Sei sulla pagina 1di 329

CONTEMPORARY

MATHEMATICS
503

"i>-Vi>`
>V>-i
qx]n
i
ii]i`i]/i ii>`
->ii
viiVivi
v i>
}iv>i>V

>Vi`ii
-i}i-i

>->
/>>
`

American Mathematical Society

Operator Structures and


Dynamical Systems

This page intentionally left blank

CONTEMPORARY
MATHEMATICS
503

Operator Structures and


Dynamical Systems
July 2125, 2008
Lorentz Center, Leiden, The Netherlands
Satellite Conference of the
Fifth European Congress of Mathematics

Marcel de Jeu
Sergei Silvestrov
Christian Skau
Jun Tomiyama
Editors

American Mathematical Society


Providence, Rhode Island

Editorial Board
Dennis DeTurck, managing editor
George Andrews

Abel Klein

Martin J. Strauss

2000 Mathematics Subject Classication. Primary 46L55, 37Bxx, 47Lxx;


Secondary 46L08, 46L35, 46H25, 37B10, 37Fxx, 16S35, 54H20.

Library of Congress Cataloging-in-Publication Data


European Congress of Mathematics (5th : 2008 : Leiden, The Netherlands)
Operator structures and dynamical systems : Satellite conference of the fth European Congress of Mathematics, July 2125, 2008, Lorentz Center, Leiden, The Netherlands / Marcel de
Jeu . . . [et al.], editors.
p. cm. (Contemporary mathematics, ISSN 0271-4132 ; v. 503)
Includes bibliographical references.
ISBN 978-0-8218-4747-3 (alk. paper)
1. Operator theoryCongresses. 2. DynamicsCongresses. I. Jeu, Marcel de, 1962
II. Title.
QA329.E97 2008
515.724dc22
2009027923

Copying and reprinting. Material in this book may be reproduced by any means for educational and scientic purposes without fee or permission with the exception of reproduction by
services that collect fees for delivery of documents and provided that the customary acknowledgment of the source is given. This consent does not extend to other kinds of copying for general
distribution, for advertising or promotional purposes, or for resale. Requests for permission for
commercial use of material should be addressed to the Acquisitions Department, American Mathematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can
also be made by e-mail to reprint-permission@ams.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the rst page of
each article.)
c 2009 by the American Mathematical Society. All rights reserved.

The American Mathematical Society retains all rights
except those granted to the United States Government.
Copyright of individual articles may revert to the public domain 28 years
after publication. Contact the AMS for copyright status of individual articles.
Printed in the United States of America.
The paper used in this book is acid-free and falls within the guidelines


established to ensure permanence and durability.


Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1

14 13 12 11 10 09

Contents
Preface

vii

List of Participants

ix

Ane transformation crossed product type algebras and noncommutative


surfaces
Joakim Arnlind and Sergei Silvestrov

C -algebras associated with iterated function systems


Gilles G. de Castro

27

Nonself-adjoint operator algebras for dynamical systems


Kenneth R. Davidson and Elias G. Katsoulis

39

Extending representations of normed algebras in Banach spaces


Sjoerd Dirksen, Marcel de Jeu, and Marten Wortel

53

Countable bases for Hilbert C -modules and classication of KMS states


Tsuyoshi Kajiwara

73

Subshifts and C -algebras from one-counter codes


Wolfgang Krieger and Kengo Matsumoto

93

Orbit equivalence in C -algebras dened by actions of symbolic dynamical


systems
Kengo Matsumoto

121

Normalisers, nest algebras and tensor products


M. McGarvey and I.G. Todorov

141

Noncommutative geometry as a functor


Igor V. Nikolaev

151

Simple group graded rings and maximal commutativity

Johan Oinert

159

The Rohlin property for inclusions of C -algebras with a nite


Watatani index
Hiroyuki Osaka, Kazunori Kodaka, and Tamotsu Teruya

177

The C -envelope of a semicrossed product and nest representations


Justin R. Peters

197

Freeness of actions of nite groups on C -algebras


N. Christopher Phillips

217

vi

CONTENTS

Examples of masas in C -algebras


Jean Renault

259

A denition of compact C -quantum groupoids


Thomas Timmermann

267

Complex dynamical systems and associated C -algebras


Yasuo Watatani

291

On classifying monotone complete algebras of operators


J.D. Maitland Wright

307

Preface
Many dynamical systems are naturally accompanied by an associated operator
structure of some kind. A well known instance occurs when a topological group acts
on a compact Hausdor space, where one can form the associated crossed product
C -algebra. This algebra can be viewed as the counterpart of the crossed product
von Neumann algebra associated with a group of measurable transformations, and,
just as the crossed product von Neumann algebra carries information about the
measurable system, the crossed product C -algebra carries information about the
topological dynamical system. The rst attempt at a systematic investigation of
this interplay between topological dynamics and C -algebras was in a 1967 AMS
Memoir by Eros and Hahn, the dating of which, incidentally, demonstrates that
crossed product C -algebras in the topological context are younger than crossed
product von Neumann algebras in the measurable context, which go back to the
much earlier seminal papers by Murray and von Neumann.
The above manner of constructing a crossed product C -algebra in a topological
context is only a special case of a more general procedure. In a noncommutative
setting, where a group acts on a possibly noncommutative C -algebra, it is likewise
possible to construct a crossed product C -algebra. The general theory of such
algebras has developed considerably since its introduction in the second half of
the 1960s, with wide ramications in the general theory of C -algebras and of
unitary group representations. This more general theory applies to the interplay
between topological dynamical systems and crossed product C -algebras mentioned
above, but by its general nature it is sometimes hard to obtain the more detailed
statements, which can be given in the commutative setting, as corollaries from the
general noncommutative context.
From July 2125, 2008, a one week meeting at the Lorentz Center in Leiden was
organized by the undersigned editors, together with Klaas Landsman (Nijmegen)
and Sjoerd Verduyn Lunel (Leiden), with the aim of bringing together senior and
junior researchers from both elds of primary interest, i.e., people working in the
general theory of C -crossed products as well as people with an emphasis on the
more special topological dynamical context. As a result, this workshop Operator
Structures and Dynamical Systems, which was a satellite of the Fifth European
Congress of Mathematics taking place in the preceding week in Amsterdam, saw a
broad spectrum of 54 participants from 16 countries and 5 continents. All in all, 28
lectures were delivered, inspired by the central theme, and covering recent research
results as well as expository material. From the papers in these proceedings the
reader will have a fair impression of what this week was like, and it is a pleasure
to thank the AMS and its sta, and Christine Thivierge in particular, for making
their publication possible.
vii

viii

PREFACE

The editors and their co-organizers are particularly grateful to the Lorentz
Center, which supplied a considerable part of the budget, the facilities and, last
but not least, the remarkable organizational skills of Gerda Filippo, backup up by
Martje Kruk. The remaining substantial part of the budget was provided jointly by
the Royal Dutch Academy of Arts and Sciences, the Dutch national research clusters
Geometry and Quantum Theory and Nonlinear Dynamics of Natural Systems,
the Dutch research schools Mathematical Research Institute and Thomas Steiltjes Institute for Mathematics, and the Netherlands Organisation for Scientic
Research. Their support, and the support by STINT and the Crafoord Foundation
on behalf of the Swedish participants, is gratefully acknowledged.
Marcel de Jeu
Sergei Silvestrov
Christian Skau
Jun Tomiyama
June 16, 2009

List of Participants
Astrid An Huef
Sydney, Australia

Patricia Hess
S
ao Paulo, Brazil

Robert Archbold
Aberdeen, United Kingdom

Marcel de Jeu
Leiden, The Netherlands

Ola Bratteli
Oslo, Norway

Tsuyoshi Kajiwara
Okayama, Japan

Berndt Brenken
Calgary, Canada

Evgenios Kakariadis
Athens, Greece

Julian Buck
Eugene, United States

Aristides Katavolos
Athens, Greece

Toke Carlsen
Odense, Denmark

Takeshi Katsura
Yokohama, Japan

Gilles de Castro
Orleans, France

Shinzo Kawamura
Yamagata, Japan

Erik Christensen
Copenhagen, Denmark

Eberhard Kirchberg
Berlin, Germany

Kenneth Davidson
Waterloo, Canada

Edward Kissin
London, United Kingdom

Sjoerd Dirksen
Delft, The Netherlands

Erik Koelink
Nijmegen, The Netherlands

Siegfried Echterho
M
unster, Germany

Wolfgang Krieger
Heidelberg, Germany

Gwion Evans
Aberystwyth, United Kingdom

Klaas Landsman
Nijmegen, The Netherlands

Ruy Exel
Florianopolis, Brazil

Aldo Lazar
Tel Aviv, Israel

Thierry Giordano
Ottawa, Canada

Kengo Matsumoto
Yokohama, Japan

Magnus Goeng
Goteborg, Sweden

Konstantin Medynets
Kharkov, Ukraine
ix

PARTICIPANTS

Igor Nikolaev
Toronto, Canada

Yasuo Watatani
Fukuoka, Japan

Johan Oinert
Lund, Sweden

Dana Williams
Hanover, United States

Hiroyuki Osaka
Kusatsu, Japan

Daniel Worm
Leiden, The Netherlands

Ben de Pagter
Delft, The Netherlands

Marten Wortel
Leiden, The Netherlands

Justin Peters
Ames, United States

John Maitland Wright


Aberdeen, United Kingdom

Christopher Phillips
Eugene, United States
Maarten van Pruijssen
Nijmegen, The Netherlands
Ian Putnam
Victoria, Canada
Iain Raeburn
Wollongong, Australia
Jean Renault
Orleans, France
Sergei Silvestrov
Lund, Sweden
Christian Skau
Trondheim, Norway
Fedor Sukochev
Adelaide, Australia
Christian Svensson
Leiden, The Netherlands & Lund,
Sweden
Thomas Timmermann
M
unster, Germany
Ivan Todorov
Belfast, United Kingdom
Reiji Tomatsu
Leuven, Belgium
Jun Tomiyama
Tokyo, Japan
Sjoerd Verduyn Lunel
Leiden, The Netherlands

Contemporary Mathematics
Volume 503, 2009

Ane transformation crossed product type


algebras and noncommutative surfaces
Joakim Arnlind and Sergei Silvestrov

Abstract. Several classes of -algebras associated to the action of an ane


transformation are considered, and an investigation of the interplay between
the dierent classes of algebras is initiated. Connections are established that
relate representations of -algebras, geometry of algebraic surfaces, dynamics
of ane transformations, graphs and algebras coming from a quantization
procedure of Poisson structures. In particular, algebras related to surfaces
being inverse images of fourth order polynomials (in R3 ) are studied in detail,
and a close link between representation theory and geometric properties is
established for compact as well as non-compact surfaces.

1. Introduction
The interplay between representation theory of -algebras and dynamical systems or
more general actions of groups or semi-groups is an expanding area of investigation
deeply intertwined with origins of quantum mechanics, foundations of invariants
and number theory, symmetry analysis, symplectic geometry, dynamical systems
and ergodic theory and several other parts of mathematics that are fundamental
for modern physics and engineering. Over the past sixty yeas investigation of this
broad interplay have been developed to some extent independently within several
interlacing contexts. On the one hand in the area of operator algebras, the crossed
product type C -algebras and W -algebras have been associated with actions of
groups, semi-groups or dynamical systems. Properties and classications of dynamics or actions are linked to the properties or classications of the corresponding
C -algebras and W -algebras and their operator representations. Historically this
general approach can be viewed as an extension of the theory of induced representations of nite and compact groups on the one hand and as a general abstract
framework for foundations of quantum mechanics and quantum eld theory on the
other hand [E65, E81, E82, Gli61b, Gli61a, Jor88, JSW95, Mac68,
Mac76, Mac89]. In this context C -algebras, W -algebras and more general Banach algebras are often considered without making specic choices of generators
of the algebras. This can be viewed as a kind of non-commutative coordinateindependent approach to simultaneous treatment of actions and spaces on the
2000 Mathematics Subject Classication. Primary 16S35, 16G99.
Key words and phrases. representations, algebras, surfaces, dynamical systems, orbits.

JOAKIM ARNLIND AND SERGEI SILVESTROV

same level within the same general framework [AS94, BR79, BR81, Dav96,
E65, E81, E82, Gli61b, Gli61a, KTW85, Li92, Ped79, Sak71, ST02,
SSdJ07a, SSdJ09, SSdJ07b, ST08, ST09, Tak79, Tom87, Tom92]). Crossed
product type algebras and related graded structures have been also considered in
connection to the properties of the actions in pure algebra context in ring theory
in the ways similar to crossed product C -algebras and W -algebras, but typically
not taking into consideration norm or metric structures, considering to a greater
extent representations which are not necessarily -representations and working with
algebras and rings over various kinds of elds, not necessarily real or complex num

bers [Kar87, NVO04, NVO08, Pas89, OS08c,


OS08a,
OSTAV08,
OS08b,

OS09].
A more straightforward approach is based on specic choices of generators for the algebras and more direct analysis of operators representing the generators. This constructive non-commutative coordinates approach is often used
in physics and engineering models. The choice of generators can be viewed as a
choice of non-commutative coordinates. The convenient choice of the generators
(non-commutative coordinates) as in any coordinate approach is a key to success
of further analysis. Choices of generators inuence the form and complexity of the
commutation rules, with the best choice of generators being often that which makes
dynamics or actions appear explicitly in the commutation relations. Such choices of
generators often indicate that the algebra might possibly be viewed as some kind of
crossed product of a subalgebra by an action or perhaps a quotient of such crossed
product. Properties and structure of the dynamics are reected in the properties
and structure of the operator representations of the associated to the dynamics
commutation relations [OS89, OS99, Sam91, VS88, Sil95, SW96, VS90].
The algebras we will consider in this article are naturally dened by generators
and relations of a certain type closely linked to the action of general ane transformations in two dimensions (see Denition 2.1). We establish close connection
between these algebras and algebras that arise from a quantization procedure of
Poisson brackets associated to a general class of algebraic surfaces (see Denition
2.2, Proposition 3.3 and Section 4). We will mostly work in this article with nitedimensional representations, and also describe some classes of innite-dimensional
representations. The algebras we consider are closely related to crossed product
algebras of the algebra of functions in two commuting variables by the action of an
additive group of integers or a semi-group of non-negative integers via composition
of a function with powers of the ane transformation applied to the two-dimensional
vector of variables (see equation (2.7) and Proposition 3.1). Therefore, there exists
a strong interplay between representations and especially -representations of these
commutation relations and dynamics of the ane transformation (see Sections 5
and 6). Especially the orbits play important role for all nite-dimensional representations and also for some classes of innite-dimensional representations as these
representations can be described explicitly in terms of orbits or parts of orbits. Another way of expressing this and the structure of representations is using graphs. In
this paper, representations of the algebras connected to ane transformation and
their structure is studied using both the orbits and the graphs of iterations of the
ane transformation.
One of our main goals in this paper is to establish and investigate the interplay of representations of these parametric families of commutation relations and
algebras with the geometry of the corresponding parametric families of algebraic

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

surfaces. In Sections 4 and 7 we investigate what happens with representations


when a change in the parameters results in a change of properties of the corresponding surface; e.g. from compact to non-compact, from genus 0 to genus 1,
changes in the number connected components etc. These and various other aspects
of the interplay between geometry and representations are studied in detail.
2. Two algebras related to an ane map
be a free associative algebra on four letters
Definition 2.1. Let CS, T, E, E
over the complex numbers. Let I be the two-sided ideal generated by the relations
(2.1)
(2.2)
(2.3)
(2.4)
(2.5)

+ uS SE = 0
ES + ES
+ vS S E
=0
ES + ES
+ uT ET = 0
T E + T E
+ vT ET
=0
T E + T E
EE
EE
= 0,
[E, E]

where , , , , u, v R. We dene AL to be the quotient algebra CS, T, E, E/I.

We can also consider AL to be a -algebra by dening S = T , T = S, E = E


= E,
since the set of relations (2.1)(2.5) is invariant under this operation.
and E
Let us dene an ane map L : R2 R2 by
   

u
x

,
+
(2.6)
L(x) = Ax + u
v
y

with , , , , u, v R. Note that the dening relations (2.1)(2.5) of the algebra
AL can be written in the following form when using block matrix notation







E
E
S 0
S
=L

0 S
E
E
 





E
E 
T 0
(2.7)
T
L
=

0 T
E
E
EE
= 0.
EE
This way of writing the relations indicates a close connection of the algebra AL to
crossed product type algebras and hence interplay with dynamics of iterations of
the algebra AL (see Proposition 3.1).
Let us now dene a second algebra related to the ane map L.
Definition 2.2. Let C W, V  be a free associative algebra on two letters over
the complex numbers, and let L be the ane map on R2 dened by Lx = Ax + u.
For any a R, let Ia be the two-sided ideal generated by the relations
(2.8)

W 2 V = aW (det A) V W 2 + (tr A) W V W

(2.9)

W V 2 = aV (det A) V 2 W + (tr A) V W V.

We then dene CL,a to be the quotient algebra C W, V  /Ia . We can also consider
CL,a to be a -algebra by dening W = V and V = W , since the set of relations
(2.8)(2.9) is invariant under this operation.

JOAKIM ARNLIND AND SERGEI SILVESTROV

In order to relate these algebras, we want to construct a homomorphism from


AL to CL , by setting
(S) = W ;

(T ) = V

(E) = k1 + mW V + nV W
+ mW
= k1
(E)
V +n
V W.
To obtain a homomorphism, we must require that elements that are equivalent to
0 in AL are mapped to elements equivalent to 0 in CL,a . This requirement gives
rise to the following system of equations


 


tr A det A
m n
m n

=0

1
0
m
n

m
n




   
  
k

u
a
m n
12
.

=

v
0
m
n

k
General solutions to this system of equations are given in Appendix A, but whenever
a = 0, a particularly simple solution is given by



1
(E) =
uW V + v u V W
a




= 1 vW V + u v V W .
(E)
a

The fact that ([E, E]) = 0 is guaranteed by the following proposition.


Proposition 2.3. In CL,a it holds that [W V, V W ] = 0.
Proof. Multiplying (2.8) by V from the left and (2.9) by W from the right
gives V W 2 V = W V 2 W , i.e. [W V, V W ] = 0.

The map will in general not be an isomorphism since, e.g., the element E k1
mST nT S (which is non-zero in AL ) is mapped to 0 in CL,a .
3. The center of AL and CL,a
denote the subalgebra of AL generated by E nd E,
and let P[D, D]

Let P[E, E]
= V W . In this section
denote the subalgebra of CL,a generated by D = W V and D
we will gather a couple of results that concern central elements in AL and CL,a .
it holds that
Proposition 3.1. For any p P[E, E]


= p En , E
n S n
S n p(E, E)


n = T n p En , E
n
p(E, E)T
n ) = Ln (E, E).

where L(x, y) = (x + y + u, x + y + v) and (En , E


it holds that
Proposition 3.2. For any p P[D, D]


= p Dn , D
n W n
W n p(D, D)


n = V n p Dn , D
n
p(D, D)V


y) = (tr A)x (det A)y + a, x and (Dn , D
n) = L
n (D, D).

where L(x,
From
these
p(x, y) =

 propositions it is clear that any polynomial
 p, satisfying

y) and any polynomial q, satisfying q(x, y) = q L(x, y) generate central
p L(x,
elements of CL,a and AL respectively. In particular, we have the following result

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

Proposition 3.3. Let Cr,s,t denote the following element in CL,a :








+s D+D
2+t DD
2.
Cr,s,t = r D + D
Then Cr,s,t commutes with W and V if and only if we are in one of the following
two situations:
(1) det A = 1, which implies that






+ (2 tr A) D + D
2 + (2 + tr A) D D
2
C = 4a D + D
(3.1)
commutes with W and V ;
(2) det A = 1, tr A = 0 and a = 0, in which case Cr,s,t commutes with W
and V for all r, s, t R.
4. Relation to noncommutative surfaces
In [ABH+ 09, Arn08b] noncommutative C-algebras of Riemann surfaces were
constructed and a particular case of spheres and tori was studied in detail. It turns
out that the classical transition from spherical to toroidal geometry corresponds
to a change in the representation theory of the noncommutative algebras. This
correspondence will later be described in detail. Let us briey recall how to obtain
algebras from a given surface.
Let C(x, y, z) be a polynomial and let = C 1 (0). One can dene a Poisson
bracket on R3 by setting


(4.1)
{f, g} = C f g ,
for smooth functions f, g. This Poisson bracket induces a Poisson bracket on by
restriction. The idea is to start from the coordinate relations
(4.2)

{x, y} = z C

(4.3)

{y, z} = x C

(4.4)

{z, x} = y C

and then construct a noncommutative algebra on X, Y, Z by imposing the relations




[X, Y ] = i z C
(4.5)


[Y, Z] = i x C
(4.6)


[Z, X] = i y C
(4.7)
where is an ordering map from polynomials in three variables to noncommutative
polynomials in X, Y and Z. In case this algebra is non-trivial, its representations
will provide an approximating sequence (in the sense of [BHSS91]) for the Poisson
algebra of polynomial functions on as  0 (see [Arn08a] for details).
Let us now consider the following polynomial
 1  2
2 1
0  2
1
C(x, y, z) =
(4.8)
x + y2 +
x + y 2 + z 2 c0 ,
2
4
2
2
which, by using the above Poisson bracket, gives rise to
(4.9)

{x, y} = z

(4.10)

{y, z} = 0 x + 1 x(x2 + y 2 )

(4.11)

{z, x} = 0 y + 1 y(x2 + y 2 ).

JOAKIM ARNLIND AND SERGEI SILVESTROV

We will choose an ordering of the right hand sides in terms of the complexied
variables W = X + iY and V = X iY (cp. [Arn08b]) 1
[X, Y ] = iZ








i  2
2
2
2

[Y, Z] = i0 X +
1 V W + V W + 1 V W V + W V W + 1 W V + W V
2




 2


i 
2
2
2

1 V W V W + 1 W V W V W V + 1 W V W V
[Z, X] = i0 Y +
2i
for any choice of 1 , 1 , 1 such that 1 +
1 +1 = 1 . By eliminating Z = [X, Y ]/i,
one can write the second two relations entirely in terms of W and V






1 + 22 1 W 2 V = 20 2 W 22 1 + 1 V W 2 + 2 22 1 W V W






1 + 22 1 W V 2 = 20 2 V 22 1 + 1 V 2 W + 2 22 1 V W V.
This algebra is isomorphic to CL,a if
a=

20 2
1 + 22 1

and L is an ane map such that


2 22 1
1 + 22 1
tr A =
.
1 + 22 1
1 + 22 1
Hence, the relation to the original parameters of the polynomial is
1 + 22 1
1 + 22 1
0 = a

1
22
22
where = 1 + det A tr A.
dened in (3.1) when det A = 1, by writing it in
Let us study the Casimir C,


= 2 X 2 + Y 2 and D D
= 2Z, we obtain
terms of X, Y and Z. Since D + D




2


C = 8a X 2 + Y 2 + 4 2 tr A X 2 + Y 2 + 42 2 + tr A Z 2 .
det A =

When the algebra CL,a arises from a surface, we can express tr A in terms of
0 , 1 , 1 , 1 , 1 to obtain

 2
 1  2

1 + 22 1
1 2
2
2
2 2
2

C
=

X
 1 Z .
X
+
+
Y
+
Y
+
1
+
2

0
1
162
2
2
In this way we see that the Casimir C is a noncommutative analogue of the embed
ding polynomial C(x, y, z). In any irreducible representation , the element (C)
will be proportional to the identity. Let us dene two constants c0 and c1 through
the following relations:
1 + 22 1
= 4
(C)
c1 1
and
c0 =
c1 .
42
In the procedure of constructing noncommutative algebras from a given polynomial,
information about the constant c0 is lost since the construction only depends on
partial derivatives of C(x, y, z). As we will see, dierent values of c0 correspond
1Note that in [Arn08b], the parameter is not present (although it is implicitly present in
1
[ABH+ 09], taking the value 1/2). This is an additional freedom in the choice of ordering that
can not be extended to higher order algebras without breaking the commutativity of W V and
V W.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

to, for instance, dierent topologies of the surface, and this rises a problem if we
want to study geometry in the algebraic setting. However, since (when det A = 1)
the central element C is a noncommutative analogue of the polynomial C(x, y, z),
we will identify c0 and c0 in an irreducible representation; this gives us a way to
determine the topology of a representation.
In the following we will compare the geometry of the surface, for all values of
0 , 1 , c0 , with the representation theory for the corresponding irreducible representations of CL,a when c0 = c0 .
5. -representations of CL,a
From the viewpoint of noncommutative surfaces, one is interested in representations
in which X, Y and Z are self-adjoint operators. This requirement is transferred to
CL,a by considering -representations. By a -representation we mean a representation such that (A ) = (A) . Clearly, writing W = X + iY and V = X iY ,
for hermitian matrices X, Y , implies that W = V .
The (-)representation theory of CL,a was worked out in [Arn08b], but let us
recall some details in the construction. Let us for simplicity denote (W ) by W and
(V ) by V in a nite dimensional -representation of CL,a . By Proposition 2.3 the
= W W will be two commuting hermitian matrices,
matrices D = W W and D
and therefore they can always be simultaneously diagonalized by a unitary matrix.
Let us assume such a basis to be chosen and write
D = diag(d1 , d2 , . . . , dn )
= diag(d1 , d2 , . . . , dn ).
D
In components, the dening relations of CL,a (together with the associativity con can then be written as
dition DW = W D)





Wij tr A di det A di + a dj = 0


Wij di dj = 0,
Thus, either Wij = 0 or





dj = tr A di det A di + a
dj = di .

By introducing the notation xi = (di , di ) and the ane map L


   
  
a
x
x = tr A det A
,
+
L
0
y
1
0
y
xi ) whenever Wij = 0. Let us now show how
we can write this relation as xj = L(

the representation theory can be described as a dynamical system generated by L


acting on a directed graph.
Let GW = (V, E) be the directed graph of W , i.e. the graph on n vertices with
vertex set V = {1, 2, . . . , n} and edge set E V V , such that
(i, j) E

Wij = 0.

By assigning the vector xi to the vertex i, it follows that for a graph corresponding
xi ) whenever
to the matrix W in a representation of CL,a , it holds that xj = L(

JOAKIM ARNLIND AND SERGEI SILVESTROV

there is an edge from i to j. The dynamical system on the graph can therefore be
depicted as in Figure 5.
2 (x)
L
x

x)
L(
3 (x)
L
2 (x)
L
3 (x)
L

acting on the directed graph of a representation.


Figure 1. The ane map L
One immediate observation is that if the graph has a loop (i.e. a directed cycle)
on k vertices, then the ane map must have a periodic orbit of order k. If the
ane map does not have any periodic points, then loops are excluded from all
representation graphs. It is a trivial fact that any nite directed graph must have
at least one loop or at least one string, i.e. a directed path from a transmitter to
a receiver. Hence, if the graph can not have a loop, it must contain a string. Due
= W W , one gets the following condition for
to the fact that D = W W and D
vertices being transmitters or receivers.
Lemma 5.1 ([ABH+ 09]). The vertex i is a transmitter if and only if di = 0.
The vertex i is a receiver if and only if di = 0.
Thus, for a string on k vertices to exist, there must exist a vector x = (d, 0) such
We call this a k-string of the ane map L.
k1 (x) = (0, d).
We also note
that L

that since the matrices D and D are non-negative, all vectors {xi } must lie in
R20 = {(x, y) R2 : x, y 0}. The natural question is now: Which graphs
correspond to irreducible representations of CL,a ? The answer lies in the following
theorem.
Theorem 5.2 ([Arn08b]). Let be a locally injective -representation of CL,a .
Then is unitarily equivalent to a representation in which (W V ) and (V W ) are
diagonal and the directed graph of (W ) is a direct sum of strings and loops. A
representation corresponding to a single string or a single loop is irreducible.
is injective on the set
Remark 5.3. A representation is locally injective if L
{x1 , x2 , . . . , xn }. A representation whose graph is connected and contains a loop
is invertible, then any repwill always be locally injective [Arn08b]. Clearly, if L
resentation is locally injective.
Furthermore, one can show that every k-string in R20 and every periodic orbit in
R2>0 = {(x, y) R2 : x, y > 0} induce an irreducible representation of CL,a ; distinct
orbits/k-strings induce inequivalent representations. In this way, the representation
theory of CL,a is completely determined by the dynamical properties of the ane

map L.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

n (x1 ) = x1 and xk = L


k1 (x1 ) = (dk , dk ) R2>0
For instance, assume that L
for k = 1, . . . , n 1. Then an n-dimensional -representation of CL,a is constructed
by setting

d1 0

0
0
0
d2
0
0

..
..
..
.
.
.
.
(5.1)
(W ) = .

.
.
.

 .
0
dn1
0

i
e
dn
0

0
0
for any R.
5.1. Innite dimensional representations. There are two classes of innite dimensional representations of CL,a that can be easily constructed. They come
in the form of innite dimensional matrices with a nite number of non-zero elements in each row and column. This assures that the usual matrix multiplication
is still well-dened.
The rst type is one-sided innite dimensional representations. In this case
the basis of the vector space is labeled by the natural numbers. The second type
is two-sided innite dimensional representations; and the basis vectors are labeled
by the integers.
A one-sided representation of CL,a can be constructed by choosing x0 = (d0 , 0)
k (x0 ) R2>0 for k = 1, 2, . . .. A one-sided
(with d0 > 0) such that xk = (dk , dk ) = L
representation is then obtained byletting (W ) be an innite dimensional matrix
with non-zero elements Wk,k+1 = dk for k = 0, 1, . . ..
to be invertible, two-sided representations can be constructed
If we assume L
k (x0 ) R2 for k Z. We then
by choosing x0 R2>0 such that xk = (dk , dk ) = L
>0

set the non-zero elements of (W ) to be Wk,k+1 = dk for k Z.


5.2. Representations when det A = 1. Let us now turn to the question
concerning when dierent kinds of representations can exist, if we x a specic
Thus, we assume that det A = 1 and that the
value of the central element C.
irreducible representation is such that
+ (2 tr A)(D + D)
2 + (2 + tr A)(D D)
2 = 4
4a(D + D)
(5.2)
c1 1.
are diagonal, this constrains the vectors xi = (di , di ) to lie in the
Since D and D
set dened by all (r, s) R2 such that
(5.3)

p(r, s) 4a(r + s) + (2 tr A)(r + s)2 + (2 + tr A)(r s)2 4


c1 = 0

We dene = p1 (0) and call this set the constraint curve of an irreducible representation. When tr A = 2, we can write the constraint curve in the following
convenient form


2 + tr A
2
2
(r s) 4
p(r, s) = (r + s 2) +
c

where = 2 tr A, = a/ and c = 2 + c1 /. By Proposition 3.2, is


Moreover, if has several disjoint
invariant under the action of the ane map L.

components, L leaves each of them invariant.

In the case when consists of one or two disjoint curves, one can check that L
will preserve the direction along the curves; i.e. we can parametrize each curve by

10

JOAKIM ARNLIND AND SERGEI SILVESTROV

and denote points on the curve by x(), such that if we dene 1 and 2 through
x(1 ) = L(x(1 )) and x(2 ) = L(x(2 )) then 1 1 , and 1 2 implies that
1 2 .
Let us now prove a few results leading to Proposition 5.8, that tells us when
there are no non-trivial (i.e. of dimension greater than one) nite dimensional
representations.
denote the ane map dened by L(x,
y) = (x tr A
Proposition 5.4. Let L
2
y det A + a, x) and assume that (tr A) = 4 det A and tr A = 1 + det A. Then it
holds that
  
 n
 
 n+1
1
x
q
+ n 
+ n+1
n x


=
L
n1
y
y
q n1

n+ n
+
+

 1n
n 
1
+ 1+
1
a
+

+
.
1n
1n
+

1+
1


where = tr A (tr A)2 4 det A /2.
has no periodic points
Lemma 5.5. Assume that det A = 1 and 0. Then L
other than x-points.
Proof. When tr A = 2 and det A = 1, a direct computation shows that there
are only periodic points when a = 0, and these points are x-points.
When tr A > 2 and det A = 1, the eigenvalues of the matrix


tr A 1
M=
1
0
are real and distinct; furthermore, they are both dierent from 1. Since no eigen is equivalent to the linear map M around some
value equals 1, the ane map L
is equivalent to nding periodic points of
point. Thus, nding periodic points of L
M . Moreover, since the eigenvalues of M are distinct, the matrix is diagonalizable.
exist
In total, this means that periodic points (of period greater than one) of L
if and only if one of the eigenvalues of M is an nth root of unity. But this is
has no
impossible since both eigenvalues are real and dierent from 1. Hence, L
periodic points except for the possible x-points.

Lemma 5.6. Assume that det A = 1, < 0 and a 0. For any integer n 1,
n (x, 0) = (0, y).
there are no x, y > 0 such that L
Proof. When det A = 1 and tr A > 2, the relations (tr A)2 = 4 det A and
n (x, 0) = (0, y) is
tr A = 1 + det A are fullled. Therefore, by Proposition 5.4, L
equivalent to
 n+1



+ n+1
(5.4)
x + a + F+n Fn = 0

 n



+ n x + a F+n Fn = (+ )y,
(5.5)




where F+n = 1 n+ /(1 + ) and Fn = 1 n /(1 ). In the current case,
0 < < 1 and + > 1, which implies that F+n > Fn > 0. Thus, when a 0 we
must have x 0 by equation (5.4).

Lemma 5.7. Assume that det A = 1, = 0 and a 0. For every integer n 1
n (x, 0) = (0, y).
there exist no x, y > 0 such that L

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

11

Proof. When det A = 1 and tr A = 2, the nth iterate of the ane map can
easily be calculated as
 


  
an n + 1
x
1 + n n
n x

+
L
,
=
y
n
1n
y
2 n1
n (x, 0) = (0, y) implies that x 0 since a 0.
and one sees directly that L

Proposition 5.8. Assume that det A = 1 and 0. If is an irreducible


nite dimensional -representation of CL,a in one of the following situations
(1) a 0,
(2) < 0, a < 0 and c 0,
(3) < 0, a < 0, c > 0 and / c 1,
then is one-dimensional.
Proof. In all three cases, Lemma 5.5 implies that there can be no non-trivial
(i.e. of dimension greater than one) loop representations.
In Case 1, Lemma 5.6 and Lemma 5.7 imply that there are no non-trivial string
representations. In Case 2 one can explicitly check that there is no component of
intersecting both positive axes. Thus, no non-trivial string representations can
exist. In Case 3, there are constraint curves with a component that do intersect
both positive axes. However, one can explicitly check that iterations of the point
of intersection with the positive r-axis (where a string must start) increases the
r-coordinate. Thus, one can never hit the positive s-axis (where a string must end)
which implies that no non-trivial string representations can exist.

In [ABH+ 09], a special case of CL,a was considered where it holds that 2 <
tr A < 2. Then tr A can be parametrized by setting tr A = 2 cos 2. This makes it
corresponds to a rotation by 2 on the constraint
obvious that the ane map L
curve, which will be an ellipse. The same kind of parametrization can be done when
tr A > 2, in which case it is convenient to set tr A = 2 cosh 2 with > 0. Let us
gather the formulas one obtains in the following proposition.
Proposition 5.9. Assume that
> 0. If we set


x1 () = c +
c


x2 () = c
c

det A = 1, c > 0 and tr A = 2 cosh 2 for some



cosh
cosh( 2)
, +
,
cosh
cosh
c

cosh
cosh( 2)
,
,
cosh
cosh
c

then the following holds


(1)
(2)
(3)

= {x1() : R} {x2 () : R},


xi () = xi ( + 2) for i = 1, 2,
L
n1 (x, 0) = (0, x) if and only if
L
x=

2 sinh sinh(n 1)
.
cosh n

12

JOAKIM ARNLIND AND SERGEI SILVESTROV

6. -representations of AL
To study the representation theory of AL , we will use the same techniques as for
the representation theory of CL,a ; we will again see that the dynamical properties
of an ane map is of crucial importance. Since there exists a homomorphism :
AL CL,a , every representation of CL,a induces a representation of AL . However,
in general there are representations that can not be induced from CL,a .
will be two comIn any (nite dimensional) -representation , (E) and (E)
muting hermitian matrices. Therefore, any such representation is unitarily equiva are diagonal. Let us assume such a basis to
lent to one where both (E) and (E)
be chosen and write
E = diag(e1 , e2 , . . . , en )
= diag(
E
e1 , e2 , . . . , en ).
For matrices in this basis, the dening relations of AL reduce to


Sij ei +
ei + u ej = 0


Sij ei +
ei + v ej = 0,
are diagonal. There are two ways of fullling these equations:
since S = T and E, E
Either Sij = 0 or
  
   
 
ej
ei
e

u
=L i ,
=
+
ej
ei
ei

v
and by dening vi = (ei , ei ) we write this as vj = L(vi ).
Let GS = (V, E) be the directed graph of S. If (i, j) E (i.e. Sij = 0) then a
necessary condition for a representation to exist is that vj = L(vi ). On the other
hand, given a graph G and vectors {vk } such that vj = L(vi ) if (i, j) E, then any
matrix whose digraph equals G denes a representation of AL . Hence, the set of
representations can be parameterized by graphs allowing such a construction.
Definition 6.1. A graph G = ({1, 2, . . . , n}, E) is called L-admissible if there
exists vk R2 for k = 1, 2, . . . , n, such that vj = L(vi ) if (i, j) E. An Ladmissible graph is called nondegenerate if there exists such a set {v1 , . . . , vn } with
at least two distinct vectors; otherwise the graph is called degenerate.
By this denition, the digraph of S in any representation is L-admissible, and
every L-admissible graph generates at least one representation. Clearly, given an
L-admissible graph, there can exist a multitude of inequivalent representations
associated to it. If L has a x-point (ef , ef ), then any graph is L-admissible and
= ef 1 and S an arbitrary matrix.
this representation corresponds to E = ef 1 and E
However, not all graphs will be nondegenerate L-admissible graphs.
Let us now show that in the case when the representation is locally injective
(cp. Remark 5.3), we can bring it to a convenient form. Let G = (V, E) be an Ladmissible connected graph (if it is not connected, the representation will trivially
be reducible, and we can separately consider each component) and let S be a
matrix with digraph equal to G, such that the representation is locally injective.
Furthermore, let v1 , . . . , vk be an enumeration of the pairwise distinct vectors in
the set {v1 , . . . , vn } such that vi+1 = L(
vi ), and dene Vi V as follows
Vi = {l V : vl = vi }

for i = 1, . . . , k.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

13

Since the representation is assumed to be locally injective, we can only have edges
from vertices in the set Vi to vertices in the set Vi+1 (identifying k + 1 1). Hence,
the vertices of the graph can be permuted such that the matrix S takes the following
block form:

0 S1 0
0
0
0 S2
0

..
.
..
..
..
..
(6.1)
S= .

.
.
.

0
0
0 Sk1
Sk 0
0
0
where each matrix Si is a |Vi | |Vi+1 | matrix. Thus, the representations of AL are
generated by the ane map L in the following way: Any point v1 R2 gives rise
to the points vi = (ei , ei ) = Li1 (
v1 ); by setting

e1 1n1
e1 1n1

..
..
=
E=
E

.
.
ek 1nk

ek 1nk

together with any matrix of the form (6.1), with Si a ni ni+1 matrix, one obtains
a representation of AL of dimension n1 + + nk . Unless x1 is a periodic point
of order k we must set Sk = 0. Moreover, distinct iterations of L (i.e, at least
one of the points dier) can not give rise to equivalent representations since the
will be dierent.
eigenvalues of E and E
7. Representations and surface geometry
We will now study the relation between the geometry of the inverse image =
C 1 (0) and representations of the derived algebra CL,a . More precisely, the geometry of C 1 (0), for dierent values of 0 , 1 , c0 will be compared with the representations of CL,a with c0 (the value of the central element) being equal to c0 , and
a, tr A, det A related to 0 , 1 , 1 , 1 , 1 ,  as in Section 4. Furthermore, the comparison will be made for small positive values of . When det A = 0, the ane map
will be invertible, and therefore Theorem 5.2 applies, i.e. all nite dimensional
L
-representations can be classied in terms of loops and strings.
Let us rewrite the polynomial C(x, y, z), as dened in (4.8), to a form which
makes it easier to identify the topology of the surface in the case when 1 = 0

 2
2

0
2 2
0
2c0
1
2
2
x +y +
+
z
+
C(x, y, z) =
.
4
1
1
12
1
but if 1 > 0 the genus of the
If 1 < 0 the inverse image will be non-compact,

surface will be determined by the quotient / c where


=

0
1

and

c=

02
2c0
+
.
2
1
1

If 1 < / c < 1 the inverse image is a compact surface of genus 0, and if / c > 1
the surface has genus 1 (see [ABH+ 09] for details and proofs). When 1 = 0, the
polynomial becomes
 1
0  2
1
C(x, y, z) =
x + y 2 + z 2 c0 ,
2
2
2

14

JOAKIM ARNLIND AND SERGEI SILVESTROV

and the smooth inverse images consist of ellipsoids and (one or two sheeted) hyperboloids. A complete table of the dierent geometries can be found in Appendix B.
We note that when the algebra CL,a arises from a surfaces, then 2 + 2
c0 /1 = c.
By introducing
1 + 22 1 12 2 1
02
2
c0
t2 =
and
c

=
+
2
2
4
1
1
one can rewrite the dening equation of the constraint curve as

20
2 8t2
(7.1)
+
(r s)2 4
c
p(r, s) = r + s +
1
1
0
(r + s) + t2 (r s)2 c0
p(r, s) =
(7.2)
2
when 1 = 0 and 1 = 0 respectively. Since we only consider small values of , we
can assume that t2 > 0.
Note that we will use the parameters of the algebra and the parameters of the
surface interchangeably, and they are assumed to be related as in Section 4.
7.1. The degenerate cases. Let us take a look at the cases when the inverse
image is not a surface (P.1 P6, Z.1 Z.4), by studying some examples. For instance,
in case P.4, will be the empty set, and we easily see that there are no non-negative
are
(r, s) on the constraint curve . Therefore, since the eigenvalues of D and D
non-negative, no representations can exist. In case P.2 one gets = {(0, 0, 0)},
and the only non-negative point on is (0, 0). Therefore, all representations must
= 0, which implies that W = 0.
satisfy D = D
By considering all degenerate cases, one can compile the following table:
= C 1 (0)

Irreducible -representations

None

{(0, 0, 0)}

(W ) = 0

{(x, y, 0) : x + y = |0 |/|1 |} (W ) = |0 |/|1 |
2

In particular, we note that all irreducible representations are one-dimensional.


7.2. Compact surfaces. We will focus on the compact surfaces for which
1 > 0 (P.7 P.10), as the only other compact surface (Z.5) can be treated analogously. When 1 > 0 and c > 0, the constraint curve will be an ellipse symmetric
around the line /4 and centered at (, ). The analysis of the corresponding nite
dimensional representations was done in [ABH+ 09] but we will recall some basic
facts.
can
Let us introduce (0, /2) such that 2 cos 2 = tr A. The action of L
then be written as
  

  
x
4 sin2
x = 2 cos 2 1
+
L
y
1
0
y
0
and one can understand it as a rotation
by an angle 2 on the ellipse. One
can easily show that R2>0 when / c > 1/ cos ; thus, by Lemma 5.1, no
representation in this region can contain a string, and
therefore all irreducible representations must consist of a single loop. When / c 1 no loop representations
2
2
can exist, since
a too large part of the ellipse is contained in R \R>0 . In the small
region 1 < / c 1/ cos (cos 1 as  0) both strings and loops can exist.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

15

Figure 2. The constraint curve and the points of iterations for


an 11-dimensional loop representation when 0 1.99, 1 3.15
and c = 1.
We call surfaces in this region critical tori ; these surfaces have a very narrow hole
through them.
However, representations do no exist for all values of and the following conditions must be fullled for a n-dimensional representation to exist:
Loop:
String:

e2in = 1

c cos n + cos = 0

String (Z.5):

c0 =

02 2 (n2 1)
.
4(1 + 22 1 )

In the case when / c > 1/ cos one can have two-sided innite dimensional
representations by letting be an irrational multiple of ; this is not possible
for the sphere. Let us summarize the representations for compact surfaces in the
following table:
= C 1 (0)

Irreducible -representations

Sphere

String representations

Critical torus

String and loop representations.

(Non-critical) torus

Loops, two-sided innite representations.

As an example, let us construct an 11-dimensional loop representation when the


surface is a torus. More precisely, we set = /11,  = tan(), tr A = 2 cos 2,
a = 1/2, c = 1 and 1 = 1/2, which corresponds to 0 1.99, 1 3.15 and
c = 1. In Figure 2 one nds the corresponding constraint curve and the 11 points
Let x1 (e.g. (1.56, 1)) be an initial point on the
of iteration of the ane map L.
curve and let xk = (dk , dk ) = Lk1 (x1 ) be its iterations. A -representation of CL,a
is then constructed by setting


0
d1 0

0
0
1.25
0

0
0
1.46
0
0
d2
0

..

.
.
.. .
.
.
.
.
.
.
..
..
..
..
..
..
.. ..
(W ) = .
.

0
0

0 0.79
d10
0

1.00
0

0
0
d11
0

0
0

16

JOAKIM ARNLIND AND SERGEI SILVESTROV

7.3. Non-compact surfaces. The remaining surfaces will have one or two
non-compact components (except for the surfaces in Section 7.4, which has both
a compact and a non-compact component) and we will show that innite representations always exist, whereas all nite dimensional representations are onedimensional. By looking at the tables in Appendix B, one sees that non-compact
surfaces appear only when 1 0 which, for small , is equivalent to 0.
We can now prove the following result about the relation between geometry and
representations.
Proposition 7.1. Let CL,a be an algebra corresponding to a surface where
each component is non-compact, and assume that at least one of 0 , 1 , c0 is different from zero. Then the following holds:
(1) All nite dimensional irreducible representations have dimension one.
(2) If has two components then there exists two inequivalent one-sided innite dimensional irreducible representations, but no two-sided representations.
(3) If is connected and non-singular, then there exists a two-sided innite
dimensional irreducible representation; if a 0, or a > 0 and c > 2 (1 +
||/4), then no one-sided representations exist. If a > 0 and c 2 (1 +
||/4) then one-sided representations exist.
Proof. Statement 1 follows immediately from Proposition 5.8. Statement 2
can be proved in the following way: By examining all cases in Appendix B where
C 1 (0) has two non-compact components (Z.8, N.1, N.2, N.5, N.8), one sees that
the components of which intersect R20 has the following form

In the rst case, two inequivalent one-sided representations can be constructed but
no two-sided representations can exist because backward or forward iterations of
any point will eventually reach outside R2>0 . In the second case, it holds that the
lower left tip of the curve intersecting R20 has strictly negative coordinates and
the curve intersects the positive axes exactly once. This immediately allows for
a construction of two inequivalent one-sided innite dimensional representations.
Now, can we have two-sided representations? Any constraint curve will cross the
positive r-axis in the following points:


1
(7.3)
r = a a2 + 4
c1
2
If there is only one strictly positive intersection-point, it must hold that r+ > 0
but r 0. Actually r < 0 since the lower left tip of the curve is not in R20 .
A necessary condition for a two-sided representations to exist is that there exists a
is contained
point on the curve such that all backward and forward iterations by L
2

in R>0 . This means that (since L preserves the direction of the curve) when we
to the point of intersection with the s-axis, we must obtain a point in R2
apply L
0

(otherwise no point is able to jump the negative part of by the action of L).

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

17

Figure 3. A connected non-compact surfaces which allows for


one-sided representations.
But this does not happen since
  
  
0 = r+ + a = r
L
/ R20 .
r+
0
0
Let us now prove the statement 3. When a 0 the are only two cases which
give a connected non-singular non-compact surface, namely N.7 and N.10. In both
cases, one can check that does not intersect the axes, and that at least one
component is contained in R2>0 . Hence, no one-sided representations exist but
two-sided representations exist.
When a > 0 (Z.6, N.4) and c > 2 (1 + ||/4) then the component of that
intersects R20 is contained in R2>0 , which implies that no one-sided representations
exist, but two-sided representations exist. When a > 0 and c 2 (1 + ||/4) one
component of the constraint curve will have the following form

In particular, it intersects the positive axes at least once. Thus, one-sided representations can be easily dened, but what about two-sided representations? We will
now show that the backward and forward iteration of the point at the lower left
tip both lie in R2>0 . We consider only the case Z.6 as the other case (N.4) can be
treated analogously. The lower tip of the ellipse has coordinates (r0 , r0 ) for some
r0 > 0. We calculate

  

  
r0 + a
r0
r0
2
1 r0

=
R>0 and L
=
R2>0 ,
L
r0
r0
r0
r0 + a
since a > 0. Hence, we can dene a two-sided representation by starting at (r0 , r0 )

and considering all backward and forward iterations by L.



The connected non-compact surfaces in Proposition 7.1 which allow for one-sided
innite dimensional representations correspond to surfaces with a narrow throat
as in Figure 3. This kind of tunneling is analogous to what happens for the
compact surfaces. For a compact torus with a narrow hole, string representations

18

JOAKIM ARNLIND AND SERGEI SILVESTROV

can continue to exist; but as the hole grows wider they cease to exist. For noncompact surfaces, one-sided representations continue to exist even though the two
components have been joined together. As an example, let us consider a surface
with 0 = 1, 1 = 1 and c = 1.02. Choosing  = 0.3, 1 = 1 = 1/4 and
1 = 1/2 gives us the constraint curve in Figure 4. On the left curve, the rst
iterations of a one-sided innite dimensional representations are plotted; on the
right curve one nds iterations corresponding to a two-sided representation. The
representations that are dened by these two gures the following form:

0 0.41

0
0.74

0
1.11
one-sided (W )

0
1.53

..
..
.
.

..

two-sided (W )

..
0

.
0.45
0

0.01
0

0.01
0

0.45
0

0.79
..
.

..

Figure 4. Constraint curves and iterations corresponding to a


non-compact one-sheeted surface with 0 = 1, 1 = 1 and
c = 1.02.
Let us make a remark about the surface that has been excluded from Proposition
7.1, namely Z.2 with c0 = 0. The polynomial reduces to C(x, y, z) = z 2 and the
inverse image describes the x, y-plane. The constraint curve is the line r s = 0
Hence, all irreducible representations
and every point (r, r) is a x-point of L.
are one-dimensional and the inequivalent representations are parametrized by the
non-negative real numbers.
7.4. A surface with both compact and non-compact components. In
the cases N.11 and N.12, the surface consists of two components: a non-compact
surface and a compact surface of genus 0 (which collapses to a point when / c = 1).
The constraint curve will have the form as in Figure 5. Thus, there is one compo-

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

19

Figure 5. The constraint curve corresponding to a surface with


both compact and non-compact components.
nent which allows for the construction of a nite dimensional string representation,
and one component
that induces a two-sided innite dimensional representation.

When / c = 1, the lower component will intersect R20 in the point (0, 0), which
allows for a one-dimensional trivial representation.
As for the compact surfaces, string representations do not exist for all values
of  (if we x c). In the notation of Proposition 5.9, the condition for the existence
of a n-dimensional string representation is

cosh c cosh n = 0.
7.5. Singular non-compact surfaces. The surfaces Z.7, N.3, N.6 have a
singularity at one point (which arises as two sheets come together) and the surface
N.9 is a limit case of N.12, where the sphere touches the non-compact surface. The
corresponding constraint curves will have one of the forms in Figure 6, where the

Figure 6. Constraint curves of singular non-compact surfaces.


left picture corresponds to Z.7 and N.3, and the right picture corresponds to N.6
and N.9. The left constraint curve clearly allows for two one-sided representations,
but no two-sided representation can exist (cp. proof of Proposition 7.1). The right
constraint curve allows for two dierent two-sided representations; it is easy to check
is invertible and (0, 0) is a x-point, all iterations of a point on the
that since L
curve in R2>0 stay in R2>0 . That is, iterations approach the origin but never reach
it. Furthermore, we note that no nite dimensional representations of dimension
greater than one can exist.
7.6. Correspondence between geometry and representation theory.
In [ABH+ 09], the representation theory of CL,a was compared with the geometry

20

JOAKIM ARNLIND AND SERGEI SILVESTROV

of the inverse image for a class of compacts surfaces of genus 0 and 1. We have now
extended this analysis to inverse images of general rotationally symmetric fourth
order polynomials. Apart from recovering earlier results, we have shown that the
representation theory respects the geometry of the surface to a high extent. Namely,
in all cases where = C 1 (0) is empty, no representations exist. When is not a
surface, then no irreducible representations of dimension greater than one exist. In
the case when is non-singular non-compact, the correspondence is as follows: if
has two sheets then there exists two inequivalent one-sided innite dimensional
representations, and if has one sheet there is a two-sided innite dimensional
representation. In all non-compact cases no nite-dimensional representations of
dimension greater than one exist.
Acknowledgement
This work was partially supported by the Swedish Research Council, the Crafoord
Foundation, the Swedish Royal Academy of Sciences and the Swedish Foundation of
International Cooperation in Research and Higher Education (STINT). J. A. would
also like to thank the Sonderforschungsbereich SFB647 as well as the Institut des

Hautes Etudes
Scientiques for nancial support and hospitality.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

21

Appendix A Solutions to system of equations


The general solution to the four equations

 


tr A
m n
m n

1
m
n

m
n


det A
=0
0

is given by
n = m
m
n
= m m.

