that
It follows, for 1= L caXa, 9 = L dbXb that

a b
where
1#9 = L Cadbeboa/2Xa+b.
a,b
Hence,
(f#9f = L Cadbeboa/2-la+bI3 /2 Xa+ b •
a,b
It is easy to check that
1cnB
Xc d/3 - -{O
1
c#o
C = O.
It follows that
and, moreover, that
1ca 1d/3 = 1CS

Id/3 = co·
Now, for A in CCR(Cn ), we consider the Berezin transform (of oper-

ators) (see [BC3 ], [BlD
Since Tfk - A uniformly for some sequence of trigonometric polynomials

{/k}, it is clear that
ik = Tfk - A
uniformly so that A is a smooth almost-periodic function. We can now
define
tr{A) = [ A d{3.
lenB
Note that
Itr{A)1 ~ IIAlloo ~ IIAII
and that tr is a positive linear function on CCR{C n ) with tr{J) = l.
Letting T9k - B with gk trigonometric polynomials, we see that
{/k#gkf - {ABf
{gk#/kf - (BAf
uniformly so that
Hence tr{AB) = tr{BA).

Finally, suppose A ~ 0 and tr{A) = O. Then A ~ 0 and
[ Ad{3 = 0
len B
so that A == O. It follows from a well-known property of the Berezin

transform [Bl, BC3 ] that A = O.
References
[Bd F. A. Berezin, Covariant and contravariant symbols of operators, Math.

USSR. Jzv., 6 (1972), 1117-115l.
[B 2 ] F. A. Berezin, Quantization, Math. USSR. Jzv., 8 (1974), 1109-1163.
[B3 ] F. A. Berezin, Quantization in complex symmetric spaces, Math. USSR.
Jzv., 9 (1975), 341-379.
[BBCZ] D. Bekolle, C. A. Berger, L. A. Coburn, K. H. Zhu, BMO in the
Bergman metric on bounded symmetric domains, J. Funct. Anal.,
93(1990), 310-350.
[BC l ] C. A. Berger and L. A. Coburn, A symbol calculus for Toeplitz oper-
ators, Proc. Nat. Acad. Sci. USA, 83 (1986), 3072-3073.
[BC2 ] C. A. Berger and L. A. Coburn, Toeplitz operators and quantum me-
chanics, J. Funct. Anal., 68 (1986),273-299.
108 L. A. Coburn
[BC3] C. A. Berger and L. A. Coburn, Toeplitz operators on the Segal-

Bargmann space, 7rans. Amer. Math. Soc., 301 (1987), 813-829.
[BC4 ] C. A. Berger and L. A. Coburn, Heat flow and Berezin-Toeplitz esti-
mates, to appear in Amer. J. Math.
[BCZ1] C. A. Berger, L. A. Coburn, and K. H. Zhu, BMO on the Bergman
spaces of the classical domains, Bull. Amer. Math. Soc., 17 (1987),
133-136.
[BCZ2 ] C. A. Berger, L. A. Coburn, and K. H. Zhu, FUnction theory on Cartan
domains and the Berezin-Toeplitz symbol calculus, Amer. J. Math.,
110 (1988), 921-953.
[BKLR] D. Borthwick, S. Klimek, A. Lesniewski, and M. Rinaldi, Super Toeplitz
operators and non-perturbative deformation quantization of super-
manifolds, Commun. Math. Phys., 153 (1993), 49-76.
[BLU] D. Borthwick, A. Lesniewski, and H. Upmeier, Non-perturbative defor-
mation quantization of Cartan domains, J. Funct. Anal., 113 (1993),
153-176.
[C] L. A. Coburn, Deformation estimates for the Berezin-Toeplitz quan-
tization, Commun. Math. Phys., 149 (1992) 415-424.
[KL] S. Klimek and A. Lesniewski, Quantum Riemann Surfaces I, the unit
disc, Commun. Math. Phys., 146 (1992), 103-122.
[R] M. A. Rieffel, Deformation quantization for actions of Rd, preprint.
[S] A. J. Sheu, Quantization of the Poisson SU(2) and its Poisson ho-
mogeneous space, the 2-sphere, Commun. Math. Phys., 135 (1991),
217-232.
SUNY at Buffalo
Buffalo, NY 14214 USA
Normal Elements ofa Simple C*-Algebra
George A. Elliott
Abstract. Evidence is assembled supporting the conjecture that a normal element of a

simple C· -algebra is determined up to approximate unitary equivalence by the following
elementary invariants: the spectrum of the element, the measure on its spectrum arising
from each trace (or quasi trace) on the algebra, the Ko-class of the spectral projection
associated to each compact component of the spectrum together with the information
whether the sum of these projections is the unit of the C·-algebra (if there is one), and,
finally, the Kl-class of the resolvent of the element over each bounded component of the
complement of the spectrum. (In the case of a self-adjoint element the last invariant is
trivial, and in the case of a unitary element it is the Kl-class of the unitary itself.)
1. The purpose of this note is to draw attention to the possibility that

embeddings of a separable amenable C·-algebra in an arbitrary C·-algebra
may be determined, up to approximate unitary equivalence, by quite simple
invariants.
Since invariants for an embedding naturally must take into account
the ideal structure of the two C· -algebras, the first case to consider is that in
which at least one of the two C·-algebras is simple. (The opposite extreme
case, that in which both C· -algebras are commutative, is trivial: two maps
of any C·-algebra into a commutative one are equal if, and only if, they
are equal when composed with each character.)
While such a classification of embeddings would indeed be remarkable,
it must be noted that the evidence for it is already substantial.
The theory of extensions of Brown, Douglas, and Fillmore and of
Kasparov ([8], [9], [21], and [22]), and its development over the last twenty
years, can be interpreted as providing a special case of such a result.
In the case considered by Brown, Douglas, and Fillmore (at least ini-
tially), the larger algebra was the Calkin algebra and the subalgebra was
commutative. In the case, for instance, that the commutative algebra is
singly generated, they showed that a complete invariant for unitary equiva-
lence (not just approximate unitary equivalence) is given by the spectrum,
together with the K1-class relative to the Calkin algebra (i.e., the index)
associated to each hole in the spectrum. Of course, one must also keep
track of whether the embedding is unital-if the commutative algebra has
a unit.
More generally, for any unital separable amenable C·-algebra A and
any stable C·-algebra B with countable approximate unit, the unitary
equivalence classes of unital embeddings of A into the quotient multiplier
C·-algebra M(B)/B which are what is called absorbing (this property is
=
automatic when B IC, and quite possibly also whenever M(B)/ B is sim-
ple) are in natural bijection with the Kasparov group K Ko(A, M(B)/ B).
110 G.A. Elliott
In particular, an embedding which is the (Brown-Douglas-Fillmore) sum of
an arbitrary embedding and an embedding of the form
A '-+ M(K) '-+ M(B ® K) ~ M(B) -+ M(B)/B
is absorbing (in fact, every absorbing embedding is such a sum), and is

therefore determined by its K K-class-up to unitary equivalence.
It should perhaps be noted that while the K K -class of a homomor-
phism is invariant under unitary equivalence (since, after stabilization, uni-
tarily equivalent maps are homotopic), it is presumably not in general in-
variant under approximate unitary equivalence. Thus, the present article is
just concerned with two homomorphisms which are known to have the same
KK-class. Any result stating that such homomorphisms are approximately
unitarily equivalent is likely to be useful in the classification of amenable
C·-algebras. (Cf. [12], [14-17], [28], [29].) It is conceivable that two ho-
momorphisms with the same K K-class which are approximately unitarily
equivalent are also homotopic-and even by a continuous family of unit aries
indexed by the half-open interval [0,1[.
A striking recent result, similar in form to that stated above, but with
the larger algebra almost unrestricted, and the subalgebra simple, is that of
R~rdam in [28] (based on a deep result on non-commutative Rokhlin towers
obtained in [5]), which states that a unital embedding of the Cuntz algebra
02n, in a C·-algebra with the property that any unitary with trivial K 1-
class can be connected to 1 by a path of fixed length, is determined up to
approximate unitary equivalence by its K K -class.
Note that, by the universal coefficient theorem ([6], [27]), which holds
at least when A belongs to a large class of separable amenable C·-algebras
(including the Cuntz algebras), the K K-class of a map from A to a unital
C·-algebra B is determined by data involving only K.A and K.B. (In the
case of a Cuntz algebra one can see from this (Cf. Theorem 3.6 of [28]) that
the K K-class is in fact invariant under approximate unitary equivalence.)
In the two results mentioned so far, B is necessarily infinite. When
B is finite, in addition to the K K -class of a map from A to B, say assumed
to be unital, one also obtains an invariant by considering how it transforms
tracial states. As we shall now see, there are a number of cases in which
this additional invariant can be shown (by more or less known methods) to
be sufficient to characterize approximate unitary equivalence.
2. Self-adjoint Elements Let us consider the case of an embedding

of A in B when A is generated by a single self-adjoint element-say A =
C([O,I]).
If B is the Calkin algebra, the embedding is unique, up to unitary
equivalence. (Cf. above.)
Theorem. Let B be a simple C· -algebra, and let I{) and t/J be two injective
homomorphisms of A into B, where A = C([O, 1]). Suppose that B is unital,
Normal elements of a simple C*-algebra III
and that for every quasitrace T on B, the compositions Ttp and TtP are equal.
Assume that B satisfies one of the following three conditions:
(i) B is purely infinite;
(ii) B is of real rank zero and stable rank one, and the ordered group KoB
(see Theorem 3.1 of [14]) is weakly unperforated ([lJ, Definition 6.7.1;
for the non-simple case see 2.1 of [13]);
(iii) B is the inductive limit of a sequence of finite direct sums of ma-
trix algebras over commutative C* -algebras, the spectra of which have
bounded dimension (or slow dimension growth in the sense of [2]).
It follows that tp and tP are approximately unitarily equivalent (to
within an arbitrarily small tolerance).
Proof. Consider first the case (i). Note that, by [31], a purely infinite
simple C* -algebra has real rank zero. It is therefore sufficient to prove, for
each fixed integer n > 0, that two self-adjoint elements h and k of B with
spectrum the finite subset {m/n; 0 ~ m ~ n} ~ [0,1] are approximately
unitarily equivalent to within tolerance 3/n.
Since B is purely infinite (i.e., every non-zero projection is infinite),
by Theorem 1.4 of [10] the set of equivalence classes of non-zero projections
in B, with the natural addition, is a group, and the natural map into KoB
is an isomorphism. It follows in particular that any non-zero projection
in B is equivalent to part of any other non-zero projection, and also that
cancellation holds for equivalence classes of non-zero projections. Using
these two facts, it is possible to construct, step by step, as follows, a unitary
u with Ilk - uku*1I < 3/n.
First, with e(S) and f(S) denoting the spectral projections of hand
k corresponding to the set S, choose a partial isometry Uo from e( {O}) to
a subprojection of f( {O}), choose a partial isometry Vl from the comple-
ment of this projection inside f({O,I/n}) to a subprojection of e({I/n}),
choose a partial isometry U2 from the complement of this projection inside
e({I/n,2/n}) to a subprojection of f({2/n}), choose a partial isometry V3
from the complement of this projection inside f( {2/n, 3/n}), and continue
in this way as long as possible, ending with either a partial isometry be-
tween subprojections of e( {(n-I)/n, I}) and f( {I}) or of f( {(n-I)/n, I})
and e( {I}), depending on whether n is even or odd. It remains to note that
the first subprojection is a proper one, and that we may suppose that the
second subprojection is a proper one. Denote the relative complements of
these subprojections by p and q. The sum of uo, U2,··· and vi, vj,··· is a
partial isometry between I-p and I-q. Since both p and q are non-zero, by
cancellation for non-zero projections they are equivalent, and adding some
partial isometry in B taking ponto q (or q onto p) we obtain a unitary
u taking each minimal spectral projection e( {min}) of h into a subpro-
jection of the spectral projection f( {(m - I)/n, min, (m + I)/n}) of k. It
follows that Ilk - uhu*1I ~ 3/n. (For each spectral projection e({m/n}),
lI(k-uhu*)ue({m/n})u*1I ~ I/n, and if eO,el, and e2 denote the sum ofall
112 G.A. Elliott
projections e({m/n}) with m congruent to 0, 1, or 2 respectively modulo 3,

by orthogonality also lI(k - uhu*)ueiu*1I ~ lin, whence Ilk - uhu*1I ~ 3/n
as eo + el + e2 = 1.)
Consider next the case (ii). In this case (using stable rank one) we
have cancellation of equivalence classes of arbitrary projections (6.5.1 of
[1]). Furthermore, the proof of Theorem 1.4 of [11], which states that,
in an unperforated ordered group with order unit, a non-zero element is
positive if it is strictly positive on all states, is also valid if the ordered
group is just weakly unperforated. (The proof is as follows: Denoting the
order unit by u, let 9 be an element with strictly positive values on all
°
states, and consider the subgroup generated by 9 and u, with the relative
order. If there exists n < such that mg + nu ~ 0, then mg > 0, and
°
by weak unperforation 9 > 0, as desired. If there is no such n, i.e., if
mg + nu ~ implies n ~ 0, then mg + nu 1-+ n is a state of the sub
ordered group, which extends to a state on the whole ordered group by
the Goodearl-Handelman Hahn-Banach theorem (Cf. [11]), and which is
zero on g, in contradiction with the hypothesis; therefore this case does not
arise.) (The following remark concerning terminology should perhaps be
° °
made here: The definition of weak unperforation given by Blackadar in [1],
that ng > for some n > implies 9 > 0, is presumably intended only for
a simple ordered group. (It happens that the ordered group KoB in the
present case is simple.) For a general ordered abelian group, the weak form
of unperforation defined in [13] agrees with that of Blackadar in the simple
case, and perhaps should constitute the definition of weak unperforation in
the non-simple case.)
It follows from these two facts, together with the result of [3] that in a
unital C*-algebra every positive functional on (the pre-ordered group) Ko
arises from a quasitrace on the C* -algebra, that if e and I are projections
in A such that T( e) < T(f) for every quasi trace T of A, then e is equivalent
to part of I.
It is now possible to proceed in very much the same way as above.
Let h and I: be self-adjoint elements of B with spectrum [0,1], such that
T(hr) = T(l: r ) for each r = 1,2"" and for each T E QT(B), i.e., for
each quasitrace T of B with T(I) = 1. Then as B is of real rank zero we
may approximate hand k by self-adjoint elements ho and 1:0 with finite
°
spectrum. We may suppose that, for some n, the spectra of ho and 1:0 are
both equal to {min; ~ m < n} ~ [0,1].
We may suppose, furthermore, that for each m with 1 ~ m ~ n -
1, and for each T E QT(B), T(eo([O,m/n]» < T(/o([O,(m + 1)/n])) and
T(fo([O, min]» < T( eo([O, (m + 1)/n])), where eo(S) and 10(S) denote the
projection-valued spectral measures of ho and 1:0, This is not obvious, but
can be seen in the following way.
First, note that if p and q are projections of B such that p ~ 9 and
g' ~ q for some g,g' E B such that p and 9 commute, g' and q commute,
and T(g) = T(g') for all T E QT(B), then
T(p) ~ T(g) = T(g') ~ T(q), T E QT(B) .

Nonnal elements of a simple C*-algebra 113
Furthermore, as B is simple and of real rank zero, T(Z) > 0 if Z > O. (To
see this, note that by considering the commutative sub-C· -algebra of B
generated by Z we may suppose that zy = yz = y for some y > 0, and
then passing to a projection in the hereditary sub-C· -algebra of B (which
still has real rank zero by [25]) we may suppose that y is a projection. But
since the set of projections on which T is zero is closed under equivalence,
orthogonal sum, and passing to subprojections, it is the set of all projections
in a closed two-sided ideal of B. Since T(l) = 1, and B is simple, this ideal
must be zero. Therefore, T(Y) > 0, and since z and y commute and z ~ y,
T(Z) > O. We conclude from this that, with P :$ 9 :$ q as above, if 9 is
distinct from either P or q, then in fact T(p) < T(q), T E QT(B).) Hence,
if, with P :$ 9 and g' :$ q as above, either of these inequalities is strict, then
T(p) < T(q) , T E QT(B) .
Now, for each r with 0 $ r:$ n - 1 consider the continuous function

on [0,1] which is 0 on the intervaI[O, r/n], Ion the interval [(r+ l)/n, 1], and
linear in between; denote by gr the corresponding element of the C·-algebra
generated by h, and by g~ the analogous element of the C*-algebra gener-
ated by k. Then grgr+l = gr+l and g~g~+l =
g~+l' r =
0,··· ,n - 2, and
hence by [7], as B is of real rank zero, there exist projections Pl, ... ,Pn-l
and ql, . .. ,qn-l in B such that
grPr+l = Pr+l, Pr+lgr+l = gr+l ,

, = qr+l,
grqr+l "
qr+lgr+l = gr+l ,
r =0,1,··· ,n - 2. Set ~ L:~-l Pi =

ho and ~ L:~-l qi =
k o . Then
IIh-holl :$ ~ and IIk-koll :$ ~. (We have L:~-l gi :$ L:~-l Pi :$ L:~-2 gi :$
L:~-l gi = nh, whence h - ~ $ ho $ h, and similarly k - ~ $ ko $ k.)
Furthermore, by the preceding paragraph, with Pr+l, gr, g:, and qr in place
ofp,g,g', and q, we have
T(Pr+d < T(qr) , TE QT(B) , r=1,···,n-2.
Similarly,
T(qr+d < T(Pr) , T E QT(B) , r = 1, ... ,n - 2 .
Hence (Cf. case (i), above), we may construct as follows a unitary

u such that IIko - uhou·1I :$ 3/n, so that then Ilk - uhu*1I :$ 5/n. First,
choose a partial isometry Un-l from Pn-l to a subprojection of qn-2. Then,
choose a partial isometry Vn -2 from the complement of this projection
inside qn-2 to a subprojection of Pn-3 - Pn-l. Next, choose a partial
isometry Un -3 from the complement of this projection in Pn-3 - Pn-l to
a subprojection of qn-4 - qn-2. Continue in this way [using that at every
stage the target projection (which it is desired should contain the range
projection) has strictly greater value on each T E QT(B) than the initial
114 G.A. Elliott
projection], until one has a partial isometry U2 from a subprojection of

P2 - P4 to a subprojection of q1 - q3, or a partial isometry V2 from a
subproject ion of q2 - q4 to a subprojection of P1 - Pa, depending on whether
n is odd or even. The sum of Un-l + U n -3 + ... and U~_2 + U~_4 + ...
is then a partial isometry from a subproject ion of Pl, containing P2, to
a subprojection of ql, containing q2. By cancellation, this extends to a
unitary u E B, necessarily taking Pr into a subprojection of qr-l for each
r = 2,3, ... ,n-l, containing qr+1 when r < n-l. It follows that (as in the
case (i» u takes each minimal spectral projection of ho into a subprojection
of the sum of the corresponding minimal spectral projection of ko and the
two adjacent minimal spectral projections of ko (or single adjacent one, at
the two ends). Hence (as in the case (i», IIko - uhou*1I ~ 3/n , and so
Ilk - uhu*1I ~ Sin, as desired.
Finally, consider the case (iii). In the special case that the inductive
limit decomposition of B (assumed to be simple and unital) is with respect
to a sequence of direct sums of matrix algebras over C([O, 1]) the theorem
follows-as was pointed out by Thomsen in [30]-from the proof of the
classification theorem for such inductive limits given in [15]. Indeed, if !P
and tP are two unital maps of A.... = C([O, 1]) into such an inductive limit
= =
Co. -algebra B lim_ Bi, such that T!p TtP for every tracial state T of B,
then (since A ..... = C([O, 1])) there are approximately commutative diagrams
(in the sense of [14])
-"
! !P1
-
!P2
B1 B2 - B,
-" -
! tPl tP2
B1 B2 - B.
The hypothesis that !P and tP determine the same map from Aff T A
to Aff TB, together with the decomposition Aff TB = lim- Aff TBi ,
implies that the maps from Aff T A to Aff T Bi determined by !Pi and tPi are
arbitrarily close on any finite subset of Aff TA for sufficiently large i. This
together with the fact that, as B is simple, each fixed non-zero positive
element of A must give rise under !P and tP to an element of Aff T B which
°
is strictly positive, and hence must give rise under !Pi and tPi to an element
of Aff T Bi which is larger than some fixed 6 > for all sufficiently large
i, implies that the hypotheses of the uniqueness result (Theorem 6) of [15]
hold for !Pi and tPi for all sufficiently large i, and in such a way that, for
large i, !Pi and (Adu)tPi are close on finitely many given elements of A for
Normal elements of a simple C*-algebra 115
some unitary u in Bi. Therefore, also !P and (Adu),p are close on these
elements, as desired.
Note that, except for the quite straightforward use of simplicity of
B, the only use of the proof of the classification theorem made here is
the application of the uniqueness result for maps between basic building
blocks, Theorem 6 of [15]. Furthermore, it is clear that, to prove the present
theorem in the full generality of case (iii), in which the spectra of the C·-
algebras Bl, B2,··· are compact metric spaces of bounded dimension (or
of slow dimension growth in the sense of [2]), it is sufficient to establish a
suitable generalization of Theorem 6 of [15]. Such a generalization (with a
somewhat less sharp estimate) is as follows.
Let Bo be a finite direct sum of matrix algebras over unital com-
mutative C·-algebras, the spectra of which are compact metric spaces of
dimension at most d < 00. Let !PI and !P2 be homomorphisms of A into Bo.
Suppose that !PI(I) and !p2(1) have the same Ko-class in B o, and that the
images of !PI(I) and !P2(1) in any primitive quotient of B have both dimen-
sion and codimension at least d12. Let 6 > 0 and n = 1,2,··· be such that,
with h (= t 1-+ t) the canonical self-adjoint central generator of A, the im-
age of both !PI (h) and !P2( h) in each primitive quotient of Bo have at least
fraction 6 of their eigenvalues in each of the canonical open subintervals of
[0,1] of length lin. Suppose also that the absolute number of eigenvalues
in each such interval is at least 3d + 2. Suppose that for each of the n
continuous functions on [0,1] which are linear and increasing from 0 to 1
on the canonical subintervals of length lin, and constant elsewhere, the
corresponding function of h has approximately the same value, to strictly
within 6, on the functionals T!PI and T!P2 for each tracial state T of Bo. It
follows that !pI(h) and !p2(h) are approximately unitarily equivalent in B o,
to within 57In.
This uniqueness result is seen as follows. Exactly as in the proof of
Theorem 6 of [15], one has that the eigenvalues ofthe images of !PI (h) and
!P2(h) in each primitive quotient of B o, when considered in increasing order,
counting multiplicity, are within 3/n of each other.
By Theorem 8.1.5 of [20], the hypothesis on the dimension and codi-
mension implies that !PI(I) and !P2(1) are unitarily equivalent. We may
=
therefore pass to the case that !Pl(l) !P2(1) 1. =
By Lemma 1.5 of [26], perturbing !PI (h) and !P2(h) by small amounts,
we may suppose that any eigenvalue of the image of !PI (h) or !P2( h) in any
primitive quotient of Bo has multiplicity at most d.
Passing to a direct summand of B o, we may suppose that all primitive
quotients of Bo are isomorphic.
Numbering the eigenvalues of !PI (h) as ~I' ~2' ... in increasing order,
with multiplicity, consider the eigenvalues ~l' ~l+P' ~1+2p, ... , where p =
3d+2. The distance between any two of these is at most 2/n (since at least
3d + 2 eigenvalues are contained in each canonical open subinterval of [0,1]
oflength lin, and therefore in any subinterval oflength 2/n). Furthermore,
the number of eigenvalues strictly less than ~(Hl)p is at least (k + l)p - d,
116 G.A. Elliott
and the number of eigenvalues less than or equal to "\ip is at most kp+d-1.
Hence, as (5d + 1)/2 < p, with 1= (k + l)p - d we have that I is less than
or equal to the dimension oUhe spectral projection ~+1(t) of 'Pl(h)(t) for
the interval] - 00, ..\(Hl)p(t)[, and I is strictly greater than (d + 1)/2 plus
the dimension of the spectral projection Qi(t) of 'Pl(h)(t) for the interval
] - 00, ..\ip(t)]. It follows by the projection selection theorem of [4] that
there exists an I-dimensional projection Ri in Bo contained between Qi
and PH1. The projections Si = Ri - Ri-l (where no = 0) are then
orthogonal, and enlarging the last Si by the complement of the sum of all
of them, we have
E
lI'Pl(h) -
i
"\ip Sill < 4/n .
Similarly, also 'P2(h) is within 4/n of an orthogonal central combina-

tion of projections. Note that, in either case, the values of the coefficients
at each point are distinct. By the Weyl spectral variation inequality these
values are also 6/n-dense in [0,1].
Adjusting the projections for each of the approximating elements, as
in the proof of Theorem 2(ii) , using that by Theorem 8.1.5 of [20] can-
cellation holds for equivalence classes of projections in Bo of dimension at
least d/2 in each primitive quotient of Bo, and that by Theorems 8.1.2 and
8.1.5 of [20] (Cf. Theorem 6.10.3 of [1]) any projection in Bo is equivalent
to part of any other projection the dimension of which in every primitive
quotient of Bo is greater than that of the first by at least d,-this involves
a perturbation of each element by at most 18/n-, we may suppose that
corresponding projections for the two elements (counting from the bottom)
are equivalent.
It follows that (on conjugating 'Pl(h) by a unitary) we may pass to
the case that 'Pl(h) and 'P2(h) are both approximated, to within 18/n, by
central combinations of the same orthogonal family of projections in Bo,
with the (self-adjoint) central coefficients in each case having distinct val-
ues at each point of the spectrum of Bo. By the Weyl spectral variation
inequality, the eigenvalues of these approximating elements, labelled in in-
creasing order (counting multiplicity), are then within 18/n respectively of
the eigenvalues of 'Pl(h) or 'P2(h), in each primitive quotient of Bo. Since
these ·have been seen to be within 3/n of each other, it follows that the
eigenvalues of the approximating elements are within 21/n of each other,
so that (as the orthogonal projections in the central combinations are the
same) the elements themselves are also within 21/n of each other. It follows,
finally, that 'Pl(h) and 'P2(h), after the unitary transformation introduced
above, are within distance 18/n + 21/n + 18/n = S7/n of each other, as
desired.
3. Unitary Elements Let us now consider the case of an embedding

of A in B when A = G(T).
If B is the Calkin algebra, the embedding is determined up to unitary

equivalence by the Kt-class of the image in B of the canonical unitary
generator of C('ll') (which is an integer). (Cf. above.)
Theorem. Let B be a simple unital C· -algebra, and let ip and", be two

injective unital homomorphisms of A into B, where A = C('ll'). Suppose
that ip and", agree on KIA = Z. Suppose that for every trace T on B,
the compositions Tip and T'" are equal. Assume that B satisfies one of the
following three conditions:
(i) B is purely infinite;
(ii) B is of real rank zero and stable rank one, the tracial states of B
separate elements of KoB, and B is the inductive limit of a sequence
of finite direct sums of matrix algebras over commutative C· -algebras;
(iii) B is the inductive limit of a sequence of finite direct sums of matrix
algebras over C('ll').
It follows then that ip and", are approximately unitarily equivalent.
Proof. Consider first the case (i). Suppose, first, that the (common) K I -
class of u = ip(z) and v = ",(z) is zero, where z (= z 1-+ z) denotes the
canonical unitary generator of A. Then, by [23], as B has real rank zero
(Cf. proof of Theorem 2(i», both u and v can be approximated by unitary
elements with finite spectrum. The proof may then be completed as in the
case of Theorem 2(i). (Alternatively, the conclusion may be deduced from
that theorem.)
Now let the K I-class of U and v be non-zero. As B has real rank zero,
for any n = 1,2,··· u and v are arbitrarily close to direct sums UI $ U2
and VI $ V2 with SpU2 = SpV2 = {e 2... im / n jO $ m < n}. (It is enough
to consider the case n = 1. By [7], there is a projection in B separating
the spectral projections of u in B·· corresponding to a small open arc in
'll' containing 1 and an open arc disjoint from this and of length close to
211". If the length of the large arc is close enough to 211" then this projection
approximately commutes with u and its product with u has spectrum close
to 1. Hence u is close to a unitary UI $U2 with Sp U2 = {I}.) Let us choose
the approximations so that U and v are within lIn of UI $ U2 and VI $ V2.
Using the result ofCuntz that any projection in B is equivalent to part
of any non-zero projection (Cf. proof of Theorem 2(i», choose a projection
p such that the support of U2 is the sum of p and projections equivalent to
the supports of UI and VI, so that we have a unitary
Since the spectrum of ui $ Vt $ p is 'll' but this partial unitary has K t-class
equal to zero, by the first paragraph above ui $ til $ P is approximately
unitarily equivalent to U2 = u2(n) to within tolerance 2/n. Similarly, writ-
ing
118 G.A. Elliott
and noting that by cancellation for non-zero projections (Cf. proof of The-
orem 2(i)) the support of ut ED UI ED P is equivalent to the support of V2,
we may identify these supports and apply the first paragraph above to
conclude that ut ED UI ED P is approximately unitarily equivalent to V2 =
v2(n) to strictly within tolerance 2/n. In other words, w is approximately
unitarily equivalent both to UI ED U2 and to VI ED V2 to within tolerance
2/n. It follows that U is approximately equivalent to V to within tolerance
l/n + 2/n + 2/n + l/n = 6/n. Since n is arbitrary the assertion is proved.
Consider next the case (ii). Suppose, first, that the (common) K 1-
class of U = /f>(z) and v = .,p(z) is zero. Then by [23], U and v can be
approximated by unitary elements Uo and Vo with finite spectrum. We may
suppose that, for each tracial state T of B, the value of T is approximately
the same on each of finitely many continuous functions acting on Uo and on
Vo. Hence, by an argument similar to that in the proof of Theorem 2(ii),
we may suppose that, for any given !, for each T E T(B) and each open arc
S~,]["
r(eo(S)) < T(fO(Se)) ,

r(fo(S)) < T(eo(Se)) ,
where Sf denotes the arc in '][' consisting of all points in '][' within the open !-
neighborhood of Sin C, and eo and /0 denote the projection-valued spectral
measures of Uo and Vo. Since Aff T(B) = lim ..... Aff T(B;), it follows that,
on choosing uo and Vo to belong to some B;, the preceding inequalities hold
for T E T(Bj ) if j ~ i is sufficiently large. In particular specializing to an
extreme tracial state of B j , by Proposition B of [19] we conclude that Uo
and Vo are approximately unitarily equivalent, to within tolerance £, in any
primitive quotient of B. Since Uo and Vo have finite spectrum, B j has a
direct sum decomposition such that, in each direct summand, Uo and Vo
are linear combinations of projections of constant dimension over primitive
quotients. The assumption that elements of KoB are separated by tracial
states of B implies that two projections in Bj have the same Ko-class in B
if they have the same dimension in each primitive quotient of Bj (because
then they have the same value on each tracial state of Bj, and hence on each
tracial state of B- since each tracial state of B restricts to a trace of Bj).
It follows by cancellation (Cf. proof of Theorem 2(ii)) that two projections
in Bj that are equivalent in each primitive quotient of Bj are equivalent
in B. The proof of Proposition B of [19] then shows that Uo and Vo are
approximately unitarily equivalent to within £, not only in each primitive
quotient of Bj, but also in each of the direct summands considered above
(for each of them consider the eigenprojections of uo and Vo in this direct
summand, as multiples up to equivalence in B of minimal projections in
the algebra of matrices in this direct summand with scalar entries, instead
of the eigenprojections in the primitive quotients).
Now let the Kl-class of U and v be non-zero. As in the case (i), as

B has real rank zero, for any n = 1,2" .. U and v are arbitrarily close to
direct sums UI ED U2 and VI ED V2 with
Choose some unitary w with support strictly less than one half of each
spectral projection of U2 or V2 (inside the support of U2 or V2, that is) on
each tracial state of B, and with the same Kl-class as U and v (and UI and
vt). Using that a projection in B is equivalent to a subprojection of another
if it is strictly smaller on every tracial state (Cf. proof of Theorem 2(ii»,
and that w· ED w can be approximated by a unitary with finite spectrum
inside its support in B ([23]), we have that UI ED U2 is within 21r/n of
UI ED w· ED w ED u~ where Spu~ = {e2rim/n j 0 $ m < n}, and hence is within
41r/n of w ED Wi where Spw' = {e2rim/njO $ m < n}. Similarly, VI ED V2 is
within 41r/n ofwEDw" where Spw" = {e2rim/njO $ m < n}. Since finitely
many functions of UI ED U2 and VI ED V2 agree approximately on T B, this is
also true for w ED Wi and w ED wIt if n is large, and hence also for Wi and
wIt. By the case of zero K I-class considered above, Wi and wIt are close to
being unitarily equivalent if n is large. Hence so also are the pairs wED w'
and wED wIt, UI ED U2 and VI ED V2, and, finally, U and v.
Now, consider the case (iii). This follows, as we shall see, from the
proof of the classification theorem for simple unital inductive limits of direct
sums of matrix algebras over G('!r), given in [16], in much the same way
as Theorem 2(iii), in the special case that B is an inductive limit of direct
sums of matrix algebras over G([O, 1]), follows from [15] (which considers
the special case of the classification theorem with [0,1] in place of 1r). The
proof of the classification theorem in the case of circles is more intricate than
in the case of intervals, and the reduction of the present uniqueness result
to the uniqueness result for maps between basic building blocks (Theorem
4 of [16]) is not quite as immediate as in the case of Theorem 2(iii). (A
preliminary step is required.) The reduction is carried out as follows.
First, as in the proof of Theorem 1 of [16], express B as the inductive
limit of a sequence BI -+ B2 -+ " ' , with each Bi a finite direct sum
of matrix algebras over G(T), in such a way that each partial map from
a minimal direct summand of Bi to a minimal direct summand of Bi+l
has the following property: it takes the canonical unitary central generator
into a (partial) unitary which either has constant determinant (inside its
support) or, in the case that its K I-class is non-zero, is approximately
constant (in the sense of 5.2 of [14]), to within a tolerance to be specified.
As in the proof of Theorem 2(iii) , we have approximately commutative
diagrams
120 G.A. Elliott
A
1 <PI
-- --
'\. <P2
Bl B2 -- B,
--
1 t/Jl '\.
--
t/J2
Bl B2 -- B.
Inspection of the proof of Theorem 1 of [16] shows that we may per-

turb each ofthe maps <Pi, t/Ji, and Bi - Bi+l by arbitrarily small amounts to
ensure that each of the maps <Pi and t/Ji composed with the map Bi - Bi+1
has the same property as the map Bi - Bi+l, described above. Hence,
replacing <Pi and t/Ji by these composed maps, we may suppose that <Pi and
t/Ji themselves have this property. The hypothesis that <P and t/J determine
the same map from Aft' T A to Aft' T B, together with the decomposition
Aft' T B = liIlL. Aft' T Bi, implies that the maps from Aft' TA to Aft' T B
determined by <Pi and t/Ji are arbitrarily close on any finite subset of Aft'
TA for sufficiently large i. This together with the fact that, as B is simple,
each fixed non-zero positive element of A must give rise under 0 for all sufficiently large i, implies that the hypotheses of the unique-
ness result (Theorem 4) of [16] hold for <Pi and t/Ji for all sufficiently large
i, and in such a way that, for large i, <pi and (Adu)t/Ji are close on finitely
many given elements of A for some unitary u in Bi. Therefore, also <P and
(Adu)t/J are close on these elements, as desired.
4. Normal Elements Let us, finally, consider the case of an embed-

ding of A in B when A is generated by a single normal element.
If B is the Calkin algebra, the embedding is determined up to unitary
equivalence by the spectrum, together with the class in KIB (an integer)
associated to each hole in the spectrum. (Cf. above.)
Theorem. Let B be a simple unital C· -algebra, and let <P and t/J be
two injective unital homomorphisms of A into B, where A = C(X) with
X a conn~cted compact subset of C. Suppose that <P and t/J agree on KIA.
Suppose that B is purely infinite.
It follows that <P and t/J are approximately unitarily equivalent.
Proof. Suppose, first, that Kl(<p) = Kl(t/J) = O. In this case, by Theorem

4.4 of [24], <P and t/J can both be approximated by direct sums of point
Nonnal elements of a simple C*-algebra 121
evaluations, by which is just meant that <p(z) and 1/J(z) are approximible
by normal elements with finite spectra, contained in X. Furthermore, the
points of X in question must be arbitrarily close to being dense in X. It
follows that there are close approximations to <p and 1/J involving the same
finite subset of X. Adding some extra points if necessary (splitting some
points), we may number this finite set in such a way that consecutive points
are close. The proof may now be carried out in the same way as in the case
considered in Theorem 2(i) (in which the finite set is {min; 0 ~ m ~ n},
in case the number of points is n + 1).
To deal with the case of arbitrary Ki-map, note first that, expressing
C(X) as the inductive limit of a sequence C(Xt} -+ C(X2) -+ ... where
each Xi is a connected finite CW-complex of dimension at most two em-
bedded in C (Cf. Theorem 1.13.5 of [18]) we may suppose that X is already
such a space. (One may pass to inductive limits in all the uniqueness re-
sults in this paper.) The purpose of this reduction is in order to construct
another map C(X) -+ B which has the opposite action on Ki - so that
we may finish the proof as in the case of Theorem 3(i).
Let, therefore, X be as in the preceding paragraph, and let us con-
struct an injective map C(X) -+ B giving rise to an arbitrary map
p: Ki(C(X)) -+ KiB. We may start with one of the given maps, and mod-
ify it (we do not have to start from scratch). Replace the given map <p by
the map (: Z 1-+ v<p( Z )v· where v is a proper isometry in B, and note that
«1) is a proper projection. For each hole in X, note that the boundary is
homeomorphic to 1l', and add a direct summand to ( which factors through
a map from the algebra of continuous functions on this boundary, which
maps the canonical generator of Ki into the element of KiB associated by
p to this hole, minus the K i-class of B associated by ( to this hole. (Recall
that K l(C(X)) consists of a direct sum of copies of~, one for each hole in
X.) The resulting map, ( plus the direct sum of orthogonal maps, one for
each hole, chosen in this way, has the desired Ki class.
The proof may now be finished in the same way as for the analogous
Theorem 4(i) (with ( in place of the map C(1l') -+ B corresponding to ui).
Acknowledgements. This article was prepared during the tenure of

a Royal Society Guest Professorship at the University College of Swansea,
and a Canada-J apan Bilateral Scientific Exchange Fellowship at Hokkaido
University, awarded by the Natural Sciences and Engineering Research
Council of Canada and the Japan Society for the Promotion of Science.
The author is indebted to David Evans and Akitaka Kishimoto for their
hospitality, and for conversations on the subject of this article.
References
1. B. Blackadar, K-Theory for Operator Algebras, MSRI Publication

Series 5, Springer-Verlag, Heidelberg, 1986.
122 G.A. Elliott
2. B. Blackadar, A. Kumjian, and M. ~rdam, Approximately central

matrix units and the structure of noncommutative tori, K-Theory 6
(1992), 267-284.
3. B. Blackadar and M. ~rdam, Extending states on preordered semi-
groups and the existence of quasi traces on C* -algebras, J. Algebra
152 (1992), 240-248.
4. O. Bratteli and G. A. Elliott, Small eigenvalue variation and real rank
zero, preprint.
5. O. Bratteli, A. Kishimoto, M. ~rdam, and E. Stj21rmer, The crossed
product of a UHF algebra by a shift, Ergodic Theory Dynamical
Systems, to appear.
6. L. G. Brown, The universal coefficient theorem for Ext and quasi-
diagonality, Operator Algebras and Group Representations, I (edited
for INCREST by Gr. Arsene et al.), Monographs and Studies in
Math. 17, Pitman, London, 1984, pp. 60-64.
7. L. G. Brown, Interpolation by projections in C*-algebras ofreal rank
zero, J. Operator Theory 26 (1991), 383-387.
8. L. G. Brown, R. G. Douglas, and P. A. Fillmore, Extensions of C*-
algebras, operators with compact self-commutators, and K-homology,
Bull. Amer. Math. Soc. 79 (1973), 973-978.
9. L. G. Brown, R. G. Douglas, and P. A. Fillmore, Extensions of C*-
algebras and K-homology, Ann. of Math. 105 (1977), 265-324.
lO. J. Cuntz, K-theory for certain C*-algebras, Ann. of Math. 113
(1981), 181-197.
11. E. G. Effros, D. E. Handelman, and C.-L. Shen, Dimension groups and
their affine representations, Amer. J. Math. 102 (1980), 385-407.
12. G. A. Elliott, On the classification of inductive limits of sequences of
semisimple finite-dimensional algebras, J. Algebra 38 (1976), 29-44.
13. G. A. Elliott, Dimension groups with torsion, International J. Math.
1 (1990), 361-380.
14. G. A. Elliott, On the classification of C*-algebras of real rank zero,
J. Reine Angew. Math. 443, to appear.
15. G. A. Elliott, A classification of certain simple C*-algebras, Quantum
and Non-commutative Analysis (editors, H. Araki et al.), Kluwer, to
appear.
16. G. A. Elliott, A classification of certain simple C* -algebras, II, preprint.
17. G. A. Elliott and G. Gong, On the classification of C* -algebras ofreal
rank zero, II, preprint.
18. R. Engelking, Dimension Theory, North-Holland, Amsterdam, 1978.

19. R. H. Herman and A. Ocneanu, Spectral analysis for automorphisms
of UHF C·-algebras, J. Funct. Anal. 66 (1986), 1-10.
20. D. Husemoller, Fibre Bundles, McGraw-Hill, New York, 1966.
21. G. G. Kasparov, A generalized index for elliptic operators, Funkcional.
Anal. i Prilozen 7 (1973), 82-83.
22. G. G. Kasparov, The operator K-functor and extensions of C·- alge-
bras, Isv. Akad. Nauk USSR, Ser. Mat. 44 (1980), 571-636; English
translation: Math. USSR Izv. 16 (1981), 513-572.
23. H. Lin, Exponential rank of C·-algebras with real rank zero and the
Brown-Pedersen conjectures, J. Funct. Anal. 114 (1993),1-11.
24. H. Lin, Approximation by normal elements with finite spectra in C·-
algebras of real rank zero, preprint.
25. G. K. Pedersen, The linear span of projections in C·-algebras, J.
Operator Theory 4 (1980), 289-296.
26. N. C. Phillips, Reduction of exponential rank in direct limits of C·-
algebras, preprint.
27. J. Rosenberg and C. Schochet, The Kiinneth and the universal coef-
ficient theorem for Kasparov's generalized K-functor, Duke Math. J.
55 (1987), 431-474.
28. M. R(ljrdam, Classification of inductive limits of Cuntz algebras, J.
Reine Angew. Math. 440 (1993), 175-200.
29. M. R(ljrdam, Classification of Cuntz-Krieger algebras, preprint.
30. K. Thomsen, Inductive limits of interval algebras: The simple case,
Quantum and Non-commutative Analysis (editors, H. Araki et al.),
Kluwer, to appear.
31. S. Zhang, A property of purely infinite simple C·-algebras, Proc.
Amer. Math. Soc. 109 (1990), 717-720.
Department of Mathematics,
University of Toronto,
Toronto, Canada M5S 1A1
Mathematics Institute,
Universitetsparken 5,
2100 Copenhagen 0, Denmark
A GELFAND·NAIMARK THEOREM FOR C*-ALGEBRAS
ICHIRO Fu JIMOTO
Introd uction.
Since the memorial work by I.M. Gelfand and M.A. Naimark [22], there have
been various attempts to generalize the beautiful representation theorem for
non-commutative C*-algebras (see e.g. [12] and its bibliography). Among them,
one notable direction was the approach from the convexity theory, initiated by
R.V. Kadison finding a functional representation of a C*-algebra on the w*-
closure of the pure states as an order isomorphism in the context of Kadison's
function representation theorem (cf. [23]). This motivated the abstract Dirichlet
problem for the extreme boundary of compact convex sets (cf. [2,3]), and the
duality arguments on the pure states for C*-algebras by F.W. Shultz [28] in the
context of Alfsen-Shultz theory, which was further developed in C.A. Akemann
and F.W. Shultz [1] based on Takesaki's duality theorem (cf. [8], [30]). Another
important approach was initiated by J.M.G. Fell [14] using the fiber bundle
theory, and culminated in the Dauns-Hofmann theorem [11] (cf. also [13]),
where they showed that every C*-algebra is *-isomorphic to the C*-algebra of
all continuous sections of a C*-bundle over the spectrum of the center of its
multiplier algebra.
Our purpose in this paper is to present a generalized Gelfand-Naimark theo-
rem for non-commutative C*-algebras on the extreme elements of the CP-state
space in the sense of CP-convexity, or the irreducible representations of the al-
gebra, which interpolates the above mentioned approaches and recovers both of
the algebraic and affine geometric aspects of the duality. As applications, we
discuss the abstract Dirichlet problem for the CP-extreme boundary, and pro-
pose a generalized spectral theory for non-normal operators, which show that
our duality generalizes the same situation of the commutative case.
The detailed version of this paper has been submitted for publication elsewhere.
A Gelfand-Naimark theorem for C*-algebras 125
1. Preliminaries.
We first review some basic definitions and results on CP-convexity which

were introduced in [15-17]. Let A be a C*-algebra and H be a Hilbert space,
and let C P( A, B( H)) be the set of all completely positive maps from a C*-
algebra A to B(H) (see [5], [27], [29J, [31J for CP-maps). We denote by QH(A)
the unit ball of C peA, B(H)) and call it the CP-state space of A for H, i.e.,
QH(A) := {'¢ E CP(A, B(H)); II'¢II :::; I}, which generalizes the quasi-state
space Q(A) in the scalar convexity theory. Then, '¢ E QH(A) is defined to be a
CP-convex combination of ('¢"')"'EA C QH(A), if
",EA "'EA
where the sum converges in the BW-[resp. BS-] topology (Le., the pointwise
weak [resp. strongJ convergence topology), which we abbreviate in the following
by
'¢ = CP - L S:,¢",S",.
"'EA
We note that this CP-convex combination is a natural notion in the C*-algebraic
approach in quantum physics to describe interactions of physical systems (cf. e.g.
[4], [10], [25], [26]).
We denote by Rep(A) [resp. Repc(A),Irr(A)J the set of all [resp. cyclic,
irreducible] representations, and we write Rep(A : H) etc to specify the Hilbert
space H in which the representations are confined. We define the support H",
of'¢ = V*1I"V E CP(A,B(H)) by the support of'¢(A) in H, i.e., H",:= [IVIH],
and denote by p", the projection of H onto H ",. We also define the cyclic [resp.
irreducible] dimension of A by
Qc(A) [resp. Qi(A)] := sup {dim Hrr; 11" E Repc(A) [resp. Irr(A)]}.
Note that, if dimH ~ C1 c(A), then QH(A) = CP - convRepc(A: H) (cf. [17;

Prop. 1.4]).
A function I : QH(A) -+ B(H) is defined to be CP-affine, if
126 1. Fujimoto
I is called bounded if 11/11 = sup{lI!( 1/J )11; 1/J E Q H(A)} < 00. We denote by
AB(QH(A), B(H)) the set of all bounded CP-affine functions from QH(A) to
B(H), and by AC(QH(A), B(H)) the set of all BW-w (or BS-s) continuous
elements. Then the CP-duality theorem [17; Th.2.2] states that, if dimH >
Qc(A), then
A ~ AC(QH(A), B(H)) (* -isomorphism),
where the product of CP-affine functions is defined on Repc(A : H), or more

precisely, if 1/J E QH( A) can be expressed as
1/J = CP - L V;7l"o,va with 7l"a E Repc(A: H) (cL [17; Prop.1.4]),
then, for 11,/2 E AC(QH(A),B(H)),
which is shown to be well-defined (cL [17] for details). This CP-duality theo-
rem generalizes Kadison's function representation theorem with recovering the
full C*-structure. If A and Bare C*-algebras and H is a Hilbert space with
dimH ~ sup{Qc(A), Qc(B)}, then QH(A) and QH(B) are CP-affine BW (or
BS) homeomorphic if and only if A and Bare *-isomorphic ([17; Th.3.2]). This,
together with Kadison's result on state spaces with scalar convexity in [24], can
be illustrated as
CP-convexity [resp. scalar convexity] <-> C*-algebra [resp. Jordan algebra]
2. CP-extreme states.
We first note that the scalar convexity is not suitable for the extremal problem
of the CP-state space Q H(A), as it can be easily seen that every representation
7l" E Rep(A: H) is an extreme point ofQH(A), while 7l" may have a direct sum de-
composition 7l" = EB a 7l"a = 2:,,,,Pa7raP,,, where Pa := P"''' (cL [19]). This indicates
that there are many decompositions of CP-maps which cannot be described by
scalar convex combination, but can be described by CP-convex combination.
We define the extreme elements of the CP-state space in the sense of CP-
convexity as follows.
Definition 2.1. A nonzero CP-state 7/J E QH(A) is defined to be a CP-extreme

state if7/J = CP- L-aS;7/JaSa with 7/Ja E QH(A) implies that 7/Ja is unitarily
equivalent to 7/J, i.e., 7/Ja= Ua7/JU~, and Sa = caUa (ca E q with a partial
isometry Ua such that U~Ua = P1/J and UaU~ = P1/Ja·
We denote the set of all CP-extreme states of A for H by DH(A). From

the above definition, it can be easily seen that every CP-extreme state 7/J E
DH(A) is in particular an extreme point of QH(A) in scalar convexity, and also
7/J E DH(A) is a pure CP-map, because 7/J cannot be decomposed into a sum,
hence a CP-convex sum, of elements in QH(A) without each summand being a
scalar multiple of 7/J. We shall denote by PH(A) the set of all pure elements in
CP(A,B(H)), while we write peA) for the pure states as usual. (Recall that
7/J = V*1rV E PH (A) if and only if 1r E Irr(A). cf. [5].) Then, the CP-extreme
states can be characterized as follows.
Theorem 2.2. Let A be a C*-algebra and H be a Hilbert space. Then,
DH(A) ={U*1rU E PH(A); u*u = IH with dimH < dimH" , No}
U {u*1rU E PH(A);uu* = IHT with dimH;::: dimH,,}.
This theorem shows that the CP-extreme boundary DH(A) consists of pure
approximately unital CP-maps (u'u = IH with dimH < dimH",No) and ir-
reducible representations of A on H (uu* = IHT with dimH ;:::: dimH,,). The
following result shows that our CP-convexity indeed interpolates the gap be-
tween algebraic theory and convexity theory.
Corollary 2.3. Let A be a C*-algebra and H be a Hilbert space.

(i) If dimH ;:::: inf{ai(A), No}, then DH(A) = Irr(A: H).
(ii) If dimH = 1, then DH(A) = peA).
Remarks. 1. There are some other different ways to define CP-extreme states
D H( A) (cf. [18, 20]). Among them, our definition in this note has a convenient
property characterized in Corollary 2.3, which is useful for our purpose for the
CP-Gelfand-Naimark theorem.
2. We note that Choquet's boundary measure representation theorem for
the state space can be generalized for the CP-state space in the context of
128 I. Fujimoto
CP-convexity if A and H are separable, Le., every CP-state in QH(A) can be

represented by a normal QH(B(H)) valued measure, which can be considered
heuristically as a measure generalization of the CP-coefficients "S;(· )Sa", sup-
ported by the CP-extreme boundary DH(A) (cf. [19]).
3. CP-Gelfand-Naimark theorem.
Definition 3.1. Let A be a C*-algebra and H be a Hilbert space, and X C

Rep(A: H) be an invariant subset under the operation of unitary equivalence.
Then a function / : X - B(H) is called equivalent if /(u*7ru) = u*/(7r)u for
7r E X and for all partial isometry u such that uu* ~ p",.
We denote by AE(X, B(H)) the set of all bounded equivalent functions from
X to B(H), and by A~(X, B{H)) the set of all c-continuous elements, where c
denotes the weak (c = w), strong (c = s) or strong* (c = SO) topology in B(H),
and note that the pointwise weak, strong and strong* convergence topologies
coincide on Rep(A : H) (cf. [8], [30]). In particular, we write A~(X, B(H)) for
the set of all uniformly c-continuous elements.
The following theorem was first proved by Akemann-Shultz
,.
[1] for larger
Hilbert space Hand sO-continuous functions, which can be improved using L.G.
Brown's recent result on Shultz's theorem in [9] as follows.
Theorem 3.2. Let A be a C*-algebraand H be an infinite dimensional Hilbert

space with dimH ~ oc{A). Then,
(*-isomorphism),
where Irr(A : H)c denotes the BW-closure of Irr{A: H) in Repc{A : H).
We note here that, since dimH ~ sup{oc{A), No} ~ inf{o;(A), No}, we have
DH(A) = Irr(A: H) by Corollary 2.3. (If A is separable, we can take an infinite
dimensional separable Hilbert space H, i.e., dimH = No.) We can now state our
main theorem, Le., the generalized Gelfand-Naimark theorem for C*-algebras
on the CP-extreme boundary DH(A) = Irr(A: H).
Theorem 3.3. Let A be a C*-algebra and H be an infinite dimensional Hilbert
space with dimH ~ oc(A). Then,
A ~ A~(Irr(A: H), B(H)) (*-isomorphism).

Sketch of Proof. Let 11" E Irr(A : H)c, then there exists a net (11"",) in Irr(A: H)
such that 11"", -+ 11". For / E A~(Irr(A: H), B(H)), (/(11"",)) is a bounded Cauchy
net in B(H) in the c-topology, hence by the uniform boundedness principle there
exists a limit in B(H) in the c-topo!ogy, which we denote by c -lim", /(11"",). Let
us define the extension of / by
We can easily see that this definition of i' does not depend on the choice of
the net (11"",) converging to 11", and that i' is c-continuous on Irr(A: H( The
fact thati' is equivalent can be proved using techniques in [8]. Obviously, the
extension map / E A~(Irr(A : H),B(H)) ...... i' E A~(Irr(A: H)c,B(H)) is
one to one and onto, and is a *-isomorphism, hence our assertion follows from
Theorem 3.2. 0
We note that the classical Gelfand-Naimark theorem for commutative C*-

algebras can be recovered from Theorem 3.3 by the functions of the unitary
equivalence classes of one dimensional CP-extreme states (representations) of A
onH.
The following result is a direct consequence of Theorem 3.3 and refines [17;
Th.3.2] (cf. §1), and could be compared with the result on the duality for pure
states in [28].
Theorem 3.4. Let A and B be C*-algebras and H be an infinite dimensional

Hilbert space with dimH ~ sup{oc(A),oc(B)}. Then, there exists an isomor-
phism f1 : Irr(A : H) -+ Irr(B : H) which is BW- (or BS-) uniformly homeo-
morphic and equivalent (i.e., f1(u*1I"u) = u*f1(1I")u for 11" E Irr(A: H) and for all
partial isometry u with uu* ~ p,,) if and only if A and B are *-isomorphic.
4. The Dirichlet problem for the CP-extreme boundary.
For a commutative C*-algebra A, every continuous function f E Co(P(A))

on the pure states (characters) P( A) vanishing at the infinity, or equivalently
every uniformly continuous function f E Cu(P(A)) on peA), can be uniquely
extended to a w*-continuous affine function 1 E Ao(Q(A)) over the whole quasi-
state space Q(A) vanishing at 0, which connects the Gelfand-Naimark theorem
130 I. Fujimoto
with Kadison's function representation theorem. (This property is characterized

as the fact that the state space of A is a Bauer simplex if A is unital. cf. [3].)
By Theorem 3.3, this situation can be generalized in our setting of CP-convexity
as follows.
Theorem 4.1. Let A be a C*-algebra and H be an infinite dimensional Hilbert

space with dimH 2:: Qc(A). Then, every I E A~(Irr(A : H), B(H)) can be
uniquely extended to l' E AC(QH(A),B(H)).
In the convexity approach for the general Stone-Weierstrass problem, some

important classes of C*-algebras have been studied, notably perfect C*-algebras
([1], [28]) and semi-perfect C*-algebras ([7]). These algebras can be simply
characterized in the abstract Dirichlet problem in the setting of CP-convexity
as follows.
Proposition 4.2. Let A be a C*-algebra and H be an infinite dimensional

Hilbert space with dimH 2:: Qc(A). Then,
0) A is perfect if and only if everY''! E A~ (I rr( A : H), B( H)) can be uniquely
extended to l' E AC(QH(A), B(H)), or equivalently A~(Irr(A : H), B(H)) =
A~(Irr(A: H),B(H)).
If A is a separable unital C*-algebra, then
Oi) A is semi-perfect if and only if every I E A~(Irr(A : H), B(H)) can be
uniquely extended to l' E AB(Q H(A), B(H)) preserving its norm.
This approach to the general Stone- Weierstrass problem in CP-convexity the-

ory has been developed into [21].
5. CP-spectral theorem.
Let A be a C*-algebra, and let a E A be an arbitrary element, and let C*(a)

be the C*-algebra generated by a. Assume first that C*(a) is commutative, e.g.,
consider the case that a is normal. We define the (nonzero) spectrum 0"( a) of a
by
O"(a):= a(P(C*(a))) = {w(a); w E P(C*(a))}.
Then a : w E P( C*( a)) -+ w( a) E 0"( a) is a homeomorphism which maps
the infinity of P(C*(a)) to 0 E C in the limit, so that by the Gelfand-Naimark
A Ge1fand-Naimark theorem for C*-algebras 131
theorem we have C*( a) 3! Co(0"( a)) 3! Cu( 0"( a)). We shall generalize this spectral
theorem to the non-commutative case.
Definition 5.1. Let A be a C*-algebra, COCa) be the C*-algebra generated by

a E A, and let H be a Hilbert space with dimH ~ G e ( C*( a)). Then, we define
the operator spectrum O"H(a) C B(H) of a by
O"H(a):= a(Irr(C*(a) : H)) = {1l'(a); 1l' E Irr(C*(a) : H)}.
It is not difficult to see that the map a: 1l' E Irr(C*(a) : H) -> 1l'(a) E O"H(a)
is uniformly homeomorphic with the strong* topology in O"H(a).
Definition 5.2. A function f : O"H(a) -> B(H) is defined to be equivalent if,

for any b = 1l'(a) E O"H(a) with 1l' E Irr(C*(a) : H),
f(u*bu)=u*f(b)u for any partial isometry u such that uu*~p".
We shall denote by C(O"H(a)) [resp. CU(O"H(a))] the set of all B(H)-valued

continuous [resp. uniformly continuous] equivalent functions on O"H(a) with the
strong* topology in O"H(A) and c-topology in B(H). Then, by Theorem 3.3, we
can show the following CP-spectral theorem.
Theorem 5.3. Let A be a C*-algebra, COCa) be the C*-algebra generated

by a E A, and let H be an infinite dimensional Hilbert space with dimH ~
Gc(C*(a)). Then
(*-isomorphism),
where a corresponds to the identity map in O"H( a).
It would be interesting to see the relation of this result with some other
attempts toward non-commutative spectral theory, such as matricial range and
spectrum which was initiated in [6].
Remark. If we define the extended operator spectrum PH(a) of a E A by
PH(a) := a(Irr(A: H)c) = {1l'(a) : 1l' E Irr(A: H)c},
then, we can show
C*(a) c,; C(PH(a)) (*-isomorphism).