If = 1 + det A tr A = 0 then the system




   
  
k

u
a
m n
12
(7.4)

=

v
0
m
n

k
Whenever a = 0, we can always solve (7.4) by
has a unique solution for k and k.
setting


1
1
m=
( 1)k + k + u
m
=
k + ( 1)k + v .
a
a
If = a = 0 there are two cases. When A = 12 it is necessary that u = v = 0,
in which case (7.4) is identically satised and the ane map L will be the identity
map. If A = 12 we get the following conditions

1 
u + k
= 1 :
if ( 1)v = u then k =
1

1
= 0 :
if ( 1)v = u then k = v + ( 1)k



1
v + k
= 1 :
if ( 1)u = v then k =
1

1
= 0 :
if ( 1)u = v then k = u + ( 1)k .

22

JOAKIM ARNLIND AND SERGEI SILVESTROV

Appendix B Inverse images of C(x, y, z)

c
/ c C 1 (0)
R20

1 > 0

P.1

<0

P.2

{(0, 0, 0)}

P.3

<0

{(0, 0)}


|0 |
{(x, y, 0) : x + y = |0 |/1 }
1 (1, 1)

P.4

>0

P.5

>0 >0

< 1

P.6

>0 >0

{(0, 0, 0)}

{(0, 0)}

P.7

0 >0

> 1

Sphere

{Ellipse} R+
0

P.8

<0 >0

<1

Sphere

{Ellipse} R+
0

P.9

<0 >0

Surface with singularity

{Ellipse} R+
0

P.10

<0 >0

>1

Torus

{Ellipse} R+
0

1 = 0

c0

C 1 (0)

Z.1

<0

Z.2

R20

{(x, y, c0 )} {x, y, c0 } Non-compact.

Z.3

>0 <0

Z.4

>0

{(0, 0, 0)}

{(0, 0)}

Z.5

>0 >0

Sphere

Compact.

Z.6

<0 <0

One sheeted hyperboloid

Non-compact.

Z.7

<0

Singular hyperboloid

Non-compact.

Z.8

<0 >0

Two sheeted hyperboloid

Non-compact.

1 < 0

/ c

N.1

<0 0

Two sheeted cone.

N.2

<0 >0

< 1

Two sheeted cone.

N.3

<0 >0

N.4

<0 >0

> 1

One sheeted cone.

C 1 (0)

One sheeted singular cone.

N.5

<0

Two sheeted cone.

N.6

One sheeted singular cone.

N.7

>0

One sheeted cone.

N.8

>0 <0

Two sheeted cone.

N.9

>0

One sheeted cone sphere (singular).

N.10

>0 >0

<1

N.11

>0 >0

N.12

>0 >0

>1

One sheeted cone.


One sheeted cone {(0, 0, 0)}.
One sheeted cone sphere.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

23

References
[ABH+ 09]

[Arn08a]
[Arn08b]
[AS94]
[BHSS91]
[BR79]

[BR81]

[Dav96]
[E65]
[E81]

[E82]

[Gli61a]
[Gli61b]
[Jor88]

[JSW95]

[Kar87]

[KTW85]

[Li92]
[Mac68]
[Mac76]

[Mac89]

[NVO04]

Joakim Arnlind, Martin Bordemann, Laurent Hofer, Jens Hoppe, and Hidehiko Shimada. Noncommutative Riemann surfaces by embeddings in R3 . Comm. Math. Phys.
(to appear), 2009. E-print: arXiv:0711.2588.
Joakim Arnlind. Graph Techniques for Matrix Equations and Eigenvalue Dynamics.
PhD thesis, Royal Institute of Technology, 2008.
Joakim Arnlind. Representation theory of C-algebras for a higher-order class of
spheres and tori. J. Math. Phys., 49(5):053502, 13, 2008.
R. J. Archbold and J. S. Spielberg. Topologically free actions and ideals in discrete
C -dynamical systems. Proc. Edinburgh Math. Soc. (2), 37(1):119124, 1994.
Martin Bordemann, Jens Hoppe, Peter Schaller, and Martin Schlichenmaier. gl()
and geometric quantization. Comm. Math. Phys., 138(2):209244, 1991.
Ola Bratteli and Derek W. Robinson. Operator algebras and quantum statistical mechanics. Vol. 1. Springer-Verlag, New York, 1979. C - and W -algebras, algebras,
symmetry groups, decomposition of states, Texts and Monographs in Physics.
Ola Bratteli and Derek W. Robinson. Operator algebras and quantum-statistical mechanics. II. Springer-Verlag, New York, 1981. Equilibrium states. Models in quantumstatistical mechanics, Texts and Monographs in Physics.
Kenneth R. Davidson. C -algebras by example, volume 6 of Fields Institute Monographs. American Mathematical Society, Providence, RI, 1996.
Edward G. Eros. Transformation groups and C -algebras. Ann. of Math. (2), 81:38
55, 1965.
Edward G. Eros. Polish transformation groups and classication problems. In General topology and modern analysis (Proc. Conf., Univ. California, Riverside, Calif.,
1980), pages 217227. Academic Press, New York, 1981.
Edward G. Eros. On the structure theory of C -algebras: some old and new problems. In Operator algebras and applications, Part I (Kingston, Ont., 1980), volume 38
of Proc. Sympos. Pure Math., pages 1934. Amer. Math. Soc., Providence, R.I., 1982.
James Glimm. Locally compact transformation groups. Trans. Amer. Math. Soc.,
101:124138, 1961.
James Glimm. Type I C -algebras. Ann. of Math. (2), 73:572612, 1961.
Palle E. T. Jorgensen. Operators and representation theory, volume 147 of NorthHolland Mathematics Studies. North-Holland Publishing Co., Amsterdam, 1988.
Canonical models for algebras of operators arising in quantum mechanics, Notas de
Matem
atica [Mathematical Notes], 120.
P. E. T. Jorgensen, L. M. Schmitt, and R. F. Werner. Positive representations of
general commutation relations allowing Wick ordering. J. Funct. Anal., 134(1):33
99, 1995.
Gregory Karpilovsky. The algebraic structure of crossed products, volume 142 of
North-Holland Mathematics Studies. North-Holland Publishing Co., Amsterdam,
1987. Notas de Matem
atica [Mathematical Notes], 118.
Shinz
o Kawamura, Jun Tomiyama, and Yasuo Watatani. Finite-dimensional irreducible representations of C -algebras associated with topological dynamical systems.
Math. Scand., 56(2):241248, 1985.
Bing Ren Li. Introduction to operator algebras. World Scientic Publishing Co. Inc.,
River Edge, NJ, 1992.
George W. Mackey. Induced representations of groups and quantum mechanics. W.
A. Benjamin, Inc., New York-Amsterdam, 1968.
George W. Mackey. The theory of unitary group representations. University of
Chicago Press, Chicago, Ill., 1976. Based on notes by James M. G. Fell and David
B. Lowdenslager of lectures given at the University of Chicago, Chicago, Ill., 1955,
Chicago Lectures in Mathematics.
George W. Mackey. Unitary group representations in physics, probability, and number theory. Advanced Book Classics. Addison-Wesley Publishing Company Advanced
Book Program, Redwood City, CA, second edition, 1989.
Constantin N
ast
asescu and Freddy Van Oystaeyen. Methods of graded rings, volume
1836 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 2004.

24

[NVO08]

JOAKIM ARNLIND AND SERGEI SILVESTROV

Erna Nauwelaerts and Freddy Van Oystaeyen. Introducing crystalline graded algebras. Algebr. Represent. Theory, 11(2):133148, 2008.
[OS89]
V. L. Ostrovski and Yu. S. Samolenko. Unbounded operators satisfying non-Lie
commutation relations. Rep. Math. Phys., 28(1):91104, 1989.
[OS99]
V. Ostrovskyi and Yu. Samoilenko. Introduction to the theory of representations of
nitely presented *-algebras. I, volume 11 of Reviews in Mathematics and Mathematical Physics. Harwood Academic Publishers, Amsterdam, 1999. Representations by
bounded operators.

[OS08a]
Johan Oinert
and Sergei D. Silvestrov. Commutativity and ideals in algebraic crossed
products. J. Gen. Lie Theory Appl., 2:287302, 2008.

[OS08b]
Johan Oinert
and Sergei D. Silvestrov. Commutativity and ideals in pre-crystalline
graded rings. Preprints in Mathematical Sciences 2008:17, Centre for Mathematical
Sciences, Lund University, 2008. To appear in Acta Applicandae Mathematicae.

[OS08c]
Johan Oinert
and Sergei D. Silvestrov. On a correspondence between ideals and commutativity in algebraic crossed products. J. Gen. Lie Theory Appl., 2(3):216220,
2008.

[OS09]
Johan Oinert
and Sergei D. Silvestrov. Crossed product-like and pre-crystalline graded
rings. In V. Abramov, E. Paal, Sergei D. Silvestrov, and A. Stolin, editors, Generalized
Lie theory in Mathematics, Physics and Beyond, chapter 24, pages 281296. SpringerVerlag, Berlin, Heidelberg, 2009.

[OSTAV08]
Johan Oinert,
Sergei D. Silvestrov, T. Theohari-Apostolidi, and H. Vavatsoulas. Commutativity and ideals in strongly graded rings. Preprints in Mathematical Sciences
2008:13, Centre for Mathematical Sciences, Lund University, 2008. To appear in Acta
Applicandae Mathematicae.
[Pas89]
Donald S. Passman. Innite crossed products, volume 135 of Pure and Applied Mathematics. Academic Press Inc., Boston, MA, 1989.
[Ped79]
Gert K. Pedersen. C -algebras and their automorphism groups, volume 14 of London
Mathematical Society Monographs. Academic Press Inc. [Harcourt Brace Jovanovich
Publishers], London, 1979.
[Sak71]
Sh
oichir
o Sakai. C -algebras and W -algebras. Springer-Verlag, New York, 1971.
Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 60.
[Sam91]
Yu. S. Samolenko. Spectral theory of families of selfadjoint operators, volume 57
of Mathematics and its Applications (Soviet Series). Kluwer Academic Publishers
Group, Dordrecht, 1991. Translated from the Russian by E. V. Tisjachnij.
[Sil95]
Sergei D. Silvestrov. Representations of Commutation Relations. A Dynamical Systems Approach. PhD thesis, Departement of Mathematics, Ume
a University, 1995.
Published in Hadronic Journal Supplement vol. 11 116 pp, 1996.
[SSdJ07a] Christian Svensson, Sergei Silvestrov, and Marcel de Jeu. Dynamical systems and
commutants in crossed products. Internat. J. Math., 18(4):455471, 2007.
[SSdJ07b] Christian Svensson, Sergei D. Silvestrov, and Marcel de Jeu. Dynamical systems associated with crossed products. Preprint in Mathematical Sciences 2007:22, Centre for
Mathematical Sciences, Lund University, 2007. E-print: arXiv:0807.2940. To appear
in Acta Applicandae Mathematicae.
[SSdJ09]
Christian Svensson, Sergei D. Silvestrov, and Marcel de Jeu. Connections between
dynamical systems and crossed products of Banach algebras by Z. In J. Janas,
P. Kurasov, A. Laptev, S. Naboko, and G. Stolz, editors, Methods of Spectral Analysis
in Mathematical Physics, volume 186 of Operator Theory: Advances and Applications, pages 391401. Birkh
auser, Basel, 2009.
[ST02]
Sergei D. Silvestrov and Jun Tomiyama. Topological dynamical systems of type I.
Expo. Math., 20(2):117142, 2002.
[ST08]
Christian Svensson and Jun Tomiyama. On the commutant of C(X) in C -crossed
products by Z and their representations. Preprint in Mathematical Sciences 2008:21,
Centre for Mathematical Sciences, Lund University, 2008. E-print: arXiv:0807.2940.
To appear in J. Funct. Anal.
[ST09]
Christian Svensson and Jun Tomiyama. On the Banach -algebra crossed product
associated with a topological dynamical system. Leiden Mathematical Institute report
2009:03, Leiden Mathematical Institute, 2009. E-print: arXiv:0902.0690.

AFFINE CROSSED PRODUCT ALGEBRAS AND NONCOMMUTATIVE SURFACES

[SW96]

[Tak79]
[Tom87]

[Tom92]

[VS88]

[VS90]

25

Sergei D. Silvestrov and Hans Wallin. Representations of algebras associated with a


M
obius transformation. J. Nonlinear Math. Phys., 3(1-2):202213, 1996. Symmetry
in nonlinear mathematical physics, Vol. 2 (Kiev, 1995).
Masamichi Takesaki. Theory of operator algebras. I. Springer-Verlag, New York, 1979.
Jun Tomiyama. Invitation to C -algebras and topological dynamics, volume 3 of
World Scientic Advanced Series in Dynamical Systems. World Scientic Publishing
Co., Singapore, 1987.
Jun Tomiyama. The interplay between topological dynamics and theory of C algebras, volume 2 of Lecture Notes Series. Seoul National University Research Institute of Mathematics Global Analysis Research Center, Seoul, 1992.
E. E. Vasleb and Yu. S. Samolenko. On the representation of the relations AU =
U F (A) by an unbounded selfadjoint operator and a unitary operator. In Boundary value problems for dierential equations (Russian), pages 3052, ii. Akad. Nauk
Ukrain. SSR Inst. Mat., Kiev, 1988.
E. E. Vasleb and Yu. S. Samolenko. Representations of operator relations by unbounded operators, and multidimensional dynamical systems. Ukrain. Mat. Zh.,
42(8):10111019, 1990.

hlenberg 1, D-14476 Golm, Germany.


Albert Einstein Institute, Am Mu
E-mail address: joakim.arnlind@aei.mpg.de
Center for Mathematical Sciences, Box 118, S-22100 Lund, Sweden.
E-mail address: sergei.silvestrov@math.lth.se

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

C*-ALGEBRAS ASSOCIATED WITH ITERATED


FUNCTION SYSTEMS
Gilles G. de Castro
Abstract. We review Kajiwara and Watatanis construction of a C*-algebra
from an iterated function system (IFS). If the IFS satises the nite branch
condition or the open set condition, we build an injective homomorphism from
Kajiwara-Watatani algebras to the Cuntz algebra, which can be thought as the
algebra of the lifted system, and we give the description of its image. Finally,
if the IFS admits a left inverse we show that the Kajiwara-Watatani algebra
is isomorphic to an Exels crossed product.

1. Introduction
In [16], Kajiwara and Watatani dened a C*-algebra dened from an iterated
function system (IFS). Although their paper was entitled C*-algebras associated
with self-similar sets, they gave examples of dierent iterated function systems
which give rise to the same self-similar set but which associated algebras are not
isomorphic. So their algebra depends not only on the self-similar set but on the
dynamics of the iterated function system.
If the IFS satises the strong separation condition, then the system can be
interpreted as the inverses branches of a local homeomorphism. For an arbitrary
IFS we can lift it to a new one that satises the strong separation condition [1].
Ionescu and Muhly suggested in [13] the construction of a C*-algebra from an IFS
by lifting it and using Renault-Deaconu construction [6], [19] of a groupoid C*algebra from a local homeomorphism. As we will see this local homeomorphism we
nd is topologically conjugate to the left shift on {1, . . . , d}N and the algebra we
nd is the Cuntz algebra Od .
From the relations between an IFS and its lifted system, we will build a natural homomorphism from the Kajiwara-Watatani algebra to the Cuntz algebra and
thus connect Kajiwara anda Watatanis construction with Ionescu and Muhlys
suggestion. We show that this homomorphism is injective and show that its image
is generated by the algebra of the self-similar set associated to the IFS and an
isometry S similar to a crossed product description.
2000 Mathematics Subject Classication. Primary 46L55, 37B99; Secondary 28A80, 37B10,
46L08.
Key words and phrases. Iterated function systems, C*-algebras, crossed products, groupoids.
Partially supported by CAPES.

1
27

28
2

GILLES G. DE CASTRO

The interplay between IFS and Cuntz algebras have been deeply studied by Jorgensen and collaborators. Among other thing, they established some relationships
between certain representations of Cuntz algebras coming from IFS and certain
wavelet basis. See [2], [14] and references therein for details and other results.
It may happen that the IFS admits a left inverse which is not necessarily a local
homeomorphism. In this case Renault-Deaconus construction no longer works and
we have dierent approaches to build a C*-algebra. In this paper, we show that
under some assumptions, the algebra considered by Kajiwara and Watatani can be
seen as an Exels crossed product [8].
2. Iterated function systems
In this section, we review some of the basic theory of iterated function systems
and self-similar sets (see for instance [1], [7] and [11]). Fix (X, ) be a compact
metric space.
Definition 2.1. We say that a function : X X is
a contraction if c (0, 1) such that ((x), (y)) c(x, y);
a proper contraction if c1 , c2 (0, 1) such that c1 (x, y) ((x), (y))
c2 (x, y);
a similarity if c > 0 such that ((x), (y)) = c(x, y).
Definition 2.2. An iterated function system (IFS) over X is a nite set of
d
continuous functions {i : X X}i=1 . We say that the IFS is hyperbolic if all
functions are contractions.
Throughout this paper we will always assume that the system is hyperbolic
unless stated otherwise.
Proposition 2.3. Given an IFS {i }di=1 , there is a unique compact nonempty
subset K of X such that
K = di=1 i (K).

(2.1)

We will call this set the attractor of the IF S and say it is self-similar.
Note that because of (2.1) the attractor is invariant by all i and we can restrict
the IFS to its attractor. From now on, we assume that X = K.
d

Definition 2.4. We say that an IFS {i }i=1 satises:


the strong separation condition if the union in (2.1) is a disjoint union;
the open set condition if U K open and dense such that
d
U i=1 i (U )

where represents the disjoint union.


N

Lets denote = {1, . . . , d} with the product topology, : the left


shift and i : the function given by
i (i0 , i1, . . .) = (i, i0 , i1, . . .)
where i {1, . . . , d}.

C*-ALGEBRAS ASSOCIATED WITH ITERATED FUNCTION SYSTEMS

29
3

Proposition 2.5. Let {i }di=1 be an IFS and K its attractor then there is a
continuous surjection F : K such that F i = i F . This map is given by
the formula
F (i0 , i1, . . .) = lim i0 in (x)
n

for an arbitrary x K. If the IFS satises the strong separation condition F is a


homeomorphism.
Remark 2.6. Note that under the strong separation condition, we can dene
the function = F F 1 and in this case the functions i are exactly the inverse
branches of . Moreover F gives us a topological conjugacy between and the shift
.
For an arbitrary IFS {i }di=1 , we can always build a new one that satises the
strong separation condition and which share some properties with the original one
 = K and dene the functions
 X
 by
[1]. We dene X
i : X
i (x, ) =

(i (x), i ()). Let K = {(x, ) X |F () = x} then

 = d i (K).
K
i=1
 K}
 d satisfy the strong separation condiAnd its easily checked that {
i : K
i=1
tion.
Definition 2.7. The IFS {
i }di=1 as above is called the lifted system of {i }di=1 .
3. C*-algebras associated with an IFS
We start this section by giving a description of the Cuntz algebra as a groupoid
C*-algebra which will be useful in some proofs. We review some of the key elements
of Cuntz-Pimsner algebras which will be used to build Kajiwara-Watatani algebras.
We build the homomorphism from Kajiwara-Watatani algebras to the Cuntz algebra from a very natural covariant representation. We give some basic properties
of this homomorphism. Finally, in the last subsection, we compare the KajiwaraWatatani algebra to a crossed product construction.
3.1. Cuntz algebras.
Definition 3.1. [5] For d N\{0}, the Cuntz algebra Od is the C*-algebra
d
generated by d isometries satisfying the relation i=1 Si Si = 1.
For the sake of some proofs, we review the construction of the Cuntz algebra
as a groupoid C*-algebra [6], [19]. Let
G = {(, m n, ) Z : m, n N; m () = n ( )}
with the product and the inverse given by
(, m n, )(, k l, ) = (, (m + k) (n + l), )
(, m n, )1 = (, n m, ).
We give a basis for the topology on G by the sets
B(U, V, m, n) := {(, m n, ) G : U, V }
where m, n N and U and V are open subsets of such that m |U , n |V are
homeomorphisms with m (U ) = n (V ). With this topology G is etale and so
admits a Haar system by the counting measures.

30
4

GILLES G. DE CASTRO

The multiplication in Cc (G) is given by



(p q)(, m n, ) =
p(, k l, )q(, (n + k) (m + l), )
where the sum is taken over all k, l N and such that k () = l () and
n+k () = m+l ( ); and the involution by
p (, m n, ) = p(, n m, )
for p, q Cc (G).
We refer to [19] for the construction of a norm in Cc (G). For us, it suces to
know that there exists a norm in Cc (G) such that its completion with respect to
this norm is Od .
Finally, given h C() we dene a function h Cc (G) by
h(, m n, ) = [m = n][ = ]h()
where [] is the boolean function that gives 1 if its argument is true and 0 otherwise.
We also dene a function S Cc (G) by
S(, m n, ) = [m n = 1][() = ].
And we note that if i is the characteristic function of the cylinder i := { :
0 = i} then Si = d1/2 i S for i {1, . . . , d} are d isometries that satises the
Cuntz relation and generates C (G).
3.2. Cuntz-Pimsner algebras. We briey recall the key elements for the
construction of Cuntz-Pimsner algebras ([15], [18]) that will be used throughout
the paper. For that x A a C*-algebra.
Definition 3.2. A (right) Hilbert C*-module over A is a (right-)A-module E
with a sesquilinear map ,
: E E A such that:
(i) , a
= ,
a;
(ii) ( ,
) = ,
;
(iii) ,
0;
(iv) E is complete with respect to the norm ||||2 = || ,
||1/2
for a A and , E. We say that E is full if E, E
is dense in A.
Let E be a Hilbert C*-module and denote by L(E) the space of adjointable
operators in E. We note that L(E) is a C*-algebra. For , E we dene an
operator , : E E by , () = ,
. This is an adjointable operator and we
denote by K(E) the closed subspace of L(E) generated by all , .
Definition 3.3. A C*-correspondence over A is a Hilbert C*-module E together with a C*-homomorphism : A L(E).
Let (E,) be a C*-correspondence over A and for simplicity suppose that is
faithful. We denote by JE the ideal 1 (K(E)).
Definition 3.4. A pair (, ) of maps : A B, : E B, where B is a
C*-algebra and a C*-homomorphism, is said to be a covariant representation of
E if:
(i) ((a)b) = (a)()(b);
(ii) () () = ( ,
);
(iii) (, )(1) ((c)) = (c) where the function (, )(1) : K(E) B is given by
(, )(1) (, ) = ()() ,

C*-ALGEBRAS ASSOCIATED WITH ITERATED FUNCTION SYSTEMS

31
5

for a, b A, , E and c JE .
For a C*-correspondence (E,), there exists an algebra O(E) and a covariant
representation (kA , kE ) that is universal, in the sense that if (, ) is a covariant
representation of E in a C*-algebra B, there is a unique C*-homomorphism :
O(E) B such that = ( ) kA and = ( ) kE
Definition 3.5. The algebra O(E) is called the Cuntz-Pimsner algebra of E.
d
3.3. Kajiwara-Watatani algebras. Let = {i }i=1 be an iterated function
system and K its attractor. We recall the C*-correspondence dened in [16]. We
let A = C(K), E = C(G) where

G = di=1 Gi
with
Gi {(x, y) K K : x = i }
being the cographs in the terminology of [16]. The structure of C*-correspondence
is given by
((a)b)(x, y) = a(x)(x, y)b(y)
and
,
A (y) =

d


(i (y), y)(i (y), y)

i=1

for a, b A and , E.
Proposition 3.6 ([16]). (E = C(G), ) is a full C*-correspondence over A =
C(K) and : A L(E) is faithful and unital. Moreover, the Hilbert module norm
is equivalent to the sup norm in C(G).
Definition 3.7. The Kajiwara-Watatani algebra O associated to is the
Cuntz-Pimsner algebra associated to the C*-correspondence dened above.
Regarding Od as C (G) as in subsection 3.1 and recalling the code map F given
in proposition 2.5, we dene : A Od by
(a)(, m n, ) = [m = n][ = ]a(F ())

(3.1)
and : E Od by
(3.2)

()(, m n, ) = [m n = 1][() = ](F (), F ( ))

Note that if () = , then 0 ( ) = and F () = 0 (F ( )) so that


(F (), F ( )) G and is well dened.
Before showing that this give us a Cuntz-Pimsner covariant representation, let
us recall some denitions and results from [16] and [17].
Definition 3.8. Let = {1 , ..., d } be an IFS, we dene the following sets
B(1 , ..., d ) := {x K|y K i = j : x = i (y) = j (y)};
C(1 , ..., d ) := {y K|i = j : i (y) = j (y)}.
I(x) := {i {1, ..., d}; y K : x = i (y)}.
We call the points of B() branched points and the points of C() branched values.
And we say that satises the nite branch condition if C() is nite.

32
6

GILLES G. DE CASTRO

Then B(1 , ..., d ) is a closed set, because


B(1 , ..., d ) = i=j {x i (K) j (K); i1 (x) = j1 (x)}
and each of the union is clearly closed.
Lemma 3.9. In the above situation, if x K\B(1 , ..., d ), then there exists an
open neighborhood Ux of x satisfying the following:
(i) Ux B = ;
(ii) If i I(x), then j (i1 (Ux )) Ux = for j = i;
(iii) If i
/ I(x), then Ux i (K) = .
Lemma 3.10. If satises the nite branch condition or the open set condition
then JE = {a A = C(K); a vanishes on B(1 , ..., d )} where JE = 1 (K(E)) as
in the previous subsection.
Remark 3.11. In the following proof, we will need an explicit description of
(a) for certain elements in JE . We do as in [16]. Let B = B(1 , ..., d ) and take
a A such that Y := supp(a) K\B. Clearly a JE .
For each x Y choose an open neighborhood Ux as in lemma 3.9. Since Y
is compact, there exists a nite set {x1 , . . . , xm } such that Y m
k=1 Uxk . Let
is
an
open cover
Uk = Uxk for k = 1, . . . , m and Um+1 = K\Y , then {Uk }m+1
k=1
of K. Let {k }m+1

C(K)
be
a
partition
of
unity
subordinate
to
k=1

 this open
cover. Dene
m k , k C(G) by k (x, y) = a(x) k (x) and k (x, y) = k (x) then
(a) = k=1 k ,k (the summation goes to m only because m+1 = 0).
Remark 3.12. Because of the lemma 3.10, our results will need that the IFS
satises the nite branch condition or the open set condition, but we note that
these conditions are independent.
Proposition 3.13. If the IFS satises the nite branch condition or the
open set condition, then the pair (, ) dened by equations (3.1) and (3.2) is a
Cuntz-Pimsner covariant representation of (A, E) in Od .
Proof. Most calculations are very similar so we only show some of them. Let
a A and E. We have
(3.3)

(a)(, m n, ) = [m n = 1][() = ]a(F ())(F (), F ( ))

On the other hand


((a) ())(, m n, ) =

(a)(, k l, )()(, (m + l) (n + k), ) =

a(F ())()(, m n, )

(3.4)

where the second equality is true due to the fact that (a) is zero unless k = l and
= . We can easily see then that (3.3) and (3.4) coincide.
For the A-valued scalar product, let , E. Then
( ,
A )(, m n, ) = [m = n][ = ] ,
A (F ()) =
(3.5)

[m = n][ = ]

d


(i (F ()), F ())(i (F ()), F ())

i=1

and on the other hand


(() ())(, m n, ) =

() (, k l, )(g)(, (m + l) (n + k), ) =

C*-ALGEBRAS ASSOCIATED WITH ITERATED FUNCTION SYSTEMS

33
7

()(, l k, )(g)(, (m + l) (n + k), ) =



[m = n][ = ]
(F (), F ())(F (), F ())

(3.6)

()=

and we note that () = i = i () for some i = 1, . . . , d and in this case


F () = i (F ()). It follows that we can rewrite (3.6) as (3.5).
show that (, )(1) ((a)) = (a) for a JE where JE =
 Finally, we have to 
a A : a|B(1 ,...,d ) = 0 by lemma 3.10. We take a JE such that Y := supp(a)
K\B and k and k as in remark 3.11, then

((k ) (k ) )(, m n, ) =
(, )(1) ((a))(, m n, ) =


(k )(, k l, )(k ) (, (m + l) (n + k), ) =

[m = n][() = ( )]


k

(3.7)

[m = n][() = ( )]

k (F (), F (()))k (F ( ), F (())) =






a(F ()) k (F ()) k (F ( )).

Note that F () = 0 (F (())) and () = ( ) implies that F ( ) = 0 (F (()))


where 0 , 0 are the coordinates zero of and respectively. Now if F () Uxk
then 0 I(xk ) because of property (iii) of lemma 3.9 and F (() 1
(Uxk ).
0
(U
))
and
if

then
by
property
(ii)
of
lemma
We have that F ( ) 0 (1
x
0
0
k
0
3.9, we have F ( )
/ Uxk . Since the support of k is contained in Uxk and if 0 = 0
then = , we have from (3.7) that
(, )(1) ((a))(, m n, ) = [m = n][ = ]a(F ()) =
(a)(, m n, ).
As the elements a C(K) such that supp(a) K\B are dense in JE , the equality

(, )(1) ((a)) = (a) holds for an arbitrary a JE .
Lemma 3.14 ([12]). Suppose that (, ) is an isometric covariant representation
of E into a C*-algebra B. Then is faithful if and only if is faithful and there is
a (strongly continuous) action : T Aut(B) such that z = and z = z
for all z T.
Proposition 3.15. If the IFS satises the nite branch condition or the open
set condition, then the homomorphism given by the covariant representation
dened by (3.1) and (3.2) is faithful.
Proof. Given a C(K),
((a) (a))(, m n, ) = [m = n][ = ]|a(F ())|2
and as F is surjective, we have that is faithful. Let : T Aut(C (G)) be the
gauge action given by
z (f )(, m n, ) = z mn f (, m n, )
then z ((a)) = (a) because (a) is zero for m = n; and for E, z (()) = z()
because () is zero for m n = 1.

We conclude with this proposition that O is a subalgebra of Od .

34
8

GILLES G. DE CASTRO

Remark 3.16. As G is a closed, and therefore compact, subset of K K, all


continuous functions in G can be seen as restrictions of continuous functions in
K K. And viewing C(K K) = C(K) C(K), we have that every continuous
function in G can be written as a limit of sums of elements of the type a b
where a, b C(K). We can do this both with respect to the sup norm and to the
Hilbert-module norm because of proposition 3.6.
We also note that the code map F : K dened in proposition 2.5 induces
an injection of C(K) in C().
Proposition 3.17. If the IFS satises the nite branch condition or the
open set condition, then O is the sub-C*-algebra of Od generated by C(K) and S.
Proof. As a Cuntz-Pimsner algebra is generated by copies of elements of the
algebra and copies of elements of the module, we have that O is generated by all
elements a C(K) and C(G). It suces to note that (1) = S where 1 is the
identity of C(G) and (a b) = (a)(1)(b) for a, b C(K).

We recall a denition from [16] and give a dierent proof of the isomorphism
between O and Od for a certain class of IFS.
Definition 3.18. We say that the IFS {i }di=1 satises the cograph separation
condition if the cographs
Gi = {(i (y), y) K K : y K}
are disjoint.
We note that the cographs of an IFS are always closed subsets of G and if
the IFS satises the cograph separation condition then they are also open. In this
case, there are no branched points and in particular, it satises the nite branch
condition.
Proposition 3.19. If the IFS {i }di=1 satises the cograph separation condition
then O  Od .
Proof. If Gi is the characteristic function of Gi then it belongs to C(G) and
we note that (Gi ) = i S where i is the characteristic function of the cylinder
i. As weve seen in subsection 3.1, the elements Si = d1/2 i S = d1/2 (Gi ) are
d isometries that satises the Cuntz relations and generates Od .

3.4. The case of inverse branches of a continuous function. In this
subsection we suppose that there exists a continuous function : K K such
that i = id for all i {1, . . . , d}. Our goal is to show that if the IFS satises
the nite branch condition than we can see O as an Exels crossed product by
endomorphism [8].
We note that needs not to be a local homeomorphism and in this case we
cannot use the construction by Renault [19] and Deaconu [6]. But when it does,
their construction is isomorphic to Exels one [10].
We begin by recalling the ingredients to build Exels crossed product. Let A
be a unital C*-algebra and suppose were given:
An unital injective endomorphism : A A.
A transfer operator L : A A for , that is, a positive continuous linear
map such that L((a)b) = aL(b) for a, b A. We suppose that L(1) = 1.

C*-ALGEBRAS ASSOCIATED WITH ITERATED FUNCTION SYSTEMS

35
9

Let T (A, , L) be the universal C*-algebra generated by a copy of A and an


element S with relations:
 = (a)S,

(i) Sa


(ii) S aS = L(a),
for a A. Note that the canonical map from A to T (A, , L) is injective.
Definition 3.20. A redundancy is a pair (a, k) A ASS A such that abS =

kbS for all b A.
Definition 3.21. The Exels crossed product A ,L N is the quotient of
T (A, , L) by the closed two-sided ideal generated by the set of dierences a k
for all redundancies (a, k).
In our case, let A = C(K) and : A A be given by
(a) = a .
Then L : A A dened by
1
a i
d i=1
d

L(a) =
is a transfer operator for .
d

Theorem 3.22. Let = {i }i=1 be an IFS satisfying the nite branch condition
or the open set condition, and let A, and L be as above then A,L N is isomorphic
to O .
Proof. The steps of the proof are similar to what we have done last subsection.
Let (A = C(K), E = C(G)) be the C*-correspondence given in last subsection. We
start by giving a covariant representation of (A, E) in A ,L N.
Let : A A ,L N be the canonical inclusion and : E A ,L N be
given by
(a b) = aSb

for a, b A. To show that is well dened in all C(G), we let j aj bj be a
nite sum where aj , bj A, then
 


 
 
 
 


 

 
 


 








=
aj bj  
aj Sbj  
aj (bj )S  
aj (bj )
=

  j
  j


  j
j


2 1/2 








d 











= sup
aj (x)bj ((x)) sup
a
(
(x))b
(x)
=
a

j i
j
j
j



xK j
xK i=1 j


 j

2

where 2 is the norm in C(G) thinking of C(G) as an A-Hilbert module. To justify
that the second inequality above holds, we note that because K is self-similar, for
any x K there is y K such that x = i (y) for some i = 1, . . . , d.
We have to show that (, ) is a Cuntz-Pimsner covariant representation, i.e.,
it satises conditions (i)-(iii) of denition 3.4. Condition (i) is easily veried. For
(ii), it suces to show for monomials a b, e f C(G) because of linearity and
continuity. We have

a b, e f
(y) =
a(i (y))b(y)e(i (y))f (y) = b (y)L(a e)(y)f (y)
i

36
10

GILLES G. DE CASTRO

and then
( a b, e f
) = b L(a e)f = b S a eSf = (a b) (e f ).
Finally, for condition (iii), we take a JE such that supp(a) K\B and k , k
as in remark 3.11. Then



(k )(k ) =
a k SS k
(, )(1) ((a)) =
k

and we have to show that this equals a inside A ,L N. For that, we show that


the pair (a, a k SS k ) is a redundancy. We let b A, then






a k SS k bS =
a k ( L)( k b)S.
To show that the pair above is a redundancy, it suces to show that


b(x) =
k (x)( L)( k b)(x)
for x supp(a). For such x, we have that x
/ B and hence, there is a unique i0
and a unique y such that i0 (y) = x. If k (x) = k (i0 (y)) = 0 then i0 I(xk )
because of (iii) of lemma 3.9 and because of (ii) we have that i (y)
/ Uxk for i = i0 .
It follows that
d




k (x)( L)( k b)(x) =


k (x)
k (i ((x)))b(i ((x))) =
i=1
d




k (x)
k (i (y))b(i (y)) =
k (x) k (i0 (y))b(i0 (y)) =
i=1

k (x)b(x) = b(x).

By the universality of O , we have a homomorphism : O A ,L N.


Since A ,L N is generated by A and S, and (A) = A, (1) = S, where 1 is the
unity of C(G), we have that is surjective.
To show that is injective, we rst note that is faithful [9]. Then we see
that : T A ,L N given by z (a) = a and z (S) = zS is an action of the circle

in A ,L N [9] which clearly satises the conditions of lemma 3.14.
Remark 3.23. It was pointed out by the referee two references close related
to this work. In [3], its studied the construction of an algebra from a shift space
using Exels crossed product. The dynamics of a IFS is closed related to a symbolic
dynamic (proposition 2.5) and the Cuntz algebra can be thought as the algebra
associated to the full shift. In [4], its proved that some dynamical properties can
be translated into algebraic ones in the case of cover maps, which in our context is
related to IFS that have no branched points. A result similar to theorem 6 in [4]
to the context of IFS is an interesting problem for future work.
Acknowledgements. The author would like to thank his three advisors: Ruy
Exel, Artur Lopes and Jean Renault. The author would also like to thank the
Universite dOrleans and the MAPMO for their hospitality.

C*-ALGEBRAS ASSOCIATED WITH ITERATED FUNCTION SYSTEMS

37
11

References
1. M. Barnsley, Fractals Everywhere, Academic Press, Inc., 1988.
2. O. Bratteli and P. Jorgensen Wavelets through a looking glass. The world of the spectrum.,
Applied and Numerical Harmonic Analysis, Birkh
auser, 2002.
3. T. Carlsen and S. Silvestrov, C*-crossed products and shift spaces, Expo. Math. 25 (2007),
275-307.
4. T. Carlsen and S. Silvestrov, On the Exel crossed product of topological covering maps,
arxiv:[math.OA]/0811.0056.
5. J. Cuntz, Simple C*-algebras generated by isometries, Commun. Math. Phys. 57 (1977), 173185.
6. V. Deaconu, Groupoids associated with endomorphisms, Trans. Am. Math. Soc. 347 (1995),
1779-1786.
7. G. Edgar, Measure, topology, and fractal geometry. 2nd ed., Undergraduate Texts in Mathematics, Springer-Verlag, 2008.
8. R. Exel, A new look at the crossed-product of a C*-algebra by an endomorphism, Ergodic
Theory Dyn. Syst. 23 (2003), 1733-1750.
9. R. Exel, Crossed-products by nite index endomorphisms and KMS states, J. Funct. Anal.
199 (2003), 153-188.
10. R. Exel and A. Vershik, C*-algebras of irreversible dynamical systems, Can. J. Math. 58
(2006), 39-63.
11. K. Falconer, Fractal geometry. Mathematical foundations and applications. 2nd ed., Wiley,
2003.
12. N. Fowler, P. Muhly and I. Raeburn, Representations of Cuntz-Pimsner algebras, Indiana
Univ. Math. J. 52 (2003), 569-605.
13. M. Ionescu and P. Muhly Groupoid methods in wavelet analysis, preprint, [arXiv:0709.2294v1].
14. P. Jorgensen Analysis and probability. Wavelets, signals, fractals., Graduate Texts in Mathematics, Springer-Verlag, 2006.
15. T. Katsura, A construction of C*-algebras from C*-correspondences, Contemp. Math. 335
(2003), 173-182.
16. T. Kajiwara and Y. Watatani, C*-algebras associated with self-similar sets, J. Oper. Theory
56 (2006), 225-247.
17. T. Kajiwara and Y. Watatani, KMS states on C*-algebras associated with self-similar sets,
preprint [arXiv:math/0405514v1].
18. M. Pimsner, A class of C*-algebras generalizing both Cuntz-Krieger algebras and crossed
products by Z, Fields Inst. Commun. 12 (1997), 189-212.
19. J. Renault A groupoid approach to C*-algebras, Lecture Notes in Mathematics, vol. 793,
Springer-Verlag, 1980.
matiques, Universit
eans, B.P. 6759, 45067 Orl
eans
D
epartement de Mathe
e dOrl
cedex 2, France
Current address: Instituto de Matem
atica, Universidade Federal do Rio Grande do Sul, Av.
Bento Goncalves, 9500, 91509-900 Porto Alegre, RS - Brazil
E-mail address: gillescastro@gmail.com

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Nonself-adjoint operator algebras for dynamical systems


Kenneth R. Davidson and Elias G. Katsoulis
Abstract. This paper is a survey of our recent work on operator algebras associated to dynamical systems that lead to classication results for the systems
in terms of algebraic invariants of the operator algebras.

1. Introduction
There is a long history of building operator algebras from dynamical systems
going back to von Neumanns construction of a group von Neumann algebra. The
use of nonself-adjoint operator algebras is more recent, but still goes back 40 years
to Arvesons paper [2]. His algebra was closed in the weak operator topology. We
will mostly be interested in norm closed (generally separable) algebras, but these
algebras are the same vintage, beginning with work of Arveson and Josephson [3].
For the most part, we consider a dynamical system to be a locally compact
Hausdor space X together with one or more proper maps i , 1 i n, of X
into itself, which we write as (X, ). A proper map of X into itself induces an
endomorphism (f ) = f of C0 (X). So more generally, we could consider an
arbitrary semigroup of endomorphisms of an arbitrary algebra. This is too general
a setting, but we will consider some variants of this kind. In particular, we will
consider a C* dynamical system (A, ), where A is a C*-algebra with a single
automorphism .
Two systems (X, ) and (Y, ) are said to be conjugate provided that there is a
homeomorphism of X onto Y so that i = i for 1 i n. Similar notions may
dened for endomorphisms of other operator algebras as well. The main question
is whether the operator algebra encodes enough information intrinsically so that
the dynamical system can be recovered up to conjugacy or some similar relation
between dynamical systems.
Our operator algebras can be considered in the abstract, in the spirit of the
BlecherRuanSinclair Theorem [7]. Generally they are dened by a universal
property dealing with representations. So the question of an sucient family of
such representations is moot. See [38, 6] for more on this viewpoint.
2000 Mathematics Subject Classication. Primary 47L65, 46L40.
Key words and phrases. conjugacy algebra, semicrossed product, dynamical system.
First author partially supported by an NSERC grant.
Second author partially supported by a summer grant from ECU.
c Mathematical
0000
(copyright Society
holder)
c
2009
American

1
39

40
2

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

2. Semicrossed Products
Arveson [2] and ArvesonJosephson [3] were the rst to consider nonselfadjoint operator algebras constructed from dynamical systems. They considered
a homeomorphism of a compact Hausdor space X which preserves a probability measure satisfying (O) > 0 for every non-empty open set O X. Given
such a system, one can dene a representation of C(X) on L2 () as multiplication
operators, and dene a unitary U f = f . Let A(X, ) denote the norm-closed
subalgebra of B(L2 ()) generated by U and the multiplication operators. Their
main result is that, under a further technical condition, that the algebraic structure of the algebra determines the dynamical system:
Theorem 2.1 (ArvesonJosephson [3]). Suppose that for i = 1, 2, i is a homeomorphism of a compact Hausdor space Xi which preserves a probability measure
i satisfying i (O) > 0 for every non-empty open set O of Xi . Moreover suppose
that the set of periodic points has measure 0. Then the following are equivalent:
(i) (X1 , 1 ) and (X2 , 2 ) are conjugate.
(ii) A(X1 , 1 ) and A(X2 , 2 ) are isometrically isomorphic.
(iii) A(X1 , 1 ) and A(X2 , 2 ) are algebraically isomorphic.
In 1985, Peters [39] introduced an abstract construction of an operator algebra
associated to the system (X, ). He did not require the map to be a homeomorphism, only that it be a proper map so that it induces an endomorphism (f ) = f
of C0 (X). Moreover, he does not require an invariant measure.
Suppose that X is compact. The idea is to construct a universal operator
algebra which contains C(X) as a C*-subalgebra, that is generated by C(X) and a
single isometry s which encodes the dynamics via the covariance relation
f s = s(f )

for all

f C0 (X).

Indeed, consider a covariant representation of this system as a pair (, S) where


is a -representation of C(X) and S is an isometry satisfying (f )S = S(f ) for
all f C(X). It is not dicult to show that there is a unique operator algebra,
denoted C(X) Z+ , satisfying these properties together with the additional property that for every covariant representation (, S), there is a completely contractive
representation of A Z+ such that |C(X) = and (s) = S. This is called the
semi-crossed product algebra.
In the non-compact case, we do not include s in the algebra. C0 (X) Z+
is the universal algebra generated by C0 (X) and the elements sf for f C0 (X).
The isometry s still belongs to the multiplier algebra however, and the covariance
relations make sense if we multiply on the right by a function g C0 (X). It has
the property that for every covariant representation (, S), there is a completely
contractive representation of A Z+ such that |C(X) = and (sf ) = S(f )
for every f C0 (X).
Every dynamical system (X, ) has a natural family of covariant representations
called orbit representations. For any point x X, build a representation x on 2
by
x (f ) = diag(f (x), f ((x)), f ( 2(x)), . . . )

and x (s) = S+ , the unilateral shift. Peters shows that = xX x is a faithful
completely isometric representation of the semicrossed product C0 (X) Z+ . So
it has a rather concrete spatial representation.

NONSELF-ADJOINT OPERATOR ALGEBRAS

41
3

Peters was able to signicantly extend the ArvesonJosephson Theorem. He


still assumes that X is compact, but works with an arbitrary continuous map of
X into itself, subject to the assumption that there are no xed points.
Theorem 2.2 (Peters [39]). Let (Xi , i ) be compact dynamical systems with
no xed points. Then the following are equivalent:
(i) (X1 , 1 ) and (X2 , 2 ) are conjugate.
(ii) C(X1 )1 Z+ and C(X2 )2 Z+ are completely isometrically isomorphic.
(iii) C(X1 ) 1 Z+ and C(X2 ) 2 Z+ are algebraically isomorphic.
In 1988, Hadwin and Hoover [20] considered a much more general class of
conjugacy algebras which contain C(X) and an element s satisfying the covariance
relations. They do not even require the algebras to be closed. Their methods
weakened the condition on xed points to saying: {x X1 : 12 (x) = 1 (x) = x}
has no interior. With this hypothesis, they reach the same conclusion. Moreover,
they show that the algebraic isomorphism of any single conjugacy algebra for the
system (X1 , 1 ) with any conjugacy algebra for the system (X2 , 2 ) is enough to
imply conjugacy of the systems, and hence completely isometric isomorphism of
the semicrossed products.
Another result along these lines is due to Power [43]. He loosens the condition
on X to be locally compact, but requires the map to be a homeomorphism. The
key advance is that he recovers the ArvesonJosephson conclusions without any
hypothesis on xed points.
In 2006, the authors removed all extraneous hypotheses on the dynamical system to obtain:
Theorem 2.3 (DavidsonKatsoulis [12]). Let Xi be locally compact Hausdor
spaces, and let i be a proper continuous map of Xi into itself. Then the following
are equivalent:
(i) (X1 , 1 ) and (X2 , 2 ) are conjugate.
(ii) C(X1 )1 Z+ and C(X2 )2 Z+ are completely isometrically isomorphic.
(iii) C(X1 ) 1 Z+ and C(X2 ) 2 Z+ are algebraically isomorphic.
We will discuss some of the ideas that go into the proof. Hadwin and Hoover
study the characters of C(X) Z+ . The restriction of a character to C0 (X)
will be a point evaluation map x for some x X. The rest of is determined
by (s) = z. As is contractive, one has |z| 1. Call this character x,z . It is
not dicult to show that x,0 always exists, and that this is the only possibility
if (x) = x. When (x) = x, then x,z exists for all z D. We write M for the
character space, and Mx for those characters which restrict to x on C0 (X).
We introduce the notion of an analytic set in the character space. This is the
image of a continuous non-constant map F of a domain C into M such that
fa (w) = F (w)(a) is analytic on for all a C(X) Z+ . A subset of M is a
maximal analytic set if it is an analytic set which is not properly contained in any
other. A crucial observation is that {x,z : z D} are maximal analytic sets for
every xed point x = (x), and there are no others.
The other important device is the notion of a nest representation. A representation of an algebra into B(H) is said to be a nest representation provided that the
lattice of invariant subspaces of its range is a nest (complete chain) of subspaces.
Nest representations and their kernels were originally introduced by Lamoureux

42
4

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

[29] as a generalization for primitive ideals. The second author, in joint work with
Peters [25] and Kribs [22], applied the concept of a nest representation to the
classication theory for non-selfadjoint operator algebras, including graph algebras
[22] and limit algebras [25]. (See also [46].) For our purposes, nest representations
into the 2 2 upper triangular matrices are sucient.
In our case, such a 2 2 nest representation of C(X) Z+ will have diagonal entries which are characters x,z and y,w , and a 1, 2 entry which is linearly
independent of the diagonal (in order that there be no other invariant subspace).
The dynamical system is recovered by showing that this implies that (x) = y.
A complication occurs when (y) = y. This is the case that forced Hadwin and
Hoover to include an additional hypothesis. We circumvent that by considering an
analytic family of nest representations.
3. Function Algebra Systems
Peters construction was actually developed in greater generality. If A is any
operator algebra and is a completely contractive endomorphism, one denes the
semicrossed product A Z+ in an analogous manner. This is the universal operator
algebra which is generated by a completely isometric copy of A and an isometry s
satisfying the covariance relation
as = s(a) for all

a A.

Given any covariant representation (, S) consisting of a completely contractive


representation of A and an isometry S, there is a completely contractive representation of A Z+ such that |A = and (s) = S. Note that when
A is a C*-algebra, the completely contractive representations are precisely the representations.
There is a natural analogue of the orbit representations for this semicrossed
product. Starting with a completely contractive representation of A on H, form
a representation on 2 (H) by setting
(f ) = diag((a), ((a)), (2(a)), . . . ) and (s) = S+ IH .
Again the direct sum of these representations yields a faithful completely isometric
representation of A Z+ .
To illustrate how the ideas of the previous section can be applied in another
setting, we consider certain function algebras. Let G be a Cauchy domain in C,
namely a bounded open subset such that G consists of a nite union of disjoint
Jordan curves. Let K = G. Then A(K) is the algebra of continuous functions on K
which are analytic on G. Suppose that A(K) is a non-constant function such
that (K) K. Then (f ) = f yields a completely contractive endomorphism
of A(K). Thus we can form the semicrossed product A(K) Z+ .
We say that two such maps i : Ki Ki are analytically conjugate if there is a
biholomorphic map of G1 onto G2 such that 2 = 1 . (Any biholomorphic map
between bounded domains of C extends to a homeomorphism of the closures.)
Example 3.1 (BuskePeters [9]). Let be an elliptic Mobius map of D onto
itself (i.e. has a single xed point, and it lies in the open disk D). Then is
analytically conjugate to to a rotation (z) = e2i z. They show that A(D) Z+
is isometrically isomorphic to the subalgebra Alg{U, V } of the rotation C*-algebra
A . It follows that there is an automorphism interchanging U and V which has the

NONSELF-ADJOINT OPERATOR ALGEBRAS

43
5

eect of ipping the automorphism to 1 . So A(D) Z+ and A(D) 1 Z+


are completely isometrically isomorphic. In particular, the dynamics cannot be
completely recovered from the semicrossed product in this case.
It turns out that this is the only thing which can complicate matters.
Theorem 3.2 (DavidsonKatsoulis [12]). For i = 1, 2, let Gi be Cauchy domains in C with Ki = Gi . Then the following are equivalent:
(i) A(K1 ) 1 Z+ and A(K2 ) 2 Z+ are algebraically isomorphic.
(ii) A(K1 )1 Z+ and A(K2 )2 Z+ are completely isometrically isomorphic.
(iii) (i) 1 and 2 are analytically conjugate, or
(ii) Ki are simply connected, i are homeomorphisms with a unique xed
point which lies in the interior Gi , and 2 is analytically conjugate to
11 .
When one applies the same analysis to A(K), one nds an additional maximal analytic set in the character space, namely {x,0 : x G}. This causes no
diculties when G is not simply connected because it must be identied with the
corresponding non-simply connected analytic set of the other algebra. However
when G is conformally equivalent to the disk, there can be some interplayand
this is precisely what occurs in the BuskePeters example. Only the elliptic case
is complicated, as otherwise the xed point(s) are on the boundary, and topological considerations show that the new analytic disk is uniquely determined. In the
elliptic case, one needs to work harder to show that an algebraic isomorphism is
automatically continuous. This makes use of ideas from a paper of Sinclair [45].
Finally, we mention that the case of = id actually requires a special result
from function theory. In this case, A(K) Z+ A(K D). We use the fact that
any biholomorphic map of one product space onto another actually decomposes as
a product map [30].
4. C*-dynamical Systems
In this section, we will consider the (nonself-adjoint) semicrossed product constructed from a pair (A, ) where A is a C*-algebra and is a -automorphism.
In this case, A Z+ is the subalgebra of the crossed product C*-algebera A Z
generated by A and the unitary U implementing .
Two C*-dynamical systems (A, ) and (B, ) are conjugate if there is a isomorphism of A onto B so that = . In the non-commutative setting, this
is too strong because one can modify by an inner automorphism of A without
aecting the algebra structure of the crossed product. Therefore we say that these
systems are outer conjugate if there is a unitary element u A so that = adu .
Automorphisms of C*-algebras have been widely studied. One useful tool is
the Connes spectrum. Let H (A) denote the set of all hereditary sub-C*-algebras
of A which are -invariant. Dene a subset of the circle T by
() = BH (A) (|B ).
This is in some sense analogous to the Weyl spectrum of a bounded operator. The
following result of Olesen and Pedersen [37] characterizes the important special
case of full Connes spectrum, () = T.
denote the spectrum of A, and let
We write A
be the induced action of on

A by ([])

= [].