132 I. Fujimoto
REFERENCES
1. C.A. Akemann and F.W. Shultz, Perlect C*-algebms, Memoirs Amer. Math. Soc. 326
(1985).
2. E.M. Alfsen, On the Dirichlet problem 01 the Choquet boundary., Acta Math. 120 (1968),
149-159.
3. E.M. Alfsen, Compact convex sets and boundary integmls, Springer Verlag, Berlin-Heidelb-
erg-New York, 1971.
4. R. Alicki and K. Lendi, Quantum Dynamical Semigroups and Applications, Lecture Notes
in Physics, No. 286, Springer-Verlag, Berlin-Heidelberg, 1987.
5. W.B. Arveson, Subalgebms 01 C*-algebms, Acta Math. 123 (1969), 141-224.
6. W.B. Arveson, Subalgebms 01 C*-algebms II, Acta Math. 128 (1972), 271-308.
7. C.J.K. Batty, Semi-perlect C*-algebms and the Stone- Weierstmss problem, J. London
Math. Soc. (2) 34 (1986), 97-110.
8. K. Bichteler, A genemlization to the non-sepamble case 01 Takesaki's duality theorem lor
C*-algebms, Inventiones Math. 9 (1969), 89-98.
9. L.G. Brown, Complements to various Stone- Weierstrass theorems lor C*-algebms and a
theorem 01 Shultz, Commun. Math. Phys. 143 (1992), 405-413.
10. E.B. Davis, Quantum Theory 01 Open Systems, Academic Press, London, 1976.
11. J. Dauns and K.H. Hofmann, Representation 01 rings by sections, Mem. Amer. Math. Soc.
83 (1968).
12. R.S. Doran and V.A. Victor, Chamcterizations 01 C*-algebms; The Gelland-Naimark
theorems, Marcel Dekker Inc., New York and Ba.sel, 1986.
13. M.J. Dupre and R.M. Gillette, Banach bundles, Banach modules, and automorphisms 01
C*-algebms, Pitman, London, 1983.
14. J.M.G. Fell, The structure 01 algebms 01 opemtor fields, Acta Math. 106 (1961), 233-280.
15. 1. Fujimoto, CP-convexity and its applications, Ph.D. dissertation, University of Florida,
1990.
16. I. Fujimoto, CP-duality lor C*- and W*-algebras, In W.B. Arveson and R.G. Douglas
(eds.), Operator Theory/Operator Algebras and Applications, Proc. Symposia in Pure
Math. 51 (1990), Part 2, 117-120.
17. I. Fujimoto, CP-duality lor C*- and W*-algebms, J. Operator Theory, to appear.
18. I. Fujimoto, A Gelland-Naimark theorem lor C*-algebms, submitted for publication.
19. 1. Fujimoto, Decomposition 01 completely positive maps, submitted for publication.
20. 1. Fujimoto, CP-extreme and lacial structure 01 CP-state spaces, in preparation.
21. 1. Fujimoto, The geneml Stone- Weierstrass problem in CP-convexity theory, in prepara.-
tion.
22. 1.M. Gelfand and M.A. Na.imark, On the embedding 01 normed linear rings into the ring
olopemtors in Hilbert space, Mat. Sbornik 12 (1943), 197-213.
23. R.V. Kadison, A representation theory lor commutative topological algebm, Memoires
Amer. Math. Soc. 7 (1951).
24. R.V. Kadison, A genemlized Schwarz inequality and algebmic invariants lor opemtor
algebms, Ann. of Math. 56 (1952), 494-503.
25. K. Kraus, Geneml state changes in quantum theory, Ann. of Physics 64 (1971), 311-335.
26. K. Kraus, States, Effects, and Operations, Lecture Notes in Physics, No. 190, Springer-
Verlag, Berlin-Heiderberg, 1983.
27. V.1. Paulsen, Completely bounded maps and dilations, Longman Scientific & Technical,
Essex, 1986.
28. F.W. Shultz, Pure states as a dual objects for C*-algebras, Commun. Math. Phys. 82
(1982), 497-509.
29. W.F. Stinespring, Positive functions on C*-algebras, Proc. Amer. Math. Soc. 6 (1955),
211-216.
30. M. Takesaki, A duality in the representation theory of C*-algebras, Ann. of Math. 85
(1967), 370-382.
31. M. Takesaki, Theory of operator algebras I, Springer-Verlag, New York-Heidelberg-Berlin,
1979.
Department of Mathematics, University of Florida, Gainesville, FL 32611

The Generalized Weyl-von Neumann
Theorem and C* -algebra Extensions
Huaxin Lin
An important problem in the C* -algebra theory is to classify the exten-

sIOns
O-tA-tE-tB-tO
of B by A, where A, B, E are C* -algebras. We will assume that both E and
B are unital and the surjective map from E to B is also unital. Furthermore,
we assume that extensions are essential, i.e. A may be viewed as an essential
ideal of E. The BDF-theory classifies those extensions when A = ]{ and
B = C(X), where ]{ is the C*-algebra of compact operators on an infinite
dimensional and separable Hilbert space and X is a compact metric space
([BDF1] and [BDF2]). Since early 1970's, the C*-algebra extension theory
and the ]{]{ -theory have been developed rapidly ( we are not attempting to
give a complete list of references but refer the reader to [BI] for references).
With the Universal Coefficient Theorem (see [RS,1.17]), for example, one can
compute Ext(A, B) in many cases. However, unlike the original BDF-theory,
in general, Ext(A, B) does not provide enough information for classifying
these extensions. In fact, in general,
(1) there are many (often infinitely many) unitary equivalence classes of
trivial extensions;
(2) if[T) = 0 in Ext(A, B), it is not known that T is in fact trivial (what we
know is that there is a trivial extension TO such that T + TO is trivial);*
(3) if A is not stable, Ext(A, B) tells very little about the extensions.
* We have found examples of non-trivial extensions T such that [Tl 0 in

Ext(A, B) ([Lin 9)).
The generalized Weyl-von Neumann theorem 135
In this note, we will give a complete classification of the extension of B

by A for certain C" -algebras.
Recall that an extension T is trivial if the exact sequence 0 -+ A -+ E -+
B -+ 0 splits. To give a complete picture of these extensions, it is essential
to obtain more information about trivial extensions. For example, we want
to know when two trivial extensions are unitarily equivalent.
Let A be a C" -algebra. The multiplier algebra A is
M(A) = {x E A"": xa,ax E A},

where A"" is the enveloping von Neumann algebra of A. Here are three ex-
amples:
(i) If A = Co(X), where X is a locally compact Hausdorff space, then
M(A) ~ Ch(X) ~ C(f3X).
(ii) If A = K, then M(A) = B(H).
(iii) Let A = B ® K, where B is a unital C"-algebras. Set
00
HB = {{an} : an E B, ~ a~an converges in norm}.

n=l
The space HB is called the standard Hilbert module over B and A may be
identified with the "compact" operators on HB. Then M(A) is the set of all
bounded module maps on HB with adjoints.
We have the following commutative diagram:
0 -+ A -+ E -+ B -+ 0
II t tT
0 -+ A -+ M(A) -+ M(A)/A -+ 0
Since the extensions are assumed to be essential, the map T : B -+ M(A)/A

is injective. These extensions are uniquely determined by the map T. This is
the reason that we need to know more about M(A) and M(A)/A in order
to have a better understanding of extensions. In the BDF -theory, it is the
136 R.Lin
Weyl-von Neumann- Berg Theorem that gives the needed information about
trivial extensions when A = J( and B = C(X). In the first part of this talk,
we will present some generalized Weyl-von Neumann Theorems.
Let T be a self-adjoint operator on an infinite-dimensional, separable
Hilbert space. The classical Weyl-von Neumann Theorem says that for f > 0,
there are a diagonalizable self-adjoint operator D and an operator k E J( such
that
T = D + k with IIkll < f.
We say that the generalized Weyl-von Neumann theorem holds in M(A),
if for any self-adjoint element x E M(A) and any f > 0, there exist an
(increasing) approximate identity {en} for A consisting of projections and a
bounded sequence {An} of real numbers such that
00
x=LAn(en-en-d+a (eo = 0)
n=l
where a E A and lIall < f.

At this point, we would like to remind the reader that a C·-algebra A has
real rank zero (denoted by RR(A) = 0), if the set of self-adjoint elements
with finite spectra is dense in A •. a . (see [BP]). It is known ([M] and [Zl, 3.1])
that the generalized Weyl-von Neumann Theorem holds for M(A) if and only
if RR(M(A)) = O.
Is it true that RR(M(A)) = 0 for every (1-unital AF-algebra?
This question was first raised by G. A. Elliott in 1978 at Tohoku and
more recently by L. G. Brown and G. K. Pedersen in 1988 ([BP, 3.22]). It
is proved by L. G. Brown and G. K. Pedersen ([BP, 3.21]), S. Zhang ([Z2,
2.8 and 2.10]) and N. Higson and M. Rc/Jrdam ([HR, 4.4]) that if A is a (1-
unital simple AF-algebra with a unique trace (up to scalar multiples), then
RR(M(A)) = O.
In fact, we have
Theorem 1 ([Lin3, 3.7])
RR(M(A))=O for every u-unital AF-algebra. Equivalently, the general-

ized Weyl-von Neumann Theorem holds for every u-unital AF -algebra.
In [BP, 3.22], there are two more conjectures:
(1) RR(M(A)) = 0 for every u-unital C·-algebra A for which RR(A) = 0
and I<l(A) = O.
(2) If A is a u-unital C·-algebra with RR(A) = 0, then RR(M(A)/A) = o.
By studying the exponential rank of C·-algebras of real rank zero and
the unitary groups of their multiplier algebras, and using the techniques in
[Lin3], we are able to prove the following:
Theorem 2 ([Lin7, 10]) If we assume that A has stable rank one, then
the conjecture (1) is correct.
Theorem 3 ([Lin7,15])If we assuem that A is simple and has stable rank
one, then the conjecture (2) is correct.
It should be noticed that both conjectures are correct for purely infinite
simple C·-algebras ([Z2, 2.6]). It is not known that there are simple C·-
algebras which have stable rank greater than one but are not purely infinite.
Now we go back to the extensions. We will restrict our attention to
the case that B = C(X) and that RR(M(A)) = o. To distinguish trivial
extensions, we define
Definition 4([Lin6, 1.4] An extension 7 : C(X) -+ M(A)/A is said to
be totally trivial ifim7 C C C M(A)/A, where C is a (unital) abelian AF-
algebra.
We notice that C ~ C(Y) for some compact, totally disconnected space
Y.
Proposition 5([Lin6,1.5]) Suppose that 7 : C(X) -+ M(A)/A is totally
trivial. Then there exist an (increasing) approximate identity {en} for A
consisiting of projections and a dense subset {An} in X such that
00
7(/) = 11"0:= f(An)(e n - en-I))

n=l
for each f E C(X), where 11" is the quotient map from M(A) to M(A)/A.
138 R.Lin
Let Ext( C(X), A) be the unitary equivalence classes of extensions of

C(X) by A and T( C(X), A) be the unitary equivalence classes of totally
trivial extensions. If A is stable, then both Ext( C(X), A) and T( C(X), A)
are abelian semi-groups. We use the notation Ext( C(X), A) for the quotient.
From the chapter II of [Lin6], we know that Ext(C(X),A) is a group. One
reason that we study Ext( C(X), A) instead of Ext( C(X), A) is that it is
much easier to deal with totally trivial extensions than to deal with trivial
extensions. In the case that RR(M(A)) = 0, trivial extensions are likely to
be totally trivial. Eventhough we do not know this is true in general. One
of the results in [Lin6] is the follwoing version of BDF-theorem.
Theorem 6([Lin6, 8.1])
Let A be a u-unital purely infinite simple C*-algebra with KI(A) = 0 and
X be a compact subset of the plane. Then
Ext( C(X), A) = Ext( C(X), A) 9! H om(KI (C(X)), Ko(A)).

Notice that Ext(C(X), A) is the set of (unitary) equivalence classes of exten-
sions not the quotient group Ext( C(X), A). So Theorem 6 gives a complete
classification of these extensions. In this case, all trivial extensions are uni-
tarily equivalent and they are totally trivial. Furthermore, if T+TO is (totally)
trivial for some (totally) trivial extension TO, then T is (totally) trivial. We
also notice that C*-algebras On satisfy the conditions in Theorem 6 (see [Cu,
3.8]).
Definition 7. Let T : B -+ C be a monomorphism, where Band C are
C* -algebras. Then T induces two maps
and
'YO(T) : Ko(B) -+ Ko(C).
Let Homo(I(*(B),K*(C)) be the set of those homomorphisms from Ki(B)
to Ki(C) (i = 0,1) which maps the image of the identity of B in Ki(B)
to the image of the identity of C in K;(C). In particular, in the case that

C = M(A)/A, we have a map
r: Ext(C(X),A) - t Homo(K.(C(X)),K.(M(A)/A))
defined by r( T) = (-yo( T), /1 (T)) for T E Ext( C(X), A).
We now consider non-unital (but u-unital) simple AF-algebras with unique
normalized trace ( for example, non-unital finite matroid algebras ([Dix])).
Our result for these C·-algebras is the following:
Theorem 7.([Lin9]) Let A be a non-unital but u-unital simple AF -algebra
with unique normalized trace and X be a compact subset of the plane. Then
the map
r: Ext(C(X), A) - t Homo(K.(C(X)),K.(M(A)/A))
is a bijection. Furtheremore, an extension T E Ext(C(X), A) is trivial if and
only if /1 (T) = 0, and every trivial extension is totally trivial.
Remark 8. we would like to make the following remarks:
(i) The C·-algebra A in Theorem 7 is not stable.
(ii) Suppose that TI, T2 E Ext(C(X), A). Then Tl is (unitarily) equivalent to
T2 if and only if /1(Td = /1(T2) and ,0(Td = 'oh). So the map /0 is used
to distinguish trivial extensions.
One may ask how much we know about K.(M(A)/A) when A is not
stable. Fortunately, in this case, we can compute K.(M(A)/A). Set
T(A) = {mT(p): m E Z,P E Proj(A)}.

We show in [LinS, 11) that if A is a non-unital but u-unital simple AF-algebra
with unique normalized trace, then
Ko(M(A)/A) ~ R/T(A).
Let inf(Ko(A) denote the set of infinitesmal elements of Ko(A). Then we
proved (in [LinS, 12]) that
Kl(M(A)/A) ~ inf(Ko(A)).
140 R.Lin
Corollary 9. Let A be a finite non-unital matroid algebra and X be a

connected compact subset of the plane. Then there is only one equivalence
class in Ext(C(X),A).
Remark 10. If A is a finite matroid algebra and A (8) K is the stablized
algebra. Then Ext(C(X), A (8) K) = Hom(KJ(C(X)), Ko(A)). So there are
nontrivial extensions if KJ (C (X)) =/: O. Furthermore, there are many different
equivalence classes of trivial extensions. Therefore, nonstable cases are quite
different from stable ones and that is why we say that Ext(C(X), A (8) K)
gives no information to Ext(C(X), A).
Example 11 Let G = Q E9 Q be the simple dimension group with the
strict order from the first coordinate and A be a finite non-unital simple
AF-algebra with Ko(A) = G([Elll]). Suppose that X is an annulus. Then,
from Theorem 7, we have
Ext( C(X), A) = H omo(K.( C(X)), K.(M(A)/ A))
~ Homo(Z E9 Z,R/Q E9 Q) ~ Q.
In particular, there is only one unitary equivalence class of (totally) trivial
extensions. However, if X is the Contor set, then every extension is totally
trivial, but
Ext( C(X), A) = T( C(X), A) = H omo( C(X, Z), R/Q).
Acknowledgements This note is for the talk given at GPOTS 1992

when the author was visiting University of Victoria. The author would like
to thank the organizers for their arrangement. He would also like to thank
John Phillips and Ian Putnam for their support during his stay at Victoria.
References
[Bl 1Blackadar, B., K-Theory for

Operator Algebras, Springer-Verlag New
York/Berlin/Heidelberg/London/Paris /Tokyo 1986.
[BDFt 1Brown, L. G., Douglas, R. and Fillmore, P., Unitary equivalence

modulo the compact operators and extensions of C* -algebras, Proc.
Conf. on Operator Theory, Springer Leture Notes in Math., 345 (1973),
58-128.
[BDF2 1Brown, L. G., Douglas, R. and Fillmore, P., Extensions of C*-

algebras and K-homology, Ann. of Math., 106 (1977) 265-324.
[BP 1Brown, L. G. and Pedersen, G. K., C*-algebras of real rank zero, J.

Funct. Anal., 99 (1991), 131-149.
[eu 1Cuntz, J., K-theroy for certain C*-algebras, Ann. of Math., 113 (1981),
181-197.
[Dix 1Dixmier, J. On some C*-algebras considered by Glimm, J. Functional

Analysis, I (1967), 182-203,
[Eff 1Effros, E, Dimensions and C*-algebras, CBMS Reginal Conf. Ser. in

Math., no.46, Amer. Math. Soc., Providence,R.I., 1981.
[Ellt 1Elliott,G. A., On the clssification of inductive limits of sequences

semisimple finite-dimesinal algebras, J. Algebra 38 (1976),29-44.
[Ell2 1 Elliott, G. A.,The classification of C*-algebras of real rank zero,

preprint.
[HR 1Higson, N. and R4>rdam, M., The Weyl-von Neumann Theorem for
multipliers of some AF-algebras, Can. J. Math., 43 (1991) 322-330.
[Lint 1Lin, H., Ideals of multiplier algebras of simple AF C* -algebras, Proc.

Amer. Math. Soc., 104 (1988), 239-244.
[Lin2 1Lin, H., Simple C*-algebras with continuous scales and simple corona
algebras, Proc. Amer. Math. Soc., 112 (1991), 871-880.
142 R.Lin
[Lin3 1Lin, H., Generlized Weyl-von Neumann theorems, Inter. J. Math. 2

(1991), 725-739.
[Lin4 1Lin, H., Generlized Weyl-von Neumann theorems (II), preprint

[Lin5 1Lin, H., Extensions by C*-algebras of real rank zero, Inter. J. Math.,
to appear.
[Lin6 1Lin, H., C*-algebra Extensions of C(X), preprint.

[Lin7 1 Lin, H., Exponetial rank of C*-algebras with real rank zero and
Brown-Pedersen's conjecture, J. Funct. Anal., to appear.
[Lin8 1Lin, H., Notes on K-theory of multiplier algebras and corona alge-
bras, preprint
[Lin9 1Lin, H., Extensions by C* -algebras of real rank zero II, in prepara-
tion.
[LZ 1Lin, H. and Zhang, S., Infinite simple C*-algebras, J. Funct. Anal.,lOO
(1991), 221-231.
[M 1Murphy, G. J., Diagonality in C*-algebras, Math. Z., 199 (1988) 279-

284.
[PI 1Phillips, N. C., Simple C*-algebras with property weak (FU), Math.
Scand., 69 (1991), 127-181.
[P2 1Phillips, N. C., Approximation by unit aries with finite spectrum in

purely infinite simple C*-algebras, to appear.
[RS 1Rosenberg, J. and Schochet, The Kunneth theorem and the univer-
sal coefficient theorem for Kansparov's generalized K - functor, Duke
Math., 55 (1987), 431-474,
[Zl 1Zhang, S., /(l-groups, quasidiagonality, and interpolation by multiplier

projections, Trans. Amer. Math. Soc. 325 (1991), 793-818.
[Z2 1Zhang, S., Certain C·-algebras with real rank zero and their corona
and multiplier algebras, Part I, Pacific J. Math., 155 (1992),169-197.
[Z3 1Zhang, S., Trivial /(l-f!.OW of AF algebras and finite von Neumann
algebras, J. Funct. Anal. 92 (1990) 77-91.
[Z4 1Zhang, S., A Riesz decomposition property and ideal structure of mul-
tiplier algebras, J. Operator Theory 24 (1990), 209-225.
The State University of New York
at Buffalo
Buffalo, New York 14214
U.S.A.
The cb-norm of a derivation
MARTIN MATHIEU
Mathemati.che. 1"IIit"t tler U"it/er.it.t Tti6i",."

A,,! tler Mo,.,e"II.lle 10. D-7!076 Tti6i",e". Germa",
ABSTRACT
The completely bO\Ulded nonn of an inn.,.. derivation of a C"-algebra is determined in terms
of the central Haagerup tensor nonn_ As a consequence, it is equal twice the distance of an
implementing element to the center of the local multiplier algebra.
1. Introduction
Let us start this paper with a warning to the reader. Albeit indicated by the
title, we will not attempt to calculate the completely bounded norm (cb-norm) of an
arbitrary derivation 6 of a C*-algebra A on a Hilbert space H, nor to decide when
6 is completely bounded. The latter is well known to be equivalent to the possibility of
extending 6 to the whole of B(H), and this, so far, can only be done under additional
assumptions on A, see e.g. [10]. Rather than that we shall assume that 6 maps A
into A again, which makes it completely bounded automatically (as it is inner in A"),
and even more restrictive, we will suppose that 6 is inner in the multiplier algebra
M(A) of A. In this case, however, the cb-norm of 6 can be determined by the central
Haagerup norm of 1 ® a - a ® 1, where a E M(A) is an element implementing 6.
Henceforth, 6" will denote the inner derivation z H za - az (z E A). One of
the first who calculated the norm of 6" seems to have been Stampfli [20] showing that
=
116" II 2 d( a, C) where d( a, C) denotes the distance of a to the scalar multiples of 1,
if A = B(H). It follows immediately that this formula remains true if A is a primitive
C*-algebra and a E M(A). This can be pushed to the case of prime C*-algebras by
standard methods, see [12, Proposition 2.23], but now reaching the utmost since, even
when replacing d(a, C) by the distance to the center Z(M(A)), in general, 116,,11 will
be smaller than 2d(a,Z(M(A))) [16]. However, if the center is well- behaved, things
=
improve again. It was proved by Zsid6 [21] that 116,,11 2d(a,Z(A)) whenever A is
a W*-algebra (with partial results obtained earlier in [13], [14], and [16]), and this was
extended to AW*-algebras by Halpern, see [11]. In Theorem 3.6 below, we shall obtain
This paper is part of a research project supported by the DAAD. It is in final fonn and no version
of it will be published elsewhere.
1991 Mathem"tic. Subject Cl"..ific"tio,,: 46L57; 47B47.
The cb-norm of a derivation 145
a common generalization of these results to boundedly centrally closed C*-algebras. Our

method of proof relies on the canonical representation of the central Haagerup tensor
product M(A) ®z h M(A) as completely bounded operators on A, which is isometric
if A is boundedly centrally closed by the results in [6]. This enables us to calculate
the cb-norm of 60' which, as is well known, coincides with 116011, see e.g. [11], [15]. For
that reason - that our methods naturally yield the cb-norm rather than the norm-
we have chosen the (possibly misleading) title of the present paper; stressing thus the
complete boundedness point of view.
In a recent paper [19], Somerset obtained a different description of the norm of
an inner derivation 60' if a is self-adjoint. Building on work by Archbold [1], he
proved that 2 dCa, Z(A» ~ Orc(A) 116011 whenever A is a non-commutative unital
C*-algebra, where Orc(A) E (0,00] is a constant arising from the primitive spectrum
of A. By the results below, Orc(A) = 1 if A is boundedly centrally closed.
Acknowledgements. We are indebted to Larry Fialkow and Roger Smith for a dis-
cussion on the topic of this paper. The results herein were obtained during a visit to the
Universitat Autonoma de Barcelona where the author enjoyed a warm hospitality by
the Departament de Matematiques and the Centre de Recerca MatemMica del Institut
d'Estudis Catalans.
2. Representing the central Haagerup tensor product

As we shall use the main result of [6]- which was presented at GPOTS 92 by the
present author - to obtain our description of the cb-norm of 60' we devote this section
to explaining its contents and terminology.
Throughout A will denote a C*-algebra with primitive spectrum A and M(A)
denotes its multiplier algebra. There is a canonical contractive representation of the
Haagerup tensor product M(A) ®" M(A) as completely bounded operators on A by
O(a®b) =axb (x E A, a,b E M(A»;
for some basic properties of the Haagerup norm 11·11" and the Haagerup tensor prod-
uct see [8] and [11], for example. This representation 0 is isometric into the space
(CE(A), 11·11<6) if and only if A is prime (the reason for this is that the prime C*-
algebras are precisely those for which the center Z of the local multiplier algebra (to
be defined below) is one-dimensional).
We define two ideals J and i as the ideal respectively the closed ideal generated by
l®z-z®l, z E Z(M(A», in M(A)®M(A) respectively M(A) 0" M(A). The central
Haagerup tensor product M(A) 0z" M(A) is then defined to be M(A) 0" M(A)/i,
which coincides with the completion of M(A) 0 M(A) / J with respect to the quotient
norm induced by the Haagerup norm, if J is closed in M(A) 0 M(A). In any
case, M(A) 0 M(A) /J is nothing but the tensor product M(A) 0Z(M(A)) M(A) as
Z(M(A»-bimodules. Since 0 is a Z(M(A»-bimodule map, it induces a bimodule
homomorphism
Oz: M(A) 0z" M(A) --+ CE(A)
which again will be contractive.

146 M. Mathieu
In fact, as examples show, {}z is not always isometric [9]. However, the following
was obtained in [6, Theorem 3.7].
Theorem 2.1. For every boundedly centrally closed C*-algebra A the canonical
homomorphism (}z is an isometry from M(A) ®z Ia M(A) into CB(A).
This result had been proven for von Neumann algebras on a separable Hilbert space
as well as for C*-algebras whose primitive spectrum is Hausdorff in [9, Theorems 2.4
=
and 3.1] extending previous work by various authors (in particular the case A B(H)
which was first treated by Haagerup). Rather than giving all the pertinent references
here, we refer the reader to [6].
In order to explain the notion of a boundedly centrally closed C*-algebra we need
to introduce the concept of the local multiplier algebra. If 1t.I2 are closed essential
ideals in A such that 12 ~ It then, by the universal property of the multiplier algebra
of h there is a unique injective *-homomorphism P12:M(It) -> M(I2) extending
the inclusion 12 ~ It. We may think of P12 as "restricting the multipliers to 12 ".
The direct limit ~ M(I) of this system {M(I)jP12, 12 ~ It} of C*-algebras and
isometric *-homomorphisms indexed by the directed set of all closed essential ideals in
A is called the local multiplier algebra of A and denoted by Mloc(A). The direct
limit ~ Z(M(I» of the centers Z(M(I» of M(I) with corresponding isometric
*-homomorphisms will be denoted by Z, and it coincides with the center Z(Mloc(A»
of Mloc(A), by the local Dauns-Hofmann theorem [5, Theorem 1].
We define the bounded central closure C A of A by AZ, which is a C*-subalgebra
=
of Mloc(A), and call A boundedly centrally closed if A cA. There are the following
characterizations of these C*-algebras, see [6].
Proposition 2.2. The following conditions on a C*-algebra A are equivalent.

(a) A is boundedly centrally closed.
(b) A is extremally disconnected.
(c) Z(M(A» = Z.
To illustrate this concept further, we list a number of examples of boundedly cen-
=
trally closed C*-algebras: every prime C*-algebra (with Z C 1), every AW*-algebra,
the bounded central closure C A [6, Proposition 3.10], and the local multiplier algebra
Mloc(A) itself [5, Theorem 2]. Moreover, every closed ideal in a boundedly centrally
closed C*-algebra inherits this property, but not every quotient does (loo/co provides
an example).
An important step towards the proof of Theorem 2.1 is the observation that, for a
boundedly centrally closed C*-algebra A, the kernel of {} coincides with the ideal J,
which therefore is closed, as well as the following description of the central Haagerup
norm lIullz" of u E M(A) ®z M(A):
lIuliz II = sup
. 111I"(u)lI" (1)
where the supremum is taken over all irreducible representations 11" of A and 1I"(u)
is an abbreviation for (11" ® 1I")J(u), the induced representation of M(A) ® M(A) /J
evaluated at u [6, Theorem 3.5].
3. Application to derivations
The results in the previous section immediately yield the following description of
the cb-norm of an inner derivation.
Theorem 3.1. Let A be a boundedly centrally closed C*-algebra. For each a E M(A)
we have that
lIc5a ll cb = 1110z a - a 0z 111z h = sup 1110 lr(a) - lr(a) 0 111h.

.. (2)
To obtain further information, one may apply Stampfli's result which yields
1Ic5.. (a)lIcb = 2 d( lr( a), C) for all lr together with the isometric property of () in the
case of primitive C*-algebras, i.e. 1Ic5.. (a)lIcb = 1110lr(a) - lr(a) 0 111h.
Corollary 3.2. II A is boundedly centrally closed, then, lor each a E M(A),
lIc5a ll cb = 2 sup d(lr(a), C) . (3)

.
However, one can circumvent the application of Stampfli's theorem by a direct calcula-
tion of the Haagerup norm of 10 a - a 01 (which, nevertheless, follows the lines of a
proof of Stampfli's result proposed in [12]).
Lemma 3.3. For each a E B(H) we have 1110 a - a 0 111h =2 dCa, C).
PROOF. For all ,\ E C,
from which 1110 a - a 0 111h ~ 2 dCa, C) follows. To obtain the reverse inequality take
a unit vector e E H and let u E B(H) be the unitary defined by ue = -e,
U7] 7] =
for all 7]..1 e. Picking 7]..1 e with 11'111 = 1 we put I(x 0 y) = (xye 17]) and obtain
a functional IE (B(H) 0h B(H))* with 11/11 ~ 1. Then
1110a-a0111h ~ lI(u0 u·)(10 a - a 0 1)lIh

= lIu0 u•a -ua 0u·llh
~ I/(u 0 u·a - ua 0 u·)1
= I(uu·ae 1'1) - (uau·e I '1)1
= I(ae I 7]) + (ae I 7])1
= 21(ae 17])1·
By Ando's formula [1), [2, Theorem 12.59), d(a,C) = sup {I(ae I 7])11 e,7] E H, lIell =
117]11 = 1, e..1 7]} wherefore, by the above, 1110 a - a 0111h ~ 2 dCa, C). 0
Remark. By the injectivity of the Haagerup norm, it is sufficient to compute
111 0 a - a 0 111h with respect to the operator space lin{l, a}; however, it seems
advantageous here to use all of B(H).
As a consequence of the Lemma, we derive Corollary 3.2 as well as the next result
without the use of Stampfli's theorem.
148 M. Mathieu
Corollary 3.4. If A is prime, then for each a E M(A)
116.. 11.6 = 2d(a, Z(M(A»). (4)
PROOF. We have that 116.. lIcb = 111 ®z a - a ®z 111n = 111 ® a - a ® lib. = 2d(a, C)
and clearly, Z(M(A» = C 1. 0
Our final aim is to establish (4) for all boundedly centrally closed C*-algebras. The
following is the last ingredient missing.
Theorem 3.5. Let A be a boundedly centrally closed C*-algebra. For each a E M(A)
we have that
d(a,Z) = sup d(1I"(a),C),
. (5)
where the supremum is taken over all irreducible representations of A.

PROOF. Since all extensions of an irreducible representation 11" of A to its multiplier
algebra are equivalent, it is irrelevant in the following which extension of 11" (again
denoted by 11") we take. If z E Z - which equals Z(M(A» by Proposition 2.2 -,
. .
then 1I"(z) E C 1 wherefore lIa - zll = sup 111I"(a) -1I"(z)1I ~ sup d(1I"(a), C) and thus
d(a,Z) ~
.
sup d(1f(a), C) .
. °implies that a E Z,
For the reverse inequality note at first that sup d( 11"(a), Cl =
.
whence we may assume that sup d(1f(a), C) = 1 in the sequel. Then, sup d(a" C) = 1
,~
as well, where a, is the image in M(A)/ker1l" if t=ker1l"nA. It now suffices to show

that, for each e > 0, there is z E Z such that lIa - zll ~ 1 + e.
°
For each tEA, the distance to the scalars, d(at,C), is attained at some A, E C.
For each I' E C, r> let B(I', r) denote the open ball with radius r and center 1',
and let B(I', r) be its closure. Since IA,I ~ d(a" _C) + lIa,lI ~.1 + II all , all A,r belong to
B(O, 1 + lIalD. Let e> ° and take A..... ,Ar
_
E B(O, 1 + lIall) such that U B(Aj,e)
j=1
cover B(O, 1 + lIalD. For every A E C, the function t ...... lI(a - A),II = lIa, - All is
lower semi-continuous, and hence
is open. If tEA is chosen, there is j E {I, ... , r} such that IAj - Ad < e. In this
case,
so that t cannot belong to UAj" This shows that nU

r
j=1
Aj = 0. As A is extremally
disconnected by Proposition 2.2, it follows that nU

r
j=1
A·
1
= 0 as well. Therefore,
{V; 11 ~ j ~ r} with V; = A \ UAj forms a covering of A by clopen subsets. Putting
nU
j-l
W1 = V1 and Wj = V; n A, for 2 ~ j ~ r we obtain from this a covering by
(=1
pairwise disjoint clopen subsets Wj ~ V;.
The cb-nonn of a derivation 149
We now put ej =
Xw;, 1 :$ j :$ r which is continuous on A and hence a
projection in Z (by the Dauns-Hofmann theorem). Finally, putting % Ej=1 >'jej =
we obtain
r
lIa - %11 =sup lIat -
t
%tll =sup lIat - E >'j(ej)tll :$ 1 + e
t j=1
as eiej = 6ijej and each t belongs to precisely one Wj ~ V; ~ {t e A Illat - >'jll :$

1 + e}. This accomplishes the proof. 0
Theorem 3.6. Let 6 be IJ derivIJtion on IJ boundedly centmlly closed C*-IJIgebm A

which is implemented by a e M(A). Then
11611.6 = 2 d(a, Z(M(A»). (6)
PROOF. We only have to combine Theorems 3.1 and 3.5 with Lemma 3.3. 0
Let us note some of the consequences, the first two of them being well known.
Corollary 3.1. For every AW*-IJIgebm A IJnd eIJch a eA we hIJve thIJt 1160 1106 =
2 d(a, Z(A».
Corollary 3.S. For every derivIJtion on IJ C*-IJIgebm, the norm IJnd the cb-norm
coincide.
PROOF. If 6: A -+ A is a derivation, then its second adjoint 6** is an inner derivation
on A**, say 6** =60 with a e A*·. We have
11611<6 = 116**11<6 = 116011.. =2d(a,Z(A**»

=2 sup. d("II"(a) , C) =sup. 116..(0)11
= 116011 = 11611,
where the supremum is taken over all irreducible representations "II" of A**. 0
Corollary 3.9. The set of inner derivations of a boundedly centmlly closed C*-algebm
is closed in the set of all derivations.
PROOF. By Theorem 3.6, the mapping a 1-+ i60 is an isometry from the Banach space
M(A)/Z onto the set of inner derivations, which therefore is closed; cf. [16, Theorem
5.3]. 0
Note also that the above results yield a nice description of the central Haagerup
norm of 1 ®z a - a ®z 1, whenever A is boundedly centrally closed, viz.
111 ®z a -a®z 111z/a =2 d(a, Z), (7)
and if the distance to the center is attained at %0 (which always holds if a is self-adjoint
by [19, Theorem 3.3]), then the following orthogonality relation is valid
111 ®z (a - %0) - (a - %0) ®z 111z/a = 111 ®z (a - %o)lIz/a + lI(a - %0) ®z 111z/a (8)
which in fact characterizes points at which d(a, Z) is attained.

150 M. Mathieu
4. Open problems
What happens with the cb-norm of .50 if A is not boundedly centrally closed?
There is at least one situation to which the same reasoning as in the previous section
applies, the one when A is Hausdorff. In this case, Hz again is an isometry [9, Theorem
= =
3.1], wherefore 11.50 11.6 111 ®z a - a ®z ll1z A sup 111 ® w(a) - w(a) ® lilA the latter
If
equality being proved in a similar way as (1). Since A if Hausdorff, the function
t 1-+ lIa, - ~II is actually continuous, thus, using the Dauns-Hofmann theorem, it is easy
to conclude that d(a, Z(M(A))) = sup d(w(a), C). Putting this together we obtain the
If
following result which, although we don't have a reference for it, is undoubtedly known.
Theorem 4.1. Let A be a C*-algebra such that its primitive spectrum is Hausdorff.
For each a E M(A) we have that 11.50 1106 = 2 d(a, Z(M(A))).
Observe that a C*-a1gebra A with A Hausdorff is boundedly centrally closed
precisely if Z(M(A» is an AW*-a1gebra. A class of C*-a1gebras containing both the
boundedly centrally closed ones as well as the C*-a1gebras with Hausdorff spectrum is
the following (by [4, Corollary 2.6]): for each two closed ideals 11,12 in A with
= =
1112 0 there is z E Z such that zh =F 0 and zI2 O. In this case we say that Z
separates orthogonal ideals. However, none of the steps needed to answer the following
question - such as Theorems 2.1 and 3.5 as well as (1) - are available so far.
Problem 4.2. Suppose that A is a C*-algebra such that Z separates orthogonal

ideals. Does, for each a E M(A), the equality 11.50 11 .. = 2d(a,Z(M(A))) hold?
Archbold proved in [7, Theorem 5.1] that (6) holds for unital homogeneous C*-
algebras. As these are type I C*-a1gebras and each type I C*-algebra A contains an
essential closed ideal I which has continuous trace (in particular, j is Hausdorff), the
following question is near at hand.
Problem 4.3. Suppose that the C*-algebra A contains an essential closed ideal I
such that 11.501/ 11.6 = 2 d(a, Z(M(I))) for all a E M(I). When does it follow that (6)
holds for all a E M(A)?
Note that Z(M(A» = Z(M(I» n M(A) and, by the proof of [7, Proposition 3.1],
11.50 11.6 = 11.501/ 1106 ~ d(a, Z(M(A))) for all a E M(A) under the hypothesis of 4.3.
As to quotients, it is known that (6) remains true for quotients of W*-a1gebras [3,
p. 1167], and for quotients of AW*- algebras the result is announced in [19]. We noticed
above that quotients of boundedly centrally closed C*-a1gebras need not be boundedly
centrally closed; in fact, [7, Example 3.3] yields a primitive C*-algebra with a quotient
not satisfying (6).
Problem 4.4. Let I be a closed ideal of a C*-algebra A such that (6) holds. When
is (6) inherited by AI H
Now let A be an arbitrary C*-a1gebra, and .5: A --+ A be a derivation. Since .5
leaves eacII closed ideal invariant, we have, for each essential closed ideal I, a restricted
derivation .51 which extends to a derivation .5M (/) on M(I), and 11.511 = =
11.5/ 11
II.5M (I)II· Passing this to the direct limit we obtain an extension of .5 to Mloc(A), again
denoted by .5, under preservation of the norm.
Theorem 4.5. Let 6 be a derivation on the C*-algebm A. For each a E Mloc(A)

such that 6 = 6,., we have that 11611c6 = 2 d( a, Z).
PROOF. This follows immediately from Theorem 3.6 and the fact that Mloc(A) is
boundedly centrally closed [5, Theorem 2]. 0
Pedersen proved in [18] that every derivation on a separable C*-algebra is in fact
inner in the local multiplier algebra. Hence, the formula in Theorem 4.5 applies to all
separable C*-algebras. Let us renew Pedersen's conjecture from [18].
Problem 4.6. Is every derivation on a C*-algebm inner in the local multiplier algebm?
Note that an affirmative answer to this question would imply the results for innerness
of derivations on simple C*-algebras and AW*-algebras.
Assuming that 6 is inner in M(A) as before, we also obtain a description of its
cb-norm for arbitrary C*-algebras.
Corollary 4.7. For eve.ry derivation 6 on a C*-algebm A and a E M(A) such

that 6 = 6,. we have that 11611c6 = 2 d(a, Z).
PROOF. Observe that the extension of 6 to Mloc(A) leaves CM(A) invariant: if
Z E M(A), z E Z, then 6(zz) = z6(z) + 6(z)z = 6(z)z E M(A)Z since 6 vanishes on
the center. Hence, we have an inner derivation 6,. on the boundedly centrally closed
C*-algebra C M(A) whose cb-norm equals 11611c6. Thus, the result follows by Theorem
3.6 and the fact that Z(CM(A» =
Z [6, Proposition 3.10]. 0
Note added in proof. The recent preprint "Inner derivations and primal ideals of
C*-algebras" by Somerset contains a characterization of those unital C*-algebras for
which (6) holds in terms of properties of A. We are grateful to Douglas Somerset for
providing us with his preprints as well as for a couple of remarks improving Section 4;
in particular, he pointed out that Theorem 4.1 also is a consequence of [11, Corollary
2.7].
References
1. T. Ando, Di.tance to the let oj thin operator., preprint, 1972.

2. C. Apostol, L. A. Fialkow, D. A. Herrero and D. Voiculescu, "Approximation of Hilbert space
operators", II, Research Notes in Math. 102, Pitman, Boston 1984.
3. C. Apostol and L. Zsid6, Ueal. in W*.alge6ra. and the Janction 'I oj A. Brown and C. Pearc"
Rev. Roum. Math. Pure Appl. 18 (1973), 1151-1170.
4. P. Ara, The e:&te"d.d .. ntroid oj C*·alge6ra., Arch. Math. 54 (1990),358--364.
5. P. Ara and M. Mathieu, A local "er,io" oj the Dau".·HoJmann theorem, Math. Z. 208 (1991),
349-353.
6. P. Ara and M. Mathieu, On the central Haagerap ten.or product, Proc. Edinburgh Math. Soc.
(1994), in press.
152 M. Mathieu
7. R. J. Archbold, 0 .. th • .. orm
Phil. Soc. 84 (1978), 273-291.
0' a .. i.. ner d.rivatio .. 0' a C*·algo6ra, Math. Proc. Cambridge
8. D. P. Blecher and V. I. Paulsen, T .... or prod".to

262-292.
0' op.rator 'pa ... , J. F\mct. Anal. 99 (1991),
9. A. Chatterjee and R. R. Smith, Th. ce ..tral Haag.np t ...lOr prod".t a ..d map. b.tw.... vo ..
N."ma .... alg.bTU, J. F\mct. Anal., to appear.
10. E. Christensen, E:t:I ....io ... 0' derivation., II, Math. Scand. 50 (1982), 111-122.
11. G. A. Elliott, 0 .. d.rivatio ... 0' AW*·alg.bra., Tohoku Math. J. 30 (1978), 263-276.
12. L. A. Fia!kow, Stnd"ral prop.rti .. 0'
.lem ... taTJ op.rator., in: "Elementary operators and
applications", Proc. Int. Workshop, Blaubeuren, June 1991; World Scientific, Singapore 1992,
55-113.
13. P. Gajendragadkar, Norm
170 (1972),165-170.
0' a d.rivatio .. 0' avo .. N."ma .... alg.bra, Trans. Amer. Math. Soc.
14. A. A. Hall, Ph. D. Thesis, Univ. Newcastle upon Tyne, 1972.

15. 0'
B. E. Johnson, Characterizatio .. a..d .. orm. derivatio ...... v ... N."man .. algebra., in: "Algebres
d'operateurs", Lect. Notes Math. 725, Springer-Verlag, Berlin 1979, 228-235.
16. R. V. Kadison, E. C. Lance, and J. R. Ringrose, Derivation. a..d a"tomorphi.m.
algebras, II, J. F\mct. Anal. 1 (1967), 204-221.
0' operator
17. V. 1. Paulsen, Thre. te ... or norms 'or operator .pac.. , in: "Mappings of operator algebras", (H.
Araki, R. V. Kadison, edo.), Birkhauser, Boston 1990.
18. G. K. Pedersen, Approzimating derivation. 0.. id.al. 0' C*.alg.bra., Invent. Math. 45 (1978),
299-305.
19. D. W. B. Somerset, The i.. n.r d.rivations a..d the primitive ideal .pa ••
Theory, to appear.
0' a C*-algebra, J. Oper.
20. J. G. Stampfli, The n.rm 0' a derivati ... , Pac. J. Math. 33 (1970), 737-748.
21. L. Zsid6, Th ....rm ., a d.rivati ... in a W*.alg.bra, Proc. Amer. Math. Soc. 38 (1973),
147-150.
Hopf C* -Algebras And Their Representations
Xiu-Chi Quan
Abstract: In this paper, we study the finite dimensional unitary rep-

resentations of Hopf C' -algebras. First of all, we study the characters of
representations, and show the orthogonal relations for characters of finite
dimensional irreducible representations and the matrix coefficients of finite
dimensional irreducible representations, these are generalizations of the re-
sults about repersentations of compact groups; also, we consider the repre-
sentation algebras of Hopf C' -algebras, and prove a Peter-Weyl type theorem
for representation algebras. Finally, we pose some open problems on Hopf
CO-algebra theory for possible future development.
1 Preliminaries
In this section, we recall the terminologies we are going to use throughout
this paper. Let us begin by recalling the definition for Hopf CO-algebras.
Let A be a CO-algebra with a dense *-subalgebra A, : A ~ A @ A a
C* -homomorphism, we say that (A, <1» is a Hopf C:* -algebra if the following
diagram commutes.
154 x.-c. Quan
A A®A
lif>®I
A®A®A
IP is called the comultiplication of (A, IP). If A is a von Neumann algebra, we

call A a Hopf- von Neumann algebra. By an involution of (A, IP), we mean a
*-anti isomorphism k : A --t A of period 2 such that the following diagram
COlllllllltes
k
A -------+-. A - - - - + -
k(6)k
A (x) A - - - - - - - - . .
where T : A0A --t A0A is the flip automorphism, T(aQ?Jb) = bQ?Ja, Va, b EA.
We say that e : A --t C) is a cOllnit of A if e is a C*-homomorphism and
the following diagram commutes
We say that (A,IP,k,e) IS a compact quantum group if the following

diagram commutes
Hopf C*-algebras 155
e
A C A
«)
ACi9A
I k@[ .
1m
A@A
Two Hopf C*-algebras (AI, «)l), (A2' «)2) with units are said to be isomor-
phic if there exists a *-isomorphism () : Al ---t A2 such that the diagram
()
Al • A2
1 <{)l ()@()
<{)2
Al @AI . A 2 @A 2
commutes. If (AI, <()t, k l ), (A2' <()2, k2) are involutive Hopf C*-algebras, we
require that () 0 kl = k2 0 (). If (AI, <()t, ell, (A2' <()2, e2) are Hopf C*-algebras
with counits, we also need that el = e2 0 ().
The Hopf C* -algebras theory grows out of the desires of understanding the
deformation of compact matrix groups. Because one parameter deformation
of matrix groups are not groups anymore, but they are Hopf C* -algebras.
Woronowicz studied the twist SU(n) group.
Definition. Let (A, «)) be a Hopf C*-algebra, a E A* a positive linear
functional, we say that a is a left Haar measure on A if
x*· a =< x*, [ > 'a, Vx* E A*.
Similarly, a is called a right Haar measure on A if
a· x* =< x*, [ > 'a, Vx* E A*.
Up to now, for all known examples of Hopf C*-algebras, there exist Haar
measures on them, we refer to [Q2] and [W2] for more details.
156 X.-C. Quan
2 Representations of Hopf C* -algebras
In this section, we study the characters of finite dimensional representa-

tions of Hopf C* -algebras and the orthogonal relations for characters and the
matrix coefficients of finite dimensional irreducible representations of Hopf
C* -algebras. We start by recalling comodules. Let A be a coalgebra, M a
linear space, and a linear map
1/J : M ----t A C$) M,
which makes the following diagrams commute.
A®A®M
r /).@I
A®M M
The pair (M, 1/;) is called a left A-comodule, and 1/; is said to be its structure
map. A right A-comodule can be defined similarly. For more information
about comodules, we refer to [AJ.
Let M be a left A-comodule with structure map 1/;, a subspace Ml C M
is said to left invariant if 1/;( M 1 ) c A (9 M1 ; we say that M is an irreducible
left A-comodule if M is the only nonzero left invariant subspace of M.
Let (A, <1» be a Hopf C*-algebra with a dense *-subalgebra A, V a finite
dimensional left A-comodule with structure map 1/; : V ----t A (9 V. If {e;}i=l
is a basis of V, and
The matrix (aij) is called the coefficient matrix of 'lj; respect to {ei};~l. Then
the comodule property implies that
Definition. The character X(V) of the comodule V is defined as
Note that X(V) does not depend on the choose of basis, i.e., for isomorphic
comodules, they have the same character.
If A =C(G) for some compact group G, then there is a one-to-one
correspnodence between the characters of G and the characters of C(G).
A left unitary representation 7r of A on ~. Hilbert space H is a comodule
map from H into A C8l H such that
< 7r(x), 7r(Y) >=< x, y >, \:Ix, y E H.

Let (A, cI>, k, e) be a Hopf C* -algebra, (J' a positive linear functional, we
say that (J' is left Haar measure if for all x* E A *, we have
x* . (J' =< x*, I > (J'.
Theorem 1. Let (A, cI» be a Hopf C* -algebra with left Haar measure (J', V
a finite dimensional left Hilbert A-comodule. Then V has an inner product
< .,. > such that V is a unitary left Hilbert A-comodule with respect to
<.,.>.
Theorem 2. (1) Let V, W be nonisomorphic finite dimensional irre-
ducible left A-comodules. Then the following is true:
(1.1) For any BE B(V, W), we have (J'. B = 0;
(1.2) For any BE B(V, V), we have (J'. B E C· I.
158 x.-c. Quan
(2) Let (A, ell, k, e) be an involutive Hopf C*-algebra, u the unital left
Haar measure on A. For nonisomorphic finite dimensional irreducible left
A-comodules V, W, we have
u(X(V)k(X(W))) = 0, u(X(V)k(X(V))) = 1.
Note that if 0 is a compact group, A = 0(0), then Theorem 2 is the
usual orthogonal relations for the characters of irreducible representations.
Theorem 3. (1) Let (A, ell, k, e) be an involutive Hopf O*-algbera with
unital left Haar measure u, V be an irreducible finite dimensional left A-
comodule with struccture map L, and dim V = d, let {el,""" et} be an
orthonormal set in V. Then we have
for j, k, 1, mE {I, 2""", t}.

(2) Let E denote the set of all unitary equivalent classes of finite dimen-
sional irreducible left unitary representation of A. For each 7r E E, fixed an
orthonormal basis {e},·,·,·,ell } of V1!"' n = dimV1!"' Let {aij: i,j = l,·,·,·,n}
be the coefficient matrix of V1!" with respect to the basis. Then the set of
elements n1/2aij E A, i, j E {I, 2, " ., " n}, is an orthonormal set in £2(A, u).
In particuar, we have
Theorem 4. (1) Let 7r E E, aij(l ::; i,j ::; n = dimV1!") be the coefficient
matrix of 7r with respect to a fixed orthonormal basis of V1!"' For any x* E A*,
we have
(2) a'Ik . x* = E~=l x*( a;k)a'Ir;

In particular, if 7r' E E, and n' = dim V1!"I, 1, m E {I""", n'}, then
(:J) ajk*ar,:. = (1/n)81!"'1!"18klaj~., whereforx,y E A,x*y = (uI8lI)[ell(y)(x*18l
1)).
Hopf C*-a1gebras 159
(4) Let V, W be finite dimensionalnonisomorphic irreducible unitary left

A-comodules with coefficient matrices {aij}, {b kl } be their coefficient matrices
with respect to the fixed bases, then we have
x(V) * X(V) = (l/dimV)x(V), X(V) * aij = aij * X(V) = (l/dirnV)aij

x(V) * X(W) = 0, X(V) * bkl = bkl * X(V) = O.
Theorem 5. (1) Let V be a finite dimensional unitary left A-comodule,
and
where \11,.,.,., V. are nonisomorphic irreducible unitary left A-comodules.

Then we have
if V" = Vj, j = 1, ... 8

O"(x(V)(x(V,,))*) = {
0, ortherwise
The decomposition is unique in the following sense. If V is isomorphic to

nj Uj ffi ... ffi ntUt as left A-comodule, where Uj , •.. , Ut are nonisomorphic
irreducible unitary left A-comodules, then 8 = t, Ui is isomorphic to V"(i)
and ni = m,,(i) for some permutation 7r of {l, ... ,.5}.
(2) Let V, \11, ... , V. and mj, ... ,m s as above, then
O"(x(V)x(V)*) = mi + ... + m;,

and V is irreducible iff O"(X(V)X(V)*) = 1.
Let (A, <1>, k, e) be a Hopf CO-algebra with a dense *-subalgebra A, ~ be
the set of all irreducible finite dimensional unitary left representations of A.
For any 7r E ~, let (aij) be the matrix coefficient of 7r. We say that A has
Peter- Weyl property (PWP) if {ai; : 7r E ~} is a basis for A.
Examples: (1) C(G), where G is a compact group. (Peter-Weyl Theo-
rem)
(2) Compact matrix quantum group in the sense of Woronowicz. (Defi-
nition)
160 x.-c. Quan
Theorem 6. Let (A, cI>, k, e) be a Hopf C*-algebra with PWP, then

(1) The left regular representation of A can be decomposed into direct
sum of finite dimensional irreducible representations;
(2) If (A, cI>, k, e) also has a faithful left Haar measure, then every ir-
reducible left unitary representation of A is finite dimensional and unitary
equivalent to a component of the left regular representation.
3 Open problems
In this section, we formulate some open problems about Hopf C*-algebra
theory, we hope that the understanding of these problems will help us to
study the quantum group and mathematical physics, also we expect them to
have applications on mathematics itself.
I. In the previous section, we study the finite dimensional irreducible
unitary representations of Hopf C* -algebras, so the natural question is: un-
der what conditions, every irreducible unitary representation is finite dimen-
siOilaI?
II. In the compact group case, the finite dimensional irreducible unitary
representations determine the group, this is the well-known Tannaka-Krein
duality theory. So we can ask: How about the Hopf C* -algebras'? More
precisely, what is a necessary and sufficient condition for a Hopf C*-algebra
to have Peter-Weyl property. Is there an analogue duality theory for Hopf
C*-algebras as the Tannaka-Krein duality theory for compact groups'? In
[Wa], Woronowicz formulated a Tannaka-Krein duality theory for compact
matrix pseudogroups by using category theory, we will study a Krein duality
theory for Hopf C*-algebras in a separate paper.
III. Establish Fourier and harmonic analysis on Hopf C* -algebras. This
is a potentially very important research area, it will have applications to
mathematical physics and quantum groups, just as in the compact group
case.
IV. Formulate Differential calculus and differential geometry on Hopf C*-
algebras. One of the main problems in this field is to formulate de Rham
theory.
V. Quantum Lie groups and quantum Lie algebras. We know that Hopf
(;'* -algebra with unit (in some sense) is a deformation of compact Lie group,
now a natural question arises: Can we formulate a Lie algebra from each
Hopf C* -algebra such that there is a one-to-one correspondence between the
representations of the Hopf C* -algebra and the representations of the Lie
algebra? Some work in this direction has been done by Woronowicz in the
twisted SU(2) case [WI], we expect that a very systematic study of this
subject is possible.
VI. Recently the study about the deformations of the classical polynomials
is very active, most the results are dealing with the orthogonality about the
deformed polynomials, so we can ask: Can we apply the orthogonal relations
in this paper to study the orthogonality of the deformations of the classical
polynomials?
ACKNOWLEDGMENTS
The author likes to take this opportunity to thank Prof. PaUe Jorgensen
for his constant support and encouragement, thank Prof. Tuong Ton-That
for his useful suggestions and comments, also thanks the refree for the sug-
gestions of the improvement of this paper. Work supported in part by NSF.
162 x.-c. Quan
References
[A] E. Abe, Hopf Algebras, Cambridge University Press, 1980.
[DR] S. Doplicher and J. Roberts, Endomorphisms of C*-algebras, cross
products and duality for compact groups, Ann. Math. 1:30 (1989), 75-119.
[D] V. Drinfeld, Quantum groups, Proc. I. C. M. (1986), 798-820.
[HR] E. Hewitt, K. Ross, Abstract Harmonic Analysis II, Springer-Verlag
Berlin 1970.
[L] R. G. Larson, Characters of Hopf Algebras, 17(1971) :352-368.
[Ql] X. Quan, Compact Quantum Groups and Group Duality, Acta App!.
Math. Vo!' 25, No.3 (1991) 277-299.
[Q2] X. Quan, Haar measures on Hopf C*-algebras, preprint.
[T] M. Takesaki, Duality and von Neumann Algebras, Lecture Notes in
Mathematics 242 (1972) 665-779.
[WI] S. Woronowicz, Twisted 8U(2) group. An example of a noncom-
mutative differential calculus, Pub!. RIMS, Kyoto Univ. 23 (1987) 117-181.
[W2] S. Woronowicz, Compact matrix pseudogroups, Comm. Math.
Phys. III (1987),61:3-665.
[Wa] S. Woronowicz, Tannaka-Krein duality of compact matrix pseu-
dogroups. Twisted 8U(n) groups, Invent. Math. 93 (1988) :35-76.
Institute of Fundamental Theory

University of Florida
Gainesville, FL :32611.
QUASI-DIAGONALIZING UNITARIES AND
THE GENERALIZED WEYL-VON NEUMANN THEOREM
SHUANG ZHANG*
ABSTRACT. We factor a quasi-diagonal unitary in the multiplier algebra M(A ®

K:) as a product of a diagonal unitary and a perturbation of the identity by an
element of A ® K:. This, combined with a break-through result of H. Lin [21],
yields that the generalized Weyl-von Neumann theorem holds in M(A®K:) if and
only if Kt{A) = 0 and every unitary in M(A ® K:) is quasi-diagonal. In turn,
RR(M(A ® K:» = 0 iff Kl (A) = 0 and for each unitary u of M(A ® K:) there
exists an approximate identity {en} of A ® K: consisting of projections such that
lien - uenu·1I - o.
1. Around Brown-Pedersen Conjecture
Throughout, we assume that A is au-unital C*-algebra. Suppose that the

following short exact sequence of C*-algebras is exact:
O--A--E--8--0.
Some matters have been clarified about the real ranks of these three C*-algebras.
If RR(E) = 0, then RR(8) = 0 and RR(A) = O. Assuming RR(A) =
0, then
RR(E) = 0 if and only if RR(8) = 0 and every projection in 8 lifts to a
projection of E ([7,3.14], [39,2.4]). In case A is stable, every projection
in 8 lifts if and only if Kl(A) = 0 ([45,2.10]); thus RR(E) =
0 if and only
if RR(A) = 0, RR(8) = 0, and Kl(A) = O. However, if one only assumes
RR(A) = 0, both RR(8) and RR(E) remain unknown even for the following
special short exact sequence
o --A - - M(A) -- M(A)/A-- o.

Brown-Pedersen conjecture states [7,3.22] that the condition 'RR(A) = 0'
together with the condition 'K1(A) = 0' would imply that RR(M(A)) = O. For
the specific case when A is an AF algebra G. A. Elliott also conjectured that
RR(M(A)) = 0 (in this case RR(A) = 0 and Kl(A) = 0 automatically).
Initial data supporting the Brown-Pedersen conjecture were, first, that RR(A)
is zero, since A is a closed ideal of M(A) ([30], [4]), and that the projection
lifting is also necessary in order RR(M(A)) = 0 ([45,2.4]; and, secondly, that
1991 Mathematics Subject Classification. Primary 46L05.