44
6

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

Theorem 4.1 (OlesenPedersen [37]). Let A be a separable C*-algebra, and


x an automorphism of A. Then the following are equivalent:
(i) has full Connes spectrum, () = T.
which is -invariant
(ii) there is a dense subset of the spectrum A

on
which
acts freely.
:
(iii) {[] A
n ([]) = []} has no interior for all n 1.
This was used by Muhly and Solel to show the following:
Theorem 4.2 (MuhlySolel [36]). Suppose that (A, ) and (B, ) are C*dynamical systems such that () = T. Then A Z+ and B Z+ are completely
isometrically isomorphic if and only if (A, ) and (B, ) are outer conjugate.
We say that an automorphism is universally weakly inner with respect to
irreducible representations if for every irreducible representation of A, there exists
a unitary W (A) so that (A) = W (A)W . Kishimoto [27] proves the
remarkable result that if A is a simple separable C*-algebra, then every universally
weakly inner automorphism is actually inner.
We were able to avoid the condition on Connes spectrum by modifying the
arguments used in the commutative case.
Theorem 4.3 (DavidsonKatsoulis [14]). Let (A, ) and (B, ) be C*-dynamical systems, and assume that the semicrossed products A Z+ and B Z+ are
completely isometrically isomorphic. Then there is a C*-isomorphism : A B
so that 1 1 is universally weakly inner with respect to irreducible
representations.
Then using Kishimotos result, we obtain that
Corollary 4.4. If A is a separable simple C*-algebra, then A Z+ and
B Z+ are isometrically isomorphic if and only if (A, ) and (B, ) are outer
conjugate.
Our methods yield a new proof of the MuhlySolel theorem. They suggest
that isometric isomorphism of the semicrossed product may imply outer conjugacy
under quite general hypotheses.
5. Multivariable dynamical systems
A multivariable dynamical system is a locally compact Hausdor space X together with a collection 1 , . . . , n of proper continuous maps of X into itself. We
will seek an appropriate analogue of the semicrossed product. To this end, we seek
an operator algebra which contains C0 (X) and operators s1 , . . . , sn satisfying the
covariance relations
f si = si (f i )

for

1 i n and f C0 (X).

Again, in the non-compact case, we do not include the si in the algebra, but do
include the elements si g.
Since we do not require any relations between the maps, the natural semigroup
that arises is the free semigroup F+
n of all words in an alphabet of n letters (including
the empty word). If w = i1 . . . ik is an element of F+
n , we write sw = si1 . . . sik .
Similarly, we write w = i1 ik .

NONSELF-ADJOINT OPERATOR ALGEBRAS

45
7

The issue of norming the elements si leads to a couple of natural choices


that yield dierent algebras. The simplest condition is just to insist that each
si 1. The universal algebra subject to this constraint will be called the semicrossed product, denoted C0 (X) F+
n . This has the universal property that given
(, S1 , . . . , Sn ), where is a -representation of C0 (X) on a Hilbert space H, and
S1 , . . . , Sn are n contractions on H satisfying the covariance relations, then there
is a completely contractive representation of C0 (X) F+
n such that |C0 (X) =
and (si f ) = Si (f ) for 1 i n and f C0 (X).


The other reasonable option is to require that S = S1 . . . Sn be a row
contraction (as an operator from H(n) to H). We call the universal algebra obtained
in this manner the tensor algebra A(X, ).
You may notice that we did not require the operators to be isometries. However
such a requirement would make no dierence. The reason is that there is a dilation
theorem showing that any contractive covariant representation of (X, ) dilates to
one in which each Si is an isometry; and each row contractive covariant representation dilates to a row isometric covariant representation. One obvious advantage
of this formulation is that we now know more about the (completely contractive)
representations of these algebras.
The tensor algebra turns out to be more tractable in general. One reason is
that there is a natural analogue of the orbit representations. Fix x X and build
+
a representation on Fock space 2 (F+
n ), with orthonormal basis {w : w Fn }, by
setting
x (f )w = f (w (x))w
x (si )w = iw

for w F+
n.

As in the one variable case, the direct sum of all orbit representations yields a
faithful, completely isometric representation of A(X, ). The semicrossed product
does not appear to have a nice family of norming representations that can be
explicitly described.
In either case, the universal property leads to the existence of gauge automorphisms, namely automorphisms z for z T of our algebra with z |C0 (X) = id and
z (si f ) = zsi f . In the standard manner, integration over the unit circle yields a
completely contractive expectation onto C0 (X). This leads to an automatic continuity result that any isomorphism of A(X, ) onto A(Y, ) or of C0 (X) F+
n onto
C0 (Y ) F+
is
automatically
norm
continuous.
n
There are no labels on our maps, or on the isometries si . So an isomorphism
can permute the maps arbitrarily. It is less obvious, but still true, that in some
circumstances, one can change from one permutation to another. This leads to our
denition of piecewise conjugate systems. Say that (X, ) is piecewise conjugate to
(Y, ) if there is a homeomorphism : X Y and an open cover {O : Sn } of
X so that
i |O = (i) |O for Sn .
To appreciate this notion, consider two maps 1 and 2 that map [0, 1] into
itself and coincide on an interval (a, b). Then one can construct 1 which agrees
with 1 on [0, b) and with 2 on (a, 1]; and similarly 2 agrees with 2 on [0, b) and
with 1 on (a, 1]. Then ([0, 1], 1 , 2 ) and ([0, 1], 1 , 2 ) are piecewise conjugate. On
the other hand, if the two maps only agree at a point {a}, then one can still dene
the maps 1 and 2 as above, but the new system will not be piecewise conjugate

46
8

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

because there is no neighbourhood of a on which we can match up the two pairs of


functions.
This appears to be a new notion in dynamics. There is a parallel with the full
group introduced by Dye [18] in his analysis of group actions on von Neumann
algebras. One begins with a group {g : g G} of measure preserving automorphisms of a measure space M. Dye considers all automorphisms which are pieced
together by a countable partition of the space into measureable sets Pg such that
g1 is the identity on Pg M. So the notion of mixing and matching maps occurs
here, but in a rather dierent context.
The main result of our paper [13] is the following:
Theorem 5.1 (DavidsonKatsoulis [13]). Let (X, ) and (Y, ) be two multivariable dynamical systems. If there is an algebra isomorphism of A(X, ) onto
+
A(Y, ) or of C0 (X) F+
n onto C0 (Y ) Fn , then (X, ) and (Y, ) are piecewise
conjugate.
The proof follows the ideas of the n = 1 case, but non-trivial complications
arise. In particular, one must be able to count the number of maps in the system
which send a point x to a point y. The key is again an analytic structure on the
set of nest representations. The ability to count the number of maps relies on the
well-known, but non-trivial, fact from several complex variable theory [19] that
the zero set of an analytic function mapping Ck into Cl , for l < k, has no isolated
points.
In the case n = 1, the converse direction was trivial. But here there are dicult
issues about how to intertwine the isometries to form the new ones. This appears
to be possible in the case of the tensor algebra. We have no idea how this could be
accomplished in the semicrossed product case. A partial converse is the following:
Theorem 5.2 (DavidsonKatsoulis). Suppose that at least one of the following
holds:
n 3, or
X has covering dimension at most 1, or
{x : |(x)| < n} has no interior.
Then the following are equivalent:
(i) (X, ) and (Y, ) are piecewise conjugate.
(ii) A(X, ) and A(Y, ) are algebraically isomorphic.
(iii) A(X, ) and A(Y, ) are completely isometrically isomorphic.
We conjecture that the converse holds in complete generality. It was reduced
to a technical conjecture about the existence of a nice map from the polytope
with vertices indexed by Sn into the unitary group Un which takes the vertices
to the corresponding permutation matrix, and satises some strict compatibility
conditions on the various faces. Chris Ramsey, a student at the University of
Waterloo, has been making progress on this conjecture [44]. He has extended the
converse to the case n = 4.
It would be interesting to study the ideal structure of these algebras. Peters
[40] has made progress on this in the case n = 1. A natural test question is
to determine when the algebra is semisimple, and more generally, to identify the
radical. This has been accomplished in the case n = 1 by Donsig, Katavolos and
Manoussos [16], with earlier work by Muhly [31]. We can answer the question
about semisimplicity, but have little to say about the radical.

NONSELF-ADJOINT OPERATOR ALGEBRAS

47
9

Call an open subset U X a (u, v)wandering set if


1
uwv
(U ) U =

for all

w F+
n.

A generalized wandering set is a (u, v)-wandering set for some pair (u, v). If there
are no wandering sets, then necessarily each i is surjective.
Wandering sets have a parallel notion of recurrence. Say that x X is (u, v)
recurrent if for every open set U x, there is some w F+
n so that uwv (x) U . In
the metrizable case, there are no non-empty generalized wandering sets if and only
if the (u, v)recurrent points are dense for all pairs (u, v). This is in turn equivalent
to the surjectivity of each i and the density of the (, v)recurrent points for each
v F+
n.
Theorem 5.3 (DavidsonKatsoulis). The following are equivalent:
(i) A(X, ) is semisimple.
(ii) C0 (X) F+
n is semisimple.
(iii) There are no non-empty generalized wandering sets.
6. C*-envelopes
Arvesons seminal paper [4] proposes that to study a nonself-adjoint operator
algebra, there should be a canonical minimal C*-algebra that contains it (completely isometrically). This C*-algebra, Cenv (A), is the proposed analogue of the
Shilov boundary in the function algebra case, and is called the C*-envelope of A.
Let j0 be the completely isometric imbedding of A into Cenv (A). Then Cenv (A) is
determined by the universal property that whenever j is a completely isometric isomorphism of A into another C*-algebra C (j(A)), there exists a -homomorphism
of C (j(A)) onto Cenv (A) such that j = j0 .
Unlike the other universal constructions mentioned in this paper, it is not at
all apparent that the C*-envelope exists. Arveson constructed it for a large family
of examples, but left the existence in general as a conjecture. This was veried a
decade later by Hamana [21]. A new proof was found a few years ago by Dritschel
and McCullough [17]. Their proof is based on the notion of a maximal dilation.
A representation of A on H is maximal if any dilation of to a completely
contractive representation on a larger space K (meaning that (A) = PH (A)|H )
has the form =  on K = H H . It is not particularly dicult to show that
any completely contractive representation can be dilated to a maximal one. The
point is that maximal representations extend to -representations of the enveloping
C*-algebra of A and factor through the C*-envelope. Some of these ideas were
already known due to work of Muhly and Solel [33]. This new proof provides a
tangible way to get hold of the C*-envelope. One starts with a completely isometric
representation , dilates it to a maximal representation , and Cenv (A) = C ((A)).
In [13], we provide two views of the C*-envelope of the tensor algebra A(X, ).
The rst is a rather abstract approach. Pimsner [42] developed a construction of a
C*-algebra from a C*-correspondence, a Hilbert C*-module with a compatible left
action, now known as the CuntzPimsner algebra of the correspondence. Muhly
and Solel [34, 35] developed an extensive theory of an associated nonself-adjoint
tensor algebra of a C*-correspondence. They show that when the left action is
faithful, the C*-envelope of the tensor algebra is the CuntzPimsner C*-algebra
of the C*-correspondence. Katsura [26] extended this theory, dening the Cuntz
Pimsner algebra for more general left actions which need not be faithful. The

48
10

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

second author and Kribs [24] used Katsuras work to generalize the MuhlySolel
theorem to describe the C*-envelope of the tensor algebra in full generality.
We show explicitly [13] that the tensor algebra of a multivariable dynamical
system is the tensor algebra of a naturally associated C*-correspondence. Consequently, by the results in the previous paragraph, we have a description of the
C*-envelope as a CuntzPimsner algebra. Unfortunately, because this algebra is
a quotient of the CuntzToeplitz algebra by Katsuras ideal, this is not a very
concrete description. We were looking for something more explicit.
The rst step, carried out in [13], is to describe the maximal dilations of the
orbit representations. Notice that if x = i (y), then the orbit representation x
can be obtained as the restriction of y to an invariant subspace. Hence x dilates
to y . One can repeat this procedure, and when the system is surjective, construct
an innite chain of orbit representations, each being a dilation of the previous one.
The inductive limit of this procedure yields a family of maximal representations.
When the system is not surjective, this procedure stops if we arrive at a point y
which is not in the range of any map. It turns out that the orbit representation
of such a point is also maximal. Since the direct sum of all orbit representations
is completely isometric on the tensor algebra, it follows that the direct sum of all
of these maximal dilations is also completely isometric. Hence the C*-envelope is
given as the algebra generated by this large representation. This still is not very
explicit, so we seek to develop this some more.
In the case n = 1 when is surjective and X is compact, this was accomplished
of the system (X, )
by Peters [41]. The idea is to take the projective limit X

X X X . . . X.
There is canonical projection p of Y onto X, and a map
of Y onto itself such that
p
= p. Moreover,
is always a homeomorphism. Consequently, one can form
Z. There is a natural injection of C(X) Z+
the C*-crossed product C(X)
into this algebra by sending f to f p and sending s to the canonical unitary of the
crossed product. Peters shows that this map is a complete isometry, and that the
image generates the crossed product as a C*-algebra. Then with a bit more work,
one obtains:
Theorem 6.1 (Peters [41]). Let X be a compact Hausdor space, and let be

a surjective continuous map of X onto itself. Construct (X,
) as above. Then

Cenv (A(X, )) C(X) Z.


The rst author and Jean Roydor [15] extended Peters construction to the
multivariable
setting. First assume that (X, ) is surjective in the sense that X =
n

(X).
One
can again construct a projective limit system. An innite tail is
i
i=1
an innite sequence i nN , where n = {1, . . . , n); say i = (i0 , i1 , . . . ). One considers
of all pairs (i, x) nN X N such that i (xs+1 ) = xs . There is again
the set X
s


onto X and maps
a natural map p(i, x) = x0 of X
j (i, x) = (j, i), (j (x0 ), x)
that satisfy p
j = j p for 1 j n. These maps are no longer homeomorphisms.
j of
j consists of all points (i, x) such that i0 = j. These
However the range X
onto X
j . The
are pairwise disjoint clopen sets, and
j is a homeomorphism of X
1
j is a local homeomorphism. The tensor algebra
inverse map given by |X j =

A(X, ) imbeds completely isometrically into A(X,
), and they have the same
C*-envelope.

NONSELF-ADJOINT OPERATOR ALGEBRAS

49
11

This leads to a more concrete description of the C*-envelope, because the new
system is much simpler to handle. One description is that this is the groupoid
) in the sense of Deaconu [10]. Another is that it is
C*-algebra of the system (X,
the crossed product of a certain inductive limit B of homogeneous C*-algebras by
an endomorphism .
Theorem 6.2 (DavidsonRoydor [15]). Let X be a locally compact
 Hausdor
space, and let 1 , . . . , n be proper maps of X into itself such that X = ni=1 i (X).

Construct (X,
) as above. Then


) B Z + .
Cenv (A(X, )) Cenv (A(X,
)) C (X,
When (X, ) is not surjective, there is
na well-known technique from graph
algebras of adding a tail. Let U = X \ i=1 (X). Form X T = X T where
T = {(u, k) : u U , k < 0}. Extend i to maps iT by setting
iT (u, k) = (u, k + 1) for k < 1,

and

iT (u, 1) = u.

It is shown that the natural imbedding of A(X, ) into (X T , T ) is a completely


isometric isomorphism. Moreover, the C*-envelope of A(X, ) is a full corner of
Cenv (A(X T , T )).
One consequence is a characterization of when the C*-envelope is simple. When
X is compact, the system (X, ) is called minimal if there are no proper closed invariant subsets of X.
Theorem 6.3 (DavidsonRoydor [15]). Let (X, ) be a compact multivariable
dynamical system. Then Cenv (A(X, )) is simple if and only if (X, ) is minimal.
References
[1] M. Alaimia and J. Peters, Semicrossed products generated by two commuting automorphisms
J. Math. Anal. Appl. 285 (2003), 128140.
[2] W. Arveson, Operator algebras and measure preserving automorphisms, Acta Math. 118,
(1967), 95109.
[3] W. Arveson and K. Josephson, Operator algebras and measure preserving automorphisms II,
J. Funct. Anal. 4 (1969), 100134.
[4] W. Arveson, Subalgebras of C*-algebras, Acta Math. 123 (1969), 141224.
[5] W. Arveson, The noncommutative Choquet boundary, J. Amer. Math. Soc., (2008).
[6] D. Blecher and C. LeMerdy, Operator algebras and their modules-an operator space approach,
Clarendon Press, Oxford, 2004
[7] D. Blecher, Z.-J. Ruan and A. Sinclair, A characterization of operator algebras, J. Funct.
Anal. 89 (1990), 188201.
[8] D. Buske, Hilbert modules over a class of semicrossed products, Proc. Amer. Math. Soc. 129
(2001), 17211726.
[9] D. Buske and J. Peters, Semicrossed products of the disk algebra: contractive representations
and maximal ideals., Pacic J. Math. 185 (1998), 97113.
[10] V. Deaconu, Groupoid associated with endomorphisms, Trans. Amer. Math. Soc. 347 (1995),
17791786.
[11] L. DeAlba and Peters, J. Classication of semicrossed products of nite-dimensional C*algebras, Proc. Amer. Math. Soc. 95 (1985), 557564.
[12] K. Davidson and E. Katsoulis, Isomorphisms between topological conjugacy algebras, J. Reine
Angew. Math. 621 (2008), 2951.
[13] K. Davidson and E. Katsoulis Operator algebras for multivariable dynamics, preprint, 2007.
[14] K.R. Davidson and E.K. Katsoulis, Semicrossed Products of Simple C*-algebras, Math. Annalen 342 (2008), 515525.
[15] K.R. Davidson and J. Roydor, C*-envelopes of tensor algebras for multivariable dynamics,
preprint, 2008.

50
12

K.
R. DAVIDSON AND
AND E.G.
E. G.KATSOULIS
KATSOULIS
K.R.DAVIDSON

[16] A. Donsig, A. Katavolos, A. Manoussos, The Jacobson radical for analytic crossed products,
J. Funct. Anal. 187, (2001), 129145.
[17] M. Dritschel and S. McCullough, Boundary representations for families of representations
of operator algebras and spaces, J. Operator Theory 53 (2005), 159167.
[18] H. Dye, On groups of measure preserving transformations I, Amer. J. Math. 81 (1959),
119159.
[19] K. Fritzsche and H. Grauert, From holomorphic functions to complex manifolds, Graduate
Texts in Mathematics, 213, Springer-Verlag, New York, 2002.
[20] D. Hadwin and T. Hoover, Operator algebras and the conjugacy of transformations, J. Funct.
Anal. 77 (1988), 112122.
[21] M. Hamana, Injective envelopes of operator systems, Publ. Res. Inst. Math. Sci. 15 (1979),
773-785.
[22] E. Katsoulis and D. Kribs, Isomorphisms of algebras associated with directed graphs, Math.
Ann. 330 (2004), 709728.
[23] E. Katsoulis and D. Kribs, The C*-envelope of the tensor algebra of a directed graph, Integral
Equations Operator Theory 56, (2006), 401414.
[24] E. Katsoulis and D. Kribs, Tensor algebras of C*-correspondences and their C*-envelopes,
J. Funct. Anal. 234 (2006), 226233.
[25] E. Katsoulis and J. Peters, Compact operators and nest representations of limit algebras,
Trans. Amer. Math. Soc. 359 (2007), 27212739
[26] T. Katsura, On C*-algebras associated with C*-correspondences, J. Funct. Anal. 217 (2004),
366401.
[27] A. Kishimoto, Outer automorphisms and reduced crossed products of simple C*-algebras,
Comm. Math. Phys. 81 (1981), 429435.
[28] M. Lamoureux, Nest representations and dynamical systems, J. Funct. Anal. 114 (1993),
467492.
[29] M. Lamoureux, Ideals in some continuous nonselfadjoint crossed product algebras. J. Funct.
Anal. 142 (1996), 211248.
[30] E. Ligocka, On proper holomorphic and biholomorphic mappings between product domains,
Bull. Acad. Polon. Sci., Ser. Sci. Math. 28 (1980), 319323.
[31] P. Muhly, Radicals, crossed products, and ows, Ann. Polon. Math. 43 (1983), 3542.
[32] P. Muhly and B. Solel, Hilbert modules over operator algebras, Mem. Amer. Math. Soc. 117,
no. 559, Amer. Math. Soc., Providence, RI, 1995.
[33] P. Muhly and B. Solel, An algebraic characterization of boundary representations, Nonselfadjoint operator algebras, operator theory, and related topics, Oper. Theory Adv. Appl. 104,
189196, Birkhauser, Basel, 1998.
[34] P. Muhly and B. Solel, Tensor algebras over C*-correspondences: representations, dilations
and C*-envelopes, J. Funct. Anal. 158 (1998), 389457.
[35] P. Muhly and B. Solel, Tensor algebras, induced representations, and the Wold decomposition,
Can. J. Math. 51 (1999), 850880.
[36] P. Muhly, B. Solel, On the Morita equivalence of tensor algebras, Proc. London Math. Soc.
(3) 81 (2000), 113168.
[37] D. Olesen and G. Pedersen, Applications of the Connes spectrum to C*-dynamical systems
III, J. Funct. Anal. 45 (1982), 357390.
[38] V. Paulsen, Completely bounded maps and operator algebras, Cambridge Studies in Advanced
Mathematics 78, Cambridge University Press, Cambridge, 2002.
[39] J. Peters, Semicrossed products of C -algebras, J. Funct. Anal. 59 (1984), 498534.
[40] J. Peters, The ideal structure of certain nonselfadjoint operator algebras, Trans. Amer. Math.
Soc. 305 (1988), 333352.
[41] J. Peters, The C*-envelope of a semicrossed product and nest representations, preprint, 2006.
[42] M. Pimsner, A class of C -algebras generalizing both Cuntz-Krieger algebras and crossed
products by Z, Free probability theory (Waterloo, ON, 1995), 189212, Fields Inst. Commun.,
12, Amer. Math. Soc., Providence, RI, 1997.
[43] S. Power, Classication of analytic crossed product algebras, Bull. London Math. Soc. 24
(1992), 368372.
[44] C. Ramsey, Algebraic characterization of multivariable dynamics, Masters thesis, University
of Waterloo, 2009.
[45] A. Sinclair, Homomorphisms from C0 (R), J. London Math. Soc. 11 1975, 165174.

NONSELF-ADJOINT OPERATOR ALGEBRAS

51
13

[46] B. Solel, You can see the arrows in a quiver operator algebra J. Aust. Math. Soc. 77 (2004),
111122.
Pure Math. Dept., U. Waterloo, Waterloo, ON N2L3G1, CANADA
E-mail address: krdavids@uwaterloo.ca
Dept. Mathematics, East Carolina U., Greenville, NC 27858, USA
E-mail address: KatsoulisE@mail.ecu.edu

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Extending representations of normed algebras in Banach


spaces
Sjoerd Dirksen, Marcel de Jeu, and Marten Wortel
Abstract. Let X be a non-degenerate left Banach module over a normed
algebra A having a bounded approximate left identity. We show that, if A is
a left ideal of a larger algebra, then this representation can be extended to a
representation of the larger algebra. Based on this result, we study in detail the
existence and properties of representations of the various centralizer algebras
of A which are compatible with the original representation of A. As a special
case we obtain that, if A embeds as a topological algebra into the bounded
operators on X, then the left centralizer algebra of A embeds as a topological
algebra as the left normalizer of the image, and the double centralizer algebra
of A embeds as a topological algebra as the normalizer of the image. We
also consider ordered and involutive contexts, and cover the right-sided cases,
which are not always the obvious analogues of the left-sided cases, in detail as
well.

1. Introduction
This paper is concerned with the possibility of extending a given Banach representation of an ideal of a normed algebra to the whole algebra, and also, given a
Banach representation of a normed algebra, with the possibility of dening representations of the various centralizer algebras of that algebra which are compatible
with the given representation. These two issues are strongly related.
Our interest in this problem arose from the study of covariant representations
of Banach algebra dynamical systems. In [5] a crossed product Banach algebra is
constructed from a Banach algebra dynamical system and a collection of covariant Banach space representations thereof, and the question is, roughly speaking,
whether all representations of the crossed product are integrated forms of the given
covariant representations of the original dynamical system. For C -dynamical systems and Hilbert representations the answer is armative, see, e.g., [12]. The
standard method to establish the result in that case is to extend a given representation of the C -crossed product to its multiplier algebra, after which the sought
covariant representation of the C -dynamical system can be found with the aid of
the canonical maps of the group and the algebra into the multiplier algebra. In this
2000 Mathematics Subject Classication. Primary 46H25; Secondary 46H10, 46H15.
Key words and phrases. Normed algebra, Banach algebra, centralizer algebra, approximate
identity, representation, Banach module.
c
2009
American
c Mathematical
0000
(copyright Society
holder)

1
53

54
2

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

method, critical use is made of the standard result that a non-degenerate Hilbert
representation of a closed two-sided ideal of a C -algebra extends to the algebra.
For our case we needed a similar result for Banach representations, and with an
upper bound for the norm of the extensions. With future applications in representation theory in Banach lattices in mind we were also interested in such results
which take ordering into account.
Somewhat to our surprise, we were unable to nd the results as we needed them
in the literature. In the presence of a bounded approximate left identity, the seminal
paper on centralizer algebras [8] contains material on Hilbert representations of
centralizer algebras which are compatible with the original representation of the
algebra, see [8, Section 9], but for more general spaces completeness assumptions
on the image are made, cf. [8, Theorem 20]. Likewise, when it is proved in [3,
Theorem 2.9.51] that, in the presence of a bounded two-sided approximate identity,
a Banach bimodule structure extends to the double centralizer algebra, it is assumed
that the algebra is a Banach algebra, in order to be able to use the Cohen-Hewitt
factorization theorem [2, p. 61], [3, Theorem 2.9.24]; the same holds for [9, p. 17].
Such completeness assumptions are, however, not necessary. As it turns out, it is
possible to develop a theory for compatible non-degenerate Banach representations
of centralizer algebras without using the factorization theorem, assuming only that
the original algebra is a normed algebra with a suitable approximate identity. The
key idea is essentially already in the proof of [9, Theorem 21], but it appears that
this has not yet been exploited systematically. This is done in the present paper,
which can perhaps be regarded as a representation-theoretical supplement to the
general material on centralizer algebras collected in [11].
It deserves to be mentioned at this point that, whereas the results in this
paper on the existence of compatible representations of centralizer algebras of a
normed algebra can, by passing to its completion, easily be derived from their
versions for the centralizer algebras of a Banach algebra, this is no longer the case
for our results on embeddings of centralizer algebras. The reason is simply that
the centralizer algebras of a normed algebra may be strictly smaller than those of
its Banach algebra completion. Thus our consideration of normed algebras rather
than Banach algebras does not only make manifest that the factorization theorem
is not needed, but it also avoids being unnecessarily restrictive as to the scope of
the results.
The results on compatible representations of centralizer algebras in this paper
rest on a basic theorem concerning extending a Banach representation of an ideal
of a normed algebra to the algebra itself, cf. the rst part of Theorem 1.1. In
spite of its elementary proof and its general relevance, we have not been able to
nd a reference for this result. It has some bearing even on the well known case
of C -algebras, where, for Hilbert representations, such a results is usually stated
for closed ideals and proved using GNS-theory, cf. [6, Proposition 2.10.4], [10,
Theorem 5.5.1]. There is an alternative approach to be found for the C -case
which uses an approximate identity and which is close to ours, cf. [1, II.6.1.6], [4,
Lemma I.9.14], but still the results in these sources are formulated under hypotheses
which are more stringent than necessary: actually, the ideal needs only to be a left
ideal, not necessarily closed, with a bounded left approximate identity for itself.
Since they nd it necessary to include a proof, the authors of [1] and [4] also seem
to be unaware of a reference for such a general extension result.

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

55
3

As a an illustration of what is implied by the mere presence of a bounded


left approximate identity, we include the following excerpt of Theorem 3.1 and
Theorem 5.3 in this introduction. The notions guring in it will be properly dened
later on, to avoid all possible misunderstanding, but they are the obvious ones. The
representations in the formulation are norm continuous homomorphisms from the
normed algebra in question to the algebra of bounded operators. Note that it is
not assumed that the algebra is complete, nor that the ideal is closed.
Theorem 1.1. Let A be a normed algebra, and let X be a Banach space, with
algebra of bounded operators B(X).
(1) If J is a left ideal of A containing a bounded approximate left identity
for itself, and if : J B(X) is a non-degenerate representation, then
extends uniquely to a representation of A. If, in addition, A is an
ordered algebra, X is an ordered vector space with a closed positive cone,
J contains a positive approximate left identity for itself, and is positive,
then the extended representation is positive. Alternatively, if, in addition,
A has an involution which leaves J invariant, X is a Hilbert space, and
is involutive, then the extended representation is involutive.
(2) Suppose A has a bounded approximate left identity, and : A B(X) is
a non-degenerate faithful representation which is an isomorphism of topological algebras between A and (A). Then, as abstract algebras, A and
its double centralizer algebra M(A) both embed canonically into the left
centralizer algebra Ml (A) of A. After identication one has A M(A)
Ml (A), and extends uniquely to a representation : Ml (A) B(X).
Moreover, is an isomorphism of topological algebras between Ml (A) and
the left normalizer of (A) in B(X), and its restriction to M(A) (with its
own norm) is an isomorphism of topological algebras between M(A) and
the normalizer of (A) in B(X). If, in addition, A is an ordered algebra
and has a positive bounded approximate left identity, X is an ordered vector space with a closed positive cone, and is an isomorphism of ordered
algebras, then the isomorphisms for Ml (A) and M(A) are isomorphisms
of ordered algebras. Alternatively, if, in addition, A has a bounded involution, X is a Hilbert space and is involutive, then M(A) has a bounded
involution, and the isomorphism for M(A) is an isomorphism of involutive algebras.
The relation of the second part of Theorem 1.1 with known results about centralizer algebras of the algebra of compact operators and about double centralizer
algebras of C -algebras is discussed in Remarks 5.6 and 5.7.
For general A, it need not be the case that the canonical homomorphism of
A into Ml (A) is injective, so that one cannot properly speak about extending
representations from A to Ml (A) as in the second part of Theorem 1.1, and the
statements then need to be phrased in terms of commutative diagrams expressing
the compatibility of the original representation of the algebra and the representation
of the centralizer algebra. The theorems in Section 4 contain such results, including
upper bounds for the norms of the various maps. These results are valid for nondegenerate modules which are not necessarily faithful. In fact, the existence of
a faithful non-degenerate module is equivalent to the injectivity of the natural
homomorphism of A into Ml (A), see Proposition 5.1.

56
4

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

When formulating our results, we have attempted to give as precise statements


as possible, under minimal hypotheses. Following the statement for general Banach spaces, we have systematically also covered the ordered and Hilbert contexts.
All this detailed information makes the statements rather long, but this seemed
unavoidable.
We have also covered the case of non-degenerate right modules over a normed
algebra with a bounded right approximate identity. We emphasize that one does
not obtain the results for the right-sided case from the left-sided case by simply
replacing left with right, and homomorphism with anti-homomorphism.
As an example, the canonical image of A in Ml (A) is a left ideal, but the canonical
image in the right centralizer algebra Mr (A) is not, in general, a right ideal: it
is a left ideal. Furthermore, a left module over A becomes a left module over
both Ml (A) and M(A), but a right module over A becomes a left (not: right)
module over Mr (A) and a right module over M(A). To continue, the right-sided
analogue of the embedding result of Ml (A) and M(A) in Theorem 1.1 is that
an anti-embedding of A into B(X) yields an embedding (not: anti-embedding) of
Mr (A) as the left (not: right) normalizer of the image, and an anti-embedding of
M(A) as the normalizer of the image. The obvious adaptations of the results for
the left-sided case to the right-sided case are therefore not the correct ones, and,
although it adds to the length of the paper, it is for this reason that we felt it would
be a disservice to the reader not to include the precise statements for the right-sided
case in full. Thus Theorems 4.1 and 5.3, and Corollary 5.4, are concerned with the
left-sided case, and Theorems 4.3 and 5.8, and Corollary 5.9, are concerned with
the right-sided case. Theorems 3.1 and 4.5, and Proposition 5.1, are concerned
with both cases. Naturally, the proofs for the right-sided case have been omitted,
as they are completely similar to the left-sided case.
Although the lack of symmetry between the left-sided and the right-sided case
may come as a surprise, there is an underlying reason for it: the standard terminology for algebras of linear maps has a left bias. One almost always and certainly
always in this paper considers A to be a left module over B(A), and X to be a
left module over B(X). This asymmetry, baked into the standard terminology, is
what causes the discrepancies later on. If, for an algebra with a bounded right
approximate identity and right A-modules, one would use the opposite algebras of
B(A) and B(X), then the symmetry in the statements would be restored. We felt,
however, that using such formulation would be counterproductive. The whole phenomenon becomes perhaps most obvious in Corollary 5.5, where tradition almost
seems to oppose the mere idea of formulating a right-sided version. The authors,
at least, were in this case content with only the left-sided result.
This paper is organized as follows.
In Section 2 we introduce the basic terminology. This is standard, but including
it makes the paper self-contained and also gives the opportunity to be precise about
conventions concerning unitality, etc. We also include some remarks on a largest
non-degenerate submodule and preservation of the set of invariant closed subspaces
and of the set of intertwining operators.
Section 3 contains the basic result about extending a representation from an
ideal to the algebra.
Section 4 starts with collecting some material on centralizer algebras, and then
proceeds, in the general setting of non-degenerate Banach modules, to develop the

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

57
5

results about the existence and properties of representations of these centralizers


algebras which are compatible with representations of the original algebra.
In Section 5 the results of Section 4 are strengthened when the module is
faithful. If the algebra embeds, then so do the appropriate centralizer algebras.
2. Basic terminology and preliminary remarks
We start by recalling some standard terminology and introducing notation.
Assume that A is a normed algebra over the eld F, where F is either R or C.
We do not require that A is unital, nor that a possible identity element has norm
1, but only that the norm is submultiplicative.
Suppose that X is a normed space over F and that l : A B(X) is a bounded
algebra homomorphism from A into the algebra B(X) of bounded linear operators
on X, thus providing X with the structure of a normed left A-module. We do
not assume that l is unital if A has an identity element. If the span of the
elements l (a)x, for a A and x X, is dense in X, then X is said to be a
non-degenerate normed left A-module, the homomorphism l being understood.
Similarly, if r : A B(X) is a bounded algebra anti-homomorphism, so that X
is a normed right A-module, and if the span of the elements r (a)x, for a A and
x X, is dense in X, then X is said to be a non-degenerate normed right A-module.
If l and r are a bounded algebra homomorphism, resp. a bounded algebra antihomomorphism, such that l (a1 ) and r (a2 ) commute, for all a1 , a2 A, then X is
a normed A-bimodule, which is called a non-degenerate normed A-bimodule if the
span of the elements l (a1 )r (a2 )x, for x X and a1 , a2 A, is dense in X. The
latter density is equivalent to X being both a non-degenerate normed left A-module
and a non-degenerate normed right A-module.
If (Ti )iI is a net in B(X) which converges in the strong operator topology to
T B(X), i.e., if limi Ti x = T x for all x X, then we will write T = SOT- limi Ti .
If m > 0, then an m-bounded approximate left identity for A is a net (ei )iI
in A, such that ei  m, for all i I, and limi ei a a = 0, for all a in A.
Similarly one denes an m-bounded approximate right identity and an m-bounded
two-sided approximate identity and one has obvious notions of bounded left, right,
and two-sided approximate identities.
If V is a vector space over F, then a cone in V is a non-empty subset C such
that 1 c1 + 2 c2 C whenever 1 , 2 0 and c1 , c2 W . Declaring that x y
whenever x y C introduces an ordering in V with the usual properties. Note
that, as in, e.g., [7], we do not assume the properness C (C) = {0} of the positive
cone C, so that the relation need not be anti-symmetric. A map : V1 V2
between two ordered vector spaces is called positive whenever, for all v1 V1 ,
v1 0, implies (v1 ) 0.
An ordered algebra is an algebra which is an ordered vector space with the
additional property that, for all a1 , a2 A, a1 , a2 0 implies a1 a2 0. If X is an
ordered normed space then the set of positive operators in B(X) is a cone, so that
B(X) becomes an ordered normed algebra. If A is an ordered algebra and X is an
ordered normed space which is a normed left A-module via l , we say that it is an
ordered normed left A-module if l is positive, i.e., when positive elements of A act
as positive operators in X. The analogous right-sided notions is obvious.
If F = C, an involution on an algebra A is a map : A A which is a
conjugate linear anti-homomorphism of order 2; if F = R, an involution is a linear

58
6

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

anti-homomorphism of order 2. We note explicitly that, when A is normed, an


involution is not required to be bounded. If A and B are two involutive algebras,
then a map : A B is called involutive if (a ) = (a) , for all a A.
Remark 2.1. Suppose X is a normed left A-module. If A has a bounded
approximate left identity, then there exists a largest non-degenerate normed left
A-submodule Xnd of X. Indeed, let Xnd be the closed linear span of the elements
l (a)x, for a in A and x in X. Surely any non-degenerate normed left A-submodule
is contained in Xnd . Moreover, if (ei )iI is a bounded approximate left identity in
A, then l (ei )iI is a norm bounded subset of B(X) and using this one sees easily
that SOT- limi l (ei ) Xnd = idXnd . In particular, Xnd is a non-degenerate normed
left A-submodule as required. There are obvious right-sided and two-sided versions
for this. In the results below we will repeatedly encounter the assumption that X is
a non-degenerate normed A-module and from the present discussion we see that we
can always pass from X to the largest left, right or two-sided submodule satisfying
this hypothesis.
Remark 2.2. In the subsequent sections, new representations will repeatedly
be dened as SOT-limits of given ones. This implies that, under the new module structure, the set of closed invariant subspaces will remain unchanged. It also
implies that the bounded intertwining operators between two such new representations will coincide with those for the two original representations. For reasons of
space we make this general observation here once and for all, rather than add it on
every separate occasion.
3. Extending from ideals
In this section we establish the basic theorem concerning extension of module
structures initially dened for ideals. It is not necessary that the algebras are
complete or that the ideals are closed, but we do need that the spaces they act
on are complete. In the next section we will apply the results in the context of
centralizer algebras.
Theorem 3.1. Let A be a normed algebra, and let X be a Banach space.
(1) If J is a left ideal in A containing an m-bounded approximate left identity
for itself, and if the homomorphism l : J X provides X with the
structure of a non-degenerate normed left J-module, then there exists a
unique homomorphism l : A B(X) extending l . This extension is,
in fact, bounded with  l  ml , so that X becomes a non-degenerate
normed left A-module. For a A, l (a) B(X) is the unique bounded
operator such that l (a)l (j) = l (aj), for all j J. If (ei )iI is any
bounded approximate left identity in J for itself, then SOT- limi l (ei ) =
idX and, for all a A, l (a) = SOT- limi l (aei ). If a A acts from the
left on J as the identity, then l (a) = idX .
If, in addition, A is an ordered algebra, X is an ordered space with
closed positive cone, J contains a positive bounded approximate left identity for itself, and l is positive, then l is positive.
Alternatively, if, in addition, A has an involution which leaves J invariant (so that J is a two-sided ideal), X is a Hilbert space, and l is
involutive, then l is involutive.

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

59
7

(2) If J is a right ideal in A containing an m-bounded approximate right


identity for itself, and if the anti-homomorphism r : J X provides X
with the structure of a non-degenerate normed right J-module, then there
exists a unique anti-homomorphism r : A B(X) extending r . This
extension is, in fact, bounded with  r  mr , so that X becomes a
non-degenerate normed right A-module. For a A, r (a) B(X) is the
unique bounded operator such that r (a)r (j) = r (ja), for all j J.
If (ei )iI is any bounded approximate right identity in J for itself, then
SOT- limi r (ei ) = idX and, for all a A, r (a) = SOT- limi r (ei a). If
a A acts from the right on J as the identity, then r (a) = idX .
If, in addition, A is an ordered algebra, X is an ordered space with
closed positive cone, J contains a positive bounded approximate right identity for itself, and r is positive, then r is positive.
Alternatively, if, in addition A has an involution which leaves J invariant (so that J is a two-sided ideal), X is a Hilbert space, and r is
involutive, then r is involutive.
(3) If J is a two-sided ideal in A containing both a bounded approximate left
identity and a bounded approximate right identity for itself, and if the homomorphism l : J B(X) and the anti-homomorphism r : J B(X)
provide X with the structure of a non-degenerate normed J-bimodule, then
the maps l and r from the rst two parts make X into a non-degenerate
normed A-bimodule.
Proof. As to the rst part, let (ei )iI be an m-bounded approximate left
identity in J for itself. Fix a A. Now l (aei ) ml a, for all i I,
and using this uniform bound a 3-argument easily implies that the set {x X :
(l (aei )x)iI is a Cauchy net} is a closed linear subspace. Since (ei )iI is an approximate left identity for J, this set clearly contains all elements of the form l (j)y
for y X and j J, hence is equal to X by the non-degeneracy of X as a normed
left J-module. This enables us to dene, for all x X,
(3.1)

l (a)x = lim l (aei )x.


i

It is obvious that l (a) B(X) and that  l (a) ml a. Using once more
the fact that (ei )iI is an approximate left identity for J, as well as the nondegeneracy of X as a normed left J-module, one sees that l extends l and also
that l (a)l (j) = l (aj), for a A and j J. By the non-degeneracy as a normed
left J-module, the latter equation determines l (a) uniquely as an element of B(X).
Since, for a, b A and j J, one has l (a)l (b)l (j) = l (a)l (bj) = l (abj) =
l (ab)l (j), we conclude from the non-degeneracy as a normed left J-module that
l is a homomorphism. This establishes the statements in the rst part of the
proposition regarding general X.
In the involutive context, for x, y H and a A, j J, we compute that
 l (a)l (j)x, y = l (aj)x, y = x, l (aj) y = x, l ((aj) )y
= x, l (j a )y = x, l (j a )y = x, l (j ) l (a )y
= x, l (j ) l (a )y = l (j)x, l (a )y =  l (a ) l (j)x, y .
Hence, by the non-degeneracy of X as a normed left J-module again, l (a) =
l (a ) , so that l is involutive.

60
8

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

The statement in the ordered context is clear from (3.1) and the fact that the
positive cone is closed.
The proof of the rst part is now complete and the proof of the second part
is similar. The third part follows from the rst two parts since, for all a1 A,
resp. for all a2 A, the operator l (a1 ), resp. r (a2 ), is an element of the strong

operator closure of l (J), resp. r (J), and hence these operators commute.
Remark 3.2.
(1) Note that in the rst and second part it is not required that the involution
is bounded.
(2) If A as in Theorem 3.1 is a Banach algebra, then in the rst part one
does not have to require that (a) is a bounded linear operator, because
as a consequence of the Cohen-Hewitt factorization theorem [2, p. 61],
[3, Theorem 2.9.24] the requirement l (a)l (j) = r (aj), for all j J,
already determines r (a) as a map from X into itself. Linearity and
boundedness are then automatic. A similar remark applies to the second
part.
4. Module structures for centralizer algebras: general case
In this section we are concerned with the possibility of nding module structures
for centralizer algebras of an algebra A which are compatible with a given module
structure for A. We cannot directly apply Theorem 3.1 because, although A maps
canonically onto an ideal in its various centralizer algebras, such maps need not be
injective. However, for non-degenerate modules the initial (anti)-representations of
A does, in fact, descend to the images in the centralizer algebras, and subsequentely
Theorem 3.1 can be applied to that situation.
We start with the necessary preparations. Suppose A is a normed algebra.
Let Ml (A) = {L B(A) : L(ab) = L(a)b for all a, b A} be the left centralizer
algebra of A. It is sometimes called the right centralizer algebra, which is perhaps
more logical since the operators in Ml (A) commute with (i.e.: centralize) all right
multiplications rather than the left ones, but we adhere to Johnsons choice of
terminology in his seminal paper [8]. Likewise, Mr (A) = {R B(A) : R(ab) =
aR(b) for all a, b A} is the right centralizer algebra of A. A pair (L, R) with
L Ml (A) and R Mr (A) is called a double centralizer if aL(b) = R(a)b for
all a, b A. Clearly Ml (A) and Mr (A) are unital closed subalgebras of B(A).
Dening (L1 , R1 ) (L2 , R2 ) = (L1 L2 , R2 R1 ) makes M(A) into a unital algebra
over F, which becomes a normed algebra if one puts (L, R) = max(L, R).
If : A A is a bounded involution and L Ml (A), then the map L :
A A dened by L a = (L(a )) is a right centralizer. This yields a bounded
unital homomorphism : Ml (A) Mr (A) (which is conjugate linear if F = C),
inverse to the similarly dened bounded unital homomorphism : Mr (A) Ml (A)
(which is conjugate linear if F = C). Combining these yields a bounded involution

: M(A) M(A), dened as (L, R) = (R , L ). Thus M(A) is a unital normed


algebra with bounded involution.
If A is an ordered algebra, then so are Ml (A) and Mr (A). Furthermore, M(A)
then also becomes an ordered algebra by dening (L, R) M(A) to be positive if
L 0 and R 0.

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

61
9

There is a canonical contractive homomorphism : A Ml (A), dened by


(a)b = ab, for a, b A. Since L (a) = (L(a)), for a A and L Ml (A),
(A) is a left ideal in Ml (A). If A has an (m-bounded) approximate left identity
(ei )iI , then ((ei ))iI is an (m-bounded) approximate left identity in (A). If A
is an ordered algebra, then is positive, and positive approximate left identities in
A yield positive approximate left identities in (A).
Likewise, we have a canonical contractive anti-homomorphism : A Mr (A),
dened by (a)b = ba, for a, b A, and since R (a) = (R(a)), for all a A and
R Mr (A), the image (A) is a left ideal in Mr (A). If A has an (m-bounded)
approximate right identity (ei )iI , then ((ei ))iI is an (m-bounded) approximate
left identity in (A). If A is an ordered algebra, then is positive, and positive
approximate right identities in A yield positive approximate left identities in (A).
The map : A M(A) which is dened, for a A, by (a) = ((a), (a)),
is a contractive homomorphism. For a A and (L, R) M(A) one computes
that (L, R)((a), (a)) = ((L(a)), ((L(a))) and similarly ((a), (a))(L, R) =
((R(a)), (R(a))); hence the image (A) is a two-sided ideal in M(A). If A has
an (m-bounded) approximate left, resp. right, identity (ei )iI , then ((ei ))iI is
an (m-bounded) left, resp. right, approximate identity in (A). If A is an ordered
algebra, then is positive, and positive left, resp. right, approximate identities in A
yield positive left, resp. right, approximate identities in (A). If A has a bounded
involution, then, for a A, one has (a) = (a ) and (a) = (a ). In that case
: A M(A) is a contractive involutive homomorphism.
Retaining the L-part in (L, R) gives a unital contractive homomorphism l :
M(A) Ml (A) which maps (A) onto (A), and retaining the R-part yields a
unital contractive anti-homomorphism r : M(A) Mr (A) which maps (A) onto
(A). If A is an ordered algebra, then both l : M(A) Ml (A) and r : M(A)
Mr (A) are positive.
Suppose that the homomorphism l : A B(X) provides the normed space
X with the structure of a non-degenerate normed left A-module. From the nondegeneracy it is clear that l (Ker ) = 0, hence there is a unique map
l : (A)
B(X) such that
(4.1)

/ B(X)
w;
w
ww

w
w
 ww l
(A)
A

is commutative; it is in fact a homomorphism. If A has an m-bounded approximate left identity, then


l is a bounded homomorphism making X into a nondegenerate normed left (A)-module, and 
l  ml . To see this, let (ei )iI be
such an m-bounded approximate left identity. We know from Theorem 3.1 that
SOT- limi l (ei ) = idX . Hence, for a A and x X, one has
l ((a))x =
l (a)x = limi l (a)l (ei )x = limi l (aei )x = limi l ((a)ei )x. Since we have
l ((a)ei )x l (a)ei x ml (a)x, the conclusion follows. If
A is an ordered algebra which has a positive bounded approximate left identity, if
X is ordered with a closed positive cone and l is positive, then
l is positive. This
follows from the equation
l ((a))x = limi l ((a)ei )x derived above.
Likewise, if the anti-homomorphism r : A B(X) makes X into a nondegenerate normed right A-module, then there is a unique map
r : (A) B(X)

62
10

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

such that
(4.2)


(A)

/ B(X)
x;
xx
x
xx
xx r
r

is commutative; it is in fact a homomorphism. If A has an m-bounded approximate


right identity, then
r is a bounded homomorphism making X into a non-degenerate
normed left (A)-module, and 
r  mr . If A is an ordered algebra which has
a positive bounded approximate right identity, if X is ordered with a closed positive
cone and r is positive, then
r is positive.
We are now in the position to apply Theorem 3.1 to centralizer algebras. The
left-sided version is as follows.
Theorem 4.1. Let A be a normed algebra with an m-bounded approximate left
identity, and let X be a Banach space.
If l : A B(X) provides X with the structure of a non-degenerate normed
left A-module, then there exist a unique map
l and a unique homomorphism l :
Ml (A) B(X) such that the diagram
(4.3)

/ B(X)
v: O
v
v
vv
l

vv l
v
 v
/ Ml (A)
(A) 
A

is commutative. All maps in the diagram are bounded homomorphisms, and l is


unital. One has  1, i = 1 if (A) = 0, 
l  ml , and  l  ml .
In particular, X becomes a non-degenerate normed left Ml (A)-module.
The image l (A) is a left ideal in l (Ml (A)). In fact, if L Ml (A) and a A,
then l (L)l (a) = l (L(a)).
If (ei )iI is any bounded approximate left identity for A, then SOT- limi l (ei ) =
idX , and, if L Ml (A), then l (L) = SOT- limi l (L(ei )).
If, in addition, A is an ordered algebra with a positive bounded approximate left
identity, if X is ordered with a closed positive cone, and if l is positive, then all
algebras in the diagram are ordered and all maps are positive.
Proof. We know from the discussion of centralizer algebras that (A) is a left
ideal in Ml (A), and that ((ei ))iI is an (m-bounded)(positive) approximate left
identity in (A) if (ei )iI is an (m-bounded) (positive) approximate left identity in
A. Furthermore, in the results surrounding diagram (4.1) we have already observed
that
l : (A) B(X) is the unique map making the upper triangle commutative,
and that it is in fact a bounded homomorphism with 
l  ml .
Hence the rst part of Theorem 3.1 applies to this situation and it provides the
unique homomorphism l making the lower triangle commutative. It also yields
that, for any bounded approximate left identity (ei )iI in A, and, for all L Ml (A)
and x X, l (L)x = limi
l (L (ei ))x = limi
l ((L(ei )))x = limi l (L(ei ))x.
It also shows that l is bounded and that  l  m
l  m2 r , but this
is not optimal: choosing an m-bounded approximate left identity (ei )iI for A
one sees immediately from l (L)x = limi l (L(ei ))x that, in fact,  l  ml .

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

63
11

Since (A) is a left ideal in Ml (A), the same holds for the images l ((A)) =
l (A) and l (Ml (A)). In fact, for L Ml (A) and a A, we have l (L)l (a) =
l (L) l ((a)) = l (L (a)) = l ((L(a))) = l (L(a)).
The remaining statements are either clear or follow from the rst part of Theorem 3.1.

Remark 4.2. It is also true that l : Ml (A) B(X) is the unique homomorphism making the square in diagram (4.3) commutative. Indeed, for a homomorphism with this property one sees that, for a A, x X, and L Ml (A),
l (L)l (a)x = l (L) l ((a))x = l (L (a))x = l ((L(a)))x = l (L(a))x. Thus
l (L) is uniquely determined, as a consequence of the non-degeneracy of X.
The right-sided version of Theorem 4.1 is not obtained by replacing left with
right and homomorphism with anti-homomorphism. Instead, it reads as follows.
Theorem 4.3. Let A be a normed algebra with an m-bounded approximate right
identity, and let X be a Banach space.
If r : A B(X) provides X with the structure of a non-degenerate normed
right A-module, then there exist a unique map
r and a unique homomorphism
r : Mr (A) B(X) such that the diagram
(4.4)

/ B(X)
v: O
v
v

vv
r
vv r
v
 v
/ Mr (A)
(A) 
A

is commutative. Then r and are bounded anti-homomorphisms,


r , i and r are
bounded homomorphisms, and r is unital. One has  1, i = 1 if (A) = 0,

r  mr , and  r  mr . In particular, X becomes a non-degenerate
normed left Mr (A)-module.
The image r (A) is a left ideal in r (Mr (A)). In fact, if R Mr (A) and
a A, then r (R)r (a) = r (R(a)).
If (ei )iI is any bounded approximateright identity for A, then SOT-limi r (ei ) =
idX , and, if R Mr (A), then r (R) = SOT- limi r (R(ei )).
If, in addition, A is an ordered algebra with a positive bounded approximate
right identity, if X is ordered with a closed positive cone, and if r is positive, then
all algebras in the diagram are ordered and all maps are positive.
Remark 4.4. Analogously to Remark 4.2, r : Mr (A) B(X) is the unique
homomorphism making the square in diagram (4.4) commutative, as it must satisfy
r (R)r (a)x = r (R(a))x, for all R Mr (A), a A, and x X.
Proof. We know from the discussion of centralizer algebras that (A) is a left
ideal in Mr (A), and that ((ei ))iI is an (m-bounded)(positive) approximate left
identity in (A) if (ei )iI is an (m-bounded) (positive) approximate right identity in
A. Furthermore, in the results surrounding diagram (4.2) we have already observed
that
r : (A) B(X) is the unique map making the upper triangle commutative,
and that it is in fact a bounded homomorphism with 
r  mr . Hence the
rst part of Theorem 3.1 applies to this situation, and the proof is completed as
for Theorem 4.1.

Next we turn to the associated M(A)-modules.

64
12

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

Recall the unital contractive homomorphism l : M(A) Ml (A), dened by


retaining the L-part, and the unital contractive anti-homomorphism r : M(A)
Mr (A), dened by retaining the R-part. As we have seen in Theorem 4.1, under
suitable hypotheses a left A-module X becomes a left Ml (A)-module through a
homomorphism l : Ml (A) B(X). Hence X will become a left M(A)-module
through the homomorphism l l : M(A) B(X). Likewise, from Theorem 4.3
we see that under suitable hypotheses a right A-module will become a right M(A)module through the anti-homomorphism r r : M(A) B(X) (recall that
r : Mr (A) B(X) is a homomorphism. The details are contained in the following
result. The maps
l and
r guring in the statements are again dened in the
diagrams (4.1) and (4.2), and l , resp. r , is the restriction of l , resp. r , to (A).
Since M(A) is an involutive algebra when A has a bounded involution, there are
now also statements on Hilbert representations included, a new feature compared
with Theorems 4.1 and 4.3.
Theorem 4.5. Let A be a normed algebra, and let X be a Banach space.
(1) If A has an m-bounded approximate left identity, and l : A B(X)
provides X with the structure of a non-degenerate normed left A-module,
then
l l : (A) B(X) is the unique map and l l : M(A) B(X)
is the unique homomorphism such that the diagram
(4.5)

/ B(X)
w; O
w
w
w
w
l l

ww
 w w l l
/ M(A)
(A) 
A

is commutative. All maps in the diagram are bounded homomorphisms,


and l l is unital. One has  1, i = 1 if (A) = 0, 
l l 
ml , and  l l  ml . In particular, X becomes a non-degenerate
normed left M(A)-module.
The image l (A) is a two-sided ideal in ( l l )(M(A)). In fact, if
(L, R) M(A) and a A, then ( l l )((L, R))l (a) = l (L)l (a) =
l (L(a)), and l (a)( l l )((L, R)) = l (a) l (L) = l (R(a)).
If (ei )iI is any bounded approximate left identity for A, and (L, R)
M(A), then ( l l )((L, R)) = SOT- limi l (L(ei )).
If, in addition, A is an ordered algebra with a positive bounded approximate left identity, if X is ordered with a closed positive cone, and if
l is positive, then all algebras in the diagram are ordered and all maps
are positive.
Alternatively, if, in addition, A has a bounded involution, X is a
Hilbert space and l is involutive, then all algebras in the diagram have a
bounded involution and all maps are involutive.
(2) If A has an m-bounded approximate right identity, and r : A B(X)
provides X with the structure of a non-degenerate normed right A-module,
then
r r : (A) B(X) is the unique map and r r : M(A) B(X)
is the unique anti-homomorphism such that the diagram

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

(4.6)

65
13

/ B(X)
w; O
w
w
w
w
r r

ww
 w w r r

/ M(A)
(A)
A

is commutative. Then the maps r ,


r r and r r are bounded antihomomorphisms, and i are bounded homomorphisms, and r r is
unital. One has  1, i = 1 if (A) = 0, 
r r  mr , and
 r r  mr . In particular, X becomes a non-degenerate normed
right M(A)-module.
The image r (A) is a two-sided ideal in ( r r )(M(A)). In fact, if
(L, R) M(A) and a A, then ( r r )((L, R))r (a) = r (R)r (a) =
r (R(a)), and r (a)(r r )((L, R)) = r (a) r (R) = r (L(a)).
If (ei )iI is any bounded approximate right identity for A, and (L, R)
M(A), then ( r r )((L, R)) = SOT- limi r (R(ei )).
If A is an ordered algebra with a positive bounded approximate right
identity, if X is ordered with a closed positive cone, and if r is positive,
then all algebras in the diagram are ordered and all maps are positive.
Alternatively, if, in addition, A has a bounded involution, X is a
Hilbert space and r is involutive, then all algebras in the diagram have a
bounded involution and all maps are involutive.
(3) If both the left-sided and the right-sided hypotheses apply, then l l and
r r provide X with the structure of a non-degenerate normed M(A)bimodule.
Proof. As to the rst part, from the surjectivity of there is at most one diagonal map making the upper triangle commutative, and we see from the discussion
surrounding (4.1) that the bounded homomorphism
l l : (A) B(X) has this
property. Since l is contractive, this discussion also shows that 
l l  ml .
Furthermore, we know from the discussion of centralizer algebras that (A) is
a two-sided ideal in M(A), and that ((ei ))iI is an (m-bounded)(positive) approximate left identity in (A) if (ei )iI is an (m-bounded) (positive) approximate left identity in A. We are now once more in the situation of the rst
part of Theorem 3.1, and we conclude that there is at most one homomorphism
from M(A) into B(X) making the lower triangle commutative, and obviously
l l meets this requirement. Certainly  l l  ml . Furthermore, the
strong limit in the rst part of Theorem 3.1 translates into ( l l )((L, R)) =
SOT- limi (
l l )((L, R)((ei ), (ei )) = SOT- limi (
l l )(((L(ei )), (R(ei )))) =
SOT- limi (
l )(L(ei )) = SOT- limi l (L(ei )).
All non-involutive statements in the rst part are now clear, except the claim
that l (a)(l l )((L, R)) = l (a) l (L) = l (R(a)). The rst equality holds by
the denition of l . As to the second we compute, for b A and x X, that
l (a)l (L)l (b)x = l (a)l (L(b))x = l (aL(b))x = l (R(a)b)x = l (R(a))l (b)x.
Hence the second equality follows from the non-degeneracy X as a left A-module.
Turning to the involutive case in the rst part, we note that (A) is a twosided ideal of M(A) which is invariant under the involution of M(A). In fact,
((a), (a)) = ((a ), (a )). Together with the fact that l is involutive this

66
14

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

implies that
l l is involutive, and then the rst part of Theorem 3.1 asserts that

l l is involutive. The proof of the rst part is now complete.


As to the second part, it is again clear that the diagonal map must be the
r r  mr . Thus
anti-homomorphism
r r : (A) B(X), and that 
X becomes a non-degenerate normed right (A)-module. Furthermore, we know
from the discussion of centralizer algebras that (A) is a two-sided ideal in M(A),
and that ((ei ))iI is an (m-bounded)(positive) approximate right identity in (A)
if (ei )iI is an (m-bounded) (positive) approximate right identity in A. Therefore
the second part of Theorem 3.1 applies and shows that there is a unique antihomomorphism from M(A) into B(X) making the lower triangle commutative.
Obviously r r has this property. The rest of the second part is then proved
analogously to the rst part.
The third part is clear.

Remark 4.6. As in Remarks 4.2 and 4.4, l l is the unique homomorphism
making the square in diagram (4.5) commutative. Indeed, if l : M(A) B(X) is
such a homomorphism, then one sees easily that, for (L, R) M(A), a A, and
x X, one must have l ((L, R))l (a)x = l (L(a))x. Likewise, if r : M(A)
B(X) is an anti-homomorphism making the square in diagram (4.6) commutative,
then it is determined by the requirement that r ((L, R))r (a)x = r (R(a))x, for
all (L, R) M(A), a A, and x X.
5. Module structures for centralizer algebras: faithful case
We will now consider normed modules which are not only non-degenerate, but
also faithful. In that case, the associated modules for centralizer algebras are also
faithful. If the faithful module is a topological (anti-)embedding of the original
algebra, then the same holds for the centralizer algebras, which (anti-)embed as
appropriate normalizers of the image of the algebra. The details follow. As a
preparation, we show that the mere existence of a non-degenerate faithful module
is strongly related to the injectivity of various maps between the algebra and its
centralizer algebras.
Proposition 5.1. Let A be a normed algebra.
(1) If A has a left approximate identity, then the following are equivalent:
(a) There exist a normed space X and an injective homomorphism l :
A B(X) providing X with the structure of a non-degenerate faithful
normed left A-module;
l

(b) In A M(A) Ml (A), the canonical maps and l are both


injective homomorphisms;
(c) The canonical map : A Ml (A) is an injective homomorphism.
(2) If A has a right approximate identity, then the following are equivalent:
(a) There exist a normed space X and an injective anti-homomorphism
r : A B(X) providing X with the structure of a non-degenerate
faithful normed right A-module;
r

(b) In A M(A) Mr (A), the canonical map , resp. r , is an


injective homomorphism, resp. an injective anti-homomorphism;
(c) Thecanonical map : A Mr (A) is an injective anti-homomorphism.
Proof. We prove only the rst part, the second being proved similarly.

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

67
15

Suppose that X is a non-degenerate faithful left A-module. Let (L, R) M(A)


and suppose l ((L, R)) = 0, i.e., L = 0. Then, for a, b A, x X, and (L, R)
M(A), we have l (R(a))l (b)x = l (R(a)b)x = l (aL(b))x. Hence, if L = 0, then
l (R(a)) = 0 by non-degeneracy, implying R(a) = 0 by the injectivity of l . Hence
R = 0, and l is injective. If a A and (a) = 0, then certainly (a) = 0. In
that case, for a, b A, and x X, we have l (a)l (b)x = l ((a)b)x = 0. By
non-degeneracy, l (a) = 0, so that a = 0 by the injectivity of l . This shows that
(a) implies (b).
Since = l , it is trivial that (b) implies (c).
Assuming (c), it is sucient to take X = A and l = : A B(A). The
module is faithful by assumption, and non-degenerate by the existence of a left
approximate identity.

Remark 5.2. Note that is not assumed that the ones-sided approximate identity is bounded. Also, it is remarkable how little is needed to show that (a) implies
(b) (and hence trivially also (c)). In the left-sided case, if A is an abstract algebra,
X is a topological vector space, and l : A B(X) is an injective homomorphism
from A into the continuous linear maps from X into itself such that the elements
l (a)x, for a A and x X, span a dense subspace of X, then, by the same proof,
and l are injective, where M(A) and Ml (A) are then dened purely algebraically.
Before stating the left-sided version of the main result on embedding of centralizer algebras, we introduce the necessary notation. If X is a normed space, and S
B(X), then let Nl (S, B(X)) = {T B(X) : T S S for all S S}, Nr (S, B(X)) =
{T B(X) : ST S for all S S}, and N (S, B(X)) = Nl (S, B(X))Nr (S, B(X)).
If S is a subalgebra of B(X), then Nl (S, B(X)), resp. Nr (S, B(X)), resp.N (S, B(X))
carries two natural norms: the norm from B(X) and the norm from Ml (S), resp.
Mr (S), resp. M(S).
Remarkably enough, if X is a normed space, A is a normed algebra with a
bounded left, resp. right, approximate identity, and if : A B(X) provides X
with the structure of a non-degenerate left, resp. right, A-module, then, in both the
left-sided and right-sided case, the two canonical norms on each of Nl ((A), B(X)),
Nr ((A), B(X)), and N ((A), B(X)), are equivalent. Introducing some notation
to make this precise, if T Nl ((A), B(X)), we let (T ) denote the norm of
the left multiplication with T as an element of Ml ((A)). Clearly (T ) T .
Similarly, if T Nr ((A), B(X)), we let (T ) denote the norm of the right
multiplication with T as an element of Mr ((A)), and clearly (T ) T . If
T N ((A), B(X)), let (T ) = max((T ), (T )), so that (T ) T .
To see the equivalences, let (ei )iI be an m-bounded left, resp. right, approximate
identity for A. As observed in Remark 2.1, SOT limi (ei ) = idX in both the leftsided and right-sided case. Hence, if T Nl ((A), B(X)), and x X, then T x =
limi T (ei )x. Since T (ei ) (T )(ei ), we see that T  m(T ).
Likewise, T Nr ((A), B(X)), and x X, then T x = limi (ei )T x, which implies
that T  m(T ). Finally, if T N ((A), B(X)), then clearly T 
m(T ).
Theorem 5.3. Let A be a normed algebra with an m-bounded approximate left
identity, and let X be a Banach space. Suppose that l : A B(X) provides X with
the structure of a non-degenerate faithful normed left A-module. Then all maps in

68
16

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

the diagram
(5.1)



 

(A)

/ B(X)
O

?
/ Ml (A)

from Theorem 4.1 are injective homomorphisms. The canonical homomorphism


l : M(A) Ml (A) is also injective.
Suppose, for the remainder of this Theorem, that l : A l (A) has a bounded
inverse l1 : l (A) A.
If Nl (l (A), B(X)) carries the norm from either B(X) or Ml (l (A)), then l is
a bounded algebra isomorphism between Ml (A) and Nl (l (A), B(X)), with  l 
: Nl (l (A), B(X)) Ml (A) is also
ml  in both cases, and the inverse map 1
l
1





in
both
cases.
bounded, with  1
l
l
l
Likewise, if N (l (A), B(X)) carries the norm from either B(X) or M(l (A)),
then l l : M(A) B(X) yields a bounded algebra isomorphism between M(A)
and N (l (A), B(X)), with  l l  ml  in both cases, and the inverse ( l
l )1 : N (l (A), B(X)) M(A) is also bounded, with ( l l )1  l l1 
in both cases.
If, in addition, A is an ordered normed algebra with a positive bounded approximate left identity, X is ordered with a closed positive cone, and l is an isomorphism of ordered algebras between A and l (A), then l : Ml (A) B(X)
yields an isomorphism of ordered algebras between Ml (A) and Nl (l (A), B(X)),
and l l : M(A) B(X) yields an isomorphism of ordered algebras between
M(A) and N (l (A), B(X)).
Alternatively, if, in addition, A has a bounded involution, X is a Hilbert space
and l is involutive, then l l : M(A) B(X) yields an isomorphism of involutive
algebras between M(A) and N (l (A), (X)).
Proof. By Proposition 5.1, and l are injective. From Theorem 4.1 we know
that l (L)l (a) = l (L(a)), for all L Ml (A) and a A, so that the injectivity of
l implies that l is injective.
For the remainder assume, then, that l1 : l (A) A is bounded.
As observed in Theorem 4.1, it follows from l (L)l (a) = l (L(a)), for all
L Ml (A) and a A, that l maps Ml (A) into Nl (l (A), B(X)). We know from
Theorem 4.1 that  l  ml  if Nl (l (A), B(X)) carries the norm from B(X),
and hence this upper bound also holds when it carries the norm from Ml (l (A)).
:
To construct the inverse of l we dene, in an anticipating notation, the map 1
l
1
(T
)(a)
=

(T

(a)),
for
T

N
(
(A),
B(X))
Nl (l (A), B(X)) B(A) as 1
l
l l
l
l
is a bounded homomorand a A. Clearly, for both norms on Nl (l (A)), 1
l
1
phism, and  1
l  l l . For a, b A, and T Nl (l (A), B(X)), we have
1
1
l (T )(ab) = l (T l (ab)) = l1 ((T l (a))l (b)) = l1 (T l (a))b = 1
l (T )(a)b,
1
(T
)
is
a
left
centralizer.
Hence,
in
fact,

maps
N
(
(A),
B(X))
into
so that 1
l
l
l
l
Ml (A). For T Nl (l (A), B(X)), a A, and x X, we have ( l 1
(T
))
(a)x
=
l
l
1
1
1
l ( 1
(T
))
((a))x
=

(
(T
)(a))x
=

((
(T
)(a)))x
=

(
(T
)a)x
=
l
l
l
l l
l
l
l
T l (a)x. Since X is non-degenerate, we conclude that l 1
l (T ) = T . Further l )(L)(a) = l1 ( l (L)l (a)) =
more, if L Ml (A), and a A, then ( 1
l
1
1
l (l (L(a))) = L(a). Hence ( l l )(L) = L. This concludes the proof of the

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

69
17

statement that l is a topological embedding of Ml (A) in B(X) and of the upper


bound for the norm of its inverse, for both norms on Nl (l (A)).
Now we turn to the statements on M(A). We have already observed in the
rst part of Theorem 4.5 that l l maps M(A) into N (l (A), B(X)). To construct its inverse, we dene the additional map l : Nr (l (A), B(X)) B(A)
as l (T )(a) = l1 (l (a)T ), for T Nr (l (A), B(X)), and a A. Clearly l
is a bounded anti-homomorphism, and l  l l1 , for both norms on
Nr (r (A), B(X)). For a, b A and T Nr (l (A), B(X)) we have l (T )(ab) =
l1 (l (ab)T ) = l1 (l (a)(l (b)T )) = al (T )(b). Hence l (T ) is a right centralizer and we have, in fact, a bounded anti-homomorphism l : Nr (l (A), B(X))
Mr (A).
Suppose now that T N (l (A), B(X)). Then the pair ( 1
l (T ), l (T )) is a dou(T
)

M
ble centralizer. Indeed, we already know that 1
l (A) and that l (T )
l
1
Mr (A), and furthermore, for a, b A, we have a( l (T )(b)) = al1 (T l (b)) =
l1 (l (a)T l (b)) = l1 (l (a)T )b = (l (T )(a))b. Thus we obtain a map l :
N (l (A), B(X)) M(A) which is dened, for T N (l (A), B(X)), by l (T ) =
1
( 1
l (T ), l (T )). Clearly l is a bounded homomorphism, and l  l l 
1
for both norms on N (l (A), B(X)), since both l and l satisfy this estimate in
two cases.
We proceed by showing that l : N (l (A), B(X)) M(A) and l l :
M(A) N (l (A), B(X)) are inverse to each other. It is immediate from the
denitions that ( l l ) l is the identity on N (l (A), B(X)). In the other direction, let (L, R) M(A). Then (l ( l l ))((L, R)) = (L, l ( l (L))). Now,
for a A, l ( l (L))(a) = l1 (l (a) l (L)). Hence l (l ( l (L))(a)) = l (a) l (L).
On the other hand, we had already observed in the rst part of Theorem 4.5 that
l (a)l (L) = l (R(a)). By the injectivity of l we conclude that l ( l (L))(a) =
R(a), and hence l ( l l ) is the identity on M(A). This concludes the proof of
the statements concerning M(A).
We now turn to the ordered situation. We know from Theorem 4.1 that l
is positive. Since we have assumed that l1 : l (A) A is positive, it is im1
1
mediate from 1
is
l (T )(a) = l (T l (a)), for T Nl (l (A), B(X)), that l
positive. Hence l : Nl (l (A)) Ml (A) is an isomorphism of ordered algebras. Likewise, we know from the rst part of Theorem 4.5 that l l is positive. Since l (T )(a) = l1 (l (a)T ), for T Nr (l (A), B(X)), the assumption
that l1 is positive shows that l is positive. Hence this is also true for l , and
l l : M(A) N (l (A), B(X)) is an isomorphism of ordered algebras.
As to the involutive situation, we know from the rst part of Theorem 4.5 that
l l is involutive. Hence so is the image, and since the inverse of an involutive
map is necessarily involutive, we are done.

Suppose that, in Theorem 5.3, l : A l (A) has a bounded inverse l1 :
in the proof shows that, after identifying A
l (A) A. Then the denition of 1
l
with its image l (A), Ml (A) is to be identied with all left multiplications by elements of Nl (l (A), B(X)), and, likewise, M(A) is to be identied with all pairs consisting of a left and a right multiplication by the same element of N (l (A), B(X)).
Therefore the following excerpts from Theorem 5.3 hold.
Corollary 5.4. Let A be a normed algebra with a bounded approximate left
identity, and let X be a Banach space. Suppose that l : A B(X) provides X with

70
18

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

the structure of a non-degenerate faithful normed left A-module, and that l is an


embedding of A as a topological algebra. Then Ml (A) is canonically isomorphic,
as a topological algebra, with Nl (l (A), B(X)), where Nl (l (A), B(X)) can carry
either the norm from B(X) or the equivalent norm from Ml (l (A)). Likewise,
M(A) is canonically isomorphic, as a topological algebra, with N (l (A), B(X)),
where N (l (A), B(X)) can carry either the norm from B(X) or the equivalent norm
from M(l (A)).
If, in addition, A is an ordered normed algebra with a positive bounded approximate left identity, X is ordered with a closed positive cone, and l is an isomorphism
of ordered algebras between A and l (A), then the above two canonical isomorphisms
are isomorphisms of ordered algebras.
Alternatively, if, in addition, A has a bounded involution, X is a Hilbert space,
and l is involutive, then the above canonical isomorphism between M(A) and
N (l (A), B(X)) is an isomorphism of involutive algebras.
Corollary 5.5. Let X be a Banach space, and suppose that A is a (not necessarily closed) subalgebra of B(X), acting non-degenerately on X. If A has a
bounded approximate left identity, then Ml (A), resp. M(A), is canonically isometrically isomorphic with Nl (A, B(X)), resp. N (A, B(X)), where the latter algebra is
supplied with the norm from Ml (A), resp. M(A).
If, in addition, X is ordered with a closed positive cone, and A has a positive
bounded approximate left identity, then the above two canonical isomorphisms are
isomorphisms of ordered algebras.
Alternatively, if, in addition, X is a Hilbert space, and A is involutive, then
the canonical isomorphism between M(A) and N (A, B(X)) is involutive.
Remark 5.6. If A is a normed algebra with a bounded left approximate identity, then it follows from Corollary 5.5 that, for each isometric non-degenerate
embedding into B(X), for some Banach space X, the left normalizer of the image,
resp. the normalizer of the image, is always canonically isometrically isomorphic
with Ml (A), resp. M(A). If, in addition, A is involutive with an isometric involution, then, for each involutive isometric non-degenerate embedding into B(X), for
some Hilbert space X, the normalizer of the image is always isometrically involutively isomorphic with M(A). As a special case, we retrieve the well known fact
that the multiplier algebra of a C -algebra is C -isomorphic with the normalizer of
the image in any faithful non-degenerate involutive Hilbert representation.
Remark 5.7. If X is a Banach space, let K(X) denote the compact operators
on X. For T B(X), it is easily checked that the norm of the corresponding left
and right multiplication on K(X) is in both cases equal to the norm of T as an
element of B(X). If K(X) has a bounded left approximate identity (e.g., if X has a
Schauder basis), then Corollary 5.5 therefore asserts that Ml (K(X)) and M(K(X))
are both canonically isometrically isomorphic with B(X). More is true, however:
these isometric isomorphisms are both valid without any assumption on K(X), see
[8, p. 313]. In the same paper, it is also shown [8, p. 314] that Mr (K(X)) is always
isometrically isomorphic with the algebra of bounded operators on the dual space
of X. We refer to [11, Section 1.7.14] for further results in this vein.
The right-sided version of Theorem 5.3 reads as follows. The statement that
the right centralizer algebra of A is isomorphic with the left normalizer of the image
is not a mistake.

EXTENDING REPRESENTATIONS OF NORMED ALGEBRAS IN BANACH SPACES

71
19

Theorem 5.8. Let A be a normed algebra with an m-bounded approximate


right identity, and let X be a Banach space. Suppose that r : A B(X) provides
X with the structure of a non-degenerate faithful normed right A-module. Then in
the diagram
r

/ B(X)
(5.2)
A
O

 

(A)

?
/ Mr (A)

from Theorem 4.3, r and are injective anti-homomorphisms, and i and r are injective homomorphisms. The canonical anti-homomorphism r : M(A) Mr (A)
is also injective.
Suppose, for the remainder of this Theorem, that r : A r (A) has a bounded
inverse r1 : r (A) A.
If Nl (r (A), B(X)) carries the norm from either B(X) or Ml (r (A)), then
r is a bounded algebra isomorphism between Mr (A) and Nl (r (A), B(X)), with
 r  mr  in both cases, and the inverse map 1
r : Nl (r (A), B(X)) Mr (A)
1
is also bounded, with  1





in
both
cases.
r
r
r
Likewise, if N (r (A), B(X)) carries the norm from either B(X) or M(r (A)),
then r r : M(A) B(X) yields a bounded algebra anti-isomorphism between
M(A) and N (r (A), B(X)), with  r r  mr  in both cases, and the inverse
map ( r r )1 : N (r (A), B(X)) M(A) is also bounded, with ( r r )1 
r r1  in both cases.
If, in addition, A is an ordered normed algebra with a positive bounded approximate right identity, X is ordered with a closed positive cone, and r is an antiisomorphism of ordered algebras between A and r (A), then r : Mr (A) B(X)
yields an isomorphism of ordered algebras between Mr (A) and Nl (r (A), B(X)),
and r r : M(A) B(X) yields an anti-isomorphism of ordered algebras between M(A) and N (r (A), B(X)).
Alternatively, if, in addition, A has a bounded involution, X is a Hilbert space
and r is involutive, then r r : M(A) B(X) yields an anti-isomorphism of
involutive algebras between M(A) and N (r (A), (X)).
Proof. The proof is similar to that of Theorem 5.3, and uses Theorem 4.3
and the second part of Theorem 4.5. The rst step is to prove that the homomor1
phism 1
: Nl (r (A), B(X)) Mr (A), dened by 1
r
r (T )(a) = r (T r (a)),
for T Nl (r (A), B(X)) and a A, is the two-sided inverse of r : Mr (A)
Nl (r (A), B(X)). The role of the anti-homomorphism l in the previous proof is
taken over by the anti-homomorphism r : Nr (r (A), B(X)) Ml (A), dened by
r (T )(a) = r1 (r (a)T ), for T Nr (r (A), B(X)), and a A. These combine to
the anti-homomorphism ( r )1 : N (l (A), B(X)) M(A), which is given by
( r )1 (T ) = (r (T ), 1

r (T )), for T N (r (A), B(X)).
Corollary 5.9. Let A be a normed algebra with a bounded approximate right
identity, and let X be a Banach space. Suppose that r : A B(X) provides X with
the structure of a non-degenerate faithful normed right A-module, and that r is an
anti-embedding of A as a topological algebra. Then Mr (A) is canonically isomorphic, as a topological algebra, with Nl (r (A), B(X)), where Nl (r (A), B(X)) can
carry either the norm from B(X) or the equivalent norm from Ml (r (A)). Likewise,

72
20

SJOERD DIRKSEN, MARCEL DE JEU, AND MARTEN WORTEL

M(A) is canonically anti-isomorphic, as a topological algebra, with N (r (A), B(X)),


where N (r (A), B(X)) can carry either the norm from B(X) or the equivalent norm
from M(r (A)).
If, in addition, A is an ordered normed algebra with a positive bounded approximate right identity, X is ordered with a closed positive cone, and r is an antiisomorphism of ordered algebras between A and r (A), then the above canonical isomorphism between Mr (A) and Nl (r (A), B(X)) is an isomorphisms of ordered algebras, and the above canonical anti-isomorphism between M(A) and N (r (A), B(X))
is an anti-isomorphism of ordered algebras.
Alternatively, if, in addition, A has a bounded involution, X is a Hilbert space,
and r is involutive, then the above canonical anti-isomorphism between M(A) and
N (r (A), B(X)) is an anti-isomorphism of involutive algebras.
References
[1] B. Blackadar, Operator Algebras. Theory of C - Algebras and Von Neumann Algebras, Encyclopaedia of Mathematical Sciences, Vol. 122, Operator Algebras and Noncommutative
Geometry, Vol. III, Springer-Verlag, Berlin, 2006.
[2] F.F. Bonsall, J. Duncan, Complete normed algebras. Ergebnisse der Mathematik und ihrer
Grenzgebiete, Band 80. Springer-Verlag, New York-Heidelberg, 1973.
[3] H.G. Dales, Banach Algebras and Automatic Continuity, London Math. Soc. Monographs 24,
Oxford Univ. Press, New York, 2000.
[4] K.R. Davidson, C -algebras by Example, Fields Institute Monographs Vol. 6, American Mathematical Society, Providence, RI, 1996.
[5] S. Dirksen, M. de Jeu, M. Wortel, Beyond the stars: crossed products of Banach algebras, to
appear.
[6] J. Dixmier, C -algebras, North-Holland Mathematical Library, Vol. 15, North-Holland Publishing Co., Amsterdam-New York-Oxford, 1977.
[7] G.J.O. Jameson, Ordered Linear Spaces, Lecture Notes in Math., Vol. 141, Springer-Verlag,
Heidelberg, 1970.
[8] B.E. Johnson, An introduction to the theory of centralizers, Proc. London Math. Soc. (3) 14
(1964), pp. 299320.
[9] B.E. Johnson, Cohomology in Banach algebras, Mem. Amer. Math. Soc. 127, American Mathematical Society, Providence, RI, 1972.
[10] G.J. Murphy, C -algebras and operator theory, Academic Press, Boston, MA, 1990.
[11] T.W. Palmer, Banach algebras and the general theory of -algebras, Vol. 1: Algebras and
Banach algebras, Encyclopedia of Mathematics and its Applications, 49. Cambridge University
Press, Cambridge, 1994.
[12] D.P. Williams, Crossed products of C -algebras, Mathematical Surveys and Monographs, 134,
American Mathematical Society, Providence, RI, 2007.
Sjoerd Dirksen, Delft Institute of Applied Mathematics, Delft University of Technology, P.O. Box 5031, 2600 GA Delft, The Netherlands
E-mail address: s.dirksen@tudelft.nl
Marcel de Jeu, Mathematical Institute, Leiden University, P.O. Box 9512, 2300 RA
Leiden, The Netherlands
E-mail address: mdejeu@math.leidenuniv.nl
Marten Wortel, Mathematical Institute, Leiden University, P.O. Box 9512, 2300
RA Leiden, The Netherlands
E-mail address: wortel@math.leidenuniv.nl

Contemporary Mathematics
Volume 503, 2009

Countable bases for Hilbert C -modules and classication of


KMS states
Tsuyoshi Kajiwara
Abstract. Rational functions on the Riemann sphere or some self-similar
maps on compact metric spaces have branch points, and then their graphs
or co-graphs are considered as typical examples of branched covering. We
construct countable bases for Hilbert C -modules explicitly for rational function case and self-similar map case. Using the bases we construct, we compute
the explicit form of the Perron-Frobenius type operators for these case, and
present classication of KMS sates on Cuntz-Pimsner algebras constructed
from rational functions and self-similar maps with respect to the gauge action.

1. Introduction
or some self similar maps on
Rational functions on the Riemann sphere C
compact metric spaces have branch points, and then their graphs or co-graphs are
considered as typical examples of branched covering. We can construct Hilbert
C -modules or Hilbert C -bimodules from these graphs or co-graphs without excluding branched points (Kajiwara-Watatani [8], [9],[11]).
Finite bases for nitely generated Hilbert C -modules are constructed naturally and are useful for investigating many examples. But Hilbert C -modules
constructed from the above branched covering are not nitely generated and do
not have nite bases. If a Hilbert C -module is countably generated, the existence
of countable bases follows directly from the Kasparovs stabilization trick. But
the explicit constructions of countable bases for Hilbert C -modules which are not
nitely generated are not known in general.
In this paper, we construct countable bases for Hilbert C -modules associated
with rational functions on the Riemann sphere and some self-similar maps explicitly using imaginary roots of unity. Using these countable bases, we compute the
explicit form of the Perron-Frobenius type operators for Hilbert C -bimodules associated with rational functions and self-similar maps, and classify KMS states on
Cuntz-Pimsner C -algebras constructed from the Hilbert C -bimodules associated
with rational functions on the Riemann sphere and associated with some self similar
maps with the aid of the theorem of Laca-Neshveyev [14]. In particular, we present
1991 Mathematics Subject Classication. Primary 46L08, 37F10.
Key words and phrases. C -algebras, Hilbert C -bimodule, KMS state, countable basis.
The author was supported in part by Grants-in-Aid for Scientic Research (19540218) from
Japan Society for the Promotion of Science.
1
73

c
2009
American Mathematical Society

74
2

TSUYOSHI KAJIWARA

a complete classication result of KMS states on Cuntz-Pimsner C -algebras associated with rational functions on the Riemann sphere.
The construction of countable bases for rational function case and self-similar
map case are original and have not yet been published anywhere. We believe that
the explicit form of countable bases for Hilbert C -modules associated with the
correspondences with branch points are not known yet. The results in section
4 for the calculation of the explicit form of the Perron-Frobenius type operators
are obtained in Izumi-Kajiwara-Watatani [6] without using countable basis. Classication results of KMS states are obtained in [6], but we present a exposition of
classication results for both cases in a manner slightly dierent from [6].
This paper is based on the joint work with Y. Watatani and with M. Izumi.
2. Preliminary
2.1. Hilbert C -bimodules and Cuntz-Pimsner algebras. Let A be a
C -algebra. We give the denition of Hilbert C -modules.

Definition 2.1. A linear space X is called a Hilbert A-module if the followings


hold.
(1) There exist an A-valued hermitian, positive denite inner product (|)A
and a right action of A.
(2) X is complete with respect to the norm x = (x|x)A 1/2 .
(3) (x|ya)A = (x|y)A a for x, y X and a A.
Definition 2.2. A countably innite subset {ui }
i=1 of X is called a countable
basis (or frame) of X over A if
x=

ui (ui |x)A

i=1

holds for each x X and the right hand side is norm convergent.

We note that the right hand side is independent of the ordering


 of {ui }i=1
([7]), and we can write the basis as {ui }i , and can write as x = i ui (ui |x)A ,
where is a nite or countably innite set. If X is countably generated over A, a
countable basis always exists due to the Kasparovs stabilization trick. But explicit
forms of countable bases are not known for specic examples in general.
We denote by L(X) the set of linear operators on X which is adjointable with
respect to the A-valued inner product. For x and y X, put x,y z = x(y|z)A for
z X. We denote by K(X) the norm closure of the linear span of {x,y |x X, y
Y } in L(X). If there exists an injective nondegenerate *-homomorphism from
A to L(X), then we call the pair (X, ) (or simply X) a Hilbert C -bimodule (or
C -correspondence) over A. Put IX = 1 ((A) K(X)). Then IX is an ideal of
A.
Cuntz-Pimsner algebra OX is the universal C -algebra generated by {a, Sx | a
A, x X} satisfying

aSx = S(a)x ,
lim a

n

i=1

(Sx )a = Sxa ,
aSui (Sui )  = 0

(Sx ) Sy = (x|y)A ,
a IX ,

a A,

x, y X,

COUNTABLE BASES FOR HILBERT C -MODULES

75
3

where {ui }
i=1 is a countable basis for X. There exist isometric representations A
and X of A and X into OX such that
A ((x|y)A ) = X (x) X (y),

X ((a)xb) = A (a)X (x)A (b)

for x, y X and a, b A. We identify a A with A (a) in OX and x X with


X (x) in OX . There exists an action of one dimensional torus group T on OX
by the relations
t (a) = a,

t (x) = eit x

for a A, x X.

This is called the gauge action.


(n)
There exists an isometric representation K of K(X n ) into OX for each n
(n)
such that K (x1 xn ,y1 yn ) = x1 xn yn y1 . We also identify K(X n )
with its isometric image in OX for each n. Let F (n) be A + K(X) + K(X 2 ) + +
K(X n ) OX with convention F (0) = A. F (n) is a C -subalgebra and K(X n )
is a closed two sided ideal in F (n) . {F (n) }n=0,1,2,... is an increasing ltration and


.
F = n=1 F (n) is equal to the xed point subalgebra OX
2.2. Rational functions. We refer [8] for this section. Let R be a rational

function. We consider the dynamical system given by R on the Riemann sphere C.


n
Let R(z0 ) = w0 . If R(z) = w0 + c(z z0 ) + (higher term) for some local chart at
z0 , n = e(z0 ) is called the branch index of R at z0 . Put
| e(z) 2},
B(R) = {z C

| w = R(z), z B(R)}.
C(R) = {w C

Points in B(R) are called branch points and points in C(R) are called branch values.
we denote by [z] the set of w C
such that Ri (z) = Rj (w) for some i,
For z C,
is called an exceptional point if [z] is a nite set. We denote
j N. A point z C
by ER the set of exceptional points of R. Exceptional points of rational functions
are classied in Beardon [1] as follows:
(1)
(2)
(3)
(4)

ER
ER
ER
ER

= .
= {z0 }.
= {z0 , z1 } (z0 = z1 ) and [z0 ] = {z0 }, [z0 ] = {z1 }.
= {z0 , z1 } (z0 =
 z1 ) and [z0 ] = [z1 ] = {z0 , z1 }.
which is an ordinary graph of R. Put A = C(C)

Put CR = {(z, R(z)) | z C},


and XR = C(CR ). We dene a right A-action on XR and a right A-inner product
for XR by

( a)(z, R(z)) = (z, R(z))a(R(z)), (|)A (w) =
e(z)(z, w)(z, w),
zR1 (w)

where , XR , a A. Then XR itself is a Hilbert A-module. We dene


a left action of A on XR by ((a))(z, R(z)) = a(z)(z, R(z)). Then (XR , )
is a Hilbert C -bimodule over A. We denote by OR the Cuntz-Pimsner algebra
constructed from (XR , ).
2.3. Self-similar maps. We refer [9] for this section. Let (K, d) be a compact
metric space. A continuous map on K is called a proper contraction if there exists
constants 0 < c1 c2 < 1 such that
c1 d(y1 , y2 ) d((y1 ), (y2 )) c2 d(y1 , y2 )

y1 , y2 K.

76
4

TSUYOSHI KAJIWARA

Let N be an integer greater than 1, and let = (1 , . . . , N ) be a family of proper


contractions on (K, d).

Definition 2.3. K is called self-similar with respect to when K = N
i=1 i (K).
Dene subsets B() and C() of K by
B() = {x K | x = j (y) = j  (y) for some y K and j = j  }
C() = {y K | j (y) = j  (y) for j = j  }.
We call points in B() branch points, and points in C() branch values. We dene
a branch index e(x, y) where x = j (y) for some j by
N

e(x, y) = #{j {1, . . . , N } | j (y) = x}.

Put C = j=1 {(j (y), y) | y K}. This C is called the co-graph of . Put
A = C(K) and X = C(C ). We dene a right A-action and A-inner product of
X by
( a)(j (y), y) = (j (y), y)a(y)
(|)A (y) =

N


(j = 1, . . . , N ),

(j (y), y)(j (y), y),

j=1

where , X and a A. Then X is a Hilbert C -module over A. We dene


a left action of A on X by ((a))(j (y), y) = a(j (y))(j (y), y). Then (X , )
is a Hilbert C -bimodule over A. We denote by O the Cuntz-Pimsner algebra
constructed from (X , ).
2.4. KMS states on Cuntz-Pimsner algebras. Let A be a C -algebra,
be an automorphic action of one dimensional torus T on A and be a non-negative
real number. Put A(n) = {a A | t (a) = eint a} for n Z. We note that A(0)
is equal to the xed point subalgebra A . Aanal denotes
 the set a A such that
t t (a) has an analytic extension to C. Put E(a) = T t (a) dt. Then E is the
conditional expectation from A to A given by .
Definition 2.4. Let > 0. A state of A is called a -KMS state on A with
respect to if
(xit (y)) = (yx)
holds for x A and y in some dense *subalgebra D contained in Aanal .
Lemma 2.5. Let be a -KMS state on A with respect to . Then the restriction of to A(0) is a tracial state such that
(x y) = en (yx )

x, y A(n) .

For a tracial state on A(0) satisfying the above condition, the -KMS state on A
whose restriction to A(0) coincides with is given by E.
Let A be a C -algebra and I be a closed two sided ideal, {e } be an approximate unit in I. For a positive linear function on I, there exists a positive linear
functional on A satisfying (a) = lim (ae ) for a A, which is called the
natural extension of .

COUNTABLE BASES FOR HILBERT C -MODULES

77
5

Lemma 2.6. (Exel-Laca [3] Proposition 12.5) Let B be a unital C -algebra, A


be a C -subalgebra of B containing unit and I be a closed two sided ideal of B such
that B = A + I. Let be a state on A, be a positive linear functional on I such
that = on A I and on A. Then there exists a state on B such that
|A = and |I = .
Let A be a C -algebra, (X, ) be a Hilbert C -bimodule over A, and OX be
the Cuntz-Pimsner algebra constructed from (X, ).
Lemma 2.7. (Fowler-Muhly-Raeburn [4]) F (n1) K(X n ) = K(X n1 )
K(X n ) in OX holds.
Lemma 2.8. (Katsura [12]) A K(X) = IX in OX .

We assume
n the existence of a countable basis {ui }i=1 for X over A such that
N = supn  i=1 (ui |ui )A  < .
The following theorem is proved in Laca-Neshveyev [14]. We give an alternative
simple and self-contained proof here.

Theorem 2.9. (Laca-Neshveyev. [14]) Let be a -KMS state on OX with


respect to the gauge action. Then the restriction of to A corresponds to a tracial
state on A satisfying the following:

(1)
(ui mui ) = e (a) for a IX .
i=1

+
(2)
i=1 (ui aui ) e (a) for a A .
The correspondence is bijective and conserves the set of extreme points.
n
Proof. Let be a -KMS state on OX . Let a IX . Since { i=1 ui ,ui }
n=1
is an approximate unit in K(X), we have







(ui aui ) = e
(aui ui ) = e a lim
ui ui = e (a).
i=1

i=1

i=1

Then |A satises (1). Let a A . Then we have



 n


n
n




1/2

1/2
(ui aui ) = e
(aui ui ) = e a
ui ui a
e (a).
+

i=1

i=1

i=1

for each n, and then |A satises (2).


Conversely for a tracial state satisfying the above conditions (1) and (2), we

construct a tracial state on OX


satisfying the condition in Lemma 2.5 whose
restriction to A coincides with . If satises the condition (2), we can dene a
bounded tracial linear functional (n) on K(X n ) ([14] Theorem 1.1) by

(n) (T ) = en
((ui1 uin |T (ui1 uin ))A ).
(i1 ,...,in )Nn
n
) is also given by the same formula. This
The natural extension (n) of (n) to L(X

is shown using the approximate unit { (i1 , ,in )F ui1 uin ,ui1 uin }{F : niteNn }

in K(X n ). We note that (n) is also a bonded tracial linear functional.


We construct a tracial state (n) on F (n) for each n such that (0) = ,
(n)
|F (n1) = (n1) , and (n) |K(X n ) = (n) inductively. First, we put (0) = .
We assume that there exists (n) on F (n) such that (n) |K(X n ) = (n) and (n)
(n) on F (n1) . When x is contained in F (n) K(X n+1 ) = K(X n ) K(X n+1 ),

78
6

TSUYOSHI KAJIWARA

using uin ui1 xui1 uin A K(X) = IX (Lemma 2.8) and the condition (1),
we have

(n+1) (x) = e(n+1)
(uin+1 uiin ui1 xui1 uin uin+1 )
(i1 , ,in ,in+1 )Nn+1

= en

(uiin ui1 xui1 uin ) = (n) (x) = (n) (x).

(i1 , ,in )Nn

We note that for x F (n)


(n+1) (x) = e(n+1)

by the condition (2), we have



(uin+1 uin ui1 xui1 uin uin+1 )

(i1 ,...,in ,in+1 )Nn+1

en

(uin ui1 xui1 uin ) = (n) (x).

(i1 ,...,in )Nn

We write x F (n) as x = y + z where y F (n1) and z K(X n ). By the


assumption of induction, (n) (y y) (n) (y y) holds. On the other hand, from
y z + z y + z z K(X n ), we have (n) (y z + z y + z z) = (n) (y z + z y + z z).
From these, we have
(n+1) (x x) = (n+1) ((y + z) (y + z))
(n) ((y + z) (y + z))
= (n) (y y + y z + z y + z z)
= (n) (y y) + (n) (y z + z y + z z)
= (n) (y y) + (n) (y z + z y + z z)
(n) (y y) + (n) (y z + z y + z z)
= (n) (y y + y z + z y + z z) = (n) (x x).
By Lemma 2.6, there exists a state (n+1) on F (n+1) such that (n+1) |F (n) =
(n) and (n+1) |K(X n+1 ) = (n+1) . Moreover (n+1) |F (n) + (n+1) holds. By
mathematical induction argument, there exists a desired sequence { (n) }n=1,2,...

of states. We dene on n=0 F (n) by |F (n) = (n) , and extend it to F by
continuity. Since (a + k1 + + kn ) = (a) + (1) (k1 ) + + (n) (kn ) for a A
and ki K(X i ), does not depend on the choice of the basis {uk }
k=1 we have
used.
Since F (n) is contained in L(X n ), (n) is a trace for each n. Then is a tracial
state on F. From (x1 , ,xn ,y1 , ,yn ) = en ((y1 yn |x1 xn )A ),
satises the condition of Lemma 2.5, and extends to a -KMS state of OX .

For A , put
F ( )(a) =

((ui |(a)ui )A )

for a A.

i=1

Then F is a linear map on A . We call F a Perron-Frobenius type operator. It is


important to calculate the explicit form of F for specic examples. We will do the
calculation for rational function case and self-similar map case in section 4.

COUNTABLE BASES FOR HILBERT C -MODULES

79
7

3. Construction of basis
In this section, we construct countable bases for Hilbert C -modules associated
and associated with self similar maps. We state the
with rational functions on C
detail for rational function case and omit the detail for self-similar map case. We
assume that every neighborhood is connected. We use notation U = {z C | |z|
} for a positive number . U is a compact connected neighborhood of 0.
3.1. Rational functions. We construct a countable basis for XR for each
rational function R in four steps.
1st step. Let V = U for some > 0. Then V is a compact neighborhood
of 0 in C. Put C = {(z, z N ) | z N V } (N 1). Then X = C(C) is naturally a
Hilbert A-module where A = C(V). We construct a basis for X. If N = 1, {u1 }
constitutes a basis, where u1 (z, z) = 1 for z V .
We assume N 2. We dene continuous functions ri (x) on [0, ) by r0 (x) = 0
and for i 1

1
i x
2i
L
L
ri (x) =
x

1
L
2i x i

L
0
0 x 2i
.
where L is a positive constant.
ri (x) ri+1 (x), we put vi (x) = (ri (x)
 Noting
2
have
v(0)
=
0, and that for each > 0 there exist an
ri1 (x))1/2 . Then we 
i=1
m
integer m0 such that i=1 vi (x)2 = 1 holds for m m0 for each x with x.
For 1 p N 1, we dene a functions p (z) on C\{0} by
p
z
1
p (z) =
.
N |z|
We dene continuous functions uk k = 1, 2, . . . on C by
1
u1 (z, z N ) = ,
N

uk (z, z N ) = p (z)vi (|z|),

where k = 1 + (N 1)(i 1) + p for i 1, 1 p N 1.



N
We put fM (z, z N ) = M
k=1 uk (uk |f )A (z, z ) for a positive integer M .
Proposition 3.1. As M tends to , fM converges to f uniformly on C.
Proof. We take an f C(C). We assume M 2. We write M = 1 + (N
1)(m 1) + n , where m is a positive integer and 1 n N 1. Then fM is
expressed as follows:
fM (z, z N ) =
(3.1)

N 1
m N 1 N 1
1 
1   p j
f ( j z, z N ) +
( ) f ( j z, z N )vi (|z|)2
N j=0
N i=1 p=1 j=0


n N 1
1  p j
+
( ) f ( j z, z N )vm+1 (|z|)2 ,
N p=1 j=0

using = e2 1/N .
We x > 0, and put  = /(N + 1). There exists > 0 such that for every
|z| < we have
|f (z, z N ) f (0, 0)| <  .

80
8

TSUYOSHI KAJIWARA

For |z| < , we have


|fM (z, z N ) f (z, z N )|
|f (z, z N ) f (0, 0)| + |fM (z, z N ) f (0, 0)|
1 
|f ( j z, z N ) f (0, 0)|
N j=0








m N
n N
1
1
1 N







1
1
p j
2
p j
2
+
( ) f (0, 0)vi (|z|) +
( ) f (0, 0)vm+1 (|z|)


N i=1 p=1 j=0
N p=1 j=0

N

< +

m N 1 N 1
n N 1
1  
1 
+
|f (z, z N ) f (0, 0)|vi (|z|)2 +
|f (z, z N ) f (0, 0)|vm+1 (|z|)2
N i=1 p=1 j=0
N p=1 j=0

 +  + 0 + 0 + (N 1)

m+1


vi (|z|)2

i=1

(N + 1) = .
N 1
We used j=0 ( p )j = 0, which follows from that p s are imaginary N -th roots
m+1
of 1 for 1 p N 1, and used i=1 vi (|z|)2 1 for every m.
m0Take the above > 0, and assume |z| . Then we can take m0 such that
i=1 vi (|z|) = 1 for all |z| . Using (3.1), for M = 1 + (N 1)(m 1) + n with
m m0 , we have
N 1
N 1 N 1
1 
1   p j
f ( j z, z N ) +
( ) f ( j z, z N )
N j=0
N p=1 j=0


N 1
N
1

1 
j p
1+
( ) f ( j z, z N )
=
N j=0
p=1


N
1
N
1


1
N
j p
1+
( ) f ( j z, z N )
= (1 + (N 1))f (z, z ) +
N
p=1
j=1

fM (z, z N ) =

=f (z, z N ).
We used that j is imaginary N -th root of 1 and 1 +
N 1.
Then if M 1 + (N 1)m0 , we have

N 1
p=1

( j )p = 0 for 1 j

|fM (z, z N ) f (z, z N )| <


for every z C.

2nd step Let R be a rational function, a B(R) and b = R(a) C(R).


Put N = eR (a). By the argument using Boettchers theorem ([15]), there exists a
compact neighborhood U of a, compact neighborhood V of b such that R1 (V ) = U
and a homeomorphism 1 from U to a compact neighborhood U  = U for some
> 0 of 0 and 2 from V to a compact neighborhood V  = UN of 0 such that,
N
(2 R1
on U  . For this situation, we say that R|U is conjugate to z N .
1 )(z) = z
Let C = {(z, R(z)) | z U }, X = C(C) and A = C(V ). Let C = {(z, z N ) | z U  }
and A = C(V  ).