• Partially supported by an NSF Young Investigator grant at the Texas A&. M University.
164 s. Zhang
the affirmative answers were given for the following subclasses of C*-algebras of
real rank zero:
(1) A is any matroid C*-algebra or the stabilizations of any type Ih factor
([7,3.21J, later in [39,2.1,2.2J with a slightly general setting, and [16]);
and
(2) A is a purely infinite, simple C*-algebra (in particular, the stabilizations
of all Cuntz algebras On for 2 ::; n ::; +00 and all type III factors [39,2.6].
A striking progress was made recently by H. Lin ([21, Theorem 9], [19,3.4])
who proved
(3) RR(M(A» = 0 whenever RR(A) = 0, tsr(A) = 1, and K 1 (A) = 0; here
tsr(A) = 1 iff A has the cancellation property ([3]).
Lin's result (3) includes (1), but is independent of the result (2); furthermore,
Lin's result (3) has covered many classes of finite C*-algebras in the public
interests; for example, all u-unital AF C*-algebras, the stabilizations of irrational
rotation C*-algebras and of all Bunce-Deddens algebras, Elliott's basic building
block algebras, among others. However, the general conjecture seems to resist
all efforts and all techniques available. The main difficulty appears in dealing
with C*-algebras neither with cancellation nor purely infinite.
As the title indicates, the condition 'RR(M(A» = 0' has a lot to do with
the generalized Weyl-von Neumann theorem and other internal properties of
M(A). Various conditions equivalent to 'RR(M(A» = 0' have been derived in
[29], [45,§3], [42], [21,Corollary 6J, [7,2.6J, and in [6J, exploiting the
structure of M(A) on one hand, and attacking the Brown-Pedersen conjecture
from different angles, on the other hand. Let us spend a few more lines on some
of the equivalent conditions. In the exact sequence of the first paragraph, an
element x E E is said to be quasidiagonal (with respect to A), if there exists
an approximate identity {In} of A consisting of a sequence of projections such
that IIxfn - fnxll ----> 0 as n ----+ 00. The terminology 'quasidiagonal' is a
natural extension of the classical one, since, equivalently, x can be written as
L:~1 ai +a for some mutually orthogonal elements ai E A ( [45J); in particular,
a self-adjoint element x is quasidiagonal iff x = L:~1 tiTi+b, where b E A, {Ti} is
a sequence of mutually orthogonal projections in A, {til is a bounded sequence
of real numbers. If every self-adjoint element of E is quasidiagonal, then we say
that the generalized Weyl-von Neumann theorem holds in E ([45J). In case A
is taken to be the algebra I(. of all compact operators over a separable Hilbert
space 1£ and E is taken to be the algebra of all bounded operators C(ll), the
above reduces precisely to the classic Weyl-von Neumann Theorem ([39,3.1]).
It was shown that the general Weyl-von Neumann theorem holds in E whenever
RR(E) = 0 ([29]). The converse is certainly not valid in general; a familiar
counterexample is when A = 1(., B = C(Sl) and E is taken to be the Toeplitz
algebra. But for the particular case when E = M(A), the following conditions
are equivalent:
(a) RR(M(A» = O.
(b) The generalized Weyl-von Neumann theorem holds in M(A) ([29], [45J).
(c) The generalized Olsen's theorem holds in M(A) ([6 ,Corollary 4J);
i.e., Whenever x, y E M(A) such that xy E A, there exists a projection
Quasi-diagonalizing unitaries 165
P E M(A) such that x(I - p),py E A.

(d) [21] Every unitary of M(A) in the identity path component is approx-
imated by unit aries with finite spectrum (Le., M(A) has the weak (FU)
property, [33]).
(e) [42,1.3] Any two projections in M(A) are simultaneously quasi-diagonal
(Le., with respect to the same increasing approximate identity of A).
The equivalence between the condition (a) and (d) is a particular case ofthe
following general result of H. Lin [21, Theorem 5, Corollary 6]:
(f) A C*-algebra has real rank zero if and only if it has the weak (FU)
property.
There are many additional equivalent versions of the generalized Weyl-von
Neumann theorem that we cannot present here due to the page limitation. The
reader is referred to [29], [45], [42], [21], [7,2.6], and [6].
2. The Main Result

The only result in this short note is the following equivalence between quasi-
diagonalizing all self-adjoint elements and quasi-diagonalizing all unit aries with
a triviality of Kl(A) standing in between.
1. Theorem. Let A be au-unital C*-algebra. Then RR(M(A®K» if and =°

only if RR(A) = 0, Kl(A) = 0, and every unitary of M(A®K) is quasidiagonal.
The prooffor Theorem I heavily relies on the above Lin's result (f). Actually,
the following lemma is the additional ingredient besides the Lin's result (f) and
some results in [45, §2].
2. Lemma. Let u be a unitary element of E where A and E are as in the
short exact sequence above. Then u is quasidiagonal if and only if there exists
an approximate identity {en} of projections of A and a unitary v E Uo(A) such
that
vu = diag(wl, W2, ••• ,Wn, ... ),
where Wn for each n ~ I is a unitary in the corner (en - en-t}A(e n - en-I).
Proof. We first look at the direction if. Clearly, each unitary v in A can be
written as ~ +a for some number ~ with I~I = I and some element a E A. Then
vu = ~u + au = diag(~wl' ~W2' ••• , ~wn' ... ) + au.

Then
lIue n - enull
=1I,X«vu - au)e n - en(vu - au»11
=1I(au)e n - en(au)1I - + o.
Thus, u is quasidiagonal.
We now show the direction only if. Let € be any positive number. The
following argument uses the idea in the proof [13,2.4] for matroid algebras.
166 s. Zhang
Since Ilulnu· - Inll = lIuln - Inull --- 0, take nt such that lIuln, u· - In, II < 8t .
Then there exists [13,2. 1J a unitary Vt E A such that
(let 8t be small enough). Let Vt = .At + at for some at E A and some number
I.AII = 1. Now
II VI u(fn -
In, )u·vi - Un - In, )11
=IIVI ulnu'vi - Inll
:::;lIuln - Inull + II(aIU)/n - In(alu)lI--- o.
Then there exist n2 > nl and a unitary v2 in (1 - In, )A(1 - In,) such that
Proceeding this way, one find a sequence of unitaries {Vi} in Uo(A) such that
€
IIv; - 111 < 2i' and
ViVi_I ...VZVIUlnj = InjViVi-I ...V2VIU VI:::; j:::; i.
If m < n, then
IIVnVn-I ... V2 VI - VmVm-I ... V2 VIII
=IIVnVn-I ... Vm+1 - 111
n-m-l
: :; L II(Vn-k - 1)(Vn-k-l ... vm+dll

k=O
n-m-l n-m-l
= L IIVn-k - 111 < L 2 n€-k --- 0 as n > m - t 00.

k=O k=O
Thus, one sees by the Cauchy criterion that the sequence of products
converges in norm to a unitary V in Uo(A} It is readily checked that vuln; =

=
In; vu for any i 2: 1. Set ei In;; and set W; = (ei - e;_t}vu( e; - ei-d for i 2: 1,
as desired. This completes the proof.
3. Proof of Theorem 1.
First, assume that RR(M(A ® K» = O. Then every projection of M(A ®
K)/A ® K lifts to a projection in M(A ® K) ([45,2. 2J), and hence K l (A) = 0
([45,2.10J). Furthermore, 'RR(A®K) = 01 implies 'RR(A) = 01 ([4,2.7J,
[30 ,14J, [7]). We need only to show that every unitary of M(A ® K) is
quasidiagonal.
Combined H. Lin's result (f) above ([21,Corollary 6J) with a known fact
that the unitary group of M(A ® K) is contractible ([28J and [l1J), one sees
that that every unitary of M(A ® K) can be approximated by unitaries with
finite spectra (i.e., M(A ® K) has the (FU) property in the terminology in
[33J). The same standard arguments (for self-adjoint elements) as in [45,§3J
can be adopted to work with unitaries. For the convenience of the reader, we
outline a proof as follows.
Let u be any unitary of M(A ® K), and let {on} be a sequence of numbers in
tl
[0, decreasing to O. Choose a unitary Un in M(A ® K) with finite spectrum
such that Ilu - unll < On. Write
mn
Un = L AinPin
i=1
for some numbers {Ain} C SI and some mutually orthogonal projections {Pin}~nl
in M(A ® K) such that L:::nl Pin = 1. Since A is a-unital, each PinAPin is also
a-unital. For each 1 ~ i ~ mn take any sequential approximate identity of
pinAPin of projections, say {Tin(j)}~l' Set qj( n) := L:::n1 Tin(j) for j ~ 1 and
n ~ 1. Then {qj( n)} ~1 is an approximate identity of A consisting of projections
(for each fixed n ~ 1). Using a 'diagonal selection' as in the proof [45,3. 6J,
one can choose {qjJn)}~=1 which is an approximate identity of A such that
lIuqjn (n) - qjn (n )ull ----+ 0;
where {qjJn)}~=1 may not be increasing, however. Using the same arguments
as in [45,3. 4J, one can adjust it to an increasing approximate identity of A
consisting of projections with all other properties kept. Therefore, u is quasidi-
agonal.
We now assume that RR(A) = 0, Kl (A) = 0 (this is equivalent to the
condition that every projection in M(A®K)/ A ® K lifts to projections in M(A®
K», and every unitary of M(A ® K) is quasidiagonal. Since every projection in
M(A ® K)/ A ® K lifts, we need only to show that the corona algebra M(A ®
K)/A®K has real rank zero ([7,3.14J, [45,2.4J). In turn, by the result (f)
of H. Lin mentioned above again it suffices to show that M(A ® K)/ A ® K has
the weak (FU) property. Let u be any unitary in the identity path component
of M(A ® K)/ A ® K). Since the identity path component of M(A ® K)/ A ® K
is precisely the image of the unitary group of M(A ® K) under the quotient
map 11' [40,2. 3J. Take a unitary u of M(A ® K) such that 11'( u) = U. Then u
is quasidiagonal by assumption. It follows from Lemma 2 that there exists an
approximate identity {en} of A and a unitary v E Uo(A 0 K) such that
vu = diag( WI, W2, ... , W n , ... ),
168 s. Zhang
where Wn is a unitary in the corner (en - en-d(A ® ~)(en - en-d for each
n ~ 1. Since K 1 (A) = 0 and RR(A ®~) = 0 ([7]), it follows from the above
result of H. Lin once more that A ® ~ has the (FU) property. It follows from
[19,1.4] that each corner peA ® ~)p has the (FU) property, where p can be
any projection of A ®~. This, in particular, applies to each W n • One gets a
unitary w~ of (en - en-d(A ® ~)(en - en-I) with finite spectrum such that
IIwn - w~1I < 2~ 'v' n ~ 1.

Set u' = diag( w~, w~, ... , w~, ... ). Then vu -
u' := a E A ®~. Then u' can be
approximated in norm by unit aries of M(A®K) with finite spectra. To see this,
one takes a subdivision of the unit circle, say
Si := {eiS : 8i < 8:S; 8i+l}' i = O,l, ... ,no,
where {8;} is a subdivision of[O,211"). Write w: =
:Ei~1 >'ijTij for some complex
numbers >'ij E S1 and some mutually orthogonal projections {Tij} in (ei -
ei-d(A ® ~)( ei - ei-d· Set Pk = :E Tij. Then PkPI = 0 if k :f:. I. Set
Poj ES.}
no
.,~ •..= """
~
eiS.p k·
k=O
Then U Ehas a finite spectrum and lIuE-u'li < £ (if the subdivision is fine enough).
Therefore, U = 7r(vu) = 7r(u') can be approximated in norm by unitaries with
finite spectra. The proof is completed.
4. Corollary. (cf. [21,Corollary 12]) Suppose that A is au-unital C*-

algebra (may not be stable) with RR(A) = 0 and the (PU) property. Then
RR(M(A» = 0 if and only if every unitary in Uo(M(A», the identity path
component of the unitary group of M(A), is quasidiagonal.
Proof. H RR(M(A» =
0, then M(A) has the weak (FU) property. Hence, the
same arguments in the proof of Theorem 1 apply to all unitaries in Uo(M(A».
Thus, every unitary in Uo(M(A» is quasidiagonal.
Let u be any unitary in Uo(M(A». Then Lemma 2 implies that there is a
unitary v in Uo(A) such that
vu =diag( Wlo Wlo ••• , W n , •.• ),
where Wn is a unitary of (en -en-1)A(e n -en-d for each n ~ 1, and {en} is an
approximate identity of A consisting of a sequence of projections. Since A has
(FU), then every corner (en - en-dA(e n - en-d has (FU) ([19,1.4]). Now
use the same arguments as in the last paragraph of the proof of Theorem 1, one
sees that u is approximated in norm by unit aries with finite spectra. Therefore,
M(A) has the weak (FU) property, and hence RR(M(A» O. =
3. Two Problems
Problem 1. Assume that RR(A) = 0 and K1(A) = o. For each unitary in

M(A 0/C), is there an approximate identity {en} of projections of A 0/C such
that lien - ue n u*lI----+ 0 as n ---> oo?
By Theorem 1 in this note, the answer for this problem is positive if and
only if RR(M(A 0 /C)) = 0, and again if and only if the generalized Weyl-von
Neumann theorem holds in M(A 0/C).
Problem 2. Assume that A is a a-unital, non-unital, non-stable C*-algebra

with RR(A) = O. Is the unitary group U(M(A)) of M(A) connected? What is
Kl(M(A))?
If A is an AF algebra, then U(M (A)) is connected ( [15 ,4. 1] ); using the same
technique as in [13,2.4] for matroid algebras; that is, to approximate each
unitary of M(A) by a product of two exponentials. More generally, U(M(A)) is
connected whenever RR(A) = 0 and tsr(A) = 1; this was proved by combining
the technique in [13,2.4] and the result [21, Corollary 6], and again, H.
Lin's result (f) stated earlier is the key additional ingredient in order Elliott's
argument works. In particular, this includes an earlier result of L.G. Brown that
M(A) is connected when A is stably isomorphic to a Bunce-Deddens algebra (a
private communication with the author in 1988). On the other direction, if A
is purely infinite and simple but non-unital, then A ~ A 0 /C ([39, §1]); thus,
M(A) is even contractible ([28], [11]). Consequently, Kl(M(A)) = 0 if A is
one of the above C*-algebras mentioned in this paragraph.
For an arbitrary C*-algebra A of real rank zero, however, G. A. Elliott's
argument in [13,2.4] can no longer be adapted. New techniques seem to be
required.
Acknowledgment. The author wishes to thank H. Lin for several helpful

communications on the subject.
REFERENCES
1. J. Bunce and J. Deddens, A family of simple C·-algebras related to weighted shift opera-
tors, J. Funct. Anal. 19 (1975), 13 - 24.
2. B. Blackadar and A. Kumjian, Skew products of relations and structure of simple C·-
algebras, Math. Z. 189 (1985), 55 - 63.
3. B. Blackadar, K-theory for operator algebras, Springer-Verlag, New York Berlin Heidelberg
London Paris Tokyo, 1987.
4. B. Blackadar, Notes on the structure of projections in simple algebras, Semesterbericht
Funktionalanalysis, Tubingen Wintersemester (1982/83).
5. L.G. Brown, Stable isomorphism of hereditary subalgebras of C·-algebras, Pacific J. Math.
71 (1977), 335 - 348.
6. L.G. Brown, Interpolation by projections in C·-algebras of real rank zero, preprint (1992).
7. L.G. Brown and G.K. Pedersen, C·-algebras of real rank zero, J. Funct. Anal. 99 (1991),
131 - 149.
8. M-D. Choi and G.A. Elliott, Density of the self-adjoint elements with finite spectrum in
an irrational rotation C·-algebra, Math. Scand. (to appear).
9. J. Cuntz, K-theory for certain C·-algebras, Ann. of Math. 131 (1981), 181-197.
10. J. Cuntz, Simple C· -algebras generated by isometries, Comm. math. phys. 57 (1977),
173-185.
170 S.Zhang
11. J. Cuntz and N. Higson, Kuiper's theorem for Hilbert modules, Operator algebras and
mathematical physics, Amer. Math. Soc., Vol. 62, Proceedings of a Summer conference.
(1985), 17 - 24.
12. E.G. Effros, Dimensions and C"-algebras, CBMS Regional Conference Series in Mathe-
matics, A.M.S 46 (1981).
13. G.A. Elliott, Derivations of matroid C"-algebra. II, Ann. of Math. 100 (1974), 407-422.
14. G.A. Elliott, On the classification of C"-algebras of real rank zero, preprint.
15. G. A. Elliott, Automorphi.ms determined by multipliers on ideals of a C"-algebra, J.
Funct, Anal. 23 (1976), 1 - 10.
16. N. Higson and M. R,prdam, The Weyl-von Neumann theorem for multipliers of Bome AF
algebra, Can. J. Math (to appear).
17. G. G. Kasparov, Hilbert C"-modules: theorems of Stinespring and VoiculeBcu, J. Operator
Theory 3 (1980), 133 - 150.
18. H. Lin, Simple C"-algebras with continuous scales and simple corona algebras, Proc. Amer.
Math. Soc. 112 (1991), 871-880.
19. H. Lin, Generalized Weyl-llon Neumann theorems" Inter. J. Math. 2 (1991), 725-739.
20. H. Lin, Generalized Weyl-llon Neumann theorems, II, preprint.
21. H. Lin, Exponential rank of C· -algebras with real rank zero and the Brown-Pedersen
conjectures, J. Funct. Anal. (to appear).
22. H. Lin, Approximation by normal elements with finite spectrum in simple AF-algebras, J.
Operator Theory (to appear).
23. H. Lin, C"-algebra extensions of C(X), preprint.
24. H. Lin, Extensions by C"-algebras with real rank zero, I" Int. J. Math. (to appear).
25. H. Lin, Extensions by C"-algebras with real rank zero, II" preprint.
26. H. Lin and S. Zhang, On infinite simple C"-algebras, J. Funct. Anal. 100 (1991), 221 -
231.
27. H. Lin and S. Zhang, Certain Simple C·-algebras with nonzero real rank whose corona
algebras have real rank zero, Houston J. Math. 18 (1992), 57-71.
28. J .A. Mingo, K -theory and multipliers of stable C· -algebras, Trans. Amer. Math. Soc. 299
(1987), 255 - 260.
29. G. J. Murphy, Diagonality in C"-algebras, Math. Z. 199 (1988), 279 - 284.
30. G.K. Pedersen, The linear .pan of projections in simple C"-algebras, J. Operator Theory
4 (1980), 289-296.
31. G.K. Pedersen, SA W"-algebras and corona C"-algebras, contributions to non- commuta-
tive topology, J. Operator Theory 15 (1986), 15-32.
32. N. C. Phillips, Approximation by unitaries with finite spectrum in purely infinite C"-
algebras, preprint.
33. N. C. Phillips, Simple C"-algebras with the property weak (FU), Math. Scand. (to appear).
34. M. R,prdam, On the structure of simple C"-algebras tensored with a UHF algebra, J.
Funct. Anal. (to appear).
35. M. R,prdam, Ideals in the multiplier algebra of a stable algebra, J. Operator Theory (to
appear).
36. D. Voiculescu, A non-commutative Weyl-von Neumann theorem, Rev. Roum. Math. Pour
et Appl. 21 (1976), 97-113.
37. D. Voiculescu, Around quasi-diagonal operators, Centre de Researches Methematiques,
preprint (1991).
38. S. Zhang, On the structure of projections and ideals of COrona algebras, Can. J. Math. 41
(1989), 721 - 742.
39. S. Zhang, Certain C· -algebras with real rank zero and their corona and multiplier algebras,
Part I, Pac. J. Math 155 (1992), 169 - 197.
40. S. Zhang, Certain C· -algebras with real rank zero and their corona and multiplier algebras,
Part II, K-theory 6 (1992), 1 - 27.
41. S. Zhang, C"-algebras with real rank zero and the internal structure of their COrona and
multiplier algebras, Part III, Can. J. Math. 62 (1990), 159-190.
42. S. Zhang, C"-algebras with real rank zero and the internal structure of their corona and
multiplier algebras, Part IV, Inter. J. Math. 3 (1992), 309 - 330.
43. S. Zhang, A Riesz decomposition property and ideal structure of multiplier algebras, J.
Operator Theory 24 (1990), 209 - 225.
44. S. Zhang, A property of purely infinite simple C·-algebras" Proc. Amer. Math. Soc. 109
(1990), 717 - 720.
45. S. Zhang, K l-groups, quasidiagonality and interpolation by multiplier projections, Trans.
Amer. Math. Soc. 325 (1991), 793 - 818.
46. S. Zhang, Trivial Kl-flow of AF algebras and finite von Neumann algebras, J. Funct.
Anal. 92 (1990), 77 - 91.
DEPARTMENT OF MATHEMATICAL SCIENCES, UNIVERSITY OF CINCINNATI, CINCINNATI,

OHIO 45221-0025
E-mail address: szhang@ucbeh.san.uc.edu
Section IV
VON NEUMANN ALGEBRAS
AND SUBFACTORS
ON THE STRUCTURE OF FINITE DEPTH SUBFACTORS
DIETMAR BISCH I
Mathematical Sciences Research Institute

1000, Centennial Drive
Berkeley, CA 94720
USA
May 1992
ABSTRACT. We study the basic construction of a pair of reduced subfactors

Np C pMp, pEN' n M a projection and provide a method of computing
the graphs of the reduced subfactors using the Bratteli diagrams of the
original inclusion N eM. Then we describe what we mean by a fusion
algebra associated to a subfactor and explain how these fusion algebras can
be computed explicitly. In particular this provides an explicit method of
calculating the graphs of the reduced subfactors. We present the results of
+
these computations for various subfactors such as for the 3 V3 subfactor
and for Haagerup's smallest finite depth subfactor of the hyperfinite III
factor with index above 4.
1. Introduction
Given a sub factor N of the hyperfinite Ih factor M with finite Jones

index [M : N] < 00, we obtain the standard invariant for N C M in the
following way: let N C M C Ml C M2 C ... be the Jones' tower of factors
constructed by iterating the basic construction for N C M ([Jol]). The
double sequence of inclusions of finite dimensional centralizer algebras
N' n M,. C N' n MHl C ...
U U
M' n M,. C M' n MHl C ...
which form a sequence of commuting squares or paragroups ([Pol],[Ocl],
see also [GHJ]), is called the standard invariant for N eM. Popa intro-
duced recently a concept of amenability for an inclusion N C M and proved
that precisely the amenable subfactors of the hyperfinite III factor Rare
classified by their standard invariant ([Po2]' [Po3]). Finite depth subfac-
tors of R are those amenable subfactors NCR for which the width of
the Bratteli diagram describing the inclusion {N' n M,. C N' n MHlh
(resp. {M' n M,. C M' n M"+1},.) does not grow anymore from a certain
by an MSRI fellowship through NSF grant DMS 8505550

1 supported
1991 Mathematics Subject Classification. 46L10.
176 D. Bisch
point on. They are classified by an initial commuting square (paragroup)

([Pol],[Oc1))
N' n Mko C N' n Mk o+!

U U
M' n Mko C M' n Mk o+l
with ko large enough. The simplest finite depth subfactors are obtained
by letting a finite group G act on R by properly outer automorphisms
(which is possible for every finite group G) and considering the inclusion
R C R )<I G. The classifying commuting square for these inclusions and
variations of it (such as R)<I HeR )<I G, H eGa subgroup) can be writ-
ten down explicitly ([Oc2],[KY)). More interesting finite depth subfactors
are the Jones' subfactors ([Jol],[J02)) of index < 4 and Wenzl's braid type
subfactors ([Wel],[We2)) obtained from braid group representations via rep-
resentations of quantum groups at roots of unity. The canonical commuting
square for Wenzl's Hecke algebra subfactors was written down in ([EK)),
using explicit solutions to the Yang-Baxter equation by Jimbo-Miwa-Okado
computed in the context of certain solvable lattice models in statistical me-
chanics. Furthermore, the canonical commuting squares for subfactors of
index < 4 have been worked out in ([B-N],[I2],[Ka],[0c1],[Oc2]).
One can construct new finite depth subfactors from old ones by reducing
N C Mk by a projection pEN' n Mk, i.e. by considering the reduced
subfactors Np C pMkP. In this way, one obtains for instance Wenzl's in-
dex values ([WeI)) from reduced Jones' subfactors. However, these reduced
subfactors are not isomorphic to 'Wenzl's: they have different standard in-
variants (as remarked by Wenzl in unpublished work). The computation
of the standard invariant for a reduced subfactor N p C pM p, pEN' n M
a projection, contains as a first step the computation of the Bratteli dia-
grams (or inclusion graphs) for the inclusions of the higher relative com-
mutants of Np C pMp. This can be done explicitly by adapting Ocneanu's
combinatorial description of finite depth subfactors in terms of bimodules
([0c1],[Oc2],[Po4)): the computation of the inclusion graphs of reduced
subfactors amounts to computing certain fusion algebras associated to the
inclusion graphs (or principal graphs) of the original inclusion N eM (see
[Bi2)). This point of view was used in ([Il],[SV)) to show that certain graphs
such as the Coxeter-Dynkin graphs E7 and Dodd cannot appear as inclusion
graphs for higher relative com mutants coming from a sub factor and it pro-
vides an effective method of excluding graphs as possible principal graphs
for subfactors ([HaJ) (see [Ka] for a different method).
In the first chapter of this paper we study the basic construction for
reduced subfactors N pC pM p, pEN' n M a projection, which is probably
well-known to specialists. In particular, we get an algorithm which de-
scribes the higher relative commutants of the reduced subfactors in terms
of the higher relative commutants of N eM. This provides some informa-
tion on the principal graphs of N p C pM p.
The second chapter recalls the notion of a fusion algebra (or hypergroup)
and we describe how the explicit computations of the fusion algebras as-
Finite depth subfactors 177
sociated to various subfactors listed below are performed (the details can
be found in [Bi2]). Our method actually allows to compute fusion alge-
bras associated to bipartite graphs. We give then a list of fusion alge-
bras associated to subfactors including the fusion algebras of Haagerup's
(5 + v'13)/2 subfactor and the Goodman-de la Harpe-Jones 3 + v'3 subfac-
tor. Some other fusion algebras related to subfactors have been computed
previously. For instance, it is shown in ([GW]) that the fusion algebras for
Wess-Zumino-Witten models can be obtained from braid type subfactors
by replacing the basic construction by a construction using the shift on
the hraid elements (see [GW] for details, see also [GN]). Both construc-
tions coincide for the B, C, D subfactors ([We2]), Jones' subfactors ([Jol])
and more generally for selfconjugate subfactors. The computations for the
Hecke algebra subfactors ([WeI]) can be found in ([GW]). Various fusion
algebras are computed in ([11], [Ha], [SuI], [Su2]) and there is some overlap
between their calculations and our work.
Fusion algebras are of course abundant in physics, they appear for in-
stance iIi algebraic QFT where they describe the sector structure of two
dimensional QFT's (nonabelian fusion algebras may occur here) ([DHRl],
[DHR2],[Frl],[Fr2],[FRS],[Lol],[L02],[R],[RSD or in rational conformal field
theory ([MS],[Ve]) ("Verlinde algebra"), where they contain the informa-
tion on how often the primary field <l)k appears in the operator product
expansion of the fields <l)i x <l)j.
The classification of finite dimensional (abelian) fusion algebras is an out-
standing open problem and of particular interest in physics. We hope that
the explicit examples computed below from subfactors have some physical
interpretation.
2. The basic construction for reduced subfactors

Let N C M be an inclusion of III factors with finite Jones index [M :
N]. We denote by L2(M, trM) the closure of M in the Hilbert norm
IIxll2 = trM(x*x)1/2 induced by the unique trace trM on M. Let e~ :
L2(M, trM) -+ L2(N, trN) be the orthogonal projection and let
JM : L 2(M,trM) -+ L2(M,trM) be the canonical conjugation defined by
JM(X) = x*, xE M viewed as a vector in L 2 (M,trM). The algebra
MI = vN(M, e~) = {M, e~}, i.e. the von Neumann algebra generated
by M and e~ in B(L2(M, trM» is called the basic construction for N C M
([Jol]). We recall ([Jol]):
(1) e~xe~ = E~(x)e~, where EN : M -+ N is the unique trace
preserving conditional expectation from M onto N.
(2) N = {e~}' n M.
(3) e~ (M, e~}e~ = Ne~ ':::!. N.
(4) (M,e~) = JMN'JM.
(5) [JM, e~] = 0:
(6) There is a unique trace trMl on Ml such that trMl (xe~) = [M :
N]-ltrM(x) for all x E M.
178 D. Bisch
(7) M = spanN e~ N W ([PiPo1]).

We will use at various instances the following simple abstract characteri-
zation of the basic construction ([PiPo2]): Let N C M be an inclusion of III
factors with [M : N] < 00 and let P be a III factor containing M and a pro-
=
jection p such that [p, N] 0 and EL(p) =
[P : M]-11M =
[M : N]-11 M ,
then P is the basic construction for N eM, i.e. there is an isomorphism
from M1 onto P leaving M pointwise fixed and carrying e~ to p.
The following two lemmas can be found in [Bi1].
LEMMA 2.l.Let N C M be an extremal inclusion of Ih factors, [M :

N] < 00 and let N C M C M1 = (M, e~) be the basic construction.
Suppose we have projections pEN' n M and q E M' n M1 with trM(p) =
trM1(q) such that trMl(pqe~) = [M : N]-ltrM(p). Then
Npq C (pMp)q C pqM1pq (1)
is the basic construction for the pair (Npq C (pMp)q) ~ (Np C pMp).
Remark 2.2. Note that if NeAl is extremal, given a projection p as

in the lemma, the projection q := JMpJM E JM(N' n M)JM = M' n M1
satisfies all the required conditions.
LEMMA 2.3 Let N C M be III factors, [M : N] < 00, pEN a projection
and N C M C M1 C ... the basic construction. Then
pN p C pM p C pM1P C ...
is that basic construction for pN p C pA-I p.

We prove now a result, which allows us to describe explicitly the higher
relative commutants for an inclusion of reduced subfactors in terms of the
higher relative commutants of the original inclusion. As usual we denote
by N C M C M1 ... the iterated basic construction for N eM.
PROPOSITION 2.4.Let N C M be an extremal inclusion of Ih fac-

tors, [M : N] < 00, let q1 E N' n M be a projection. Then the ba-
sic construction for N q1 C q1M q1 is obtained in the following way: Set
Ao := N q1, A1 := q1M q1 and let Ao C A1 C A2 C ... be the ba-
sic construction. Then there exist projections q2, ... , qk with q2 E M' n
M 1, trM 1(q2) = trM(qd, trMl(q1q2e~1) = [M : N]-ltrM(qd, and qr E
=
(q1 ... qr-2 M r-2q1 ... qr-2)' n q1 ... qr-2 M r-1q1 ... qr-2 with trM r _ 1(qr)
trM r _ 3 (q1 ... qr-d, trM r _ 1(q1q2 ... qre~:=~) = trM(q1y-1[M : N]-l, 2 ::;
r ::; k, such that
c (qIq2MIqIq2)qa ... qk C ... C qIq2·· .qk M k-IqIq2 ... qk).

Proof. The proof pro cedes by induction using lemma 2.1 and remark
2.2.
We prove that (Ao CAl C A 2) ~ (NqIq2 C (qIMqI)q2 C qIq2MIqIq2):
Consider the basic construction N C M C MI and let q2 E M' n MI be
= =
such that trMl (q2) trM(ql), trMl (qIq2e~1) [M : N]-ltrM(ql) (remark
2.2). Then the claim follows from lemma 2.1.
For clarity of exposition let us also describe the next step of the in-
duction. We want to find Al C A2 C Aa. Since qi E M, qIq2 E
(qIMql)' n qlMIql and qlMql C qlMIql C qI M 2qI is the basic construc-
tion, we can apply lemma 2.1 with N +-+ qlMql, M +-+ qIMIql, p +-+ qIq2:
since qlM qi C qlMIql is clearly extremal (N C M is extremal iff Me MI
is extremal), there is a projection qa E (qIMIqI)' n qIM2qI such that
trq1M* (qa) = trq1M1ql (qIq2) (3)
trqlM2ql(qlq2qae:~~~~1) = tf"q 1M1ql(qlq2)[qIM l ql : qIMql]-1 (4)

Note that e:~~~~l = qIe~lq!, [qIMIql : qlMql] = [MI : M] = [M : N],
thus (3) ¢} (3)' : trM2(qa) = trM 1(qI)2 and (4) ¢} (4)': trM2(qIq2qae~1) =
trMl(qt}2[M: N]-I. Since (AI C A 2) ~ «q I MqI)q2 C qIq2MIqIq2), we get
by lemma 2.1
(AI C A2 C Aa) ~ «qIq2 M qIq2)qa C (qIq2 M IqIq2)qa C qIq2qaM2qIq2qa).

(5)
Note that (Ao C Al C A 2) ~ (NqIq2 C qI MqIq2 C qIq2MIqIq2) ~
(NqIq2qa C qIq2M qIq2qa
C qlq2Mlqlq2qa) since qa commutes with everything. This proves the claim
for k=3.
Now suppose the lemma is true for k. We want to find Ak-l C Ak C
Ak+1' Since we have projections ql E N' n M, q2 E M' n M I , ... ,
qk-I E (ql ... qk-aMk-aql ... qk-a)' n ql ... qk-aM k-2ql ... qk-a, qk E
(ql ... qk-2M k-2ql ... qk-2)' n qi ... qk-2Mk-Iql ... qk-2, one shows induc-
tively that qlq2 ... qr, 1 :5 r :5 k (Mo := M) are projections and that
qIq2 ... qk E (ql ... qk-I M k-2ql ... qk-I)' n qi ... qk-IMk-Iql ... qk-l. We
know that
qi ... qk-I M k-2ql ... qk-I C qi ... qk-lMk-lql ... qk-l C

C ql·· .qk-IMkql .. ·qk-I
is the basic construction (since ql ... qk-I E Mk-2). Apply now again
lemma 2.1 with N +-+ ql .. . qk-lMk-2ql ... qk-l,
M +-+ qi ... qk-IMk-Iql ... qk-I and p +-+ ql ... qk, i.e. let
qk+1 E (qi ... qk-IMk-Iql ... qk-I)' n ql ... qk-lMkql ... qk-I such that
180 D. Bisch
t ( Q l ... q"-l M"-lql ... q"-l)
rQ1 ... Q"_lM"Ql ... Q"_1 ql·· .qk+leQl ...Q"_lM,,_2Ql ... Q"_1 = (7)
= trQ1 "'Q"_l M"_lQl ...Q"_1 (ql ... qk)[M : N]-l.
It is an easy computation to show that
and
(7) ¢> (7)': trM,,(ql" .qk+le~:=~) = trM(qI)k[M: Ntl.
Lemma 2.1 says then that
(ql ... qk-l M k-2ql ... qk-l)qkqk+1 C (ql ... qkMk-lql ... qk)qk+1 C
C ql ... qk+lMkql ... qk+l (8)
is the basic construction. Since we have by induction hypothesis that
(Ak-l c Ak) ~ (ql .. . qk M k-2Ql .. ·qk C ql·· .qkMk-lql ... qk) (9)
~ (ql .. . qk M k-2ql ... qk C ql" .qkMk-lql.' .qk)qk+b
(the last isomorphism holds since qk+l E (ql ... qkMk-lql ... qk)'), the basic
construction Ak-l C Ak C Ak+1 is isomorphic to (8). This together with
(9) completes the induction and hence the proof. Q.E.D.
The proposition tells us precisely how the higher relative commutants

of the reduced inclusions can be obtained. The following corollary can be
easily read off its proof.
COROLLARY 2.5.Let N C M and Ao := Nql, Al := qlMql be as in

proposition 2.4. With the notation there we have:
(10)
and
(11)
1 ~ r ~ k.
Remark 2.6. The corollary implies in particular that reduced subfactors

N p C pMp, pEN' n M a projection, of a finite depth inclusion N C M
are again of finite depth. The above algorithm provides information on
the principal graphs of the reduced subfactors by looking at the Bratteli
diagrams for N eM. Note also that proposition 2.4 can be used to give
another proof of theorem 2.6 in [Bi1].
We would like to find a method that allows us to compute the standard
invariant of a reduced subfactor explicitly. In the next section we describe
how the principal graphs of reduced subfactors can be computed and give
examples where this computation has been carried out.
3. Fusion algebras associated to subfactors
We describe in this chapter some of our computations in [Bi2]. Let us

recall first the definition of a fusion algebra ([Fr],[GW],[MS],[Ve],[Sul]).
DEFINITION 3.1. Let I be a countable set with a distinguished element
oE I and assume I has an involution - : i E I - z E I. A fusion algebra A
is a unital associative algebra over &:: with generators ao = 1, ai, i E I, i :f. 0
and structure constants (the fusion constants) {Ni~ E NU {O}, i,j,k E I},
such that
(1) The generators ai, i E I form a linear basis for the &::-module A.
(2) ao is the unit for A.
(3) The structure constants {Ni~ E NU{O}, i,j, k E I} satisfy aiaj =
LkEI Ni~ak.
(4) ai := a. defines an involution on A.
(5) ao = aD = ao·
(6) Ni~ = 8i,j, for all i, j E I.
Clearly, these axioms impose several conditions on the structure con-

stants Ni~j' Let us collect these conditions in the following lemma.
LEMMA 3.2.The structure constants {Ni~ E N U {O}, i, j, k E I} of a

fusion algebra A satisfy the following conditions:
(1) (existence of the unit ao) N/'o =
Nti =
8f, i, k E I, where 8f denotes
as usual the Kronecker delta.
(2) (associativity of A) LrEI Nij N: k = LrEI NJkN/r, for all i, j, k, I E
I.
(3) (involution) Ni1 = Nf" for all i, j, k E I.
(4) Nli = Ni~ for all i,j,k E I.
Proof. The proof is a straightforward computation using the linear in-

dependence of the generators. (4) follows from (6) in definition 3.1 and
associativity. Q.E.D.
There is some special interest in abelian fusion algebras which are finite
dimensional &::-modules since these algebras appear in algebraic quantum
field theory and conformal field theory. In other words, we are interested in
=
fusion algebras A with generators ao 1, al, ... , an such that aiaj ajai, =
182 D. Bisch
for all i, j E I, which can equivalently be expressed in terms of the structure

constants as
(12)
If we consider the matrices Ni := ((Ni1)j,kEI), i E I, then No is clearly

the identity matrix and the Ni'S satisfy
i,j E I, (13)
if the fusion algebra is abelian (this relation follows from (2) in lemma
3.2 and commutativity of the fusion algebra). The matrices Ni, i E I,
are commonly called the matrices of fusion rules. Abelian fusion algebras
can be obtained from the representation theory of compact groups, where
I is the set of irreducible unitary representations of the group and the
Ni1 are the multiplicities with which the representation k appears in the
tensor product representation i ® j. The involution is given by taking the
contragredient of a representation and the trivial representation plays the
role of the distinguished element in I. Further examples of abelian fusion
algebras arise from the representation theory of quantum groups at roots
of unity, for instance if we consider the quantum group Uq (sl(2)) at q =
exp 27ri/n + 1, then I labels the irreducible representations in the truncated
Weyl chamber and the fusion constants are given by Nt; = 1, for Ii - jl <
k < min{i + j, 2(n + 1) - i - j} and N j j = 0 otherwise ([GW],[Re]).
k
We will now describe a simple method which allows to compute fu-

sion algebras associated to principal graphs which are obtained from an
inclusion of III factors N C M with finite Jones' index [M : N]. Let
N C M C Ml C M2 C ... be the Jones' tower of factors associated to
N C M ([Jo1]) and consider the two sequences of higher relative commu-
tants obtained from it, namely C = N' nNe N' n MeN' n Ml c ...
and C = M' n M C M' n Ml C M' n M2 C .... These inclusions of finite
dimensional C" -algebras are described by Bratteli diagrams which can be
recaptured from bipartite graphs with a distinguished vertex * correspond-
ing to the copy of C at the beginning of each Bratteli diagram. These
graphs are the principal graphs ([GHJ], [Jo1]) associated to the inclusion
N C M, the even vertices are by definition the ones with the same color as
the distinguished vertex, the other ones are the odd vertices. Equivalently,
the Bratteli diagrams can be described by two matrices with non-negative
integer entries A = (akl)kEK,IEL and B = (bk'I')k'EK',I'EL' with a distin-
guished index ko E K and kh E K' ([Pol]). The connection between both
descriptions is the following: K labels the even vertices of the principal
graph, ko is the label of the distinguished vertex *. akl gives the num-
ber of edges between even vertex k and odd vertex I. Similarly for Band
the second principal graph. Each even vertex of the principal graph for
C = M' n M C M' n Ml C M' n M2 C ... represents a simple summand
in the decomposition of the centralizer algebra M' n M 2r , which can be
viewed as the space of intertwiners of the M -M bimodules L2(Mr). If we
pick a minimal projection Pk, k E K in each simple summand of M' n M 2r
the M-M bimodules PkL2(Mr) will be irreducible. Different choices of

minimal projections in the same simple summand give unitarily equivalent
irreducible M-M bimodules. We can therefore think of the even vertices of
the principal graph as being labelled by irreducible M-M bimodules ([Ocl],
[Oc2]); we denote them by Qk, k E K (assume for simplicity that we start
with a finite depth inclusion N C M and that r is beyond the depth of
N eM). The index set K has an involution coming from the taking the
contragredient of an M-M bimodule. Furthermore, we can multiply two
irreducible M-M bimodules Qkl and Qk, in the obvious way, namely we can
form the tensor product M-M bimodule Qkl ®MQk" which decomposes into
a direct sum of M-M bimodules Qk, k E K, with certain (finite) multiplic-
ities (by properties of the basic construction). As in the case of irreducible
representations of a compact group as described above, these multiplicities
form matrices which generate a (not necessarily abelian) fusion algebra.
Furthermore, these matrices describe the graphs of the reducedsubfactors
M Pk C Pk Mr Pk, Pk E M' n Mr (see [Bi2] for details). A similar description
can be given for the odd vertices a" I E L, which represent the irreducible
M-N bimodules. For instance a minimal projection q, E M' n M 2r +1 gives
the irreducible M-N bimodule qI L2(Mr ), I E L. The natural contragredi-
ent map on these "odd" bimodules transforms them into N-M bimodules.
We can then form various bimodule tensor products such as Qkl ®M Qk"
kl' k2 E J( (M-M bimodule), Qk ®M a, a,
(M-N bimodule), ®M Qk (N-M
a'
bimodule), a'l ®N 2 (M-M bimodule) etc. and determine their decom-
position into irreducible M-M resp. M-N bimodules, i.e. determine the
fusion rules. The fusion algebra associated to a subfactor consists there-
fore of two "Zz-graded" fusion algebras, one associated to each principal
graph, which have to be compatible (this condition can be read off the
inclusions of the higher relative commutants in the commuting square). -
Let us describe the computations of the fusion algebra associated to one
principal graph. The even part (decomposition of tensor products of M-M
bimodules Qkl ®M Qk2 is generated by tensor powers of the M -M bimodule
ML<M1,trMI)M ==M L<Ml,trMI)M1®M1M1L(Ml,trMJM and one has to
decompose these tensor powers into irreducible M-M bimodules, which can
be done by solving simple matrix equations ([Bi2]). Similarly for the odd
part. However, if the principal graph has a triple vertex (or higher valency),
we may get an underdetermined system of equations even if we take the
consistency of the multiplication between even and odd part into account
(this imposes some additional conditions on the equations). For instance
in the example of Haagerup's (5 + V13)/2 subfactor we obtain an abelian
and a nonabelian fusion algebra for the second principal graph (see below)
in this way and only the condition of compatibility of this fusion algebra
with the fusion algebra associated to the first principal graph (see bekow)
excludes the abelian one and hence determines the whole fusion algebra as-
sociated to this subfactor. In the case of the principal graph of the 3 + J3
subfactor our method provides five possible fusion algebras associated to
this graph and it is at this point not clear which one is the one coming from
184 D. Bisch
the 3 + vasubfactor in ([GHJ]). We point out that our method of comput-

ing these fusion algebras can be applied to any bipartite graph, however
there may be no solutions. For instance for the Dynkin diagram E7, which
does not appear as principal graph from a subfactor, we can still compute
a fusion algebra for the even vertices. However there is no solution for the
odd part of the fusion algebra associated to this graph (and therefore E7
cannot appear as principal graph). Similar phenomena occur for the Dodd
graphs.
As mentioned above, not all fusion algebras arising from subfactors are
abelian (in the sense that the even part of the fusion algebra associated to
one principal graph need not be abelian), for instance if we consider the
inclusion Me M)cI G with G a finite group, then one (even) fusion algebra
will be just the group algebra. If the irreducible bimodules corresponding
to the even vertices of the principal graph are self-contragredient, then the
(even) fusion algebra will be abelian. Below we list some examples of fu-
sion algebras associated to subfactors. The multiplication tables contain
the above described bimodule tensor products. For the A, D, E subfactors,
these computations were done previously in ([11]), see also ([Su1], [Su2]).
Since they are rather implicit there, we list them here for the sake of com-
pleteness.
The most interesting fusion algebras are those associated to the principal
graphs of the 3 + va ([GHJ],[Ok]) and Haagerup's (5 + V13)/2 ([Ha]) sub-
factors since these fusion algebras should shed some light on the connection
between these subfactors and those coming from quantum groups via braid
group representations ([We1],[We2]) or the connection to conformal field
theories and algebraic QFT.
EXAMPLES OF FUSION ALGEBRAS 3.3.
1. The fusion algebra associated to the subfactor with principal graphs

A4 (the involution on the even part of the fusion algebra is trivial).
Principal graph:
I 0{
V\ (\ b
Multiplication table for even-even multiplication (this fusion algebra ap-
pears in RCFT ([Ve])):
A4
1 a
1 1 a
a a l+a
Graph of 0' (i.e. graph of the corresponding reduced subfactor):

a=
Multiplication tables for even-odd and odd-odd mulitplication:
a b
a
A4
a b
a 1+a a
b a 1
2. The fusion algebra associated to the sub factor with principal graphs
As (the involution on the even part of the fusion algebra is trivial).
Principal graph:
b
Multiplication tables for even-even multiplication:
As
1 a f3
1 1 a f3
a a 1+a+f3 f3
f3 {3 {3 1
Graphs (of reduced subfactors):
a=
/3= • •
Multiplication tables for even-odd and odd-odd multiplication:

186 D. Bisch
As
a b
a a+b a+b
{3 b a
As
a: b
a 1+01 a+{3
b a+{3 1+01
3. The fusion algebra associated to the subfactor with principal graphs

E6 (the involution on the even part of the fusion algebra is trivial).
Principal graph: I
b c
E6
1 a {3
1 1 a {3
a a 1 + 201 + {3 a
f3 f3 a 1
O!=
• •
Es
a b c
a a+b+c a+c a+b+c
f3 c b a
Es
a b c
a 1+0 0 0+13
b 0 1+13 0
c 0+13 0 1+0
4. Although there is no subfactor with principal graph E7, we still obtain

a fusion algebra (with trivial involution) associated to E7: Bipartite graph:
I il<. f! t"
Mu!t;p!;"t;on t ' b ! e V f \
E7
0 13 "I
0 1+0+13+"1 0+"1 0+13+"1
13 0+"1 1+"1 0+13
"I 0+13+"1 0+13 1+0+"1
5. The fusion algebra associated to the sub factor with principal graph
Es (the involution on the even part of the fusion algebra is trivial).
Principal graph:
q b c
Multiplication table for even-even multiplication:
Es
1 0 13 "I
1 1 0 13 "I
0 0 1+0+13 0+ 213 + "I 13
13 13 0+ 213 + "I 1 + 20 + 313 + "I 0+13
"I "I 13 0+13 1+"1
Q:'=
,t-I-~:f---"'l---i:
f3=
188 D. Bisch
"1=
• • • •
Es
a b c d
01 a+b a+b+c+d b+d b+c+d
(3 b+c+d a+ 2b+ c+ 2d a+b+c+d a+ 2b+c+d
"( d b+c b a+d
Es
a b c d
a 1+01 01+(3 (3 (3+"(
b 01+(3 1 + 01 + 2(3 + "( 01+(3+"( 01 + 2(3
c (3 01+(3+"( 1+(3 01+(3
d (3+"( 01 + 2(3 01+(3 1+01+(3+"(
6. Haagerup's 5+fI3 subfactor, the smallest irreducible finite depth

subfactor ofthe hyperfinite III factor with index above 4 (the fusion algebra
was also computed by Haagerup ([Hal)).
a) The fusion algebra associated to the first principal graph:
~ q b c ~
Multiplication table for even-even multiplication (trivial involution):
I~
1 01 (3 "(
1 1 01 (3 "(
01 P ~+0I+(3+"( 01+"( 01+(3+2"(
(3 ~ 01+"( 1+"( 01+(3+"(
"( ~ 01+(3+2"( 01+(3+"( + 201 + (3 + 2,,(
a=
{3=
:: II
r=
i~
a b c d
et a+b a + 2b + c+ d b+c b+d
{3 b a+b+c+d b b
'Y Ib+c+d a +3b+c+ d a+b+d a+b+c
~
a b c d
a l+et et+i3+'Y 'Y 'Y
b et+{3+'Y + 2et + (3 + 3'Y et+{3+'Y et+{3+'Y
c 'Y et+{3+'Y l+et 'Y
d 'Y et+{3+'Y 'Y l+et
b) The fusion algebra associated to the second principal graph:

I '" ,1 i & £
\.i. bed.
Multiplication table for even-even multiplication (the even part of this
fusion algebra is noncommutative and has the following involution: I = 1,
= = =
a a, i3 {3, 7 r, 8 = {, =
f 8).
I~
1 et f3 'Y 6 €
1 1 et {3 'Y 6 €
et et l+et+{3+'Y et+{3+'Y+€ et+f3+'Y+ 6 'Y f3
f3 {3 et+f3+'Y+ 6 l+et+f3+'Y et+f3+'Y+€ et 'Y
'Y h' et+f3+'Y+€ et+f3+'Y+ 6 l+et+f3+'Y f3 et
6 6 f3 'Y et € 1
€ € 'Y et f3 1 6
The graphs for a, {3, r, 8, { can be read off the multiplication table.
190 D. Bisch
~
a b c d
01 rz+ b a + 2b+ c+ d b+d b+c
f3 b+ c a+2b+c+ d a+b ~+d
'"I ~+d a+2b+c+ d b+c rz+ b
8 c b d a
f d b a c
~
a b c d
a 1 + 01 01 + f3+'"1 f3 + f '"1+ 8
b p+f3+'"1 ~ + 201 + 2f3 + 2'"1 + 8 + f p+f3+'"1 ~+f3+'"1
c f3+8 01 + f3+'"1 1+'"1 01 + f
d '"I + f 0I+f3+'"I 01 + 8 1 + f3
REMARK. As mentioned above, there is also an abelian fusion algebra

associated to this principal graph. However, this fusion algebra does not
appear in the subfactor setting.
6. Fusion algebras associated to the graph of the 3+v'3 subfactor ([GHJ],

[Ok)). The even parts of all five fusion algebras listed below are commu-
tative and have the same odd-odd and even-odd multiplication tables. At
this point we don't know which of these fusion algebras are the ones coming
from the Goodman-Jones-de la Harpe subfactor2 .
Principal graph:
t fl( 11 i ~
c
a) First table (a fusion algebra with trivial involution):
3+v3
01 f3 '"I 8 f
01 1+0I+f3+'"I+ 8 01+'"1+ 8 0I+f3+8 0I+f3+'"I+8+f 8
f3 01+'"1+ 8 + 2f3 + f 01+8 01+'"1+8 ~
'"I 01+.8+8 01+8 ~+2'"1+f 0I+f3+8 I"Y
8 0I+f3+'"I+8+f 01+'"1+8 0I+f3+c5 1+0I+f3+'"I+ 8 p
f 8 f3 '"I 01 1
b) Second table (a fusion algebra with involution a = a, i3 = 'Y, t = (3,

8 = 6, f = t").
2y. Kawahigashi showed in recent work that the first and fifth table are the correct
ones
3+73
a f3 "I 8 e
a ~+a+f3+-y+5 a+-y+5 ~+f3+5 a+f3+-y+5+e 5
f3 a+-y+5 a+-y+e 1+f3+5 a+f3+8 rr
"I a+f3+.5 1+f3+.5 a+-y+e a+-y+.5 16
8 a+f3+-y+8+e ia+f3+8 a+-y+8 1+a+f3+-y+8 ~
( 8 "I f3 a 1
c) Third table (a fusion algebra with involution a = a, /3 = /, r = j3,

8 = 8, (= E).
f3+73
a f3 "I 5 e
a Il+a+f3+-y+.5 a+-y+.5 a+f3+.5 a+f3+-y+.5+e .5
f3 a+-y+8 a+f3+{ 1+"1+ 8 a+f3+8 "I
"I a+f3+8 1+"1+ 8 a+f3+{ a+-y+8 lB
8 a+f3+-y+8+{ ia+f3+8 a+-y+.5 1+a+f3+-y+8 Ia
( 8 "I f3 a 1
d) Fourth table (a fusion algebra with trivial involution):
3+73
a f3 "I .5 (
a +a+f3+-y+8 a+-y+5 ~+f3+8 a+f3+-y+.5+{ .5

f3 a+-y+8 1+"1+ 8 a+f3+{ a+f3+8 h'
"I a+f3+8 a+f3+{ 1+"1+ 8 a+-y+8 lB
8 a+f3+-y+.5+e 1a+f3+.5 1a+-y+8 1+a+f3+-y+8 Ia
e 8 "I f3 a 1
e) Fifth table (a fusion algebra with trivial involution):
f3+73
a f3 "I .5 {
a +a+f3+-y+5 a+-y+5 1a+f3+5 a+f3+-y+5+{ 5

f3 a+-y+5 1+f3+5 a+-y+e a+f3+5 h'
"I a+f3+.5 a+-y+e 1+f3+.5 a+-y+5 16
8 a+f3+-y+.5+e 1a+f3+8 1a+-y+5 l+a+f3+-y+.5 Ia
{ .5 "I f3 a 1
The graphs can be read off the multiplication table.
The multiplication tables for even-odd and odd-odd multiplication (they

are compatible with all five even-even multiplication tables above):
3+V3
a b c
a a+b ~ + 3b + c
b+c
f3 b a + +
2b c b
"I b a+ 2b + c b
.5 b+c ~+3b+c f>+b
€ c b a
192 D. Bisch
3+...,(3
a b c
a 1+0 0+13+,),+0 0+(
b p+13+')'+o + 30 + 213 + 2')' + 30 + ( C}+13+')'+o
c 0+( 0+13+,),+0 1+0
References
[Bi1] D. BISCH, A note on intermediate subfactors, to appear in Pacific

Journal of Math ..
[Bi2] D. BISCH, Fusion algebras associated to subfactors, in preparation.
[B-N] J. BION-NADAL, An example of a subfactor of the hyperfinite lit
factor whose principal graph invariant is the Coxeter graph E 6 , in "Current
Topics in Operator Algebras", World Scientific Publishing, 1991, pp. 104-
113.
[DHR1] S. DOPLICHER, R. HAAG, J.E. ROBERTS, Local observables and
particle statistics I, Commun. Math. Phys. 23 (1971), 199-230.
[DHR2] S. DOPLICHER, R. HAAG, J.E. ROBERTS, Local observables and
particle statistics II, Commun. Math. Phys. 35 (1974),49-85.
[KE] D. EVANS, Y. KAWAHIGASHI, Orbifold subfactors from Heeke alge-
bras, preprint 1992.
[FRS] K. FREDENHAGEN, K.H. REHREN, B. SCHROER, Superselection
sectors with braid group statistics and exchange algebras: I. General theory,
Commun. Math. Phys. 125 (1989),201-226.
[Fr1] J. FROHLICH, Braid statistics in three dimensional local quantum
theory, preprint ETH-TH/89-36.
[Fr2] J. FROHLICH, Statistics of fields, the Yang-Baxter equation, and
the theory of knots and links, in "Nonperturbative quantum field theory",
Proceedings Cargese 1987, London, Plenum Press 1988, p. 7l.
[GHJ] F. GOODMAN, P. DE LA HARPE, V.F.R. JONES, Coxeter graphs
and towers of algebras, MSRI publications 14, Springer, 1989.
[GN] F. GOODMAN, T. NAKANISHI, Fusion algebras in integrable systems
in two dimensions, Physics Letters B 262 (1991),259-264.
[GW] F. GOODMAN, H. WENZL, Littlewood-Richardson coefficients for
Heeke algebras at roots of unity, Adv. in Math. 82 (1990), 244-265.
[Ha] U. HAAGERUP, in preparation.
[11] M. IZUMI, Application of fusion rules to classification of subfactors,
Publ. RIMS Kyoto Univ. 27 (1991),953-994.
[12] M. IZUMI, On flatness of the Coxeter graph E s , preprint 1992.
[Jo1] V.F.R. JONES, Index of subfactors, Invent. Math. 72 (1983), 1-25.
[Jo2] V.F.R. JONES, Heeke algebra representations of braid groups and
link polynomials, Ann. Math 126 (1987), 335-388.
[Ka] Y. KAWAHIGASHI, On flatness of Ocneanu's connections on the
Dynkin diagrams and classification of subfactors, preprint 1990.
[KY] H. KOSAKI, Y. YAMAGAMI, Irreducible bimodules associated with

crossed product algebras, preprint 1992.
[Lo1] R. LONGO, Index for subfactors and statistics of quantum fields I,
Commun. Math. Phys. 126 (1989),217-247.
[L02] R. LONGo, Index for subfactors and statistics of quantum fields II,
Correspondences, braid group statistics and Jones polynomial, Commun.
Math. Phys. 130 (1990), 285-309.
[MS] G. MOORE, N. SElBERG, Classical and quantum conformal field
theory, Commun. Math. Phys. 123 (1989), 177-254.
[Oc1] A. OCNEANU, Quantized groups string algebras and Galois theory
for operator algebras, in Operator Algebras and Applications 2, London
Math. Soc. Lect. Notes Series 136 (1988), 119-172.
[Oc2] A. OCNEANU, Quantum symmetry, differential geometry of finite
graphs and classification of subfactors, University of Tokyo Seminary Notes
45, 1991 (Notes recorded by Y. Kawahigashi).
[Ok] S. OKAMOTO, Invariants for subfactors arising from Coxeter graphs,
in "Current topics in operator algebras", World Scientific Publishing
Comp., 1991, p. 84-103.
[PiPo1] M. PIMSNER AND S. POPA, Entropy and index of subfactors,
Ann. Scient. Ec. Norm. Sup. 19 (1986),57-106.
[PiP02] M. PIMSNER AND S. POPA, Iterating the basic construction,
Trans. AMS 310, No. 1 (1988), 127-133.
[Pol] S. POPA, Classification of subfactors: reduction to commuting
squares, Invent. Math. 101 (1990), 19-43.
[P02] S. POPA, Sur la classification des sous-facteurs d'indice fini du facteur
hyperfini, C. R. Acad. Sci. Paris, Serie 1311 (1990), 95-100.
[P03] S. POPA, Classification of amenable subfactors and their automor-
phisms, preprint 1991.
[P04] S. POPA, Correspondences, Increst preprint 1986.
[R] K.-H. REHREN, Field operators for anyons and plektons, Commun.
Math. Phys. 145 (1992), 123-148.
[RS] K.H. REHREN, B. SCHROER, Einstein causality and Artin braids,
Nuclear Physics B 312 (1989), 715-750.
[Re] N. RESHETIKHIN, Quantized universal enveloping algebras, the Yang-
Baxter equation and invariants of links I,ll, LOMI preprint 1987.
[SuI] V.S. SUNDER, III factors, their bimodules and hypergroups, Trans.
of the AMS 330 (1992), 227-256.
[Su2] V.S. SUNDER, From hypergroups to subfactors, preprint 1992.
[SV] V.S. SUNDER, A.K. VIJAYARAN, On the non-occurrence of the Cox-
eter graphs E 7 , D 2n +1 as principal graphs of an inclusion of III factors,
preprint 1991.
[Vel E. VERLINDE, Fusion Rules and modular transformations in 2D con-
formal field theory, Nuclear Physics B300 FS22 (1988), 360-376.
[WeI] H. WENZL, Heeke algebras of type A and subfactors, Invent. Math.
92 (1988), 345-383.
194 D. Bisch
[We2] H.WENZL, Quantum groups and subfactors of type B, C and D,

Commun. Math. Phys. 133 (1990), 383-432.
Current Address: Dietmar Bisch, Department of Mathematics, UC Berke-

ley, Berkeley, CA 94720.
Universally bounded operators on
von Neumann algebras of type III
BY
ERIK CHRISTENSEN
MATEMATISK INSTITUT
K0BENHAVNS UNIVERSITET
DK - 2100 COPENHAGEN
DENMARK
Abstract. Continuous linear operators, on a von Neumann algebra M of

type Ih, with a faithful finite normal trace tr, are called universally bounded
if they are bounded with respect to the trace-norm too.
The algebra of all universally bounded operators has a natural structure
as a Banach *-algebra. As a consequence of Proposition 2.3 we get that this
algebra has a natural faithful representation on the Hilbert space L2(M, tr).
Proposition 2.9 show that the Haagerup tensorproduct Msa®hMsa can be
embedded into the algebra of universally bounded operators and Corollary
2.12 show that positive definite functions on discrete groups yield univer-
sally bounded operators on the group von Neumann algebra.
1. Introduction.
Von Neumann algebras of type III have a rich structure which make
them nice and tractable in many ways. On the other hand the theory of
index initiated by V. Jones show that even for the approximately finite
dimensional III factors, there are lots of open questions and lots of deep
connections to other branches of mathematics. In the past 7 years I have
been working together with Allan M. Sinclair, trying to understand some
problems in the Hochschild cohomology for von Neumann algebras. We
have had some success and have been able to prove that all the contin-
uous Hochschild cohomology groups for a von Neumann algebra M with
coefficients in M vanish, whenever M has no summand of type III, [4,5].
Although the structure for III von Neumann algebras is very nice, it does
not help in settling these sorts of questions. In order to try to compute
the Hochschild cohomology for III von Neumann Algebras it seems to be
necessary to have a better understanding of the linear operators on a von
Neumann algebra M of type III' When attacking this question, it becomes
clear that the finiteness of M is an important aspect, because it sometimes
makes it possible to extend an operator on M to a bounded operator on
L 2 (M, tr) where tr is a finite faithful trace in M. (We are only consider-
ing the factor case here.) One example of this interplay between operators
on M and operators on the underlying Hilbert space is found in the basic
construction used in index theory [3,9], which we will describe briefly. Let
M be a finite factor of type III in standard position on a Hilbert space H
and let ~ in H be a cyclic unit trace vector for M. If N is a von Neumann
196 E. Christensen
subalgebra of M then there is a conditional expectation 7rN of M onto N

which is determined by
(7rN(x)n~,~) = (xn~,O x E M,n EN.
Since w~ is a trace and N* = N ~ M we get
Let PN denote the orthogonal projection from M~ onto N1, then the above
equality shows that 7rN(X)~ = PNX~ so the operator 7rN : M -+ M can
naturally be extended to an operator on H. The reader may recall that
the basic construction consists in producing the von Neumann algebra Ml
which is generated by M and PN. In the cohomology questions we have
encountered operators ¢ : M -+ M such that ¢ is also continuous with
respect to the trace norm" IlIon M given by Ilmll! = tr((m*m)1/2).
Such an operator will be called universally bounded and it turns out that
a universally bounded operator also implements a bounded operator on H.
We have not been able to characterize universally bounded operators by
other properties, but we are quite sure that some better knowledge about
their properties and the properties of the algebra of all universally bounded
operators will turn out to be fruitful in the theory of continuous cohomology
for von Neumann algebras. An obvious question to which we so far have
only obtained trivial or no answers is the following: Suppose ¢ : M -+ M is
universally bounded and T : H -+ H is the corresponding operator on H.
What kind of conditions on ¢ will ensure that the von Neumann algebra
generated by M and Tq, is finite. Except for the conditional expectation
example already mentioned which sometimes yields a finite von Neumann
algebra we can get a finite algebra if a : M -+ M is an automorphism and
the fixed point algebra MG: has finite index in M.
The rest of the paper consists of some preliminary results and some ques-
tions on universally bounded operators.
2. Universally bounded operators on finite von Neumann alge-
bras.
Our aim is to get a better understanding of the spaces of continuous linear
operators on a von Neumann algebra M and to deepen our knowledge
of the possible properties the individual operators may have. Since this
area of research by no means has been overcrowded so far and very few
general results exist, we think that it is reasonable just to stick to the
simplest cases. Hence we will here only consider mappings on a finite factor,
say M, of type III with norm separable pre dual M*.A finite factor M
of type III, has a unique tracial state, tr, [8, ch.8]. The trace defines a
prehilbert space structure on M by (x,y) = tr(y*x) and the Hilbert space
completion is denoted L2(M, tr) in analogy with the commutative case
where M = LOO(X,Jl) and Jl is a probability measure. In the sequel we will
let H denote the Hilbert space L2(M, tr), then H contains M and as usual
we define an anti linear isometric involution J on H as the completion of
Universally bounded operators 197
the operator Jo on H which has M as domain of definition and is given

by Jom = m*. In order to distinguish between the element I in M(as I in
Loo (X, J.l) and the element I in H = L2 (M, tT) we will denote the latter
by the letter ~. This may seem superfluous, but the author finds it easier
this way. For an element m in M we will let m~ denote the corresponding
element in H. Following Segals extension of the theory of integration to the
noncommutative setting [1,14] we know that there is an algebra C of closed
densely defined operators affiliated with M which generalizes the algebra
of measurable functions on (X, J.l). The product and sum in C is defined
as the closure of the usual product and sum of unbounded operators. This
sum and product is denoted =f and 7. It is possible to extend the trace to
the positive operators in C such that tT : C+ ---. [0,00] and one then defines
the space Ll(M, tT) of integrable operators as those measurable operators
T for which tT((T*T)l/2) < 00. The integrable operators Ll(M, tT) form
a Banach space when equipped with the norm IITlll := tT((T*T)l/2). For
min M and T in Ll(M,tT),mT and Tm are in £l(M,tT) and we get
actually an isometry of £l(M, tT) onto M*, the predual of M, by
T ---. CPT : CPT(m) := tT(Tm) = tT(m7T).
Finally we want to recall that H can be thought of as L 2 (M, tT). For an

element x in H we will let L2 be defined on M~ by L2m~ := Jm* Jx, and
Lx is the closure of this operator. It turns out that for operators Lx, Ly
in L 2(M, tT) we get Lx:Ly is in Ll(M, tT), and any measurable operator
T, which is square integrable in the sense that T*:T belongs to Ll(M, tT),
must have the form T = Lx for some x in H. Having this identification
between Hand L 2(M, tT), we can define a norm II 112 on L 2(M, tT) by
IILxl12 = IIxll· We will end this tour into non commutative integration by
mentioning that inside the algebra of measurable operators we have these
inclusions
M ~ L 2(M, tT) ~ Ll(M, tT)
and the inequalities for m in M and x in H.
The separability assumption on M*, i.e. Ll(M, tT) turns H into a sep-
arable Hilbert space. In the rest of the paper we will let M, tT, L 2(M, tT),
Ll(M, tT) be present as the frame in which the definitions and results are
presented.
2.1 DEFINITION.
Let : M ---. M be a bounded operator. If is bounded with respect
to the norm 11111 also, then is said to be universally bounded, i.e.
3c> 0 s.t. 1I(m)lll ::; cllmlll for all m in M.
If : M ---. M is a universally bounded operator then can be extended

to Ll(M, tT) by continuity since M is dense in Ll(M, tT) with respect to
198 E. Christensen
the norm Illh. The extended operator will be denoted CI> too, so whenever
we talk about a universally bounded operator CI> on M we mean automati-
cally the restriction to M of a continuous operator CI> on Ll(M, tr) which
maps Minto M and is continuous with respect to 111100 on M. It is now
clear that a universally bounded operator CI> naturally has 2 norms associ-
ated to it namely the norms 1ICI>111 and 1ICI>1I00 corresponding to the norms
as operator on the spaces Ll(M, tr) and M respectively.
Since Ll(M, tr) can be identified with the predual of M we can - for a
universally bounded operator CI> : Ll(M, tr) ~ £1(M, tr) define the adjoint
operator CI>* : M ~ M by
'IT E Ll(M, tr), '1m EM: tr(CI>*(m)T) = tr(mCI>(T*)*). (1)
2.2 LEMMA.
If CI> : M ~ M is universally bounded then so is CI>* and
PROOF: Let us define 'l1(T) = CI>(T*)*. Then 'l1 is universally bounded with
1I'l11100 = 1ICI>1100 and 1I'l1l1l = 1ICI>lh. Furthermore CI>* = 'l1 t , the transposed of
'l1. It is clear from the way the duality Ll(M, tr)* = M is obtained, that
CI>* = 'l1t : M ~ M and 1ICI>*1I00 = 11'l1ll l . In order to estimate 11CI>*(m)lIl we
have to consider the relation (1) but only for T in the unitball (111100) of M.
From (1) we then find 11CI>*(m)lh ::; 1ICI>lloo IImll l . Hence CI>* is universally
bounded and 1ICI>*lh = 1ICI>1I00·
The non commutative integration theory also introduces the spaces
LP(M, tr) for 1 < p < 00, by defining a measurable operator T to be in
LP(M, tr) iftr((T*T)p/2) < 00. It turns out that for 1 ::; p < 00, LP(M, tr)
is a Banach space when equipped with the norm IITllp = tr((T*T)p/2).
2.3 PROPOSITION.
Let CI> : M ~ M be a universally bounded operator and let 1 to LP is continuous with respect to the norm II lip·
Let 1ICI>lIp denote the norm of this operator in LP(M, tr) then
PROOF: . Take an m in M, let m = vh and CI>(m) = wk be the polar

decompositions of m and CI>(m) and choose abelian von Neumann subalge-
bras A, B in M such that h is in A and k is in B. The restriction of tr
to A induces a regular probability measure say p, on the spectrum .A of A.
Similarly v is obtained as a probability measure on the spectrum of B. It is
a well known fact that there exists a conditional expectation 1C' of M onto
B and that 1C' is given by the relations
1C'(X) E B tr(bx) = tr(b1C'(x)) , x E M,b E B.

We can now construct an operator T from A = £00(.4, /-L) into B = £00(13, v)

by defining
T(a) = 7T(w*<1I(va)).
Some easy computations show that for 1 :5 P :5 00 Ilmlip = IIhllp, II<1I(m)IIp =
IIkilp and T(h) = k.
It is clear that IIT(a)IIoo :5 II<1IIIooliali oo so IITiloo :5 11<111100. The relation
(*) yields 117T(X)IIl = sup{tr(b7T(x)) I b E B, IIbil oo :5 1} :5 IIxII! and hence
IIT(a)III :5 II<1I(va)lh :5 II<1IIIlllalh so IITIII :5 11<11111. We can now apply the
Riesz-Thorin Theorem [6, vol II p.142] with Po = 1, PI = 1, qo = 00, ql = 00,
in order to obtain
Hence II<1I(m)IIp :5 II<1III~/PII<1lll~I/Plimilp and the proposition follows:

2.4 REMARK.
The case P = 2 yields 11<11112 :5 II<1III~/2 II<1III~2 so <11 can be considered an
element in B(£2(M,tr)). It should be noted that the restriction of <11* to
£2(M, tr) is the usual adjoint of the restriction of <11.
As the following example shows there are no natural relations between
universally boundedness and complete boundedness.
2.5 EXAMPLES.
Let f be a singular state on M then the mapping M 3 m ~ f(m)IM
is completely positive (and hence also completely bounded). Since f is
singular we can find a sequence of projections (en)nE from M such that
en ! 0, tr(e n ) :5 2- n and f(e n ) = 1. Hence IIenill :5 2- n and II<1I(en)III = 1
so <11 is not universally bounded.
If <11 : M -+ M is an anti automorphism, then <11 is not completely
bounded but <11 is an isometry for both II 1100 and II III so <11 is universally
bounded.
2.6 DEFINITION.
The linear space of all universally bounded operators on M is denoted
UB(M).
For a universally bounded operator <11 we define
Having Lemma 2.2 we get immediately:

2.7 PROPOSITION.
Equipped with the usual composition of mappings as product, the invo-
lution * and the norm II IIub, U B(M) becomes a Banach *-algebra with an
isometric involution.
The extension of the operators in U B(M) to bounded operators on
£2(M, tr) is a self-adjoint faithful bounded representation which we call
200 E. Christensen
the trivial representation. The C* -algebra generated by the image of the

trivial representation is denoted C~b(M). For operators a and m in M the
mappings m -+ am and m -+ ma are in U B(M), and when these operators
are extended to L2(M, tr) we get a and Ja* J respectively. Hence C~b(M)
contains M as well as the commutant M' = JMJ. In particular C~b(M)
is irreducible on L2(M, tr). Left and right multiplication by elements from
M will then generate a very "big" subalgebra inside UB(M) so we will
study the operator rab in U B(M) which is determined by a pair of ele-
ments a, b from M by rab(m) = ambo Clearly rab is in U B(M) and the
computation below show that Crab)* = ra-b-·
trCr~b(m)n) = tr(m(an*b)*) = tr(mb*na*) = trCra*b*(m)n). (2)
Since we are in the situation where M is a factor, we get the following

result.
2.8 LEMMA.
Let a, b be elements in M, then
PROOF:
Let f > O. Let e be a spectral projection for a*a such that a*ae ;?:
lIall~ (1 - f)e and let f be a spectral projection for bb* such that fbb* ;?:
IIbll~ (1 - f)f. Since M is a factor either e »- f or f »- e. Now, either way
makes no difference for the argument so we may as well assume that f »- e
and that v is a partial isometry in M such that v*v :S f and vv* = e. We
then get
lIrab(v)lloo = lIavbb*v*a*II~2 ;?: (1- f)l/2I1bII00Ilaea*II~2

= (1- f)l/2I1bIl001Iea*aell~2 ;?: (1- f)llbllooliali oo .
Hence lIrablioo = lIalloollbll oo. By the previous remarks lIrablll = Ilra-b-lioo =
lIa* 1100 Ilb* 1100 = Iialloo Ilblloo.
The result in Lemma 2.8 does not carryover to sums of elementary
operators. Suppose for instance that {Vi I i = 1, ... ,n} is a set of partial
isometrics such that ViVi = ViVl and vivivjvj = DijVjvj then the operator
<P on M given by <P = L~=l rv,v: satisfies
n
11<p1100 = 1, <p* = L rv:v, , ,1I<p11I = 11<p*1100 = n, and
i=l
11<p*lll = 11<p1100 = 1.
The example is apparently based on the orthogonality between vivi and
Vjvj. If we only look at linear combinations of rhk when hand k are self-
adjoint operators from M we find that we have a manageable structure.
In order to describe our result we remind the reader of the Haagerup ten-
sor product [2,12J. Let Msa denote the real vector space of self-adjoint
operators in M. The real algebraic tensor product Msa 18> Msa can be
equipped with the Haagerup norm through the following definition. For x
in Msa 18> Msa we define:
The completion of Msa 18>h Msa with respect to this norm is called the
Haagerup tensor product and denoted Msa 18>h Msa·
We will define a mapping r : Msa 18>h Msa -+ UB(M) by first defining r
on Msa 18> Msa and then extending r by continuity. For x = E~=l ai 18> bi
we define
n n
r(L ai 18> bi) = L "Ya;b;, ai, bi E Msa· (3)
;=1 i=l
For m in M and vectors 17, pin L2(M, tT) we get
I (~"Ya;b;(m)17'P) I ::; ~ IImll oo llbi17l1l1 a ipll

::; IImIl00(~b?17,17)1/2(~a?p,p?/2.
so
n n
II L "Ya;b; 1100 ::; II L ai 18> bill h
i=l i=l
since II E~=l "Ya;b; 111 = II E~=l "Y~;b; 1100 = II E~=l "Ya;b; 1100 we have
n n
IIr<L ai 18> bi)lIub ::; II L ai 18> bill h .
i=l i=1
The operator r can now be extended to Msa 18>h Msa. The extension is
also denoted r and we have proved the following proposition.
2.9 PROPOSITION.
There exists a unique contraction r of Msa 18>h Msa into U B(M) such
that for all a, bin M sa , r(a 18> b) = "Yab.
It would be interesting to know whether r is injective on Msa 18>h MBa'
We have not been able to prove this, but we think there might be a good
chance, that the following lemma can be extended.
2.10 LEMMA.
The mapping r : Msa 18> Msa -+ U B(M) is injective on the algebraic
tensor product.
PROOF:
202 E. Christensen
Let x = r::l ai ® bi be in Msa ® Msa and suppose r::l aimbi = 0 for

n
all m in M. We can then consider the Hilbert space K = EB H and let the
i=l
algebra Mn(M) of n x n matrices act on K in the canonical way. Let A, B
be the column operators in Mn(M) given by
A(O"i) = (aWl), B(Pi) = (biPl), for (O"i), (Pi) in K,
and let M denote the diagonal operators given by
mE M m(O"i) = (mO"i).
Finally we define P to be the orthogonal projection of norm one from K
onto the closed linear span of the set {MAK}. It follows that P is a
projection in the commutant of M, so P is in Mn(M') on K. On the
other hand M ~ Mn(M) and A E Mn(M) so P is a supremum of range
projections for operators in Mn(M).
In particular P is a projection in Mn(M) as well so since M is a factor
P is a projection in Mn( /), say P = (Pij) where Pij E . By construction
PA = A and A*P = A*. The assumption f(x) = 0 implies that for all m
in M and any vectors O",p in K(A*mBp,O") = O. This means that PB = O.
Now let P denote the scalar nxn matrix obtained from P by element wise
conjugation. P ij = Pij. Since ai, bi E Msa we get P A = A A * P = A * and
P B = 0 so with R = 1/2(P + P) we have a symmetric scalar matrix with
real entries such that A * R = A * and RB = 0, and we get
n
L ai ® b = L (r:
i j rji aj) ® bi = L L aj ® (rjibi) = o.
i=l j
The lemma follows.

lt seems possible that the result is valid for infinite sequences (ad, (b i ) as
well and hence that f is injective on Msa ®h Msa. On the other hand the
change of the order of summation in the argument could not be justified
for infinite sums.
We will end by showing yet another way to obtain universally bounded
operators. According to Haagerups article [7] on the metric approximation
property for the C*-algebra generated by the left regular representation of
a free group we have the following result, [7, Lemma 1.1]:
2.11 LEMMA.
Let G be a discrete group for which the nontrivial conjugacy classes are
infinite (ICC) and let ¢ be a positive definite function on G. There is a
unique completely positive map Mq, : C~(G) --+ C~(G) such that
Mq,>.(s) = ¢(s)>.(s).
The norm of this map is IIMq,11 = ¢(e).
Let now G be a discrete ICC group and let M denote the von Neumann
algebra generated by the left regular representation of G then M is a finite
factor of type Ill. Using the terminology of Lemma 2.11 we can get more
examples of universally bounded operators.
2.12 COROLLARY.
Let <fJ be a positive definite function on G. There is a unique universally
bounded operator M", on M such that
M",>'(s) = <fJ(s)>.(s).
The norm IIM",llub = <fJ(e).

PROOF:
The lemma tells us that M", is bounded on C~(G) with respect to the
norm 111100 on C~(G). Let I,g denote functions on G with finite support
and let x,Y denote the operators x = ~I(s)>'(s), y = ~g(s)>'(s). We have
the following general rules
IIyll1 = sup{1 1 * g(e) Illxll oo :::; I},

IITyl11 = sup{1 1 * (<fJg)(e) IlIxll oo :::; I}.
Let 't/J : G --+ be given by 't/J( s) = <fJ( S-1 ), then 't/J is positive definite, so
by the lemma IIM1/Jxll oo :::; 't/J(e) Ilxlioo' We can then estimate IITyll 1 and we
get for an arbitrary x = ~I(s)>'(s) with Ilxlloo :::; 1.
11* (<fJg)(e) I =1 LI(s)<fJ(s-l)g(s-l) I

=1 L't/J(s)l(s)g(S-l) I
s
Hence M", is continuous for the norm 11111 as well on the functions with
finite support and the corollary follows.
REFERENCES
[1]. S.K. Berberian, Regular ring of a finite AW*-algebra, Ann. Math. 65 (1957),
224-239.
[2]. A. Chatterjee, A.M. Sinclair, An isometry from the Haagerup tensor product into
completely bounded operators, J. Operator Theory (to appear).
[3]. E. Christensen, Subalgebras of s finite algebra, Math.Ann. 243 (1979), 17-29.
[4]. E. Christensen, E.G. Effros, A.M. Sinclair, Completely bounded multilinear maps
and C*-algebraic cohomology, Invent.Math. 90 (1987), 279-296.
[5]. E. Christensen, A.M. Sinclair, On the Hochschild cohomology for von Neumann
algebras, Preprint, Copenhagen.
[6]. R.E. Edwards, "Fourier Series," Holt, Rinehart and Winston Inc., New York, 1967.
[7]. U. Haagerup, An example of a non nuclear C*-algebra which has the metric ap-
proximation property, Invent. Math. 50 (1979), 279-293.
[8]. R.V. Kadison, J.R. Ringrose, "Fundamentals of the theory of operator algebras,"
Academic Press, Orlando, 1986.
[9]. V.F.R. Jones, Index for subfactors, Invent.Math. 72 (1983), 1-25.
[10]. V. Paulsen, Completely bounded maps and dilations, Pitman Research Notes in
Math. 146; Longman Sci.& Tech., Harlow UK, (1986).
204 E. Christensen
[11). S. Popa, On a problem by R. V. Kadison on maximal abelian *-subalgebras in

factors, Invent. Math. 65 (1981), 269--281.
[12). F. Pop, R.R. Smith, Schur product and completely bounded maps on finite von
Neumann algebras, Preprint, Texas A&M ..
[13). F. Pop, R.R. Smith, Cohomology for certain finite factors, Preprint Texas A&M.
[14). I.E. Segal, A non commutative extension of abstract integration, Ann. Math. 57
(1953), 401--457.
Non occurrence of star graphs as principal graphs
SaidA. Rida
School of Mathematics, The University of New South Wales

Kensington NSW 2033, Australia
Introduction
Let N C M be an irreducible, finite depth, finite index inclusion of hyperfi-
nite lIt-factors. The question of which graphs occur as principal graphs has
been settled by various authors when the index is ~ 4 and is still open for
the remaining values. In an unpublished work, Ha.a.gerup proved that the
smallest irreducible, finite depth subfactor of the hyperfinite lIt -factor with
index above 4 has index (5 + .[13)/2 and principal graph T(4,4,4) (cf. def
1.3). Independantly from Haagerup's result, we use the gaps in Popa's set
A(M,N) (c!. [4]) to prove that star graphs do not occur as principal graphs
when the index is in (4,2 + V5).
1 Preliminaries
In [3], Jones defined the polynomials Pn{x) recursively as follows:
Po == 1 nEN
Define for n = 1,2, ... ,00 the function:
An : [0,1/4] --+ R
by
These functions will be the main tool in this paper. The next lemma
outlines some of their properties.
lemma 1.1 a) An(T) ~ 1/2 \In E N \IT E [0,1/4]

b) An is strictly increasing for every n = 1,2, ... ,00.
c) The sequence (An(T»n~t is strictly increasing and Aoo(T) = limn An{T)
206 s. A. Rida
for every T E [0,1/4]

d) The sequence (An+I(T)-An(T))n>l is strictly decreasing and the sequence
(An(T)/An+I(T))n~l is strictly increasing in n.
theorem 1.2 ([4]) Let N c M be an inclusion of type Ill-factors with fi-

nite index [M : N] ~ 4. Let T = [M : Ntl ,t = Aoo(T) and define:
A(M,N) = {o I EN(e) = 0.1 for some projection e in M}

We have:
A(M,N) n [O,t] = {OJ U {TPk(T)/Pk+l(T) , k ~ O} U {t}

A(M, N) n [1- t, 1] = {I - t} U {Pk+I(T)/ Pk(T) , k ~ I} U {I}
definition 1.3 ([1]) An n-star-graph T(Pt,P2, ... ,Pn), Pl ~ P2 ~ ... ~

Pn ~ 00 is a graph with the following properties:
1) There exists one and only one vertex of degree n, and every other vertex
has degree ~ 2.
2) It contains no cycles.
3) It has n branches and the ith branch of the graph has Pi - 1 vertices.
An example of such graph is T(2, 2, 3, 4) given below:

o
1
Recall that the norm of a graph is defined to be the operator norm of its
adjacency matrix. ([1])
theorem 1.4 Let r = T(pt.P2, ... ,Pn) where Pi = 2,3,···,00 , T =11

r 11- 2 Then T is the only solution in the interval [O,lJ of the equation
•
API ( T) + An ( T) + ... + APn ( T) 1.
=
Proof
cf [6].
definition 1.5 ([1]) An H-graph H(p, q, r, s; I) is a graph with the following

properties:
Non occurrence of star graphs 207
1) There exist two and only two vertices of degree 3, and all other vertices
have degree :5 2.
2) It contains no cycles.
3) The four free branches of the graph have p-1, q-1, r-1 and s-l vertices.
4) There are 1-1 vertices between the two degree 3 vertices.
(We will usually assume p :5 q , r :5 s , q:5 s.)
Note that this definition differs from that given in [1] in terms of the number
of vertices between the degree 3 vertices.
An example of such graph is H(2, 3, 3, 4; 3):
2 Bratteli diagrams and finite depth

Let N C M be finite dimensional von Neumann algebras, let (Pi)i=l, ... ,n
(resp. (qj )j=l, ... ,m ) be the minimal central projections in N (resp. M ).
The inclusion matrix of N in M is the n x m matrix A = (aij) where aij is
the number of simple components of a simple Mqj-module viewed as a N pi -
module. The normalized trace tr on M (resp. N) is given by the column
vecto, f = ( :~ ) (....p. if = ( :~ )) whe,e t; (,..p. ';) is the ',ore of a
minimal projection in M qj (resp. N pi ).

Let m~. = dim M qj n~i = dim N pi ' The vectors iii = (m q1 , ••• , m qm )
and ii = (n p1 , " " npn ) are called dimension vectors. If we assume that
the unit of M belongs to N, then the trace vectors, dimension vectors and
inclusion matrix are related by the following formulae:
A.t=s
208 s. A. Rida
n
where At is the transpose of A. We will often gather these data in a diagram
called the Bratteli diagram as in the following example:
M~M3(C)$C N~ {( ~ ~ $6 a,6EC}"C$C
we have n= ( ~) m= ( ~ ) s = ( :: ) r = ( !: )
A=(~ ~)
3, tl 1, t2
and we will adopt the convention that white vertices correspond to the
smaller algebra and black vertices to the bigger one.
Let N C M be an inclusion of III-factors with finite index, N C M C
MI C M2 C ... the Jones' tower of subfactors and ... C N3 C N2 C NI C
N C M a choice of the tunnel (cf [4]). Consider the infinite Bratteli diagram
corresponding to the chain of inclusions
C = N' nNe Ni nNe N~ nNe· ..
By theorem 4.6.2 [l],the Bratteli diagram of N{. nNe N{.+l n N is obtained
by a reflection of that of N{._l nNe N{. n N and a possible addition of
edges and vertices. The graph r remaining after deleting the parts corre-
sponding to the reflection on each level is called the principal graph and the
distinguished vertex on the first level (the one corresponding to N' n N) is
denoted *. (This is the same as ko in Popa's standard matrix in [5, page 20]).
remark 2.1 Theorem 4.6.2 [1} treats the chain of higher relative commu-
tants (N' n Mi)i, but we know from [5) that N' n Mi is antiisomorphic to
NIH n N, which justifies the previous paragraph.
When r is finite, we say that N C M has finite depth, and in that case, we
have [M : N] =11 r 112 (cf also [1]).
The next proposition gives a method of computing values in A(M,N)
and will be crucial in proving the main result of this paper.
proposition 2.2 Let N c M be a finite index, finite depth inclusion of

type I It -factors, Nand M finite dimensional algebras such that the following
square:
M C M
u U
N C N
is a commuting square (cf [1] for example). Let (Pi)i=t, ... ,n,(qj)j=t,... ,m,s,r
=
and A (aij) be as in the beginning of this section, h = =
{j 1, ... , mlaij =I
O} and denote by EN the normalized trace preserving conditional expectation
from M onto N. Define
Api(M, N) = {a E [0,1]1 EN(e) = api for some projection e in M}
We have:
{ "L...J -
(jtj. ' (j E {0,1, .. . ,aij} } C Api(M,N) C A(M,N)
- -
jEJi s.
Proof
For j E Ji, we have Piqj = rt + r2 + ... + r Oij
where rk is a projection in M qj with rk.r, = Okl.rk and tr(rk) = ni.tj
= =
Claim: set a tj/Si, then EN(rk) api, 1 ~ k ~ aij
Proof of claim:
Let x E N pi , then x = Eq,.x
EN(rk) = api {:} rk - api ..L Npi(with respect to tr)
=
{:} tr« rk - api).x) 0 '<Ix E Npi
tr«rk - api).x) = tr«rk - api). E, q,.x)
=tr(rkxqj - a E,Piq/X
=Tr(x).tj - aTr(x) E, ai,.t,
= Tr(x)[tj - aSi] = 0 since Si = E, ai,.t,
Where Tr is the usual trace of matrices.
By taking different combinations of the rk's, we get projections that expect
on (jtj/Si, (j E {O,l, ... ,aij} since the rk's are mutually orthogonal.
Taking projections from different M qj 's, j E Ji yields the proposition.
3 Perron-Frobenius eigenvectors
In this and the next sections, we will write An and Pn instead of An(r) and
Pn(r) whenever no ambiguity is possible.
210 S. A. Rida
The calculations in the following lemma are due to Schou/Haagerup [7]. We

state their result in a form compatible with our notations.
lemma 3.1 Let Nand M be as in the beginning of section 2. Assume that

the Brattdi diagram is given by an n-star graph r = T(pt,P2,'" ,Pn) where
the multiple vertex is white (i.e. corresponds to N) and that s (resp. i)
is a Perron-Frobenius eigenvector for AAt (resp. At A) with the eigenvalue
T- 1 =\1 A \1 2 • Let ri = [Pi;-l] , [Ii = Pf] (the bracket stands for the integer
part) and (sii, ... ,sm (resp. (tii, ... ,tfti») the components of s (resp. i)
corresponding to the p~h - branch. Denote by S~i the component of the mul-
tiple vertex. We have:
a) if Pi = 21 + 1
Pi _ 1-;+1 P. . Pi/PPi- 1 1 ::; j ::; ri
{ s; - T . • 2)-2. S 0
t ;Pi = T 1-)+1 •p.2;-I.S0Pi/Pp;-1 1 ::; j ::; Ii
b) if Pi = 21
SPi - T'-; . p..
{ ; - 2)-1' 0
sPifP
Pi-1
= A2j+l..' ...J·• Apj SoPi 1 ::; j ::; ri
t J~i -- T ' -;+1 • P.2·3-·
2 SPi/P .
0 p,-l =
A2J.· ... Api SPi
T' J 0 1 ::; j ::; Ii
c) (S}i); and (t}i); are increasing inj.
lemma 3.2 Let N c M be an inclusion of type I II -factors whose principal

graph is a finite 3-star graph. Then the * vertex (cf. section 2) is the one
at the end of the longest branch.
Proof
Let r = T(p,q,r) , p::; q::; r < 00 and assume without loss of generality
that q < r. Since N' n M = C, the * vertex is an end vertex and by theo-
rem 3.8 [5], * should have the smallest component of the Perron-Frobenius
eigenvector. Give r a bicoloration so that * is white. We have to consider
four cases depending on whether the multiple vertex is black or white and
whether q is even or odd. Let's assume that the multiple vertex is white
and that q is odd, the remaining cases being essentially similar. Let S., Sm
and Sq be the components of the Perron-Frobenius eigenvector correspond-
ing respectively to *, the multiple vertex and the outmost vertex on the
q-branch (by outmost we mean away from the multiple vertex). We want to
prove that s. is the smallest component ofthe Perron-Frobenius eigenvector
and since the components are increasing on each branch by lemma 3.1 c),
it suffices to compare s. to Sq. Since the 3 above mentioned vertices are all
white by assumption, rand q are odd.
Let r = 21 + 1 , q = 2m + 1 , m 1
by lemma 1.1 a).
4 Non occurence of star graphs

Let N C M be a finite depth, irreducible inclusion of hyperfinite IIl-factors
with index [M: N] E (4,2+ VS). Let ... C N3 C N2 C Nl C N C M be
a choice of the tunnel for which Ui( Nt n M) W = M and Ui( Nt n N) W = N
(such a tunnel always exists when N C M has finite depth (c!. [5])).
In what follows, we construct the Bratelli diagram of the two chains (NL n
N)" and (NLnM)" and study the "horizontal" inclusions NLnN c NLnM.
Recall that [x] denotes the integer part of x E R and En the parity
function on N, i.e. En = 0 if n is even, En = 1 if n is odd.
lemma 4.1 Let N c M be an irreducible inclusion of hyperfinite III -factors
with principal graph r = T(p, q, r) where p ::; q ::; r < 00, and let T = [M :
Ntl. In what follows, Q~ 's are some matrix algebras, and t~ is the trace
of a minimal projection in Qj. We have the following:
1) If k :5 r - 1 and k' = [k/2], then:
i) NL n N ~ ej~~l Qj.
ii) Q~'+1 = C.
"'Jt"j = T "'-j+lp.2j-2+£".
In
2) i) N; n N ~ ej~2 Qj where r' = [r/2].
ii) Q~'+l = Q~'+2 = C.
"'Jtir = Tr'-j+lp.2j-2+£r ,J. <'
UI _ r, t rr'+l = I\q
\ p.r-l! t r'+2
r
= I\p\ p.r-l'
3) i) N;+1 n N ~ ej:t 2Qj+1, where r" = [( r + 1)/2].
II"J Qr+l
r"+l = r"+2 = C .
Qr+l
212 S. A. Rida
")t j +
In
r 1 = T r"-j+lp.2j-2+<r+l' J. <
_ r" r+1
, t r"+l = AqAq_l
\ \ P.r-l! t rr"+2
+1 =
ApAp-lPr-l.
Note that Q~t~l (resp. Q~t~2) is 0 if p = 2 (resp. q=2).
Proof
The decomposition of the algebras Nk n N is a consequence of theorem 4.6.2
[1].
1) We have seen that the * vertex is at the end of the longest branch (lemma
3.2). It follows that up to k = r - 1, the Bratteli diagram behaves in the
same way as the one in [3, p. 19]. The computation of the traces follows
the same technique.
2) The inclusion N:_ 1 nNe N: nN is given by the following sub diagram
of T(p, q, r):
t~'+1 t~'+2
Since the traces on the Bratteli diagram get multiplied by T down every
vertical column, (theorem 4.6.2 i)[I]) the trace on N: n N is given by a
multiple of the Perron-Frobenius eigenvector of T(p, q, r).
Also, 8 m = tf(~~1)/21+1 = Pr-l by i), and the traces on the long branch of
the above diagram are given by lemma 3.1 a), i.e.
t jr = T r'-j+lp.2j-2+<r 1 ~ j ~ r'
t~'+1 and t~'+2 correspond to vertices on the p and q branches and lemma
3.1 gives:
P.
r
t r '+2 = -p:-.8
TPq- 1
q
m
\
= Aq. r-l
3) The diagram of N: nNe N:+1 n N is as follows:

N:nN
N:+1 nN A/"'~ T p. t r+1 t r+1

The (tj+1 )j~r' are obtained in a similar fashion as in 2). r-l r"+1 r"+2
By lemma 3.1, we have
p. \ \ P.
r+1
tr"+l = -Aq_lAq
- - T r - l = Aq_lAq r-l
T
and similarly t~t;2 = Ap - 1Ap Pr -l'

Next is a copy of the previous lemma where the principal graph is an H-
graph. The L~'s are some matrix algebras and t~ is the trace of a minimal
projection on L~.
lemma 4.2 Let N c M be an irTeducible inclusion of h1lperfinite III -factors

with principal graph r = H(2,q,2,sjl), and let T = [M : Nt1, {J =
1 - T - As. We have the following:
1) If k ~ s - 1 and k' = [k/2], then:
i) N~ n N ~ EBj~~l Lj.
=
ii) L~'+1 C.
iii)tj = T k '-j+1 P2j-Hf".
2) i) N! n N ~ EBj~2 Lj.
ii) L:'+1 = L:'+2 = C.
iii)tj = T 8 '-j+1 P2j-Hf. for j ~ s' t:'+1 = {JPs- 1 t:'+2 = T Ps- 1 '
3) If I > 1, we have
•.) N's+1 n N ~ a::. 8 "+1 L s+1
Wj=1 j h
,were = s + 1)/2] .
s"[(
..) L s +1 C
II 8"+1 = .
iii)tj+1 = T a"-j+1 P2j-Hf.+l for j ~ s" t:t;l = ({J - T)Ps- 1.
If I = 1, we have
i'lI N'a+l n N '" a::.~"+2 L~+l
- W)=1 J '
where s" = [(s + 1)/2].
"'1 L 8 +1 L a+1 C
n I a"+1 = ."+2 = .
or 'J <
""lt a+1 = T a"-;+Ip.2j-2+f'+1 f
III I j _ S",
t 8s"+1 s+1 =
+1 = TJJf.lPa-It t s"+2
({J - T - TfJ)Pa- 1 •
Before proving the main result, we need a few lemmas about graphs.
The proofs can be found in [6].
Define the Hoffman's sequence I'q =11 T(2,q, 00) Wfor q E N (cf. also [2]).
This is a strictly increasing sequence and limq J.tq = 2 + V5 (cf. [1D.
lemma 4.3 For c ~ 4 we have the following:

i) II T(3,3,c) 112< I'e'
ii) II T(2,s + l,s + 1) 11=11 T(2,s,s + 3) II.
iii)1I T(2,s + 1, s + 1) 112< 1'•.
iv) II T(2,s,s) 11<11 T(3,3,s) II·
=
lemma 4.4 Let T [M : Nt 1, then
[M : N] < 2 + .f5 <=> (1;;)3 < T.
214 s. A. Rida
lemma 4.5 Let r = =

T(a,b,c) , a ~ b:5 c and n H(p,q,r,sj/) , p :5
q , r ~ S , q ~ s. Assume that r :F T(3,4,4) and that II r 112=11 n 112E
(4,2 + v'5). Then c ~ S + 1.
The next theorem is the main result of this paper. Here is an outline of
the techniques used in its proof:
1: Let N C M be an inclusion of finite dimensional algebras with N con-
taining the unit of M. Then every summand in the decomposition of M
contains at least one summand from the decomposition of N . (Note that
all these inclusions have multiplicity 1 as this is the only case of interest to
us).
2: Let (Q, s) be a summand in the decomposition of N along with its trace of
a minimal projection, and assume that (Q,s) C (Li,ti), i = I, ... ,n where
(Li,tSs are summands of M. Then I'i = tds E A(M,N) Vi = I, ... ,n
and Ef=ll'i = 1 (cf. section 2).
3: One dimensional summands of M contain only one dimensional sum-
mands of N (This is obvious).
4: Let N C M be hyperfinite III-factors with index [M : N] = r-l. The
smallest value :F 0 (resp. largest value :F 1 ) in A(M,N) is r (resp.
1 - r). Moreover, A(M,N) has gaps between ~n(r) and ~n+I(r) for all
n. By exhibiting values in these gaps, we will show that certain inclusions
can't occur.
theorem 4.6 No star graph occurs as a principal graph 0/ an irreducible

inclusion 0/ finite depth sub/actors with index in (4,2 + v'5).
Proof
Let N C M be such an inclusion, r = [M : Ntl its index, r = T(a,b,c)
its principal graph and n its dual graph (Le. principal graph for M C M},
where Ml is the result of the basic construction performed on N eM,
also note that II r 11=11 n ID. We know from lemma 1.1.12 [7] that n :F r,
therefore, either n = H(2,q,2,sjl) for some 2 ~ q ~ s, I ~ 1, or n =
T(p,q,r» (p,q,r) :F (a,b,c). Let's assume that r :F T(3,4,4) (this case
uses similar techniques, and will therefore be omitted).
Case 1 n = H(2,q,2,sjl). Set (J = 1- r - ~8.
By lemma 4.5, we have c ~ S+ 1. Consider the inclusion N;nN C N;nM. By
lemma 4.1 i), the decomposition of N; n N has one I-dimensional summand,
call it Q, where the trace of a minimal projection is p•. Let s" =
[~]
and (Lj+I)j:1 the first s" summands in the decomposition of N; n M. Let
(T, ... , T2 p.-3, Tp._ t) be the corresponding traces as given by lemma 4.2
iii). (Note that the indexing of the tunnel of M C M1 is different from that
of N C M j it is shifted down by 1).
Claim: Vj ~ sIt - 1 Q rt. Lj+1.
Proof of claim: Each inclusion of summands gives rise to a nonzero value
in A(M,N). Assume Q c L:t~l' then p = . .2~_3 E A(M,N).
On one hand
on the other hand, lemma 4.3 ii) implies

T;l =11 T(2,s + l,s + 1) 112< P. <II n 112= T- 1
Since Ts + 2A,,+1(T,,) = 1 and T < Ts , we get
I-T
As+1(T) < -2- j (2)
Combining (1) and (2), we get I' < (1;;)3 < T by lemma 4.4, a contradic-
tion since T is the smallest nonzero value in A(M,N), and this completes
the proof of the claim.
Since the (tj+1)j:l1 is increasinginj, all the other inclusions Q C Lj+1, j ~
sIt - 1 yield values smaller than T and therefore can't occur.
Now, let's distinguish the cases I > 1 and 1= 1.
I> 1 : In this case we are left with 2 possible inclusions:
Q c L:t1 which yields
P1 = --p;-
TPs - 1 = A.+1 E A( )
M,N
Qc L:t!l which yields
_ ({3 - T)Ps- 1 _ ({3 - T)AS+1 _ 1- A. - 2T A(M N)

1'2- - - E,
p. T 1- As
Since both P1 and 1'2 are -F 1, they both have to be in A( M, N) and satisfy
P1 + 1'2 = 1 (because N; n N contains the unit of N; n M), but this is a
contradiction since
1- As - T
P1 + P2 = 1 _ As < 1
216 s. A. Rida
1 = 1 : In this case, there are 3 possible inclusions:
Q C L:t1 which yields
JJI = -p;-
TP.- 1 = >'.+1 E A( )
M,N
QC L:t;l which yields
JJ2 =
T{JP.-
PII
1
= (J >'.+1 E A(
M, )
N
Q C L:;t;2 which yields
_ ({J - T - T{J)PII-l _ ({J - T - T{J)>'II+1 A(M N)

~-
p.
- T
E,
For the same reason as above, we get a contradiction since

1-T->'.
JJI +JJ2 + JJ3 = 1 _ >'. < 1
Finally, we conclude that n cannot be an H-graph.

Case 2 n = T(p,q,r).
Assume c = r. Since r and n are different graphs, one of them (say n) is
a T(2,q,r) and the other is a T(3,3,r). But this is impossible in virtue of
lemma 4.3 iv) since
II T(2,q,r) 11$11 T(2,r,r) 11<11 T(3,3,r) II

Now assume r $ c - 1, then N: n N has one 1-dimensional summand Q
where the trace of a minimal projection is Pr , and N: n M has at most two
1-dimensional summands Ll and L2 with respective traces >'p-l>'pPr - 1 and
>'q-l>'qPr- 1 •
Since p and q cannot be both equal to 2 (otherwise II n 112=11 T(2,2, r) 112<
4), one of them, say q, is ~ 3. Since both Ll and L2 contain Q, we get the
following value
Now if q = 3, then
p. = -r>'3;r±l
= >'3>'r+1
< >'3
but this implies ~3~r+1 = r, which is a contradiction since ~3 = r/(1 - r)

and this forces 1/2 > ~r+l = 1 - r > 1/2.
If q ~ 4, then p = 2 (recall that r", T(3,4,4».
Claim: p. < r
Proof of claim: Suppose this is not the case, then
p. ~ r <=> '\g-1:,'\rt 1 ~ r
<=> ~q-l~r+1 ~ r(1 - ~q-l)
<=> "\q-l("\r+l + r) ~ r
<=> ~r+l + r ~ 1 - ~q-2
<=> ~r+l + + +
r ~ r ~q ~r - ~q-2
<=> ..\r+l - ..\r ~ ..\q - "\q-2 > ..\q - "\q-l
which is impossible since q < r + 1 and (..\n+1 - ..\n)n is strictly decreasing

by lemma 1.1 d). Therefore the claim is proved and yields a contradiction.
Now if c :5 r - 1, the previous proof applies to the inclusion M C Ml and
concludes that n cannot be a T(p,q,r) and this means that r = T(a,b,c)
cannot be a principal graph.
References
[1] F. Goodman, P. de la Harpe, V. Jones, Coxeter graphs and towers of
algebras. MSRI Publications 14, Springer Verlag.
[2] A. J. Hoffman, On limit points of spectral radii of nonnegative sym-
metric integral matrices. Lecture notes in Math. 303: 165-172, 1972.
[3] V.F.R Jones, Index for subfactors. Invent. Math. 72:1-25, 1983.
[4] S. Popa, Relative dimension, towers of projections and commuting
squares of subfactors. Pac. J. Math. 137 number 2: 95- 122, 1989.
[5] S. Popa, Classification of subfactors: reduction to commuting squares.
Invent. Math. 101:19-43, 1990.
[6] A.S.Rida, On subfactors with the generating property and graphs. PhD
Thesis, University of California, Berkeley, 1992.
[7] J. K. Schou, Commuting squares and index for subfactors. PhD Thesis,
Odense Universitet, Denmark, 1990.
Depth 2 Subfactors and Hopf Algebra
Crossed Products
Wojciech Szymanski
University of Cincinnati
Cincinnati, OH 45221
Abstract
We discuss finite dimensional Hopf *-algebra structures related to
depth 2 subfactors of I h -factors.
1. PRELIMINARIES ON HOPF ALGEBRAS AND THEIR ACTIONS
In the late sixties Sweedler introduced concepts of a Hopf algebra, its action,
and the crossed product (see [9]). Interest in Hopf algebras grew in the
eighties after the discovery of quantum groups. Hopf *-algebras came into
play and their relation to the theory of operator algebras was clarified (see
[12]). Since then, many applications of Hopf *-algebras to operator algebras
have been found. In particular, several concepts known in the case of group
actions (such as spectral subspaces, spaces of spherical functions) have been
translated into the context of finite dimensional Hopf *-algebra actions on
CO-algebras (see for example [6]). In this article we want to discuss some
applications of Hopf algebra techniques to subfactors.
A (finite dimensional) Hopf *-algebra H is a (finite dimensional) C'-
algebra equipped in linear maps; ~ : H ~ H ® H, comultiplication, c :
H ~ C, counit, and S : H ~ H, antipode (see [9, 12]). Both ~ and c
are CO-algebra homomorphisms and S is a *-preserving antimultiplicative
involution. Moreover, the following identities are required; (~181 id)~ =
Depth 2 subfactors 219
(id®~)~, (c:®id)~ = id = (id®c:)~, m(S®id)~ = c: = m(id®S)~, where

m : H ® H --+ H denotes the multiplication. A simpliest example is the
complex group algebra C[G] of a finite group, with ~(g) = 9 ® g, c:(g) = 1,
and S(g) = g-t.
HO, the vector space of linear functionals on H, is again a Hopf *-algebra
with the following structure; (4),,p)(h) = (4) ® "p)~(h), 4>*(h) = 4>(S(h*»,
~O(4))(h®k) = 4>(hk), c: 0 (4)) = 4>(1), SO(4))(h) = 4>(S(h». This notion gener-
alizes the duality for finite abelian groups. The dual to a group algebra of G
will be denoted by C( G). It is an abelian algebra with minimal projections
{Pg I 9 E G}, comultiplication ~(pg) = L,hEGPh ®Ph-Ig, counit C:(Pg) = beg
(e the neutral element of G), and antipode S(pg) = Pg-1. One should notice
that in order to define a comultiplication in HO we used only multiplication
in H. Thus the dual to an algebra is a co algebra (see [9]).
An action (see [6, 9]) of H on a unital *-algebra L is a bilinear map
. : H xL --+ L satisfying the following conditions; h . I = c:(h), I· x = x,
hg· x = h· (g. x), h . xy = L,;(hf . x)(hr· y), (h· x)* = S(h*) . x*, for all
h,g E H, x, Y E L. Here we denoted ~(h) by L,; hf ® hr. In case of a group
action D: : G --+ Aut(L) the bilinear map is just (L,9EG tgg)·x = L,9EG tgD:g(x).
The crossed product L >4 H is a (unital) *-algebra that as a vector space
coincides with L ® H, and with multiplication and *-operation defined as
follows; (x ® h)(y ® g) = L,i x(hf· y) ® hrg, (x ® h)* = L,i(hf)*· x* ® (hr)*.
The fixed point algebra LH is defined as all such x E L that h . x = c:( h)x
for any h E H. If all elements of L are fixed, then the action is trivial. There
is always an (inner) action of H on itself defined as h . 9 = L,i hfgS(hr).
The most interesting for our further discussion is the dual action of HO on
the crossed product L )</ H, defined as 4>. (x ® h) = x ® (id ® 4>)(~(h».
Suppose that L is a factor von Neumann algebra. If (LH)' n L = CI and
L is a faithful H-module, then it is natural to call the action outer.
2. HOPF ALGEBRAS AND SUBFACTORS
Towers of I It-factors provide particularly interesting objects for applications

of Hopf algebra techniques. In what follows we consider a pair N C M
of I It-factors containing a common identity with trivial relative commutant
and finite Jones index. We also denote M t = (M, eN) and M2 = (M t , eM),
the Jones extensions (see [1, 2]).
220 W. Szymafiski
If H C G are finite groups acting outerly on a I II-factor M, Hopf algebra

techniques may be applied to the investigation of the derived tower for M XI
HeM XlG.
This can be easiest illustrated on the following example, when we take
H = (e), G abelian and N = MG. It is clear that MI can be identified
with M XI G and M2 with MI XI G, with the dual action of G. We denote
N'nMI = A, M'nM2 = Band N'nM 2 = C. We have A ~ C[G),
B ~ C[G) and C ~ M1G1(C). Moreover, the above description remains valid
even if we no longer require that G be abelian, when one replaces G with
C(G)-the Hopf *-algebra dual to C[G).
It is a remarkable fact that the above model characterizes towers of 111-
factors such that C simple and B abelian (see [8]). A natural question arises;
what if C is a factor, but B is not abelian? As indicated by Ocneanu ([5),
see also [1]), in this case one must consider a crossed product related to a
Hopf algebra action rather than to a group action. A model for such a case
is provided by the following tower (see also [11])
MH eM c M XI He (M XI H) XI HO,
with a finite dimensional Hopf *-algebra H acting outerly on a type I II factor

M.
The author gave in [10) a detailed proof of the following theorem.
Theorem 1 If N c M are Ill-factors such that [M : N) < 00, N'nM =

CI and C = N'nM 2 simple, then B = M'nM 2 has a natural Hopf *-
algebm structure and acts on MI in such a way that the resulting crossed
product is naturally isomorphic to M 2 .
Longo has recently obtained a similar result for infinite factors (see [4]).
The idea of the proof is based on the duality between A and B established
by a bilinear form (-, .) : A x B - t C defined as
for a E A, b E Band T the canonical trace on M 2 • This allows us to view

B as a vector space dual to A and thus introduce a comultiplication in B
as dual to the multiplication in A, as shown in the previous section. It is
not trivial, however, to check that this leads to a Hopf *-algebra. The most
Depth 2 subfactors 221
difficult part of the proof is to show that ~ preserves multiplication. For this
we make use of the following observation.
A map T : B®B -+ C defined as T : b1 ® ~ ....... b1 eNS(b2 ), where S
denotes the antipode of B, maps ~(B) onto A. S: B -+ B is the unique
map satisfying (a, S(b)) = (a*, b*) for all a E A, bE B.
The crossed product decomposition of M2 is based on the following ob-
servation.
There is an isomorphism of vector spaces M1 ® B -+ M2 given by x®b .......
xb. This map becomes an isomorphism of von Neumann algebras M1 )4 B -+
M2 if the crossed product is related to the action of B on M1 defined as
b· x ~ [M: N]E(bxeM),
with E the trace preserving conditional expectation from M2 onto M 1.

Theorem 1 applies to depth 2 subfactors only. However, assuming irre-
ducibility and finite index, one may consider the following question.
Problem 2 If N c M are 111 factors such that [M : N] < 00 and

n
N' M = eI, is there a "natural" Hopf *-algebm structure on N' M1 ? n
Although, unless the depth of N in M is 2, there is no such a crossed product
decomposition as described in Theorem 1, it seems that at least in some
cases the answer is positive. The author also suspects that a rigorous proof
of Problem 2 might lead to new conjugacy invariants for subfactors.
References
[1] F. M. Goodman, P. de la Harpe, V. F. R. Jones, Coxeter graphs and tow-
ers of algebms, Math. Sci. Research Inst. Publ. n. 14, Springer-Verlag,
1989.
[2] V. F. R. Jones, Index for subfactors, Invent. Math. 72 (1983),1-25.
[3] H. Kosaki, S. Yamagami, Irreducible bimodules associated with crossed

product algebms, Internat. J. Math. 3 (1992).
[4] R. Longo, A duality for Hopf algebms and for subfactors, preprint, 1992.
222 W. Szymanski
[5] A. Ocneanu, "Quantized groups, string algebras and Galois theory for
algebras," in Operator algebras and applications, vol. 2, London Math.
Soc. Lecture Notes, vol. 136, Cambridge Univ. Press, 1988,119-172.
[6] C. Peligrad, W. Szymanski, Saturated actions of finite dimensional Hopf
*-algebras on C*-algebras, preprint, 1992.
[7] M. Pimsner, S. Popa, Entropy and index for subfactors, Ann. Sci. Ec.
Norm. Sup. 19 (1986),57-106.
[8] M. Pimsner, S. Popa, Finite dimensional approximation of pairs of al-
gebras and obstructions for index, J. Funct. Anal. 98 (1991),270-291.
[9] M. Sweedler, Hopf algebras, New York, W. A. Benjamin, 1969.

[10] W. Szymanski, Finite index subfactors and Hopf algebra crossed prod-
ucts, Proc. Amer. Math. Soc., to appear.
[11] W. Szymanski, Actions of finite dimensional Hopf *-algebras on C*-

algebras and von Neumann algebras, preprint, 1993.
[12] S. L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys.

111 (1987),613-665.
Section V
REPRESENTATIONS OF GROUPS AND
ALGEBRAS ON HILBERT SPACE
Generalized Characters of Uoo
Robert P. Boyer
Section 1. Introduction
Recently, G. I. Ol'shanskii introduced a new class of semifinite factor
representations and their associated matrix coefficients, called generalized
characters, which include the essential features of finite character theory
for the infinite dimensional classical groups. They are related to the con-
cept of total positivity [Kar] which has important applications in analysis,
combinatorics, and to the theory of operator nodes and their characteristic
functions which were introduced to study non self-adjoint Hilbert space
operators. In [04], Ol'shanskii classified generalized characters associated
to the infinite symmetric groups. In this situation, they are related to the
Brauer semigroups which, together with their q-analogues, have important
applications in knot theory and in the study of the Yang-Baxter equation
(see [04] for references). The association of all these ideas suggest that the
investigation of generalized characters may provide new insight in these
areas. In this paper, we use c· -algebraic methods developed in (Boy2] to
begin the classification of certain generalized characters associated to the
unitary group.
We now give a brief review of [03] and relate these results with previous
work. We examine the pair (L, K), where L is a topological group with a
closed subgroup K. The subgroup K acts naturally on L via conjugation.
A normalized continuous positive definite function p on L is K -central if
p(kxk-l) = p(x), for all x E L, k E K. The collection of all K-central
functions for (L, K) forms a convex set whose extreme points are called
generalized characters, denoted by GC(L, K). They can be identified with
a matrix coefficient of certain semifinite factor representations. For K = L,
the generalized characters reduce to the usual finite characters of L.
For many classical pairs (L, K) (see [02] for the complete list), the
space LK of K-orbits admits a binary operation giving a semigroup. This
algebraic structure, absent in finite dimensions, provides the infinite di-
mensional theory its unique features. For example, a K-central function
gives rise naturally to a function on LK which is extremal if and only if
it is multiplicative, relative to multiplication of K-orbits. An important
example to consider is L = U(oo), the inductive limit unitary group, with
K = L, then the K-orbits are identical with the conjugacy classes 6.(L) of
L, given by unordered families of complex numbers {Zj} such that IZjl = 1
and Zj #- 1, for only finitely many indices. Two such families, say C = {Zj}
and C' = {zj}, multiply to yield Co C' = {Wj} where W2j = Zj and
W2j+l = zj. Since a L-central function p yields a function p on 6.(L), p is
a finite character (that is, extremal) if and only if p(C 0 C') = p(C)p(C').
226 R.P. Boyer
Generalized characters are interesting examples of spherical functions

for the pair (L, K). Recall a function p on L is K-biinvariant if it is nor-
malized, continuous, positive definite, and p(kxk') = p(x), for all x E L
and k, k' E K. The set of all K-biinvariant functions for (L, K) is a con-
vex set, whose extreme points are called spherical functions. The spherical
functions for the classical pairs (L, K), that correspond to the infinite Rie-
mannian symmetric spaces L/ K are discussed in [Pl,2] and [01]. The
generalized characters p for (L, K) are in natural correspondence with the
spherical functions <Pp of (L x K, diagK), where <pp(g, k) = p(gk- 1 ). [04]
The principal pair we shall study is (U (1 + 00), U (00 )), whose gen-
eralized characters correspond to spherical functions on a non-symmetric
homogeneous space. This pair is formed as follows. Given an orthonormal
basis {ej }~1 for the complex Hilbert space H, let H(m,n) denote the linear
span of {em, em+1, ... , en} c H, where m S n; HN = H(1,N). Define the
unitary groups LN = U(HN) and KN = U(H(2,N»). Let L = U(I+oo) de-
note the inductive limit of the unitary groups L N, while K = U (00) is the
limit formed via K N . Observe that L1 = K2 = U(I) and L1 x KN C LN.
We write V E LN as a (2 x 2) operator matrix V = (~ ~), relative to
the decomposition H = H1 $ H(2,N).
For (U(1 + 00), U(oo)), there is a natural semigroup structure where
the product of the K-orbits containing V = (: ~) E Lm and V' =
(~; ~,) E Ln .. the K-o,bit containing the usual p",duct of ( ~ i ~)

(
a'
o 0
1 0
b')E L1+m+n C L. This semigroup can be identified with the
c' 0 d'
product of two semigroups: D.(U(oo)) x , where is the semigroup of
rational inner functions, that is, a rational function J(z) on the complex
plane such that IJ(z)1 S 1, for Izl < 1, and IJ(z)1 = 1, for Izl = 1. [03]
The list of generalized characters is similar to the Voiculescu list [V] of
finite characters of U(oo). These generalized characters can be realized as
matrix coefficients of tensor products of metaplectic and spinor represen-
tations. Let <p+(z;o) = (1 +0 -az)-1, where 0 S 0 < 00, associated with
the metaplectic representation; <p-(z; f3) = 1 + f3 - f3z, where 0 S f3 S 1,
associated with the spinor representation; and <Pr(z; >.) = e,\(z-1), where
OS>' < 00. These are the generators for the finite characters. To simplify
notation, we restrict the attention to non-negative signatures. Such finite
characters have the formp(V) = det[J(V)], where J(z) = e,\(z-1) TI~1(1+
f3j - f3jz)/(1 + OJ - aZj), such that L~1(aj + f3j) < 00. These meromor-
phic functions are precisely the generating functions for the one-sided to-
tally positive sequences {Cn}~=o (see [Boy2, KV, Kar]), where a sequence
Generalized characters of U= 227
is totally positive if every finite minor from the infinite matrix {Ci-j }i,j=o
is non-negative. All known spherical functions for the classical pairs have
forms closely related to these unitary characters.
The generalized characters require an additional factor, associated
with the subsemigroup ~ of inner functions. Let p+(V; a, k, m) = [a +
ab(1 + a - ad)-lc]kdet(4)+(d;a))m, where 0::; k < 00 and 1 ::; m < 00;
p_(V;{3,k,m) = [a-{3b(I+{3-{3d)-lc]kdet(4>_(d;{3))m, where °: ;
k::; m;
and Pr(V;A) =det(4)r(d;A)). The list given in [03] for non-negative sig-
natures has the form: p(V) = anpr(V; A) IT;:l p+(V; aj, kj, mj)p_(V; (3j,
kj,mj), where 'L;:l(mjkj + mjkj) < 00. The task of classification
remains open.
We now examine the generalized characters associated with the spinor
representation using ideas from C* -algebra theory; in particular, with the
form of the CAR-algebra and its associated quasi-free representations de-
veloped in [SVl,2]. This C*-algebra naturally supports the representations
of U(1 + 00) realized on the antisymmetric tensors. Further, a class of
positive definite functions PA(V) = det(I - A + AV), °: ; A ::; I and
V E U(1 +00), were examined and classified according to factoriality, type,
and quasi-equivalence. Now, PA is L-central if and only if A = {3I, for
°: ; {3 ::; 1; moreover, these L-central functions are finite characters (see
section 2). On the other hand, PA is K-central if and only if A has the
form: (~ ~). Write P>.,{3 for such PA.
It is natural to determine which P>.,{3 are generalized characters. By
direct calculation, we find P>',{3 = (1 - A)PO,{3 + AP1,{3; by extremality, to
°
be a generalized character, it is necessary that A = or 1. (In section 2,
we shall give a proof independent of [03] that this condition is, in fact,
sufficient.) Again, by direct calculation, we find PO,{3 and P1,{3 are among
the generalized characters on Ol'shanskii's list: PO,{3(V) = p-(V;{3,O, 1)
and P1,{3(V) = p-(V; (3, 1, 1). Moreover, by [SV2], PO,{3 and P1,{3 induce
°
finite type II factor representations, for < (3 < 1. In the remaining cases,
Po,o and P1,O give infinite dimensional traceable irreducible representations,
while PO,l(V) = 1 and P1,1(V) = det(V), for V E U(1 + 00). Finally, we
record that P>',{3 is quasi-equivalent to PA',{3' if and only if {3 = {3'.
These more detailed results allow us to refine the nature of the clas-
sification of generalized characters: find those of type 100 , Ill, and 11 00 •
Further, the generalized characters fail to distinguish their factor represen-
tations up to quasi-equivalence. What further invariants are needed?
Section 2. Generalized Characters in the Antisymmetric Tensors
We shall sketch a self-contained argument to classify the characters on
the antisymmetric tensors by identifying them with the moment sequences
of probability measures on [0, I] (compare with [Pet, SS]). This method is
independent of [Boy2, KV]. Then we classify the generalized characters by
identifying them with the usual characters. This gives evidence of a close
228 R.P. Boyer
connection between characters and generalized characters.