COUNTABLE BASES FOR HILBERT C -MODULES

81
9

If N = 1, the basis consists of singleton. We assume N 2. Let {


uk }
k=1 be
the countable basis constructed for C in step 1. We put
uk (z, R(z)) = u
k (1 (z), 2 (R(z))).
Then for z U , we have
M


(uk (uk |f )A )(z, R(z)) =

k=1

M

k=1

M


uk (z, R(z))

uk (
z , R(z))f (
z , R(z))

R(
z )=R(z)

u
k (, N )

N )f (1 (),
1 ( N )))
u
k (,
1
2

N = N

k=1

1
=f (1
1 (), 2 ())

= f (z, R(z))

z ). This shows that {uk }


where = 1 (z) and = 1 (
k=1 is a countable basis for
X.
3rd step Let b C(R) and R1 (b) = {a1 , a2 , , am }. Let V be a suciently
small compact neighborhood of b such that there exist mutually disjoint compact
neighborhoods Ui of ai such that R(Ui ) = V and each R|Ui is conjugate to z Ni in the
sense of 2nd step. Put CV = {(z, R(z)) | R(z) V } and CVi = {(z, R(z)) | z Ui }.
i
Then CV = {(z, R(z)) | R(z) V } is a disjoint union of
CmV (i i = 1, . . . , m). Put
i
i
AV = C(V ) and XV = C(C ). Then we have XV =
i=1 XV as AV modules.
If N 2, we can construct a countable basis for XVi for each i as in step 2. If
N = 1, the basis consists of singleton. By gathering these bases, we can construct
a countable basis for XV .
4th step Let C(R) = {b1 , b2 , , bs }. We take suciently small open neighborhoods Vi s such that Vi s are mutually disjoint and the argument using Boettchers
theorem in step 2 works well. Let XVi be as in step 3 for each i. Let Wi be an open
s Wi . Then
neighborhood of bi such that W i Vi for each i. Put V0 = C\
i=1
and V0 does not contain elements in C(R).
{Vi }si=0 is a nite open covering of C
s+t
We can nd open sets Vs+1 , Vs+t such that V0 i=s+1 Vi and there exist continuous sections ik (1 k N ) of w = R(z) at Vi such that if k = l ik (w) = il (w)
uij }ji for XVi (s + 1 i s + t). Let
for w Vi . We can take a nite basis {
i
{
uj }ji (1 i s) be the countable basis for XVi constructed in step 3.
s+t
Let {i }s+t
i=1 be a partition of unity associated with the open covering {Vi }i=1
Since f (z, R(z))i (z) = 0 for z
of C.
/ Vi for f X, we may write
(f i )(z, R(z)) =

u
ij (
uij |f i )A (z, R(z)).

ji

We dene

uik (z, R(z))

u
ik (z, R(z))(i (R(z)))1/2
0 for z
/ Vi .

for z Vi

82
10

TSUYOSHI KAJIWARA

Put I(z) = {i | R(z) Vi }. Then we have


s+t 


uik (uik |f )A (z, R(z)) =

i=1 ki

 

1/2

u
ik i

1/2

(
uik i

|f )A (z, R(z))

iI(z) ki

 

iI(z)

s+t


u
ik (
uik |f i )(z, R(z))

ki

(f i )(z, R(z)) = f (z, R(z)).

i=1

Convergence is uniform on CR .
We dene a countable discrete set = {(i, k) | 1 i s + t, k i }.
Theorem 3.2. Let R be a rational function. Then {uik }(i,k) constructed
above is a countable basis for XR over A.
3.2. Self-similar maps. We construct a countable basis for self-similar map
= (1 , . . . , N ) on a compact metric space (K, d) under the condition that C()
is nite. For the convenience of construction, we permit the case N = 1. The result
of this section is contained in the unpublished preprint Kajiwara-Watatani [10].
If N = 1, the basis consists of {u1 ((y), y)} where u1 ((y), y) = 1. We assume
N 2.
1st step In 1st step we consider the following situation for the convenience
of construction. Let K1 and K2 be compact metric spaces and i (i = 1, , N )
be proper contractions from K1 to K2 . We use the same terminology proper
contraction for maps from K1 to K2 . We assume that there exists a c K1
such that 1 (c) = = N (c) and i (y)s are dierent for y = c. We put C =
{(i (y), y)|y K1 , i = 1, . . . , N }, A = C(K1 ) and X = C(C). Then X is a Hilbert
A module in a canonical way. We say that such a module is of N -branch class.
We use the same family of functions {ri (x)}
given in section 3.1,
i=0 on [0, ) 

that i=1 v(0)2 = 0, and


and put vi (x) = (ri (x) ri1 (x))1/2 for i 1. We note 
m
that for each > 0 there exist an integer m0 such that i=1 vi (x)2 = 1 holds for
m m0 and foreach x with x.
Let = e2 1/N . Since j (y) (j = 1, , N ) are mutually dierent for y = c
and vi (0) = 0, we can do the following denition. For k 1, we dene a family of
continuous functions uk (k = 1, 2, . . . ) in X as follows:
1
1
u1 (x, y) = , u1+(N 1)(i1)+p (j (y), y) = pj vi (d(y, c)),
N
N
where i 1 and 1 p N 1.

For M , we put fM = M
k=1 uk (uk |f )A .
We take arbitrary small and put  = /(N + 1). There exists > 0 satisfying
the following: If d(y, c) < , then we have
|f (j (y), y) f (b, c)| < 
for each j. We assume d(y, c) < . By the calculation similar to rational function
case using the property of imaginary roots of 1 in section 3.1, we have
|fM (j (y), y) f (j y, y)| < (N + 1) = .
For the above , we can choose m0 > 0 satisfying the following: For y with
d(y, c) and k 1 + (N 1)(m 1) where m m0 , we have uk (j (y), y) = 0

COUNTABLE BASES FOR HILBERT C -MODULES

83
11

for each j. Put M0 = 1 + (N 1)(m0 1). Let M M0 . For each j, by the the
calculation similar to rational function case in section 3.1, we have
fM (j (y), y) = f (j (y), y).
Then we conclude that for every M M0 ,
|fM (j (y), y) f (j (y), y)| <
for all y K1 . We conclude that {uk }
k=1 as above is a basis for X.
2nd step Let c C() and {j (c) | 1 j n} = {b1 , . . . , bs }, and U be a compact neighborhood of c which does not contain other elements of C(). We put C =
{(x, y) | x = k (y), for some k , y U } and Ci = {(x, y)|x = k (y), where k (c) =
bi , y U }. We put A = C(U ), X = C(C) and Xi = C(Ci ) for each
si. We can
construct a basis for each Xi as an A-module by step 1. Since X = i=1 Xi as A
module, we can construct a basis for X over A.
3rd step Since C() is a nite set, we write C() = {c1 , . . . , cm }. We take
suciently small open neighborhoods Ui of ci such that C() U i = {ci } for
1 i m and U i U i = for i = i . We take anopen neighborhood Vi of
m
ci such that V i Ui for each i. We put U0 = K\ m
i=1 V i . Then {Ui }i=0 is
an open covering of K. We put Ai = C(U i ), Ci = { (x, y) C | y U i } C
and Xi = C(Ci ), i = 0, , m. Then Xi s are Hilbert Ai -modules in step 2. Let
{
uik }ki be the basis for Xi dened in step 2, where i is a countable set. For
XU0 , we can construct a nite basis {
u0k }N
k=1 because U0 contains no point in C().
m
Let {i }i=0 be a partition of unity associated with the open covering {Ui }m
i=0 of
K.
For 0 i m, we dene

u
ik (j (y), y)i (y)1/2 for y Ui
i
uk (j (y), y) =
0 for y
/ Ui .
m
Put = i=0 {(i, k)) | k i }. As in 4th step in section 3.1, we can show the
following theorem.
Theorem 3.3. ([11]) Let (K, d) be a compact metric space, = (1 , . . . , N )
be a system of proper contractions on K. We assume that C() is a nite set. Then
{uki }(i,k) constructed above is a basis for X .
4. Computation for the Perron-Frobenius type operators
We present the calculation of the explicit form of Perron-Frobenius type operators for rational function case and self-similar map case using countable bases
constructed in section 3. In this section and the next section, we identify a Borel
measure on a compact metric space and the corresponding tracial linear functional
on a commutative
C -algebra C() for notational convenience. We use notation

(a) for a() d(), where a C().
we dene the Borel function
4.1. Rational function case. For a A = C(C)
a
by

a
(w) =
a(z).
zR1 (w)

If the degree of R is greater than 1, a


is not continuous. We note that 1 N 1,

and |
a| N a 1, where 1(z) = 1 for z C.

84
12

TSUYOSHI KAJIWARA

Let {uk }
k=1 be the countable basis for XR constructed in the section 3.1.
we have
Proposition 4.1. For a C(C),

(uk |auk )A (w) = a


(w).

k=1

The convergence of the left hand side is monotone for positive a.


Proof. First, we consider 1st step in section 3.1. Let V = U for some > 0,
and U = {z C | z N V } = U1/N . Let {uk }
k=1 be the basis as in 1st step in
section 3.1. We consider X = C(C) as a left C(U )-module. For a C(U ), we have

(uk |(a)uk )A (w) =

e(z)uk (z, w)a(z)uk (z, w)

k=1 zR1 (w)

k=1

e(z)|uk (z, w)|2 a(z)

k=1 zR1 (w)

e(z)|u1 (z, w)|2 a(z) +

zR1 (w)

e(z)|uk (z, w)|2 a(z).

k=2 zR1 (w)

If z = 0, then e(z) = N and uk (z, w) = 0 for k 2. Then the last expression is


equal to N (1/N )a(0) = a(0). If z = 0, then e(z) = 1. The last expression is
1
N
=

1
N

a(z) +

zR1 (w)

a(z) +

zR1 (w)

N 1
1 
N i=1 p=1

N 1
N

vi (|z|)2 a(z)

zR1 (w)

vi (|z|)2 a(z) =

zR1 (w) i=1

a(z).

zR1 (w)

In both cases, it coincides with a


(w).
We consider 2nd step in section 3.1. The case N = 1 is trivial. We assume
N 2. Let a B(R) and U , V , 1 , 2 , U  and V  be as in 2nd step in section
3.1, and R|U be conjugate to z N for some N 2. Let {uk }
uk }
k=1 and {
k=1 be
as in 2nd step in section 3.1. We consider X = C(C) as a left C(U )-module. For
a C(U ), we have

(uk |(a)uk )A (w) =

e(z)uk (z, w)a(z)uk (z, w)

k=1 zR1 (w)

k=1

e(1 (z))
uk (1 (z), 2 (z))a(z)
uk (1 (z), 2 (z))

k=1 zR1 (w)



k=1

e()
uk (, )b()
uk (, ) = b() = a
(w).

n =

We used notation = 1 (z), = 2 (w) and b = a 1


1 .
We consider 3rd step in section 3.1. Let c C(R) and R1 (c) = {b1 , , bm }.
Let V be a compact neighborhood of c satisfying the following: Let Ui be a compact
neighborhood of bi such that R(Ui ) = V , R|Ui is conjugate to z Ni for some Ni 1
in the sense of 2nd step in section 3.1 and Ui s are mutually disjoint.

COUNTABLE BASES FOR HILBERT C -MODULES

85
13

i
We put Ri = R|Ui . Let {uik }
k=1 be a basis of XV constructed in 2nd step in
i

we
for each i. For a C(C),
section 3.1. We consider XV as a left C(C)-module
have

m 


(uik |(a)uik )A (w) =

m


a(z)

i=1 z(Ri )1 (w)

i=1 k=1

a(z) = a
(w).

zR1 (w)

For 4th step, we use the partition of unity {i }s+t


i=1 in section 3.1. We x w C

and put I(w) = {i | w Vi }. Let {uk }k=1 and {


uk }k=1 be as in 4th step in section

3.1. Then for a C(C),


s+t 


(uik |(a)uik )A (w) =

i=1 ki

s+t 


1/2

(
uik i

1/2

|(a)
uik i

)A (w)

i=1 ki

 

1/2

(
uik i

1/2

|(a)
uik i

)A (w)

iI(w) ki

 

iI(w)

(
uik |(a)
uik )A (w)i (w)

ki

a
(w)i (w) =

s+t


a
(w)i (w) = a
(w).

i=1

iI(w)


By Proposition 4.1, we have

sup 
n

k=1 (uk |uk )A

n


= 1 N 1. Then we have

(ui |ui )A  N.

i=1

We can rewrite the condition (1) and (2) in Theorem 2.9 for rational function
case.
Proposition 4.2. ([6]) Let R be a rational function. For a -KMS state
corresponds to a
on OR with respect to the gauge action, its restriction to C(C)

probability Borel measure on C satisfying the following:


(
a) = e (a)

(a|B(R) = 0)

(
a) e (a)

(a A )

(3)

(4).

The correspondence is bijective and conserves extreme sets.


Proof. The conclusion follows from IX = {f A | f (z) = 0 for z B(R)}
and Proposition 4.1.

Proposition 4.2 is obtained in [6] without using countable basis. We have
presented a direct and simple proof using countable basis we have constructed in
section 3.1.

86
14

TSUYOSHI KAJIWARA

4.2. Self-similar map case. Let be a self-similar map for a compact metric
space (K, d). We put j (a)(y) = a(j (y)). For a A, we dene a Borel function a

by
N

1
a(j (y)).
a
(y) =
e(j (y), y)
j=1
We note that if C() is not empty, a
is not continuous.
Proposition 4.3. We assume that C()
is a nite set. For the basis {uk }k
for X constructed in section 3.2, we have k (uk |auk )A (y) = a
(y). We note
that the left hand side is monotone convergent for a positive a A.
Proof. We consider 1st step in section 3.2. Let X be of N -branch class. Let
{uk }
k=1 be the basis constructed in 1st step in section 3.2. We show

(uk |auk )A (y) = a


(y)

k=1

for every y K1 .
If N = 1, we have (u1 |au1 )A (y) = a(1 (y)). We assume N 2. We consider
X as a left C(K2 )-module. Let a C(K2 ). Then we have

(uk |(a)uk )A (y) =


N


uk (j (y), y)a(j (y))uk (j (y), y)

k=1 j=1

k=1


N


|uk (j (y), y))|2 a(j (y)))

k=1 j=1

N
N
N 1 
1 
a(j (y)) +
vi (d(j (y), y))2 a(j (y)).
N j=1
N i=1 j=1

N
N
The last expression is equal to j=1 a(j (y)) if y = c and equal to N1 j=1 a(j (y))
if y = c. In any case, this is equal to a
(y). If a A+ , the left hand side is monotone
convergent for a positive a.
We consider 2nd step in section 3.2. Let {uik }ki be the basis in 2nd step in
section 3.2. We consider Xi as a left C(K)-module. Let a C(K). Then, from
step 1,
 s
s 

a(bi ) for y = c
i
i

i=1
(uk |(a)uk )A (y) =
s
for y = c.
i=1
j,j (c)=bi a(j (y))
i
i=1
k

This is equal to a
(y).
We consider 3rd step. By taking a partition of unity associated with the open
covering in 3rd step in section 3.1 and doing calculation similar to rational function
case in section 4.1, we can prove the conclusion.

As in section 4.1, we have
sup 
n

n

i=1

(ui |ui )A  N.

COUNTABLE BASES FOR HILBERT C -MODULES

87
15

Proposition 4.4. ([6], [11]) Let = (1 , , N ) be a self-similar map on


(K, d). We assume C() is a nite set. Then (1) (2) in Theorem 2.9 are rewritten
as
(
a) = e (a)

(a|B()) = 0)

(
a) e (a)

(a A )

(3)
(4).

5. Classication of KMS states


In this section, we give a classication result for KMS states on C -algebras
given by rational functions or self-similar maps using countable bases constructed
in section 3. The results of this section are already obtained in [6], we present only
outline of them in a slightly dierent manner.
for rational function case or a compact metric space K for selfLet be C
similar map case. We x > 0. We dene F by
F ()(a) = e F ()(a)
for each Borel measure on and each a C(). Then F is a positive linear
map on C() . We denote by N the degree of rational function or the number of
contraction for self-similar map.
Lemma 5.1. If > log N , F  < 1.
Proof. Let be a signed Borel measure on . Then = v||, where v =
(d)/(d||). We note v 1. From F ()(a) = e ||(v
a), we have
F () = sup |F ()(a)| = sup e |||(v
a)|
a 1

a 1

sup e
a 1

||(|v||a|1) sup a e ||(1)


a 1

e N ||(1) = e N ,
and taking sup for  1, we have F  e N < 1.

5.1. Rational function. In this section we refer the condition (3) and (4) in
Proposition 4.2 simply as (3) and (4).
First we consider the case > log N . Let be a Borel probability measure on
satisfying the condition (3), (4). If > log N , then F  < 1 by Lemma 5.1.
C
We put 0 = F (). Then we have
n

lim
(F )j (0 ) = lim (F )n () =
n

j=0

Since B(R) is nite set,


By the condition (3), 0 vanishes on a C(C\B(R)).

is
given
by
a
point
measure
on
B(R).
Let
B(R)
=
{b
0
1 , , bs }. For each bi ,

j
j=0 (F ) (bi ) is norm convergent. Put
bi , =
where m

bi ,

1
mbi ,


(F )j (bi ),
j=0

is the normalizing positive constant.

Proposition 5.2. When > log N , bi , satises the condition of (3) and (4)
in Proposition 4.2, and dene a -KMS state bi , on OX .

88
16

TSUYOSHI KAJIWARA

Proposition 5.3. ([6]) When > log N , the set of extreme points of -KMS
states on OX is {b, }bB(R) .
satisfying (3) and
Proposition 5.4. Let be a Borel probability measure on C
does not have point mass at B(R) and C(R). Then the equation (3) holds for every
a A, and must be equal to log N .
Proof. Let {an }
n=1 be a sequence in A such that an |B(R) = 0 and an a
monotone increasingly outside B(R). Since does not have point mass at B(R)
and C(R), we have
lim (an ) = (a)

lim (
an ) = (
a).

If the equation (3) holds for all a A, we can substitute a = I, and then =
log N .


For a A, we dene F by F (a) = e zR1 (w) e(z)a(w). If a measure on
has no point mass on B(R) and C(R), and F () = , then satises (3) for
C
all a A.
R and each n N, we dene
Definition 5.5. (Lyubich [13]) For any w C\E
w
by
a probability measure n on C

N n e(z)e(R(z)) e(Rn1 (z))z .
w
n =
xRn (w)

L
L
is
The sequence (w
n )n=1 converges weakly to a measure . The measure
L
independent of the choice of w. This is called the Lyubich measure.

By the denition, L is F -invariant. It is known that L has no point mass.

Then the condition (3) holds for all a C(C).


gives a log N -KMS
Proposition 5.6. ([6]) The Lyubich measure L on C
L
state on OR with respect the gauge action.
Proposition 5.7. ([6]) We assume = log N . If a Borel probability measure
satises (3) and does not have point mass at B(R) and C(R), then must
on C
be the Lyubich measure L .
We denote by {z} the characteristic function of the singleton {z}. Put c (z) =
({z}). Then c (z) is a point mass of at z.
We assume that satisfy (3) and (4) for > 0.
Proposition 5.8. For z
/ B(R), we have c (R(z)) = e c (z). For z B(R),

we have c (R(z)) e c (z).


Proof. We substitute a = {z} to (3) and to (4).

The following proposition is proved in [6] using Proposition 5.8, and is the key
point for the classication.
If z
Proposition 5.9. ([6]) Let be a Borel probability measure on C.
/ ER
and c (z) > 0, then > log N .

COUNTABLE BASES FOR HILBERT C -MODULES

89
17

which satises the


Let log N . Let be a Borel probability measure on C
condition (3). Put

1 =
c (c)c .
cER

Then 1 satises the condition (3) in Theorem 4.2 and 1 has no point mass. If
< log N , 1 = 0 holds, and if = log N , 1 is a constant multiple of L by
Proposition 5.4 and Proposition 5.7.
Theorem 5.10. (Izumi-Kajiwara-Watatani [6]) The set of extreme points of
-KMS states on OR with respect to the gauge action are classied as follows.
1. When > log N , the set of extreme points of -KMS states are {b, }bB(R) .
2. When = log N , the set of extreme points of -KMS states are {L }
{c, }cER .
3. When 0 < < log N , the set of extreme points of -KMS states are {c, }cER .
The measures c, where c ER are given as follow. When ER = {w1 , w2 }
and [w]i = {wi } for each i, {c, | c ER } = {w1 , w2 }. When ER = {w1 , w2 } and
[w1 ] = [w2 ] = {w1 , w2 }, {c, | c ER } = {1, , 2, }, where
1, =

e
1
w +
w ,
+ 1 1 e + 1 2

2, =

1
e
w1 +
w .
+1
e +1 2

When ER = {w1 }, { c, | c ER } = {w1 }. The -KMS states {c, | c ER } on

OR are determined by these probability measures c, on C.


5.2. Self-similar maps. We present classication results for some examples
of self-similar maps. We assume that C() is a nite set. In this section, we refer
the condition (3) and (4) in Proposition 4.4 simply as (3) and (4).
weconsider the case log > N and we write as B() = {b1 , , bs }. Then
First
k
e
k=0
(j1 ,...,jk ){1,...,N }k jk j1 (bi ) is norm convergent for each i. Put
bi , =

1
mbi ,


k=0

ek

jk j1 (bi ) ,

(j1 ,...,jk ){1,...,N }k

where mbi , is the positive normalizing constant. Then bi , s satises (3) and (4).
Let bi , be the corresponding -KMS state on O . In a similar manner as rational
function case in section 5.1, the following proposition holds.
Proposition 5.11. If log > N , the set of extreme points of -KMS states
on O with respect to the gauge action is {b, }bB() .
Let G be the map from C(K) to C(K) dened by
G(a) =

N
1 
(a).
N i=1 i

Proposition 5.12. (Huchinson [5]) There exists a unique G -invariant probability measure H on K.
We call H the Huchinson measure associated with .
Proposition 5.13. If C() is nite and H has no point mass, H satises
(3) and (4) of Proposition 4.4, and dene a log N -KMS state of O .

90
18

TSUYOSHI KAJIWARA

In Izumi-Kajiwara-Watatani [6], the set of extreme points of -KMS states of


O are classied under some condition. In this paper, we present the classication
only for some examples.
Example 5.14. Let K = [0, 1], 1 (y) = 12 y and 2 (y) = 12 (y + 1). Then K
is self-similar with respect to . In this case, B() = and C() = , and O
is Cuntz algebra O2 . A KMS state exists only for = log 2 and corresponds the
Lebesgue measure on [0, 1].
Example 5.15. ([6]) Let K = [0, 1], 1 (y) = 12 y, and 2 (y) = 1 12 y. Then K
is self-similar with respect to . For this example B() = { 12 } and C() = {1}. In
this case O is Cuntz algebra O . This example is called tent map.
The set of extreme points of -KMS states on O with respect to the gauge
action is as follows:
(1) When = log 2, H is the unique -KMS state, where H corresponds
to the Lebesgue measure on [0, 1].
(2) When > log 2, 1/2, is the unique -KMS state, where 1/2, corresponds to the measure 1/2, on [0, 1] given by

1/2,

= (1 e


k=0

ek

jk j1 (1/2) .

(j1 ,...,jk ){1,2}k

Example 5.16. ([9], [6]) We present a classication result for the CuntzPimsner C -algebra associated with Sierpinski Gasket introduced in Kajiwara
Watatani [9]. Let be a regular triangle in R2 with three vertexes c1 = (1/2, 3/2),
c2 = (0, 0) and c3 = (1, 0). The middle point of c1 c2 is denote by b1 , the middle
point of c1 c3 is denoted by b2 and the middle point of c2 c3 is denoted by b3 . We
dene proper contractions i (i = 1, 2, 3) by




x y 
x 1 y
x 1 y
3
+ , +
, 2 (x, y) =
,
, 3 (x, y) =
+ ,
.
1 (x, y) =
2 4 2
4
2 2
2 2 2
Let be the rotation by the angle . We put 1 = 1 , 2 = 2/3 2 and
3 = 2/3 3 . We denote by S with the metric d induce from R2 the self-similar
set determined by = (1 , 2 , 3 ). We note that ci and bi i = 1, 2, 3 are contained
in S. In this case, we have B() = {b1 , b2 , b3 } and C() = {c1 , c2 , c3 }. We denote
by OS, the Cuntz-Pimsner algebra constructed from S and .
Let > log 3. For y B(), we dene a probability measure y, on S as
follows:


e 3  k
y,
e
jk j1 (y) .
=
e
k
k=0

(j1 ,...,jk ){1,2,3}

Then satises (3) and (4), and extends to a -KMS state y, on OS, . -KMS
states on OS, with respect to the gauge action exist only for log 3 and are
classied as follows:
y,

(1) When = log 3, H is the unique KMS state on OS, .


(2) When > log 3, the set of extreme points of -KMS states on OS, with
respect to the gauge action is {y, | y = b1 , b2 , b3 }

COUNTABLE BASES FOR HILBERT C -MODULES

91
19

References
[1] A. F. Beardon, Iteration of rational functions, GTM 132, 1991, Springer New York
[2] M. Enomoto, M. Fujii and Y. Watatani, KMS states for gauge action on OA , Math. Japon.
29(1984), 607619
[3] R. Exel and M. Laca, Partial dynamical systems and the KMS condition, Comm. Math. Phys.
232(2003), 223277
[4] N. J. Fowler, P. S. Muhly and I. Raeburn, Representations of Cuntz-Pimsner Algebras, Indiana Univ. Math. J. 52(2003), 569605.
[5] J. E. Huchinson, Fractals and self-similarity, Indiana Univ. Math. J. 30 (1981), 713747.
[6] M. Izumi, T. Kajiwara and Y. Watatani, KMS states and branched points, Ergod. Th. &
Dynam. Sys. 27(2007), 18871918
[7] T. Kajiwara, C. Pinzari and Y. Watatani, Jones index theory for Hilbert C -bimodules and
its equivalence with conjugation theory, J. Funct. Anal. 215(2004), 149
[8] T. Kajiwara and Y. Watatani, C -algebra associated with complex dynamical systems, Indiana
Univ. Math. J. 54 No. 3 (2005),755778
[9] T. Kajiwara and Y. Watatani, C -algebras associated with self-similar sets, J. Operator
Theory, 56(2006), 225247
[10] T. Kajiwara and Y. Watatani, C -algebras associated with algebraic correspondences, to
appear J. Operator Theory
[11] T. Kajiwara and Y. Watatani, KMS states on C -algebras associated with self-similar sets,
[axXiv:math.OA/0405514v1]
[12] T. Katsura, On C -algebras associated with C -correspondences, J. Funct. Anal. 217(2004),
366-401.
[13] M. Y. Lyubich, Entropy properties of rational endomorphisms of the Riemann sphere, Ergodic Th. & Dynam. Sys. 3 (1983), 351385
[14] M. Laca and S. Neshveyev, KMS states of quasi-free dynamics on Pimsner algebras, J.
Funct. Anal. 211(2004) 457482
[15] J. Milnor, Dynamics in one complex variable, Annals of Math. Studies, Princeton Univ.
Press, 2006
Department of Mathematics for Environmental Study, Okayama University, Tsushima,
Okayama, Japan 700-8530
E-mail address: kajiwara@ems.okayama-u.ac.jp

This page intentionally left blank

Contemporary Mathematics
Contemporary Mathematics
Volume 503, 2009

Subshifts and C -algebras from one-counter codes


Wolfgang Krieger and Kengo Matsumoto
Abstract. We introduce a class of subshifts under the name of standard
one-counter shifts . The standard one-counter shifts are the Markov coded
systems of certain Markov codes that belong to the family of one-counter
languages. We study topological conjugacy and ow equivalence of standard
one-counter shifts. To subshifts there are associated C*-algebras by their graph systems. We describe a class of standard one-counter shifts with the
property that the C*-algebra associated to them is simple, while the C*-algebra
that is associated to their inverse is not. This gives examples of subshifts that
are not ow equivalent to their inverse. For a family of highly structured
standard one-counter shifts we compute the K-groups.

1. Introduction
Let be a nite alphabet. We use notation like
x[i,k] = (xj )ijk ,

x Z , i, k Z, i k,

and we denote by x[i,k] also the word that is carried by the block x[i,k] . The length
of a word a is denoted by (a). On the shift space Z there acts the shift by
x = (xi )iZ Z .

x (xi+1 )iZ ,

A closed shift-invariant subset of Z is called a subshift. For an introduction to


the theory of subshifts see [10, 13]. A word is called admissible for a subshift if
it appears in a point of the subshift. We denote the language of admissible words
of a subshift X Z by L(X) and set Ln (X) = {a L(X) | (a) = n}, n N. A
subshift X Z is uniquely determined by L(X). For a subshift X Z and for
I , I+ Z, I < I+ , one has a topological conjugacy
x (x[i+I ,i+I+ ] )iZ ,

(x X)

of X onto the higher block system X [I ,I+ ] of X.


2000 Mathematics Subject Classication. Primary 37B10; Secondary 68Q45, 46L80.
Key words and phrases. subshifts, Markov codes, Markov coded systems, standard onecounter shifts, C*-algebras, -graph systems, K-theory, ow equivalence.
This work was supported by JSPS Grant-in-Aid for Scientic Reserch (N0. 20540215).
1
93

94
2

WOLFGANG KRIEGER AND KENGO MATSUMOTO

Among the rst examples of subshifts are the topological Markov shifts. Using
a matrix (A(,  )), ,
A(,  ) {0, 1},

,  ,

that has in every row and every column at least one entry that is equal to 1 as a
transition matrix one obtains a topological Markov shift tM (, A) by setting
tM (, A) = {(i )iZ Z | A(i , i+1 ) = 1, i Z}.
For n > 1 the n-block system (Z )[1,n] of the shift on Z is a topological Markov
shift with a transition matrix A(n) that is given by

1 if a(1,n] = a[1,n) ,
(n)

,
a, a n .
A (a, a ) =
0 if a(1,n] = a[1,n)
A subshift X Z is said to be of nite type if there is a nite set F of words in
the alphabet such that (i )iZ X precisely if no word in F appears in (i )iZ .
A subshift is topologically conjugate to a subshift of nite type if and only if it is
of nite type [10, 13]. We formulate this theorem equivalently as:
Theorem 1.1. Let X Z be a subshift that is topologically conjugate to a
topological Markov shift. Then there exists an n N such that
X [1,n] = tM (Ln (X), (A(n) (a, a ))a,a Ln (X) ),

n n .

The coded system [2] of a formal language C in a nite alphabet is the


subshift that is obtained as the closure of the set of points in Z that carry biinnite concatenations of words in C. C can here always be chosen to be a prex
code. The property of being coded is an invariant of topological conjugacy. We
denote the coded system of a code C by sc(C). More generally a Morkov code
(see [8]) is given by a formal language C of words in a nite alphabet together
with a nite index set and, mappings s : C , t : C and a transition
matrix (A(,  )),  , A(,  ) {0, 1}, ,  . From a Markov code (C, s, t)
one obtains the Markov coded system scM (C) as the subshift that is the closure of
the set of points x Z such that there are indices ik Z, k Z, ik < ik+1 , k Z
such that x[ik ,ik+1 ) C, k Z, and such that
k Z.

A(t(x[ik1 ,ik ) ), s(x[ik ,ik+1 ) )) = 1,

With the alphabet {an | 1 n N } { , + }, N N, consider the codes


(N )

k m
Creset ={
+ an | 1 n N, m, k N, m k}

and with the alphabet {bn | 1 n N } { , + }, N N, consider the codes


(N )

k k
+ bn | 1 n N, k N}.
Ccounter = {
(N )

(N )

(N )

(N )

The coded systems sc(Creset ), sc(Ccounter ) and sc(Creset ) sc(Ccounter ) serve us as


prototypes for a class of subshifts that we will call standard one-counter shifts.
(Compare here [1, Example 1, p. 561], [8, Example II, p. 449], [11, Example
6.1, p. 896]). We arrive at a description of this class of subshifts by observing the
(N )
(N )
(N )
(N )
behavior of sc(Creset ), sc(Ccounter ) and of sc(Creset )sc(Ccounter ) and by abstracting
the essential structural properties that these coded systems are to share with the
(N )
(N )
(N )
standard one-counter shifts. sc(Creset ) and sc(Creset ) sc(Ccounter ) are prototypes

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

95
3

of what we will call standard one-counter shifts with reset. To every standard onecounter shift X Z there is associated a unique Markov code C (X) such that
X = scM (C (X) ) and such that a version of Theorem 1.1 holds. A formal language
is called a one-counter language if it is recognized by a push down automaton with
one stack symbol [4, 5, 7]. The Markov code C (X) that is associated to a standard
one-counter shift X X is a one-counter language.
Given a subshift X X a word v L(X) is called synchronizing if for u, w
L(X) such that uv, vw L(X) also uvw L(X). A topologically transitive subshift
is called synchronizing if it has a synchronizing word. Before turning in Section
3 to the standard one-counter codes we introduce in Section 3 auxiliary notions
for synchronizing subshifts. We introduce strongly synchronizing subshifts as the
subshifts in which synchronizing symbols appear uniformly close to synchronizing
words, and we introduce suciently synchronizing subshifts as the subshifts that
have a strongly synchronizing higher block system.
-graph systems (as introduced in [15]) are labeled directed graphs that are
equipped with a shift like map . A -graph system L gives rise to a C -algebra
OL . To a subshift X there is invariantly associated a future or predicting -graph
system XL that is based on the future equivalences of the pasts in X(,0] (as in
[11]) and there is invariantly associated a past or retracing -graph system LX
that is based on the past equivalences of the futures in X[0,) (as in [15]). The
predicting and the retracing -graph systems of a subshift are time symmetric to
one-another: the predicting -graph system of a subshift is identical to the retracing
-graph system of its inverse and vice versa. For a standard one-counter shift X we
will see that OXL is simple if and only if X has reset and that OLX is not simple.
Since the stable isomorphism class of OLX is an invariant of ow equivalence [17],
a standard one-counter shift with reset is not ow equivalent to its inverse. For the
(N )
(N )
one-counter shifts sc(Creset ), sc((Creset )rev ), we will show that
K0 (Osc((C (N )

rev )

K1 (Osc((C (N )

rev )

reset )
reset )

)
= K0 (Osc(C (N ) ) )
= Z/N Z Z,
reset

) = K1 (O
)
(N )
= 0.
sc(Creset )

(N )

(N )

The one-counter code Ccounter is equal to its reversal (Ccounter )rev . The K-groups
of the C -algebra have been computed in [12] as
K0 (Osc(C (N )

counter )

K1 (Osc(C (N )

counter )

)
= Z/N Z Z2 ,
)
= Z.

For another computation of K-groups of one-counter shifts see [14, 21].


For a subshift X Z we set
X[i,k] = {x[i,k] | x X},

i, k Z,

i k.

We set also
+
k (a) = {b X(n,n+k] | (a, b) X[m,n+k] },
+
(a)

= {x X(n,) | (a, x ) X[m,) },


+

k N,
n, m Z, m < n, a X[m,n] .

has the time symmetric meaning.



Z , and a topological conjugacy
We recall that, given subshifts X Z , X

:X
[L,L] , such
: X X, there is for some L Z+ a block mapping

96
4

WOLFGANG KRIEGER AND KENGO MATSUMOTO

that
x[iL,i+L] ))iZ ,
(
x) = ((

x
X.

We set
a) = ((
a[jL,j+L] ))i+LjkL,
(

[i,k] ,
a
X

i, k Z,

k i > 2L.

We use similar notation for words.


2. Strong synchronization
The rst lemma is well known. We include the proof for completeness.

Z , X Z be subshifts and let : X
X be a
Lemma 2.1. Let X

topological conjugacy. Let L, L Z+ be such that [L, L] is a coding window for


x = (
X,
x) and I , I+ Z, I
and [L, L] is a coding window for 1 . Let x
I+ . Let x[I ,I+ ] be synchronizing. Then x
[I LL,I
is synchronizing.

+
L+L]
+
Proof. Let
x[I LL,I
),
y

(
+ +L+L]
y+ +
x[I LL,I
),

(
+ +L+L]
+
+
and let y
(x[I ,I+ ] ), y (x[I ,I+ ] ), be given by

 y , x
[I LL,I
) = (y , x[I ,I+ ] ),
(

+ +L+L]
x
(
, y+ ) = (x[I ,I+ ] , y + ).

[I LL,I
+ +L+L]
One has (y , x[I ,I+ ] , y + ) X and
y , x
[I LL,I
, y+ )
1 (y , x[I ,I+ ] , y + ) = (

+ +L+L]
and the lemma follows.

For a subshift X Z , we denote the set of its synchronizing symbols by


synchro (X).

Z , X Z be subshifts and let a topological conjugacy
Lemma 2.2. Let X

. Let L Z+ be such that 1


: X X be given by a one-block map :
x[L,L] ) = (
x0 ), x
[L,L] . Then
has coding window [L, L] and set (
[L,L] X
[L,L] ).
1 (synchro (X)) synchro (X
Proof. Apply Lemma 2.1.

We say that a synchronizing subshift X Z is strongly synchronizing if there


exists a Q Z+ such that the following holds: if x X and I , I+ Z, I < I+ are
such that x[I ,I+ ] is synchronizing, then there exists an index i, I Q i I+ +Q
such that xi is a synchronizing symbol.
The higher block systems of a strongly synchronizing subshift are also strongly
synchronizing. We say that a subshift X Z is suciently synchronizing if it has
strongly synchronizing higher block systems.
Proposition 2.3. Sucient synchronization is an invariant of the topological
conjugacy of subshifts.

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

97
5

Proof. To prove the lemma it is by Lemma 2.2 enough to consider the case

Z and of a topological conjugacy : X X
that is
of subshifts X Z , X

given by a one-block map : such that


synchro (X)
synchro (X)
1 (

(2.1)

strongly synchronizing and to show that X is strongly synchronizing. Let


X
Z+ be
Z+ be such that 1 has the coding window [L, L] and let Q

[I ,I+ ] is synchronizing, one


that for x
X, I , I+ Z, I < I+ such that x

i is a synchronizing symbol. Then


has an i, I Q i I+ + Q such that x
one has for x X, I , I+ Z, I < I+ such that x[I ,I+ ] is synchronizing, by
Lemma 2.1 that (x)[I ,I+ ] is synchronizing. It follows that there exists an i Z,
i I+ + L + Q
such that (xi ) is a synchronizing. By (2.1), then xi
I L Q
is synchronizing.

with
L
such

3. Standard one-counter shifts


3 a. The structure of standard one-counter shifts
Let X Z be a topologically transitive subshift. We call a pair (( )iZ , (+ )iZ )
of xed points of X a characteristic pair, if it is the unique pair of xed points that
satises the following conditions (a), (b) and (c ), and a condition (c+ ) that is
symmetric to condition (c ):
(a) X has a unique orbit OX that contains all points that are left asymptotic to
( )iZ and right asymptotic to (+ )iZ , and that do not contain a synchronizing
word.
(b) X has a point that is left asymptotic to (+ )iZ and right asymptotic to
( )iZ and that contains a synchronizing word.
(c ) There exists a K N such that the following holds: If x X and
I , I+ Z, I I+ , are such that x is right asymptotic to ( )iZ , and x[I ,I+ ]
is synchronizing, and x(I+ ,I+ +k] , is not synchronizing, k N, then there exists an
index i, I < i I+ + K, such that xj = , j i.
Proposition 3.1. Let X Z be a topologically transitive subshift with a
characteristic pair (( )iZ , (+ )iZ ) of xed points, and let be a topological

Z onto X. Then (1 (( )iZ ), 1 ((+ )iZ )) is a
conjugacy of a subshift X

characteristic pair of xed points of X.


Proof. Conditions (a), (b), (c ), (c+ ), being satised by (( )iZ , (+ )iZ ),
the proposition follows by means of Lemma 2.1.

We introduce notation that we use for a synchronizing subshift X Z , that
has a characteristic pair (( )iZ , (+ )iZ ) of xed points. For synchro (X)
we denote by D( , ) the set of words d L(X) such that

k
(
),
d
kN
+

and for
such that

+
synchro (X) we denote by D(
, + ) the set of words d
+ L(X)
+
d+


kN

k
(+
).

98
6

WOLFGANG KRIEGER AND KENGO MATSUMOTO

We set
(X) = { synchro (X) | D( , ) = },
+
+
+
(X) = { synchro (X) | D( , + ) = }.

, ) and + (X),
D(+ , + , ), D( , +
+ (X) have the symmetric meaning.

Lemma 3.2. For a strongly synchronizing subshift X Z that has a characteristic pair (( )iZ , (+ )iZ ) of xed points, the sets (X) and + (X) are
not empty and the sets D( , ), (X) and D(+ , + ), + + (X) are
nite.
Proof. We show that (X) is not empty, and that the sets D( , ),
(X) are nite. By condition (b) there exists an x X that contains a synchronizing word and that is right asymptotic to ( )iZ . The assumption that X
is strongly synchronizing implies that x contains a synchronizing symbol. Let
i Z be such that xi synchro (X), xi+K  synchro (X), K N. If here
xi+K = , K N, then the empty word is in D( , ) where = xi . Otherwise let j > i be given by xj = , xj+K = , K N and have x[i,j] D( , ).
The niteness of D( , ), (X) follows from condition (c ).

Let X Z be a subshift with a characteristic pair (( )iZ , (+ )iZ ) of xed
points. Let x OX . If for some i Z,
xi = ,
xi = + ,

i i ,
i > i

then set cX equal to the empty word. Otherwise determine i , i+ Z, i < i+ , by


xi = ,

i < i ,

xi = ,
 + ,
x i+ =
xi = + ,

i > i+ ,

and set cX equal to the word x[i ,i+ ] .


We also set
+ (X) = {d+ + | + + (X), d+ D(+ , + )},

(X) = {d+
+ | + + (X), d D( , + )}.
+
+
We denote by +
reset (X) the set of d + (X) such that there is a D Z+
such that
k
k +D
cX ++ d+ + L(X),
k , k+ N,

the smallest such D to be denoted by D(d+ + ). We say that X has reset if


+
reset (X) = . We set
+
+
+
counter (X) = (X)\reset (X).

We set
k +D(d+ + ) +

reset (X) = {cX +

d + | d+ + +
reset (X)},

and we say that X satises the reset condition if (X)\


reset (X) is a nite set.

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

99
7

For a strongly synchronizing subshift X Z that has a synchronizing symbol


denote by C(X) the set of admissible words of X that begin with a synchronizing
symbol, that have no other synchronizing symbol and that can be followed by a
synchronizing symbol. For c C(X) set t(c) equal to the set of synchronizing
symbols that can follow c and set s(c) equal to the singleton set that contains the
rst symbol of c. With the set of subsets of as index set and with a transition
matrix A whose positive entries are given by
A( , {}) = 1,

{t(c) | c C(X)},

synchro (X),

C(X) is a Markov code and


X = scM (C(X)).
Given a strongly synchronizing subshift X Z with a characteristic pair
(( )iZ , (+ )iZ ) of xed points such that +
(X) = , that satises the reset
condition, we set

C (+
) = { d d+
| (X), d D( , ),
k

(X)

(3.1)

d+
+ (X)\reset (X), k N},

(X)

(3.2)

+

+ (X),

C (+
),
(X)

+ (X)


t(c) = {+
+ (X) | c C (+ )},
(X)

(3.3)

(X)

c C ,

and, given M , M+ Z+ and mappings


d D ( d ) Z+ ,

(X), d D( , ),

d+ + D+ (d+ + ) Z+ ,

d+ + +
reset (X),

we set
(3.4)
(X)

Creset (D , M , M+ , D+ ; + )
k +D(d+ + ) +

={ d cX ++
k

d | (X), d D( , ), d+ + +
reset (X),

k , k+ N, D ( d ) + k + M M+ + k+ + D+ (d+ + )}, + + (X),



(X)
(X)
(3.5)
Creset (D , M , M+ , D+ ; + ),
Creset (D , M , M+ , D+ ) =
+ + (X)
(X)

(3.6)

t(c) = {+ + (X) | c Creset (D , M , M+ , D+ ; + )},


(X)

c Creset (D , M , M+ , D+ ).

and given J , J+ Z+ , and mappings


d ( d ) Z+ ,

(X), d D( , ),

d+ + + (d+ + ) Z+ ,

d+ + +
counter (X),

100
8

WOLFGANG KRIEGER AND KENGO MATSUMOTO

we set
(X)

Ccounter ( , J , J+ , + ; + )
={ d cX ++ d+ | (X), d D( , ),
k

(3.7)

(3.8)

d+ + +
counter (X), k , k+ N,
( ( , d ) + k + J ) (J+ + k+ + + (d+ + )) = }, + + (X)

(X)
(X)
Ccounter ( , J , J+ , + ; + ),
Ccounter ( , J , J+ , + ) =
+ + (X)
(X)

t(c) = {+ + (X) |c Ccounter ( , J , J+ , + , + )},

(3.9)

(X)

c Ccounter ( , J , J+ , + ).

By (3.1-9) there is dened a Markov code


(X)

(X)

(X)

Creset (D , M , M+ , D+ ) Ccounter ( , J , J+ , + ).

We dene a standard one-counter shift as a strongly synchronizing subshift


X Z that has a characteristic pair of xed points, such that +
(X) is empty,
such that X satises the reset condition, and such that there exist I Z+ and
parameters D , M , M+ , D+ , , J , J+ , + such that
(3.10)
{c C(X) | (c) > I}
(X)

={c C

(X)

(X)

Creset (D , M , M+ , D+ ) Ccounter ( , J , J+ , + ) | (c) > I},

where the equality is understood as an equality of Markov codes. If (3.10) holds then
we say that I, D , M , M+ , D+ , , J , J+ , + are parameters of the standard
one-counter shift X Z . The parameters , J , J+ , + can be missing, and in
the case that X has no reset the parameters D , M , M+ , D+ are missing.
For a standard one-counter shift X Z denote the smallest I Z+ such
that (3.10) holds by IX , and denote by D (X), M (X), M+ (X), D+ (X), (X),
J (X), J+ (X), + (X) the uniquely determined parameters for X that satisfy the
normalization conditions
min (M , M+ ) = min (J , J+ ) =
=
=
(N )

min

d+ + + (X)

min

(X),d D( ,d )

D(d+ + ) =
min

D( d )

min

(X),d D( , )

{+ + (X),d+ D(+ ,+ )|d+ + +


counter (X)}

( d )

+ (d+ + ) = 0.

(N )

E. g. for sc(Creset Ccounter ) the normalized parameters are given by I = M =


M+ = J = J+ = 0, the range of D and D+ being {0}, and the range of and
+ being {0}. We associate with a standard one-counter shift X Z the Markov
code
C (X) ={c C (X) | (c) IX }

(X)
(X)
{c C Creset (D (X), M (X), M+ (X), D+ (X))
(X)

Ccounter ( (X), J (X), J+ (X), + (X)) | (c) > IX },

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

101
9

and nd that X has a distinguished presentation as the Markov coded system of


C (X) .
There is a development that can be considered, at least partially, as the converse. One takes as a starting point a nite alphabet , a proper subset synchro
of and symbols , +  . One also has to provide for some I Z+ a Markov
code all of whose words begin with a symbol in synchro , with the remaining symbols in \synschro and that have length less than or equal to to I, and one has
to provide the other components, , + , D , M , M+ , D+ , , J , J+ , + that
are needed for the construction of a Markov code C according to rules that imitate
the content of (3.1-9). An additional requirement is that the symbols in synchro
are the only synchronizing symbols in scM (C) for which there is a test. One arrives
in this way at a standard one-counter shift scM (C) such that C(scM (C)) = C, and
one observes a perfect reciprocity between a class of Markov codes and the class of
standard one-counter shifts.
3 b. Behavior under topological conjugacy
In this subsection we assume that we are given a strongly synchronizing subshift
X Z with a characteristic pair (( )iZ , (+ )iZ ) of xed points and a subshift
X that is given by a
X Z together with a topological conjugacy : X
:
such that
one-block map
synchro (X).

(3.11)
1 (synchro (X))
We introduce notation that we use in this situation. We set (
)iZ = 1 (( )iZ ),
1
(
+ )iZ = ((+ )iZ ). [L, L] will denote a coding window of 1 and will be
that gives 1 . Q N will be chosen such that
a block map : L2L+1 (X)
+
+

+
for a synchronizing word a of X and for a
Q (a), a Q (a) the word a aa
contains a synchronizing symbol.
(X),
b (
For

Q+L ), we can set by Lemma 2.1 and by (3.11)
b
(
) = b (b
) (b
)a (b
),


where the words b (b
) and a (b
) and the symbol (b
) are uniquely

under the condition that (b


) is synchronizing and that
determined by b
a (b
) does not contain a synchronizing symbol. We set
) = (a (b
)).
I (b

d ) that is in D( , ),
Denoting by d (b
d ) the longest prex of a (b
)(
we have a mapping
d ),
d D(
,
).
: d d (b

Denote by Db ( ) the set of d D( (b


), ) such that the prex of
Q+L

b
) (b
)d
is equal to (
) and
length Q + L + 1 of the word b (b

Q+2L

such that the prex of length Q+2L+1 of the word (b (b


) (b
)d
)
is a sux of b
. We use corresponding symbols with a time symmetric meaning.
We dene H , H+ Z+ by
(3.12)

) = H cX H+ .
(c

+
X

(X),
b (
Lemma 3.3. For

Q+L ), the mapping
b
is a bijec ) onto Db ( ).
tion of D(
,

102
10

WOLFGANG KRIEGER AND KENGO MATSUMOTO

Proof. By construction
b (D(
,
)) Db ( ),
and one conrms that the inverse of b is given by the mapping that assigns to
a d Db ( ) the word that is obtained by removing the prex of length Q + 1
2L+1
from the longest prex of the word (b (b
) (b
)d
) that does not

end in
.
+
Lemma 3.4. Let +
(X) = . Then also (X) = .
+
+

Proof. If there were a

,
+ ), then one would
(X) and a d+ D(

) that
have for a b Q+L (
+
b (d
), + ).
+ ) D( (b


+
+
+ +
+ ), one has
Lemma 3.5. For d+
reset (X), b L+Q (

d+ (d+
+b+ )+ (
+b+ ) +
reset (X).
Proof. One has
+b+ )+ (
+b+ )) D+ (d+
+ ) + I+ (
+b+ ) + (d+ ) (d+ (d+
+b+ )).
D+ (d+ (d+


b+ + (
+
(X),
Lemma 3.6. Let d+

L+Q + ), and let

d+ + (
+b+ ) +
reset (X)

(3.13)
and
(3.14)

k+ > 2L

be such that
k +D+ (d+ + (
+
b+ )) +

+ b+ ) = cX +
d+
(d
+
+

(3.15)
Then

d .

d+
+

reset (X).

Proof. By (3.13)
L+l +H

b+ )) +
l +k+ +D+ (d+ + (
+

d + (
+b+ ) L(X),

cX ++

l , l+ N,

and by (3.14) and (3.15) there is a k+ N such that the word


L+l +H

b+ )) +
l +k+ +D+ (d+ + (
+

d + (
+b+ )b+ (
+b+ ))

cX ++

contains the word


l

as a subword for l , l+ N.

+ +
l +k
+
d

++

cX

b+
Lemma 3.7. For d+
+ + (X),
+

d
+

+
+ ),
L+Q (


one has

+
reset (X)

if and only if
+b+ )+ (
+b+ ) +
d+ (d+
reset (X).

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

Proof. This follows from Lemma 3.5 and Lemma 3.6.

103
11

also satises the reset


Lemma 3.8. Let X satisfy the reset condition. Then X
condition.
Proof. It follows from Lemma 3.6 that there is a bound on the length of the

words in (X)\

reset (X).
We note that the converse of Lemma 3.8 also holds.
has reset if and only if X has reset.
Proposition 3.9. X
and b+ + (
+ + (X)
Proof. Let d+ + + (X). To obtain
L+Q + )
+
+
+
+

such that d + D+b+ (+ ), let a L+Q (+ ) and let


+ b equal to the rst
subword of length Q + L + 1 of
2L+1 +
(+
d + a + )

that begins with a synchronizing symbol. Apply Lemma 3.3 and Lemma 3.7.

is also a standard
Lemma 3.10. Let X be a standard one-counter shift. Then X
one-counter shift.

Proof. By Lemma 3.4 +


(X) is empty and by Lemma 3.8 X satises the reset
condition. Let I, D , M , M+ , D+ , , J , J+ , + be parameters for X. Let
I >I + 2Q + +6L + M + M+ + J + J+ + (CX )
+ 2 max{( d ) | (X), d D( , )}
(3.16)

+ 2 max{(d+ + ) | d+ + + (X)}

) | d+ + (X)\
+ max{(d+ +
reset (X)}

+ max (X),d D( , ) ( d )
+ max d+ + +

counter (X)

(3.17)
(3.18)

= M + H ,
M
J = J + H ,
(
D
d ) =

(3.19)

+ = M+ + H+ ,
M
J+ = J+ + H+ ,
{D ( (b
)d (b
d ))

d )) + (d ) + I (b
)},
(d (b
d D(

(X),
,
),
+ (d+
D
+ ) =

(3.20)

max

b
)
Q+2L (

+ (d+ + ),

min

{I+ (
+b+ ) + (d+ ) (d+ (d+
+b+ ))

b+ +
+ )
L+Q (

+b+ )+ (
+b+ ))},
+ D+ (d+ (d+

d+
+ + (X),
reset

104
12

WOLFGANG KRIEGER AND KENGO MATSUMOTO

and
(
d ) =

( (b
)d (b
d )) + (d )

)
b
Q+L (

(3.21)

+ (d+
+ ) =

d )) + I (b
)),
(d (b

(X),
d D(


,
),

I+ (
+b+ ) (d+ (d+
+b+ )) + (d+ )

b+ +
+ )
L+Q (

(3.22)

+b+ )+ (
+b+ )),
+ + (d+ (d+

+ d+ +
(X).
counter

We prove that
| (

{
c C(X)
c) = I}

(X)

(X)
(X)
, M
, M
+, D
+ ) Ccounter
, J , J+ ,
+ ).
Creset (D
(

of length I,
let
+ t(
c).
Given a word c C(X)
be the rst symbol of c and let
Also let
b (
b+ + (
Q+L ),
L+Q + ),
and let a word c L(X) be given by
b c
(
+b+ ) = b (b
)c+ (
+b+ )b+ (
+b+ ).
By (3.16)
(X)
(X)
c C (+ (
+b+ )) Creset (D , M , M+ , D+ ; + (
+b+ ))
(X)

+b+ )).
Ccounter ( , J , J+ , + ; + (
In the case that
(X)

+b+ )),
c C (+ (

(3.23)

one has + (
+b+ )
+ (X), and there are
), ),
d D( (b

d+
+b+ )),
D( , + (

and k N such that


d+
+b+ ) (X)\
reset (X),
+ (

k
c = (b
)d d+
.

By (3.16) k > 2L, and it is seen from the action of that one has, setting
(d ),
d = 1

1
d+
(d+ ),
=

b+
+

and
(3.24)

k = k I (b
) + (d ) (d ) H ,

that

k
c =
d
d+
.

Here

+ (X)\
d+

reset (X),

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

105
13

for otherwise one would have by Lemma 3.5 a contradiction to (3.23). This means
that

(X)
c C (
+ ).
In the case that
(X)
cC
(D , M , M+ , D+ ; + (
+b+ )),
reset

there are

), ),
d D( (b
and k , k+ N such that

k
k
c = (b
)d cX ++ d+ ,

+b+ ) +
d+ + (
reset (X),
(3.25)

d+
+b+ )),
D(+ , + (

D ( (b
)d ) + k + M M+ + k+ + D+ (d+ + (
+b+ )).