Let P be a central function on L supported on the antisymmetric
tensors. Since p is determined by its restriction to L N , plLN = PN =
Ef=o ct)X~N), where X~N)(V) = Tr[Ai(V)], the trace on the j-th exterior
power of V E LN. The restriction rules
X~N+1)ILN = X~N) + x~~L 0 < j < N + 1; quadx~N+1)ILN = X~N), (2.1)
for j = 0, N + 1, imply that PN+1ILN = Ef=0(c~N+1) + c~~t1»)X~N) =

"N (N) (N) H
L.Ji=O ci Xi . ence,
c~N) = c~N+1) + c~~t), 0 ~ j ~ N. (2.2)
So, P is uniquely determined by the sequence {~N)}N=O. We shall show

that this sequence is the moment sequence of a unique probability measure
on [0,1]. In particular, the convex set of central functions supported on
antisymmetric tensors is affinely isomorphic to the convex set of probability
measures on [0,1].
The moment sequences {aN }rJ=o are uniquely determined by the con-
ditions ao = 1 and (_1)1e Aan ~ 0, where A is the forward difference oper-
i
ator AaN = aN+1 - aN [F, p. 224] We find AdoN ) = (_1)le c N+k) ~ 0, by
(2.2). Hence, doN) = Jo1t N dJ.L(t), for a unique probability measure 1'. fur-
ther, a character will correspond to a point mass; so, doN) = >..N, for some
>.. such that 0 ~ >.. ~ 1. We note for the character p(V) = det(1 - [3 + [3V)
that>.. = 1 - [3, since piLl = (1 - (3) + [3xP) = dol)
+ cP)X~l).
The identity det(I + T) = Ef=o 'fr(Ai(T)), T E End(HN)' implies
det[l- [3 + [3V] = Ef=o [3i(l- (3)N-iTr(Ai(T)). A simple induction shows
ct) = [3i(1- (3)N-i if doN) = [3N. Hence, every character P supported on
the antisymmetric tensors has the form: det[1 - [3 + [3V].
We next reduce the classification of the generalized characters to the
above character problem. Let f be a K-central function on L supported
on the antisymmetric tensors. We first show that f = >"fA + (1 - >")fB,
where 0 ~ >.. ~ 1, where fA and fB are K-central with fAIL1 = 1 and
fBIL1 = XP), respectively. By [DJ (where K-central functions are called
partial traces), (fILN)(V) = Ef=oat)x~N)(A~N)V) + bt)xt)(Bt)V),
where a~N), b~N) ~ o. Here, A~N) is the projection of Ai (HN) onto the
subspace generated by the monomials ~l A ~2 A ... A ~i' where 2 ~ i1 <
i2 < ... ~ N, while Bt) is the projection onto the subspace generated by
eil AW) = BaN) = o.
A et 2 A ... A ei i , with i1 = 1. Set
The restriction rules are almost identical to those of (2.1):
~N+1)(A~N+1)V)IL ~N)(A~N)V)
N= X + X)-l
~N)(A~N)V) 0< . < N,
{ X) ) )) )-1 , _ J -
X~N+1)(Bt+1)V)ILN = X~N)(Bt)V) + X~~l(B~~~V), 1 ~ j ~ N + 1.

Generalized characters of U 00 229
These rules yield the same relations as (2.2) :
a~N) = a~N+1) +a~~tl), 0 ~ j ~ N -1; b~N) = b~N+1) +b~~t1), 1 ~ j ~ N.

(2.3)
The K-central functions IA and IB are given by their restrictions, when
ab1) > 0 or b~l) > 0 as:
N-l (N) N (N)

IA (V )1 LN -- '"' aj
LJ (ifXj(N)(A(N)V)
j f ()I
,B V LN -_ '"' bj (N)( Bj(N) V,
LJ (IfXj ).
~o ~ j=l~
ub
otherwise, take IA = 1 and IB = 1. Since 1) + bP) = 1, I is a convex
combination of I A and lB.
For I to be a generalized character, it must be an extremal K-central
function; hence, either IIL1 = 1 or IIL1 = X~l). By (2.3), the convex sets
of all K-central functions I that satisfy IILl = 1, respectively IILl = X~l),
are affinely isomorphic to the set of usual characters. Hence, the general-
ized characters I with IILl = 1, respectively, IILl = X~l), are uniquely
ab
determined by the value 2), respectively, b~2). They can be identified with
ab
the generalized characters, given in section 1, Po,,8(V), with 2 ) = 1 - {3,
respectively, P1,,8(V), with b~2) = 1 - {3.
Further,the same method used for finite characters shows Po,,8(V) =
Ef:;/ {3j(1 - {3)N-j-1Tr(A~N) Ai (V)), where V E LN. By [SV1, pp. 112-
117], we have P1,,8(V) = Ef=1{3j-1(1- {3)N-jTr(Bt)Ai(V)), V E LN.
Hence, every generalized character P supported on the antisymmetric ten-
sors have the form det[I - A + AV], where A = (~ ~), where A = 0 or
1 andO~{3~ 1.
In section 3, we interpret these ideas in their C" -context to derive
partial classification results.
Section 3. Classification Results
We need the fundamental construction of Stratila and Voiculescu [SVl]
of the C* -algebra associated with a direct limit G of separable compact
groups G N. It is defined as the C* -completion of the direct limit of the
Banach *-algebras £(N), where £(N) is the Banach *-subalgebra generated
by the £l(GN) C M(GN) and M(GN) is the Banach *-algebra of finite
Borel measures on G N. The natural inclusion M (G N) c M (G N+1) induces
the embedding of £N into £(N+1). We denote this C"-algebra by C*(G).
It is an AF-algebra.
It is convenient to work with the primitive quotients AJ = C*(U(oo))j
J, where J is some primitive ideal. The space of primitive ideals of
C* (U (00)) is parametrized by a (double) signature {Ujj L j }f=1' where
Uj , -Lj E Z U {oo} and Uj ~ Uj+l ~ Lj+1 ~ Lj . [SV1] To simplify
230 R.P. Boyer
notation, we shall assume for the remainder of this section that the lower
signature {Lj } is identically o. Further, we single out those primitive quo-
tients AJ = C*(U(oo))jJ that admit faithful type III factor representa-
tions. Their signature entries have at most one distinct finite value. We
call these signatures type III [Boy2].
We can interpret the proof in section 2 by using the Stratila-Voiculescu
dynamical system associated to the AF-algebra AJ = A(O, r), where 0 is a
compact Hausdorff space (consisting of paths through the Bratelli diagram)
and r is a countable group of homeomorphisms (consisting of finitely sup-
ported path permutations). J is the primitive ideal with U1 = U2 = ... = l.
We give a briefreview of this system. For quotients of C·(U(1 + 00)), the
nodes DN of the Brattali diagram D at level N are a certain subset of f;.
Two nodes A E DN and p. E DN+1 are connected if and only if p.ILN > A.
The space 0, then, is the set of all paths (WN) through D with WN E DN.
Further, finite characters are given by the r-invariant and ergodic probabil-
ity measures on O. By [SVl], this path space for the antisymmetric tensors
is given explicitly as infinite paths through the usual Pascal triangle.
The generalized characters found in section 2 are also described as
measures. This does not automatically follow from [SVl,2] since the op-
erator A, giving the KMS-function PA for the generalized character, does
not satisfy ker(A(I - A)) = {O}. The only functions PA that are directly
given by probability measures on 0 are those such that A is diagonal and
A(I - A) is injective. We need to use a subdynamical system. Let Ox,
(x = 0,1) be the subset of 0 of all paths through the Pascal triangle such
that (WN) E Ox only if W1 = x, where x = 0,1. Let ro be the subgroup
of r that preserves Ox. Then the generalized characters are given by the
ro-invariant, ergodic measures on Ox. Further, the two dynamical systems
(0, r) and (00, ro) are isomorphic. This is the method we used to find the
infinite characters of U(oo) in [Boy2].
For the generalized characters of the pair (L, K) = (U(1 +00), U(oo)),
there are three natural invariants, which identify the subdynamical system
attached to the primitive quotient of C·(U(1 + 00)). The first invariant
is the multiplicative character of U(I), or equivalently an integer n, given
by the restriction piLl. The second invariant is the finite character X of
the group K, p1K1. The third invariant is the (double) signature of the
primitive ideal of C* (U (1 +00)) given by the semifinite factor representation
associated to p. By example, each of these invariants is necessary. Consider
the two generalized characters P1(V) = adet(1 - (3 + (3tI) and P2(V) =
[a - {3b(1 - {3 + {3d)-lc] det(1 - (3 + (3d). Then P11L 1 = P21L 1 = a and
P11K = P21K. But the associated primitive ideals are different: ker(p1) has
signature: (2,1,1, ... ), while ker(p2) has signature: (1,1,1, ... ). The two
other cases are given by similar examples. The results of this section show
many incidences where these three invariants do uniquely determine the
generalized character.
We shall use repeatedly that the U(N)-central functions of (U(N +
--
1), U(N)) are linear combinations of the functions ON(A, p)(V) =
Tr(P.>.,,.A(V)), where A E U(N), p E U(N - 1) -: and P.>.,,. is the pro-
jection of the representation space of A onto the subspace that supports
the representation I' [D].
Proposition 3.1 Let p E GC(L, K) such that piLl = 1 and p has signa-
ture of type lIt. Then p (: :) = det[f(d)], where fez) is a genernting
function for a totally positive sequence.
Proof. We first determine the deco~ition of the generalized char-
acter p on restriction to LN. If A E L N, then AILl x KN = L:{m} x
{AIm}. Now, we need only the term in this sum for m = 0 since piLl =
aO = 1. For m = 0, {AIm} = {A}. Hence, plLN = L:{CN(A)ON(A,A) :
the number of parts of A < N}. Further, the coefficients CN(A) obey the
same consistency relations as a trace for U (00 ); that is, there is a unique
trace t such that tlU(N -1) = L:{CN(A)XN-l(A) : A E U(N -1) . . .}, where
XN -1 (A) denotes the character of the irreducible representation of U(N -1)
with signature A. This mapping is an affine isomorphism between these two
convex sets. In particular, this map preserves extreme points. Hence, there
is a natural bijection between the generalized characters with piLl = 1 and
finite characters t of U(oo).
One consequence of this correspondence is that any K-central
function is uniquely determined by its restriction to L2, since any cen-
tral function of U(oo) is determined by its restriction to U(l). Now, write
plL2 = L: {C2(A)02(A, A) : number of parts of A = I} or plL2 = L:~o
C2( {k} )02( {k}, {k}), where {k}, the signature with one non-zero entry k, is
realized as a representation by the k-th symmetric power of the usual rep-
resentation of U(2) on C 2 . In particular, 02( {k}, {k}) (~ ~) = dk , where
(~ ~) E U(2). (This is the matrix coefficient of the lowest weight.) Fin-

ally, we conclude that if the finite character of U (00) corresponding to p
has the form det[f(W)], then plL2 (~ ~) = fed). Since q (~ ~) =
det[f(d)] is a generalized character such that qlL2 (~ ~) = fed), we

must have q = p, by uniqueness.
It is possible to extend this argument to give an independent proof
that the K-central functions det[f(V)], where f is a generating function of
a totally positive sequence are generalized characters.
The same method of proof gives us the following proposition. The key
observation is that for the partial traces that occur in the restriction plLN
all have the form: ON(A, p), where A must have k rows and p is obtained
from A by deleting one node from each row.
232 R.P. Boyer
Proposition 3.2 Let p E GC(L, K) such that piLl = a k and p has

signature of type Ill, with (Ut, U2,"') = (k, k,·· .). Then p ( : : ) =
n;=l [a + ,Bjb(1 - ,Bj + ,Bjd)-lc) det(1 - ,Bj + ,Bjd), where 0 < ,Bj < 1.
The combination of Propositions 3.1 and 3.2 allow us to recover the
classification of generalized characters that are supported by the primitive
quotient with signature (1,1,1, ... ), that is, supported by the antisymmet-
ric tensors, given in section 2.
Both of these Propositions have a dynamical system interpretation.
Let (Ok, rk) be the dynamical system for the primitive quotient with the
type Ill-signature (k, k, .. .). Let (0, r) be the system for the primitive quo-
tient in Proposition 3.1. Let 0' be the subspace of all paths in 0 that pass
through the node (0) in 6(1) and r' be the subgroup of r that preserve the
subspace 0'. Then (O',r') is isomorphic to (Ok,rk)' For Proposition 3.2,
the subspace 0' is given by restricting paths to pass through (k) in 6(1).
Then the generalized characters correspond to the r'-ergodic probability
measures that are supported on 0'. This is an exact analogue of the finite
character situation [SVl).
We next address the problem of determining the range of the canon-
ical map from GC(L, K) --+ Prim(L) where p 1-+ ker(p), where ker(p)
denotes the kernel of the semifinite factor representation corresponding
to p. The finite characters parametrize the subset of the primitive ideal
space of U(1 + (0), whose signature entries have at most one finite value
[Boy2). Generalized characters still parametrize only a proper subset of
Prim(U(l + 00».
Proposition 3.3 Let p E GC(L, K), then the signature of p has at most
two distinct finite values.
Proof. If (Ut, U2, ... ) is the signature of p, let r = /{j : Uj = 00}1,
so 0 :::; r :::; 00. Next, let PN = pILN, then PN = E{CN('\,p)ON('\,p) :
,\ E t:;"p E K;.}. Set XN to be the support of PN C t:;, and YN,
the support of plKN C K;. Since there exists an integer n such that
piLl = a'''', any irreducible component v occuring in the restriction '\IKN
satisfies (*) 1'\1- n = Ivl.
Since plK is a finite character, its signature (U1, U2, ... ) must be of
type Ill; that is, there exists an index 0 :::; r' :::; 00 such that, if r' < 00,
U~'+1 = U~'+2 = ... < 00 and, if r' > 0, U1= '" = U~, = 00. This has
the consequence that if v E YN then (**)lj :::; Uj, where v has signature
(It, 12, ... , IN-t). It follows because of (*) that r = r'. Hence, the signatures
of P and plK agree if all the signature entries of P are infinite. So, assume
r < 00. Then there must be an integer jo such that Uj = Uj, j ~ jo and, if
r ~ 1, Uj = Uj, 1 :::; j :::; r,
We will be done if we can show that jo :::; r + 2. Since we always
have Uj :::; Uj, for all j, we need to verify that Ur +2 = U~+2' To accom-
Generalized characters of U ~ 233
plish this, consider the partial trace(A, v), where A e X N. Recall that
(JN
e YN; moreover, v must be an
(IN(A, v)IKN is the character of v, where v
irreducible component of the skew Schur function {A}/{n}. If A has sig-
nature (mb m2, ... ,mN) and 1nr+2 > U~+2' then mr+l > U~+2 = Ur+1
as well. Then, by the Littlewood-Richardson rule, the irreducible v e YN
cannot satisfy (**), since we must have lr+1 > U~+1' Contradiction. In
other words, the signatures of p and plK agree except for the (r + l)-st
entry, when r < 00.
In the next two propositions, we classify generalized characters in non-
negative signatures that correspond to infinite factor representations. We
first treat the type I case. There it is easy to exhibit examples; for example,
if 7r is the natural representation of U(l +00) on the Hilbert space H, then
p(V) = (7r(V)el' el) = a is a generalized character. We show that the other
type I generalized characters are just powers of this one.
Proposition 3.4 Suppose pe GC(L, K) such that the corresponding prim-
itive quotient AJ is type I, then p has signature (n, 0, 0, ... ) and
p (: ~) = an.
Proof. Since plK is a finite character and gives a representation of a type

I C·-algebra, plK must be a power of the determinant. In fact, plK = 1.
This follows since the signature entries for a type I quotient are finite
and eventually 0 while, for a finite quotient, all finite signature entries
are equal [Boy2). Now, the representations that support this primitive
quotient are symmetric powers of the usual action of U (N) on eN, denoted
by {m}. Recall {m} e ~ decomposes when restricted to Ll x KN as
2:7=o{m - j} x {j}. Since only the trivial representation {O} of KN can
occur in this decomposition, plLN = en(JN ( {n}, {O} ). Finally, we observe
that (IN({n}, {O}) (: ~) = an. (This is the matrix coefficient of the
highest weight.) Hence, p has the desired form.
Next we identify generalized characters of type 1100 given by a product
of generalized characters of type 100 with type Ill. These functions are
included in the positive definite functions studied in [Boy4). Again, the
technique of proof of Proposition 3.1 is sufficient.
Proposition 3.5 lip e GC(L,K) such that piLl = an andp has signature
(n, k, k, ... ), where n > k, then p ( : ~) = a n- k 0;=1 det(l - .Bj + .Bjd),
where 0 < .Bj < 1.
The subdynamical system (n', r') associated with these generalized
characters has a similar description as those ~en above. n' consists of
all paths that pass through the node (n) in U(l). r' consists of all path
permutations preserving the subspace n'. Then (n', r') is isomorphic to
(n k, rk), the system for primitive quotients with signature (k, k, ... ), just
234 R.P. Boyer
as before, and the generalized characters correspond to the r' - ergodic

probability measures on 0'. It is interesting that this family of infinite
factor representations is classified in the same manner as the finite factor
representations.
We can establish the converse of Proposition 3.3. First, we note that
the generalized character p+(V; a,O, 1) = det(l+a-ad)-l, where 0 < a <
00, has the form of the positive definite functions supported on the sym-
metric tensors studied in [Boy4]. In particular, its signature is (00,0,0, ... ).
This, together with Propositions 3.4 and 3.5, implies that the kernel of the
generalized characters a n - k n;=l det(1 + aj - a j d)-l n~=l det(1 - f3j +
f3j d) , where 0 < aj < 00 and 0 < f3j < 1, has signature (oor,n,k,k, ... ).
The signature (00,00, ... ) is the kernel of n~l det(1 +aj - a j d)-l, where
° < aj < 00 and L:~1 aj < 00. Hence, we have:
Proposition 3.6 Every primitive ideal with at most two finite signature
entries is the kernel of a generalized character.
The generalized characters we have found can be easily expressed in
terms of the usual finite and infinite characters of U( (0), that we classified
in [Boy2]. Let p be the faithful state on the primitive quotient A that
corresponds to the generalized character p. Then there exists a projection
e E A which corresponds to an irreducible representation from £1 and a
character 4J on A such that p(x) = 4J(ex)N(e). This is an exact analogue
of the finite dimensional formula for generalized characters. In general, we
do not know if this expression is either necessary or sufficient.
Further, it would be interesting to develop an asymptotic theory of
generalized characters for the infinite unitary group as was done for the
infinite symmetric group [04]. This seems to require an analogue of the
Weyl character formula and a study of K-invariant projections.
References
[Boyl] R.P. Boyer, Representation theory of the Hilbert-Lie group U2(H),
Duke J. Math. 47 (1980), 325-344.
[Boy2] - , Infinite traces of AF-algebras and characters of U(oo), J. Op-
erator Theory 9 (1983), 205-236.
[Boy3] - , Representation theory of U1(H) in the symmetric tensors, J.
Funct. Anal. 77 (1988), 13-23.
[Boy4]-, Representation theory of U1 (H), Proc. Amer. Math. Soc. 103
(1988), 97-104.
[D] J. Dieudonne, "Treatise on Analysis," Volume 6, Academic Press, 1978.
[F] W. Feller, An Introduction to Probability Theory and its Applications,
Volume II, Second Edition, John Wiley & Sons, New York, 1971.
[Kar] S. Karlin, Total Positivity, Stanford Univ. Press, 1968.
[KV] S.Kerov and A. Vershik, Characters and factor representations of the
infinite unitary group, Soviet Math. Dokl. 26 (1982), 570-574.
[01] G.!. Ol'shankii, Method of holomorphic extensions in the theory of uni-
tary representations of infinite dimensional classical groups, Funct. Anal.
Appl., 22 No. 4 (1989), 273-285.

[02] - , Unitary representations of infinite-dimensional pairs (G, K) and
the formalism of R. Howe, in "Representations of Lie Groups and Related
Topics," A.M. Vershik and D.P. Zhelobenko (editors), Gordon and Breach
Publ., New York, 1990; pp. 269-464.
[03] - , Camcteres genernlises de U(oo) et functions interieures, C.R.
Acad. Sci. Paris, t. 313, Serie I, p. 9-12, 1991.
[04]-, Unitary representations of (G,K)-pairs connected with the infinite
symmetric group 8(00), Leningrad Math. J. 1 (1990),983-1014.
[Pet] J. Peters, On trnceable factor representations of crossed products, J.
Funct. Analy. 43 (1981), 7~96.
[PI] D. Pickrell, Separnble representations for restricted groups and associ-
ated to infinite symmetric spaces, J. Funct. Anal. 90 (1990), 1-26.
[P2] - , Mackey analysis of infinite classical motion groups, Pacific J.
Math 150 (1991), 139-166.
ISS] D. Shale and W. Stinespring, States of the Clifford Algebrn, Ann.
Math. 80 (1964), 365-381.
[SVl] ~. Stratila and D. Voiculescu, Representations of AF-algebrns and of
the group U( 00), Lecture Notes in Math., vol. 486, Springer-Verlag: Berlin,
Heidelberg, and New York: 1975.
[SV2] - , On a class of KMS states for the unitary group U(oo), Math.
Ann. 235 (1978), 87-110.
[V] D. Voiculescu, Representations factorielles de type III de U(oo), J. de
Math. Pures et Appl. 55 (1976), 1-20; Sur les representations factorielles
de type Ih de U(oo) et autres groupes semblables, C.R. Acad. Sci. Paris
Sk A., 279 (1975), 945-946.
Module structures on Hochschild and cyclic
cohomology of crossed products
Ronghui Jit
Indiana. Univ.- Purdue Univ. at Indianapolis
402 N. Blackford Street, Indianapolis, IN 46202
Introduction. In his foundational work of noncommuatative

differential geometry [3] Alain Connes introduced a notion of
cyclic theory for cyclic objects in an abelian category. He also
introduced a paring theory between cyclic cohomology and
K-theory. It turns out that from its invention
noncommutative differential geometry plays a key role in the
interactions among analysis, dynamics, topology, and
geometry. Calculating cyclic cohomology of appropriate
algebras will then be a crucial step in realizing these
interactions.
Burghelea [2] following an earlier work of Karoubi [13]
calculated the cyclic cohomology of a group ring. His idea is to
t Research sponsored in part by NSF Grant DMS-9204005.

Module structures 237
decompose the Hochschild complex and the cyclic double

complex of the group ring into a direct sum of subcomplexes of
cyclic objets [4] with respect to conjugacy classes of the group.
Nistor [14] along the same scheme showed that for algebraic
crossed product there is a module structure on the cyclic
cohomology of the cyclic objects, which correspond to
conjugacy classes containing no torsion elements. Moreover,
Connes' periodicity operators can be identified with a kind of
cup product for such conjugacy classes. However, his
construction uses arguments, which involve purely algebraic
mechinery and classifying spaces of certain quasicyclic objects,
so that it is difficult to be understood from an analytic
view-point. For instance, if one wishes to apply Nistor's
results to certain topological crossed products by discrete
groups, one is forced to check the continuity of all morphisms
used in his algebraic argument. This is hardly possible due to
the complexity of the algebraic machinery, such as, spectral
sequences.
The starting point of the present paper is to try to
overcome this difficulty for analytic purposes. In order to do
so a different construction of this module structure must be
done in a purely analytical way. Fortunately, inspired by the
~8 RH
works of Burghelea and Nistor, not only we have been able to
recover Nistor's results on the module structure mentioned
above by a purely analytical construction, but also the new
construction may even be generalized to those conjugacy
classes which contain torsion. Moreover the construction can
also be generalized to much more general situations [10] which
will lead to simpler calculations of periodic cyclic cohomologies
of many examples done before by others. But in this paper we
will concentrate on the calculation of cyclic cohomology of
crossed products, algebraic or topological.
§l. Preliminary.
Throughout this paper G will denote a discrete group
and (G) the set of conjugacy classes of G. We also denote by
(G)', resp. (G)", the subsets of (G) which contain
elements of finite and infinite orders, respectively. Given a
conjugacy class x, let h be an element in x. The subgroup
Gh = { g E G I gh = hg} of G contains a central cyclic
subgroup generated by h, which we denote llh. The quotient
GhtD.h is denoted by Nh. Obviously Gh and Nh do not
depend on the element h chosen up to group isomorphisms.
We will henceforth denote them by G x and Nx ' respectively,

whenever it is convenient.
Let A be a unital (for simplicity) algebra over a field k
of characteristic zero upon which G acts as a group of

automorphism. The algebraic crossed product of A by G is
denoted by A xo.G. Elements in A xo.G are of the form
~ agg, where ag E A, g E G, and the sum is finite. The

product is the usual one with the rule that ag· bh = abggh =
abggh, where a, b E A, g, h E G, and b g = g. bg- 1 = ag(b).
For each conjugacy class x E (G), set
(1. 1) L(G, A, x) = {Ln(G, A, x), b n ', b n, t n +1},

where Ln(G, A, x, a) = linear span of the set
®n+i
{(aog o, ... , ang n) :: aog o® ... ®ang n E (Axo.G) I
a i E A, gi E G, and go' ··gn EX},
and where
b n ' (aog o, ... , ang n)
= ~~;J (_1)i (aog o, ... , aigia i +igi+i' ... , ang n),
bn(aog o, ... , ang n)
= ~~;J (_1)i (aog o, ... , a ig iai+igi+i' ... , ang n)
+ (-l)D(angnaog o, a 1g i, ... , a n-ign-1),

240 R. Ji
and
It is clear that L(G, A, x) is a cyclic object [4, 14] so
that its Hochschild and cyclic cohomologies are defined which

will be denoted by HH *(L(G, A, x)) and He *(L(G, A, x)),
respectively. We have
(1.2) HH *(AxaG) ~ IIXE(G) HH*(L(G, A, x)),

and He *(AxaG) ~ IIXE(G) He *(L(G, A, x)).
When A is (, we will simply write L(G, x) instead of

L(G, x).
§2. A reduction theorem
The key observation of the original Burghelea's

calculation for the cyclic cohomology of a group ring [2] is his
reduction theorem. This reduction theorem reduces the
calculations of HH *(L(G, x)) and He *(L(G, x)) to that of
HH *(L(G h , (h»)) and He *(L(G h , (h»)), respectively. This
reduction theorem is also valid in the case of a crossed product.
It is proved in [14] by an algebraic argument. Since our main
motivation of the present paper is to apply cyclic cohomology

to the study of crossed product C*-algebras and the
calculation in [14] is not in general valid for topological crossed
products, we proceed to give a new proof which is extremely
useful when checking the continuity of maps involved and
when one wishes to extend the results of this paper to twisted
crossed products [10].
(2.1) Theorem. Suppose hEX. Then we have

1) HH*(L(G, A, x)) ~ HH *(L(G h, A, (h»))i
2) HC *(L(G, A, x)) ~ HC *(L(G h, A, (h»)).
Proof: By Connes' long exact sequence which relates cyclic
and Hochschild cohomologies, we need only to show that the
inclusion i: L(G h, A, (h»)) Co+ L(G, A, x) induces an
isomorphism on Hochschild cohomology. To this end, we
construct a homotopy inverse wh of i at the level of
Hochschild chain complexes. We define, first of all, a map 7r.
G --I Gh (as in [8]) as follows. Let Gh\G be the set of right

cosets of Gh in G, and let s: Gh\ G --I G be the
cross-section of the quotient map such that s(g) = g, where

bars denote the cosets the elements represent in Gh\G. Then
7r{g) = g. s(g) -1. Note that 7r(g) E Gh, and that 7r(hg) =
242 R. Ji
h7r{g).
We set -n!: Ln{G, A, x) --+ Ln{G h , A, (h») on

generators by
(2.2) ~(aoho, ... , anh n)
= (a o7r{g)-lg. 7r{gtl7r{gho), al7r{gho)-lgho. 7r{ghO)-I7r{ghohl)' ... ,
a -Igh
n\ lI\gh ••. h )-1-1
0 ···hn-1)-lgh 0 ••. h n-1. -I h • •h» ,
"\g.
The homotopy Dn: Ln{G, A, x) --+ Ln+1{G, A, x» is

given by
Do{a[hoD = D o{a[g-lhg])
= (a·g- 1h7r{g), 7r{gt1g) = (a.g- l7r{gho), 7r{hog)- lh og),
where ho = g-lhg, and in general,

(2.3) Dn{aoh o, ... , anh n)
= (a o·g-l7r{gho), 7r{gh ot 1gho, a 1h 1, ... , anh n)
- (ao· g-l7r{gho), a17r{gho)-lgho. 7r{ghO)-17r{ghoh1)'
7r{ghohl)-lghohl' a~2' ... , anh n)

+ ...
+ (-l)i{a o•g-l7r{gho), a17r{gho)-lgho. 7r{ghO)-I7r{ghohJ, ... ,
ai 7r{gho· •. hi _1tlgho· .• hi -1. 7r{gho· •. hi -1) -l7r{gho· •• hi),
7r{gho·· .h i)-lgh o•• .hi, ai+1hi+1' ... , anh n)

+ ...
+ (-1)n (a o. g-l7r{gho), a l7r{gh ot 1gh o. 7r{ghO)-l7r{ghohl)' ... ,
an 7r{gho· .. hn -1) -lgho· .. hn -1. 7r{gho· .. h n -1) -l7r{gho· .. hn ),
7r{gho· .. hn ) -lgh o· .. hn )·
Note that 7r{gho··· hn ) -lgh o· .. hn = 7r{hgtlhg = 7r(gt1g since
7r{hg) = h 7r(g).
It is now a straight forward matter to check that
idn - ino~ = bn +1Dn + Dn -1b n ·

On the other hand, by the following theorem, ~ is a
morphism between the Hochschild complexes. In fact, it even
commutes with the cyclic operator t n +1. Therefore, {~} is a
homotopy inverse of the chain map
i: L(G h , A, (h}) t:.+ L(G, A, x).
This shows the desired isomorphisms.
Remark. In particular, when A = (, and ()" is trivial, this

theorem recovers Burghelea's original reduction theorem.
Therefore, ~: L(G, x) -t L(G h , (h}) on generators is given
by [8]
~(ho, ... , h n )
= (7r{g) -l7r(gho), 7r{gho) -l7r{ghohl)' ... ,

7r{gho· .. hn _lt l7r(gho· •. hn )),
244 R. Ii
(2. 4) Theorem. With the notations as above, we have
(1) 7rf 0-0 obo = boon!;

(2) 7rfo- o obo' = bo'on!;
(3) n! ot O +1 = t o+10 n!.
This is proved by a straight forward calculation.
§3. A H*(G x ; k)-module structure on HH*(L(G, A, x)).
When the field is k, the group cohomology of the group

Gx with coefficients in k is denoted by H*(G x ; k). In this
section we will show that there is an H*(G x ; k)-module
structure on HH*(L(G, A, x)) in a natural way.
Let tp be a Hochschild n-cocycle on L(G h , A, (h») and

'I/J be a group cohomological m-cocycle. Since HH*(L(G, x))
~ H*(G x ; k) [2, 8], a co cycle 'I/J on G x corresponds to a
Hochschild cocycle 'I/J on L(G, x). We define tpU'I/J a

Hochschild n+m-cocycle by
-
'l/J(1, hn+l' hn+lhn+2' ... , hn+t"· .h n+m),
In fact, this cup product is well-defined at cochain level,
and one verifies that b(cpU'I/J) = bcpU'I/J + (-l)ncpUd'I/J, where d
is the differential in the group cohomological cochain complex.
This shows that cpU'I/J is a Hochschild co cycle if cp and 'I/J are
cocycles. Moreover this cup product is associative in the sense
that if cp is a Hochschild cocycle and 7/J, 'TJ are group

cocycles, then (cpU'I/J)U'TJ = cpU ( 7/JU'TJ). Therefore, we have
(3.2) Theorem. There is an H*(G x ; k)-module structure

defined by (3.1) on HH*(L(G, A, x)).
§4. Module structures on HC*(L(G, A, x».
As the Hochschild cohomology of L(G, A, x), there is
also a module structure on HC*(L(G, A, x)). When x
contains no torsion elements, HC*(L(G, A, x)) is a module

246 R. Ji
over the cohomology ring H*(N x ; k). However, this module

structure is nontrivial. We proceed to give a construction
which is purely analytical and may be applied to topological

situations as well. To make the construction we will, first of
all, recall the sharp product in cyclic cohomology [3].
Recall first the notion of universal differential graded
algebra O(A) of a given algebra A [3]. Let

-
A be the
algebra obtained by adjoining an extra unit I to A. Let

- e
on(A) = A e A n. Let d: on(A) --1 on+l(A) be defined by
d«a o+.AoI)ea 1e ... ean) = IGJaoGJ ... GJa n. A right A-module

structure on on(A) is given by the equality
-
(aoGJaC' .ean)·a
- -
= Ej;J(-l)n-Laoe- •. GJajaj+lGJ ... GJanGJ a + aoGJ ... GJ(an-a).
As in [3], (wa)b = w(ab), for all we on, and a, b e A.
This right module structure extends to a right
A-module structure on on(A). One then defines the product:

OixOj --1 Oi+j by
-
w·(b oGJb 1GJ ..• ebj ) = wb-oeb 1GJ ... ebj , V we Oi.
This defines an associative product such that

d(wO) == (dw).O+ (-l)degWw.dO, V w, Oe o.
A cyclic n-cocycle cp on A corresponds uniquely to a

graded n-trace cp on O(A). That is, cp(a o, ... , an) -
~
cp(aodaC· .da n). It is clear that aOda l ·· . dan = aoQ)·· ·Q)an.
Recall that there is a sharp product #: HC*(A) Q)

HC*(B) -i HC*(AQ)B) defined by Connes which is an
associative product. Let cp be a cyclic n-cochain on A. We
- - Q)n -
will define cp: AQ)A - i k by cp(ao+AoI, a l, ... , an) =
~
cp(a o, a l, ... , an) = cp(aodaC·· dan)· Then there is a canonical

~ ~
map 11": O(AQ)B) -i O(A) Q) O(B), where Q) is the graded

tensor product of graded differential algebras. The element
(ao+AoI)Q)alQ)· . ·Q)an can be identified in O(A) with
(ao+AoI)da l · .. dan·
Let cp be a cyclic n-cocycle on A, and 'I/J be a cyclic

m-cocycle on B. The sharp product cp#'I/J is then defined by
(4.1) cp#¢(aoQ)b o, ... , an+mQ)b n+m)
= cp Q) 'I/J 011"(( aoQ)bo)d( a 1Q)b 1)· •• d( an+mQ)b n+m))'
where the differential operator d is the total differential on
248 R. Ji
Now consider the homomorphism p: A xo.G ~
(A xo.G)GDkG defined by p(E a~) = E aggGDg. Then p*( rp#'I/J)

e HC*(A xo.G), for cp e HC*(A xo.G) and 'I/J e HC*(kG). We
will denote by cpU'I/J the cyclic cohomology class p*( rp#'I/J).
One checks that cpU'I/J defines an associative product which
yields a right HC*(kG)-module structure on HC*(A xo.G).
(4.2) Definition. cpU'I/J = p*( rp#'I/J).
(4.3) Theorem. 1) HC*(L(G, x)) is a ring under the cup
product defined abovei

2) HC*(L(G, A, x) is a module over HC*(L(G, x)) via the
cup product defined above. Moreover, if x contains no
torsion elements, then HC*(L(G, A, x)) is a module over
H*(Nxi k).
Proof: According to the discussion above, one simply observes
that the cup product, U, so defined decomposes with respect to
conjugacy classes. When x contains no torsion elements,
HC*(L(G, x)) ~ H*(Nxi k). This isomorphism was obtained in
[2] by a topological method, but we obtain it in [8] by an

analytical construction which is along the line of this paper.
§5. Connes' periodicity operator.
The module structure constructed in the previous section

does not mean much if one can not recover the periodicity
operator S of Connes. In fact, to recover the periodicity
operator is the core of the subject ever since the work of
Burghelea [2]. To this end, we denote by u the group
cohomology cocycle on Nh associated to the group extension
(5.1) 7lh ~ Gh L Nh .
We also require that u is Nh-invariant and is normalized,
that is, u(ggo, gg1' gg2) = u(go, g1' g2)' and u(go, g1' g2) = 0,
where g, go' g1' g2 are in Gh , and where g = q(g), whenever
two entries are the same. When h is of infinite order, uoq is
cohomologous to O. Thus there is a normalized one cochain A
over Gh such that dA = uoq. Then as in [8], one can choose
in a canonical way that A(lG' h) f o. (In fact, if [u] f 0 in
H2(N h; k), then A(l G, h) :I 0 is independent of A chosen; and
if [u] = 0, then one can choose A such that A(l G, h) = 1.)
(5.2) Theorem. If x contains no torsion and hEX, let cp E

HC*(L(G h, A, (h»)) ~ HC*(L(G, A, x)), then
Scp = [41ri/A(1, h)] cpUru ,
250 R.Ji
where 1"u E (L 2(Gh, (h») * is defined in [8]

by 1"u(ho' hl' h 2) = u(I, n1, n1n2), for hohlh2 = h, and hi E
Gh ·
Proof: One notes first that 1"u is cyclic since u is normalized
[8]. For ci = aih i, i = 0, 1,
... , n+2, and h o·· .hn +2 = h, we compute that
(5.3) cpU1"u(c o, ... , cn +2) = ~1~i<j~n+2(-I)i+j-l
A
cp(c odc 1· .. dc·1-1· c··

1 dc·1+ 1··· dc·J- 1· c··
J dc·J+ 1··· dc n+ 2)·
A
1 1+1·· ·h·J - 1· dhJ.. hJ. +1··· h n+2) .

1"u(h o·· ·h·1-1 ·dh··h·
A
Note once again that cp(c odc 1·· ·dcn) = cp(c o, ... , cn), and I{J
is a graded trace on O(AxQG) of degree n.

To identify SI{J one may assume I{J is normalized, that
is, cp(c o, ... , cn) = 0 whenever for some i, Ci =1 in AxQG.
Let 1"}.(ho, h 1) = A(I, h 1), for hoh1 = h. Note that 1"}. is not
a cyclic cocycle since A is not a cocycle. Let 1/1 E
(Ln+1(G, A, (h») * be defined by
(5.4) 1/J(co, ... , cn +1)
= ~j!l (_I)j+l cp(c odc 1·· .dcj-1· cr dcj+1". ·dcn+1)·
A
{1"}.(hj +1·· .h n+1h o·· .hj _1dhj ) +

A
~nk+~ 11"'\.(hk+1···hn+1h O•• ·h·J -1dhJ.. hJ. +1·· .hk)} ,

=J +"
It is nontrivial to verify that b'I/J = cpUTu and that B'I/J

= [(n+1)(n+2)'x(1, h)j2] - cpo
In fact,
b'l/J(c o, ... , cn +2)
+ (_1)n+2_'I/J(cn +2c o, c 1' ... , cn +1)
= };o<i<j~n+2
"'/c 1 1+1- - -dc·J - l- c.-
dc 1- - -dc·1-l-c.-dc.
'f'\ O J dcJ. +1- - -dc n +2)-A 1J"
..
Case 1): j = i+1.

Ai( i+1) = (-1)i-l_(-1)i+l_{T>,.(h i +2- - -hn+2ho- - -h idh i +1)
A
+ };~!~+2T>,.(hk+( - -hn+2h o- - -hi-dhi+chi+2- - -h k)}

+ (-1 )i_ (-1 )i+l_ {T>,.(hi +2- - -hn+2hO- - -h i -1d(h ih i +1))
A
+ };~!~+2T>,.(hk+C - -hn+2h O- - -hH-d(hihi+l)-hi+2- - -hk)}

A
+ (-1) i+1_ (-1) i+1_ {T>,.(h i +1- - -h n +2hO - - -h i -1dh i)

A
+ };~!~ +2T>,. (h k+1- - -h n +2h O- - -hi -1- dh i -(hi +lhi +2) - - -h k)}
-- (- 1) ( i-1) +( i+1) {(h
T>,. i+2- - - h n+2h 0- - -
h idh i+l)
A
- T>,.(h i +2- - -hn+~o- - -h H d(hihi+l))

A
+ T>,.(h i +1- - -hn+2h o- - -h i -1dhJ}

= (_1)2i_Tu (h o-- -h·1- l-dh.-dh.
1 1+ 1-h.1+ 2- - -h n+·
2)
252 R.Ji
Case 2: i < j-l.
+ ~~!I+1T~(hk+1" ohn+2hoo ohj _1dhr h j +1" ooh k)} 0 0
+ (-l)i(-l)j{T~(hj+1" ohn+ 2h oO ohj _1dh j ) 0 0
+ ~~!I +1T~(hk+1° hn+2hO hj -1dhr hj +1 h k)}

0 0 0 0 0
0
0 0
+ (_l)j-l(-l)i+l{T~(hi+1" ohn+ 2h oO oh i _1dh i)

A
0 0
+ ~~!~+1 kfj-1T~(hk+1° ohn+2hoO Ohi_1dhi"hi+l° ooh k)} 0 0
. • 1
+ (_l)l(-l)l+ {T~(hi+1" ohn+2hoO oh i _1dh i) 0 0
+ ~~!~+1 kfjT~(hk+1" 0 ohn+~oo ohi-1dhiOhi+l° oh k)} 0 0
= (-1)(j-1) +( h1) 0
{T. . (h.l +1
1\
000 h n +2hO 0 0 0 h.1-1dh 1. h·1+ 1 0 0 0 0 (h.l -I h.))-
l
T)..«hjhj +1) 0 0 0 hn+~O 0 0 0 hi-1dhi" hi +1 0
0 0 h j -1)}
= (-l)i+j{[).(l, hi" 0 ohj ) - ).(1, h i +1o ohj )] 0 -
[).(1, hio ohj _1) - ).(1, h i +1o ohj_Un

0 0
= (_l)i+j -l T U
(h oo oh·1-1dh.oh.
1 1+ 1 oh·l -1dh.oh.
0 2)
l l +1" ohn+' 00 0
The last equality uses the facts that d)' = uo q, and that Tu
is a graded trace of degree 2. This shows that b7jJ = tpUTu'

Now let's compute B7jJ, where B is Connes' B
operator. Since cp is normalized, we have (all inferior indices
are mod n+1)
B7jJ(cO' ... , cn )
= ~Y=o (-1)ni'l/J(l, Cil ... , Cn' co, ... , Ci -1)
''''ldc-··
'f'\ 1 ·dc-1 +J-- l·C-1 +J-.dc-+-+
1 J l •· ·dc-+
1 n ).
A
{ T).(hi +j +1" .. hi +nhi" .. hi +j -ldh i +j) +

A
};k=j +IT).(h k+i +1· .• h n+i +lhi" .. hi +j -ldh i +j" hi +j +1· .. hi +kH
= };Y=o (-l)ni{};j=O (-l)j (_l)j(n-j ).
{A(l, h i +j ) + };k=j+l(A(l, hi+j"· ·h i +k) -

A(l, h i +j +1"· .h i +k))}
= };Y=O (-1 )ni{};j =0 (-l)j(-l)j (n-j) (_l)n( i+j) •
!p(co, c l' .•. , Cn)·
{A(l, h i +) + };k=j+l(A(l, h i +j ·· ·h i +k)-

A(l, h i +j +1"· .hi+k)H
= !p( Co' C1' ... , Cn)· };Y=O };j =0 {A(l, hi +j) +
};k=j+l[A(l, hi+j"· .h i +k ) - A(l, h i +j +1·· .h i +k )]}

= !P(C o, c 1' •.. , Cn)·};i=o {A(l, hi) + A(l, hih i +l) +
... + A(l, hi" .. hi+nH

= [(n+1)(n+2)A(1, h)/2]·!p(c o, c l' •.. , cn)-
Here we notice that hi" •• h i +n = h o· .. hn = h since hi E Gh,
for all i_ The last equality is a nontrivial fact which uses the
facts that dA = uoq and that u is normalized_ We leave the
proof to the reader_
254 R. Ii
This shows that Scp = [47Ti/ 'x(1, h)]. cpUru. [3, Ch. II].
(5.5) Corollary. Suppose that x E (G), and hEx is
nontorsion. If cd~Nh < CD, then PHC*(L(G, A, x)) = o.
§6. Topological crossed products and applications.
Throughout this section G will denote a finitely

generated discrete group, and l will denote a fixed word
length function. Let A be a Banach algebra upon which G
acts as a group of isometric automorphisms by a. Let
ll(A, G, a) be the completion of AxaG with respect to the
norm 11·111 defined by II E aggll 1 = E II agll, and let

C~(A, G, a) be the completion of the regular representation of
AXaG on l2(Gj H) if A is a C*-subalgebra of B(H), where
H is some Hilbert space.
Let S(Gj A) denote the lCrapid-decay functions on the

group G with values in A. That is,
(6.1) S(Gj A) = {cp: G ~ A I
EgEG II CP(g)II(1+l(g))k < CD, for all k > a}.
One can prove easily that if one equip S(Gj A) the product as
that ofAxo.G:
f(J*'If;(g) = l1tEG cp(h)ah-1( 'If;(h -lg»,
then S(Gj A) becomes an dense subalgebra of il(A, G, a). By
a similar argument as in [12) that SeA, G, a) is a smooth
subalgebra of il(A, G, a). Recall from [7) that a subalgebra
A of a Banach algebra A is called smooth if AGDMn(() is
stable under holomorphic functional calculus in AGDM n((), for
all n.
IT G is of polynomial growth [5), and A is a

C*-algebra, then S(G, A) can also be identified with a
smooth dense sub algebra of C*(A, G, a). This can be done by
a similar argument to the case when A is ( as in [7).
(6.2) Theorem. [11] If A is a C*-algebra and G is a group of

polynomial growth, then S(G, A) is a smooth and dense
sub algebra of C*(A, G, a).
A special case of this theorem where G is a finitely

generated abelian group has been obtained in [1). There is also
a more general version of this theorem in the case when a
256 R. Ji
twisted crossed product is considered.
We will denote by S(Gj A)# the cyclic object

associated to the algebra S( Gj A) in the category of Frechet
spaces. Let 8 be the projective tensor product. For a
conjugacy class x E (G), let hEx, and Ln(G, A, x) be the
A
closure of Ln(G, A, x) in S(Gj A)8(n+l). Then
S(Gj A)# ~ iXE(G)L(G, A, x),

where the bar over Ell is the notation for taking closure with
respect to the topology on S(Gj A)#. Therefore, following
Burghelea, we again have
(6.3) HH*(S(G; A)) ~ iiXE(G)HH*(L(G, A, x)),
and HC*(S(G; A)) ~ iiXE(G)HC*(L(G, A, x)),
where n denotes the restricted direct product. That is, a
cocycle nXE(G) lJ'x defines a cohomology class in the right hand
of (6.3) if and only if n lJ'x defines a continuous linear
functional on S(G; A)8n+l. Moreover, Connes' periodicity

operator S decomposes respectively.
Now suppose that G is of polynomial growth. All

results concerning L(G, A, x) in previous sections will still be
true for L(G, A, x) except that in section 2 we need 7r(g) =
gS(g)-l such that l(s(g) = min {l(g·s(g) I g E Gh}. This

latter requirement ensures that all morphisms and maps in
previous sections are continuous. However, we must keep in
mind that the usual group cohomology H*( G h; () and
H*(Nhi () must be replaced by the corresponding Schwartz
cohomology [7]. It is fortunate that when G is of polynomial
growth so are Gh and Nh, and by the results of [7] the
usual group cohomology of these groups with complex
coefficients is canonically isomorphic to the Schwartz
cohomology of the corresponding groups.
Since groups of polynomial growth have finite rational

cohomological dimensions ([7]), we see that Connes'
periodicity operator Sx on HC *(L'(G, A, x» is nilpotent
when x contains no torsion elements by the results in §5.
Therefore, we have
(6.4) Theorem. Suppose that G is of polynomial growth. If

x E (G)", then the periodic cyclic cohomology
PHC*(L'(G, A, x» = o.
We can finally say some thing about the crossed product

258 R. Ii
C*-algebras C*(A, G, a) for a C*-algebra A and a group

G of polynomial growth.
(6.5) Theorem. a) Suppose that A is a C*-algebra and G

is torsion free and is of polynomial growth acting on A as a
group of automorphisms by a. Let r be any tradal state on
C*(A, G, a), and let e = E egg be a projection in
C*(A, G, a). Then r(e) = r(e1(j)' Moreover, if A contains
no non-trivial projections then A®C*(G) contains no
non-trivial projections.
b) Similar results will be true for the £1-algebra £1(A, G, a),
where A is a Banach algebra, and G is of polynomial growth
or is word hyperbolic [6; 8] (Projections in a) should be
replaced by idempotents).
The proof of this theorem uses the fact that S(G; A) is

smooth and dense in C*(A, G, a) (resp. in £1(A, G, a) in
b)). Therefore, their K-theories are the same. For any
projection e in S(G; A) and any tradal state r on
C*(A, G, a), the pairing between cyclic cohomology and
K-theoryof S(G; A) satisfies <r, e> = < Ch[r], [e]> [3]. A
localized form of this equality was given in [7] which states
that the pairing between cyclic cohomology and homology of
S(G; A) satisfies <rx ' ex> = < [Chr]x' (Ch[eD x >' where
the subscript x is a conjugacy class of the group, and the
letter with such a subscript represents the cyclic (co )homology
class restricted to the conjugacy class x. Since the periodic
cyclic cohomology PHC*(t(G, A, x» = 0 for those

nontorsion conjugacy classes x in <G> by Theorem (6.4),
the theorem follows.
References
[1] J. Bost, Principe D'Oka and K-theorie et Systeme

Dynamiques Non-commutatifs, IHES preprint, 1989.
[2] D. Burghelea, The cyclic homology of the group rings,
Comment. Math. Helvetici, 60 (1985), 354-365.
[3] A. Connes, Non-commutative differential geometry, Pub 1.
IHES 62 (1985),41-144.
[4] , Cohomologic cyclique et foncteurs Extf, C.
R. Acad. Paris, 296 Serie 1(1983), 953-958.
[5] M. Gromov, Groups of polynomial growth and expanding
maps, Publ. IHES, 53 (1981), 53-73.
[6] , Hyperbolic groups, in "Essays in group
200 RH
theory", ed. S. M. Gersten, MSRI Publ. 8, Springer {1987},

75-269.
[7] R. Ji, Smooth dense subalgebras of reduced group
C*-algebras, Schwartz cohomology of groups, and cyclic
cohomology, J. of Funct. Anal. Vol. 107 {1992}, 1-99.
[8] _ _ _ _, Nilpotency of Connes' periodicity operator
and the idempotent conjectures, to appear in K -theory.
[9] , Trace invariant and cyclic cohomology of
twisted group C*-algebras, IUPUI preprint {1992}.
[10] , Module structures on cyclic cohomology of
certain group graded algebras, in preparation.
[11] R. Ji and L. Schweitzer, Smooth and dense subalgebras for
twisted crossed product C*-algebras, IUPUI preprint, 1999.
[12] P. Jolissaint, K-theory of reduced C*-algebras and
rapidly decreasing functions on groups, K -theory, vol. 2
{1989}, 729-796.
[13] M. Karoubi, Homologie cycliq:ue et K -theory algebrique II,
C. R. Acad. Paris, Serie 1297{198-1}, 519-518.
[14] V. Nistor, Group cohomology and the cyclic cohomology of

crossed product, Invent. Math. 99{1990}, -111--12-1.
Email-address: igesl00@indycms.iupui.edu
q-RELATIONS AND STABILITY
OF C·-ISOMORPHISM CLASSES
P.E.T. J0RGENSEN,1 L.M. SCHMITT AND R.F. WERNER2
ABSTRACT. In this paper we consider a clau of relations which is shown to

generate C·-algebras. Our emphasis is on a deformation clau which inter-
polates between Fennions and Bosons, and we give conditiona for stability
of the C·-isomorphism class, as the deformation parameter varies. The
base-point C·-algebra is the universal C· -algebra on a Hilbert space, which
is also called the Cuntz-Toeplitz-C·-algebra, and (in the finite rank case) is
an extension of the Cuntz-algebra 0,. by the compacts.
1. INTRODUCTION
The relations we study come up in the following three connections:

(i) commutation relations [6},
(ii) matrix relations [7} defined by matrix entries of classical Lie groups,
(iii) quantization of quadratic forms [3-1];
and the commutation relations may refer to the multiplicative commutator,
uvu-1v- 1, or the additive commutator uv - vu. Each of the two commuta-
tors may be subjected to one-parameter deformations, or to perturbations.
From universal algebra, we have a familiar notion of the free • -algebra based
on some given relationj but, in the present paper, we shall restrict attention
to a free C· -algebra based on the relation. Since such a C·-construction
may not work for the most general relations, we shall focus here on a lim-
ited class of relations which admits universal C· -algebra completionsj- the
most general relations which are C· -admissible have not yet been describedj
and it appears to be difficult to supply workable universal conditions on a
relation to be C·-admissible.
2. NOTATION AND DEFINITIONS
Let 8 be a set, and let s -+ s· be a bijection, which is involutory, Le.,

s·· = s, Vs E ,8. Let A = A(8) be the free algebra with 8 as a set of
generators, with unit 1 (which we may include in 8 if convenientj then
1· = 1) and with involution extending s 1---+ s·. Let the base field be C,
and let B be some subset of A, 1 rJ. B. If (B) denotes the two-sided ideal
1991 Mathematics S.bject C'a88ijicatioB. 81R50, 47A62, 461.05.

lSupported in part by the U.S. NSF, NATO, a University of Iowa Faculty Scholar
Award, and by the U.I.-Oakdale Institute for Advanced Studies
2 Supported in part by a Heisenberg fellowship and a travel grant by the DFG (Bonn)
262 P. E. T. J¢rgensen, L. M. Schmitt, and R. F. Werner
in A(5) generated by B, then the quotient C = Aj{B} serves as an algebra

on relations given by the subset.
We say that the system consisting of (5, *, B) is C·-admissible if there
is a unital C· -algebra 21, a unital homomorphism, tP : C ---+ 21 with the
following universal property: for all C· -algebras 21, and homomorphisms
'I/; : A ---+ 2i such that 'I/;(b) = 0, Vb E B, there is a unique unital homo-
morphism ~ : 21 ---+ 2i such that the composite homomorphism
is the given homomorphism '1/;. We then say that 21 is the C· -algebra on

(5, B), and that tP is the universal bounded representation of (5, B). It
follows from the definition that any two C· -algebras 2! 1 and 2!2, both satis-
fying the conditions relative to (5, B), are then isomorphic as C· -algebras
j - and it follows that functions f : 5 -+ 5 such that f(s·) = f(s)*,
"Is E 5, extend to C· -endomorphisms if it is given that the associated uni-
versal algebra mapping i E End(A) maps B into {B}. The corresponding
C· -endomorphisms will be called the structure homomorphisms, and are
simply reflecting the functoriality of the definition.
Clearly, a necessary condition for a triple (5, *, B) to be C·-admissible
is that some unital homomorphism 'I/; : C -+ 2i into a C·-algebra 21 exists,
or equivalently, that the relations B can be realized by bounded opera-
tors on some Hilbert space. Since the above map ~ : 2! -+ 21 is a C·-
homomorphism, it has norm 1, and hence for C·-admissible relations and
any x E C the norm of tP(x) E 2! in the algebra on (8, B) is an upper bound
for the norm of '1/;( x) E 2i in any realization 'I/; : C -+ 21 of the relations B
in a C· -algebra 2i. The existence of such a uniform norm bound needs only
to be checked on the generators, since it trivially extends to the algebra C.
If such a bound exists, we can use it to define a semi norm on C, and to
define 2t as the Hausdorff completion of C with respect to this seminorm.
(Note that by the universal property tP(C) c 21 is necessarily dense). We
thus arrive at the following criterion:
Lemma 2.1 [20, 29]. (8, *, B) is C· -admissible if and only if
(i) There is some unital C· -algebra 2i and a unital *-homomorphism
'I/; : Aj{B} -+ 21.
(ii) There is a function n : 5 -+ lR.+ such that for any such homomor-
phism, and all s E 5, 1I'I/;(s)1I ~ n(s).
This Lemma will allow us to conclude the C·-admissibility of most of

the examples below by inspection. However, in general both (i) and (ii)
may require a detailed investigation of a given set of relations. Even more
work is usually required to determine the C·-algebra 2! on a C·-admissible
set of relations. For C·-algebras on generators and relations, we refer the
reader to [3,6,8, 17,20,23,24,30,34].
q-Relations and stability 263
3. EXAMPLES
3.1 The Rotation Algebras. Let S have four elements u, u·, v, and v·,
with the * denoting a pairing, and let 9 be a real number, 9 E (0,1). Let
B be the set whose elements are: uu· - 1, u·u - 1, vv* - 1, v*v - 1,
and uvu*v* - ei2d l, where 1 denotes the abstract unit element. It easy
to construct unit aries with this commutation rule, and hence Lemma 2.1
immediately gives the C*-admissibility of these relations for all values of 9.
21 = 21(9) is called the rotation algebra [1, 6, 28, 30]. It is further known
[25(a), 28] that the C·-isomorphism classes are labeled by K-theory, in fact
by [0,1] n (Z + 9Z). Moreover, Sll(9) is simple iff 9 is irrational.
3.2 The Cuntz-Toeplitz Algebra. Let 1£ be a Hilbert space with inner
product (-,.), and let 1£. be the conjugate space with identification I t---+
r : 1£ -+ 1£* conjugate linear. Consider the set B indexed by 1£ x 1£.,
{fg* - (f,g)l} where the product Ig· is in the universal algebra on 1£,
and 1 is the unit element in that algebra. Again it is easy to check using
Lemma 2.1 that B is C·-admissible, and the C·-algebra on these relations
is called the Cuntz-Toeplitz algebra. It is described in [7,13,17, 25(b), 34],
and, when diml£ = 1, in [9). (We will be mainly interested in diml£ > 1.)
If 1£ is finite-dimensional, and el ... ,ed is an orthonormal basis, then we
may enlarge the set B by the element 1 - Li ei ei, i.e. divide out by the
two-sided ideal generated by this element of the Cuntz-Toeplitz algebra. It
is known, and easy to check, that this ideal is isomorphic to the (non-unital)
C* -algebra of all compact operators on a separable oo-dimensional Hilbert
space.
The corresponding C·-algebra on the extended B-set is known as the
Cuntz-algebra Od. It is simple for each d, and distinct values of d represent
different C* -isomorphism classes. We get a short exact sequence
o-+ C -+ 21 -+ Od -+ 0
in the category of C* -algebras, where C denotes the C* -algebras of the
compacts, and 21 is the Cuntz-Toeplitz algebra on d generators.
3.3 q-Commutation Relations. Let 1£ be a given Hilbert space, 1£.
as in 3.2, and let q E JR.. Consider the set B indexed by 1£ x 1£* with
elements Ig* - qg* 1- (f,g)l, and Ih - qhl, for I, h E 1£, g* E 1£*.
Equivalently, consider the possibility of dividing out by the ideal on the
combined relations:
Ig· - qgo. 1= (f,g)l , (1)

and Ih - qhl = 0 . (2)
Applying these two relations to expressions of the form Igh* - qglh* in

all possible ways shows that (1)-(2) is consistent iff q2 = 1. The two cases
q = 1 (q = -1) are the Bosonic relations (resp., the Fermionic relations),
also denoted CCR, and CAR respectively, see [2,4,8, 16,26]. The CAR-
C* -algebra is a familiar isomorphism class, represented by 200 in the Glimm
264 P. E. T. Jf/lrgensen, L. M. Schmitt, and R. F. Werner
labeling [15], but the CCR-algebra is different; it is well known that it has
no representation by bounded operators, so it is not C·-admissible. But
we still get a C·-algebra (see [6, 30]) by generalizing the construction in
3.1 above: Look at elements u, v indexed by 1£ x 1£., and given by formal
exponentials
u = e*(J-r>,
These formally satisfy the relations
(3)
which are now C·-admissible by Lemma 2.1. The generators u = u(!), v =
v(g) now depend on an argument /,g E 1£, which even for dimll = 1
takes uncountably many values, so the CCR-C·-algebra based on (3) is
non-separable.
In Section 5 ofthe present paper we shall look closer at relation (1) for
q in the interval -1 < q < 1. Since in this interval (2) is not consistent
with (1), we ignore this relation. We show that (1) is C·-admissible, and
thus provides an interpolation between the CCR and CAR. We shall refer
to the C·-algebra on (1) as the q-CCR-C·-algebra. We also show that
the C·-isomorphism class of this algebra is constant for q in the interval
Iql < v'2 - 1; and we expect, in fact, in the full interval, -1 < q < 1.
Several generalizations and modifications of the q-CCR-relations have
been considered in the literature. In [32] the Fock representation of the
relations aiaj = 6ij + qij ajai was shown to be positive, and another mod-
ification, based on the positive definiteness of functions on pair partitions
was investigated in [33]. The techniques of section 5 also apply to relations
in which the term qg. / is replaced by a general, but sufficiently small lin-
g;
ear combination of operators /i [18], or even by suitable transcendental
functions of the generators.
3.4 The C·.algebras U::c. This is the non-commutative (ne) C·-algebra
on the unitary n by n matrices, Le., the classical matrix group Un. The ma-
trix functions (Uij), 1 ~ i, j ~ n, satisfy the familiar relations EI: UtiUl:j =
EI: Ui1: ujl: = 6ijl, that is to say, (Uij) is a unitary matrix in Mn(U::C) ~
U:: c ® Mn where Mn denotes the n x n complex matrix ring. We may
take these as relations for elements in a C·-algebra. By Lemma 2.1 they
are obviously C·-admissible, so we can talk about the universal C· -algebra,
U:: c , on these relations, see [3 ,7, 13, 24, 34], U:: c • This C·-algebra was
studied recently in [24] along with a certain reduced version U::~ed' and it
was shown that U::,~ed has no non-trivial projections, and the ~morphism
classes were also found.
3.5 Woronowicz' S.,U(2). The construction in 3.4 has an analogue for
the other classical Lie groups G, so we get associated C·-algebras Gnc. This
is especially useful for the Lie groups SU(n), and the non-compact group
U(n, 1) of all (n + 1) by (n + 1) complex matrices 9 such that gJg. = J
where J = (~1 1:). This is also the group of matrices preserving the
indefinite sesquilinear form
and it was shown in [7, 34] that U( n, 1)nc may be viewed as a self-similar ver-
sion of the Cuntz-Toeplitz-algebra, see [13]. Each Gnc may alternatively be
reconsidered from the point of view of deformations and Hopf-algebras [35],
leading to the Woronowicz-Drinfeld theory [23, 35] of "quantum-groups."
The 1/ is a deformation parameter, then the C·-algebra is denoted cnC(I/),
but when the classical matrix-group G is non-compact, then the correspond-
ing 1/- relations are typically not C·-admissible, see [27, 34, 35(b)]. The
simplest case when they are C·-admissible is S"U(2) with matrix-entries
(;. -;"Y) , and relations
QQ. = 1- 1/21 • 1 , 11· = 1- Q.Q , Q1* = 1/1*Q

and Q1 = 1/1Q , 1·1 =11· .
We remark that the relations in the second row can be shown to be redun-
dant [18]. When n· is eliminated from the first two relations 'With the help
of the last, we get
QQ. = 1/2Q•Q + (1- 1/ 2 )! (4)
which is, modulo a renormalization, the same as (1) in 3.3 (for dim 1l = 1)
and with q = 1/ 2 • Hence, for q ~ 0 (in this special case), the q-CCR-
C· -algebra is the C·-algebra of the non-normal generator of Woronowicz'
Sy7U(2). The normalization in equation (4) makes these algebras into a
continuous field of C* -algebras having the abelian C* -algebra C(1I» as the
limit at q -+ 1 where l!) = {z E C : Izl < I}. (The limit at q = -1 has
been studied by [21,19], and has an interpretation as a "non-commutative
circle"). However the simple interpretation of the q-CCR-C*-algebra as
a deformed disk is special to the degenerate case dim1l = 1, and we now
turn to dim1l > I. This case turns out to the more "rigid" and more
non-commutative.
4. SYMMETRIES
We include a brief review of some of the structure-homomorphisms of

the relations from Section 3 above, with emphasis on the automorphism
groups which come from symmetries of the respective underlying relations:
Each example has the circle 'll' =
{z E C : Izl =
I} acting as a group
of automorphisms, and it is strongly continuous, as can be seen from
the C·-construction of the universal C* -algebra 2l (for each of the con-
sidered relations). This means that the action T : 'll' -+ Aut(2l) sat-
isfies: Tzw(A) = Tz(Tw(A», VA E 2l, Vz, w E 'll'j Qz(A·) =
(Qz(A»*,
Qz(AB) = Qz(A)Qz(B), (A, B E 2l)j and IIQz(A) - All -+ 0 as z -+ 1.
This action allows us to make a Fourier-series expansion for elements A
266 P. E. T. J!Ilrgensen, L. M. Schmitt, and R. F. Werner
in the C*-algebra, i.e., A ,... LnezAn where An := fTz-nrz(A)dz E 21,
with dz denoting the normalized Haar measure on 11'. The Fourier-series
is important for problems in the literature (see, e.g., [5, 6, 8, 13, 30]), for
each of the examplesj and, in particular, for the q-relations of [4, 16 ,17],
and Section 3 above.
The generators are scaled as follows in the respective examples: Let
z E 11'. In Example 3.1, the scaling is u -+ zu, v -+ ZVj but the bigger group
11'2 = 11' x 11' acts naturally as well, with: u ~ zu, v ~ WV, (z, w) E 11'2.
Thus bigger scaling-symmetry also applies to the CCRs (3) from Example
3.3. But the additive relations in Example 3.2-3.3 are scaled by: I ~ zl,
and therefore g* ~ %g*, IE 1{., 9 E 1{.*. Notice also that (3) is scaled by
the symplectic group consisting of real linear maps S : 1{. -+ 1{. such that
=
Im(SI,Sg) Im(f, g), VI, 9 E 1{.. Similarly, the relation, Ig* (/,g)1 in =
Example 3.2 is scaled by the generalized U(n, 1) group G as follows: This
is the co-dimensional group G of operator-matrices
where a E C, A : 1{. -+ 1{. is linear, and hI, h2 E 1{. are restricted by the
"isometry" conditions:
for all z E C, I E 'H. Equivalently, A*h2 = ahl.
and
It was shown in [7, 34] that this group acts as an automorphism-group on

the Cuntz-Toeplitz algebra, and on the Cuntz-algebra for every 1{., finite-
or infinite-dimensional. In the q-CCRs of Example 3.3 the whole unitary
group of 1{. acts by structure homomorphisms.
The examples in Section 3 involving a deformation-parameter show that
the presence of a deformation may reduce the "size" of the group of struc-
ture automorphisms. In Example 3.4, note that both U;:c and SU;:c carry
their doubled classical classical counterparts Un X Un (resp. SUn X SUn)
as a group of structure-automorphisms. This is because each of the sets
of relations is invariant under u ........... guh-l, when g, h E Un (resp. SUn).
On the other hand, the relations defining the one-parameter deformation
S.,U(2) of SU(2) ( see Example 3.5 ) imply those of SU;:c with the ap-
propriate identifications of matrix elements. Hence S"U(2) is a quotient of
SU;:c, by an ideal that turns out not to be invariant under all the structure
automorphisms of SU;:c. One checks easily that the only structure homo-
morphisms of SU;:c leaving invariant the relations of S"U(2) are those with
diagonal 9 and h, which correspond to multiplying the generators Q and 'Y
with arbitrary phases.
5. GRADED q-DERIvATIONS
Before describing the C* -theory of the q-CCR

Ig* - qg* 1= {f,g)1 (1)
when 11. (and conjugate 11.*) is given, we note the following covariant
representation on the full tensor algebra T(1£*) on 11.*, i.e., T(1I.*) =
Cl + 11.* + 11.* @ 11.* + ... with 1 representing a Fock-vacuum vector: Let
g* E 11.* be represented by left-multiplication on T(1I.*), i.e., A(g*)e := g*e
as product in T(1I.*), and let 0 be given as a graded derivation by (f E 11.):
0(1) = 0, o,(g*) = (f,g)l, g* E 1£*; and
0,(66) = 0,(6)6 + qdegelelO,(6)
for el, 6 E T(1I.*) such that 6 has a homogeneous degree relative to the
T(1I.*)-grading, i.e., deg(gr ... g~) = n. Then (1) shows that we have a
representation of our q-algebra, i.e., that
(5)
holds for all I E 11., g* E 1£*. Moreover, this representation is universal.
These structures can be generalized to the case that the "twist" map e 1-+
qdegee is replaced by a general linear transformation [18]. In this general
context equation (6) in the followig Lemma defines a hermitian, but not
necessarily positive semi-definite inner product.
Lemma 5.1 [4, 14, 36]. There is a unique positive definite inner product
(.,.) on T(1I.*), extending the inner product given on 11.*, such that (1,1) =
1, and
(0,6,6) = (6, )..(1*)6) (6)
for all I E 11., V'el,6 E T(1I.*). Moreover, the operators A(I*),O, are
bounded with respect to this inner product.
The representation (5) is called the Fock-representation, and (6) shows
that there is a unique (up to scale) positive definite inner product making
it a unitary representation. (A representation ?r is said to be unitary iff
?r(f)* = ?r(I*), 'II E 11..)
The following technical lemmas are basic to our stability theorem for the
q-CCRs.
Lemma 5.2 [17]. Let 1£ be a C-vector space with a sesquilinear form (', .),
let 21 be a C* -algebra, and a : 11. -+ 21 a mapping s. t.
a(f)a(g)* - qa(g)*a(f) = {f,g)1 ,
'II, g E 11.. Then it follows that (" .) is positive semidefinite (Iql < 1, q E ~),
and
lIa(f)1I2 = { (1- q)-l(f,/), 0 ~ q < 1
(f, f), -1 < q ~ 0 .
Lemmas 5.1 and 5.2 together establish the hypotheses of Lemma 2.1,
and we conclude that the q-CCR are C* -admissible. We will denote the
C·-algebra on the q-CCR by 21q(1I.) (denoted by Cq(1I.) in [17]), and by tPq
the universal bounded representation of the q-CCR in 21q(1I.).
268 P. E. T. JI/lrgensen, L. M. Schmitt, and R. F. Werner
Lemma 5.3. (A Wold-Decomposition.) Let 1l =F 0 be a complex Hilbert
space, Iql < 1, and 11" a unitary *-representation of ~f(1l) acting on a
Hilbert space /C. Consider the following subspaces of /C:
/Coo = {4>E/CIVJE1t1l"(I)4>=O}
/Co = lin { 11" (Ii ... f~) 4> I n EN, fi,··· , rn E 1l" , 4> E /Coo}
/Cl = n
nEN
lin {1I"(li ... f~)1/I1 fi,··· , f~ E 1l", 1/1 E /C}
Then /Co and /C 1 are 1I"-invariant, and /C = /Co$/C 1 • This invariant decom-
position is uniquely characterized by the property that 1I"1~o is a direct sum
of isomorphic copies of the Fock-representation, and /C 1 n /Coo = to}.
Moreover, ifdim1l = 1, and /C 1 =F to} then -vr=i1l"(f)1/C 1 is unitary for
every unit vector f E 1l.
The proof of this result is given in [18]. The Wold decomposition from
single operator theory [22] provides a canonical direct sum decomposition
for an arbitrarily given isometry as a sum of a unilateral shift (with mul-
tiplicity) and a unitary. In this analogy, the Fock representation 11"0 corre-
sponds to the unilateral shift. As in the single operator case the comple-
ment /C 1 is characterized by the property that every vector can be given an
arbitrarily long "iteration history" with respect to the operators 11"(1*).
We remark that there are, for every q E (-1,1), non-trivial representa.-
tions with /C = /Cl. In fact, if h E 1l is a vector with IIhll = 1 there is a
unique state Wh such that
Such "coherent states" were studied in the case q = 0 in [7], and, for general
q, in [19]. The GNS-representation associated with such a state contains
no Fock component /Co.
At least in a limited range of q one has the following alternative descrip-
tion of this decomposition: if Iql < 1/2, and dim1l < 00, one can show [17]
that 0 is an isolated eigenvalue of
Hence the spectral projection P of R2 belonging to to} is also in ~f(1l). On

the other hand, 1I"(P) is the projection onto /Coo. Hence 11" is a representa.-
tion with /C = /Cl, if and only if1l"(P) = O. In the case of the Cuntz-Toeplitz
algebra (see section 3.2. above), the quotient of ~0(1l) by the ideal gener-
ated by P is the Cuntz algebra, so Lemma 5.3. describes the decomposition
of a general representation of the Cuntz-Toeplitz algebra into a multiple of
the Fock representation and a representation ofthe Cuntz algebra Odim1t.
The following Theorem is based on an extension of this analysis to the
q-CCR, and identifies ~f(1l) for small q.
Theorem 5.4 [17]. Let q E JR., Iql < -12 - 1 be given, and let f/Jo be
the universal representation of the Cuntz- Toeplitz C· -algebra 2(0(1l) over
a finite-dimensional Hilbert space 1£. Let P = 1- f/Jo(E. ete.), where {e.l
is an orthonormal basis in 1£. Then 2(11(1£) = 2(0(1£), and the universal
bounded representation f/JII of the q-relations is determined by
f/J1I(f) = f/Jo(f)R for all f E 1£ ,
for a unique R E 2(0(1£) such that R 2: 0, and RP = O.

Corollary 5.5 [17]. If dim1£ < 00, and Iql < -12 -I, then there is a short
exact sequence in the category of C· -algebras
0-+ C -+ 2(11(1£) -+ 0(1£) -+ 0 ,
where C is the C· -algebra of the compact operators (on a separable Hilbert

space), and 0(1£) is the Cuntz-algebra on 1£.
Remark. The proof in [17] is based on the iteration
1
f(R) = (1- P + q L f/Jo(ei)* Rf/Jo(ej )*f/Jo(ei)Rf/Jo(ej») 2" ,
ij
which is contractive in the range of q given in the Theorem, and has R

as its unique fixed-point. The range of q determined in this way is almost
certainly not optimal. For example, Dykema and Nica [12] have shown
that, at least in Fock representation of the relations, the interval can be
extended a little bit.
Our argument also applies to other relations (see [17]) of the form
aia; = rij(at. ... ,ad), provided all bounded solutions of these relations
are bounded by the same constants (compare Lemma 2.1), the matrix rij
is bounded away from zero, and the matrix elements are Lipshitz con-
tinuous in (at, ... ,ad) with sufficiently small constant. Then there is an
isomorphism of C· -algebras, constructed as in 5.5, such that
In general, the stability interval will depend on d: For the case, aia; =
6ij 1 + qat aj, 1 :5 i, j :5 d, our Theorem in [17] gives isomorphism of the
q-dependent C·-algebras, for Iql < (v'5 - 1)j(2d). Even though there may
possibly be a bigger stability-interval, the true interval must depend on d,
since one can show [17] that apart from some discrete values of q there are
no bounded representations for q < -ljd.
270 P. E. T. J!IIrgensen, L. M. Schmitt, and R. F. Werner
ACKNOWLEDGEMENTS
The present paper is based on the joint article [17] by the co-authors,
and a lecture by one of us (RFW) at The University of Iowa-Great Plains
Operator Theory Symposium (May 1992). It represents research done while
the authors have benefited from hospitality at the following institutions:
PETJ and RFW at Oberwolfach, RFW at Universitat Tiibingen, and The
University of Iowa, and PETJ at the University of Pennsylvania and the
University of Calgary. The coauthors acknowledge discussions with B.
Kiimmerer, R. Speicher, G. L. Price, J. Cuntz, S. L. Woronowicz and B.
Brenken. We would like to thank the referee for his/her careful reading and
considered suggestions.
REFERENCES
1. J. Anderson and W. Paschke, The rotation algebra, Houston Math. J. (1992).
2. L. C. Biedenharn, The quantum group SUq (2) and a q-analogue of the Bo.on op-
erator., J. Phys. A 22 (1989), L873-878.
3. B. Blackadar, Shape theof'J for C* -alge6rtu, Math. Scand. 56 (1985), 249-275.
4. M. Boiejko and R. Speicher, An e2:ample of a generalized Brownian motion, Comm.
Math. Phys. 131 (1991),519-531.
5. O. Bratteli, G. A. Elliott, F. M. Goodman and P. E. T. Jt&rgensen, Smooth Lie
group action. on the irrational rotation algebrA, Nonlinearity 2 (1989), 211-286.
6. O. Bratteli, G. A. Elliott and P. E. T. Jt&rgensen, Decompo,ition of unbounded
derivation. into inVAriAnt And approrimAtel,l inner part., J. reine angew. Math.
(Crelle's J.) 346 (1984),166-193.
7. O. Bratteli, D. E. Evans, F. M. Goodman and P. E. T. Jt&rgensen, A dichotom,l for
derivAtion. on On, Publ. RIMS, Kyoto Univ. 22 (1986),103-117.
8. O. Bratteli and D. W. Robinson, OperAtor Algebra, And qUAntum .tatiatica.l me-
chanic. I and II, Springer-Verlag, New York, 1979,1987.
9. L. A. Coburn, The C*-Algebra generated bll an i.ometf'J I, Bull. Amer. Math. Soc.
13 (1967), 722-726.
10. A. Connes and M. A. Rieffel, YAng-Mill. for non-commutative two-tori, Contemp.
Math. 62 (1987), 237-266.
11. J. Cuntz, Simple C*-Algebra. generated by i.ometrie., Comm. Math. Phys. 51
(1977),173-185.
12. K. Dykema and A. Nica, On the FocI: repre.entation of the q-commutAtion relAtion.,
J. reine angew. Math. (Crelle's J.) 440 (1993),201-212.
13. D. E. Evans, On On, Publ. RIMS, Kyoto Univ. 16 (1980),915-927.
14. D.I. Fivel, Interpolation between Fermi And Bo.e .tAtillic. u.ing generalized com-
mutAtor., Phys.Rev.Lett. 65 (1990),3361-3364; Erratum, Phys.Rev.Lett. 69 (1992),
2020.
15. J. Gllmm, On A certain cla" of operator algebra., Trans. Amer. Math. Soc. 95
(1960), 318-340.
16. O. W. Greenberg, Particle. with .mall violation, of Fermi or Bo.e .tatidic., Phys.
Rev. D 43 (1991),4111-4120.
17. P. E. T. Jt&rgensen, L. M. Schmitt and R. F. Werner, q-Canonical commutation
relAtion. And .tahilit, of the Cuntz algebra, Pacific J. Math. (to appear).
18. P. E. T. Jt&rgensen, L. M. Schmitt and R. F. Werner, Po.itive repre.entation. of
general commutation relation, allowing Wicl: ordering, Preprint Osnabriick, 1993.
19. P. E. T. Jt&rgensen and R. F. Werner, Coherent datu of the q-ca.nonica.l commuta-
tion relation" Preprint Osnabrock, 1993; archived at funct-antbabbage .aina. it,
'9303002.
20. P. Kruszynski and S. L. Woronowicz, A non-commutatille Gelfand-Naimarlc tkeo-

rem, J.Operator Th. 8 (1982),361-389.
21. G. Nagy and A. Nica, On tke "quantum did" and a "non-commutatille circle",
Article in this volume.
22. B. Sz.-Nagy and C. Foi8lj, Harmonic ana/,l.i. of operator. on Hilbert 'pace, North-
Holland PubI., Amsterdam, 1970.
23. Yu. I. Manin, Topic. in non-commutatille geometry, Princeton U. Press, Princeton,
NJ,I991.
24. K. McClanahan, C"-algebra. generated b, element. of a unitary matN, J. FUnct.
Anal. 107 (1992),439-457.
25. M. Pimsner and D. Voiculescu, Ezact .equence. for K -group. and Em-group. of
certain cro•• -product C" -algebra., J. Operator Theory 4 (198Oa) , 93-118; K -grouPI
oj red'llced crolled product. b, free group., J. Operator Theory 8 (1982), 131-156.
26. R. T. Powers, Repre.entation. of uniJorml, kllJlerfinite algebra. and tke allociated
1I0n Neumann ring', Ann. Math. (1967), 138-171.
27. W. Pusz and S. L. Woronowicz, Twi.ted.econd quantization, Reports Math. Phys.
27 (1989), 231-257.
28. M. A. Rieffel, C" -algebrlll a6l0ciated witk irrational rotation., Pacific J. Math. 93
(1981), 425-429.
29. L. M. Schmitt and R. F. Werner, Non-commutatille Junctional calcu/uI, In prepa-
ration.
30. J. Slawny, On factor-repre.entation. and tke C"-algebra oj tke canonical commu-
tation relation., Comm. Math. Phys. 24 (1971), 151-170.
31. R. Speicher, A non-comm'lltatille central limit tkeorem, Math.Z. 209 (1992),55-66.
32. R. Speicher, Generalized .tati.tic. oj macro.copic field., Lett.Math.Phys. 27 (1993),
97-104.
33. R. Speicher and M. Boiejko, Interpolation. between bOlonic and Jermionic relation,
gillen b, generalized Brownian motionl, Preprint Heidelberg, October 1992.
34. D. Voiculescu, Symmetrie. of ,ome reduced free product C" -algebra., LNM 1132,
Springer-Verlag, New York, 1985, pp. 556--588.
35. S. L. Woronowicz, Twi.ted SU(2)-group. An ezamp/e of a non-comm'lltatille differ-
ential calc'll/ul, Publ. RIMS, Kyoto Univ. 23 (1987),117-181; Unbounded element.
affiliated witk C" -algebra, and non-compact quantum group., Comm. Math. Phys.
136 (1991),399-432.
36. D. Zagier, Realizability oj a model in infinite ,tati,ticI, Commun. Math. Phys. 147
(1992), 199-210.
P.E.T. J0RGENSEN, DEPARTMENT OF MATHEMATICS, THE UNIVERSITY OF IOWA,

IOWA CITY, IOWA 52242, U.S.A.
L.M. SCHMITT, F.B. MATHEMATIK/INFORMATIK, UNIVERSITAT OSNABRUCK, POST-

FACH 4469, OSNABRUCK, GERMANY
Addre.. after April 199!: University of Aizu, Aizu-Walcamatsu City, Fukushima
Prefecture 965, Japan
R.F. WERNER, F.B. PHYSIK, UNIVERSITAT OSNABROCK, POSTFACH 4469, Os-

NABROCK, GERMANY
E-mai/addre.. :reinwerCd08uni1.rz.Uni-Osnabrueck.DE
A test for injectivity for asymptotic
morphisms
Terry A. Loring *
Department of Mathematics and Statistics
University of New Mexico
Albuquerque, NM 87131
The purpose of this note is to prove the following result which gives a suffi-
cient condition for an asymptotic morphism to be associated to a deformation
of C* -algebras. All C· -algebras in this paper will be assumed separable. See
[1, 2] for definitions, etc.
Theorem 1 Suppose A is a C· -algebra such that the quotient map A - AI I

induces a noninjective map on K -theory for I any nonzero ideal. For any C·-
algebraB and any asymptotic morphism (cpt) : A - B, iicp.: K.(A) - K.(B)
is injective then cp is the asymptotic morphism associated to a deformation of
A toB.
This applies, in particular, to A = C(X) for X a compact, orientable

manifold. To see this, consider the fundamental class in cohomology and
the Chern character. Theorem 1 was used in [2, 3] to deduce that certain
deformations exist.
Before giving the proof we must correct a small error from [1] regarding
injectivity. Recall that (cpt): A - B is injective ifliminfllcpt(a)11 > 0 for all
a =I O. Let us call cP weakly injective iflimsup IIcpt(a)1I > 0 for all a =I O. Recall
that associated to cP is a *-homomorphism rj; : A - Boo. It is easily seen from
'partially supported by NSF grant DMS-9007347

A test for injectivity 273
the definition of the norm in Boo that cp is injective exactly when cP is weakly
injective.
To see that weak injectivity is not the same as injectivity, consider any con-
tractable C* -algebra A. Let B = A and let CPt be a path of *-homomorphisms,
parameterized by [1,00), that equals idA for t an even integer and equals the
zero map for t an odd integer.
Given an arbitrary asymptotic morphism (cpt) : A -+ B, one may mod out
by Ker cp and conclude only that cP factors as the composition of a quotient
*-homomorphism and a weakly injective asymptotic morphism. This immedi-
ately gives a proof of a weaker version of Theorem 1: injectivityon K-theory in
these cases implies weak injectivity of the asymptotic morphisms. To deduce
the existence of a deformation, one needs injectivity.
To summarize the relation between injectivity and weak injectivity we give
the following two lemmas which, modulo the above correction, are due to
Connes and Higson.
Lemma 2 For any asymptotic morphism (cpt) A -+ B, the following are

equivalent:
(a) cP is weakly injective,

(b) cp is injective,
(c) lim sup IIcpt( a )11 = lIall for all a E A.
Lemma 3 For any asymptotic morphism (cpt) A -+ B, the following are

equivalent:
(a) cP is injective,
(b) cP is the asymptotic morphism associated to a deformation of
A to B,
(c) lim II CPt (a) II = Iiall for all a EA.

274 T. A. Loring
Suppose (cpt) : A -+ B is any asymptotic morphism and let tn be any

sequence in [1,00) with tn -+ 00. Associated to this discrete parameter we
define a *-homomorphism fjJ : A -+ iJ by setting
and
00
fjJ(a) = (cpt n (a))::':1 + E9 B.

n=1
This reduction to a discrete parameter gives the following proof, due to
Connes and Higson, of (a) => (c) in Lemma 3. If liminfllcpt(a)1I < lIall, for
some a E A, then t" may be chosen so that IIfjJ(a)1I < lIali. Since fjJ is a *-
homomorphism, there must be a nonzero element x in its kernel. For this
x,
liminfllcpt(x)1I :5liminfllcp'n(x)1I = IIfjJ(x)1I = o.
t "
Therefore, injectivity for cP implies
for all a. On the other hand, for any cP and any a E A,
lim sup IIcpt(a)1I = IIt,O(a)1I :5 lIall·

t
One may also reduce to a discrete parameter if one is only interested in

a map on K-theory. To state a precise result, we identify the K-theory of iJ
,g
with a subgroup of
K.(B) / ~ K.(B).
Lemma 4 For any choice oft,,, as above, and fQr any x E K.(A),
00
fjJ.(x) = (cp.(x), CP.(x), ... ) + E9 K.(B).

n=1
Therefor~, cpo : K.(A) -+ K.(B) injective implies fjJ. injective.
Lemma 5 If (cpt) : A -+ B is not injective then there exists t" -+ 00 such

that the associated *-homomorphism fjJ is not injective.
Theorem 1 follows from Lemmas 3, 4 and 5.

A test for injectivity 275
References
[1] A. Connes and N. Higson, Deformations, morphismes asymptotiques et
K -theory bivariante, C. R. Acad. Sci. Paris, t. 310, Series I (1990) 101-106.
[2] M. Da.da.rlat and T. A. Loring, Deformations of topological spaces predicted by
E-theory, these proceedings.
[3] T. A. Loring, Deformations of nonorientable surfaces as torsion E-theory ele-
ments, C. R. Acad. Sci. Paris, t. 316, Serie. I (1993) 341-346.
On the "quantum disk"
and a "non-commutative circle"
Gabriel Nagy Alexandru Nica
Department of Mathematics Department of Mathematics
Kansas State University University of California
Manhattan, KS 66506 Berkeley, CA 94720
e-mail: nagy@math.ksu.edu e-mail: nica@cory.berkeley.edu
Abstract
Let Tq be the universal C*-algebra generated by an element z satisfying
the equation:
zz* = qz*z + (1- q)I,
where q E [-1,1] is a parameter. We show that, in contrast to Tq for
-1 < q < 1, which is known to be the Toeplitz algebra, 7-1 is a (stably)
finite C*-algebra, embeddedable into M2 ® C(D). We prove the continuity
of the field (Tq)-l~q~b by using a remarkable continuous field of states,
(4)q)-l~q~l. 4>-1 is a faithful trace-state of 7_ 1 . For q > 0, 4>q is the
restriction to Tq of the Haar measure on the quantum group S.,fiU(2).
The distribution (with respect to 4>q) of the real part of the generator z is
deformed from the semicircle law ~\..I'f='t2dt, at q = 0, to Itldt, at q = -1.
On the "quantum disk" 277
1 Introduction
By the "quantum disk" we shall understand the universal unital C* -algebra

Tq generated by an element z( = Zq) satisfying the equation:
zz* = qz*z + (1- q)I; (Ll)
here q E [-1,1) is a "deformation" parameter. For q = 1, one adds to (1.1) the

condition "llzll :s; I" - this is quite natural, since "llzll :s; I" is automatically
fulfilled by the generator of Tq for 0 :s; q < 1. It is clear that It ~ C(ll) (which
explains the name "quantum disk" given to Tq ; we mention that an other name
in use is the one of "q-deformed harmonic oscillator" - [2), [8)).
To is the C*-algebra generated by a non-unitary isometry, i.e. the Toeplitz
algebra ([4)). It is known that Tq ~ To for every q E (-1,1). For 0 < q < 1,
the argument proving this fact can be traced back to Appendix A2 of [16] (see
also [9), or [7] where a different parametrization is considered). The remark
that the "stability of Tq" also holds for -1 < q < 0 was made in [6), where a
renormalization of (1.1) is studied, as a particular case of "the deformation of
the commutation relations".
For q = -1, denoting by x and y the real and imaginary parts of z in (1.1),
we obtain that 7-1 is, equivalently, the universal unital C*-algebra generated
by two elements x and y such that:
x = x*, y = y*, x 2 + y2 = I. (1.2)
Hence 7-1 is, in some sense, a "non-commutative circle". We will show here
that:
1.1 Theorem 10 All the irreducible *-representations of 7-1 are 1- or 2-

dimensional. The spectrum of 7-1 is the quotient of II by the relation {A =
-A I IAI < I}.
2 There is an embedding of 7-1 into M2 ® C(ll) which identifies the gen-
0
erator z( = L 1) of 7-1 with:
(1.3)
278 G. Nagy and A. Nica
In particular it follows that (in contrast to Tq for -1 < q < 1) 7-1 is a

(stably) finite C* -algebra.
Now, for 0< q ~ 1, the quantum disks are closely related to the quantum
groups SI.lU(2), 0 < I' ~ 1 (see [16], [13], [2], [8]). For instance, in the description
of [16], SI.lU(2) has two generators a and 'Y, the first of which satisfies (1.1) for
q = 1'2, and actually has C*( a) ~ 7,..2. A consequence of this fact is that, for
o~ Tq has a "canonical state" </>q, obtained by restricting the "Haar
q ~ 1,
measure" of S..;qU(2) (see [16], [17]). We shall put here into evidence a natural
extension of the definition of </>q for -1 ~ q ~ O. Before describing </>q explicitly,
let us point out an important fact which can be obtained by using it:
1.2 Theorem Let E be the linear span of the cross-sections of the form:
r(q) = J(q)z;mz;, q E [-1, I], (1.4)
with J E C([-I, I]) and m,n E N (and Zq the generator of Tq). Then the
completion of E makes (Tq)-l<q<l a continuous field of C*-algebras (in the
sense of [5], Chapter 10).
The point in the last theorem is the continuity at -1. The lower-semicontinuity
will follow from the fact that the states </>q form a continuous field, and have
faithful GNS representations; after that, the upper-semicontinuity will come out
via a universality argument (see [12], or [10] where a general treatment of this
sort of situation is done).
In order to give the definition ofthe state </>q, let us first recall some basic facts
about the C*-algebra Tq, -1 ~ q ~ 1. It is immediate that sp{ z;m z; I m, n :::: O}
is a dense unital *-subalgebra of Tq. There is a natural continuous action by
automorphisms, a, of the circle Ton Tq, determined by a,\(zq) = ).Zq,). E T.
The fixed-point subalgebra for a is Vq = sp{z;mz;' 1m:::: OJ, and there is a
faithful conditional expectation Eq : Tq -+ Vq obtained by integrating a (Le.
Eq(x) = iT a'\(x)d>.). For q f: 0, Vq is the unital C*-algebra generated by the
selfadjoint element aq = 1- Z;Zq of Tq, because, as it is easy to check:
(1.5)
Hence, for q f- 0, we have Vq ~ C(a(aq)). Furthermore, a(aq) can be computed

precisely, and is found to be a "quantum interval" I q :
[-1,1], if q = -1
Iq = { {qnJn~O}U{O}, ~fqE(-I,I)\{O} (1.6)
[0, 1), If q = 1
(see for instance Proposition 1 of [6] for -1 < q < 1, and Remark 2.2 below
for q = -1; the case q = 1 is obvious). The "quantum interval" Iq has an
appropriate "Lebesgue measure", which integrates J E C(Iq) as follows:
1 Iq
~I21J(t)dt,
I~ J(t)dt,
ifq=-1
J(t)dqt = { (I-JqJ)Lk=oJqJkJ(qk), ifqE(-I,l)\{O}
if q = 1
(1.7)
(for q f- ±l, IIq J(t) dqt is an infinite Riemann sum, "approximating" the nor-
malized Riemann integral on [0,1)' for q > 0, and on [q,I], for q < 0 - compare to
[1), p.l0-11). This gives a faithful positive functional on C(Iq), thus providing
a faithful state 1/Jq on Vq (by 1/Jq(J(aq)) = II q
Jdq, J E C(Iq)). The canonical
state <pq on Tq we have in mind is simply:
(1.8)
for q > 0, this is exactly the restriction to Tq of the Haar state of S,;qU(2). The
last formula also holds for q = 0, when we take 1/Jo ; Vo = clos sp{ z~m Zo J m ~
O} -> C to be determined by 1/Jo( z~m zo) = 0 for every m ~ 1 (<Po is thus the
remarkable state of the Toeplitz algebra uniquely determined by the fact that
<Po(z(jzo) = °-see [4]).
The properties of <Pq, -1::; q::; 1, we need in Theorem 1.2 are:
1.3 Proposition ro The states <Pq, -1 ::; q ::; 1, form a continuous field, in
the sense that for every cross-section T considered in Theorem 1.2, the function
q -> <pq(r(q)) is continuous on [-1,1].
2° The GNS representation of <pq is faithful for every q in [-1,1] (and actually
<pq itself is faithful for q f- 0).
Moreover, it is interesting to note that:

1.4 Proposition We have a factorization of cf>-1 as cf>-1 o~, where ~ : 7-1 ->
M2 ® C(fi) is the embedding of Theorem 1.1, and

_ 1>.1 2
cf>-1 = Tr ® 7r vr=lIf4dA
1 -IA
(1.9)
(with Tr the normalized trace on 2 x 2-matrices, and dA = planar Lebesgue

measure on fi). In particular, cf>-1 is a trace on the non-commutative circle.
We conclude with a remark concerning the element (Zq + z;)/2. If A is a

unital C*-algebra and cf> is a state of A, and if a = a* E A, then the unique prob-
ability measure Jt on o-(a) such that I(a) = J l(t)dJt(t) for every I E C(o-(a))
is called the distribution of a. The element (zo + zo)/2 of To is called "semicir-
cular", because its distribution (with respect to cf>o) is ~v'f=t2dt on [-1,1] (see
[14]). The semicircular element plays an important role in the non-commutative
probability theory of Voiculescu (see for instance [15]); it is therefore interesting
to study (Zq + z;)/2 (in Tq with cf>q) for other values of q, too. (Such a study has
already been done in part II of [3], where Tq is considered with a different state,
"the vacuum expectation").
Now, for 0 < q ::; 1, (Zq + z;)/2 is still semicircular; this was proved in
Appendix Al of [17], by using the fact that the "representation theory of S"U(2)
is the same as the one of SU(2)". For -1 ::; q < 0, however, we encounter a
proper deformation of the semicircle law (see Remark 2.7 below); moreover, in
the case q = -1, the description of cf>-1 given in Proposition 1.4 allows us to
compute that:
1.5 Proposition The distribution of (Z-1 + z':..I)/2 (in 7_ b with respect to

cf>-.), is Itldt on [-1,1].
2 The proof of the results
2.1 Proof of Theorem 1.1 We denote by z( = Z_.) the generator of 7-1,

and by x, y, respectively, its real and imaginary parts (satisfying (1.2)). Note
that x 2 and y2 are in the center of 7-1 (for instance x 2 commutes with y because
it can be written as 1 _ y2).
Let us prove first that every irreducible *-representation IT : 7-1 -+ £(1l) of
7-1 is at most 2-dimensional. Indeed, since x2 is in the center, we must have
IT(x 2 ) = >.I for some 0 ~ A ~ 1. If, say, A = 1, then IT(y) = 0 and RanIT is the
unital C*-algebra generated by the selfadjoint unitary U = IT(x); this and the
irreducibility of IT imply dimIT = 1. The case A = 0 is similar. If 0 < A < 1,
then U = IT(x/v'X) and V = IT(y/v'f="'X) are selfadjoint unit aries and, clearly,
RanIT is the unital C*-algebra generated by the projections P = (U + 1)/2 and
Q = (V + 1)/2. It is well-known, however, that two projections can generate
a unital irreducible C*-algebra only on a space of dimension at most 2 (see for
instance [11], Section 3); hence dimIT ~ 2.
Now, let IT be a 2-dimensional irreducible *-representation. Then A =
IT(z) E £(C 2 ) satisfies:
A*A+AA* = 21. (2.1)
It is easy to check (by looking at the entries) that a 2 x 2-matrix with this
property is either of null trace or unitary. But A cannot be unitary (its spectral
projections would be invariant for IT); hence it is of null trace. Taking the
normalized trace Tr of (2.1), we get Tr(A*A) = 1.
Conjugating A E £(C 2 ) of the preceding paragraph by a unitary, such that
it becomes upper triangular, gives us something of the form B = (~ !A)
with f3 :f. 0 and 21AI2 + I/W = 2. We can actually assume f3 = J2(1-IAI2) (by
further conjugating B with an (
ifJ
eO ~ )
,8 E R, such that eifJ f3 is positive).
As a consequence, every 2-dimensional irreducible representation of 7-1 is uni-
tarily equivalent to one sending the generator z E 7-1 into a matrix as the one
appearing in (1.3).
There exists a unital *-homomorphism cfI : 7-1 -+ M2®C(D) such that
[cfI(Z)](A) is as in (1.3) for every A in D (because the matrix appearing in (1.3)
satisfies (2.1), and by universality). From the preceding paragraph it is clear
that composing cfI at the left with *-homomorphisms M2 ® C(D) -+ M2 which
evaluate at various A'S in the open disk D, we get representatives of all the classes
of equivalence of 2-dimensional irreducible representations of 7_ 1 • On the other
hand, it is also clear that composing cfI with evaluations at various A'8 in the
circle T, we obtain direct sums involving all the 1-dimensional representations of

7-1 (which are parametrized by T). Hence Ker\) is contained in the intersection
of the kernels of all the irreducible *"representations of 7-1 (which is {O}), and
\) is an embedding.
Finally, for every A E D, let 0(A) be the class of the 2-dimensional *-
representation of 7-1 sending z into (~ J2(1_-;.IAI2)) of (1.3); and for every
A E T, let 0(A) be the class of the character of 7-1 sending z into A. Then 0
is a continuous map from D onto the spectrum 7-1 of 7_1. and it obviously
factors to a homeomorphism D/P = -A IIAI < 1} ~ 7_ 1 . QED
2.2 Remark Let z( = z_t} be the generator of 7_1. and define a-1 =
1- z:1L1. We mentioned in the Introduction (relation (1.6» that the spec-
trum of a_1 is [-1,1]. This follows immediately from the embedding \) : 7-1 -+
M2 ® C(D) provided by Theorem 1.1. Indeed, we have
u(a_1) = u(\)(a_1» = U u([\)(a_1)](A»,

>'Efi
and the characteristic polynomial of [\)(a_1)](A) is (2 - (1 - IAI4), with roots

(1,2 covering [-1,1] when A runs in D.
A similar argument shows that the spectrum of (L1 + z:1)/2 is [-1,1], too.
2.3 Proof of Proposition 1.3 10 It clearly suffices to show the continuity

of [-1,1] :;l q -+ ¢q(z;mz~) for every m,n ~ O. This function is identically 0
when m # n (because Eq(z;mz~) =h A-m+nz;mz~dA = 0), and is identically
1 when m = n = 0; hence we may assume m = n ~ 1. In this case (by (1.5) and
the definition of ¢q):
¢ (z.mzm) _ { 0, if q = 0
q q q - IIq(l- t)(l- ~t) ... (1- q';_1 t)dqt, if q # 0 (2.3)
The continuity of (2.3) on [-1,1] \ {O} follows immediately from the fact that
q IIq f(t)dqt is continuous on [-1,1] \ {O} for every polynomial f (indeed, if
-+
f(t) = t k and q # ±1, then IIq f(t)dqt is (1 - Iql)/(1 - Iqlqk), continuous and
converging to t f~l tkdt for q -+ -1 and to f~ tkdt for q -+ 1). In order to prove
the continuity at 0, we can compute the integral appearing in (2.3) explicitly:
h Iq
I 1
(1 - t)(1 - -t) ... (1 - - t ) d t
q qm-I q
=
E JqJk(I - qk)(1 -
00
(1 -JqJ) qk-I) .. . (1 _ qk-m+I)

k=O
E JqJi+m(1- qi+m) ... (1 _ qi+I) =

00
(1 -Jql)
j=o
E E
00
(I-JqJ)JqJm (-I)IFI(JqJqlFl)jqllFlI,
j=O F~{1, ... ,m}
where IFJ is the cardinality of F ~ {I, ... ,m}, and JIFII is the sum of its
elements,
E (-I)lF1 qllFll 1
I-JqJqlFl'
F~{1, ... ,m}
which clearly conyerges to 0 for q -+ O.

2° For q f:. 0, <pq itself is faithful, since Eq and tPq are so. For q = 0, the
GNS representation of <Po is the representation of To as the C*-algebra of the
unilateral shift on [2(N), which is of course faithful ([4]). QED
2.4 Proof of Theorem 1.2 E is a unital *-algebra of cross-sections due to

(1.1). What needs to be proved is the continuity, for r E E, of the map:
[-1,1]3 q -+ JJr(q)JJ. (2.4)
Note that for every q E [-1,1] and every r E E, r(q) belongs to the dense *-
subalgebra ~ = sp{ z;m z; J m, n ~ O} of Tq, and conversely, for every q E [-1, 1]
and x E ~ there exists r E E such that r(q) = x.
It is important to mention at this point that, for every -1 :;; q :;; 1, the
elements (z;mz;)m,n~O spanning ~ are linearly independent. For q = 1 this
is obvious, and for -1 < q < 1 it follows from the well-known representation
of Tq sending Zq into a weighted shift (see for instance [16], Appendix A2, or
[6], Example 2). Let us also prove this assertion for q = -1. Remark first that
(z~T z~'t )m~O are linearly independent; this happens because z~T z~~.\ is a polyno-
mial of degree m in a_I (formula (1.5)), and because u(a-d = [-1,1] (Remark
2.2 above). Now, if E~,n=o Am,nz:Tz~1 = 0 for some scalars (Am,n)O~m,n~N'

then we get, for every 0::::; k ::::; N:
N
o= E_I(Z:;( L Am,nz:iz~d =
m,n=O
(with E-I : T-I - V-I the conditional expectation); hence Am,m+k = 0 for
every 0 ::::; k ::::; N, 0 ::::; m ::::; N - k. A similar argument shows that Am,n =0
when m > n, too.
Let us prove the lower semicontinuity of (2.4). For every q E [-1,1], the
faithfulness of the GNS representation of I/>q (Proposition 1.3) gives the norm
evaluation:
(see for instance [5], Section 2.4). Let us fix qo E [-1,1], TEE and £ > O. There
exists (by (2.5)) Yo E ~o such that
(2.6)
Take TI E E such that TI(qO) = Yo. The functions q - I/>q«TiT*TTI)(q)) and

q -+ I/>q«TiTd(q)) are continuous at qo (Proposition 1.3), hence
I/>q(TI(q)*TI(q)) < 1, I/>q(TI(q)*T(q)*T(q)TI(q)) > IIT(q)W - £
still holds (giving together with (2.5) that IIT(q)W > IIT(qo)W - i), for q in a
neighborhood of qo, Hence (2.4) is lower semicontinuous.
The upper semicontinuity of (2.4) comes out via a "universality" argument.
Let us fix qo, and define:
IIxll' = sup lim sup IIT(q)lI, x E ~o. (2.7)

TEE q-qo
T(qO) x =
We pause a moment to make sure that IIxll' defined in (2.7) is finite for every
x E ~o. So(qo being the one fixed above),fixalso x = E~,n=oAm,nZ;:'Z;o E ~o.
Let T be a cross-section of E which passes through x. We can write T(q) =
E~,n=ofm,n(q)z;mz;, with fm,n E G([-l,l]) for 0::::; m,n::::; M (by the defini-
tion of E, in equation (1.4)). Note that, because of the above mentioned linear
independence of (z;:,z;o)m,n~O' we must have:
f ( ) _ {Am,n, if m, n ::::; N (2.8)

Jm,n qo - 0, otherwise.
For every q E [-1, 1J we majorize:

M
IIr(q)11 = II L fm,n(q)z;mz;1I
m,n=O
M
L Ifm,n(q)llIzqllm+n ~
m,n=O m,n=O
(the bound IIzqll ~ J2 is immediate; actually, we have Ilzqll = 1 for 0 ~ q ~ 1,

and Ilzqll = vr=q for -1 ~ q ~ 0 - see Corollary 2 of [6]). As a consequence:
M
lim sup Ilr(q)11 ~ lim sup L Ifm,n(q)12(m+n)/2
q -+ qo q -+ qo m,n=O
N
L IA m,n(q)12(m+n)/2.
m,n=O m,n=O
So, taking also a supremum after r, we get

N
Ilxll' ~ L IA m ,n(q)12(m+n)/2 < 00.
m,n=O
Now, it is an immediate verification that II . II' defined in (2.7) is a C*-

seminorm on '!go, and since IIxll' ~ IIxil for every x E '!go (due to the lower
semicontinuity proved above), we see that it is actually a C*-norm. Denoting
the completion of ('!go, 11·11') by T,{o' the universality property of Tqo gives a
*- homomorphism f3 : Tqo -+ T,{o such that f3( x) = x for x E '!go. But then
IIxli' = 11f3(x)II ~ IIxll, x E '!go, and we can conclude that II '11' = 11·11. Hence
(2.7) implies that lim SUPq-+qollr(q)II ~ IIr(qo)II for every TEE, yielding the
upper-semicontinuity we need. QED
2.5 Proof of Proposition 1.4 (IAI2dA)/(1l"~) is a probability mea-

sure of full support on fi, which entails that <1>-1 of (1.9) is a faithful trace-state
of M 2 ®C(fi).
Now , by writing z*mzn
-1 -1 as z*(m-n)(z*n
1 -1 ) , when m > n , and as (z*mzm
-1 zn -1 -1 )zn-m
-1 ,
when m ~ n, and using (1.5), we immediately see that the state <1>-1 of 7-1 is
uniquely determined by:
1 if n = 0 and m is even
-1(a:'\z::1) = { Om,+l'
otherwise
(2.9)
with a_1 =I - z~1 L1. SO it will suffice to check that 4>-1 0 <l also satisfies (2.9).
This can be done by explicitly computing [<l(a~~.\z~1)](A) and then integrating
its trace after A E fi, which in turn can be done due to the relations:
The verification of (2.10) is immediate, and so is the fact that Tr[<l( a~:~.\ z~1)](A) ==
oon fi unless both m and n are even. In the latter case we have [<l(a~\z~1)](A) ==
An(1 - IAI4)m/2 I, hence:
"'_ (<l(a m zn ))
'f'1 -1 -1
= .!.7r if)
[ An(I_IAI4)m/2. vr=1Xf4
IAI2 dA =
= [1 2r ( [ (r(t(I-lr(14)m/2. Ir(1 2 d() dr

io iT Jl - Ir(14
(by passing to polar coordinates, with d( the Haar measure on T)
= 2 l rn+3 (1_ r4 )(m-1)/2dr . h CdC
which is indeed 1/(m + 1) if n = 0, and 0 otherwise. QED
2.6 Proof of Proposition 1.5 We shall use the description of 4>-1 given
by the previous proposition. We have (writing simply z instead of Lt):
[<l (Z ~ z*)] (A) = ()~~~ ~~~), A E fi,
and [<l(( Z+;{ )2)](A) == l+R;(>.2)I. It is clear that Tr[<l(( z.!{ )m)](A) == 0 (imply-
ing 4>-1(( z.!{ r) = 0 ) if m is odd. If m is even, m = 2k, we have:
4>-1 ((z+z*)m) = .!. [_ (I+Re(A 2))k. IAI2 dA =
2 7r in 2 vr=TXJ4
1 [ (I+Re()k 1
= ;;: if) 2 . 2F-l(j2
(where we performed the change of variable ( = A2, d( = 21AI 2dAj this is ob-
tained by summing up the changes of variables ( = A2 on the upper and lower
half-disks, with Jacobian 41A12)
= .!. [
7r if)
(1 + X)k . 2Jl - 1x
2 2 _ y2
dxdy
-.!...j1 (1 + X) k(j..Ji=X2 1 dy) dx

27T -1 2 -..Ji=X2 V(I - x 2 ) - y2
= -.!...j1 (I+X)k.7r dx = _1_.

27r - 1 2 k + 1
Hence the moment of order m of (z + z*)/2 is 0, if m is odd, and is I/(k + 1), if
m = 2k, so that it always equals the moment of order m of Itldt on [-1,1]. QED
2.7 Remark Let q be in (0,1), and consider the embedding of Tq into

S..;qU(2), as in the comment preceding Theorem 1.2. For every non-negative half
integer I E !N consider the "unitary representation of spin I" of S..;qU(2), which
is a unitary matrix (uLkj=-I.-1+1 ..... 1 E M21+1(S..;qU(2», and its "character"
XI = I:~=-I u!,; (see Sections 5 of [16], [17]). It is known (Appendix Al of [17])
that XI = Pz«Zq + z;)/2), I E !N, where:
1
Po(t) = 1, Pl(t) = 2t, ~+t{t) =
2
2tPz(t) - PI_1(t),
2
l2:: -2·
Also, it is not hard to see that for lEN: ul.i E Tq and Eq( u!,;) = 0, for i i- 0,
while:
ul
0.0
~(_I)kq(k+k2)/2q-k(l-k)[I]\I_a
= L.... k q
)(I-!a ) ... (I _ _
q q
I_ )ak
ql-k-1 a q q
k=O
belongs to Vq; [i) in the last relation is the q-binomial coefficient [klJ~·!klq"
(where, as usual, [i]q = l+q+ .. +qi-l, i 2:: 1, and [O]q! = 1, [j]q! = [I]q[2]q··· U]q
for j 2:: 1). Hence we have XI E Tq and Eq(XI ) = ub.o, lEN, which can be
rewritten as:
Eq ( PI ( - +-
Zq -
2
z;)) = (2.11)
I
[I] 2
"L....(-I) kq(l;+k 2 )/2 q- k(1 - k) k (1- aq )(l- 1 1
qa q ) •• ·(1- ql-k-1k
aq)aq , lEN.
°<
10=0
We have obtained (2.11) only for q < 1, but this equation is true for
-1 < q < 0, too. Indeed, it is immediately seen by direct computation that the
left side of (2.11) is of the form:
RO.I(q)I + R1.I(q)a q + ... + RI.I(q)a~, q E (-1,1) \ {O}, 12::0,
where Ro.I, . .. , RI.I are rational functions on the punctured plane. On the other
hand, it is of course sufficient to prove the equality of the two sides of (2.11) only
after we apply to them the well-known faithful representation of Tq sending Zq

into a weighted shift (see for instance [16], Appendix A2, or [6], Example 2). But
in thi, ,,",,,,,,,.tation n, b,rom" tb, di",.nal op,,,t.. ( ' q q' ... ), and
it is visible that both sides of (2.11) have analytic (operator-valued) extensions

to the punctured disk; so if they coincide on (0,1), they must also coincide on
(-1,0).
In conclusion, denoting the distribution of {Zq + z;)/2 (with respect to </>q)
by /-Lq, we obtain (by using (2.11), (1.8) and (1.7)) that for every 1 EN:
J 2:( _llq(k+k )/2 q-k(l-k) fir 1, (l_t)(I __qt ) .. . (1- ql-:-l )tk dqt.
I
PI d/-Lq = 2
k=O Iq
(2.12)
Of course, for q > 0, (2.12) is 0 for every 1 E N \ {O}, because of the "orthog-
onality relations for the representations of S,;qU(2)" (more precisely, because
ub,o and ug,o are orthogonal in the scalar product given by the Haar state of
S,;qU(2)); this explains the stability of the semicircle law in this case. For
q < 0, (2.12) is not zero, and allows a comparatively fast computation of the
even moments 0 f /-Lq: ft 2d/-Lq (t) = 4(1+q2)'
~ ft 4 d/-Lq ()
t = 8(1+q2)(1+q
(1-q)4(1+q! )' ••. The
odd moments of /-Lq are all 0, since obviously Eq ({(Zq + z;)/2)m) = 0 for m
odd.
Acknowledgement: We are grateful to Ed Effros for useful discussions

during the preparation of this paper.
References
[1] G.E. Andrews. q-series: their development and application in analysis, num-
ber theory, combinatorics, physics and computer algebra, Regional confer-
ence series in mathematics No.66, 1986 (published by the American Math.
Society).
[2] L.C. Biedenharn. The quantum group SUq(2) and a q-analogue of the boson
operators, Journ. of Phys. A, 22(1989), L873-878.
[3] M. Bozejko, R. Speicher. An example of a generalized Brownian motion,

part I - in Commun. Math. Phys. 137(1991),519-531, part II in Quantum
Probability and related topics VII, Proceedings New Delhi 1990,67-77.
[4] L. Coburn. The C*-algebra generated by an isometry, Bull. Amer. Math.

Soc. 73(1967),722-726.
[5] J. Dixmier. Les C*-algebres et leurs representations, Gauthier-Villars, 1969.
[6] P.E.T. Jorgensen, L.M. Schmitt, R.F. Werner. q-canonical commutation

relations and stability of the Cuntz algebra, to appear in the Pacific Journal
of Mathematics.
[7] S. Klimek and A. Lesniewski. Quantum Riemann surfaces I. The unit disk,
preprint 1991.
[8] A.J. Macfarlane. On q-analogues of the quantum harmonic oscillator and

the quantum group SU(2)q, Journ. of Phys. A, 22(1989),4581-4588.
[9] G. Nagy. On the Haar measure of the quantum SU(N) group, to appear in
Commun. Math. Phys.
[10] G. Nagy. Thesis, Berkeley 1992.
[11] G.K. Pedersen. Measure theory for C*-algebras II, Math. Scand. 22(1968),
63-74.
[12] M. Rieffel. Continuous fields of C*-algebras coming from group cocycles

and actions, Mathematische Annalen, 283(1989),631-643.
[13] A. L.-J. Sheu. Quantization of the Poisson SU(2) and its Poisson homoge-
nous space - the 2-sphere, Commun. Math. Phys. 135(1991),217-232.
[14] D. Voiculescu. Symmetries of some reduced free product C*-algebras, in

Operator algebras and their connection with topology and ergodic theory,
Busteni 1983 (Springer Lecture Notes on Mathematics No 1132).
290 G. Nagy andA. Nica
[15J D. Voiculescu. Free non-commutative random variables, random matrices

and the III factors of free groups, in Quantum probability and related
topics VI (L. Accardi, editor), 1991,473-487.
[16] S.L. Woronowicz. Twisted SU(2) group. An example of a non-commutative

differential calculus, Publ. RIMS 23(1987), 117-181.
[17] S.L Woronowicz. Compact matrix pseudogroups, Commun. Math. Phys.