Set again
(d ),
d = 1

(3.26)

and also set


d+ = 1
(d+ ),

+
b+

(3.27)

k
1
(c + d+ ).
d+
=

b+ X +
+

If here

+ (X)\
d+
reset (X),

(3.28)

then by Lemma 3.6 k+ 2L, and then by (3.16) k > 2L, and it is seen from the
action of that, with k given by the expression (3.24),

d+
c =
d
.
By (3.28) this means that

(X)

c C (
+ ).
If here

d+
+

reset (X),
one has by (3.16) and (3.25) that k > 2L and it is seen from the action of that,
with k given by the expression (3.24), and with
(3.29)
k+ = k+ H (d+ ) + (d+ ) I+ (
+b+ ),
that one has then

k
k
c =
d
cX
++ d+ .

By (3.25), (3.19) and (3.20)


M
+ + k+ + D
+ (d+
(
d ) + k + M
+ ),
D
and this means that

(X)

, M
, M
+, D
+;
c Creset (D
+ ).
In case that
(X)

c Ccounter ( , J , J+ , + ; + ),
there are

d D( (b
), ),

d+ D(+ , + (
+b+ )),

106
14

WOLFGANG KRIEGER AND KENGO MATSUMOTO

and

)d ),
D+ + (d+ + (
+b+ )),
D ( (b
and k , k+ N such that
(3.30)
d+ + (
+b+ ) +
counter (X),
)d cX + k+ d+ ,
c = (b
k

(3.31)

D + k + J = J+ + k+ + D+ .
By (3.16) and (3.31) k , k+ > 2L, and with d , d+ , k , k+ given by the expressions
(3.26),(3.27), (3.24), (3.29) and with
= D + (d ) (d ) + I (b
D
(3.32)
),
+ = I+ (
+b+ ) + (d+ ) (d+ ) + D+ ,
D

(3.33)

it is seen from the action of that

c =
d
cX
++ d+ .
By (3.32) and (3.33)
+,
+ k + J = J+ + k+ + D
D
and by Lemma 3.7 and by (3.30) this means that

(X)
, J , J+ ,
+;
c Ccounter (
+ ).

We prove that

(X)

{
c C

(3.34)

C(X).

| (
c) = I}

(X),
and for a word c C (X)
with the rst symbol
+
) of length I,
For

(
+

, there are

d D(
,
),
d+
+ ,
+
),
D(

and k N such that

d+
+
(X)\

reset (X),

(3.35)

d+
c =
d
.
Let

b (
Q+L ),

b+ + (
Q+L + ),

and let a word c L(X) in the symbols of be given by


(3.36)
(b c
+b+ ) = b (b
)c (
b+ ).
+

From (3.36) it is seen that there is a k N such that


k
+ b+ ).
)d (b
d ) d+
c = (b
(d
+

If here

+ b+ ) (
d+
(d
+
+ + b ) (X)\reset (X),
then by (3.16), k > 2L and if here
d+ (d+
b+ ) (
b+ ) (X),

then there are d


+

+
D(+ , +
(
+ b ))

d+ +
+
reset (X),

reset

and k+ N such that


+ b+ ) = cX k+ d+ .
d+
+
(d
+

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

107
15

By Lemma 3.6 and by (3.35) k+ 2L, and then by (3.16) k > 2L, and also
D ( (b
)d ) + k + M M+ + k+ + D+ (d+ + (
+b+ )),
and therefore
k
(X)
ccX ++ d+ Creset (D , M , M+ , D+ ; + (
+b+ )).

By (3.16) then
k

ccX ++ d+ C(X),
and it is seen from the action of that the word c is a subword of the word
k
)ccX + d+ + (
+b+ )b+ (
+b+ )) L(X),
(b (b
+

and (3.34) is conrmed.


We prove that
(3.37)

(X)

, M
, M
+, D
+ ) | (
C(X).

{
c Creset (D
c) = I}

+ (X)
and for a word
For
+

(X)
, M
, M
+, D
+;
+ )
c Creset (D

of length I with the rst symbol


there are
d D(
,
),

d+ D(
+ ,
+ ),

and k , k+ N such that


(3.38)
(3.39)

+
d+
reset (X),

(
M
+ + k+ + D
+ (d+
D
d ) + k + M
+ ),

cX
++ d+ .
c =
d
By (3.12), (3.17), (3.19) and (3.20) one can select
b (
Q+L ),

b+ + (
Q+L + )

such that
(
D
d ) = D ( (b
)d (b
d )) (d (b
d )) + (d ) + I (b
),
+ (d+
+ ) = I+ (
+b+ ) + (d+ ) (d+ (d+
+b+ )) + D+ (d+ (d+
+b+ )+ (
+b+ )),
D
and such that one has with
(3.40)
) (d (b
d )) + (d ) + k + H ,
k = I (b
(3.41)

k+ = H+ + k+ + (d+ ) (d+ (d+


+b+ )) + I+ (
+b+ ),

that
(3.42)
D ( (b
)d (b
d )) + k + M M+ + k+ + D+ (d+ (d+
+b+ )+ (
+b+ )).
By (3.38) and Lemma 3.7 and by (3.42) it follows for the word c in the symbols of
that is given by
b c
(
+b+ ) = b (b
)c+ (
+b+ )b+ (
+b+ ),
that
k
k
(X)
+
c = (b
)d (b
d ) cX ++ d+ (d+
+
+b+ )).
b ) Creset (+ (

108
16

WOLFGANG KRIEGER AND KENGO MATSUMOTO

By (3.16) then c C(X). By (3.16) and (3.42) k , k+ > 2L and from the action of
it is seen that the word c is a subword of the word
(b (b
)c+ (
+b+ )b+ (
+b+ )) L(X),
and (3.37) is conrmed.
We prove that
(3.43)

(X)

, J , J+ ,
+ ) | (
C(X).

{
c Ccounter (
c) = I}

+ (X)
and a word
For
+

(X)
, J , J+ ,
+;
+ )
c Ccounter (

of length I with a rst symbol


there are
,
),
d D(

d+ D(
+ ,
+ ),


(
D
d ),

+
+ (d+
D
+ ),

and
and k , k+ N such that

+
d+
counter (X),
+.
+ k + J = J+ + k+ + D
D

(3.44)
(3.45)

By (3.12), (3.18), (3.21) and (3.22) one can select


b (
Q+L ),

b+ + (
Q+L + ),

such that there are


D ( (b
)d (b
d )),
+b+ )+ (
+b+ )),
D+ + (d+ (d+
such that
= D (d (b
D
d )) + (d ) + I (b
),
+
+
+
+
+
+ = I+ (
+b ) + (d ) (d (d
+b )) + D+ .
D
With k , k+ N given by the expressions (3.40) and (3.41) then
(3.46)

D + k + J = J+ + k+ + D+ .

By (3.44) and Lemma 3.7 and by (3.45) it follows for the word c in the symbls of
that is given by
b c
(
+b+ ) = b (b
)c+ (
+b+ )b+ (
+b+ ),
that
k
k
(X)
+
)d (b
d ) cX ++ d+ (d+
+
+b+ )).
c = (b
b ) Ccounter (+ (

By (3.16) then c C(X). By (3.16) and (3.46) k , k+ > 2L and from the action of
it is seen that the word c is a subword of the word

)c+ (
+b+ )b+ (
+b+ )) L(X),
(b (b
and (3.43) is conrmed.
D
, M
, M
+, D
+,
, J , J+ ,
+ are parameters for
We have shown that I,

X.


SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

109
17

3 c. Shifts of standard one-counter type


One has a theorem that can be viewed as analogous to Theorem 1.1.
Theorem 3.11. Let X Z be a subshift that is topologically conjugate to a
standard one-counter shift. Then there exists an n N such that X [1,n] is a
standard one-counter shift,
X [1,n] = scM (C (X

[1,n]

n n .

),

Proof. Apply Lemma 2.2 and Lemma 3.10.

One can view the class of standard one-counter shifts as extending the class
of topological Markov shifts and one is then lead to introduce a class of subshifts
of standard one-counter type as the class of subshifts that have a higher block
system that is a standard one-counter shift. Theorem 3.11 is then equivalent to the
statement that a subshift that is topologically conjugate to a subshift of standard
one-counter type is itself of standard one-counter type.
4. -graph systems and C -algebras
Consider a -graph system L = (V, E, , ) over alphabet with vertex set
V = lZ+ Vl , edge set E = lZ+ El,l+1 , labeling map : E and shift-like
map that is given by surjective maps l,l+1 : Vl+1 Vl , l Z+ . A subset V of V
is called hereditary if all v V such that (v) V are in V, and if v V then all
initial vertices of all edges that have v as a nal vertex are also in V. A hereditary
subset V is said to be proper if V Vl = Vl for all l N.
l
Let us denote by {v1l , . . . , vm(l)
} the vertex set Vl at level l. For i = 1, 2, . . . , m(l), j =
1, 2, . . . , m(l + 1), , we put

1 if s(e) = vil , (e) = , t(e) = vjl+1 for some e El,l+1 ,
Al,l+1 (i, , j) =
0 otherwise,

1 if l,l+1 (vjl+1 ) = vil ,
Il,l+1 (i, j) =
0 otherwise.
The C -algebra OL associated with L is the universal C -algebra generated by
partial isometries S , and projections Eil , i = 1, 2, . . . , m(l), l Z+ subject
to the following operator relations called (L):

S S = 1,

m(l)

m(l+1)

Eil

= 1,

Eil =

i=1

j=1

S S Eil
m(l+1)
S Eil S

Il,l+1 (i, j)Ejl+1 ,

Eil S S ,

Al,l+1 (i, , j)Ejl+1 ,

j=1

for i = 1, 2, . . . , m(l), l Z+ , [18].


For a subshift X Z we recall the construction of its predicting -graph
system XL. The label set of XL is and its vertex set is
V (X) = lZ+ Vl (X)

110
18

WOLFGANG KRIEGER AND KENGO MATSUMOTO

where V0 (X) contains the singleton set that contains the empty word, and where

Vl (X) = {+
l (x ) | x X(,0] },

l N.

All edges of L leave a vertex in lN Vl (X), and a vertex v lN Vl (X) has an


outgoing edge that carries the label if and only if v contains a word that
begins with and the target vertex of this outgoing edge is equal to {a +
[1,l) |
a v}, l N. The mapping
X

: lN Vl (X) lZ+ Vl (X)


deletes last symbols.
Theorem 4.1. Let X Z be a standard one-counter shift with a characteristic
pair (( )iZ , (+ )iZ ) of xed points. Then
(i) V (X) has a proper hereditary subset if and only if X has no reset.
(ii) LX has a proper hereditary subset.
Proof. (i) Let +
reset (X) = . Let I, D , M , M+ , D+ be parameters for
X, where I is chosen such that scM ({c L(X) | (c) I}) is aperiodic and
topologically transitive subshift of nite type with alphabet . Let Q N be such
that for ,  synchro (X) there exists for q > Q an admissible concatenation of
words in {c L(X) | (c) I} that begins with and that can be followed by  .
With D > I such that also
D > (cX ) + M + M+ + (d ) + D ( d ),

, d D( , ),

one has for x X(,0] , that +


D (x ) contains a synchronizing symbol. Let

x X(,0] , l N. One can choose a word a L(X) of length less than l + D


such that
+

+
l (x ) = l (a)

and for y X(,0] one has that +


l+2D+Q (y ) contains a word with sux a. It
follows that V (X) has no proper hereditary subset.
l
In case that +
reset (X) = one has {+ } Vl (X), l N and it follows that the
l
set lN Vl (X)\{+ } is a proper hereditary subset of V (X).
l
}) is a proper hereditary subset of V (X).

(ii) Here lN (Vl (X)\{

Corollary 4.2. Let X be a standard one-counter shift. Then


(i) OXL is simple if and only if X has reset.
(ii) OLX is not simple.
Proof. There exists a bijective correspondence between hereditary subsets of
the vertex set V and ideals in the C -algebra OL ([18], [20]).

For the notion of ow equivalence see [3, 6, 22, 23]. For a subshift Y Z
and for ,   , we say that the subshift Y  ( {  })Z is obtained from
the subshift Y by replacing in Y by  if for every admissible word a of Y  there
is an admissible word a of Y such that a can be obtained by replacing in a the
symbol by the word  and then, if necessarily, still removing the rst symbol or
the last symbol or both. We say then also that Y is obtained from Y  by replacing
in Y  the word  by the symbol .

 Z are ow equivalent if there exists a chain of
Subshifts X Z and X
subshifts

Y [q] [q]Z , 1 q Q, Q N,
Y [1] = X, Y [Q] = X,

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

111
19

such that Y [q] is topologically conjugate to Y [q + 1] or Y [q + 1] is obtained from


Y [q] by replacing in Y [q] a symbol by the word  or Y [q] is obtained from
Y [q + 1] by replacing in Y [q + 1] a symbol by the word  . We remark at this
point that the denition of a standard one-counter shift can be given a more general
formulation in which the characteristic pair of xed points are replaced by a pair of
periodic points. In this way one arrives at a class of subshifts that is closed under
ow equivalence.
Corollary 4.3. A standard one-counter shift with reset is not ow equivalent
to its inverse.
Proof. The ideal structure of the C -algebra OXL is an invariant of ow equivalence [17]. Apply Theorem 4.1.

5. K-groups
We will compute the K-groups and the Bowen-Franks groups of the one-counter
shift
(N )

m k
sc((Creset )rev ) = sc({an +
| 1 n N, m, k N, m k}),
(N )

that is of the canonical future -graph system of sc((Creset )rev ) or, equivalently, of
(N )
the canonical past -graph system of sc(Creset ). The set up that we choose is for the
(N )
canonical future -graph system of sc((Creset )rev ). Let (M, I) = (Ml,l+1 , Il,l+1 )lZ+
(N )
be the canonical symbolic matrix system of sc((Creset )rev ) (the canonical future
(N ) rev
-graph system of sc((Creset ) )). Let (M, I) = (Ml,l+1 , Il,l+1 )lZ+ be its nonnegative matrix system. The entries of the nonnegative matrix Ml,l+1 count the
number of symbols of the corresponding entries of Ml,l+1 . We denote by m(l) the
row size of Ml,l+1 , so that the both matrices Ml,l+1 and Il,l+1 are m(l) m(l + 1)
matrices. They satisfy the following relations
Il,l+1 Ml+1,l+2 = Ml,l+1 Il+1,l+2 ,

l Z+ .

t
t
t
, l Z+ the homomorphism from Zm(l) /(Ml1,l
Il1,l
)Zm(l1)
We denote by Il,l+1
t
t
t
to Zm(l+1) /(Ml,l+1
Il,l+1
)Zm(l) induced by Il,l+1
. Then as in [15]
(N )

(5.1)

t
t
t
Il,l+1
)Zm(l) , Il,l+1
},
K0 (sc((Creset )rev )) = lim{Zm(l+1) /(Ml,l+1

(5.2)

(N )
K1 (sc((Creset )rev ))

t
t
t
= lim{Ker(Ml,l+1
Il,l+1
) in Zm(l) , Il,l+1
}.

Let ZI be the group of the projective limit lim{Zm(l) , Il,l+1 }. The sequence Ml,l+1

l
Il,l+1 , l Z+ acts on it as an endomorphism, denoted by M I. The Bowen-Franks
(N )
groups BF i (sc((Creset )rev )), i = 0, 1, are dened by
(N )

BF 0 (sc((Creset )rev )) = ZI /(M I)ZI ,


(N )

BF 1 (sc((Creset )rev )) = Ker(M I)

in
(N )

ZI .

We denote the symbols + , in the subshifts sc((Creset )rev ) now by b, c respectively. For l N, consider the following subsets {Fil }i=1,...,2l+2 of the right one-sided

112
20

WOLFGANG KRIEGER AND KENGO MATSUMOTO


(N )

shift sc((Creset )rev )[1,) .

(N )

F1l ={(xn )nN sc((Creset )rev )[1,) | x1 = b, x2 = = xl+2 = c,


xl+3 = ai for some 1 i N },
(N )

F2l ={(xn )nN sc((Creset )rev )[1,) | x1 = b, x2 = = xl+1 = c,


xl+2 = ai for some 1 i N },
..
.
(N )

l
={(xn )nN sc((Creset )rev )[1,) | x1 = b, x2 = c, x3 = ai for some 1 i N },
Fl+1
(N )

l
Fl+2
={(xn )nN sc((Creset )rev )[1,) | x1 = ai for some 1 i N },
(N )

l
Fl+3
={(xn )nN sc((Creset )rev )[1,) | x1 = c, x2 = ai for some 1 i N },
(N )

l
={(xn )nN sc((Creset )rev )[1,) | x1 = x2 = c, x3 = ai for some 1 i N },
Fl+4

..
.
(N )

l
={(xn )nN sc((Creset )rev )[1,) | x1 = xl = c, xl+1 = ai for some 1 i N }.
F2l+2

(N )

The sets {Fil }i=1,...,2l+2 are the l-past equivalence classes of sc((Creset )rev ). Put
m(l) = 2l + 2. Let vil , i = 1, . . . , m(l) be the vertex set Vl of the canonical -graph
(N )
rev
(N )
system Lsc((Creset ) ) for the subshift sc((Creset )rev ). The vertex vil is considered
l
l
to be the class [Fi ] of Fi . For a symbol , if Fjl+1 is contained in Fil , then a
labeled edge labeled from the vertex vil to the vertex vjl+1 is dened in the l
graph system. Hence there are labeled edges labeled an , n = 1, . . . , N from vl+2
to
l+1
l+1
l
vj for j = 1, 2, . . . , l + 2. There are labeled edges labeled b from vi to v2l+4i and
to vil+1 for i = 1, 2, . . . , l + 1. There are labeled edges labeled c from vil to vil+1 for
l+1
l+1
l
i = l + 3, l + 4, . . . , 2l + 2, and from v2l+2
to v2l+3
and to v2l+4
.
l+1
l+1
l
If Fj is contained in Fi , the -map is dened by (vj ) = vil . Hence we have

v1
l+1
l
(vj ) = vj1

l
v2l+2

if j = 1,
if j = 2, 3, . . . , 2l + 3,
if j = 2l + 4.

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

113
21

We will consider the symbolic matrix system Ml,l+1 , Il,l+1 on the ordered bases
l
. For i = 1, . . . , m(l) and j = 1, . . . , m(l + 1), we have
F1l , , Fm(l)

a1 + + aN if i = l + 2, j = 1, 2, , l + 2,

b
if 1 i = j l + 1,

b
if i + j = 2l + 5, 1 i l + 1,
Ml,l+1 (i, j) =

c
if l + 3 i = j 2l + 2,

c
if i = 2l + 2, j = 2l + 3, 2l + 4,

0
otherwise,

1 if i = j = 1,

1 if 2 j = i + 1 2l + 3,
Il,l+1 (i, j) =

1 if i = 2l + 2, j = 2l + 4,

0 otherwise.
Hence we have

so that

1
t
Ml,l+1
(i, j) =

if j = l + 2, i = 1, 2, , l + 2,
if 1 i = j l + 1,
if i + j = 2l + 5, 1 j l + 1,
if l + 3 i = j 2l + 2,
if j = 2l + 2, i = 2l + 3, 2l + 4,
otherwise,

1
t
t
Ml,l+1
(i, j) Il,l+1
(i, j) = 1

0 .................... 0
N
0 ....................
1 1
0 .......... 0
N
0 . . . . . . . . . . . . . . . . . . . .

0 1 1 0 . . . . . . 0
N
0 . . . . . . . . . . . . . . . . . . . .

. . . 0 1 1 0 . . . 0
N
0 . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . 0 1 1
N
0 . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . 0 1 N
0 . . . . . . . . . . . . . . . . . . . .
.

=
0 . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . 0 1 1
. . . . . . . . . . . . . . . . . . . . 0
1
0 1 1 0 . . . . . . . . . . .

. . . . . . . . . . . . . . . . 0 1
0
0
0 1 1 0 . . . . . . . .

. . . . . . . . . . . . .

.............

. . . . . . . .

.......................

. . . 0
1 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 1 1

0
1
0 .......................................... 0
1
0 ............................................... 0

t
t
Ml,l+1
Il,l+1

if j = l + 2, i = 1, 2, , l + 2,
if 2 i = j 2l + 2, i = l + 2,
if i + j = 2l + 5, 1 j l + 1,
if 2 i = j + 1 2l + 2,
otherwise,

114
22

WOLFGANG KRIEGER AND KENGO MATSUMOTO

We see that
t
t
Il,l+1
)=0
Lemma 5.1. Ker(Ml,l+1

for

2 l N.

Thus we have by (5.2),


(N )
Lemma 5.2. K1 (sc((Creset )rev ))
= 0.
(N )

We will next compute K0 (sc((Creset )rev )). Set for i = 1, . . . , 2l + 4, j =


1, . . . , 2l + 2

if j = l + 2, i = 1,

if 2 i = j 2l + 2, i = l + 2,
1
Bl,l+1 (i, j) = 1
if (i, j) = (2l + 4, 1), (2l + 3, 2),

1 if 2 i = j + 1 l + 2,

0
otherwise.
That is

Bl,l+1

0 .................... 0 N 0 ................
1 1
0 .......... 0
0 0 . . . . . . . . . . . . . . . .

0 1 1 0 . . . . . . 0
0 0 . . . . . . . . . . . . . . . .

. . . 0 1 1 0 . . . 0
0 0 . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . 0 1 1
0 0 . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . 0 1 0 0 . . . . . . . . . . . . . . . .
.

=
0 1 0 . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . 0
. . . . . . . . . . . . . . . . . . . . 0
0
0 0 1 0 . . . . . . . . .

. . . . . . . . . . . . . . . . 0 0
0
0 0 0 1 0 . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . .

...................

. . . 0
0 0 .......................... 0 0 1

0
1
0 .................................... 0
1
0 ......................................... 0

Let Pl be the (2l + 2) (2l + 2) matrix dened by setting for i, j = 1, . . . , 2l + 2,

if i = j,
1
Pl (i, j) = 1 if j = 1, i = 2, . . . , l + 1,

0
otherwise.
We know that
(5.3)

t
t
Pl+1 (Ml,l+1
Il,l+1
)Z2l+2 = Bl,l+1 Z2l+2 .

t
t
Denote by Pl+1 the induced isomorphism from Z2l+4 /(Ml,l+1
Il,l+1
)Z2l+2 onto
2l+4
2l+2
Z
/Bl,l+1 Z
. Let Jl,l+1 be the (2l + 4) (2l + 2) matrix dened by setting
for i = 1, . . . , 2l + 4, j = 1, . . . , 2l + 2,

1 if i = j = 1,

1 if i = j 1, i = 2, . . . , 2l + 3,
Jl,l+1 (i, j) =

1 if i = 2l + 4, j = 2l + 2,

0 otherwise.

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

115
23

Denote by Jl,l+1 the induced homomorphism from Z2l+2 /Bl1,l Z2l into Z2l+4 /Bl,l+1 Z2l+2 .
Lemma 5.3. The diagram
t
Il,l+1

t
t
t
t
Z2l+2 /(Ml1,l
Il1,l
)Z2l Z2l+4 /(Ml,l+1
Il,l+1
)Z2l+2

Pl+1 
Pl 

Z2l+2 /Bl1,l Z2l

Jl,l+1

Z2l+4 /Bl,l+1 Z2l+2

is commutative.
For an integer n, we denote by q(n) Z the quotient of n by N and by
r(n) {0, 1, . . . , N 1} its residue such as n = q(n)N + r(n). The following lemma
is straightforward.

z1

Lemma 5.4. Fix l = 2, 3, . . . . For z = ... Z2l+4 , put inductively


z2l+4
x1 = z2l+4 ,
x2 = z2l+3 ,
xk = zk

for k = l + 3, l + 4, . . . , 2l + 2,

xl+2 = q(z1 ),
xl+1 = zl+2 ,
xl = zl+1 zl+2 ,
xlk = zlk+1 zlk+2 zl+2 ,

for k = 1, 2, . . . , l 3.

Set
rl,l+1 (z) = r(z1 ) {0, 1, . . . , N 1},
l,l+1 (z) = z2 z2l+3 + z2l+4 ,
l,l+1 (z) = z3 + z4 + z5 + + zl+2 + z2l+3 .
Then we have

rl,l+1 (z)

l,l+1 (z)

z1
x1

..
.. l,l+1 (z)
.
. = Bl,l+1 . +
0

..
z2l+4
x2l+2

.
0

The following lemma is also direct.


2l+4
Lemma 5.5. For z = [zi ]2l+4
, one has
i=1 Z

rl,l+1 (z) = 0 in {0, 1, . . . , N 1}


if and only if there exists y =

[yi ]2l+2
i=1

and

l,l+1 (z) = l,l+1 (z) = 0 in Z

Z2l+2 such that z = Bl,l+1 (y).

2l+4
Lemma 5.6. The map l+1 : [zi ]2l+4
(rl,l+1 (z), l,l+1 (z), l,l+1 (z))
i=1 Z
{0, 1, . . . , N 1} Z Z induces an isomorphism from Z2l+4 /Bl,l+1 Z2l+2 onto
Z/N Z Z Z.

116
24

WOLFGANG KRIEGER AND KENGO MATSUMOTO

Proof. It suces to show the surjectivity of l+1 . For (g, m, k) {0, 1, . . . , N


1} Z Z, put z = [g, m, k, 0, . . . , 0]t Z2l+4 . One then sees that
rl,l+1 (z) = g,

l,l+1 (z) = m,

l,l+1 (z) = k.


We denote by l+1 the above isomorphism from Z2l+4 /Bl,l+1 Z2l+2 onto Z/N Z
Z Z induced by l+1 .
Lemma 5.7. The diagram
Jl,l+1

Z2l+2 /Bl1,l Z2l Z2l+4 /Bl,l+1 Z2l+2

l+1 
l 
Z/N Z Z Z

1 0
is commutative, where L = 0 0
0 1

0
0 .
1

Z/N Z Z Z

2l+2
Proof. For z = [zi ]2l+2
, it is direct to see that
i=1 Z

rl,l+1 (Jl,l+1 (z)) = rl1,l (z),

l,l+1 (Jl,l+1 (z)) = 0,

l,l+1 (Jl,l+1 (z)) = l1,l (z) + l1,l (z).



Therefore we conclude
Lemma 5.8. K0 (sc((Creset )rev ))
= Z/N Z Z.
(N )

Proof. By Lemma 5.1, it follows that


(N )

K0 (sc(sc((Creset)rev )) = lim{Z2l+4 /Bl,l+1 Z2l+2 , I t l,l+1 }

= lim{Z/N Z Z Z, L}

=Z/N Z Z.

(N )

(N )

As the torsion free part of K0 (sc((Creset )rev )) is not isomorphic to K1 (sc((Creset )rev )),
these types of K-groups can not appear in those of soc systems.
(N )
(N )
We next compute the Bowen-Franks groups BF 0 (sc((Creset )rev )) and BF 1 (sc((Creset )rev )).
As in [15, Theorem 9.6], one sees the following formulae of short exact sequences
of the universal coecient type theorem:
(N )

0 Ext1Z (Ki (sc((Creset )rev )), Z)


(N )

BF i (sc((Creset )rev ))
(N )

HomZ (Ki+1 (sc((Creset )rev )), Z) 0.


The sequences split unnaturally.
Lemma 5.9. BF 0 (sc((Creset )rev ))
= Z/N Z,
(N )

BF 1 (sc((Creset )rev ))
= Z2 .
(N )

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

117
25

Proof. Since for a nitely generated abelian group G, HomZ (G, Z) is the
torsion free part of G and Ext1Z (G, Z) is the torsion part of G, one gets the desired
assertions by Lemma 5.8.

(N )

(N )

As the torsion free part of BF 0 (sc((Creset )rev )) is not isomorphic to BF 1 (sc((Creset )rev )),
these types of Bowen-Franks groups can not appear in those of soc systems. We
restate Lemma 5.2, Lemma 5.8 and Lemma 5.9 as

Theorem 5.10.
(N )
K0 (sc((Creset )rev ))
=Z/N Z Z,
(N )
BF 0 (sc((Creset )rev ))
=Z/N Z,

(N )
K1 (sc((Creset )rev ))
= 0,
(N )
BF 1 (sc((Creset )rev ))
= Z2 .
(N )

We will next compute the K-groups for sc(Creset ). The computation is completely similar to the above one as in the following way. We can take the l(N )
(N )
past equivalence classes of sc(Creset ) as the similar ones to the sc((Creset )rev ).
Let (M, I) = (Ml,l+1 , Il,l+1 )lZ+ be the canonical symbolic matrix system for
(N )
sc(Creset ). We see that

a 1 + + aN

b
Ml,l+1 (i, j) =

if i = j = l + 2,
if 1 i = j l + 1,
if i + j = 2l + 5, 1 i l + 1,
if l + 3 i = j 2l + 2,
if i = 2l + 2, j = 2l + 3, 2l + 4,
otherwise.
(N )

Dierent from the symbolic matrix system for sc((Creset )rev ) is only the l + 2-th
(N )
row in Ml,l+1 . The matrix Il,l+1 is the same as the one for sc((Creset )rev ). Let
(Ml,l+1 , Il,l+1 )lZ+ be its nonnegative matrix system. Hence we have

1
t
t
Ml,l+1
(i, j) Il,l+1
(i, j) = 1

if i = j = l + 2,
if 2 i = j 2l + 2, i = l + 2,
if i + j = 2l + 5, 1 j l + 1,
if 2 i = j + 1 2l + 2,
otherwise.

By considering the kernels and cokernels of the following matrices Bl,l+1 , l N


dened by

N
if i = j = l + 2,

1
if
2 i = j 2l + 2, i = l + 2,

Bl,l+1 (i, j) = 1
if (i, j) = (2l + 4, 1), (2l + 3, 2),

1 if i = 2, j = 1,

0
otherwise,

118
26

WOLFGANG KRIEGER AND KENGO MATSUMOTO

that is

0 ............... 0 0 0 ................
1 1 0 . . . . . . . . 0 0 0 . . . . . . . . . . . . . . . .

0 0 1 0 . . . . 0 0 0 . . . . . . . . . . . . . . . .

. . . 0 0 1 0 . 0 0 0 . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . 0 0 1 0 0 . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . 0 0 N 0 . . . . . . . . . . . . . . . .
.

. . . . . . . . . . . . . . . . . . . . 0 0 1 0 . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 0 0 0 0 1 0 . . . . . . . . .

. . . . . . . . . . . . . 0 0 0 0 0 0 1 0 . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . 0 0 0 . . . . . . . . . . . . . . . . . . . . . . 0 0 1

0 1 0 ................................ 0
1 0 .................................... 0

Bl,l+1

We can similarly show that


K1 (sc(Creset ))
=0
(N )

and
(N )

K0 (sc(Creset )) = lim{Z2l+4 /Bl,l+1 Z2l+2 , I t l,l+1 }

= lim{Z/N Z Z Z, L}

=Z/N Z Z.
Therefore we have
Theorem 5.11.
K0 (sc(Creset ))
= K0 (sc((Creset )rev ))
= Z/N Z Z,
(N )

(N )

(N )
(N )
K1 (sc(Creset ))
= K1 (sc((Creset )rev ))
= 0.
(N  )

Corollary 5.12. For N, N  N, N = N  , sc(Creset ) and sc(Creset ) are not


ow equivalent to each other.
(N )

Proof. K-groups are invariants of ow equivalence ([16]).

References
1. F. Blanchard, Syst`
emes dynamiques topologiques associ
es `
a des automates r
ecurrents, Z.
Wahrscheinlichkeitstheor. Verw. Geb. 58(1981), pp. 549564.
2. F. Blanchard and G. Hansel, Syst`
emes cod
es, Theor. Computer Sci. 44 (1986) pp. 1749.
3. R. Bowen and J. Franks, Homology for zero-dimensional nonwandering sets, Ann. Math.
106 (1977) pp. 7392.
4. P. C. Fischer, Turing machines with restricted memory access, Information and Control
9(1966), pp.364379.
5. P. C. Fischer, A. R, Meyer, and A. L. Rosenberg, Counter machines and counter languages, Math. Systems Theory 2(1968), pp. 265283.
6. J. Franks, Flow equivalence of subshifts of nite type, Ergodic Theory Dynam. Systems 4
(1984) pp. 5366.
7. J. E. Hopcroft and J. D. Ullman, Introduction to Automata Theory, Languages, and
Computation, Addison-Wesley, Reading (2001).

SUBSHIFTS AND C -ALGEBRAS FROM ONE-COUNTER CODES

119
27

8. G. Keller, Markov extensions, zeta functions, and Fredholm theory for piecewise, invertible
dynamicl systems, Tans. Amer. Math. Soc. 314(1989), pp. 433497.
9. G. Keller, Circular codes, loop counting, and zeta-functions, J. Combinatorial Theory 56
(1991), pp. 7583.
10. B. P. Kitchens, Symbolic dynamics, Springer-Verlag, Berlin, Heidelberg and New York
(1998).
11. W. Krieger and K. Matsumoto, Shannon graphs, subshifts and lambda-graph systems, J.
Math. Soc. Japan 54(2002), pp. 877899.
12. W. Krieger and K. Matsumoto, in preparation.
13. D. Lind and B. Marcus, An introduction to symbolic dynamics and coding, Cambridge
University Press, Cambridge (1995).
14. K. Matsumoto, A simple C -algebra arising from a certain subshift, J. Operator Theory
42(1999), pp. 351-370.
15. K. Matsumoto, Presentations of subshifts and their topological conjugacy invariants, Doc.
Math. 4 (1999), pp. 285-340.
16. K. Matsumoto, Bowen-Franks groups as an invariant for ow equivalence of subshifts, Ergodic Theory Dynam. Systems 21(2001), pp. 18311842.
17. K. Matsumoto, Bowen-Franks groups for subshifts and Ext-groups for C -algebras, K-Theory
23(2001), pp. 67104.
18. K. Matsumoto, C -algebras associated with presentations of subshifts, Doc. Math. 7(2002),
pp. 130.
19. K. Matsumoto, Construction and pure inniteness of the C -algebras associated with graph systems, Math. Scand. 97(2005), pp. 7389.
20. K. Matsumoto, C -algebras associated with presentations of subshifts II, -ideal structures
and lambda-graph subsystems-, J. Australian Mathematical Society 81(2006), pp. 369385.
21. K. Matsumoto, A class of simple C -algebras arising from certain nonsoc subshifts,
preprint.
22. W. Parry and D. Sullivan, A topological invariant for ows on one-dimensional spaces,
Topology 14 (1975) pp. 297299.
23. J. Thamsborg, Flow equivalence between substitutional dynamical systems, Masters Thesis,
Department of Mathematics, University of Copenhagen, August 2006.
Institute for Applied Mathematics, University of Heidelberg, Im Neuenheimer Feld
294, 69120 Heidelberg, Germany
E-mail address: krieger@math.uni-heidelberg.de
Department of Mathematical Sciences, Yokohama City University, 22-2 Seto, Kanazawaku, Yokohama, 236-0027 Japan
E-mail address: kengo@yokohama-cu.ac.jp

This page intentionally left blank

Contemporary Mathematics
Contemporary Mathematics
Volume 503, 2009

Orbit equivalence in C -algebras dened by actions of


symbolic dynamical systems
Kengo Matsumoto
Abstract. A C -symbolic dynamical system is a nite family { } of
endomorphisms of a C -algebra A with some conditions. The endomorphisms
yield a C -algebra O from the associated Hilbert C -bimodule. If A is commutative, O contains a commutative C -algebra C(X ) with a transformation
. The class of dynamical systems (X , ) include both one-sided topological Markov shifts and homeomorphisms on compact Hausdor spaces. In this
paper, we will prove that orbit equivalence class of the dynamical systems
(X , ) exactly correspond to the isomorphism class of the pairs (C(X ), O )
of C -algebras.

1. Introduction
H. Dye has initiated to study of orbit equivalence of ergodic nite measure preserving transformations, who proved that any two such transformations are orbit
equivalent ([9], [10]). W. Krieger [18] has proved that two ergodic non-singular
transformations are orbit equivalent if and only if the associated von Neumann
crossed produtcs are isomorphic. In topological setting, Giordano-Putnam-Skau
[13], [14] (cf. [15]) have proved that two Cantor minimal systems are strong orbit
equivalent if and only if the associated C -crossed products are isomorphic. In more
general setting, J. Tomiyama [36] (cf. [2], [37] ) has proved that two topological
free homeomorphisms (X, ) and (Y, ) on compact Hausdor spaces are continuously orbit equivalent if and only if there exists an isomorphism between the
associated C -crossed products keeping their commutative C -subalgebras C(X)
and C(Y ), which is also equivalent to an existence of a homeomorphism h : X Y
preserving their topological full groups. Orbit equivalence of continuous maps on
compact Hausdor spaces that are not homeomorphisms are not covered by the
above Tomiyamas setting. In [29], the author has shown that similar results to the
Tomiyamas results hold for one-sided topological Markov shifts, that are continuous surjections but not homeomorphisms. He has proved that one-sided topological
Markov shifts (XA , A ) and (XB , B ) for matrices A and B with entries in {0, 1}
2000 Mathematics Subject Classication. Primary 46L55; Secondary 46L35, 37B10.
Key words and phrases. C -algebra, orbit equivalence, symbolic dynamical system, subshift,
topological Markov shift, -graph system, Cuntz-Krieger algebra.
The author was supported by JSPS Grant-in-Aid for Scientic Reserch (N0. 20540215).
1
121

c
2009
American Mathematical Society

122
2

KENGO MATSUMOTO

are continuously orbit equivalent if and only if there exists an isomorphism between
the Cuntz-Krieger algebras OA and OB keeping their commutative C -subalgebras
C(XA ) and C(XB ) ( Note that the term topological orbit equivalence has been
used in [29] instead of continuous orbit equivalence). It is also equivalent to
an existence of a homeomorphism from XA to XB preserving their continuous full
groups [A ]c and [B ]c . In [30], these results for one-sided topological Markov
shifts have been generalized to the class of topological dynamical systems of graph systems. A -graph system L is a generalization of a nite labeled graph
and presents a subshift. It yields a topological dynamical system (XL , L ) of a
zero-dimensional compact Hausdor space XL with shift transformation L , that
is a continuous surjection but not a homeomorphism. The C -algebra OL is associated with (XL , L ) such that C(XL ) is naturally embedded into OL as the
diagonal algebra of the canonical AF-algebra FL inside OL . It has been shown
that the topological dynamical systems (XL1 , L1 ) and (XL2 , L2 ) for -graph systems L1 and L2 are continuously orbit equivalent if and only if there exists an
isomorphism between the associated C -algebras OL1 and OL2 keeping their commutative C -subalgebras C(XL1 ) and C(XL2 ) ([30]). It is also equivalent to an
existence of a homeomorphism from XL1 to XL2 preserving their continuous full
inverse semigroups [L1 ]sc and [L1 ]sc .
A C -symbolic dynamical system, introduced in [27], is a generalization of both
a -graph system and an automorphism of a unital C -algebra. It is a nite family
{ } of endomorphisms of a unital C -algebra A such that the closed ideal
generated by (1), coincides with A. A nite labeled graph G gives rise to
a C -symbolic dynamical system (AG , G , ) such that AG = CN for some N N.
A -graph system L gives rise to a C -symbolic dynamical system (AL , L , )
such that AL is C(L ) for some compact Hausdor space L with dimL = 0.
A C -symbolic dynamical system (A, , ) yields a subshift denoted by over

, {u } ) with an orthogonal nite


and a Hilbert C -right A-module ( , HA

basis {u } and a unital faithful diagonal left action : A L(HA


). By
using general construction of C -algebras from Hilbert C -bimodules established

by M. Pimsner [32], a C -algebra denoted by O from ( , HA


, {u } ) has been
introduced in [27]. We call the algebra O the C -symbolic crossed product of A by
the subshift . If A = C(X) with dimX = 0, there exists a -graph system L such
that the subshift is presented by L and the C -algebra O is the C -algebra OL
associated with L. If in particular, A = CN , the subshift is a soc shift and O
is a Cuntz-Krieger algebra. If = {} an automorphism of a unital C -algebra
A, the C -algebra O is the ordinary C -crossed product A  Z.
In the rst half of this paper, a condition called (I) on (A, , ) will be studied as
a generalization of both the condition (I) on a -graph system [26](cf. [25]) (and
hence on a nite matrix of Cuntz-Krieger [7] ) and the topological freeness of a
homeomorphism on a compact Hausdor space dened by Tomiyama([35]). Under
the assumption that (A, , ) satises condition (I), the simplicity conditions of the
algebra O will be discussed in Section 4.
Let F be the xed point algebra of O under gauge action . Assume that A
is commutative. The diagonal algebra denoted by D of F is a commutative C algebra C(X ) with a continuous surjection on X . The topological dynamical
systems (X , ) contain both the class of dynamical systems (XL , L ) for -graph

ORBIT EQUIVALENCE IN C -ALGEBRAS

123
3

systems L (and hence one-sided topological Markov shifts) and the class of dynamical systems (X, ) of homeomorphisms on compact Hausdor spaces X. In the
second half of this paper, we will prove
Theorem 1.1. Let (A1 , 1 , 1 ) and (A2 , 2 , 2 ) be C -symbolic dynamical systems satisfying condition (I). Suppose that both the algebras A1 and A2 are commutative. The following are equivalent:
(i) There exists an isomorphism : O1 O2 such that (D1 ) = D2 .
(ii) (X1 , 1 ) and (X2 , 2 ) are continuously orbit equivalent.
(iii) There exists a homeomorphism h : X1 X2 such that h[1 ]sc h1 =
[2 ]sc , where [i ]sc denotes the continuous full inverse semigroup of i
consisting of partial homeomorphisms on Xi .
The following are special cases.
Corollary 1.2 ([29]). Let A and B be irreducible square matrices with entries in {0, 1} satisfying condition (I) in [7]. Then their respect right one-sided
topological Markov shifts (XA , A ) and (XB , B ) are continuously orbit equivalent
if and only if there exists an isomorphism : OA OB between their respect
Cuntz-Krieger algebras such that (DA ) = DB , where DA and DB are the diagonal
algebras isomorphic to C(XA ) and C(XB ) of the canonical AF algebras inside OA
and inside OB respectively.
Corollary 1.3 ([36] (Tomiyama)). Let (X, ) and (Y, ) be topological free
homeomorphisms on compact Hausdor spaces X and Y respectively. Then the
dynamical systems (X, ) and (Y, ) are continuously orbit equivalent if and only
if there exists an isomorphism : C(X)  Z C(Y )  Z between their respect
C -crossed products such that (C(X)) = C(Y ).
In what follows, we will denote by Z+ and N the set of nonnegative integers
and the set of positive integers respectively.
2. C -algebras associated with -graph systems
Throughout the paper, denotes a nite set with its discrete topology, that
is called an alphabet. Each element of is called a symbol or a label. Let Z be
the innite product space of over Z endowed with the product topology. The
transformation on Z given by ((xi )iZ ) = (xi+1 )iZ is called the full shift over
. Let be a shift invariant closed subset of Z i.e. () = . The topological
dynamical system (, | ) is called a two-sided subshift, written as for brevity.
Let L = (V, E, , ) be a -graph system over with vertex set V = lZ+ Vl
and edge set E = lZ+ El,l+1 that is labeled with symbols in by a map : E ,
and that is supplied with surjective maps (= l,l+1 ) : Vl+1 Vl for l Z+ . Here
the vertex sets Vl , l Z+ are nite disjoint sets. Also El,l+1 , l Z+ are nite
disjoint sets. An edge e in El,l+1 has its source vertex s(e) in Vl and its terminal
vertex t(e) in Vl+1 respectively. Every vertex in V has a successor and every vertex
in Vl for l N has a predecessor. It is then required that for vertices u Vl1
and v Vl+1 , there exists a bijective correspondence between the set of edges
e El,l+1 such that t(e) = v, (s(e)) = u and the set of edges f El1,l such
that s(f ) = u, t(f ) = (v), preserving thier labels ([24]). We henceforth assume
that L is left-resolving, which means that t(e) = t(f ) whenever (e) = (f ) for

124
4

KENGO MATSUMOTO

l
e, f El,l+1 . Let us denote by {v1l , . . . , vm(l)
} the vertex set Vl at level l. For
i = 1, 2, . . . , m(l), j = 1, 2, . . . , m(l + 1), , we put

1 if s(e) = vil , (e) = , t(e) = vjl+1 for some e El,l+1 ,
Al,l+1 (i, , j) =
0 otherwise,

1 if l,l+1 (vjl+1 ) = vil ,
Il,l+1 (i, j) =
0 otherwise.

The C -algebra OL associated with L is the universal C -algebra generated by


partial isometries S , and projections Eil , i = 1, 2, . . . , m(l), l Z+ subject
to the following operator relations called (L):

S S = 1,

m(l)

m(l+1)

Eil

= 1,

Eil =

i=1

j=1

S S Eil
m(l+1)
S Eil S

Il,l+1 (i, j)Ejl+1 ,

Eil S S ,

Al,l+1 (i, , j)Ejl+1 ,

j=1

for i = 1, 2, . . . , m(l), l Z+ , . If L satises -condition (I) and is irreducible, the C -algebra OL is simple and purely innite ([26], [25]).
Let AL,l be the C -subalgebra of OL generated by the projections Eil , i =
1, . . . , m(l). We denote by AL the C -subalgebra of OL generated by the all projections Eil , i = 1, . . . , m(l), l Z+ . We denote by : AL,l AL,l+1 the natural
inclusion. Hence the algebra AL is the inductive limit limAL,l of the inclusions.

For , put

L
for x AL .
(x) = S xS
L
Then
{L
} yields a family of -endomorphisms of AL such that (1) = 0,

L
L
(1) 1 and for any nonzero x AL , (x) = 0 for some .

3. C -symbolic dynamical systems and their crossed products


Let A be a unital C -algebra. In what follows, an endomorphism of A means
a -endomorphism of A that does not necessarily preserve the unit 1 of A. For an
alphabet , a nite family of nonzero endomorphisms , of A is said to
be essential if the closed ideal generated by (1), coincides with A. It is
said to be faithful if for any nonzero x A there exists a symbol such that
(x) = 0.
Definition ([27]). A C -symbolic dynamical system is a triplet (A, , ) consisting of a unital C -algebra A and an essential and faithful nite family { }
of endomorphisms of A.
Two C -symbolic dynamical systems (A, , ) and (A ,  ,  ) are said to be
isomorphic if there exist an isomorphism : A A and a bijection : 
such that = () for all . For an automorphism of a unital

ORBIT EQUIVALENCE IN C -ALGEBRAS

125
5

C -algebra A, by setting = {}, = the triplet (A, , ) becomes a C symbolic dynamical system. A C -symbolic dynamical system (A, , ) yields a
subshift over such that a word 1 k of is admissible for if and only
if k 1 = 0 ([27, Proposition 2.1]). We say that a subshift acts on a
C -algebra A if there exists a C -symbolic dynamical system (A, , ) such that
the associated subshift is .
Let G = (G, ) be a left-resolving nite labeled graph with underlying nite
directed graph G = (V, E) and labeling map : E (see [23, p.76]). Denote
by v1 , . . . , vN the vertex set V . Assume that every vertex has both an incoming
edge and an outgoing edge. Consider the N -dimensional commutative C -algebra
AG = CE1 CEN where each minimal projection Ei corresponds to the vertex
vi for i = 1, . . . , N . Dene an N N -matrix for by

1 if there exists an edge e from vi to vj with (e) = ,
G
A (i, , j) =
0 otherwise
N
for i, j = 1, . . . , N . We set G (Ei ) = j=1 AG (i, , j)Ej for i = 1, . . . , N, .
Then G , dene endomorphisms of AG such that (AG , G , ) is a C -symbolic
dynamical system such that the subshift G is the soc shift G presented by G.
Conversely, for a C -symbolic dynamical system (A, , ), if A is CN , there exists
a left-resolving labeled graph G such that A = AG and = G the soc shift
presented by G ([27, Proposition 2.2]).
More generally let L be a -graph system (V, E, , ) over . We equip each
vertex set Vl with discrete topology. We denote by L the compact Hausdor space

with dimL = 0 of the projective limit V0 V1 V2 as in [25, Section


2]. Since the algebra C(Vl ) of all continuous functions on Vl is the commutative
nite dimensional algebra AL,l , the C -algebra AL is the commutative AF-algebra
C(L ). We then have a C -symbolic dynamical system (AL , L , ) such that the
subshift L coincides with the subshift L presented by L. Conversely, for a C symbolic dynamical system (A, , ), if the algebra A is C(X) with dimX = 0, there
exists a -graph system L over such that the associated C -symbolic dynamical
system (AL , L , ) is isomorphic to (A, , ) ([27, Theorem 2.4]). We remark that
another generalization of soc shifts are in [22] by using endomorphisms of nite
dimensional linear spaces.
The C -algebra O has been originally constructed in [27] as a C -algebra by
using the Pimsners general construction of C -algebras from Hilbert C -bimodules
[32] (cf. [16] etc.). It is called the C -symbolic crossed product of A by the subshift
, and realized as the universal C -algebra C (x, S ; x A, ) generated by
x A and partial isometries S , subject to the following relations called ():

S S = 1,
S S x = xS S ,
S xS = (x)

for all x A and . Assume that A is commutative. Then we know ([27,


Theorem 4.2])
(i) If A = C, the subshift is the full shift Z , and the C -algebra O is
the Cuntz algebra O|| of order ||.
(ii) If A = CN for some N N, the subshift is the soc shift G presented
by a left-resolving labeled graph G, and the C -algebra O is the CuntzKrieger algebra OG associated with the labeled graph G.

126
6

KENGO MATSUMOTO

(iii) If A = C(X) with dimX = 0, there exists a -graph system L such that
the subshift is the subshift presented by L and the C -algebra O is
the C -algebra OL associated with the -graph system L.
Let be an automorphism of a unital C -algebra A. Put = {} and = .
The C -algebra O for the C -symbolic dynamical system (A, , ) is the ordinary
C -crossed product A  Z.
4. Condition (I) for C -symbolic dynamical systems
The notion of condition (I) for nite square matrices with entries in {0, 1}
has been introduced in [7]. The condition gives rise to the uniqueness of the associated Cuntz-Krieger algebras under the canonical relations of the generating
partial isometries. The condition has been generalized by many authors to corresponding conditions for generalizations of the Cuntz-Krieger algebras, for instance,
innite directed graphs ([21]), innite matrices with entries in {0, 1} ([11]), Hilbert
C -bimodules ([16]), etc. (see also [17], [33], etc.). The condition (I) for -graph
systems has been also dened in [25], [26] to prove the uniqueness of the C -algebra
OL under the canonical relations of generators. In this section, we will introduce
the notion of condition (I) for C -symbolic dynamical systems to prove the uniqueness of the C -algebras O under the relation (). The condition is needed to show
the main result (Theorem 1.1) of this paper. In [16], a condition called (I)-free has
been introduced. The condition (I)-free is similar condition to our condition (I).
The discussions given in [16] is also similar ones to ours in this section. We will give
complete descriptions in our discussions for the sake of completeness. In what follows, for a subset F of a C -algebra B, we will denote by C (F ) the C -subalgebra
of B generated by F .
Let (A, , ) be a C -symbolic dynamical system over and the associated
subshift . We denote by Bk () the set of admissible words of with length
|| = k. Put B () =
k=0 Bk (), where B0 () denotes the empty word. Let
S , be the partial isometries in O satisfying the relation (). For =
(1 , . . . , k ) Bk (), we put S = S1 Sk and = k 1 . In the
algebra O , we set
F

= C (S xS : , B (), || = ||, x A),

Fk

= C (S xS : , Bk (), x A) for k Z+ ,

= C

(S xS


and

: B (), x A).

identity S xS = S (x)S
for x A
k
the algebra F is embedded into the algebra Fk+1

The
and , Bk () holds so
such that kZ+ Fk is dense
that
in F . The gauge action of the circle group T = {z C | |z| = 1} on O is dened
by z (x) = x for x A and z (S ) = zS for . The xed point algebra of O
under is denoted by (O ). Let E : O (O ) be the conditional expectaton
dened by

E (X) =
z (X)dz,
X O .
zT

It is routine to check that (O ) = F .