111(1987), 613-665.
LANDSTAD DUALITY FOR COACTIONS ON C*-ALGEBRAS
JOHN C. QUIGG
ABSTRACT. Landstad's duality theorem for C·-dynamical systems characterizes

algebras arising as crossed products by actions of a given group. We state a dual
version, characterizing cocrossed products by coactions.
INTRODUCTION
Coactions of locally compact groups on C* -algebras were invented to give

nonabelian versions [IT], [Kat], [Val] of Takai's duality theorem [Tak] for crossed
products by abelian groups. When G is abelian, coactions of G correspond to
actions of the Pontryagin dual group G.
Let 6 be a coaction of G on a C* -algebra A, and let A ~ 6 G be the cocrossed
product (see the next section for the definitions). Roughly speaking, by a "du-
ality" for this situation we mean some method of "recovering" A and 6 from
A ~6 G. For example, Katayama's duality theorem [Kat, Theorem 8] shows how
to get A and 6 up to stable conjugacy by forming the reduced crossed product
of A ~6 G by a "dual" action of G. Similarly, Imai and Takai's duality theorem
[IT] (see also [Val]) does the corresponding job starting with an action rather
than a coaction.
A different type of duality was invented by Landstad [Lan2]. Starting with
an action O! of G on A, Landstad shows how to recover A and O! up to conjugacy
from the reduced crossed product A ~cx,r G. More than that, he characterizes
CO-algebras of the form A ~cx,r G in terms of a dual coaction and an equivariant
homomorphism from G to the multiplier algebra M(A ~cx,r G).
Our main result is a version of Landstad duality characterizing C* -algebras
of the form A ~ 6 G and showing how to recover A and 6 up to conjugacy from
A ~ 6 G given the dual aotion and an equivariant homomorphism from Co(G) to
M(A ~6 G). We mention that Landstad himself [LanI, Theorem 1) has proven
the special case where G is compact.
1. COACTIONS AND COCROSSED PRODUCTS
Let G be a locally compact group, with (full) group CO-algebra C*(G), and
let UG: G --+ M (C*( G)) denote the canonical homomorphism. As a bounded
strictly continuous function from G to M(C*(G)), UG may be regarded as a
unitary element of M(Co(G) ® C*(G)), and we write WG for UG when we have
this interpretation in mind. Although it does not matter in this case, we will
1991 Mathematics Subject Classification. Primary 46L55.

292 J. C. Quigg
give all C*-tensor products the minimal CO-tensor norm. Recall that if A and
Bare CO-algebras a homomorphism 7r: A --> M(B) is called nondegenerate if
7r(e;) --> 1 strictly for some (hence every) bounded approximate identity {e;}
for A. The homomorphism s ~ UG(s) I)$) UG(s) determines a nondegenerate
homomorphism 6G: C*(G) --> M(C*(G) I)$) C*(G» which is a comultiplication,
i.e., (6G I)$) t) 0 6G = (t I)$) 6G) o6G.
Definition 1.1. A coaction of G on A is an injective nondegenerate homomor-
phism 6: A --> M(A I)$) C*(G» such that
(i) 6(A)(C I)$) C*(G» c A I)$) C*(G)j
(ii) (6 I)$) t) 06= (t I)$) 6G ) 0 6.
We call the triple (A, G, 6) a cosystem. 6 gives rise to a Banach representation

of the Fourier-Stieltjes algebra B( G) = C*( G)* on A via 6f = (t I)$) f) 0 6 for
J E B( G). Most of the time we only need the restriction of this representation
to the Fourier algebra A( G). A cosystem (A, G, 6) is called nondegenerate if A is
nondegenerate as an A(G)-module. From the context it is always clear whether
nondegeneracy of 6 is meant in this latter sense or in the sense of homomorphisms
into multiplier algebras. Is every coaction automatically nondegenerate? This
is the automatic nondegeneracy problem, originally posed by Landstad [Lan2).
Automatic nondegeneracy has been proven when G is amenable [Lan2, Lemma
3.8), [Kat, Proposition 6) or discrete [BS).
Let B be a CO-algebra. A covariant representation of (A,G,6) in M(B)
is a pair (7r,I'), where 7r: A --> M(B) and 1': Co(G) --> M(B) are nonde-
generate homomorphisms satisfying the covariance identity (7r I)$) t) 0 6( a) =
Ad I' I)$) t( wG)( 7r( a) I)$) 1) for all a E A. In this case, C*( 7r, 1') = 7r( A )1'( Co( G»
(where juxtaposition of subspaces of an algebra denotes the linear span of the
set of products) is a C*-subalgebra of M(B), and 7r and I' are non degenerate
homomorphisms into M( C*( 7r, 1'». We call (C*( 7r, 1'), 7r, 1') a cocrossed prod-
uct of (A, G, 6) if every covariant representation (p, 1/) factors through (7r, 1'),
i.e., there is a homomorphism p X 1/ of C*( 7r, 1') such that (p X 1/) 0 7r = P and
(p X 1/) 0 I' = 1/. We denote a generic cocrossed product by (A ~o G,jA,jG)'
Two cosy stems (A, G, 6) and (B, G, () are called conjugate if there is an isomor-
phism 8: A --> B such that ( 0 8 = (8 I)$) t) 0 6. As expected, conjugate cosystems
have (essentially) the same covariant representations, hence the same cocrossed
product [QR, Proposition 1.3). The dual action h of G on A ~o G is determined
by
s E G,
where (s·f)(t) = J(ts). Katayama's duality theorem [Kat, Theorem 8) states
that if 6 is nondegenerate then
(A ~o G) ><I 6,r G = A I)$) K,
where K denotes the CO-algebra of compact operators on L2(G).

We are using the "full" coact ions of [Rae1) and [Rae2) rather than the "re-
duced" coactions of [Kat) and [LPRS). As shown in [Rae2) (see also [Quill), if
6 is a coaction of G on A in the sense of Definition 1.1, then (t I)$) >.) 0 6 (where
Landstad duality 293
A is the left regular representation of G) factors through a reduced coaction 6r

on AI ker( £ ® A) 0 6 having the same covariant representations as 6. Conversely,
in [Qui3] we show that if 0 is a nondegenerate reduced coaction of G on A, then
o = (£ ® A) 0 61 for a unique full coaction 0' (which is also nondegenerate),
again having the same covariant representations as o. Consequently, full and
reduced coactions are almost the same concept, with reduced coactions being
a little more general only if they can be nondegenerate. Even if nondegener-
ate coactions exist, every coaction has a canonically associated nondegenerate
coaction with the same covariant representations [Qui3]. The author believes
that the solution of the automatic nondegeneracy problem for coactions will be
in the affirmative, and in any event the coactions (either full or reduced) aris-
ing from all currently known constructions are guaranteed to be non degenerate.
Moreover, one of the most common uses of coactions, namely for crossed prod-
uct duality, requires nondegeneracy. Full coactions offer greater flexibility than
reduced ones; for example, a full coaction of G can be restricted to GIN for any
normal subgroup N, whereas a reduced coaction can only be restricted if N is
amenable. Besides, it seems more natural to use C*( G) than C;( G). Finally, the
techniques of proof are often essentially the same with full or reduced coactions.
The author therefore feels justified in preferring to use full coactions.
Definition 1.1 differs from [Rae1, Definition 1] in two aspects: requiring 0 to
be injective, and choosing the minimal tensor product A ®min C*( G) over the
maximal tensor product A ®max C*(G). As discussed in [Qui1] and [Rae1], if
o is allowed to be noninjective, then A is a semidirect sum of the ideal ker6
and the quotient 61 (A), and OIOl(A) is an injective full coaction with the same
covariant representations as O. Hence, no significant flexibility is lost by requiring
injectivity of 6. Regarding ®min versus ®max, as pointed out in [Qui1] and
[Rae2], if 0': A -+ M(A ®max C*(G» is a full coaction in Raeburn's sense, then
composing with the natural map from ®max to ®min produces a full coaction in
the sense of Definition 1.1 (except that it may be noninjective) which has the
same covariant representations as fl. However, we remarked in [Quil] that it was
unclear whether every full coaction using ®min factors through ®max. We have
recently been able to settle this question negatively in [Qui3]. Full coactions
using ®min have advantages over those using ®max; for example, in the present
theory, if (7r,JJ) is a covariant representation of (A,G, 0), then Ad JJ®£(wG)(·®l)
gives a full coaction on 7r(A) [Quil, Proposition 2.4], but we also show in [Qui3]
that this is not always the case using Raeburn's definition. Hence, it seems
preferable to use the more manageable ®min.
The Landstad duality for coactions is:
Theorem 1.2. [Qui2, Theorem 3.3] A C*-algebra B is of the form A ~6 G if
and only if there are an action a of G on B and a G-equivariant non degenerate
homomorphism JJ: Co(G) -+ M(B).
Moreover, in this case 0 may be chosen non degenerate, and then it is uniquely
determined up to conjugacy by 6 = a and iG = JJ.
2. IDEAS OF THE PROOF

First, by wishful thinking, JJ should be iG. Given a cosystem (A,G,o), the
294 J. C. Quigg
map Adja 0 t(wa)(· 01) defines a coaction on jA(A) which is conjugate to {j.
So, our strategy should be to try to get A in the form jA(A), and then (jA,ja)
will be (t, p,). A handle on the problem is afforded by the observation that the
dual action 6 is trivial on jA(A). This leads us to the strategy of getting A by
averaging over G-orbits. Naively, we would like an "averaging map" E: B -> A
given by E(b) = Jaas(b)ds. A substantial amount of analysis is required by
the fact that E is actually unbounded and densely defined. It turns out that
this all works, however, and we let A be the C*-algebra generated by the range
of E. Then we finish by verifying the following facts:
(1) AdJ-L 0 t(wa)(· 01) gives a coaction {j of G on A;
(2) (t,J-L) is a covariant representation of (A,G,{j) in M(B);
(3) B = C*(AJ-L(Co(G)));
(4) every covariant representation of (A,G,{j) factors through (t,J-L).
Of course, the heart of the matter is the structure of the averaging map E.
We dualize the method of Landstad [Lan2] and follow a construction of Olesen
and Pedersen [OP1, Section 2) and [OP2, note added in proo~: the set
is a dense G-invariant *-subalgebra of B, and we have:

Lemma 2.1. For all bE Bo there exists E(b) E M(B) such that for all ¢> E B*,
(E(b),¢» = fa (as(b),¢» ds.
The connection between the coaction {j and the homomorphism J-L is:
Lemma 2.2. Define {j: B -> M(B 0 C*(G)) by {j(b) = AdJ-L 0 t(wa)(b 01).
Then {j is a coaction ofG on B, and for all a E A and /,g E Cc(G) we have
Dualizing Green's theory of twisted actions, Phillips and Raeburn [PR) con-
structed a theory of twisted coactions and twisted co crossed products. Using
[Qui2) and Echterhoff's characterization [Ech, Theorem) of induced C*-algebras,
Raeburn and the author [QR, Theorem 5.2) were able to give a twisted version
of Landstad duality for coactions: twisted cocrossed products are characterized
by the presence of an action of a normal subgroup N of G and an N -equivariant
homomorphism of Co(G).
It would be interesting to dualize this to give a characterization of Green's
twisted crossed products. Also, it is tempting to imagine that some sort of
recognizable structure must be present if one has an action of G and a G-
equivariant homomorphism of Co(G / N).
Landstad duality 295
REFERENCES
[BS] S. Baaj and G. Skanda.lis, C·-algebres de Hopf et theorie de Kasparov equivariante,

K -theory 2 (1989), 683-721.
[IT] S. Imai and H.Takai, On a duality for C·-crossed products by a locally compact group,
J. Math. Soc. Japan 30 (1978), 495-504.
[Kat] Y. Katayama, Takesaki's duality for a non-degenerate co-action, Math. Scand. 66
(1985), 141-151.
[Lanl] M. B. Landstad, Duality for dual C·-covariance algebras over compact groups (un-
published manuscript, 1978).
[Lan2] ___ , Duality theory for covariant systems, Trans. Amer. Math. Soc. 248 (1979),
223-267.
[LPRS] M. B. Landstad, J. Phillips, I. Raeburn, and C. E. Sutherland, Representations of
croued products by coactions and principal bundles, Trans. Amer. Math. Soc. 299
(1987), 747-784.
[OPl] D. Olesen and G. K. Pedersen, Applications of the Connes spectrum to C·-dynamical
,ystems, J. Funct. Anal. 30 (1978), 179-197.
[OP2] ___ , Applications of the Connes spectrum to C·-dynamical systems. II, J. Funct.
Anal. 36 (1980), 18-32.
[PRJ J. Phillips and I. Raeburn, Twisted crossed products by coactions (preprint).
[Quil] J. C. Quigg, Full C·-crossed product duality, J. Austral. Math. Soc. Ser. A 60 (1991),
34-52.
[Qui2] ___ , Landstad duality for C·-coactions, Math. Scand. (to appear).
[Qui3] ___ , Full and reduced C·-coactions (in progress).
[QR] J. C. Quigg and I. Raeburn, Induced C·-algebras and Landstad duality for twisted
CO-coactions (preprint).
[Rael] I. Raeburn, A duality theorem for crossed products by nonabelian groups, Proc. Centre
Math. Anal. Austral. Nat. Univ. 16 (1987), 214-227.
[Ra.e2] ___ , On crossed products by coactions and their representation theory, Proc. Lon-
don Math. Soc. (3) 64 (1992), 625-652.
[Talc] H. Takai, On a duality for crossed products of CO-algebras, J. Funct. Anal. 19 (1975),
25-39.
[Val] J.-M. Vallin, CO-algebres de Hopf et CO-algebres de Kac, Proc. London Math. Soc. (3)
60 (1985), 131-174.
DEPARTMENT OF MATHEMATICS, ARIZONA STATE UNIVERSITY, TEMPE, ARIZONA 85287

E-mail address:quigg@math.la.asu.edu
QUANTIZATION OF POISSON SU(2)
Albert Jeu-Liang Sheu 1

Department of Mathematics, University of Kansas
Lawrence, Kansas 66045-2142, U. S. A.
O. Introduction
In this paper, we describe and compare two different C*-algebraic de-
formation quantizations (in RiefFel's sense [Rl, R2, R3]) ofthe 'multiplica-
tive' Poisson structure on the Lie group SU(2), which are 'compatible' with
Woronowicz's C*-algebraic quantization [Wol, W02] ofthegroup structure
of SU(2) by pseudogroups (or quantum groups). D.ue to the space limita-
tion, in this paper we only describe the results and leave their proof to a
separate paper [Sh3].
In [Shl], it is shown that there exists a deformation quantization of
the multiplicative Poisson structure on SU(2) which is compatible with
Woronowicz's deformation quantization of SU(2) by the pseudogroups
C(S"U(2)), in the sense that the C*-algebras obtained in these two pro-
cesses are isomorphic.
However, it is highly desirable to have a C*-algebraic deformation of
the Poisson SU(2) which, when restricted to the two prominent generators
Q, 'Y of C(SU(2)), satisfies Woronowicz's commutation relations on the
two corresponding generators of C(S"U(2)) . But it is not clear (and the

answer is very likely to be negative) whether the deformation found in [Shl]
has this property.
On the other hand, it can be proved [Sh3] that the Weyl transforma-
tion studied in [D) also gives a C*-algebraic deformation quantization of
the Poisson SU(2) and its restriction to the two prominent generators Q, 'Y
of C(SU(2)) gives exactly the two corresponding generators of C(S"U(2)).
So a more direct link between RiefFel's and Woronowicz's deformation the-
ories on the Poisson Lie group SU(2) is established. In fact, the Weyl
transformation is shown in [D) to be a coalgebra isomorphism from regular
functions on SU(2) to elements of the algebra C(S"U(2» which 'preserves'
the two generators.
A more detailed comparison of these two different quantizations of the
Poisson SU(2) is given later in this paper.
The author would like to dedicate this paper (and the related work
[Sh3]) to the memory of Professor John W. Bunce, an honorable colleague
and mathematician with important contribution in the theory of operator
algebras.
Ipartially supported by NSF-Grant DMS-9002895
Quantization of Poisson SU(2) 297
1. Poisson SU(2)
A Poisson manifold [Li, We] is a smooth manifold M endowed with a
Lie bracket structure
such that, for all /,g, h E COO(M),
{f,gh} = {f,g}h + g{f, h}.

Each / E COO(M) determines a unique Hamiltonian vector field HI on
M, such that
{f,g} = Hr 9
for all 9 E COO(M). A Poisson 2-tensor 7r on M, i.e. a contravariant
alternating 2-tensor field 7r with the Schouten bracket [7r,7r] vanishing [Li],
defines a Poisson bracket by
{f,g} = < d/ Adg,7r >
and hence HI = < d/,7r >. Integrating "all" Hamiltonian vector fields
on a Poisson manifold, we get a singular foliation, called the symplectic
foliation, such that 7r restricted to each leaf is non-singular everywhere and
hence defines a symplectic form on that leaf [We]. A Poisson structure on
a Lie group G is multiplicative [LuWe], if the multiplication operation
m:GxG-+G
is a Poisson map (i.e. preserving the Poisson 2-tensor), where G x Gis

endowed with the product Poisson structure. A Lie group endowed with a
multiplicative Poisson structure is called a Poisson Lie group.
There is a unique multiplicative Poisson structure on SU(2) up to inner
automorphisms and scalar multiplications [VSo], e.g.
[Sh2] which vanishes on
where T is the unit circle in the complex plane C. The singular folia-
tion of Poisson SU(2) by symplectic leaves consists of a circle family of
the canonical symplectic C and a circle U(l) of O-dimensional symplectic
298 A. l-L. Sheu
leaves, to which the complex planes C are glued together along their circle
boundaries at infinity. More precisely, there is a smooth diffeomorphism
4> : T x C -+ SU(2)\U(I)
such that 4> restricted to each {z} x C is a symplectomorphism onto the

symplectic leaf
of SU(2) and
lim 4>Ara) = [ao
r-+oo ~]
a E U(l).
2. Quantum SU(2)
In [Wol], Woronowicz introduced a family of Hopf C*-algebras
C(SI'U(2» with -1 < I' ~ 1 generated by a, 'Yas a deformation of the
classical Hopf algebra of continuous functions on the Lie group SU(2), with
C(SlU(2» ~ C(SU(2».
Let v be the Gaussian measure on C, i.e.
dv{z) = exp( -lzI2)dxdy,

H2(C) (called the Segal-Bargmann space) be the Hilbert space of holo-
morphic v-L2-functions on C, and P be the orthogonal projection from
L2(C, v) onto H2(C). Then T", = PM",IH2(c) is called the the Toeplitz
operator with symbol tP E LOO(C, v). Consider the symbol space C(C)
consisting of continuous functions tP on C which converge radially and uni-
formly to continuous functions tPoo on T. We define the Toeplitz algebra
T(C) = to be the C*-algebra generated by T", with tP E C(C). By results
of Berger and Coburn [BeC], we have a short exact sequence
0-+ x:; -+ T(C) ~ C(T) -+ 0,
where 0' is the symbol map sending T", to tPoo and x:; is the algebra of all
compact operators on H2(C).
From the following structure theorem of C(SI'U(2», we can view
C(SI'U(2» as a C*-algebra of the above symplectic foliation on SU(2).
Theorem I ([VSo, ShI]) . For 11'1 < 1, C(SI'U(2» is isomorphic to
{f E C(T, T(C» : O'(f(z» E C(T) is independent of z E T}.

3. Deformation quantization of Poisson structures

Let us recall the C*-algebraic deformation quantization in Rieffel's sense
[RI, R2, R3].
Definition 1 . Given a Poisson manifold (M, {., .}) and a smooth Pois-
son algebra A (of functions on M), i.e. a (sup-norm) dense Poisson *-
subalgebra of Cr(M) c Cb(M), a family of linear surjections Ph : A --+
Ah with 0 S h < Ii for some Ii > 0 is called a C*-algebraic deformation
quantization of A if
(1) each Ah is a *-subalgebra of a C*-algebra Ah with Ao = A and Po
the identity map,
(2) Ilph(f)1I is continuous in h E [0, Ii) for any fixed f E A,
(3) limh--+o IIPh(f*) - Ph(f)*11 = 0 for any f E A,
(4) limh--+o Ilph(f)Ph(g) - Ph(fg) I! = 0 for any f,g E A,
(5) limh--+o I!(ih)-l[ph(f), Ph(g)]- Ph( {f,g})11 = 0 for any f,g E A.
A deformation Ph is called *-preserving if Ph(f*) = Ph(f)* for any f E A.
One idea of quantizing the Poisson SU(2) is to quantize the non degen-
erate leaves separately and consistently as done in [Shl]. First, we quantize
the symplectic C by Weyl operators.
Consider the symbol space
SO = {a E C OO (R2) with an asymptotic expansion
a(~,x) rv ~i=o a_j(~,x)},
where ak are homogeneous functions in (~, x) of degree k. Note that SO is

closed under the bracket { , }. Given a E So, the Weyl operator Wh( a) on
L2(R) with symbol a, for h > 0, is defined by
wh(a)u(x) = (27rh)-1 f[a(e, (x + t)/2) exp(ie(x - t)/h)u(t)]dedt
for u E C~(R). Then (SO, Wh) is a *-preserving C*-algebraic deformation

quantization of SO on the symplectic manifold R2 (endowed with the sym-
plectic form dx 1\ d~. Furthermore, the C*-algebra generated by the Weyl
operators Wheal with a E SO is isomorphic to the Toeplitz algebra y(e).
Note that for any a E COO(SU(2)), we have (a 0 4>(z, .))= = aIU(l) , i.e.
lim ao4>(z,ra) = a(a)
r--+oo
for all a E T ~ U(l), where 4> is the diffeomorphism defined in section 1
and· = x + i~ E C is identified with (~,x) E R2.
Theorem 2 ([Shl]) . The maps Wheal sending z E T to wh(ao4>(z, .)) E
y(e) form a *-preserving C*-algebraic deformation quantization of
C=(SU(2)) and the completion ofCOO(SU(2))h is isomorphic to C(S/LU(2))
for 1 ;::: h ;::: 0 and {L = 1 - h.
300 A. J.-L. Sheu
Although the above deformation quantization produces C*-algebras iso-

morphic to the pseudogroups C(S,.U(2» as stated in the theorem, it is not
clear (and the answer is very likely to be negative) whether the deformed
generators Wh(a) and Wh(-r) satisfy Woronowicz's commutation relations
on the generators of C(S,.U(2». In fact, it is a challenge to investigate this
problem, since it involves technically very difficult exact (not just modulo
compact operators) calculation of the composition of pseudo differential op-
erators.
4. Continuous field of quantum SU(2)

Let A be the universal C*-algebra generated by a, ",(, and v, subject
to the relations
a*a + ""rr* = 1, +
aa* v 2 "'("'(*= 1,
a"'( = v"'(a, a"'(* =
v"'(* a,
"'("'(* = "'(* "'( , va =av, =
v"'( "'(v ,
v* = v, vv* ~ 1.
Clearly C*(v) ~ C([-I, 1]) is a central sub algebra of A.

Let I,. = {f E e([-I,I]) I f(p.) = O} for p. E [-1,1]. Then the
surjective homomorphism 1r,. : A --+ C(S,.U(2» sending a, "'(, and v
to a, "/, and p., respectively, factors through A/(I,.A) and induces an
isomorphism
1r~ : A/(I,.A) --+ C(S,.U(2».
Using ideas and results of [R4], we get
Theorem 3 ([Sh3D . The universal C*-algebra A is identified with a
maximal algebra of cross sections of {C(S,.U(2»}"E[-l,l] by the map
A ~ {1r,.(A)},.E[-l,l].
The family {C(S,.U(2»},.E[_l,l] is an upper semi-continuous field of C*-

algebras over [-1,1] with
p. ~ 111r,.(A)1I
continuous in p. E (-1,1] for any A E A.
Since our goal is to find a deformation quantization of SU(2), we are
only interested in the behavior of C(S,.U(2» for p. close to 1, and so we
shall fix an f > 0 and add v ~ f to the commutation relations. The theorem
holds for this with [-1,1] replaced by [f,I]. Note that the generator v is
now invertible in A.
5. Weyl transformation
We denote by A ®z A the tensor algebra A ® A modulo the ideal

generated by II ® 1 -1 ® II. Then by the universality of A we can define the
comultiplication <b : A ~ A ®z A on the algebra A as the C*-algebra
homomorphism determined by
<b(Q) = ex. ® ex. - 11"'(* ® "'(
and
<b("'() = "'( ® ex. + Q* ® "'(.
Then for n ~ 0 and 0 ~ k ~ n, <b( ex.n-k",(*k) is of the form
L ex.n-i"'(*i ® wik
OSiSn
where wik is a finite Z-linear sum of monomials in the elements II, 11- 1, Q,
Q*, ",(, and "'(* in A.
The set {1l",,(wik) I n ~ 0, 0 ~ i, k ~ n} consisting of the matrix el-
ements of all the canonical irreducible quantum group representations of
C(S"U(2)) is used by Dubois-Violette [D] to define the Weyl transforma-
tion between the smooth algebras of regular functions
W h : C(SU(2))OO ~ C(S"U(2»OO
with p. = e- h / 2 as follows. For any regular function Ion SU(2),
Wh(f) = L iii.1l",.(wik)'
n,i,k
where
hi. = 111l"1 (Wik) 11;2 isU(2) 11l"1(wik)
and 1= En,i,k .hk1l"1 (Wik) is the Fourier expansion of I.
Theorem 4 rD, WoIJ. The Weyl transformation
is a coalgebra isomorphism lor each h = - 2In(p.). In particular, W 1 is the

identity map on C(SU(2))oo.
Remark. Dubois-Violette uses the term, Weyl transformation, in a more

restrictive way, namely, when wik are replaced by the matrix elements uik of
irreducible unitary representations. Note also that W h sends the generators
a, 'Y of C(SU(2»OO to the generators a, 'Y of C(S"U(2».
6. Weyl quantization of Poisson 8U(2)

Using the explicit Poisson 2-tensor on SU(2) and the commutation
relations, we can prove
302 A. J.-L. Sheu
Theorem 5 [Sh3]. With J.t = e- h/2 , the Weyl transformations
W h : C(SU(2))OO __ C(SI'U(2))OO
form a C*-algebraic deformation quantization of the Poisson algebra

C(SU(2))OO of regular functions on the Poisson Lie group SU(2) and the
completion ofWh(C(SU(2))OO) is isomorphic to C(SI'U(2)).
Remark. (1) The author learned recently that A. Bauval had also
obtained similar results [Ba) , using a different method. (2) The above
theorem still holds when we replace wile's by u?le's in defining the Weyl
transformations.
This result strengthens the claim that the group structure and the
Poisson structure on SU(2) are not only compatible on the classical level,
but also compatible on the quantum level, since now they are not only
compatible globally (i.e. as algebras) but also locally (Le. for individual
functions on SU(2)).
We now compare the above two deformation quantizations of SU(2).
The deformation W h is obtained by a 'leaf-wise' quantization through
pseudo-differential operators (with respect to the foliation of SU(2) by
symplectic leaves) and hence is 'leaf-preserving', in the sense that if f = g
on a symplectic leaf L of SU(2) then 'll"L(Wh(J)) = 'll"L(Wh(g)) where 'll"L is
the irreducible representation of the C*-algebra C(SI'U(2)) naturally as-
sociated with the leaf L [VSo). This quantization gives some kind of a
singular foliation C*-algebra and fits well with the more general theory of
external Rd-actions developed in [R5). But it does not seem to work well
with the internal SU(2)-action and so its relation with the comultiplica-
tion is not clear. On the other hand, the deformation Wh (using either
wile or uile) is compatible with Woronowicz's quantization of group struc-
ture, and with the internal SU(2)-action and hence the comultiplication
on C(SI'U(2)). But it applies to a smaller algebra (consisting of regular
functions) than the algebra of smooth functions quantized by Wh and it
is not 'leaf-preserving'. In fact, we can actually find explicitly a regular
function f on SU(2) vanishing on a specific symplectic leaf L such that
'll"L(Wh(J)) =f 0 for all h =f O.
7. Universal Poisson algebras

In proving the above result, the concept of a universal Poisson algebra
[Sh3) turned out to be handy. Usually, a Poisson algebra refers to a com-
mutative algebra P with a Lie bracket {.,.} : P x P - - P such that {a,·}
and {.,a} are derivations of P, Le.
{a,bc} = {a,b}c+ b{a,e}, and {ab,e} = {a,e}b+ a{b,e},

for all a, b, c E P. But this definition does not need the commuta~ivity of
P in order to make sense. In fact, a (noncommutative) algebra A endowed
with the commutator bracket [a, b] = ab - ba for a, b E A is an example.
Definition 2 . A universal Poisson algebra peS) generated by elements

in a set S (with no additional relations imposed) is a Poisson algebra en-
dowed with an embedding (i.e. an injection) t : S --+ peS) such that
any function from S to a Poisson algebra A extends uniquely to a Poisson
algebra homomorphism from P( S) to A.
Such a universal Possion algebra is clearly unique up to isomorphism.

Actually it exists and can be constructed in the following way. We take
one representative from each congruence class of morphisms f : S --+ Aj,
where f is a function, A J is a Poisson algebra which has no proper Poisson
sub algebra containing f(S), and two such morphisms f, g are congruent
if there is a Poisson algebra isomorphism h : A J --+ Ag such that g =
h 0 f. Let P be the Poisson algebra direct product I1[Jl A J with exactly
one AI from each congruence class [fl. Then we define peS) to be the
smallest Poisson subalgebra of P containing all I1[Jl f(s), s E S. Because
the free (noncommutative) algebra generated by elements of S is such an
AI containing a faithful copy of S, the Poisson algebra peS) is well-defined
and clearly satisfies the universality condition.
It is an interesting unanswered question how such a universal Poisson
algebra can be constructed from free Lie algebras and the enveloping alge-
bras of Lie algebras.
References
[Ba] A. Bauval, Deformation du groupe de Lie-Poisson SU(2),
preprint.
[Bee] C. A. Berger and L. A. Coburn, Toeplitz operaotrs and quantum

mechanics, J. Func. Anal. 68 (1986), 273-299.
[D] M. Dubois-Violette, On the theory of quantum groups, Lett.

Math. Phys. 19 (1990), 121-126.
[Li] A. Lichnerowicz, Les varieties de Poisson et leurs algebres de

Lie associees, J. Difr. Geom. 12 (1977),253-300.
[LuWe] J. H. Lu and A. Weinstein, Poisson Lie groups, dressing

transformations and Bruhat decompositions, J. Difr. Geom. 31
(1990), 501-526.
304 A. l-L. Sheu
[R1] M. A. Rieffel, Deformation quantization and operator algebras,

in "Proc. Symp.Pure Math., Vol. 51", AMS, Providence, 1990,
pp. 411-423.
[R2] _ _ , Deformation quantization of Heisenberg manifolds,

Comm. Math. Phys. 122 (1989), 531-562.
[R3] _ _, Lie group convolution algebras as deformation quantiza-

tion of linear Poisson structures, Amer. J. Math. 112 (1990),
657-686.
[R4] _ _, Continuous fields of C*-algebras coming from group co-

cycles and actions, Math. Ann. 283 (1989),631-643.
[R5] _ _ , Deformation quantization for actions of R d , preprint.
[Sh1] A. J. L. Sheu, Quantization of the Poisson SU(2) and its Pois-

son homogeneous space - the 2-sphere, Comm. Math. Phys. 135
(1991),217-232.
[Sh2] _ _, Quantum Poisson SU(2) and quantum Poisson spheres,

in "Contemporary Mathematics, Vol. 134", AMS, Providence,
1992, pp. 247-258.
[Sh3] _ _, The Weyl quantization of Poisson SU(2), preprint.
[VSo] L. L. Vaksman and Ya. S. Soibelman, Algebra of functions on

the quantum group SU(2), Func. Anal. Appl. 22 (1988), 170-
181.
[We] A. Weinstein, The local structure of Poisson manifolds, J. Diff.

Geom. 18 (1983),523-557.
[Wo1] S. L. Woronowicz, Twisted SU(2) group: an example of a non-

commutative differential calculus, Publ. RIMS. 23 (1987), 117-
181.
[Wo2] _ _ , Compact matrix pseudogroups, Comm. Math. Phys. 111

(1987), (j13-665.
Section VI
GEOMETRY AND TOPOLOGY
"Vector bundles" over quantum Heisenberg manifolds
Beatriz Abadie!
Abstract: We construct, out of Rieffel projections, projections in cer-

tain algebras which are strong-Morita equivalent to the quantum Heisen-
berg manifolds D~v. Then, by means of techniques from the Morita equiv-
alence theory, we get finitely generated and projective modules over the
algebras D~v. This enables us to show that the group Z + 2fLZ + 2vZ is
contained in the range of the trace on Ko(D~v).
§0 Preliminaries.
Let G be the Heisenberg group,
and, for a positive integer c, let Hebe the subgroup of G obtained

when x, y, and cz are integers. The Heisenberg manifold Me is the
quotient G / He·
N on-zero Poisson brackets on Me that are invariant under the
action G on Me by left translation can be parametrized by two real
numbers fL and v, with fL2 + v 2 =f:. 0 ([Rf3]).
For each positive integer c and real numbers fL and v as above,
Rieffel constructed ([Rf3]) a deformation quantization {D~:hER of
Me in the direction of the Poisson bracket A!-,,,.
Since D~': is isomorphic to Di~.n", and we will not need to keep
track of the Planck constant h, we absorb it from now on into the
parameters fL and v. Thus we will use D~" to denote D~·~.
As shown in [Rf3], the algebra D~" can be described as the gen-
eralized fixed-point algebra of the crossed-product Co(R x T) x>. Z,
where '\p(x,y) = (x + 2PfL,y + 2pv), for all P E Z, under the action
p of Z defined by
(Pk<P)(X, y,p) = e(ckp(y - pv))<p(x + k, y,p),

IThe material of this work is part of the author's Ph.D dissertation, submitted
to the University of California at Berkeley ill May 1992.
308 B. Abadie
where k, p E Z, C) E Cc(R x T x Z), and, for any real number
x, e(x) = exp(271"ix). (The action p defined above corresponds to
the action p defined in [Rf3, p.539], after taking Fourier transform
in the third variable to get the algebra denoted in that paper by
An, and viewing An as a dense *-subalgebra of Co(R x T) x>. Z via
the embedding J defined in [Rf3, p.547].) Equivalen~ly, D~v is the
closure in the multiplier algebra of Co( R x T) x,\Z of the *-subalgebra
Do consisting of functions C) E C(R x T x Z) which have compact
support on Z and satisfy
C)(x + k, y,p) = e( -ckp(y - pv»C)(x, y,p),
for all k,p E Z, and (x,y) E R x T (Do is the image under the
embedding J mentioned above of the sub algebra denoted by CP in
the proof of [Rf3, Thm.5.4]).
There is a faithful trace ([Rf3]) TD on D~v defined for C) E Do,
by
TD(C) = [ C)(x, y, O)dxdy.
JT 2
Now ([Ab2]), the algebra D~v is strong-Morita equivalent to the gen-
eralized fixed-point algebra E~v of the crossed product Co(Rx T) x uZ
under the action 'Y of Z, where O"k(X,y) = (x - k,y) and
(-rpc)(x, y, k) = e(-ckp(y - pv»c)(x - 2pl', Y - 2pv, k),
for k, p E Z and c) E Cc(R x T x Z).

The algebra EZ v can also be described (see [Ab2]) as the closure
in the multiplier algebra of Co(R x T) XU Z of the *-algebra Eo
consisting of functions c) E C(R x T x Z), with compact support on
Z and such that
C)(x - 2pl',y - 2pv,k) = e(ckp(y - pv»C)(x,y,k),
for all k, p E Z, (x, y) E R x T. The equivalence D~v-EZv bimodule

X constructed in [Ab2] is the completion of Cc(R x T) with respect
to either one of the norms induced by the D~v and E~v-valued inner
products given by
< /,g > D (x, y,p) = Le(ckp(y-pv»/(x+k, y)g(x-2pl'+k, y-2pv)

keZ
and </,g>E(x,y,k)=
= L e(-ckp(y - pv»/(x - 2pp.,y - 2pv)g(x - 2pl' + k,y:- 2pv),
peZ
"Vector bundles" over quantum Heisenberg manifolds 309
respectively, where /,g E Co(RxT), ~ E Do, \) E Eo, (z,y) E RxT,

and k,p E Z.
§1 Projective modules.
In what follows we produce finitely generated and projective mod-

ules over the algebras D~/J' To do this we apply to the Morita equiv-
alence structure described above the methods for constructing pro-
jections provided by the Morita equivalence theory. Finally, by com-
puting the trace of those projections, we show that the range of the
trace at the level of Ko(D~/J) contains the subgroup Z + 2J.lZ + 211Z.
First notice that both Do and Eo have identity elements ID and
IE, respectively, defined by
ID(Z,y,P) = oo(p) and IE(Z, Y, k) = oo(k),

for (z, y) E R X T and k, p E Z.
Therefore, by [Rf2, Prop. 1.2], if P is a projection in Eo, then
X P is a projective finitely generated left module over D~/J' and the
corresponding projection in Mm(D~/J) is given by
Q =( < Y1 '.~1 > D < Ym,Zl >D )
< Yt,Zm >D < Ym,Zm >D
where, for i = 1, ... , m , Zi, Yi E X P are such that
i=m
P = L < Zi, Yi > E .
i=l
On the other hand ([Rfl, Prop. 2.2]), the trace TD on D~/J induces
a trace TE on E~/J via
TE« /,g >E) = TD« g,/ >D)·

A straightforward computation shows that for \) E Eo we have
TE(\)) = 10r 10r

lL
\)(Z, Y, O)dzdy.
Then, in the notation above, we get

i=m i=m
TD(Q) = LTD« Yi,Zi >D) = LTE« Zi,Yi >E) = TE(P).
i=l i=l
310 B. Abadie
Theorem 1 The bimodule X is a finitely generated and projective

D~v-module of trace 2p.. If II E [0,1/2], and p. > 1, then there is a
finitely generated projective D~v-submodule of X with trace 211.
Proof: Let us take P = IE, in the notation above. Then X =

X P is finitely generated and projective and its trace is TE(IE) = 2p..
We now find a projection P in Eo with TE(P) = 211, when
II E [0,1/2] and p. > 1, which ends the proof, in view of the re-
marks above.
Following the techniques in [Rfl], we will consider self-adjoint ele-
ments P of the form:
P(x,y,p) = f(x,y)ol(p) + h(x,y)oo(p) + I(x - 1,y)Ll(P),
where hand f are bounded functions on R x T and h is real-valued.

Our next step is to get functions f and h such that P is a projection
. E !-'v'
In
C
Now,
(P * P)(x, y,p) = L: P(x, y, q)P(x + q, y,P - q),

qEZ
and it follows that P *P = P if and only if, for all (x, y) E R x T:
1) I(x, y)/(x + 1, y) = 0
2) I(x, y)[h(x + 1, y) + h(x, y) - 1] = 0
3) I/(x, y)12 + I/(x - 1, y)12 = h(x, y)(1 - h(x, y)).
We also want P to be in Eo, so we require
P(x,y,p) = e(cp(y + II))P(-X + 2p., y + 211,p),

that is
4) I(x, y) = e(c(y + 1I))/(x + 2p., y + 211)
and
5) h(x, y) = h(x + 2p., y + 211).
Now, it was shown in [Rfl, 1.1] that for any ( E [0,1/2] there are
maps F, H E C(T) such that:
1)' F(t)F(t - () = 0
2)' F(t)[1 - H(t) - H(t - ()] = 0
3)' H(t)[l- H(t)] = IF(t)12 + IF(t + ()1 2
4)'
5)' °IT: :;
H = (
H(t) :::; 1 for any t E T and F vanishes on [1/2,1].
Let us assume that v E [0,1/2]' I' > 1 and let F and H be functions
satisfying 1)'-5)' for ( = v/I'.
Translation of t by ( in equations 1)'-5)' plays the same role as
translation of z by 1 in equations 1)-5), which suggests taking (z as
the variable t. However, the variable y will play an important role
in getting I and h to satisfy 4) and 5), for which we need to take
t = 1/2 + y - (z.
So let
h(z, y) = H(I/2 +y- (z),
so h is in C(R x T), and it is real-valued and bounded.
Also,
h(z+21-', y+2v) = H(I/2+y+2v-(z-2v) = H(I/2+y-(z) = h(z, y),
so h satisfies 5).
Now, for (z,y) E [0,21-'] x [0,1]' set
F(I/2 + y - (z) if y :::; z/(21-')

{
I(z,y) = e(c(y + v»F(I/2 + y - (z) if y ~ z/(21-')
To show f is continuous it suffices to prove that F(I/2+y-(z) =

when y = z/(21-'), and that follows from the fact that F vanishes on
°
[1/2,1]' and from the conditions on I' and v.
Now, since I(z, 1) = I(z, 0), f is continuous on [0,21'] X T. We want
to extend I to R x T by letting
I(z + 21', y) = e( -c(y - v»/(z, y - 2v),
so as to have I satisfy 4). We only need to show that
1(21', y) = e( -c(y - v»/(O, y - 2v) for any yET.
For an arbitrary y E R, let le, 1 E Z be such that y + le, and y - 2v + 1

are in [0,1]. Then,
1(21-', y) = F(I/2 + y + k - 2v) = F(I/2 + y - 2v), and
f(O, y - 2v) = f(O, y - 2v + I) = e(c(y - v + I»F(I/2 + y - 2v) =

= e(c(y - v»f(21-', y),
312 B. Abadie
as wanted, and f, extended to R x T as above, satisfies 4). It remains
to show that f and g satisfy 1), 2) and 3):
1) I/(z, y)/(z + l,y)1 = IF(I/2 + y - (z)F(I/2 + y - (z - (I = 0, by

1)'.
2) I/(z, y)[h(z + 1, y) + h(z, y) - 1]1 = IF(I/2 + y - (z)[H(I/2 + y-
(z - () + H(I/2 + y - (z) - 1]1 = 0, by 2)'.
3) I/(z, y)1 2 + I/(:c - 1, y)12 = IF(I/2 + y - (:c)12 + IF(I/2 + y - (:c +
01 2 =
= H(I/2 + y - (:c)[1 - H(I/2 + y - (:c)] = h(:c, y)(1 - h(:c, y)), by
3)'.
Therefore P is a projection on Eo. Besides,
Jor" (JT[ H(I/2 + y - (:c)dy)d:c = Jor" (= 2JJ( = 2v, by 5)'.
Q.E.D.
The following propositions enable us to extend the previous re-

sults by dropping the restrictions on JJ and v.
Notation. In Propositions 2 and 3 we denote by II the faithful rep-

resentation of D~II on L2(R x T x Z) obtained by restriction of the
left regular representation of the multiplier algebra of Co( R x T) x..\ Z
on L2(R x T x Z), i.e.
(IIC)e)(z, y,p) = L 4i(:c + 2PJJ, Y + 2pv, q)e(:c, y,p - q),

qEZ
for 4i E Do, e E L2(R x T x Z), and (z,y,p) E R x T x Z. Notice

that II is faithful because Z is amenable ([Pd, 7.7.5 and 7.7.7.]).
Proposition 2 There is a trace-preserving isomorphism between

D~II and D~+k,II+1' for all k, I E Z.
Proof: It is clear that 4i 1-+ C) is an isomorphism between

D~II and D~,1I+1' so let us assume 1= 0, k = 1.
Let J : D~+1,1I --+ D~II be defined at the level of functions in Do by:
(J4i)(:c, y,p) = e(c( 4p 3 v / 3 - p2y))C)(:c, y,p).

It is easily checked that J iP E D~v for all iP E D~+1,v- Besides, the
unitary operator U : L2(R x T x Z) - - L2(R x T x Z) given by
intertwines IIJ~ and n~:
(n~uo(x, y,p) = I.: iP(x + 2p(Jl + 1), y + 2pv, q)Ue(x, y,P - q) =

qEZ
= I.: e( -2pcq(y + (2p - q)v)e(c(( -4v/3)(p _ q)3 _ (p _ q)2y)).

qEZ
.iP(x + 2PJl, y + 2pv, q)e(x, y,p - q) = e(c( _ 4vp3/3 _ p2 y)).

L e(c( 4q3v/3 - q2(y + 2pv))iP(x + 2PJl, y + 2pv, q)e(x, y,P - q) =
qEZ
= (UnJ~O(x,y,p).
Also,
T(JiP) = 101 kJiP(X,y,O) = 101 iP(x,y,O) = T(iP).
Q.E.D.
Proposition 3 There lS a trace-preserving isomorphism between

D~v and D=-!-',_v'
Proof: Let J : D~v - - D:'!-',_v be defined, at the level of func-

tions, by:
(JiP)(X, y,p) = iP( -x, -y,p).
It is easily checked that JiP E D-!-',-v. Besides, the unitary
operator
U : L2(R x T x Z) - - L2(R x T x Z) defined by
(Ue)(x, y,p) = e( -x, -y,p)
intertwines n~ and nJ~:
[nJ~(U e)](x, y, p) = I.:(JiP)(x - 2PJl, y - 2pv, q )e( -x, -y, p - q) =

qEZ
= L iP( -x + 2PJl, -y + 2pv, q)e( -x, -y,p - q) =

qEZ
314 B. Abadie
= (II~e)(-x,-y,p) = (UII~e)(x,y,p).
Finally, J preserves the trace:
T( J cJ?) = 11'2 cJ? ( - x , - y, 0) = T ( cJ? ).

Q.E.D.
Theorem 4 The range of the trace on Ko(D~/I) contains the set

Z + 2p.Z + 2vZ.
Proof: We obviously have Z ~ T(Ko(D~/I))' since D~/I has an

identity element. Besides, it follows from Theorem 1 that 2p.Z ~
T(Ko(D~/I))' So it only remains to show that 2vZ ~ T(Ko(D~J).
Let k E Z be such that v' = ±v + k and v' E [0,1/2]. Then
one can find I E Z and p.' = ±p. + I such that p.' ~ 1. Thus, by
Propositions 2 and 3 we have that T(Ko(D~'vl)) = T(Ko(D~/I))'
Now, by Theorem 1 there is a projection in Mm(D~'/I') for some
positive integer m with trace 2v ' = ±2v + 2k, which ends the proof.
Q.E.D.
Remark. It can be shown ([Ab1D that the inclusion in the previous

theorem is actually an identity.
Acknowledgement. I am glad to thank my thesis advisor, Marc
Rieffel for his constant support and for many helpful suggestions and
comments.
References
[Ab1] Abadie, B. On the K-theory of non-commutative Heisenberg

manifolds.
Ph. D. Dissertation. University of California at Berkeley,
1992.
[Ab2] Abadie, B. Generalized fixed-point algebras of certain actions

on crossed products.
Submitted for publication.
[Pd] Pedersen, G. C*-algebras and their automorphism groups.

Academic Press, London/N ew York/San Francisco (1979).
[Rfl] Rieffel, M. C*-algebras associated with irrational rotations.

Pacific Journal of Mathematics Vol. 93, 2, (1981), 415-429.
[Rf2] Rieffel, M. Projective modules over higher dimensional non-

commutative tori.
Can. J. Math. Vol. XL. No 2. (1988) 257-338.
[Rf3] Rieffel, M. Deformation quantization of Heisenberg mani-

folds.
Commun. Math. Phys. 122 (1989), 531-562.
Centro de Matematica
Facultad de Ciencias
Eduardo Acevedo 1139
CP 11200
Montevideo-Uruguay.
e-mail: abadie@cmat.edu.uy
Deformations of topological spaces
predicted by E- theory
Marius Dadarlat *
University of Maryland
College Park, MD 20742
Terry A. Loring t
Department of Mathematics and Statistics
University of New Mexico
Albuquerque, NM 87131
1 Introduction
Let X be a locally compact space. By a deformation of X we mean a
continuous field {At It E [0, I]} of C*-algebras with Ao ~ Co(X), and
{At It E (0, I]} ~ B x (0,1],
for a fixed C'-algebra B. Replacing Co(X) by another C·-algebra A, we

generalize this to a deformation of one C*-algebra to another. This is a basic
interpretation of deformation-it reflects only the topology of X and omits
more general fields of algebras-but is an important one. This importance
is seen in the relation to E-theory and the examples [3, 9, 13, 14] that have
ansen.
Deformations are, in fact, very common. About the only requirement
for a C·-algebra to arise as a deformed CW-complex is that it have the
·current address: Purdue University, West Lafayette, IN 47907

t partially supported by NSF grant DMS-9007347
Deformations of topological spaces 317
correct K-theory. This fact follows from our calculations in "unsuspended"

E-theory [6]. We will explicitly describe one of the deformations predicted by
these calculations: a deformation of a three-dimensional CW complex into
a dimension-drop interval. We hope this example will further clarify the
role of the dimension-drop interval as a building block in Elliott's inductive
limits [7].
Recall, from [3], that an asymptotic morphism (cpt) : A --. B between
C* -algebras is a collection of maps CPt : A --. B for t E [1,00) such that for
a, b E A and a E C, as t --. 00,
IIcpt{ab) - cpt{a)cpt{b)11 --. 0,

IIcpt{a*)-cpt{a)*1I --. 0,
Ilcpt{aa + b) - acpt{a) - cpt{b)11 --. 0
and t 1-+ CPt{ a) is continuous. We say that (cpt) is injective if also
liminfllcpt{a)1I > 0
for all nonzero a.

Injective asymptotic morphisms correspond to deformations. Thus, we
will work in the context of asymptotic morphisms. See [3, 4, 12] for an
explanation of this correspondence and the definitions of equivalence and
homotopy for asymptotic morphisms.
The following result often gives the easiest way to show a given asymp-
totic morphism is injective. First, we recall how an asymptotic morphism
(cpt) : A --. B induces maps on K-theory. Given a projection p in A, the
class of cP*{[PJ) in Ko{B) is represented by any projection that is close to
CPt{p) for some sufficiently large value of t. For projections, and unitaries, in
Mk{A) a similar construction is used.
Proposition 1 Suppose X U {pt} is a compact, orientable manifold. If

an asymptotic morphism (cpt) : Co{X) -t B induces an injective map on
K -theory then (cpt) is injective.
318 M. Dad1irlat and T. A. Loring
Remark 2 This type of result holds more generally. In particular, it holds

for Rp 2 , the Klein bottle and the CW complex discussed in section 3. This
remark and Proposition 1 are special cases of [12, Theorem 1].
Example 3 Our first example is an asymptotic morphism (at) : CO(R2) -

K which induces an isomorphism on K-theory. This is well-known in several
contexts.
We regard CO(R2) as the universal CO-algebra generated by selfadjoint
element h and normal element N subject to the relation h = h2 + N* N so
that the generator of Ko( CO(R2)) ~ Z is just
An asymptotic morphism from A to B is given, up to equivalence, by a

*-homomorphism from A to B=, where B= is the CO-algebra described in
[3]. Therefore, if A is universal for a set of relations one need only define
the paths in B that are to be the images of the generators. In this case, we
need only define at(h) and at(N).
Let S denote the unilateral shift and, for t E [1, (Xl), let D t denote the
diagonal operator whose diagonal is the sequence
lit, 2It, . .. , [t]lt, 1, 1, ....
Set at(h) =1- D t and at(N) = JD t - DiS. Since the required relations
hold asymptotically, this determines at. The fact that this induces an iso-
morphism on Ko follows from the calculation, in [2] or [8], of the spectrum
of
See [10] for more details and a modification of this example that produces
deformations of Rp2 and the Klein bottle.
2 Unsuspended E-theory
Let A and B be C·-algebras. For convenience, we shall assume that A and
B are separable and nuclear. We will use the notation
[A, B] = homotopy classes of *-homomorphisms,

[[A, B]] homotopy classes of asymptotic morphisms.
We will use the following isomorphisms, from [3],
KK(A,B) ~ E(A,B)
~ [[SA 0 K, SB 0 K]]
'" [[S2 A 0 K, S2 B 0 K]].
We now arrive at our main result. By Xu {pt} we mean the one-point

compactification of a locally compact space X. Combined with Proposition 1
this result guarantees the existence of many deformations.
Proposition 4 If Xu {pt} is a connected, finite OW complex then the

suspenslOn map
[[Co(X), B 129 }C]]-. KK(Oo(X), B)
is an isomorphism.
The proof of this will be given in [6]. The inverse map may be described
as follows. Let
be a *-homomorphism inducing an isomorphism on K-theory. By [5, Corol-

lary 3.1.8] there exists a map
whose suspension is homotopic to f3 0 idco(x). Composition, on appropriate

sides, by f3x and the asymptotic morphism
320 M. Dlidarlat and T. A. Loring
(see Example 3) defines the inverse mapping
Using the universal coefficient theorem we obtain a corollary.
Corollary 5 If Xu {pt} is a compact, orientable manifold and
7] : K*(X) --t K*(B)
is an isomorphism then there exists a deformation of X to B ® K which

induces 7].
3 Matricial torsion
Consider the three-dimensional CW complex obtained by attaching, with
degree two, the boundary of a three-cell B3 to a two-sphere S2. Remove the
base-point (which sits in the copy of S2) and call the result X. That is,
where ( : 8B3 --t S2 has degree 2. Thus Ko(Co(X)) = 0 and Kl(CO(X)) S;!
Z/2.
Let B denote the non-unital dimension-drop interval, that is,
B = {J E CoCCO, 1], M 2 ) I f(l) is scalar }.
One may compute Ko(B) = 0 and Kl(B) S;! Z/2.

We know, by Corollary 5, that there is an asymptotic morphism
inducing an isomorphism on K-theory. This is an example of topological

torsion being "quantized" into matricial torsion. Our goal is to find .,p ex-
plicitly.
We first must be more explicit about the attaching map and the asso-
ciated *-homomorphism 8 : CO(R2) ---; CO(R2). Using the generators and
relations of Example 3, we determine 8 by setting
8(h) f(h),
8(N) 91(h)N + 92(h)N*
where f,91 and 92 are functions of the form
112 1
112
~ 1
which satisfy 9192 = ° and f(t) = f(t)2 +( (t)2 + 92(t)Z)(t - t Z).
We will also need L : /1..: ---; /1..: ® M2 gi :n by L(T) = T ® I. With the
additional notation of 01 indicating evalua .. on at 1, we have two pull-back
diagrams:
Co(X)
Co(O, 1] ® CO(R2)
/ \C (R2)
O CO(O, 1] ® A.: ® Mz
o~ ~ 01~
CO(R2)
322 M. Dadarlat and T. A. Loring
Lemma 6 There exists an asymptotic morphism (cpt) : CO(R2) --+ M 2(K)

such that, for all t,
image(~, 0 6) \; {[ TTl TElc}

which induces an isomorphism on K -theory.
We defer the proof until after we see how the lemma is used.
Consider the following commutative diagram (commuting exactly for
each t):
Here "It is the unique solution to £O"It = CPto(}. Since cP induces an isomorphism
on K-theory, "I must as well.
Since these maps are not *-homomorphisms we cannot immediately in-
voke the pull-back property. However, simply restricting idco(o,lj ® CPt pro-
duces"pt : Co(X) --+ B®K.:. Now considering the K-theory of the commuting
diagram
Co(O, 1) ® CO(R2)
t
Co(X)
it is easy to see that "p induces an isomorphism on K-theory.
Proof (Lemma 6) In order to specify (cpt) : Co(R2) --+ M 2 (K), it suffices to

specify where the generators hand N are sent. At t =n EN, we shall have
cp,,(h) = H" and cp,,(N) = N" where H" and N" are the following elements
Defonnations of topological spaces 323
h
were, lor J• -- 1, ... , 2 ,at}
L " · -- J·/2,,+1 an d R. --
/oJ} J. at} - 2
atj.
We are interested, more generally, in matrices A, BE M 2 (Mk ) such that

the following relations hold:
II[A, Blll, II[B, B*lll, II [A, B*lll :5 f
IIA-A*II :5 f
IIA2 + B*B - All :5 f (1)
f(A) E Mk I8l I
gl(A)B + g2(A)B* E Mk I8l I
It may be checked that H.. and N.. satisfy (1), for some fn ~ 0, with f .. --t 0
as n --t 00.
324 M. Dlldarlat and T. A. Loring
We will need some auxiliary matrices in M2(M2n) :
1/2
1/2
1/2
1/2
1/2
1/2
0 1/2
1/2 0
1/2 0
N,,=
0 1/2
o '.
'.1/2
1/2 0
A path of unit aries, multiplied by 1/2, in the lower-right-hand corner
will create a path satisfying (1) from H", Nn to
1/2 o 1/2
1/2 1/2 0
, .
1/2 1/2 0
1/2 o 1/2
1/2 1/2 0
1/2 1/2 0
Now, deforming the pair of scalars (1/2,1/2) to (0,0) appropriately con-

tinues this path to the pair of matrices 0, O. By this argument, we have
reduced the construction of (<pt) to being able to connect Hn EB fIn' N n EB Nn

to Hn+1,Nn+1 via pairs satisfying (1).
Let
o o
f3I 0 o
so that
Let
1/2 o 1/2
1/2 1/2 0
, En =
1/2 1/2 0
so that
fIn =
[~
-n An1,N = [~-n1.
1-
n
iJ~
By Berg's technique, [1, 11], there exists a unitary W E M 2 n+l such that,
for some constant C, independent of n,
W [~l W' - B"H "C2-"I'
and (by keeping W trivial except for vectors in a "segment" avoiding the
"last" basis vector in each copy of C 2 " )
326 M. Diidarlat and T. A. Loring
Let W= W ® 12 • It follows from above that
An An
An An
W W*~
I-An I-An
I-An I-An
and
Bn
[ ]
En Bn+!
W W·~
On B~ On+! B~+!
0 E~
Notice that the two matrices on the left satisfy (1) By taking a path of
unitaries from W to I, we get paths satisfying (1) from H" (fdln , Nn ffi fin to
this pair. Linear interpolation from this pair to the pair on the right-hand
side gives a path satisfying (1), perhaps after increasing En. Finally, it is a
simple matter to slide the scalars which are on the diagonal of
A"
A"
to connect the right-hand pair to H"+1> Nn+!. We may conclude that (IPt)
exists with the properties specified in the lemma and IPn(h) = H" and
IPn(N) = N n. 0
References
[1] I. D. Berg, On operators which almost commute with the shift, J. Operator
Theory, 11 (1984) 365-377.
[2] M.-D. Choi, Almost commuting matrices need not be nearly commuting, Proc.
Amer. Math. Soc., 102 (1988),529-533.
[3] A. Connes and N. Higson, Deformations, morphismes asymptotiques et K-

theory bivariante, C. R. Acad. Sci. Paris, t. 310, Series I (1990) 101-106.
[4] A. Connes and N. Higson, asymptotic morphisms and operator K -theory, in
preparation.
[5] M. D3:d3:rlat and A. Nemethi, Shape theory and connective K -theory, J. Op-
erator Theory, 23 (1990), 207-291.
[6] M. D3:darlat and T. A. Loring, K-homology, asymptotic representations and
unsuspended E-theory, preprint.
[7] G. A. Elliott, On the classification of C' -algebras of real rank zero, J. reine
angew Math., to appear.
[8] R. Exel and T. A. Loring, Invariants of almost commuting unitaries, J.
Funct. Anal., 95 (1991) 364-376.
[9] S. Klimek and A. Lesniewski, Quantum Riemann surfaces I, The unit disk,
Comm. Math. Phys. 146 (1992), 103-122.
[10] T. A. Loring, Deformations of nonorientable surfaces as torsion E-theory
elements, C. R. Acad. Sci. Paris, t. 316, Serie I (1993) 341-346.
[11] T. A. Loring, Berg's technique for pseudo-actions with applications to AF
embeddings, Canad. J. Math., 43 (1991) 119-157.
[12] T. A. Loring, A test for injectivity for asymptotic morphisms, these proceed-
ings.
[13] M. A. Rieffel, Deformation quantization for actions of R d, Mem. Amer.
Math. Soc., to appear.
[14] S. L. Woronowicz, Twisted SU(2) group. An example of a non-commutative
differential calculus, Publ. RIMS, Kyoto Univ., 23 (1987), 117-181.
Analyticity, Uniform Averaging and K-Theory
Paul S. Muhly
University of Iowa
Iowa City, IA 52242
Chaoxin Qiu
University of Nevada at Reno
Reno, NY 89557
Jingbo Xia
State University of New York
Baffalo, NY
ABSTRACf
Let (A, R, a) be a C*-dynamical system~ let

Hoo(a) be the subalgebra of A consisting of those
a in A such that the Arveson spectrum of a is
contained in [0, 00), and let AU be the fixed point
subalgebra of A. In this note we investigate
conditions for the existence of an R-invariant
conditional expectation from A onto AU and we
show that when such an expectation exists, then the
K-theoryof HOO(a) is isomorphic to that of AU.
We show by example that if no such expectation
exists, then the K-theories may be different.
Unifonn averaging and K-theory 329
1. Iptroduction
Let (A, R, a) be a C*-dynamical system. By this we mean
that A isaC*-algebraand a= (atl tE R isaone-parameter

group of C*-automorphisms of A such that t~~1 at(x) - x II = 0
for all x E A. We follow [Ped] for the basic facts, notatIon and
terminology about such systems. In particular, we write sPa(x)
for the Arveson spectrum of x with respect to a. (Recall that
sPa(x) =(s E R : f"(s) =° for all [E {X}L }, where {X}L = ( [

E Ll(R): x *a [ =o} and x *a [: =
R
J[(t) at(x) dt; also f"
denotes the Fourier transform of f.) If E ~ R, we write A(E) for
the closure of ( x E A: sPa(x) ~ E }. It is known that if E is
closed, then A(E) already is (x E A: sPa(x) c E} while if E
is open, A(E) is the closure of those x E A with compact spectrum
contained in E. If E = [0, 00), we write HOO(a) for A([O,oo».
It is known that A ([0, 00» is a subalgebra of A ([ArD and
the reason for calling it HOO(a) is that A([O,oo» coincides with
the collection of all x E A such that for each p in the dual of A,
A*, the complex-valued function of t, p(a_t(x», extends to be
bounded and analytic in the upper half-plane (see Lemma 2.4
below). As a result, HOO(a) is often referred to as the algebra of
analytic elements in A with resppect to a. (However, we want to
emphasize that this terminology is !lQ1 universal. Some authors call

330 P. S. Muhly, C. Qiu, and J. Xia
x E A analytic, if t -7 ~(x) extends to an entire A-valued
function. The two notions are related, but we shall not refer to the
second one again.) Furthermore, it is known and not hard to prove
(see Lemma 2.2), that A( (O}), or as we shall simply write, A(O),
coincides with the fixed point algebra of n, A<x. Thus, as is easy to
see, an element x E A and its adjoint belong to HOO(n) precisely
when x is fixed by n.
Our primary objective is to show that if there is an n-
invariant, faithful conditional expectation from A onto An, then the
K-theories of HOO(n) and An are isomorphic. To achieve this
result, we require the development of a number of equivalent
conditions for the existence of such expectations. We feel that they
are of intrinsic interest and may also prove useful elsewhere.
The reason for the interest in calculating the K-theory of

HOO(n) is that algebras of this form appear naturally in a number of
contexts and the knowledge of their K-theory has proven useful in
understanding the spectral analysis (in the sense of invertibility) of
elements in these algebras. (See [Da], [Pi 1,2] and [Qi].) Moreover,
these algebras constitute a very rich class of non-self-adjoint
operator algebras whose K-theory can be computed.
In the next section, we list some basic facts that we shall use
in our proofs. The third section is devoted to an analysis of ergodic
averages and their relation to the notion of analyticity while the

Uniform averaging and K-theory 331
fourth section is devoted to the K-theoretic computations. In the
fifth section, we attend to some examples which show, in one
direction, the richness of the theory, and, in another direction, some
of its limitations.
2. Preliminaries
We recall here for easy reference several facts that we shall
use:
Lemma 2.1 ([Fo], Lemma 2) Let x be in A and let f be
In L1(R).
(i) sPa(x *a f) ~ sPa(x) (l supp(f\
(ii) x *a f =0 if sPa(x) is contained in the interior of the
set where f" is O.