Definition . A C -symbolic dynamical system (A, , ) satises condition (I)
if there exists a unital increasing sequence
A0 A 1 A

ORBIT EQUIVALENCE IN C -ALGEBRAS

127
7

of C -subalgebras of A such that (Al ) Al+1 for all l Z+ , , the union


k
lZ+ Al is dense in A and for > 0, k, l N with k l and X0 F,l
=


C (S xS : , Bk (), x Al ), there exists an element g D Al (= {y


D | ya = ay for a Al }) with 0 g 1 such that
(i) X0 k (g) X0 ,

m

(ii) gm
(g) = 0 for all m = 1, 2, . . . , k, where (X) =
Bm () S XS .
As the element g belongs to the diagonal subalgebra D of F , the condition (I)
of (A, , ) is intrinsically determined by (A, , ) by virtue of Lemma 4.1 below.
We remark that the rst condition above is equivalent to the following equality
X0 = sup{ X0 k (g) | g D Al  , 0 g 1}.
If a -graph system L over satises condition (I), then (AL , L , ) satises
condition (I) (cf. [25, Lemma 4.1]). In [28], the author had introduced condition
(I) for (A, , ). The condition is stronger than the condition (I) dened above.
Let B be a unital C -algebra. Suppose that there exist an injective unital
-homomorphism : A B and a family s B, of partial isometries
satisfying

s s = 1,
s s (x) = (x)s s ,
s (x)s = ( (x))

for all x A and . Put A = (A) and ((x)) = ( (x)), x A. It is easy


 , ) is a C -symbolic dynamical system such that the
to see that the triple (A,
presented subshift is the same as the one (= ) presented by (A, , ). Let
O,s be the C -subalgebra of B generated by (x) and s for x A, . Let F,s
be the C -subalgebra of O,s generated by s (x)s for x A and , B ()
with || = ||. By the universality of the algebra O , the correspondence

x A (x) A,

S s ,

extends to a surjective -homomorphism


: O O,s .
Lemma 4.1. The restriction of
to the subalgebra F is a -isomorphism from
 , ).
F to F,s . Hence if (A, , ) satises condition (I), so does (A,
is injective on Fk . Let us assume that

 Proof. It suces to show that

,Bk () s (x, )s = 0 for


,Bk () S x, S F with x, A. For
, Bk (), it follows that

( (1)x, (1)) = s (
s (x, )s )s = 0.
,Bk ()

As : A 
B is injective, one has (1)x, (1) = 0 so that S x, S = 0. This

implies that ,Bk () S x, S = 0.
We henceforth assume that (A, , ) satises condition (I) dened above. Take
a unital increasing sequence {Al }lZ+ of C -subalgebras of A as in the denition
of condition (I). We set for k l
k
= C (S xS : , Bk (), x Al ).
F,l


k
k


There exists an inclusion relation F,l
F,l
 for k k and l l . Let P,s be the
-subalgebra of O,s algebraically generated by (x), s for x Al , l Z+ , .

128
8

KENGO MATSUMOTO

Lemma 4.2. For x P,s such that




x s + x0 +
s x
x=
||1

||1

where x , x0 , x P,s F,s , , B (), one has x0 x .


k
(F,l
) and the words , are all
Proof. We may assume that x , x0 , x
k
 , ) satises condition
in n=0 Bn () for some k, l with k l. By Lemma 4.1, (A,
(I). For any > 0, one may nd g
(D ) (Al ) with 0 g 1 such that
(i) x0 k (g) x0 ,


m
(ii) g m
(g) = 0 for all m = 1, 2, . . . , k, where (g) =
Bm () s gs .
1

Put h = g 2 . We then have


x k (h)xk (h)


=
k (h)x s k (h) + k (h)x0 k (h) +
k (h)s x k (h) .
||1

||1

Since h commutes with (Al ), one easily sees that k (h) commutes with x so
that k (h)x s k (h) = x k (h)s k (h). It then follows that
k (h)s k (h) 2

= k (h)s k (g)s k (h) = k (g)s k (g)s = k||


(||

(g)g)s s = 0
so that k (h)x s k (h) = 0 and similarly k (h)s x k (h) = 0. As one sees that
k (h)x0 k (h) = x0 k (g), one obtains
x k (h)xk (h) = x0 k (g) x0 .
Hence we get x x0 .

By a similar argument to [7, 2.8 Proposition], one sees


Corollary 4.3. There exists a conditional expectation E,s : O,s F,s
such that E,s
=
E .
Therefore we have
Proposition 4.4. The -homomorphism
: O O,s dened by

(x) = (x),

x A,

(S ) = s ,

becomes a surjective -isomorphism, and hence the C -algebras O and O,s are
canonically -isomorphic through
.
Proof. The map
: F F,s is -isomorphic and satises E,s
=

E . Since E : O F is faithful, a routine argument shows that the homomorphism


: O O,s is actually a -isomorphism.

Hence the following uniqueness of the C -algebra O holds.
Theorem 4.5. Assume that (A, , ) satises condition (I). The C -algebra
O is the unique C -algebra subject to the relation (). This means that if there

ORBIT EQUIVALENCE IN C -ALGEBRAS

129
9

exist a unital C -algebra B and an injective -homomorphism : A B and a


family s B, of nonzero partial isometries satisfying the following relations:

s s = 1,
s s (x) = (x)s s ,
s (x)s = ( (x))

for all x A and , then the correspondence


x A (x) B,

S s B

extends to a -isomorphism
from O onto the C -subalgebra O,s of B generated
by (x), x A and s , .
As a corollary we have
Corollary 4.6. Assume that (A, , ) satises condition (I). For any nontrivial ideal I of O , one has I A = {0}.
Proof. Suppose that I A = {0}. Hence S  I for all . By Theorem
4.5, the quotient map q : O O /I must be injective so that I is trivial.

For a C -symbolic dynamical system (A,
, ), let : A A be the completely positive map on A dened by (x) = (x) for x A. Then (A, , )
is said to be irreduible if there exists no nontrivial ideal of A invariant under .
Corollary 4.7. Assume that (A, , ) satises condition (I). If (A, , ) is
irreducible, the C -algebra O is simple.
For a homeomorphism on a compact Hausdor space X, let be the induced
automorphism of C(X) dened by
(f )(x) = f ( 1 (x)),

f C(X),

x X.

Then the pair (C(X), ) naturally gives rise to a C -symbolic dynamical system
over = { }. The notion of topological freeness of a homeomorphism has been
introduced by Tomiyama in [35]. A homeomorphism on X is said to be topologically free if the set of aperiodic points is dense in X.
Lemma 4.8. Let be a homeomorphism on a compact Hausdor space X.
If (X, ) is topological free, then the associated C -symbolic dynamical system
(C(X), , { }) satises condition (I).
k
= F = D = C(X)
Proof. Put Al = A = C(X) for all l Z+ . One has F,l


k
and D Al = C(X) C(X) = C(X). For k l, > 0 and f0 F,l
= C(X)
there exists an aperiodic point x0 X such that

|f0 (x0 )| f0
because the set of aperiodic points is dense in X. One may nd a neighborhood U
of x0 such that U n (U ) = for all n = 1, 2, . . . , k. Take g C(X) with
0 g 1 such that
g( k (x0 )) = 1,

supp(g) k (U ).

It then follows that


f0 k (g) |f0 (x0 )g( k (x0 ))| f0 .
m
As supp(m
(supp(g)) km (U ), one has gm
(g)) =
(g) = 0 for 1 m
k.


130
10

KENGO MATSUMOTO

By [36, Theorem 5.4] with Corollary 4.6, we have


Proposition 4.9. Let be a homeomorphism on a compact Hausdor space X.
Then (X, ) is topological free if and only if the associated C -symbolic dynamical
system (C(X), , { }) satises condition (I).
5. Cartan subalgebra D of O
A C -symbolic dynamical system (A, , ) for which A is commutative is called
a commutative C -symbolic dynamical system. In what follows, we will consider
such situations as A = C() for some compact Hausdor space . We denote by
the character for dened by (f ) = f (), f C(). As (1) for
is a projection in A, one sees that ( (1)) = 1 or 0. Put the clopen set
H = { | ( (1)) = 1}.
As is a character for H , there uniquely exists h () such that
= h ()

for H .

The essentiality for (A, , ) implies H = , and the faithfulness for (A, , )
implies h (H ) = . Hence we have
Proposition 5.1. A commutative C -symbolic dynamical system (A, , ) with
A = C() for some compact Hausdor space bijectively corresponds to a family
h , of continuous maps h : H from clopen sets H to such that
H = h (H ) = .
:= Hc j the complement
Let = {1 , 2 , . . . , N }. Put H1 j := Hj and H1
j

of Hj . Let V1 be the set of partitions dened by the sets H1 j , H1


, j = 1, . . . , N ,
j
that is
V1 = {Hi11 Hi22 HiNN | i1 , i2 , . . . , iN = 1, 1}.

1
}. For l N, let Vl+1
be the set of partitions
We denote V1 by {v11 , v21 , . . . , vm(1)

1 l
dened by the sets hj (vi ), i = 1, . . . , m(l), j = 1, . . . , N . We denote Vl+1
by

l+1 l+1
l+1
1
{v1 , v2 , . . . , vm(l+1) }. Since h () = H , , Vl+1 is a renement of Vl .
l
: disjoint union. For , if h (vjl+1 ) vil , dene a
Hence = v1l v2l vm(l)

directed edge from vil to vjl+1 labeled . The labeling is denoted by . We denote

such labled edged from Vl to Vl+1


. If vjl+1 vil , dene (vjl+1 ) = vil . We
by El,l+1
then have a -graph system
L = (V , E , , )
over . We note that the vertex sets V = {vil : i = 1, . . . , m(l), l Z+ } generate
the topology of if and only if C() is AF, which is also equivalent to the situation
that the C -symbolic dynamical system (C(), , ) comes from a -graph system
L.
A subset E is dened by
E = {(u, , v) | u = h (v)},
which is a compact Hausdor space in a natural topology. Put

( ) | (ui , i+1 , ui+1 ) E for i N, u1 H1 }.
X := {(i , ui )iN
iN

ORBIT EQUIVALENCE IN C -ALGEBRAS

131
11

For x = (i , ui )iN X , we write h1 (u1 ) as u0 (x). It is direct to see that


D = C(X ).
Lemma 5.2. Assume that (A, , ) satises condition (I). Then we have D 
O F .
Proof. Take an arbitrary X D  O . For B (), we put
X = E (S X),

X = E (XS ).

Suppose that X = 0 and X = 1 for some B (). For f D , as


S f S D , one sees
X S f S = f X .

(5.1)

Since X S S = X , it follows that by (5.1)




S f S = X S f S = f X .
(5.2)
X ||
(f ) = X S S
||=||
k
For > 0, take k l and X0 F,l
with X0 = 1 such that X X0 < . By
(5.2), one sees

(5.3)

||
f X0 X0 ||
(f ) f X0 X + X0 X (f ) 2 f .

By condition (I), one may take g D Al as in the denition of condition (I) for
k
k
and > 0. Put Q = k (g) D . As Q commutes with F,l
, by (5.3),
X0 F,l
one has
X0 Q X0 ||
(Q) 2 .

(5.4)
||

||

||

k
and Q (Q) = k (g (g)) = 0, one obtains
Since (Q) commutes with F,l
||
X0 Q X0 ||
(Q) = max{ X0 Q , X0 (Q) } X0 Q 1

a contradiction with (5.4) for < 14 . Hence we conclude that X = 0 and similarly
X = 0 for all B () and then X belongs to F .

Lemma 5.3. D  F = D .
Proof. For k N, let Dk be the diagonal subalgebra C (S aS ; a A,

Bk ()) of Fk . Put P = S S for Bk () and Ek (X) = Bk () P XP for
X Fk . We then have Ek (X) Dk for X Fk . This means that Ek : Fk Dk is
a conditional expectation. Since Ek+1 |Fk = Ek , the sequence Ek , k N gives rise to
a conditional expectation E : F D such that E|Fk = Ek . Now for X D  F ,
one has Ek (X) = X for all k N. This implies that E(X) = limk Ek (X) = X.
As E(F ) = D , one sees that X D .

Hence we have
Proposition 5.4. D is a maximal commutative -subalgebra of O .
A commutative C -subalgebra D of a C -algebra B is called a Cartan subalgebra of B if it satises the following properties:
(i) D is a maximal commutative -subalgebra of B,
(ii) there exists a conditional expectation from B onto D, and

132
12

KENGO MATSUMOTO

(iii) D is regular in B, which means that the normalizer semigroup


Ns (B, D) = {v B | v is a partial isometry; vDv D, v Dv D}
generate B as a C -algebra (cf. [33]).
Corollary 5.5. D is a Cartan subalgebra of O .
Proof. As in the proof of Lemma 5.3, there exists a conditional expectation
from F onto D . By composing the conditional expectation E from O onto F , we
have a conditional expectation from O onto D . The generating partial isometries
S , and the unitaries in A all belong to Ns (O , D ), which generate O . 
6. Full inverse semigroups and normalizers
For a commutative C -symbolic dynamical system (A, , ) with A = C(), let
X be the compact Hausdor space satisfying D = C(X ) dened in the preceding
section. Let : X X be the continuous map dened by ((i , ui )iN ) =
(i+1 , ui+1 )iN . For x X , the orbit orb (x) of x is dened by

k
l
orb (x) =
k=0 l=0 ( (x)) X .

Let [ ] be the set of all homeomorphisms on X such that (x) orb (x) for all
x X . Let [ ]c be the set of all in [ ] such that there exist continuous maps
k, l : X Z+ satisfying
(6.1)

k(x) ( (x)) = l(x) (x)

for all x X .

We call [ ]c the continuous full group for (X , ). Let : U V be a homeomorphism from a clopen set U X onto a clopen set V X . We call a partial
homeomorphism. Let us denote by X and Y the clopen sets U and V respectively. We denote by P H(X ) the set of all partial homeomorphisms of X . Then
P H(X ) has a natural structure of inverse semigroup (cf. [31], [33]). Let [ ]s be
the set of all partial homeomorphisms P H(X ) such that (x) orb (x) for
all x X . Let [ ]sc be the set of all in [ ]s such that there exist continuous
maps k, l : X Z+ satisfying
(6.2)

k(x) ( (x)) = l(x) (x)

for all x X .

We call [ ]sc the continuous full inverse semigroup for (X , ). The maps k, l
above are called orbit cocycles for , and sometimes written as k , l respectively.
The orbit cocyles are not necessarily uniquely determined for . It is clear that
[ ]s is a subsemigroup of P H(X ) and [ ]sc is a subsemigroup of [ ]c .
l
Recall Vl = {v1l , . . . , vm(l)
} the clopen partitions of dened in the preceding
section. Let be the subshift dened by (A, , ). For = (1 , . . . , k )
Bk (), vil , we set the cylinder set
U,vil = {(n , un )nN X | 1 = 1 , . . . , k = k , uk vil }.
Since vil is a clopen set in , U,vil is clopen in X .
Lemma 6.1. For = (1 , . . . , k ) Bk () and vil Vl with 2 k l and
U,vil = , there exists ,vil [ ]sc such that
(6.3)

,vil (x) = (x)

for x U,vil .

ORBIT EQUIVALENCE IN C -ALGEBRAS

133
13

Proof. Put = (2 , . . . , k ) Bk1 (). The map ,vil : U,vil U,vil


dened by ,vil (x) = (x) for x U,vil is a partial homeomorphism, which

belongs to [ ]sc .
Lemma 6.2. For x = (n , un )nN X , put u0 = u0 (x) . Let 0 be a
symbol such that (n1 , un1 )nN X . Then there exists [ ]sc with a clopen
set X X such that x X and (y) = (yn1 )nN for all y = (yn )nN X ,
where y0 = (0 , u0 (y)).
Proof. Let X be the clopen set U,vil for = (1 , 2 ) B2 () and vil =
V2 , where u2 = (ul2 )lZ+ , so that x belongs to X . One has (yn1 )nN
X for (yn )nN X , where y0 = (0 , u0 (y)). By setting (y) = (yn1 )nN for
y = (yn )nN X , we have ( (y)) = y for y X so that [ ]sc .


u22

For x X , put [ ]sc (x) = { (x) X | [ ]sc with X  x}.


Lemma 6.3. [ ]sc (x) = orb (x).
Proof. For any [ ]sc with X  x, one sees (x) orb (x) and hence
[ ]sc (x) orb (x). For the other inclusion relation, by the previous lemmas, for
x = (xn )nN X and x0 = (0 , u0 (x)) , there exist 1 , 2 [ ]sc such
that
2 (x) = (xn+1 )nN X
1 (x) = (xn1 )nN ,
so that both (xn1 )nN and (xn+1 )nN belong to [ ]sc (x). Since [ ]sc is a semigroup, one sees that
[ ]sc (x)  (xk , . . . , x1 , x0 , xl+1 , xl+2 , . . . , )
for all k, l Z+ . Hence [ ]sc (x) orb (x).

Let H be the Hilbert space with complete orthonormal system ex , x X . For


, dene partial isometry s on H by setting

ey if u0 (x) H ,
s ex =
0
otherwise,
where x = (i , ui )
i=1 X , y = ((, u0 (x)), (1 , u1 ), (2 , u2 ), . . . ) X . For
f C() = A, dene a bounded operator (f ) on H by setting
(f )ex = f (u0 (x))ex

for x X .

It is direct to see that the relation


s (f )s = ( (f ))
for , f C()

holds. Since one easily sees that s s = 1 and s s (f ) = (f )s s for
f C(), , the C -algebra O is faithfully represented on the Hilbert space
H if (A, , ) satises condition (I) by Theorem 4.5. We henceforth identify S , f
with s , (f ) respectively and assume that O is represented on H .
Let us denote by U(O ) the group of unitaries of O and U(D ) the group
of unitaries of D respectively. As in [29], the continuous full group [ ]c will
correspond to the normalizer N (O , D )(= {v U(O ) | vD v = D }) of D in
O . For the continuous full inverse semigroup [ ]sc , we will consider the normalizer
Ns (O , D )(= {v O | v is a partial isometry ; vD v D , v D v D }) of
partial isometries. It is easy to see that Ns (O , D ) has a natural structure of

134
14

KENGO MATSUMOTO

inverse semigroup. We will identify the subalgebra D of O with the algebra


C(X ). For a partial isometry v O , put Ad(v)(x) = vxv for x O . The
following proposition holds.
Proposition 6.4. For [ ]sc , there exists a partial isometry u Ns (O , D )
such that
Ad(u )(f ) = f 1

for f C(X ),

Ad(u )(g) = g

for g C(Y ),

and the correspondence [ ]sc u Ns (O , D ) is a homomorphism of


inverse semigroup. If in particular [ ]c , the partial isometry u is a unitary
so that u N (O , D ).
Proof. Put the subspaces
HX = span{ex | x X },

HY = span{ey | y Y }.

Since : X Y is a homeomorphism, the operator u : HX HY dened


by u (ex ) = e (x) for x X yields a partial isometry on H . By a similar manner
to the proof of [29, Proposition 4.1], one knows that u belongs to Ns (O , D ). 
For v Ns (O , D ), put the projections pv = v v, qv = vv in D , and the
clopen subsets Xv = supp(pv ), Yv = supp(qv ) of X . Then Ad(v) : D pv D qv
is an isomorphism and induces a partial homeomorphism v : Xv Yv such that
Ad(v)(f ) = f v1

for f C(Xv ),

Ad(v )(g) = g v

for g C(Yv ).

We will state that v gives rise to an element of [ ]sc in Proposition 6.8. The proof
basically follows a line of the proof of [29, Proposition 4.7] as in the following way.
Fix v Ns (O , D ) for a while. For m Z and Bn (), n N, put


t (v)e 1mt dt,


v = E(S v),
v = E(vS ).
vm =
T

Then we have
Lemma 6.5.
(i) The family vm , m Z of operators are partial isometries
in O such
that
all but nitely many vm , m Z are zero, and

(a) v = mZ vm : nite sum.

(b) vm
vm , vm vm
are projections in D for m Z.

(c) vm D vm D and vm
D vm D for m Z.


(d) vm vm = vm vm = 0 for m = m .
(e) v0 F .
(ii) For a xed n N, the operators v , v for Bn () are partial isometries in Fsatisfying the following conditions:

(a) vn = Bn () S v and vn = Bn () v S .

v and v v
are projections in D such that
(b) v v , v v , v


vn vn =
v v ,
vn vn =
S v v S ,

vn
vn =

Bn ()

Bn ()

S v
v S ,

vn vn
=

Bn ()

(c)

v D v , v D v , v D v

v v
.

Bn ()

and

v
D v

are contained in D .

ORBIT EQUIVALENCE IN C -ALGEBRAS

135
15

Lemma 6.6. For u Ns (O , D ), let u : Xu Yu be the induced partial


homeomorphism on X . Assume u F . Then there exists k N such that
u (x)m = xm for all x = (xn )nN Xu and m > k, where u (x) = (u (x)n )nN .
Proof. As u F , one may take u Fk0 for some k0 such that u u < 12 .
Suppose that for any k N there exist x Xu and N > k such that u (x)N = xN .
For the above k0 , take x Xu and N0 > k0 such as u (x)N0 = xN0 . Put y = u (x).
Let x = (xn )nN = (n , un (x))nN and y = (yn )nN = (n , un (y))nN . Since u
belongs to FN0 1 , it is written as

c, S a, S
u =
,BN0 1 ()

where c, C, a, A for , BN0 1 (). Put


b = S1 N0 1 S1 N0 1 a1 N0 1 ,1 N0 1 S 1 N0 1 S1 N0 1 A.
We then have
S1 N0 1 u S1 N0 1 = c1 N0 1 ,1 N0 1 b.
Since xN0 = yN0 , we have either N0 = N0 or uN0 (x) = uN0 (y).
Case 1: N0 = N0 .
We have S1 N u S1 N0 = c1 N0 1 ,1 N0 1 SN bSN0 = 0. Hence
0

1
.
2
= 0. On the

S1 N uS1 N0 = S1 N uS1 N0 S1 N u S1 N0 <


0
0
0
As S1 N uS1 N0 is a partial isometry, one has S1 N uS1 N0
0
0
other hand, S1 N0 e N0 (y) = y and

Ad(u)(S1 N0 S 1 N )ey = U1 N (u1 (y))ey = U1 N (x)ey = ey


0
0

where U1 N is the characteristic function on X for the cylinder set U1 N0 ,


0
so that
(S1 N uS1 N0 S 1 N u S1 N0 e N0 (y) | e N0 (y) )
0

= (Ad(u)(S1 N0 S 1 N0 )ey | ey ) = 1.
Hence S1 N uS1 N0 = 0, a contradiction.
0
Case 2: uN0 (x) = uN0 (y).
We may assume that N0 = N0 . One may take f, g C() such that 0
f, g 1, f g = 0 and
f (uN0 (x)) = 1,

f (uN0 (y)) = 0,

g(uN0 (x)) = 0,

g(uN0 (y)) = 1.

Put T = gS1 N uS1 N0 f . As N0 = N0 , it follows that


0

gS1 N0 u S1 N0 f

= gS N0 bSN0 f = S N0 bSN0 gf = 0.

Hence
T = gS1 N0 uS1 N0 f gS1 N0 u S1 N0 f u u

1
.
2

On the other hand, one notes


S1 N0 ge N0 (y) = S1 N0 g(uN0 (y))e N0 (y) = S1 N0 e N0 (y) = ey .

136
16

KENGO MATSUMOTO

As u ey = eu1 (y) = ex and f S 1 N ex = f (uN0 (x)) = ex , one has


0

(T T e N0 (y) | e N0 (y) )
= (uS1 N0 f 2 S 1 N u S1 N0 ge N0 (y) | S1 N0 ge N0 (y) )
0
= (Ad(u)(S1 N0 f 2 S 1 N0 )ey | ey )
= (f S 1 N0 ex | f S 1 N0 ex ) = 1

so that T e N0 (y) = 1, a contradiction.

Hence we have
Lemma 6.7. For a partial isometry u F satisfying
uD u D ,

u D u D ,

let u : supp(u u) supp(uu ) be the homeomorphism dened by Ad(u)(g) =


g u1 for g D u u. Then there exists ku N such that
ku (u (x)) = ku (x)

for x supp(u u).

By virtue of Lemma 6.5 and Lemma 6.7 we have


Proposition 6.8. For any v Ns (O , D ), the partial homeomorphism v
induced by Ad(v) on D gives rise to an element of the continuous full inverse
semigroup [ ]sc . If in particular v belongs to N (O , D ), then v belongs to [ ]c .
Proof. The proof is similar to that of [29, Proposition 4.7].

The unitaries U(D ) are naturally embedded into Ns (O , D ). We denote the


embedding by id. For v Ns (O , D ), the induced partial homeomorphism v on
X gives rise to an element of [ ]sc by the above proposition. Let PU(D ) be the
set of partial isometries in D . The following is easy.
Lemma 6.9. Let u, v Ns (O , D ) be partial isometries satisfying uf u = vf v
for all f D . Then we have.
(i) u f u = v f v for all f D ,
(ii) v u, u v, vu , uv PU(D ) and
(v u) (v u) = (u v) (u v) = u u = v v,
(vu ) (vu ) = (uv ) (uv ) = uu = vv
We then have
Theorem 6.10. The diagrams
id

id

1 U(D ) N (O , D ) [ ]c 1




1 PU(D ) Ns (O , D ) [ ]sc 1.
are all commutative, where three vertical arrows denoted by are inclusions. The
rst row sequence is exact and splits as group, and the second row sequence is exact
and splits as inverse semigroup.

ORBIT EQUIVALENCE IN C -ALGEBRAS

137
17

Proof. By Proposition 6.8, the map : v Ns (O , D ) v [ ]sc


denes a homomorphism as inverse semigroup such that (N (O , D )) = [ ]c .
It is surjective by Proposition 6.4. Suppose that v = id on X for some v
Ns (O , D ). This means that Ad(v) = id on D so that v commutes with all
of elements of D . By Proposition 5.4, v belongs to D . Hence the second row
sequence is exact. Similarly, the rst row sequence is exact. As in Proposition 6.4,
the partial isometries u for [ ]sc dened by u ex = e (x) , x X X give
rise to sections of the both exact sequences, which split. The commutativities of
the diagrams are all clear

7. Orbit equivalence
Let (A1 , 1 , 1 ) and (A2 , 2 , 2 ) be commutative C -symbolic dynamical systems such that A1 = C(1 ) and A2 = C(2 ) for some compact Hausdor spaces
1 and 2 respectively. If there exists a homeomorphism h : X1 X2 such that
h(orb1 (x)) = orb2 (h(x)) for x X1 , then (X1 , 1 ) and (X2 , 2 ) are said to
be topologically orbit equivalent. In this case, there exist maps k1 , l1 : X1 Z+
and k2 , l2 : X2 Z+ satisfying

k (x)
l (x)
for x X1 ,
21 (h(1 (x))) = 12 (h(x))
(7.1)
k2 (y) 1
l2 (y) 1
1 (h (2 (y))) = 1 (h (y))
for y X2 .
We say that (X1 , 1 ) and (X2 , 2 ) are continuously orbit equivalent if there exist
continuous maps k1 , l1 : X1 Z+ and k2 , l2 : X2 Z+ satisfying the equalities
(7.1). By putting
k1n (x) =
k2n (y) =

n1

i=0
n1


k1 (i 1 (x)),
k2 (i 2 (y)),

l1n (x) =
l2n (y) =

i=0

n1

i=0
n1


l1 (i 1 (x)),

x X1 , n N,

l2 (i 2 (y)),

y X2 , n N,

i=0

we have
kn (x)

21

k2n (y)
1

ln (x)

(h(n1 (x))) = 12

(h1 (n2 (y))) =

l2n (x)
2

(h(x)),
(h1 (y)),

x X 1 ,
y X 2 .

Hence one easily knows that the relations (7.1) imply


h(orb1 (x)) = orb2 (h(x))

for x X1

so that continuous orbit equivalence implies topological orbit equivalence (cf. [29,
Lemma 5.2]. Similarly to the proof of [29, Proposition 5.3, 5.4], one directly sees
the following proposition by using Lemma 6.3.
Proposition 7.1. (X1 , 1 ) and (X2 , 2 ) are continuously orbit equivalent
if and only if there exists a homeomorphism h : X1 X2 such that h [1 ]sc
h1 = [2 ]sc .
Proposition 7.2. If there exists an isomorphism : O1 O2 such that
(D1 ) = D2 , then there exists a homeomorphism h : X1 X2 such that
h [1 ]sc h1 = [2 ]sc .

138
18

KENGO MATSUMOTO

Proof. Suppose that there exists an isomorphism : O1 O2 such that


(D1 ) = D2 . By the split exact sequences
1 PU(Di ) Ns (Oi , Di ) [i ]sc 1,

i = 1, 2

 : [ ]sc [ ]sc of inverse


of inverse semigroups, one may nd an isomorphism
1
2
semigroup such that the following diagrams are commutative:
id

id

1 PU(D1 ) Ns (O1 , D1 ) [1 ]sc 1


 U (D1 )


1 PU(D2 ) Ns (O2 , D2 ) [2 ]sc 1.
Let h : X1 X2 be the homeomorphism satisfying (f ) = f h1 for f
C(X1 ). For v Ns (O1 , D1 ), take the partial homeomorphism v : Xv Yv
satisfying Ad(v)(f ) = f v1 for f C(Xv ). For g C(h(Xv )), we have
Ad(v) 1 (g) = g h v1 h1 ,

and

1
Ad((v))(g) = g (v)
.

By the identity Ad(v) 1 = Ad((v)), one has


1
g h v1 h1 = g (v)

for g C(h(Xv )).

Hence h v h1 = (v) . As [i ]sc = {v | v Ns (Oi , Di )}, i = 1, 2, one sees



that h [1 ]sc h1 = [2 ]sc .
Proposition 7.3. If (X1 , 1 ) and (X2 , 2 ) are continuously orbit equivalent, then there exists an isomorphism : O1 O2 such that (D1 ) = D2 .
Proof. The proof is essentially the same as the proof of Proposition 6.4 and
[29, Proposition 5.5]. We omit its proof.

Therefore we reach the main result of this paper.
Theorem 7.4. Let (A1 , 1 , 1 ) and (A2 , 2 , 2 ) be C -symbolic dynamical systems satisfying condition (I). Suppose that both the algebras A1 and A2 are commutative. Then the following are equivalent:
(i) There exists an isomorphism : O1 O2 such that (D1 ) = D2 .
(ii) (X1 , 1 ) and (X2 , 2 ) are continuously orbit equivalent.
(iii) There exists a homeomorphism h : X1 X2 such that h[1 ]sc h1 =
[2 ]sc .
8. Special cases
We have the following two special cases of Theorem 7.4.
Let A = [A(i, j)]N
i,j=1 be an N N matrix with entries in {0, 1}. Let us denote
by (XA , A ) the right one-sided topological Markov shift dened by A. The CuntzKrieger algebra OA for A is generated by partial isometries S1 , . . . , SN satisfying
N
N

j=1 Sj Sj = 1, Si Si =
j=1 A(i, j)Sj Sj , i = 1, . . . , N. The C -subalgebra gener

ated by projections Sn S1 S1 Sn , 1 , . . . , n {1, . . . , N } denoted by DA


is canonically isomorphic to the commutative C -algebra C(XA ). It is the diagonal
algebra of the canonical AF algebra inside OA .

ORBIT EQUIVALENCE IN C -ALGEBRAS

139
19

Corollary 8.1 ([29]). Let A and B be irreducible square matrices with entries
in {0, 1} satisfying condition (I) in [7]. Then their respect right one-sided topological
Markov shifts (XA , A ) and (XB , B ) are continuously orbit equivalent if and only
if there exists an isomorphism : OA OB between their respect Cuntz-Krieger
algebras such that (DA ) = DB .
Let be a homeomorphism on a compact Hausdor space X. Denote by
the associated automorphism on the commutative C -algebra C(X). Then condition (I) for the C -symbolic dynamical system (C(X), , { }) is nothing but the
topological freeness of the dynamical system (X, ) as in Proposition 4.9. Hence
Theorem 7.4 implies the following result proved by Tomiyama ([36], cf. [2],[35],
[37]).
Corollary 8.2 ([36] (Tomiyama)). Let (X, ) and (Y, ) be topological free
homeomorphisms on compact Hausdor spaces X and Y respectively. Then the
dynamical systems (X, ) and (Y, ) are continuously orbit equivalent if and only
if there exists an isomorphism : C(X)  Z C(Y )  Z between their respect
C -crossed products such that (C(X)) = C(Y ).
Acknowledgment. The author is grateful to Takeshi Katsura for his useful
discussions. The author also would like to deeply thank the organizers Marcel de J
eu, Sergei Silvestrov, Christian F. Skau and Jun Tomiyama.
References
1. T. Bates and D. Pask, C -algebras of labelled graphs, J. Operator Theory 57(2007), pp.
207226.
2. M. Boyle and J. Tomiyama, Bounded continuous orbit equivalence and C -algebras, J. Math.
Soc. Japan 50(1998), pp. 317329.
3. O. Bratteli, Inductive limits of nite-dimensional C -algebras, Trans. Amer. Math. Soc.
171 (1972), pp. 195234.
4. T. M. Carlsen and Silvestrov, C -crossed products and shift spaces, Expo. Math. 25(2007),
no. 4, 275307.
5. T. M. Carlsen and S. Silvestrov, On the Exel crossed product of topological covering maps,
arxiv:[math.OA]/0811.0056, to appear in Acta Appl. Math.
6. J. Cuntz, Simple C -algebras generated by isometries, Comm. Math. Phys. 57(1977), pp.
173185.
7. J. Cuntz and W. Krieger, A class of C -algebras and topological Markov chains, Invent.
Math. 56(1980), pp. 251268.
8. M. Denker, C. Grillenberger and K. Sigmund, Ergodic theory on compact spaces, Lecture
Notes in Math. 527, Springer-Verlag, Berlin, Heidelberg and New York (1976).
9. H. Dye, On groups of measure preserving transformations, American J. Math. 81(1959), pp.
119159.
10. H. Dye, On groups of measure preserving transformations II, American J. Math. 85(1963),
pp. 551576.
11. R. Excel and M. Laca, Cuntz-Krieger algebras for innite matrices, J. Reine Angew. Math.
512(1999), pp. 119172.
12. R. Fischer, Soc systems and graphs, Monats. f
ur Math. 80 (1975), pp. 179186.
13. T. Giordano, I. F. Putnam and C. F. Skau, Topological orbit equivalence and C -crossed
products, J. Reine Angew. Math. 469(1995), pp. 51111.
14. T. Giordano, I. F. Putnam and C. F. Skau, Full groups of Cantor minimal systems, Israel
J. Math. 111(1999), pp. 285320.
15. R. H. Herman, I. F. Putnam and C. F. Skau, Ordered Bratteli diagrams, dimension groups
and topological dynamics, Internat. J. Math. 3(1992), pp. 827864.

140
20

KENGO MATSUMOTO

16. T. Kajiwara, C. Pinzari and Y. Watatani, Ideal structure and simplicity of the C -algebras
generated by Hilbert modules, J. Funct. Anal. 159(1998), pp. 295322.
17. T. Katsura, A class of C -algebras generalizing both graph algebras and homeomorphism
C -algebras I, fundamental results, Trans. Amer. Math. Soc. 356(2004), pp. 42874322.
18. W. Krieger, On ergodic ows and isomorphisms of factors, Math. Ann 223(1976), pp. 1970.
19. W. Krieger, On soc systems I, Israel J. Math. 48(1984), pp. 305330.
20. W. Krieger, On soc systems II, Israel J. Math. 60(1987), pp. 167176.
21. A. Kumjian, D. Pask, I. Raeburn and J. Renault, Graphs, groupoids and Cuntz-Krieger
algebras, J. Funct. Anal. 144(1997), pp. 505541.
22. J. Kwapisz, Cocycle subshifts, Math. Z. 234(2000), pp. 255290.
23. D. Lind and B. Marcus, An introduction to symbolic dynamics and coding, Cambridge
University Press, Cambridge (1995).
24. K. Matsumoto, Presentations of subshifts and their topological conjugacy invariants, Documenta Math. 4 (1999), pp. 285340.
25. K. Matsumoto, C -algebras associated with presentations of subshifts, Documenta Math.
7(2002), pp. 130.
26. K. Matsumoto, Construction and pure infniteness of C -algebras associated with lambdagraph systems, Math. Scand. 97(2005), pp. 7389.
27. K. Matsumoto, Actions of symbolic dynamical systems on C -algebras, J. Reine Angew.
Math. 605(2007), pp. 23-49.
28. K. Matsumoto, Actions of symbolic dynamical systems on C -algebras II. Simplicity of C symbolic crossed products and some examples, preprint, math arXiv:0705.3283, to appear in
Math. Z..
29. K. Matsumoto, Orbit equivalence of topological Markov shifts and Cuntz-Krieger algebras,
preprint, math arXiv:0707.2114.
30. K. Matsumoto, Orbit equivalence of one-sided subshifts and the associated C -algebras,
preprint, math arXiv:0709.1185.
31. A. L. T. Paterson, Groupoids, inverse semigroups, and their operator algebras, Progress in
Mathematics 170, Birkh
auser, Boston, Basel and Berlin, 1998.
32. M. V. Pimsner, A class of C -algebras generalizing both Cuntz-Krieger algebras and crossed
products by Z, in Free Probability Theory, Fields Institute Communications 12(1996), pp.
189212.
33. J. Renault, A groupoid approach to C -algebras, Lecture Notes in Math. 793, SpringerVerlag, Berlin, Heidelberg and New York (1980).
34. M. Rdom, Classication of purely innite simple C -algebras I, J. Func. Anal. 131(1995),
pp. 415458.
35. J. Tomiyama, The interplay between topological dynamics and theory of C -algebras, Lecture
Note No 2, Res. Inst. Math. Seoul 1992.
36. J. Tomiyama, Topological full groups and structure of normalizers in transformation group
C -algebras, Pacic. J. Math. 173(1996), pp. 571583.
37. J. Tomiyama, Representation of topological dynamical systems and C -algebras, Contemporary Math. 228(1998), pp. 351364.
38. B. Weiss, Subshifts of nite type and soc systems, Monats. Math. 77(1973), pp. 462474.
Department of Mathematical Sciences, Yokohama City University, 22-2 Seto, Kanazawaku, Yokohama, 236-0027 Japan
E-mail address: kengo@yokohama-cu.ac.jp

Contemporary Mathematics
Volume 503, 2009

Normalisers, nest algebras and tensor products


M. McGarvey and I. G. Todorov
Abstract. We show that if A is the tensor product of nitely many continuous nest algebras, B is a CDCSL algebra and A and B have the same normaliser
semi-group then either A = B or A = B.

1. Introduction
Normalisers of selfadjoint operator algebras were introduced by Murray and
von Neumann in the 1930s and have played an important role in Operator Algebra
Theory thereafter. They are used in a fundamental way in the theory of crossed
products, a notion which provides a setting for Non-commutative Dynamics (see
[13] and [20]). Normalisers constitute a basic object in the theory of limit algebras
as well [16]. Normalisers of tensor products of von Neumann algebras were recently
considered in [4], [11], [17] and [18]. The study of the normalisers of non-selfadjoint
operator algebras, namely of nest algebras, was initiated in the 1990s [1], [7], [5].
In [14] the notion of a normaliser was generalised and studied in the context of
reexive algebras, a non-selfadjoint generalisation of von Neumann algebras. It
was shown that normalisers are closely related to ternary rings of operators, a
class of spaces studied independently in Operator Space Theory (see [3]). This
connection provided the base in [8] for the introduction of an equivalence relation
for non-selfadjoint operator algebras which later lead to the study of a more general
equivalence relation for abstract dual operator algebras linked to Morita equivalence
[9], [10].
If A is an operator algebra acting on a Hilbert space H, a normaliser of A is
a bounded linear operator on H such that
T AT A and

T AT A.

Let N (A) be the set of all normalisers of an operator algebra A. It is obvious that
N (A) is a selfadjoint semi-group of operators containing the diagonal A A of A
(here and in the sequel we let A = {A : A A}). The question to what extent
N (A) determines A was considered in [19], where A was taken from the class of
CSL algebras introduced by Arveson in his seminal work [2]. It is obvious that for
any operator algebra A we have that N (A) = N (A ). It is thus natural to ask to
1991 Mathematics Subject Classication. Primary 47L35; Secondary 47D03.
Key words and phrases. Nest algebras, tensor products, normalisers.
The rst author was supported by a grant from the Department of Education and Learning
for Northern Ireland.
1
141

142
2

M. MCGARVEY AND I. G. TODOROV

what extent the converse is true; in particular, whether for two CSL algebras A
and B the equality N (A) = N (B) implies that either A = B or A = B. Within the
classes where this holds one is able to determine (up to adjoint) the non-selfadjoint
algebras belonging to the class by using selfadjoint objects, namely their normaliser
semi-groups.
Easy examples, in which the atomic and the continuous parts of the invariant subspace lattices of the algebras are both non-trivial, show that this converse
statement fails. It is however true if A and B are totally atomic CSL algebras, as
well as when they are continuous nest algebras. Namely, the following result was
established in [19]:
Theorem 1.1. Let H be a separable Hilbert space and A and B be continuous
nest algebras acting on H. Suppose that N (A) = N (B). Then either A = B or
A = B.
A class of operator algebras larger than the class of nest algebras is that of
CDCSL algebras. It has played an important role in non-selfadjoint operator algebra theory (see [6, Chapter 23]). CDCSL algebras are characterised among CSL
algebras by the fact that the Hilbert-Schmidt operators contained in the algebra
are weakly dense in it. We note that nite tensor products of nest algebras possess
this property.
In this note we prove the following generalisation of Theorem 1.1:
Theorem 1.2. Let Hi be a separable Hilbert space, Ai be a continuous nest
algebra acting on Hi , i = 1, . . . , n, H = H1 Hn and A = A1 An B(H).
Suppose that B B(H) is a CDCSL algebra. If N (A) = N (B) then either A = B
or A = B.
The proof of Theorem 1.2 and some of its corollaries are given in Section 3. In
the next section we collect preliminary notions and results.
2. Preliminaries
All Hilbert spaces in this note will be assumed to be separable. Let H be a
Hilbert space and B(H) be the space of all bounded linear operators on H. The set
S(H) of all closed subspaces of H is a complete lattice with respect to intersection
and closed linear span. Using the bijective correspondence between S(H) and the
set P(H) of all orthogonal projections on H, we can equip P(H) with a natural
lattice structure. A subspace lattice on H is a sublattice L P(H) closed in the
strong operator topology. Given a subspace lattice L, we let
Alg L = {A B(H) : (I L)AL = 0, for each L L}
be the algebra of all operators on H leaving every projection of L invariant. Obviously, Alg L contains the identity operator, and it is trivial to check that it is closed
in the weak operator topology. Conversely, given a weakly closed unital subalgebra
A B(H), we let
Lat A = {L P(H) : (I L)AL = 0, for each A A}
be the lattice of all projections on H invariant under every operator in A. The set L
is easily seen to be a subspace lattice. A weakly closed unital subalgebra A B(H)
is called reexive if A = Alg Lat A. By virtue of von Neumanns Bicommutant
Theorem, the class of reexive algebras contains all von Neumann algebras.

NORMALISERS, NEST ALGEBRAS AND TENSOR PRODUCTS

143
3

A commutative subspace lattice (CSL) on H is a subspace lattice L


P(H) with the property that P Q = QP whenever P, Q L. An atom of a CSL
L is a non-zero projection E on H such that for every L L, either E L or
EL = 0. A CSL is called continuous if it has no atoms. A CSL algebra is
a reexive algebra A of the form A = Alg L for some CSL L; equivalently, CSL
algebras are the reexive operator algebras containing a maximal abelian selfadjoint
algebra (masa). CSLs and CSL algebras were introduced and studied in depth by
Arveson in [2].
A CDCSL algebra is a CSL algebra A with the property that the HilbertSchmidt operators belonging to A are weakly dense in A. We note that usually the
denition of a CDCSL algebra is given in terms of a strong distributivity property
of its subspace lattice; however, the denition given above is equivalent to it [6,
Theorem 23.7].
A nest is a totally ordered CSL, and a nest algebra is an operator algebra
A B(H) of the form A = Alg L for some nest L P(H). By the Erdos Density
Theorem [6, Theorem 3.11], the class of CDCSL algebras contains all nest algebras.
If L is a nest, let L = {L : L L}, where for a projection P we have set
P = I P . Then Alg L = (Alg L) . Throughout the paper, I will denote the
interval [0, 1] equipped with the Lebesgue measure, H will denote the Hilbert space
L2 (I) and D L (0, 1) will denote the corresponding multiplication masa. For
each t [0, 1], let Nt be the projection onto the subspace {f L2 (0, 1) : f (s) =
0, a.e. s < t}. The nest N = {Nt : 0 t 1} is known as the Volterra nest, and
the corresponding reexive algebra Av = Alg N is called the Volterra nest algebra.
The von Neumann algebra N  generated by N is equal to D and is in particular
a masa; nests with this property are called multiplicity free. (Here, and in the
sequel, for a subset S B(H) we let S  = {T B(H) : T S = ST, S S} be the
commutant of S.) It is well-known that every continuous multiplicity free nest is
unitarily equivalent to the Volterra nest. We refer the reader to [6] for the theory
of nest algebras.
We denote by [T ] the linear span of a subset T V of a linear space V, and
by V  W the algebraic tensor product of the linear spaces V, W. If H1 and
H2 are Hilbert spaces, we denote by H1 H2 their Hilbertian tensor product. If
A B(H1 ) and B B(H2 ) we let A B denote the weakly closed subalgebra of
B(H1 H2 ) generated by the elementary tensors A B, where A A and B B.
Tensor products of nest algebras were studied in detail in [12] where it was shown
def

that if Ni is a nest and Ai = Alg Ni , i = 1, . . . , n, then A = A1 An is a


CSL algebra. In fact, A = Alg(N1 Nn ) where, if L1 and L2 are subspace
lattices, L1 L2 is the smallest subspace lattice containing the projections of the
form P1 P2 , where P1 L1 and P2 L2 .
We denote by C2 (H) the ideal of all Hilbert-Schmidt operators on a Hilbert
space H, and by 2 the Hilbert-Schmidt norm. Let (X, ) be a standard measure
space and H = L2 (X, ). The Hilbert-Schmidt operators on H are precisely the
integral operators Th where, for a function h L2 (X X, ) we let


(Th f )(y) =

h(x, y)f (x)d(x),


X

f H, y X.

144
4

M. MCGARVEY AND I. G. TODOROV

The function h is called the integral kernel of Th . Moreover, if H = H = L2 (I)


then Th Av if and only if, up to a null set,
def

supp h v = {(x, y) I I : x y},


where supp h = {(x, y) I I : h(x, y)
= 0} is the support of the function h
(dened up to a null set).
The Hilbert space H H can be naturally identied with L2 (I n ), where



n

I n = [0, 1] [0, 1] is equipped with the (n-dimensional) Lebesgue measure .





n

For x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) I n we write x y if xi yi for each


i = 1, . . . , n. It follows from [12, Proposition 2.1] that if h L2 (I n I n ) then the
Hilbert-Schmidt operator Th belongs to the algebra Av Av if and only if,



n

up to a null set,
def

supp h = {(x, y) I n I n : x y}.


If Ai , i = 1, . . . , n, are operator algebras and A = A1 An we let
Ne (A) = {T1 Tn : Ti N (Ai ), i = 1, . . . , n}.
It is obvious that Ne (A) N (A).
3. Proof of the result
In this section we give a proof of our main result, Theorem 1.2. We will need
several auxiliary facts.
Lemma 3.1. (i) Let H = H H and A B(H) be a Hilbert-Schmidt



n

operator with integral kernel h L2 (I n I n ). Assume that the set supp h has
positive measure. Then there exist Pi N , i = 1, . . . , n, such that (P1
Pn )A(P1 Pn )
= 0.
(ii) Let Hi be a Hilbert space, Ni be a continuous multiplicity free nest on Hi ,
Ai = Alg Ni , i = 1, . . . , n, H = H1 Hn and A = A1 An . Then the
linear span of

{(P1 Pn )C2 (H)(P1 Pn ) : Pi Ni , i = 1, . . . , n}
is dense in C2 (H) A in the Hilbert-Schmidt norm.
Proof. (i) For an element t = (t1 , . . . , tn ) I n write
[t, 1] = {(s1 , . . . , sn ) I n : ti si 1, i = 1, . . . , n}
and
[0, t) = {(s1 , . . . , sn ) I n : 0 si < ti , i = 1, . . . , n}.
Let {tk }kN be a dense subset of I n , Lk = P ([tk , 1]) and Mk = P ([0, tk )), k N.
Suppose that Lk AMk = 0 for each k N. This implies that h[0,tk )[tk ,1] = 0
for each k N. Since kN [0, tk ) [tk , 1] = o (where o is the interior of )
and ( )( \ o ) = 0, it follows that the set supp h has measure zero, a
contradiction.
(ii) Since each continuous multiplicity free nest is unitarily equivalent to the
Volterra nest, we may assume that Ni = N , i = 1, . . . , n. Suppose that T

NORMALISERS, NEST ALGEBRAS AND TENSOR PRODUCTS

145
5

C2 (H)A is orthogonal to the spaces of the form P C2 (H)Q where P = P1 Pn


and Q = P1 Pn for some Pi N , i = 1, . . . , n. Thus, tr(P SQT ) = 0
for all S C2 (H) and P and Q of the above form. It follows that tr(SQT P ) = 0
for all such S, P and Q and so P T Q = 0 for all such P and Q. Suppose that
h L2 (I n I n ) is the integral kernel of T . By (i), supp h is a null subset of
I n I n . However, since T A, we have that supp h up to a null set. It follows
that supp h is a null set, and hence T = 0.

The following lemma is easily inferred from the fact that if K is a separable
Hilbert space and C B(K) is a continuous nest algebra then [N (C)] is weakly
dense in B(K) [19, Corollary 3.5 (i)]; we omit its proof.
Lemma 3.2. Let Hi be a Hilbert space, Ni be a continuous nest on Hi , Ai =
Alg Ni , i = 1, . . . , n, H = H1 Hn and A = A1 An . Then [Ne (A)] is
weakly dense in B(H).
Lemma 3.3. Let Hi be a Hilbert space, Ni be a continuous multiplicity free nest
on Hi , Ai = Alg Ni , i = 1, . . . , n, H = H1 Hn and A = A1 An . Let
A C2 (H) be a non-zero Hilbert-Schmidt operator. Then
(3.1)

2

[V AW : V, W Ne (A)]

= C2 (H).

Proof. Without loss of generality, we may assume that Hi = L2 (I) and that
Ni = N is the Volterra nest for each i = 1, . . . , n. Thus, up to unitary equivalence,
H = L2 (I n ). If H let H be the function given by (t) = (t), t I n .
Let : H H C2 (H) be the unitary operator given on elementary tensors by
( )(0 ) = (0 , ). Given T B(H) we let T B(L2 (I n )) be the operator
 = TS for all
dened by T() = T (), H. It is easy to verify that ST

S, T B(H) and that the mapping T T takes Ne (A) onto itself.
A straightforward calculation shows that
(3.2)

S()T = ((T S)), S, T B(H), H H.

Let 0 = 1 (A). By (3.2),



2
 V )(0 ) : V, W Ne (A)] .
(3.3)
1 [V AW : V, W Ne (A)]
= [(W
Denote by E the right hand side of (3.3). It follows from the previous paragraph
that E is invariant under each operator of the form T S where S, T Ne (A).
Hence, E is invariant under [Ne (A)]  [Ne (A)]. Lemma 3.2 now implies that E is
invariant under B(H H) and since E
= {0} we have that E = H H.

Let Ai be a nest algebra acting on a Hilbert space Hi , Ni = Lat Ai and Pi Ni
where i = 1, . . . , n. Set A = A1 An , P = P1 Pn and
NP (A) = {(V1 Vn ) + (W1 Wn ) :
Vi , Wi N (Ai ), Vi = Pi Vi Pi , Wi = Pi Wi Pi , i = 1, . . . , n}.
Lemma 3.4. Let Hi be a Hilbert space, Ai be a nest algebra acting on Hi ,
i = 1, . . . , n, and A = A1 An . For each i = 1, . . . , n, let Ci be either equal
to Ai or to Ai , and set C = C1 Cn . Let Pi Lat Ci , i = 1, . . . , n and
P = P1 Pn . Then NP (C) N (A).

146
6

M. MCGARVEY AND I. G. TODOROV

Proof. Let A = A1 An where Ai Ai , i = 1, . . . , n, and T NP (C).


Thus, T = V + W where V = V1 Vn , W = W1 Wn for some Vi , Wi
N (Ci ) with Vi = Pi Vi Pi and Wi = Pi Wi Pi , i = 1, . . . , n. Then T AT = V AV +
W AV + V AW + W AW . Since V, W N (A), we have that V AV , W AW A.
We will show that W AV , V AW A.
Note that
W AV = (P1 W1 P1 A1 P1 V1 P1 ) (Pn Wn Pn An Pn Vn Pn )
and, for each i, either Pi Ni or Pi Ni . If Pi Ni , then Pi Ai Pi = 0 and so
Pi Wi Pi Ai Pi Vi Pi = 0. If Pi Ni then
Pi Wi Pi Ai Pi Vi Pi Pi B(Hi )Pi Ai .
It follows that W AV A. Similar arguments show that V AW A. It follows
that T AT A. Similarly, T AT A and hence T N (A).

The next proposition shows that the normalisers of the tensor product of nitely
many copies of the Volterra nest algebra act transitively on certain Hilbert-Schmidt
operators.
Proposition 3.5. Let Hi be a Hilbert space, Ni be a continuous multiplicity
free nest acting on Hi , Ai = Alg Ni , i = 1, . . . , n, H = H1 Hn and A =
A1 An . Suppose that A A is a non-zero Hilbert-Schmidt operator and
Pi Ni , i = 1, . . . , n, are such that A = (P1 Pn )A(P1 Pn ). Then
(i) [(ST )A(ST ) : S Ne (A), T NP (A)]2 = C2 (H) A, and

(ii) [(ST )A(ST ) : S Ne (A), T NP (A)]w = A.