(iii) X *a f = x if sPa(x) is contained in the interior of the
set where f" is 1.
Lemma 2.2 ([LM], Lemma 2.2,2.3) Let x be in A and
let be in A*.
(i) sPa(x) ~ {O} if and only if ~(x) =x for every t E R.

(ii) Define f(t) =<\>(a_t(x» for t E R. Then f is a
bounded, uniformly continuous funtion on R and SPt(O C sPa(x)
n sPa.(<\» where t is the standard left translation on R.
Lemma 2.3 ([01], Prop. 2.3.3, [LM], Lemma 3.11) A(O,
00) is a two-sided ideal of HOO(a) and A(O, 00) = V ( A([c,d]) :

O<c<d<oo }.
Consider the transposed action, a*, of a on the Banach
space dual of A, A*. It is not necessarily strongly continuous. Yet
using duality techniques (see [Ar] and [Ped]), one may profitably
define and analyze the spectrum of an element in A *. The
following lemma is the main result from such an analysis that we
shall use. In it, HOO(R) denotes the usual space of (boundary value
of) bounded analytic functions in the upper half-plane.
Lemma 2.4 ([LM], Corollary 3.13)
(i) cp is analytic in (A *, R, a*) if and only if cp
annihilates A (0,00), and this happens if and only if for all x in A,
the function defined by f(t) =cp(a_t(x» belongs to HOO(R).
(ii) x E A is analytic if and only if for all <\> in A*, the
function defined by f(t) =cp(cx_t(x» belongs to HOO(R).
Let M be the double dual of A. Then M is the
enveloping von Neumann algebra of A. If {Yt l t E R is the double

transpose of {all le R' then {Ytl te R is not continuous in any
useful sense, unless {a*t1t E R happens to be strongly continuous
(and this, as we have hinted at, is a rather rare occurence).
Nevertheless, we may contemplate the discrete dynamical system

(M, R d, y), where Rd denotes the real numbers with the discrete
topology. We write D for the fixed point algebra {x EM: Yt(x)
=x }. According to ([Eb 1], or [Run, for any invariant mean 11 on
R d, there is an expectation ~ of A into D defined by <~(x), <1»
= 11t«, ~(x»), x E A, E A*. In [Ebl], Eberlein called an
element x E A er~ic if ~(x) E A and he called A or {~ll E
R er~odic if each x E A is ergodic. Since this terminology
conflicts with the more common use ofthe term "ergodic", we shall
say that A is ergodic in the sense of Eberlein if it has this property.
The point of Theorem 3.1 below is that this notion is equivalent to
what we shall call uniform averaging and th~t seems to be a more
appropriate term, particularly in view of [Ox]. The following
proposition is due to Jorgensen.
Proposition 2.5 ([Jo], Theorem 6.1) A is ergodic in the
sense of Eberlein if and only if A is equal to the closure of A(-<:IO,
0) + A(O) + A(O, 00).
Proof Even though the discussion in Jorgensen's paper is
for unital C*-algebra, we can see obviously that this assumption is

not necessary and that his argument works for non unital cases.
D
We conclude this section with a lemma due to Eberlein
([Eb2], Theorem 1).
Lemma 2.6 Let A be a C*-algebra and let (A, R, a) be a

T
C*-dynamical system. If 21T -1 <1>( ut(x)) dt converges to 0 for
T
every <p E A * as T -700 ,then 2~ -1 al(x) dt converges to 0
in the norm of A as T -700 .
3. Uniform Avera~im:
Our objective in this section is to relate Eberlein's notion of
ergodicity with the kinds of averages that appear in the ergodic
theorem.
Theorem 3.1 The following assertions are equivalent:
(1) There is an R-invariant conditional expectation <l> from
A onto A(l such that, for any x E A,

T
2~ -1 al(x) dt converges to <l>(x) in norm, as T -700.
(2) There is an R-invariant conditional expectation from

A onto Aa such that x * aPy converges to (x) in nonn, as y
~oo, for any x in A, where J.l.y is the Poisson kernel on R.
(3) There is an R-invariant conditional expectation from
A onto Aa such that <1>* from A * into A *(0) is onto.
(4) A = the closure of A(-oo, 0) + Aa + A(O, 00) .
(5) A is ergodic in the sense of Eberlein.
Proof The equavilence of (4) and (5) was established in
Proposition 2.5. We will prove the implications: (4)::::) (2) ::::) (3)
::::) (4), (1)::::) (3), and (2,3,4) ::::) (1).
(4} ::::) (2). We first show that if x e A([a,b]) with

O<a<b<oo, then x * aJ.1 y ~ 0 in norm as y ~oo.
Indeed, choose positive numbers c and d such that [a,b] ~
[c,d] ~ (0,00) and choose a function fed e C':(R) (the Schwartz
class on R) satisfying supp( f"cd) ~ [c, d], f"cd(t) =1 for t E
[a,b], and O:$; f"cd(t):$; 1 for t e [a,b] (where" denotes the
Fourier transform). By Lemma 2.1, x =x *a fed· Thus x *a J.l.y =
x *a fed *t J.l.y . It now suffices to show that fed *t J.1y converges to
o in L1(R)-norm as y ~oo. (Recall that we denote the standard
left translation on R by 't.) We will write f for fed in the rest of
the proof. Since

(f *t J,ly)"(A) = f"(A) J.1 yA(A) = f"(A) exp(-YA) e C':(R),
we see that f *t Ily e C':(R), and thus (1 +41t2t 2 )(f *t Ily)(t)e

CJ. (R). Since
00
(f *t Il y)(t) = ( 1+4~2t2) (1 +41t2t2 )(f *t Ily }(t) ,
it suffices to show that (1 +41t2t2 )(f *t Il y )(t) converges to 0 in
L2(R)-nonn as y ~oo.
d2
It is well-known that --2-fl$A) = (-41t 2t2 f(t»I\(A) for
dA
any feel.00 (R). Thus
(1 +41t 2t2f(t»I\(A)
d2
= fl\(A) - --fl\(A).
dA2
Applying this to f *t Ily' we get
Notice that since supp( fl$ C [c, d] ~ (0,00), it is clear now that
((1 +41t2t2 )(f *t ll y »)I' converges to 0 in L2(R)-nonn as y ~oo.
Therefore (1 +41t;2 )(f *t Ily)(t) converges to 0 in L2(R)-nonn as
y~oo.
Suppose now that we are given x = x' + x" + x'" with x'
e A(-oo, 0), x" E Afl, x'" E A(O, 00). By Lemma 2.3, we see that
x"'*a Ily converges to 0 in nonn as y ~oo. Similarly, x'*a Ily
converges to 0 in nonn as y ~oo. Since x" E Afl, alex) = x for

every t E R. It results that x"*a Ily = x". Therefore, x *a J..l y
converges to x" in norm as y ~oo.
Since (4) is equivalent to (5), there is an R-invariant

conditional expectation <1> from A onto Aa such that <1>(x) =
<1>(x' + x" + XIII) = x". And as A = the closure of
A(-00, 0) + Aa + A(O, 00), we conclude that x *a J..ly converges to
<1>(x) in norm as y ~oo for all x E A.

(2) ~ (3) We only need to show that <1>* is onto. Recall
that <1>*: A * ~ A*(0) is defined by (<1>*(<\»)(x) = <\>(<l>(x» and <\>
E A*(O) if and only if «(1l(x» = <\>(x), for all
x E A and t E R. Let <\> E A*(O) and x E A be given. As x *a
J..ly ~ <1>(x), we have
<\>(x) = <\>(x *a Ily ) ~ <\>(<1>(x» (as <\>(~(x» = <\>(x) ).
Thus <\>(<1>(x» = <b(x), i.e., <1>*(<\» = <\>. We conclude that <1>*: A*

~ A*(0) is onto.
(3) ~ (4) Conversely let <\> (* 0) E A* be such that 

annihilates A(-oo, 0) + Aa + A(O, 00). By Lemma 2.4, both <\>
and <\>* are analytic and thus sPa.(<\» ={O}, i.e., (1*l(<\» = <1>, i.e., <\>
(* 0) E A*(0). Since <1>*: A* ~ A*(0) is onto, there exists a y (*
0) E A* such that <1>*(y) =<\>. Thus <1>*(<\» =<\>. It results that

<\>(x) = <\>(<1>(x» = 0 (as <1>(x) E A(O» for all x E A. A
contradiction.
(1) ~ (3) Let $ E A*(O), then $(~(x» =$(x) for all x E
J
T
A and t E R. Since 2~ al(x)dt ~ <1>(x) in nonn as T ~co,
we have
$(x) = $ ( 2T
1
J T
al(x) dt ) ~ $(<1>(x».
It results that <1>*($) = $. Thus <1>*: A* -+ A*(O) is onto.
(2,3,4) ~ (1) As a first step, we will prove that
J
T
2~ alex) dt ~0 in nonn if there are positive numbers a and
b such that x E A([a,b)). By Lemma 2.6, it suffices to show that

1 T
J
cp ( 2T al(x) dt ) ~ 0 (as T ~co) for all CP E A*.
We first note that f(t) =$(a_t(x» belongs to HCO(R) and
spt(f) ~[a,bl ~(O,oo) as shown in Lemma 2.2 and 2.4. Regarding
LCO(R) as the commutative von Neumann algebra acting on L2(R),
we see that 2~ J T
'tl(f) dt ~0 (as T ~co) in the cr(LCO(R),
LCO(R).) - topology. It follows that, for any;, 11 e L2(R),
f (ir]f(S-tldt) ~(slT\(sl-dt ---) 0 (as T ---)-l·

R
We will shrink the support of ; and 11 to show that

-1 o.t(x) dt )
1 T
 (zT ---7 0 (as T ---7 00 ) for all E A*. Fix E >
O. There is a 0 > 0 such that II x - o.s(x) II < E whenever I s I < o.

Let ~(s) =;0 X[_~,~] =11(s). We see that there is a K > 0 such
that, if T > K, then
o
I ;0 f (2~]$(a1_s(x»dt) ds I
-0
f
: I (2~]f(S-I)dt) i;(s)T](s)-dt I < E .
R
It follows that
o
(o.s_t(x»dt) ds I
-0
o
+ I 1-
20
r
f (2~-1 CPt a_1(x)("o(x) - as(x)))dt) ds
-0
< E + II f 11/1 x II II x - o.s(x) liSE + II 1111 a /I E .
This shows that ~ J

1 T
(2T at(x) dt ) ~0 (as T ~oo) for all ~
E A*. Thus,
1
~ (2T J~(x)
T
dt) ~0 (as T ~oo) for all ~E A*.
By Lemma 2.6, we have 2~ JT

at(x) dt ~0 (as T ~oo) .
We may now use the same argument as in (4) ~ (2) to
conclude that
(lilT) JT
at(x) dt ~ (x)
in norm as T ~oo for any x in A. The proof is complete.
D
On the basis of Theorem 3.1, we shall refer to (A,
(atl tE R) or simply to A or {~}tE R as uniformly avera~in~ if
any of the conditions of that result are satisfied.
4 K-eroups of Analvtic Subaleebras
We tum now to computing the K-groups for an analytic
subalgebra. We will use the terminology, notation and basic facts

on K-theory found in [BI]. One of the key facts about K-theory is
that it is a homotopy invariant theory. This implies that if B is a

Banach algebra and if {1t t }lE [0,1] is a "homotopy" of
homomorphisms of B in the sense made precise in Theorem 4.1

such that 1to is the identity map and 1tl is a "projection" onto a
subalgebra, i.e., 1tl is a retraction, then the K-theory of B and the
K-theory of 1t 1(B) coincide. In our setting B is HOO(o.) and 1tl
is <1>. The fact that <1> is a homomorphism is well known (see
[LM] or [Jo]) and can be proved on the basis of Theorem 3.1, once
one recognizes that the kernel of <1> is A(O, 00), which is an ideal in
HOO(o.). Alternatively, this fact comes out of the proof of the
following theorem.
Theorem 4.1 Suppose (A, R, a.) is a C*-dynamical

system that is uniformly averaging, then
(i=O,I).
There is a similar result in [PW] for nest subalgebras of
UHF algebras.
Proof Define a map F: HOO(o.) x [0, 1] ~ HOO(o.) by
if t =0
F(a, t) ={ : * (l ~ y if t < 1 where y = tan(t1t!2)
<1>(a) if t =1
We will prove that F has the following properties:
(1) F(a, 0) = a, F(a, 1) = <1>(a), for all a E H OO(a) ;
(2) F(., t) is a homomorphism of HOO(a) for all t E
[0,1]; and
(3) F(a,.) is a continuous path from a to <1>(a) for all a
Thus F is a homotopy of homomorphisms of HOO(a) joining the
identity map to <1>. It is evident that K/HOO(a» == Kj(A(O» (i =

0, 1), if F satisfies (1) - (3).
(1) is trivial.
(2) If t = 0, then F(a, 0) = a is the identity map while if t
= 1, then F(a, 1) = <1>(a). We commented above that <1> is a
homomorphism from HOO(a) onto A(O). However, this will
follow from part 2 of Theorem 3.1 when we prove that each F(., t),
o $ t < 1, is a homomorphism. However, to show this, we need to
show that convolution with respect to the Poisson kernel is
multiplicative, i.e., we want to show that
(a b) *a Ily = (a *a Ily ) (b *a Ily ) (*)
for any a and b in HOO(a). Given a <\> E A*, we notice that the
function f(t, s) defined by f(t, s) =<\>(a_t(a) a_s(b» belongs to the
analytic subalgebra, which will be denoted by T, in the W*-
dynamical (LOO(R 2), R2, 't 2), with respect to the subsemigroup
R2 , by Lemma 2.4. (See Lemma 2.2, 2.3 in [LM] for more

+
details.) It thus suffices to show that
R
Jf(t,t)J.1 (t)dt = Jf(t,s) J.1 (t)J.1 (s) dt ds.
y
RxR
y y (**)
Recall that since T is equal to HOO(R) ® HOO(R), it suffices to
show (**) for the functions that can be written as f(t, s) = get) h(s)
with g, h in HOO(R). As the measure ~dt is multiplicative on
HOO(R), we have
J g(t)h(t)J.1y<t)dt = (J g(t)J.1y<t)dt)( J h(t)J.1y<t)dt)
=
RxR
Jg(t)h(s) J.1 (tW (s) dt ds.
y y
This proves (2).
(3) Given a e HOO(a) , we need to show that F(a, t) is
conti;~ JOUS in t. This is true for t =1 because of Theorem 3.1.

For t < I, all we need to show is that (a * aJ.1 y ) is continuous in y
and that it converges to a as y approaches to O. But this is a well-
known fact proved using a standard argument. The proof is

complete. D
5 Examples
344 P. S. Muhly. C. Qiu. and J. Xia
Theorem 4.3 begs the natural question: To what extent does
it depend on the assumption that there is an expectation <l> from A
onto Aa satisfying the conditions of Theorem 3.1? To see that
some condition is necessary, we recall the well-known fact that on
the circle, the conjugate function of a function which is continuous
except for a jump discontinuity must be unbounded [Z]. Keep this
in mind, take the Cayley transform and consider the C*-algebra C
of all bounded uniformly continuous functions on the line. Then R
acts on C through translation, giving a one-parameter group {~}t
E R of *-automorphisms of C. The fixed point algebra consists
solely of the constant functions and it is easy to see that there is no
a-invariant expectation <l> from C onto the constants satisfying the
conditions of Theorem 3.1. Indeed, if such an expectation were to
exist, then the flow on the maximal ideal space of C would be
strictly ergodic [Ox] and that, clearly, is not the case.
In this setting, HOO(a) = C (\ HOO(R) and we claim that
K}(HOO(a»;t= 0 = K](CU). (The latter equality holds since CU is
the scalars.) Let C!> be a smooth, non-negative function on R, 0 S
C!> S 1, such that c!>(x) =0, x S -1, and C!>(x) = 1, x ~ 1. (The Cayley
transform of C!> has a jump discontinuity at 1.) Define 'I'(t) to be
f
the principal value integral
1
'I'(t) = tJ,lim0 -It1 s - -)ds
c!>(s) ( -
1+s2
~O.
t-s
IIl-sll~E
Then '" is a conjugate function of <» and so must be unbounded,
as mentioned earlier. (Actually, straightforward estimates show that
",(t) =O(logl t I).) The discontinuity of '" is at 00 because the
discontinuity of the Cayley transfonn of '" is at 1. In fact, '" is
unifonnly continuous on R because ",(t) - ",(t') is the singular
integral l'-t
1t
f <» (s) ( 1
(t-s)(1'-s)
) ds. The function <» + i", extends
to be analytic in the upper half-plane through convolution with the

Poisson kernel z ~ l.
1t
f [(t) + i",(t)] ( 1m z 2) dt, and so
Iz-tl
R
u(t) := exp [<»(t) + i",(t)] is a bounded uniformly continuous
function on R that lies in H OO (a). Also, of course, 1/u lies in
HOO(a).
On the other hand, arg(u(t» - arg(u(t'» = ",(t) - ",(t') is
unbounded as t and t" vary over R, and so u cannot be written

as exp(f), for any f E C. We next show that u represents a
nontrivial element in K1(C). Suppose [u] =0 in K1(C). Then
there would be an integer n such that the nxn matrix u- = diag(u,
1, 1, ... , 1) belongs to the component of GLn(C) which contains
the identity element. Since

det: GLn(C) ~ GL 1(C) = C- 1
(where C- l denotes the group of all invertible elements in C) is a
continuous map, det(u-) would belong to the identity component of
C- l . That would imply that u = det(u-) is an exponential, contrary

to what we just noted. The contradiction shows that [u] '# 0 in
K 1(C), as desired. Since u and Ilu both lie in H CO (a), we
conclude that K1(HCO(a» '# 0, too.
We tum now to a setting in which C*-algebras with
automorphism groups satisfying the conditions of Theorem 3.1 arise
quite naturally. Let G be an amenable, locally compact, r-discrete,
principal groupoid in the sense of Renault [Re] and let 8: G ~ R
be a continuous l-cocycle with the property that 8- 1(0) = G(O). For
f E Cc(G), set at(O(y) = exp(it8(y» fey). Then {at}tE R is a
one-parameter group of *-automorphisms of Cc(G) that extends to
a strongly continuous one parameter group of C*-automorphisms
of C*(G). The unit space of G, G(O), is a elopen subset of G and

Cc(G(O» may be viewed as the subalgebra of C*(G) consisting of
all elements in C*(G) that are supported on G(O). (Recall from
[Re] that our amenability hypothesis implies that elements of C*(G)
may be viewed as continuous functions on G that vanish at co.)
The amenability assumption is equivalent to the assertion that the

map <l> from C*(G) to CO(G(O»obtained by restricting functions
in C*(G) to G(O) is a faithful conditional expectation. From the
form of a, it is evident that is a-invariant. In fact, it can be
shown that every strongly continuous one-parameter group of *-

automorphisms {at}tE R of C*(G) mapping Co(G(O» onto
Co(G(O» and satisfying <l>o~= for all t is given by a real-

valued cocycle O. The assumption that 0- 1(0) = G(O) is simply the
assumption that the fixed point algebra of (~}t E R is Co(G(O».
Let P = ( YE G : O(y) ~ 0 }. It can be shown (see [MS]
and {MSS 1,2]) that P satisfies the following conditions: p.p ~ P,

P n ~1 = G(O), where ~1 == (y-1 : yE P}, and P u ~l =G.
Furthennore, W'>O(a) is the closure in C*(G) of Cc(P) and the
set of all f E C*(G) such that sPa(f) ~ (0, 00) is the closure of
Cc(P\G(O». It follows that condition (4) of Theorem 3.1 is satisfied
and therefore, from Theorem 4.3, we conclude that the K-theory of

HOO(a) is isomorphic to that of Co(G(O).
As a concrete example of the set up we have been
describing, let r be a countable, discrete, torsion free, abelian
group, let r act freely on a locally compact Hausdorff space X,
and let G =X x r be the transfonnation group groupoid. Then r

is isomorphic to a subgroup of R. If we let 8 : r ~ R be an
injective homomorphism and if we set O(x, y) =8(y), then the

freeness of the action guarentees that G is a principal groupoid of
the preceeding type and we see that 0 is a cocycle of the type we
require.
Examples of this special type are covered also by the
analysis in [Qi], which motivated the present study.

Acknowledgment We would like to thank Palle
Jorgensen for interesting conversations.
REFERENCES
[Ar] W. Arveson, On groups of automorphisms of operator

algebras, 1. Funct. Anal., 15 (1974), 217-243.
[Bl] B. Blackadar, K-theory for Operator Algebras,

Springer-Verlag, New York, 1986.
[Da] K. Davidson, Nest Algebras, Pitman Research Notes in

Mathematical Sciences, 191, Longman Scientific and
Technical, John Wiley and Soncs, New York, 1988.
[Eb 1] W. Eberlein, Abstract ergodic theorems and weak almost

periodic functions, Trans. Amer. Math. Soc. 67 (1949),
217-240.
[Eb2] W. Eberlein, Mean Ergodic Flows, Advances in

Mathematics 21 (1976),229-232.
[Fo] F. Forelli, Analytic and quasi-invariant measures, Acta

Math. 118 (1967), 33-59.
[Jo] P. Jorgensen, Ergodic properties of one-parameter

automorphism groups of operator algebras, J. Math.
Anal. Appl. 87 (1982), 354-372.
[LM] R. Loebl, P. Muhly, Analyticity and flows in von

neumann algebras, 1. Funct. Anal. 29 (1978), 214-252.
[MSSl] P. Muhly, K-S. Saito, and B. SoleI, Coordinates for

triangular operator algebras, Ann. of Math. 127(1988),
245-278.
[MSS2] P. Muhly, K-S. Saito, and B. SoleI, Coordinates for

triangular operator algebras II, Pac. J. Math. 137(1989),
335-369.
[MS] P. Muhly, B. SoleI, Subalgebras of groupoid C*-

algebras, J. reine angew. Math. 402(1989),41-75.
[01] D. Olesen, On spectral subspaces and their applications to

automorphism groups, Physics Seminar 1973-1974, V.
E. R. Scientifique de Luminy, Vniversite d'Aix-Marseille
II, 1974
[Ox] 1. Oxtoby, Ergodic sets, Bull Amer. Math. Soc. 58

(1952), 116-136.
[Ped] G. Pedersen, C*-algebras and their Automorphism

Groups, Academic Press, London - New York - San
Francisco,1979.
[PW] 1. Peters and B. Wagner, Triangular AF algebras and nest

subalgebras of VHF algebras, 1. Operator Theory, to
appear.
[Pi 1] D. Pitts, Factorization problems for nests: Factorization

methods and characterizations of the universal factorization
property, J. Funct. Anal. 79 (1988), 57-90
[Pi2] D. Pitts, On the Ko groups of nest algebras, K-Theory,

2, 737-752, 1989.
[Qi] C. Qiu, K-theory of analytic crossed products, Rocky

Mountain Journal of Mathematics, 22(1992), 1545-
1557.
[Re] J. Renault, A Groupoid Approach to C*-algebras, Lect.

Notes in Math. 793, Berlin - Heidelberg - New York,
1980
[Ru] D. Ruelle, Statistical Mechanics, Math. Phys.

Monograph Ser., Benjamin, London, 1969.
[Z] A. Zygmund, Trigonometric Series, Cambridge

University Press, London - New York, 1959.
BOUNDARY VALUE PROBLEMS FOR FUNCTIONS
ANALYTIC ON MULTIPLY CONNECTED DOMAINS
ON SPACES WITH A GENERAL WEIGHT
D. KURTZ, YU. LATUSHKIN, AND I. SPITKOVSKY
ABSTRACT. Necessary and sufficient conditions of Fredholmness, as well as

index formulas, are obtained for differential boundary value problems with
several unknown functions analytic on multiply connected domains. The com-
plex conjugation and (forward or backward) shifts are involved in boundary
conditions of above mentioned problems, and solutions are sought in weighted
spaces with weights satisfying only the Hunt-Muckenhoupt-Wheeden condi-
tion. The results are of an explicit nature in the case of piecewise continuous
coefficients.
O. INTRODUCTION
The theory of boundary value problems for analytic functions with bound-
ary conditions that contain derivatives, complex conjugation and shift (trans-
lation) of the argument is a well-developed field in the modern function the-
ory. The exposition of the history of the subject can be found in classical
books [1,7,19]; the survey [81 contains recent references.
In the simplest case a problem mentioned above can be formulated as
follows. Find a piecewise analytic function iI1 by the boundary condition
(0.1) a(t)t/I(a(t)) + b(t)t/I(a(t)) + e(t)4>(f3(t)) + d(t)4>(f3(t)) = h(t),
on the contour L that separates the complex plain C into the domains 6
and 0 = C \ (6 U L). Here 4>(t) and t/I(t) are limit values of the function
iI1(z) as z -+ t when z E 6 and z E 0 correspondingly, bar means complex
conjugation, a, b, e, d are given coefficients, h is a known function from some
functional space F of functions on L (say, from Lp(L)), and shifts a and
f3 are diffeomorphisms of L onto itself. One of possible generalizations of
(0.1) is the boundary value problem with boundary condition that contains
derivatives:
L {a/l(t)t/lC/ll(a(t)) + b/l(t)t/lC/ll(a(t))+
m
(0.2) /l=0
e/l(t)4>C/ll(f3(t)) + d/l(t)4>C/ll(f3(t)) } = h(t).
1991 Mathematics Subject Classification. Primary 30E25, 47G20; Secondary 45E05,
47A53.
Key words and phrases. Boundary value problems, integro-differential operators, shifts,
Fredholm operators, index, Hunt-Muckenhoupt-Wheeden weights.
The third author was supported in part by NSF Grant #9101143
Boundary value problems 351
The main goal in the theory of boundary value problems like these is to
obtain the conditions of Fredholmity and calculate index of the problem in
terms of its coefficients. Let us recall that (0.1) is called Fredholm if the set
H of right-hand parts such that problem is solvable forms a closed subspace
in F, and dimension [ of the set of solutions of homogeneous problem and
co dimension [' of H in F are finite. Then index K. of the problem is defined
as K. = [- I'.
Let us stress that the conditions of the Fredholmity of (0.1) and (0.2)
depend on the fact whether a and f3 preserve or reverse the orientation of
the contour, which domain contains infinity, whether domains under consid-
erations are multi connected, etc.
The boundary value problems (0.1) and (0.2) were generalized in several
different directions. Let us note first of all that t:,. and 0 can belong to the
different copies of the plain. In this case a (corresp. f3) will map L onto
the boundary of 0 (corresp. t:,.). Moreover, t:,. and 0 can be multi connected
domains, infinity can belong to anyone of them. And, finally, shifts can
map some components of the boundaries of t:,. and 0 preserving, and others
- reversing the orientation.
One of the goals of our paper is to consider the boundary value problems
of the type mentioned above in the most general setting, that contains all
listed difficulties. Let us formulate the setting.
Let r = {r n}:;'=1 and I = hm}~=1 be systems of compound Lya-
in km
punov's contours such that r n = U r nj and 1m = U Imk are boundaries
j=1 k=1
of domains t:,.n and Om, which can be bounded or unbounded. Let also
ir
L = {L r }~=1' where Lr = U Lri, r = 1, ... ,R, be a system of Lyapunov's
i=1
N M R
contours, and q = L: jn = L: km = L: ir'
n=1 m=1 r=1
Let us introduce a mapping (shift) a transforming curves from L onto
curves from I, and a shift f3, transforming curves from L onto curves from
r. More precisely, let a(Lri) = Imk, f3(Lri) = rnj, where (m,k) = &(r,i),
(n,j) = ~(r, i), with & and ~ being bijections of R = {(r, i) : i = 1, ... ,i r ,
r = 1, ... ,R} onto M = {(m,k) : k = I, ... ,km' m = I, ... ,M} and
}If = {(n,j) : j = 1, ... ,jn, n = 1, ... ,N} correspondingly. It is supposed
in the sequel that a and f3 have nonzero continuous derivatives on all the
curves L ri , preserve orientation of some of them (forward shifts), and change
orientation of others (backward shifts). Finally, let us consider the following
boundary value problem:
L {al-'(t)1/!~)(a(t)) + bl-'(t)1/!!:) (a(t)) } +

I-'m
1-'=0
(0.3)
L
Vn
{ev(t) r/>~)(f3(t)) + dv(t) r/><;)(f3(t))} = h(t),

v=o
tEL ri , (r,i)ER, (m,k)=&(r,i), (n,j)=~(r,i).
352 D. Kurtz, Yu. Latushkin, and I. Spitkovsky
Here superscripts (Il) or (v) mean differentiation, bar - means complex con-
jugation, Ilm, Vn E IN, aI" bp , ell, d ll , h are given, and .,pm, r/>n are unknown
functions.
This problem represents a natural generalization of (0.1), (0.2), and other
boundary value problems considered in [1], [5-7],[14-15], [18-19]. In fact, in
the simplest case N = M = R = 1 and Lr = r n = 'Ym the problem (0.3)
reduces to the differential boundary value problem (0.2) with a shift and
complex conjugation, for a pair of functions analytic on a multiply connected
domain. Chapter 2 of the classical monograph [19] is devoted to the last
problem, and a generalization of this simplest problem to the case of two
multiply connected domains is a subject of the dissertation [14]. The case
of several simply connected domains (N = M > 1, R = 1, in = k m = 1)
was considered in [181. Finally, the case when boundaries of several multiply
connected domains were mapped onto a boundary of one multiply connected
domain (N = 1, M > 1, Lr = r n), was considered in a series of papers
[5-61. Such attention to those types of problems is justified, of course, by
their importance for applications (see, for example, [19, Chapter 3)).
The second point of our attention is related to the choice of the space
F. The operator approach, developed in [9-13,15]' allows one to reduce a
problem of (0.3) type to a singular integral operator with Cauchy kernel
and 2 x 2 matrix coefficients as far as Fredholm criteria and index formulaes
are concerned (see Theorems 1.1 and 1.2 below for exact statements). That
is why it seems natural to choose as F the space where these operators
are bounded. One of the possible choices is a weighted space Lp. It is
well known [4] that the singular integral operator is bounded in the space
Lp(L,p) = {J:pf E Lp(L)} if and only if p E Ap, that is, the weight p
satisfies the famous Hunt-Muckenhoupt-Wheeden condition
(0.4) s~p I~I (i Ip(t)/P Idtl) IIp (i Ip(t)l-q Idtl) 1/q < 00
Here I are arbitrary arcs of Lj III stands for the length of I. In the present
paper we consider boundary value problems in Lp with arbitrary weights p
satisfying (0.4).
And, finally, the third moment is the appropriate choice of the classes
of coefficients. This choice is subordinated to the availability of results on
Fredholm theory of singular integral operators with matrix coefficients from
the class considered. For piecewise conitnuous coefficients and spaces Lp with
power weights pet) = n!l It-tilPj the corresponding results go back to the
classical works of I. Gohberg and N. Krupnik [3]. Recently these results
were generalized [171 to the case of arbitrary weights p E Ap. The Fredholm
criteria and index formula in this case still have transparent geometrical
sense (see Theorem 1.3 below). That is why in the present paper we make
a reduction of the boundary value problem to the singular integral operator
for arbitrary bounded coefficients, and then give explicit formulas for the
piece-wise continuous case.
1. PRELIMINARY RESULTS
We will consider a differential boundary value problem (0.3) in Sobolev
spaces W~ = W~(L, p) with a weight p satisfying (0.4). It means that the
R
right side in (0.3) is chosen in a space Lp(L, p) = EO Lp(Lr. p), and un-
r=l
known functions tPn and ,pm are searched for in classes W~"n)+ (r, poP) =
N M
EO W~"n)+(r n, poP) and W~l'm)+(-y, poa) = EO W~l'm)+(-ym, poa), respec-
n=l m=l
R
tively. Coefficients of (0.3) are arbitrary functions in Loo(L) = EO Loo(Lr).
r=l
Let us denote by.No and Mo sets of such nand m, n = 1, ... , N, m =
1, ... , M that domains Don and Om correspondingly do not contain ooj a
number of such domains we'll denote by No and Mo.
The following result allows one to consider a problem
(1.1) a(t),pm(a(t)) + b(t),pm(a(t)) + e(t)tPn(p(t)) + d(t)tPn(P(t)) = h(t)

with no derivatives, instead of (0.3). Here and in what follows a, b, e, d stand
for coefficients of higher derivatives in (0.3):
(1.2)
a(t) = al'm(t), b(t) = bl'm(t), e(t) = e..n(t), d(t) = d..n(t),
t E L ri .
Theorem 1.1. Problems (0.3) and (1.1) are Fredlwlm simultaneously. The
index 1 K.D of the problem (0.3) is related to the index K. of the problem (1.1)
as follows:
K.D = K. + 4 ( L
mEMo
Jlm + L
nENo
lin) - 2 (f
m=l
k m Ilm + t
n=l
in lin) .
Theorem 1.1 follows directly from the fact that the differentiation operator
D is a compact operator (see [16]) from W~, 1> 1, into Lp. Moreover, D' is a
Fredholm operator from the space W~+ of Sobolev functions having analytic
continuations into the interior domain V, to the space Lt of Lp-functions
having such continuations. If this domain is 8-connected then (see [13]) the
index of D' equals -281 when 00 E V, and (4 - 28)/ otherwise.
An operator Tw corresponding to the problem (1.1) acts from the direct
+
sum Lt('Y,p) Lt(r,p) into Lp(L,p), where
M
Lt('Y,p) = 'L 0
Lt('Ym,p), Lt(r,p) =
m=l
(1.3)
n=l r=l
1 index in this paper is considered with regard to the field of real numbers
354 D. Kurtz, Yu. Latushkin, and 1. Spitkovsky
This operator Tw, up to a compact summand, coincides with a direct sum

of 1 x 2 block operators
Tri = [(a + bC)W",Pmk, (e + dC)WpPnj], (r, i) En,

+
acting from Lt(-rmk,P) Lt(fnj,p) into Lp(Lri,P). Here (m,k) = li(r,i),
(n,j) = ~(r, i), C is an operator of complex conjugation, a, b, c, d are oper-
ators of multiplication by corresponding functions, W'" and Wp are substitu-
tion operators: (W",)f(t) = f(OI(t)), (Wp)f(t) = f(f3(t)); Pmk and Pnj are
lliesz projections for contours /mk and f nj correspondingly. Each of these
operators Tri is Fredholm only simultaneously with a certain singular integral
2 X 2 block operator Kri. The structure of Kri depends on the orientation of
01 and f3 on Lri. Let us consider, for example, a case when both shifts Q' and f3
preserve orientation of L ri . In this case W",Pmk ~ Pri lIV"" WPPnj ~ Pnj Wp,
where "~"stands for equality up to a compact summand. One can prove (see
[10]) that a 2 X 1 block operator
Pmk W,;l ] +. +
ITri= [ PnjWp-1C : Lp(Lr;)-tLp(-rmk)+Lp(fnj)
is Fredholm, and compute its index.

A product Te = Tri ITri is a singular integral operator with conjugation
acting in the space Lp(Lri) and containing no shift operators. Using the
matrix identity of 1. Gohberg and N. Krupnik [2] one can also get rid of the
conjugation operator C in Te, passing meanwhile from Te to ](ri.
As a result, the problem (1.1) (and therefore (0.3)) is Fredholm simulta-
neously with a direct sum of operators K ri , and its index can be easily re-
calculated in terms of these operators. To state this equivalence accurately,
let us introduce the following notations:
L++ is a union of Lri on which 01 and f3 preserve direction,
L+_ is a union of Lri on which 01 preserves, and f3 reverses direction,
L_+ is a union of Lri on which 01 reverses, and f3 preserves direction,
L __ is a union of Lri on which both Q' and f3 reverse direction.
Starting with the coefficients of the problem (1.1), let us define functions
z, y, U, von L ri , (r, i) ERas follows:
z = ad - be, y = ae - bd, U = Idl2 - IW, v = lal 2- IW on L++

z = bd - ae, y = ad - be, u = IW - laJ2, v = IdJ2 - lel 2 on L+_
(1.4)
z = bd - ae, y = be - ad, u = Idl 2- Je12, v = IW - lal 2 on L_+
z = ad - be, y = bd - ae, u = lal 2- IW, v = Idl2 - lel 2 on L __
In case z-l E Loo(Lr), r = 1, ... , R, put
(1.5)
and define a singular integral operator on L~(Lr) by the formula ](r

PL, + GQL,.
Theorem 1.2. For problem (1.1) to be Fredholm it is necessary and suffi-
dent that
r = 1, ... ,R,
and all the operators K r are Fredholm, r = 1, ... ,R. If these conditions are
satisfied, then an index of(1.1) is given by the following formula:
1 R
K, = 2: E IndK r + 2(No + Mo - q).
r=l
Note that operators Kr (unlike (1.1)) do not contain any shifts or con-
jugations. The theory of singular integral operators Kr is well-understood.
Many results are available to compute explicitly their Fredholm conditions
and index.
The detailed proofs of Theorems 1.1 and 1.2 in case of unweighted spaces
Lp can be found in [13].
Let us consider now the situation when a, b, e, d are piecewise continuous
functions. According to Theorem 1.2, to calculate Fredholm conditions and
index formula for (0.3) it sufficies to obtain them for a 2 x 2 matrix singular
integral operator P+GQ with a matrix coefficient G given by (1.5) on a space
L; with a weight satisfying the Hunt-Muckenhoupt-Wheeden condition.
Let us cite a result from [17] on Fredholm properties and index formula
for singular integral operators on the space L;(f, p) with a weight p E Ap
For such p and an arbitrary T E r put
It is shown in [17, Lemma 2.1] that Tr is an open interval of a length not

greater than 27f. Let Tr=(-v_(T), 27f-V+(T)). Then
Theorem 1.3. Let G be a piecewise continuous matrix function on a Lya-

punov contour f, let Tl, ... ,Tm be all the points of discontinuity for G, and
f l , ... ,f m be all the arcs into which f is divided by Tl, ... ,Tm. Then for
operator KG = P + GQ to be Fredholm it is necessary and sufficient that
(i) det G(t) i- 0,
and
(ii) arg Ak(t) [v_(t), v+(t)] ,
~ k = 1, ... ,n.
Here Ak(t) are the eigenvalues of a matrix G-l(t - 0) G(t + 0), and arg is
chosen in [0, 27f).
If conditions (i) and (ii) are satisfied, then
IndKG
1 (m
= 27f ~ {argdetGLErj + f,;t;arg
m n
Ak(Tj)
)
-I,
where I is the number of summands arg Ak( Tj), j = 1, ... ,m, k = 1, ... ,n
such that arg Ak( Tj) > v+( Tj).
2. FREDHOLMNESS AND INDEX

Let the coefficients of problem (0.3) be piecewise continuous functions.
According to Theorems 1.1 and 1.2, to obtain Fredholm conditions and index
formula for (0.3) it suffices to consider a singular integral operator P + GQ
in a space L~(L, p) and a matrix coefficient G given by (1.5) with entries
v, u, y, z defined according to (1.4) and (1.2).
Directly from (1.4) it follows that identically on L
(2.1) u(t)v(t) = Iz(t)j2 - ly(t)j2, u(t), vet) e lR..
Let T be one of the discontinuity points of the coefficients (1.2). To make
use of Theorem 1.3 one has to compute eigenvalues Al and A2 of a matrix
G: 1G+ (here and in what follows subscripts "+" and "-" stand for one
sided limits at the point Tj arguments of functions are omitted for a sake of
brevity).
Lemma 2.1. Let entries of the matrix
G = ~ [~ ~y]
satisfy relations (2.1). Then A1,2 = fz!T~ (R ±~. where
(2.2) R= v_u++u_v++2Re(y_y+)
2Iz+z-1
Proof. One can verify directly that
G:1G+ =~ ['11 112],

z_ z+ 121 122
where
111 = v_U+ + y_y+, 121 = v+y_ - v_y+,

112 = u_y+ - u+y_. 122 = Y+Y_ + u_v+.
Therefore, A1,2 =~
2z_z+
(,ll + 122 ± "';(,11 - 122)2 + 4"Yl2/21)' Routine
calculations using (2.1) show that (,ll -,22)2+4,1 2/21 = 4(R2-1) Iz+ 12Iz_12,
which yields the desired formula. 0
Having denoted
0, R>l
z+
(2.3) 6 = arg-, Tf ={ arccos R, IRI::; 1
z_
71', R <-1.
we obtain the following three series of formulas for arg Al ,2:
arg Al = b + Tf ::; arg A2 = 271' + b - Tf forbe[O, Tf)

(2.4) arg A2 = b - Tf ::; arg Al = b + Tf for b e [Tf, 271' - Tf)
arg Al = b + Tf - 271' ::; arg A2 = b - Tf forb e [271' - Tf, 271')
Consider II± such that 0 < 11_ ::; 11+ ::; 271'.
Lemma 2.2. arg>'I,2 ~ [v_, v+] if and only if the closed line segments
L = [v_ - 0 - 1), v+ - 0 - 1)] and 1+ = [v_ - 0 + 1), v+ - 0 + 1)]
do not contain points 0, ±211":
(2.5)
Proof· It follows from (2.4) that arg >'1,2 ~ [v_, v+] iff one of conditions
(2.6) o E [0, 1)), O~ L, 211" ~ 1+

(2.7) o E [1), 211" - 1)), o ~ 1+, O~L
(2.8) o E [211" - 1), 211"), -211" ~ L, o ¢ 1+
is fulfilled.
It's clear that (2.5) yields each of the conditions (2.6)-(2.8). Let's prove
the converse.
Suppose, (2.6) is satisfied. Since v_ > 0 and 0 + T/ < 21) ~ 211', then
v_ - 0 - 1) > -211', and therefore -211' ¢ L. Since v_ > 0 and 1) - 0> 0,
then v_ - 0 + T/ > 0 and 0 ~ 1+, -211" ~ 1+. Lastly, v+ - 0 - T/ < 211' and
211' ~ L, which yields (2.5).
Suppose, (2.7) is satisfied. Since 0 < 211" - 1) and v_ > 0, then -211" <
v_ - 0 - 1) < v_ - 0 + 1) and -211' ~ L U 1+. Since 0 > 1) and v+ < 211', then
211' > v+ - 0 + T/ > v+ - 0 - 1), and 27l' ¢ L U 1+, which again yields (2.5).
Finally, let (2.8) be satisfied. Then 0 > 211" - 1) and v+ < 211", thus
v+-O-1) < 0, and 211", 0 ~ L. As 0 < 27l', v_ > 0,1) > 0, then v_ -0+1) >
-211", and -211" ~ 1+. Since 1) < 0, v+ < 27l', then v+ - 0 + 1) < 27l', and
211" ~ 1+, which completes the proof. 0
Observe that when the conditions of Lemma 2.2 are satisfied, the line
segment [v_ - 0, V+ + 0] is contained in one of open intervals (-211", -211" +
1)), (-211"+1), -1)), (-1), 1)), (1), 27l'-1)), (27l'-1) , 211"). Thus, Iv±-ol =f
1), and Iv± - 01 =f 211" - T/, and if Iv± - 01 > 1), then 0 - v+ and 0 - v_ have
the same sign.
Denote now by n the number of eigenvalues >'1,2 having argument greater
than v+. The following table for evaluating n can be obtained by direct
verification.
[0,1)) [T/,27l'-1)) [211" - 1), 27l')
(-211", -211"+1)) - - n=2
(-27l' + T/, -T/) - n=2 n=l
(-T/, T/) n=2 n=l n=O
(1),27l'-1)) n=l n=O -
(27l' - 1), 211") n=O - -
Here the first column indicates location of v± - 0, and the first row indicates
the location of O. "-" means that corresponding combination of locations
does not occur. For example, if 0 E [0, 1]), then 0 -1] < 0 < V_, and therefore
V± - 0 1. (-27r + TJ, -TJ)·
To illustrate how this table was obtained, let's verify, for example, that
n = 1 for 0 E [TJ, 27r - TJ) and V± - 0 E (-TJ, TJ)· Indeed, -TJ < V± - 0 < TJ
implies that 0 - 1'/ < V+ < 0 + TJ, and the result desired follows from (2.4).
All the other cases are treated the same way.
If conditions (2.5) are fulfilled, then a number
1, Iv± - 51 < TJ
(2.9) 1 +sgn(o - v±), Iv± - 51 < 27r -
1'/ < 1'/
1 + 2sgn(o - v±), 27r - TJ < 11J± - 51 < 27r.
is well defined.
Lemma 2.3. If the conditions of Lemma 2.2 are met, then
Proof. It follows from (2.4) that
25 - 27r(n - 1), 5 E [0, 1'/)

argAl +argA2 -27rn= { 25-27rn, 5E [TJ, 27r-TJ)
25 - 27r(n + 1), 5 E [27r - TJ, 27r)
The desired relation follows now from the table for n and the definition (2.9)
ofm. 0
Now we are ready to state the Fredholm criterion and index formula for
problems (0.3).
Let the coefficients (1.2) of higher derivatives in (0.3) be piecewise con-
tinuous on L functions, let {Tj}jEl be all their discontinuity points, and let
{Lj}jEl be the arcs into which L is divided by Tj. Denote by v± = V±(Tj)
the numbers in [0, 27r) related to the weight p at the point T = Tj as explained
before Theorem 1.3. Define functions z, y, u, v and a matrix function G on
L according to formulas (1.4)-(1.5), and introduce then R, 5, 1'/ by (2.2) and
(2.3) at all the points Tj, j E J.
Theorem 2.4.
(a) A problem (0.3) with piecewise continuous coefficients of higher deriva-
tives is Fredholm in a Sobolev space with a weight p satisfying the
Hunt-Muckenhoupt- Wheeden condition if and only if z( t ± 0) f:. 0,
tEL, and the closed line segments [v_ - 5 - 1'/, v+ - 5 - 1'/],
[v_ - 5 + 1'/, v+ - 5 + 1'/] do not contain the numbers 0, ±27r for
all the discontinuity points Tj.
(b) If these conditions are satisfied, then the index of the problem (0.3)
is given by
M
"'D =4 L 11m +4 L lin - 2 L km I1m-
mEMo nE.!Vo m=l
(2.10) 2 Lin lin + 2(No + Mo - q) + -1 LLi {argz(t)}Li +

N
n=l 7r
L (~arg ztTjj +- ~~ + m(Tj)) .

rj 7r Z
Proof. According to Theorems 1.1 and 1.2, the problem (0.3) is Fredholm
simultaneously with operators P + GQ having matrix coefficients (1.5), in
the spaces L;(Lri, p). Apply Theorem 1.3 to these operators. It follows from
(2.1) that
(2.11) det G(t) = z(t)/z(t)

and therefore the condition z(t ± 0) =i 0 is equivalent to statement (i) of
Theorem 1.3. Due to Lemma 2.2, statement (ii)of Theorem 1.3 coincides
with the second part of our conditon (a).
To prove part (b), note that the first five groups of summands in (2.10) are
due to Theorems 1.1 and 1.2, the sixth is due to Theorem 1.3 and relation
(2.11), and, finally, the last sum in (2.10) is a result of an application of
Theorem 1.3 and Lemma 2.3. 0
REFERENCES
1. F. D. Gahov, BoundaT!/ value problems, "Nauka", Moscow, 1977. (Russian)
2. I.Gohberg and N .Krupnik, On one-dimensional singular integral operators with shift,
Isv. Acad. Nauk Arm. SSR. Matematica. 8 (1973), no. I, 3-12. (Russian)
3. ___ , Systems of singular integral equations in weighted spaces L p , Soviet. Math.
Dokl. 10 (1969), 688-691.
4. R. Hunt, B. Muckenhoupt, R. Wheeden, Weighted norm inequalities for the conjugate
function and Hilbert transform, Trans. Amer. Math. Soc. 176 (1973), 227-251.
5. R. S. Isahanov, Some boundary value problems of analytic functions, Trudy Tbiliss.
Mat. Inst. Razmadze Akad. Nauk Gruzin. SSR, (Collection ofarticles on the equations
of mathematical physics, 6), vol. 52, 1976, pp. 62-80. Russian
6. ___ , A general problem for holomorphic functions, Trudy Tbiliss. Mat. Inst. Raz-
madze Akad. Nauk Gruzin. SSR, (Collection of articles on the boundary problems of
analytic functions, singular integral equations and some questions of harmonic analy-
sis), vol. 65, 1980, pp. 99-109. Russian
7. G. S. Litvinchuk, Boundary value problems and singular integral equations with a
shift, "Nauka", Moscow, 1977. (Russian)
8. ___ , On operator approach to boundary value problems with shift and conjuga-
tion, Memoirs of the anniversary seminar on boundary value problems (F. D. Gahov
anniversary VoL), Belorussian University, Minsk, 1985, pp. 69-76. (Russian)
9. N. I. Lisovets, A four-element boundary value problem with piecewise continuous co-
efficients with shift on a composite contour, Ukrain. Mat. Zh. 35 (1983), 764-767.
(Russian)
10. ___ , A boundary value problem for functions analytic in several multiply connected
domains, Izv. Vyssh. Uchebn. Zaved. Mat. 4 (1984), 75-77. (Russian)
11. ___ , Differential boundaT!/ value problem for functions analytic in several multiply
connected domains, Manuscript no. 546-Uk-D84, UkrNIINTI, Kiev, 1984. (Russian)
12. _ _ _ , A method for investigating boundary value problems for function. that are
analytic in a domain, Soobshch. Akad. Nauk Gruzin. SSR 122 (1986), no. 3, 477-480.
(Russian)
13. _ _ _ , A method of inve.tigating differential boundary value problem. for analytic
function., Differentsial'nye Uravneniya 24 (1988), no. 6, 1049-1053. (Russian)
14. A. V. Protsenko, Differential boundary value problem. for analytic function., Ph.D.
thesis, Odessa Univ., USSR, 1981. (Russian)
15. S. F. Skorohod, Noether theory of multielement boundary value problem. with a .hift
for function. analytic in a domain, Ph.D. thesis, Odessa Univ., USSR, 1984. (Russian)
16. S. L. Sobolev, Application. of functional analysis to mathematicap physics, Novosi-
birsk, 1962. (Russian)
17. I. Spitkovsky, Singular integral operators with PC symbol. on weighted Hardy spaces,
Journal of Functional Analysis 104 (1992).
18. N. E. Tovmasjan, Integral representation. of holomorphic functions by holomorphic
densities and their applications, Lecture Notes in Math., Analytic functions, Kozubnik
1979 (Proc. Seventh Conf., Kozubnik, 1979), vol. 798, Springer-Verlag, Berlin and New
York, 1980, pp. 439-445.
19. N. P. Vekua, System. of singular integral equations and boundary value problems,
"Nauka", Moscow, 1970. (Russian)
DEPARTMENT OF MATHEMATICS, ROLLINS COLLEGE, WINTER PARK, FLORIDA 32789

E-mail address: dkurtz@rollins.bitnet
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF MISSOURI, COLUMBIA, MISSOURI

65211
E-mail address: mathyl@mizzou1.missouri.edu
DEPARTMENT OF MATHEMATICS, COLLEGE OF WILLIAM AND MARY, WILLIAMSBURG,

VIRGINIA 23181
E-mail address: ilya@cs.wm.edu

Algebraic Methods in Operator Theory

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Algebraic Methods in Operator Theory

Enviado por

Direitos autorais:

Formatos disponíveis

Algebraic Methods

Springer Science+Business Media, LLC

Library of Congress Cataloging-in-PubUcation Data

Algebraic methods in operator theory I Raul E. Curto, Palle E. T.

Printed on acid-free paper

Permission to photocopy for internal or personal use of specific clients is granted by

Camera-ready copy by the Authors.

I. Single Operators and Applications

On the commutant lifting theorem and Hankel operators

II. Nonselfadjoint Algebras

Elementary operators and subalgebras

Normal elements of a simple C·-algebra

IV. von Neumann Algebras and Subfactors

On the structure of finite depth subfactors

V. Representations of Groups and Algebras on Hilbert Space

Landstad duality for coactions on C· -algebras

VI. Geometry and Topology

"Vector bundles" over quantum Heisenberg manifolds

The theory of operators stands at the intersection of the frontiers of modern

Raul Curto and Paile E. T. J¢rgensen

bounded measurable function e : T -+ 8(V, V') such that e(z) is an isometry

is clearly a positive operator in 8(1t'ffi/C). Using X to define a new inner product

Let us define T: 1t'VU/c-+T'*1t'V/C by

(2) IIT'*h' + kll~ = IIT'*h'II~, + 2Re < Bk,T'*h' >1t' +lIkllk

We ass~rt that U' in 8(i') is a minimal isometric dilation of T', AU

<Ak,h' >i'=< k,h' >i. =< k,h' >it=< Bk,h' >1(',

Finally, to show that AU

< 1Uk,h' >i;. =< Uk,h' >i;. =< Uk,h' >it

is invertible, so any vector h in if. can be uniquely expressed as h = h' + h for

Hence T can be extended to an isometry (denoted again by T) acting on the

(4) ite1t' = { - Hh + h; he 1t},

so there exists h in 1t such that

For all h' in 1t',

=< Th + DT.hbh' >it'

is invertible, so any vector h in ii can be uniquely expressed as h = h' + h for

(1-£'VT1-l)J.. = { - H Di/d + Di/d; dE V} ,

(4) iie1-£' = { - Hh + h; h E 1-£ },

Since D'jo is the orthogonal projection onto (TH.)J., 'D'jo C (T'*1{,'V1{,)J., 80

By the way we extended T to all of ii, T«'H.'VTH)J.) c (T'*1{,'V1t)J., 80 there

(9) T( -HDii2d + Dii2d) = Dii~ (zdD - H* Dii~ (z~) .

Putting together (1) and (6)-(9) we get

= T(h' + T~) + T(-HDii2d + Dii2d) + (Dii~(zd') - H* Dii~(zd'»

Hhl = Hh2 - HH* Dii~(%(d' +~» = -T'*h' - Dii~(%(d' + dD>,

11Th + Dt.h1112 = IIhll2+ II Dt. h111 2= 11- Hh1 +h111~

IIhll2+ IIDt.h1112 = IIDii1rD~h1l2 .

IIhll2+ II E U!"'-1)(U. - r)h",11 2

= IIhll2+ E IIDt. h",1I 2

JOHN FROELICH AND MICHAEL MARSALLI

Let B(H) be the algebra of bounded operators on a complex, separable

2. METRIC SPACES OF INVARIANT SETS

Let I(I:) denote the collection of norm closed convex subsets of HI

b) (I(E),d",) is a closed subspace of (e,d",).

3. TRANSITIVE OPERATOR SEMIGROUPS

In order to study ergodic properties of operator semigroups, we make

Theorem 2. Let E be a weakly closed operator semigroup on an infinite

Note that E is I-transitive if and only if J(E) = {rHl : 0 ~ r ~ I},

'fransitive Semigroup Problem. Is (B(H)h the only weakly closed

The answer is no. In fact, for each n there is a non-trivial weakly

Theorem 3. If E is a commutative operator semigroup, and dimH ~ 2,

We now construct another example of a I-transitive operator semigroup

Theorem 4. If L. is a weakly closed, i-transitive operator semigroup which

In particular, when H is finite dimensional, TI is contained in every 1-

Question 1. If L. is a i-transitive weakly closed operator semigroup, 's

Theorem 5. A set J is a minimal invariant set for L. if and only if there

(2) IIT'h' + kll~ = IIT'h'II~, + 2Re < Bk,T'*h' >1t' +lIkllk

( ) < b(T , T)h j,m, h i,n >:= « hj,m,k. hi,n

b(T, T) = L 91(T) fl(T).

b(T, T) =I - L hs(T) hs(T)

b(T, T) ~ I - P L hs(J) hs( J)P