Proof. (i) Let P = P1 Pn and Q = P1 Pn . Fix Vi , Wi N (Ai )
with Vi = Pi Vi Pi and Wi = Pi Wi Pi , i = 1, . . . , n, and set V = V1 Vn ,
W = W1 Wn and T = V + W . Since A = P AQ, we have that
(3.4)

T AT = (T P )A(QT ) = V AW .

By Lemma 1.1 (ii) of [19], for any projection E Ni we have that N (EA|EHi ) =
EN (A)|EHi . It now follows from (3.4) that if Ci = Pi Ai |Pi Hi and Bi = Pi Ai |Pi Hi ,
i = 1, . . . , n, then
{T AT : T NP (A)} = {V AW : V = V1 Vn , W = W1 Wn ,
Vi N (Ci ), Wi N (Bi ), i = 1, . . . , n}.
Here we have identied an operator B acting on a subspace H1 of a Hilbert space
H2 with the operator B 0 acting on H2 .
The algebras Ci , Bi are continuous multiplicity free nest algebras, and hence
are all unitarily equivalent to Av . Lemma 3.3 now implies that
2

P C2 (H)Q [T AT : T NP (A)]

Let Ei be a non-zero projection in Ni , i = 1, . . . , n. Then there exists a unitary


operator Ui N (Ai ) such that Ui Pi Ui = Ei , i = 1, . . . , n. It follows that, if
E = E1 En and F = E1 En , then
EC2 (H)F [(ST )A(ST ) : S Ne (A), T NP (A)]2 .
Lemma 3.1 (ii) implies that
C2 (H) A [(ST )A(ST ) : S Ne (A), T NP (A)]2 .

NORMALISERS, NEST ALGEBRAS AND TENSOR PRODUCTS

147
7

The converse inclusion follows from Lemma 3.4 and hence (i) is established.
w

(ii) follows from (i) and the fact that C2 (H) A

= A (see [6]).

We note the following corollary of Proposition 3.5.


Corollary 3.6. Let A be a non-zero Hilbert-Schmidt operator in Av and
P N be a projection such that A = P AP . Then
w

[T AT : T N (Av )]

= Av .

Lemma 3.7. Let Bi B(H) be either equal to Av or to Av , i = 1, . . . , n. Set


A = Av Av and B = B1 Bn . Then N (A) = N (B) if and only if



n

A = B or A = B.
Proof. Let rst n = 2 and B = Av Av . We show that N (A)
= N (B). Let
W B(H H) be the unitary operator given by W ( ) = , , H.
For P1 , P2 P(H) we have that W (P1 H P2 H) = P2 H P1 H. If A A and
P1 , P2 N then
W AW (P1 H P2 H) = W A(P2 H P1 H) W (P2 H P1 H) = P1 H P2 H.
Thus, W AW A. Since W = W , we have that W N (A).
On the other hand, W
N (B). Indeed, let P1 , P2 be any non-trivial projections
in N and V be a partial isometry with initial space P1 H and nal space P1 H. Then
V Av . However,
W (I V )W (P1 H P2 H)

= W (I V )(P2 H P1 H)
=

W (P2 H P1 H) = P1 H P2 H,

which shows that W (I V )W


B.
Now suppose that n is arbitrary. If A
= B and A
= B then there exist nite
(possibly empty) tensor products A1 and A2 of Volterra nest algebras such that,
up to a permutation of the factors in the tensor products,
A = (Av Av ) A1 A2 and B = (Av Av ) A1 A2 .
Let T = W I I; the rst paragraph of the proof shows that T N (A).
Assume that T N (B). It is easy to see that this implies W N (Av Av ), a
contradiction.

We are now ready to prove our main result.
Proof of Theorem 1.2. As in the proof of Proposition 4.4 of [19], we may assume
that A A is a masa, that is, that A = Alg(N1 Nn ) for some continuous
multiplicity free nests N1 , . . . , Nn acting on H1 , . . . , Hn , respectively. Suppose that
the statement holds if Ai = Av , for each i = 1, . . . , n. Recall that H = L2 (I).
There exists a unitary operator Ui : H Hi such that Ui Av Ui = Ai , i = 1, . . . , n.
def

Let U = U1 Un . Then U AU = A0 = Av Av . Let B0 = U BU .





n

It is easy to verify that for any operator algebra C and a unitary operator V
we have N (V CV ) = V N (C)V . The condition N (A) = N (B) now implies that
N (A0 ) = N (B0 ). By the assumption, A0 = B0 or A0 = B0 . It follows that A = B
or A = B . Hence, it suces to establish the statement in the case Ai = Av for
each i = 1, . . . , n.

148
8

M. MCGARVEY AND I. G. TODOROV

We identify H with L2 (I n ). The condition N (A) = N (B) easily implies that


A A = B B . For each subset G {1, 2, . . . , n}, let
G = {(x1 , . . . , xn , y1 , . . . , yn ) I n I n : xi < yi if i G, xi > yi if i
G}
and CG = C1 Cn where Ci = Av if i G and Ci = Av if i
G. We note that
the union G{1,...,n} G is a subset of I n I n of full measure.
Let B B be a Hilbert-Schmidt operator and I n I n be the support of
its integral kernel. Since ( )() > 0 there exists a subset G {1, . . . , n} such
that ( )(G ) > 0. It follows from Lemma 3.1 (i) and the fact that every
continuous multiplicity free nest is unitarily equivalent to the Volterra nest that
there exist Pi Lat Ci , i = 1, . . . , n, such that the Hilbert-Schmidt operator
A = (P1 Pn )B(P1 Pn )
is non-zero. Since the diagonals of A and B coincide, it follows that A B. We
also have that A CG . Indeed, let Li Lat Ci , i = 1, . . . , n. If Li Pi for some i
then
(L1 Ln )(P1 Pn ) = 0.
If Pi Li for each i then P1 Pn L1 Ln and hence
(P1 Pn )(L1 Ln ) = 0.
Thus, in both cases, (L1 Ln ) A(L1 Ln ) = 0 and hence A CG .
By Proposition 3.5 (ii),

CG = [(ST )A(ST ) : S Ne (CG ), T NP (CG )]w .


Using the fact that Ne (A) = Ne (CG ) and Lemma 3.4 we conclude that
w

CG [V AV : V N (A)]

= [V AV : V N (B)]

B.

Now suppose that there exists a Hilbert-Schmidt operator in B, say T , and a


subset F {1, 2, . . . , n} with F
= G such that, if is the support of the integral
kernel of T , then ()(F ) > 0. The previous paragraph implies that CF B.
Suppose that CF = E1 En where Ei = Av if i F and Ei = Av if i
F .
Since F
= G, there exists an i such that Ci = Ei . Let C Ci be a non-zero
Hilbert-Schmidt operator, C = C1 Cn where Cj = I if j
= i and Ci = C,
and S = C + C . Then
S = S CG + CF B.
Thus, S A A which implies that C + C is a Hilbert-Schmidt operator in
Av Av . If C + C = 0 then C = C Av , and hence C is a Hilbert-Schmidt
operator such that C, C Av . If h is the integral kernel of C then the integral
kernel k of C is given by k(x, y) = h(y, x), x, y I. It follows that h is supported
on {(x, x) : x I}, a set of measure zero, and hence C = 0, a contradiction. Thus,
C + C is a non-zero Hilbert-Schmidt operator in Av Av , a contradiction.
We have hence shown that every Hilbert-Schmidt operator in B has integral
kernel supported on G , and hence belongs to CG . Since C2 (H) B is weakly dense
in B, we conclude that B CG and so B = CG . It follows that N (CG ) = N (A).
Lemma 3.7 now implies that either A = B or A = B. The proof of Theorem 1.2 is
complete.
We nish this note by listing two immediate corollaries of Theorem 1.2. The
second of them includes as a special case Theorem 1.1.

NORMALISERS, NEST ALGEBRAS AND TENSOR PRODUCTS

149
9

Corollary 3.8. Let H be a Hilbert space, Ai , Bj , i = 1, . . . , n, j = 1, . . . , m,


be continuous nest algebras such that A = A1 An and B = B1 Bm
are unital subalgebras of B(H). Suppose that N (A) = N (B). Then n = m and,
moreover, either A = B or A = B.
Corollary 3.9. Let H be a Hilbert space, A B(H) be a continuous nest
algebra and B B(H) be a CDCSL algebra. Suppose that N (A) = N (B). Then
either A = B or A = B.
Acknowledgement The authors would like to thank A. Katavolos for his useful
remarks on the content of this paper.
References
[1] M. Anoussis and A. Katavolos, Unitary actions on nests and the Weyl relations, Bull.
London Math. Soc. 27 (1995) 265-272
[2] W.B. Arveson, Operator algebras and invariant subspaces, Ann. Math. (2) 100 (1974), 433532
[3] D. P. Blecher and C. Le Merdy, Operator algebras and their modulesan operator space
approach, Oxford University Press, 2004
[4] I. Chifan, On the normalizing algebra of a masa in a II1 factor, preprint
[5] K. J. Coates Normalizers of nest algebras, Proc. Amer. Math. Soc. 126 (1998), 159-165;
Errata, ibid, 2511-2512
[6] K. R. Davidson, Nest Algebras, Pitman Research Notes in Mathematics 191, Longman (1988)
[7] K. R. Davidson, Normalizers of nite multipliciy nests, Proc. Edinburgh Math. Soc. 39
(1996), 337-344
[8] G. K. Eleftherakis, TRO equivalent algebras, preprint, ArXiv: math. OA/0607488
[9] G. K. Eleftherakis, A Morita type equivalence for dual operator algebras, J. Pure Appl.
Algebra 212 (2008) no 5, 1060-1071
[10] G. K. Eleftherakis and V. I. Paulsen, Stably isomorphic dual operator algebras, Math.
Ann. 341 (2008), no 1, 99-112
[11] J. Fang, R. R. Smith, S. A. White and A. D. Wiggins, Groupoid normalisers of tensor
products, preprint
[12] F. Gilfeather, A. Hopenwasser and D. R. Larson, Reexive algebras with nite width
lattices: tensor products, cohomology, compact perturbations, J. Funct. Anal. 55 (1984), 176
- 199
[13] R.V. Kadison and J.R. Ringrose, Fundamentals of the theory of operator algebras, vol. 2,
Academic Press, 1986
[14] A. Katavolos and I.G. Todorov, Normalizers of operator algebras and reexivity, Proc.
London Math. Soc. (3) 86 (2003), 463-484
[15] A.I. Loginov and V.S. Shulman, Hereditary and intermediate reexivity of W*-algebras,
Izv. Akad. Nauk. SSSR, 39 (1975), 1260-1273; Math. USSR-Izv. 9 (1975), 1189-1201
[16] S.C. Power, Limit algebras: an introduction to subalgebras of C*-algebras, Longman, 1992
[17] A. M. Sinclair, R. R. Smith, S. A. White and A. D. Wiggins, Strong singularity of
singular masas in II1 factors, Illinois J. Math. 51 (2007), no. 4, 1077-1084
[18] R. R. Smith, S. A. White and A. D. Wiggins, Normalisers of irreducible subfactors,
preprint
[19] I. G. Todorov, Normalisers of CSL algebras, Indiana Univ. Math. J. 53 (2004), no. 4, 1013
- 1036
[20] D. P. Williams, Crossed products of C -algebras, American Mathematical Society, 2007
Department of Pure Mathematics, Queens University Belfast, Belfast BT7 1NN,
United Kingdom
E-mail address: mmcgarvey07@qub.ac.uk
Department of Pure Mathematics, Queens University Belfast, Belfast BT7 1NN,
United Kingdom
E-mail address: i.todorov@qub.ac.uk

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Noncommutative geometry as a functor


Igor V. Nikolaev
To N. I. Vulpe on his 60th birthday.

Abstract. In this note the noncommutative geometry is interpreted as a


functor, whose range is a family of the operator algebras. Some examples are
given and a program is sketched.

Introduction
A point made in this note is that some traditional noncommutative spaces (i.e
the C -algebras, Banach or associative algebras) can be viewed as a generalized
homology in the sense that there exist functors with the range in such spaces. The
domain of the functors can be any interesting category, e.g. the Hausdor spaces,
compact manifolds, Riemann surfaces, etc. We shall give examples of such functors.
The above functors have a long history, rather natural and well familiar to specialists. A foundational example is given by the Gelfand-Naimark functor, which
maps the category of the Hausdor spaces to the category of commutative C algebras. It was conjectured by Novikov and proved by Kasparov [7] & Mischenko
[8] using the operator algebras, that in many cases the higher signature of the
smooth n-dimensional manifold is a homotopy invariant of the manifold. The respective functor is known as an assembly map. In dynamics, Cuntz & Krieger [1]
constructed a functor from the category of topological Markov chains to a category
of the C -algebras (the Cuntz-Krieger algebras). There are many more examples
to add to the list.
As long as the functor is constructed, one can calculate the (noncommutative)
invariants available for that functor. On the face of it, the C -algebras are a way
more complex objects, than the abelian groups, rings or modules. However, a huge
variety of important families of the operator algebras have been recently classied in
terms of the algebraic K-theory [4] and more developments are expected down that
road. Clearly, the invariants of the above C -algebras are invariants of the objects
in the initial category. Thus, a problem of interpretation of the noncommutative
invariants in terms of the initial category stands up.
1991 Mathematics Subject Classication. Primary 46L; Secondary 18D.
Key words and phrases. Categories, functors, operator algebras.
Partially supported by NSERC.
c
2009
American Mathematical Society

1
151

152
2

IGOR V. NIKOLAEV

In relation to the traditional parts of noncommutative geometry (e.g. the index


theory, cyclic cohomology, quantum groups, etc), the functorial approach is rather
pragmatic than a romantic development, in the sense that the noncommutative
spaces are viewed as a toolkit in the study of the classical spaces. The problem
has two parts: (i) to map the given category into a family of the noncommutative
spaces and (ii) to prove that the mapping is a functor. Note that (ii) is where one
can expect most of the diculties to appear.
The note is organized as follows. In section 1 some examples of functors with
the range in a category of the operator algebras are given and their noncommutative
invariants are analyzed. In section 2, the draft of a program is sketched.
Acknowledgments. I am grateful to Wolfgang Krieger for useful discussions and
Ryan M. Rohm for helpful remarks on the rst draft of the note.
1. Three examples
In this section, some examples of the functors with the range in a family of
the operator algebras are given. In the two of three cases, the functors are noninjective. The list is by no means nal and the reader is encouraged to extend the
list.
1.1. The Gelfand and Naimark functor. A. This is a foundational example. Let X be a locally compact Hausdor space. By C(X) one understands a
commutative C -algebra of all functions f : X C, which vanish at the innity.
The norm on C(X) is the supremum norm. Recall that every point x X can
be thought of as a linear multiplicative functional x
: C(X) C. The Gelfand
transform F : X C(X) is dened by the formula x  f , where f C(X) is such
that x
(f ) = f (x).
B. Let h : X Y be a continuous map between the Hausdor spaces X and Y . It
can be easily shown that the map h = F 1 h F is a homomorphism from the
C -algebra C(Y ) to C(X). In other words, F is a contravariant functor from the
locally compact Hausdor spaces to the commutative C -algebras:
X

continuous map
-

F
?

C(X)

homomorphism


?
C(Y )

C. Note that F is an injective functor. The functor F does not produce new invariants of the Hausdor spaces, because of the following isomorphism: K alg (C(X))
=
K top (X), where K alg and K top are the algebraic and the topological K-theories,
respectively.
1.2. Anosov automorphisms of the two-dimensional torus. A. Let us
consider an application of the operator algebras to a problem in topology. Recall
that an automorphism : T 2 T 2 of the two-dimensional torus is called Anosov,
if it is given by a matrix A = (a11 a12 a21 a22 ) GL(2, Z), such that |a11 + a22 | > 2.

NONCOMMUTATIVE GEOMETRY AS A FUNCTOR

153
3

We wish to construct a functor (an assembly map) :  A , such that for every
h Aut (T 2 ) the following diagram commutes:
conjugacy
-

 = h h1

A K

isomorphism
-

?
A K,

where A is an AF -algebra and K is the C -algebra of compact operators on a


Hilbert space. In other words, if ,  are conjugate automorphisms, then the AF algebras A , A are stably isomorphic (Morita equivalent).
B. The map :  A is as follows. For simplicity, let a11 + a22 > 2. (The case
a11 + a22 < 2 is treated likewise 1 .) Note that without loss of the generality,
one can assume that aij 0 for a proper basis in the homology group H1 (T 2 ; Z).
Consider an AF -algebra, A , given by the periodic Bratteli diagram shown in Fig.
1, where aij indicate the multiplicity of the respective edge of the graph. We
encourage the reader to verify that :  A is a correctly dened function on
the set of Anosov automorphisms given by the hyperbolic matrices with the nonnegative entries. Note that is not injective, since and all its powers map to the
same AF -algebra.

b
@
a

a11

b
@
a

a11

b
@
a

a11

b ...

12@
12@
12@
b
@ a21 @ a21 @ a21 @
@b
@b
@b
@b . . .
a22
a22
a22

A = (a11 a12 a21 a22 ),

Figure 1. The AF -algebra A .


C. Let us show that if ,  Aut (T 2 ) are the conjugate Anosov automorphisms,
then A , A are the stably isomorphic AF -algebras. Indeed, let  = h h1
for an h Aut (X). Then A = T A T 1 for a matrix T GL(2, Z). Note that
(A )n = (T A T 1 )n = T An T 1 , where n N. We shall use the following criterion
([2], Theorem 2.3): the AF -algebras A, A are stably isomorphic if and only if their
Bratteli diagrams contain a common block of an arbitrary length. Consider the
following sequences of matrices: A A . . . A and T A A . . . A T 1 , which mimic
the Bratteli diagrams of A and A . Letting the number of blocks A tend to
innity, we conclude that A K
= A K.
1I am grateful to Ian Putnam for pointing out this fact to me.

154
4

IGOR V. NIKOLAEV

D. The conjugacy problem for the Anosov automorphisms can now be recast in the
terms of the AF -algebras: nd the invariants of the stable isomorphism classes of
the stationary AF -algebras. One such invariant is due to Handelman [5]. Consider
an eigenvalue problem for the hyperbolic matrix A GL(2, Z): A vA = A vA ,
(1) (2)
where A > 1 is the Perron-Frobenius eigenvalue and vA = (vA , vA ) the corre(i)
sponding eigenvector with the positive entries normalized so that vA K = Q(A ).
(1)
(2)
Denote by m = ZvA + ZvA a Z-module in the number eld K. Recall that the
coecient ring, , of module m consists of the elements K such that m m.
It is known that is an order in K (i.e. a subring of K containing 1) and, with
no restriction, one can assume that m . It follows from the denition, that m
coincides with an ideal, I, whose equivalence class in we shall denote by [I]. It has
been proved by Handelman, that the triple (, [I], K) is an arithmetic invariant of
the stable isomorphism class of A : the A , A are stably isomorphic AF -algebras
if and only if =  , [I] = [I  ] and K = K  . It is interesting to compare the
operator algebra invariants with those obtained in [10].
E. Let M be a mapping torus of the Anosov automorphism , i.e. a threedimensional manifold {T 2 [0, 1] | (x, 0)  ((x), 1) x T 2 }. The M is known
as a solvmanifold, since it is the quotient space of a solvable Lie group. It is an
easy exercise to show that the homotopy classes of M are bijective with the conjugacy classes of . Thus, the noncommutative invariant (, [I], K) is a homotopy
invariant of the compact manifold M .
1.3. Complex tori and the Eros-Shen algebras. A. Let us consider
an application of the operator algebras to a problem in conformal geometry. Let
H := {z C | Im (z) > 0} be a complex number. Recall that the quotient
space E = C/(Z + Z ) is called a complex torus. It is well-known that the complex
tori E , E  are isomorphic, whenever  mod SL(2, Z), i.e.  = a+b
c+d , where
a, b, c, d Z and ad bc = 1.
B. Let 0 < < 1 be an irrational number given by the regular continued fraction:
1
, a0 N {0} and ai N for i 1.
= a0 +
1
a1 +
a2 + . . .
By the Eros-Shen algebra [3], A , one understands an AF -algebra given by the
Bratteli diagram:
a0 a1
b
b
b
b @ @
@b @b @b

...
...

Figure 2. The Eros-Shen algebra A .


where ai indicate the number of edges in the upper row of the diagram. Recall
that A , A are said to be stably isomorphic if A K
= A K. It is known that
the Eros-Shen algebras A , A are stably isomorphic i  mod SL(2, Z).
Comparing the categories of complex tori and the Eros-Shen algebras, one cannot

155
5

NONCOMMUTATIVE GEOMETRY AS A FUNCTOR

fail to observe that for the generic objects in the respective categories, the corresponding morphisms are isomorphic as groups. Let us show that the observation is
not a mere curiosity there exists a functor, F , which makes the following diagram
commute:
E

isomorphic
-

E 

F
?
A

stably isomorphic
-

?
A

C. To construct the map F : E  A , we shall use the Hubbard-Masur homeomorphism h : H {pt} T 2 {0}, where T 2 is the space of measured foliations
on the two-torus [6]. Each measured foliation F T 2 looks like a family of the
parallel lines of a slope endowed with an invariant transverse measure (Fig.3).
a measured
If is a closed 1-form on T 2 , then the trajectories of dene

 foliation

F T 2 and vice versa. It is not hard to see that = 1 and = 2 / 1 ,
where {1 , 2 } is a basis in H1 (T 2 ; Z). Denote by N an invariant (Neron) dierential of the complex torus C/(1 Z + 2 Z). It is well known that 1 = 1 N and

2 = 2 N . Let be a projection acting by the formula (, )  . The assembly
map F is given by the composition F = h, where h is the Hubbard-Masur homeomorphism. In other words, the assembly map E  A can be written explicitly
as:


h
E = E( N )/( N )  F(1 )/( )  A( )/( ) = A .
2






 
Figure 3. The measured foliation F on T 2 = R2 /Z2 .
D. Let us show that the map F is a covariant functor. Indeed, an isomorphism
E E  is induced by an automorphism Aut (T 2 ) of the two-torus. Let
realizing such
an automorphism. From
A = (abcd) GL(2; Z) be a matrix


c+d
and  =
the formulas for F , one gets  = ( c1 +d2 N )/( a1 +b2 N ) = a+b


c+d
( c1 +d2 )/( a1 +b2 ) = a+b
. Thus, F sends isomorphic complex tori to the
stably isomorphic Eros-Shen algebras. Moreover, the formulas imply that F is
a covariant functor. Note that F is not an injective functor, since it contains a
projective map .
E. Finally, let us consider a noncommutative invariant coming from the functor
F . The ECM is said to have a complex multiplication, if the endomorphism ring
of the lattice Z + Z exceeds Z. It is an easy exercise to show (in view of the

156
6

IGOR V. NIKOLAEV

explicit fromulas for F ) that F (ECM ) = A , where is a quadratic irrationality.


In this case the continued fraction of is eventually periodic and we let r be the
length of the minimal period of . Clearly, the integer r is an invariant of the stable
isomorphism class of the Eros-Shen algebra A . To interpret the noncommutative
invariant r in terms of ECM , recall that ECM is isomorphic to a projective elliptic
curve dened over a subeld K = Q(j(ECM )) of C, see e.g. [9] p.38, Prop. 4.5
(b). It is known (the Mordell-Weil theorem), that the K-rational points of ECM
make an abelian group, whose innite part has rank R 0. We conclude by the
following
Conjecture 1.1. For every elliptic curve with a complex multiplication R =
r 1.
2. Sketch of a program
One can outline a program by the indication of: (i) an object of the study,
(ii) a typical problem and (iii) a set of the exercises. A functorial noncommutative
geometry (FNCG) can be described as follows.
The object of study. The FNCG studies the non-trivial functors from a category
of the classical objects, G, to a category of the noncommutative spaces (operator,
Banach or associative algebras), A. The functor can be non-injective. The category
A is (possibly) endowed with a good set of the invariants.
A typical problem. The main problem of FNCG is the construction of new
invariants of the objects in G from the known noncommutative invariants of A.
A reconstruction of the classical invariants from the noncommutative invariants is
regarded as a partial solution of the main problem.
Exercises. Let A be a category of:
(i) the U HF algebras;
(ii) the Cuntz-Krieger algebras OA with det (A) = 1.
Find a category G corresponding to A and solve the typical problem. (Hint: for the
Cuntz-Krieger algebras of type (ii), the category G consists of the homotopy classes
of the ber bundles MA over the unit circle S 1 with the ber an n-dimensional torus
T n and the monodromy given by the matrix A.)
References
[1] J. Cuntz and W. Krieger, A class of C -algebras and topological Markov chains, Invent.
Math. 56 (1980), 251268.
[2] E. G. Eros, Dimensions and C -Algebras, in: Conf. Board of the Math. Sciences No.46,
Amer. Math. Soc., Providence, RI, 1981.
[3] E. G. Eros and C. L. Shen, Approximately nite C -algebras and continued fractions,
Indiana Univ. Math. J. 29 (1980), 191204.
[4] G. A. Elliott and A. S. Toms, Regularity properties in the classication program for separable
amenable C -algebras, Bull. Amer. Math. Soc. 45 (2008), 229245.
[5] D. Handelman, Positive matrices and dimension groups aliated to C -algebras and topological Markov chains, J. Operator Theory 6 (1981), 5574.
[6] J. Hubbard and H. Masur, Quadratic dierentials and foliations, Acta Math. 142 (1979),
221274.
[7] G. Kasparov, Topological invariants of elliptic operators I: K-homology, Izv. Akad. Nauk
SSSR, Ser. Math. 39 (1975), 796838.

NONCOMMUTATIVE GEOMETRY AS A FUNCTOR

157
7

[8] A. S. Mischenko, Innite-dimensional representations of discrete groups and higher signatures, Izv. Akad. Nauk SSSR, Ser. Math. 38 (1974), 81106.
[9] J. H. Silverman, Advanced Topics in the Arithmetic of Elliptic Curves, GTM 151, Springer,
1994.
[10] D. I. Wallace, Conjugacy classes of hyperbolic matrices in SL(n, Z) and ideal classes in an
order, Trans. Amer. Math. Soc. 283 (1984), 177184.
The Fields Institute for Mathematical Sciences, Toronto, ON, Canada
Current address: 101-315 Holmwood Ave, Ottawa, ON, K1S 2R2, Canada
E-mail address: igor.v.nikolaev@gmail.com

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Simple group graded rings and maximal commutativity

Johan Oinert
Abstract. In this paper we provide necessary and sucient conditions for
strongly
 group graded rings to be simple. For a strongly group graded ring
Rg the grading group G acts, in a natural way, as automorphisms
R=
gG
of the commutant of the neutral component subring Re in R and of the center
of Re . We show that if R is a strongly G-graded ring where Re is maximal
commutative in R, then R is a simple ring if and only if Re is G-simple (i.e.
there are no nontrivial G-invariant ideals). We also show that if Re is commutative (not necessarily maximal commutative) and the commutant of Re is
G-simple, then R is a simple ring. These results apply to G-crossed products
in particular. A skew group ring Re  G, where Re is commutative, is shown
to be a simple ring if and only if Re is G-simple and maximal commutative
in Re  G. As an interesting example we consider the skew group algebra
C(X) h
Z associated to a topological dynamical system (X, h). We obtain
necessary and sucient conditions for simplicity of C(X) h
Z with respect to
the dynamics of the dynamical system (X, h), but also with respect to algebraic properties of C(X) as a subalgebra of C(X) h
Z. Furthermore, we show
that for any strongly G-graded ring R each nonzero ideal of R has a nonzero
intersection with the commutant of the center of the neutral component.

1. Introduction
The aim of this paper is to highlight the important role that maximal commutativity of the neutral component subring plays in a strongly group graded ring
when investigating simplicity of the ring itself. The motivation comes from the
theory of C -crossed product algebras associated to topological dynamical systems.
To each topological dynamical system, (X, h), consisting of a compact Hausdor
2000 Mathematics Subject Classication. Primary 13A02, 16S35; Secondary 17C20, 14C22.
Key words and phrases. Graded rings, Ideals, Simple rings, Maximal commutative subrings,
Picard groups, Invariant ideals, Crossed products, Skew group rings, Minimal dynamical systems.
This work was partially supported by The Swedish Research Council, The Crafoord Foundation, The Royal Physiographic Society in Lund, The Swedish Royal Academy of Sciences, The
Swedish Foundation of International Cooperation in Research and Higher Education (STINT)
and LieGrits, a Marie Curie Research Training Network funded by the European Community
as project MRTN-CT 2003-505078. The author wishes to thank Magnus Goeng, Patrik Lundstr
om, Sergei Silvestrov and in particular Christian Svensson for useful discussions on the topic
of this paper.
c Mathematical
0000
(copyright Society
holder)
c
2009
American

1
159


JOHAN OINERT

160
2

space X and a homeomorphism h : X X, one may associate a C -crossed prodC

uct algebra1 C(X) h Z (see e.g. [23]). In the recent paper [22], C. Svensson and
J. Tomiyama prove the following theorem.
Theorem 1.1. The following assertions are equivalent:
(i) (X, h) is topologically free (i.e. the aperiodic points are dense in X).
C

(ii) I C(X) = {0} for each nonzero ideal I of C(X) h Z.

(iii) C(X) is a maximal commutative C -subalgebra of C(X) h Z.


This theorem is a generalization (from closed ideals to arbitrary ideals) of a
well-known theorem in the theory of C -crossed products associated to topological
dynamical system (see e.g. [23] for details). Theorem 1.1 is very useful when
proving the following theorem, which originally appeared in [17].
C

Theorem 1.2. Suppose that X is innite. Then C(X) h Z is simple if and


only if (X, h) is minimal (i.e. each orbit is dense in X).
In [19, 20, 21] C. Svensson, S. Silvestrov and M. de Jeu proved various analogues of Theorem 1.1 for (algebraic) skew group algebras which are strongly graded
by Z. It then became natural to investigate if their results could be generalized
to other types of (strongly) graded rings and in [11, 12, 13, 14] an extensive investigation of the intersection between arbitrary nonzero ideals of various types of
graded rings and certain subrings, was carried out. Given a subset S of a ring R
we denote by CR (S) the commutant of S in R, i.e. the set of all elements of R
which commute with each element of S. In particular CR (R), i.e. the center of R,
is denoted by Z(R). In the recent paper [15], the following theorem was proven.

Theorem 1.3. If R =
gG Rg is a strongly G-graded ring, where Re is
commutative, then
I CR (Re ) = {0}
for each nonzero ideal I of R.
This implies that if R is a strongly G-graded ring where Re is maximal commutative in R, then each nonzero ideal of R has a nontrivial intersection with Re .
For skew group rings the following was shown in [14, Theorem 3].
Theorem 1.4. Let R = Re  G be a skew group ring satisfying either of the
following two conditions:
Re is an integral domain and G is an abelian group.
Re is commutative and G is a torsion-free abelian group.
Then the following two assertions are equivalent:
(i) The ring Re is a maximal commutative subring of R.
(ii) I Re = {0} for each nonzero ideal I of R.
This theorem can be seen as a generalization of the algebraic analogue of Theorem 1.1 and it is applicable to the skew group algebra which sits densely inside
C

the C -crossed product algebra C(X) h Z. In the theory of graded rings, one
C

1To avoid confusion, we let C(X)  Z denote the C -crossed product algebra in contrast to

the (algebraic) skew group algebra, which is denoted C(X) h


Z.

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

161
3

theorem which provides sucient conditions for a strongly group graded ring to be
simple is the following, proven by F. Van Oystaeyen in [25, Theorem 3.4].

Theorem 1.5. Let R =
gG Rg be a strongly G-graded ring such that the
morphism G Pic(Re ), dened by g [Rg ], is injective. If Re is a simple ring,
then R is a simple ring. (The Picard group, Pic(Re ), is dened in Section 2.2.)
A skew group ring is an example of a strongly graded ring. Given a skew
group ring R = Re  G, the grading group G acts as automorphisms of Re . The
results in [3] show that simplicity of a skew group ring R is intimately connected
to the nonexistence of G-invariant nonzero proper ideals of Re . Given a strongly
G-graded ring R, the grading group G acts, in a canonical way, as automorphisms
of CR (Re
) (see Section 2.1). This means that for an arbitrary strongly G-graded
ring R = gG Rg , one may speak of G-invariant nonzero proper ideals of CR (Re )
and try to relate the nonexistence of such ideals to simplicity of R, in a manner
similar to the case of skew group rings.
In Section 2 we give denitions and background information necessary for the
understanding of the rest of this paper. In Section 3 we generalize [15, Corollary
3] and show that in a strongly G-graded ring R each nonzero ideal has a nonzero
intersection with CR (Z(Re )) (Theorem 3.1). Furthermore, we generalize [14, Theorem 3] and show that for a skew group ring Re  G where Re is commutative,
each nonzero ideal of Re  G has a nonzero intersection with Re if and only if Re
is maximal commutative in Re  G (Theorem 3.5).
The main objective of Section 4 is to describe the connection between maximal
commutativity of Re in a strongly group graded ring R and injectivity of the
canonical map G Pic(Re ). In Section 5 we show that if A0
G is a simple
crystalline graded ring where A0 is commutative, then A0 is G-simple (Proposition
5.1). In Example 5.3 we apply this result to the rst Weyl algebra. In Section 6 we
investigate simplicity of a strongly G-graded ring R with respect to G-simplicity
and maximal commutativity of Re . In particular we show that if R is a strongly
G-graded ring where Re is maximal commutative in R, then Re is G-simple if and
only if R is simple (Theorem 6.6). We also show the slightly more general result in
one direction, namely that that if CR (Re ) is G-simple (with respect to the canonical
action) and Re is commutative (not necessarily maximal commutative!), then R
is simple (Proposition 6.5). Thereafter we investigate the simplicity of skew group
rings and generalize [3, Corollary 2.1] and [3, Theorem 2.2], by showing that if Re
is commutative, then the skew group ring Re  G is a simple ring if and only if Re
is G-simple and a maximal commutative subring of Re  G (Theorem 6.13). As
an example, we consider the skew group algebra associated to a dynamical system.
In Section 7 we consider the algebraic crossed product C(X) h Z associated
to a topological dynamical system (X, h). Under the assumption that X is innite,
we show that C(X) h Z is simple if and only if (X, h) is a minimal dynamical
system or equivalently if and only if C(X) is Z-simple and maximal commutative
in C(X) h Z (Theorem 7.6). This result is a complete analogue to the well-known
result for C -crossed product algebras associated to topological dynamical systems.
2. Preliminaries
Throughout this paper all rings are assumed to be unital and associative and
unless otherwise is stated we let G be an arbitrary group with neutral element e.


JOHAN OINERT

162
4

A ring R is said to be G-graded if there is a family {Rg }gG of additive subgroups of R such that

R=
Rg and Rg Rh Rgh
gG

for all g, h G. Moreover, if Rg Rh = Rgh holds for all g, h G, then R is said to be


strongly G-graded. The product Rg Rh is here the usual module product consisting
of all nite sums of ring products rg rh of elements rg Rg and rh Rh , and not
just the set of all such ring products. For any graded ring R it follows directly
from the gradation that Re is a subring of R, and that Rg is an Re -bimodule for
each g G. We shall refer to Rg as the homogeneous component of degree g G,
and in particular to Re as the neutral component. Let U (R) denote the group of
multiplication invertible elements of R. We shall say that R is a G-crossed product
if U (R) Rg = for each g G.

2.1. Strongly G-graded rings. For each G-graded ring R = gG Rg one
has 1R Re (see [9, Proposition 1.1.1]), and if we in addition assume that R is a
strongly G-graded ring, i.e. Rg Rg1 = Re for each g G, then for each g G there
(i)
(i)
exists a positive integer ng and elements ag Rg , bg1 Rg1 for i {1, . . . , ng },
such that
(2.1)

ng


(i)

a(i)
g bg 1 = 1R .

i=1

For every CR (Re ), and in particular for every Z(Re ) CR (Re ), and
g G we dene

(2.2)

g () =

ng


(i)

a(i)
g bg 1 .

i=1
(i)

(i)

The denition of g is independent of the choice of the ag s and bb1 s (see e.g.
[15]). For a proof of the following lemma we refer to [15, Lemma 3].

Lemma 2.1. Let R = gG Rg be a strongly G-graded ring, g G and write
ng (i) (i)
(i)
(i)
i=1 ag bg 1 = 1R for some ng > 0 and ag Rg , bg 1 Rg 1 for i {1, . . . , ng }.
ng (i) (i)
For each CR (Re ) dene g () by g () =
i=1 ag bg 1 . The following
properties hold:
(i) g () is the unique element of R satisfying
(2.3)

rg = g () rg ,

rg R g .

Furthermore, g () CR (Re ) and if Z(Re ), then g () Z(Re ).


(ii) The group G acts as automorphisms of the rings CR (Re ) and Z(Re ),
with each g G sending any CR (Re ) and Z(Re ), respectively, to
g ().
(iii) Z(R) = { CR (Re ) | g () = , g G}, i.e. Z(R) is the xed
subring CR (Re )G of CR (Re ) with respect to the action of G.
The map , dened in Lemma 2.1, will be referred to as the canonical action.

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

163
5

2.2. The Picard group of Re , Pic(Re ). We shall


now give a brief description
of the Picard group of Re in a strongly graded ring R = gG Rg . For more details
we refer to [2].
Definition 2.2 (Invertible module). Let A be a ring. An A-bimodule M
is said to be invertible if and only if there exists an A-bimodule N such that
M A N
=A
= N A M as A-bimodules.
Given a ring A, the Picard group of A, denoted Pic(A), is dened as the set of
A-bimodule isomorphism classes of invertible A-bimodules, and the group operation
is given by 
A .
If R = gG Rg is a strongly G-graded ring, the homomorphism of Rg Re Rh
into Rgh sending rg rh into rg rh for all rg Rg and rh Rh , is an isomorphism
of Re -bimodules, for any g, h G (see [4, p.336]). This implies that Rg is an
invertible Re -bimodule for each g G. We may now dene a group homomorphism
: G Pic(Re ), g
[Rg ], i.e. each g G is mapped to the isomorphism class
inside Pic(Re ) to which the invertible Re -bimodule Rg belongs.
2.3. Crystalline graded rings. We shall begin this section by recalling the
denition of a crystalline graded ring. We would also like to emphasize that rings
belonging to this class are in general not strongly graded.
Definition 2.3 (Pre-crystalline graded ring). An associative and unital ring
A is said to be pre-crystalline graded if
(i) there is a group G (with neutral element e),
(ii) there is a map u : G A, g
ug such that ue = 1A and ug = 0 for
every g G,
(iii) there is a subring A0 A containing 1A ,
such that the following conditions are satised:

(P1) A = gG A0 ug .
(P2) For every g G, ug A0 = A0 ug is a free left A0 -module of rank one.
(P3) The decomposition in P1 makes A into a G-graded ring with A0 = Ae .
Lemma 2.4 (see [10]). With notation and denitions as above:
(i) For every g G, there is a set map g : A0 A0 dened by ug a =
g (a) ug for a A0 . The map g is a surjective ring morphism. Moreover,
e = idA0 .
(ii) There is a set map : G G A0 dened by us ut = (s, t) ust for
s, t G. For any triple s, t, w G and a A0 the following equalities
hold:
(2.4)
(2.5)

(s, t)(st, w) = s ((t, w))(s, tw)


s (t (a))(s, t) = (s, t)st (a)

(iii) For every g G we have (g, e) = (e, g) = 1A0 and (g, g 1 ) =


g ((g 1 , g)).
A pre-crystalline graded ring A with the above properties
 will be denoted by
A0
G and each element of this ring is written as a sum
gG rg ug with coecients rg A0 , of which only nitely many are nonzero. In [10] it was shown that for
pre-crystalline graded rings, the elements (s, t) are normalizing elements of A0 , i.e.
A0 (s, t) = (s, t) A0 for each s, t G. For a pre-crystalline graded ring A0
G,


JOHAN OINERT

164
6

we let S(G) denote the multiplicative set in A0 generated by {(g, g 1 ) | g G}


and let S(G G) denote the multiplicative set generated by {(g, h) | g, h G}.
Lemma 2.5 (see [10]). If A = A0
G is a pre-crystalline graded ring, then the
following assertions are equivalent:
(i) A0 is S(G)-torsion free.
(ii) A is S(G)-torsion free.
(iii) (g, g 1 )a0 = 0 for some g G implies a0 = 0.
(iv) (g, h)a0 = 0 for some g, h G implies a0 = 0.
(v) A0 ug = ug A0 is also free as a right A0 -module, with basis ug , for every
g G.
(vi) For every g G, g is bijective and hence a ring automorphism of A0 .
Definition 2.6 (Crystalline graded ring). A pre-crystalline graded ring A0
G,
which is S(G)-torsion free, is said to be a crystalline graded ring.
Remark 2.7. Note that G-crossed products are examples of crystalline graded
rings. In fact, suppose that R is a G-crossed product and put A = R. For each
g G, we may pick some ug Rg U (R). Choose ue = 1R and A0 = Re . It
is now clear that Rg = A0 ug = ug A0 for each g G, and that {u
g }gG is a

basis for A as a free left (and right) A0 -module. By assumption A = gG A0 ug
with Ae = A0 . This shows that A is pre-crystalline graded. Recall that for each
g G, ug is chosen to be a unit in R and hence, from Lemma 2.4 (ii), we get
that (s, t) = us ut u1
st U (A0 ) for all s, t G. This certainly shows that A0 is
S(G)-torsion free and hence A = R is a crystalline graded ring.
The notation for G-crossed products is inherited from the crystalline graded
rings, e.g. we shall write {ug }gG for the basis elements. In particular, in the proof
of Theorem 3.5 where we consider a skew group ring, which is a special case of a
G-crossed product, we shall use this notation. However, by custom we shall write

Re 
G instead of Re G.
3. Ideals of strongly graded rings
In this section we shall improve some earlier results. We begin by making a
generalization of Theorem 1.3 ([15, Corollary 3]). The following proof is based on
the same technique as in [15], but we will make it somewhat shorter by doing a
proof by contra positivity.

Theorem 3.1. If R = gG Rg is a strongly G-graded ring, then
I CR (Z(Re )) = {0}
for each nonzero ideal I of R.
Proof. Let I be an ideal of R such that I CR (Z(Re )) = {0}. If we can
show that I = 
{0}, then the desired conclusion follows by contra positivity. Take an
arbitrary x = gG xg I. If x I CR (Z(Re )), then x = 0 by the assumption.
Therefore, suppose that I \ CR (Z(Re )) is not empty. We may choose some x
I \ CR (Z(Re )) such that N = # supp(x) = #{g G | xg = 0} Z>0 is as small
as possible. Furthermore, we may assume that e supp(x). Indeed, take any t
supp(x) and choose some rt1 Rt1 such that x = rt1 x = 0 and e supp(x ).
nt (i) (i)
at bt1 as in
It is always possible to choose such an rt1 , because if 1R = i=1

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

165
7

(i)

(2.1), then bt1 xt must be nonzero for some i {1, . . . , nt }, for otherwise we would
have 1R xt = 0 which would be contradictory (since xt = 0). Note that x I is
nonzero and since I CR (Z(Re )) = {0} we conclude that x I \ CR (Z(Re )). By
the assumption on N we conclude that # supp(x ) = supp(x) = N . Now, take an
arbitrary a Z(Re ). Then x = ax x a I but clearly e
/ supp(x ) and hence

by the assumption on N we get that x  I \ CR (Z(Re )), thus x = 0. Since a
Z(Re ) was chosen arbitrarily we get x CR (Z(Re )) which is a contradiction. 
Remark 3.2. Note that Re CR (Z(Re )). If Re is commutative, then clearly
Re = Z(Re ) and we obtain Theorem 1.3 as a special case of Theorem 3.1.
For a crystalline graded ring A0
G we obtain the following result which generalizes [14, Corollary 8].
1
Theorem 3.3. If A = A0
)
G is a crystalline graded ring with (g, g
Z(A0 ) for all g G, then
I CA (Z(A0 )) = {0}

for each nonzero ideal I of A0


G.

Proof. Let x = gG ag ug , with ag A0 for g G, be a nonzero element
of the crystalline graded ring A0
G. Pick some t G such that at = 0. For
x = x ut1 we have e supp(x ). Indeed, in degree e of x we have (at ut )ut1 =
at (t, t1 ) = (t, t1 ) at and, by Lemma 2.5 (iii), this is a nonzero element of A0 .
The rest of the proof is analogous to the proof of Theorem 3.1.

For an element r of a commutative ring A, the annihilator ideal of r in A is
dened to be the set Ann(r) = {b A | rb = 0}. The following lemma from [11,
Corollary 6] applies to G-crossed products and in particular skew group rings.
Lemma 3.4. Let Re 
G be a G-crossed product with Re commutative. The
subring Re is maximal commutative in Re 
G if and only if, for each pair (s, rs )
(G \ {e}) (Re \ {0}), there exists a Re such that s (a) a  Ann(rs ).
The following theorem is a generalization of Theorem 1.4 ([14, Theorem 3])
and the proof makes use of the same idea as in [14]. However, in this proof we
make a crucial observation and are able to make use of an important map.
Theorem 3.5. Let R = Re  G be a skew group ring with Re commutative.
The following two assertions are equivalent:
(i) Re is a maximal commutative subring of R.
(ii) I Re = {0} for each nonzero ideal I of R.
Proof. By Theorem 3.1 (i) implies (ii) for the (strongly graded) skew group
ring R. We shall now show that (ii) implies (i). Suppose that Re is not maximal
commutative in R. If we can show that there exists a nonzero ideal I of R, such
that I Re = {0}, then by contra positivity we are done. Let {ug }gG be a basis
for R as a free left (and right) Re -module, as in Section 2.3. By the assumption and
Lemma 3.4 there exists some s G \ {e} and rs Re \ {0} such that rs s (a) = rs a
for each a Re . Let us choose such a pair (s, rs ) and let I be the twosided ideal
of R generated by rs rs us . The ideal I is obviously nonzero, and furthermore
it is spanned by elements of the form ag ug (rs rs us ) ah uh where g, h G and


JOHAN OINERT

166
8

ag , ah Re . By commutativity of Re and the properties of rs we may rewrite this


expression.
ag ug (rs rs us ) ah uh

ag ug (rs ah rs s (ah ) us ) uh
  

=
=

ag ug rs ah (1R us ) uh
ag g (rs ah ) ug (1R us ) uh

ag g (rs ah ) (ugh ugsh )






b ugh b ugsh

=rs ah

:=b

(3.1)

Each element of I is a sum of elements of the form (3.1), where b Re and g, h G.


Dene a map


 : Re  G Re ,
ag ug

ag .
gG

gG

It is clear that  is additive and one easily sees that  is identically zero on I.
Furthermore, |Re , i.e. the restriction of  to Re , is injective. Take an arbitrary
m I Re . Clearly (m) = 0 since m I and by the injectivity of |Re we conclude
that m = 0. Hence I Re = {0}. This concludes the proof.

Remark 3.6. It is not dicult to see that the map  is multiplicative if and
only if the action is trivial, i.e. Re  G is a group ring. In that situation the
map  is commonly referred to as the augmentation map. However, note that the
preceding proof does not require  to be multiplicative.
4. The map : G Pic(Re ) and simple strongly graded rings
We begin by recalling a useful lemma.

Lemma 4.1 ([15]). Let R = gG Rg be a strongly G-graded ring. If a R
is such that
a Rg = {0} or Rg a = {0}
for some g G, then a = 0.
If we assume that Re is maximal commutative in the strongly G-graded ring
R, then we can say the following about the canonical map : G Pic(Re ).

Proposition 4.2. Let R =
gG Rg be a strongly G-graded ring. If Re is
maximal commutative in R, then the map : G Pic(Re ), g
[Rg ], is injective.
Proof. Let Re be maximal commutative in R. Suppose that : G Pic(Re )
is not injective. This means that we can pick two distinct elements g, h G such
that Rg
= Rh as Re -bimodules. Let f : Rg Rh be a bijective Re -bimodule
homomorphism. By our assumptions Re = CR (Re ) and hence we can use the map
: G Aut(Re ) dened by (2.2) to write
(4.1)

h (b) f (rg ) = f (rg ) b = f (rg b) = f (g (b) rg ) = g (b) f (rg )


  
Rh

for any b Re and rg Rg . (It is important to note that g (b) Re since b Re .)


The map f is bijective and in particular surjective. Hence, by (4.1) we conclude
that (h (b) g (b)) Rh = {0} for any b Re . It follows from Lemma 4.1 that

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

167
9

h (b) g (b) = 0 for any b Re . Hence g = h in Aut(CR (Re )) = Aut(Re )


and this implies g1 h = idRe . Now equation (2.3) shows that the homogeneous
component Rg1 h (= Re since g = h) commutes with Re , and hence Re is not
maximal commutative in R. We have reached a contradiction and this shows that
: G Pic(Re ) is injective.

The following proposition is a direct consequence of Theorem 1.4 and we shall
therefore omit the proof.
Proposition 4.3. Let R = Re  G be a skew group ring, where Re is a eld
and G is an abelian group. The following assertions are equivalent:
(i) The subring Re is maximal commutative in R.
(ii) R is a simple ring.
Example 4.4. Consider the group ring R = C[Z], which corresponds to the
special case of a skew group ring with trivial action. The so called augmentation
ideal, which is the kernel, ker(), of the augmentation map


 : C[Z] C,
ck k

ck
kZ

kZ

is a nontrivial ideal of R and hence R = C[Z] is not a simple ring. This conclusion
also follows directly from Proposition 4.3. Indeed, R = C[Z] is commutative and
hence R0 = C is not maximal commutative in C[Z].
The following proposition shows that in the case when Re is assumed to be
commutative, Theorem 1.5 is equivalent to Corollary 6.7 (see Section 6).

Proposition 4.5. Let R = gG Rg be a strongly G-graded ring. If Re is a
eld, then the following two assertions are equivalent:
(i) Re is maximal commutative in R.
(ii) The map : G Pic(Re ) is injective.
Proof. It follows from Proposition 4.2 that (i) implies (ii).
To prove that (ii) implies (i), let us assume that Re is not maximal commutative
in R. We want to show that is not injective and hence get the desired conclusion
by contra positivity.
By our assumptions, there exists some nonzero element rg Rg , for some
g = e, such that rg a = a rg for all a Re . Consider the set J = rg Rg1 Re .
Since rg commutes with Re and Rg1 is an Re -bimodule, J is an ideal of Re and as
rg Rg1 = {0} (this follows from Lemma 4.1 since rg = 0), we obtain rg Rg1 = Re
since Re is simple. Consequently, we conclude that there exists an sg1 Rg1
such that rg sg1 = 1R . In a symmetrical way we get Rg1 rg = Re which yields
wg1 rg = 1R for some wg1 Rg1 . Clearly wg1 = sg1 .
From the gradation we immediately conclude that Re rg Rg and Rg sg1
Re . By the equality sg1 rg = 1R we get Rg Re rg and hence Rg = Re rg .
Note that rg is invertible and hence a basis for the Re -bimodule Re rg . This shows
that Rg and Re belong to the same isomorphism class in Pic(Re ), and hence the
morphism : G Pic(Re ) is not injective. This concludes the proof.

Remark 4.6. The previous proof uses the same techniques as the proof of [25,
Theorem 3.4].

168
10

JOHAN OINERT

5. G-simple subrings of crystalline graded rings


If A is a ring and : G Aut(A) is a group action, then we say that an ideal
I of A is G-invariant if g (I) I for each g G. Note that it is equivalent to say
that g (I) = I for each g G. If there are no nontrivial G-invariant ideals of A,
then we say that A is G-simple. (Not to be confused with the term graded simple!)
Proposition 5.1. Let A0
G be a crystalline graded ring, where A0 is commutative. If A0
G is a simple ring, then A0 is a G-simple ring (with respect to
the action dened in Lemma 2.4).
Proof. Note that since A0 is commutative, the map : G Aut(A0 ) is a
group homomorphism. Let A0
G be a simple ring, and J an arbitrary nonzero

G-invariant ideal of A0 . One may verify that J


G is a nonzero ideal of A0 G.
(This follows from the fact that for each g G, A0 ug is a free left A0 -module with

basis ug .) Since A0
G is simple, we get J G = A0 G. Therefore A0 J G,
and from the gradation it follows that
A0 J A0
and hence A0 = J, which shows that A0 is G-simple.


Corollary 5.2. Let R =
gG Rg be a G-crossed product, where Re is
commutative. If R is a simple ring, then Re is a G-simple ring (with respect
to the canonical action).
Cx,y
Example 5.3. It is well-known that the rst Weyl algebra A = (xyyx1)
is
simple. The rst Weyl algebra is an example of a crystalline graded ring, with
G = (Z, +) and Ae = A0 = C[xy] (see e.g. [10] for details). Note that C[xy] is
not a simple ring. However, by Proposition 5.1 we conclude that A0 = C[xy] is in
fact Z-simple. As a side remark we should also mention that one can show that
A0 = C[xy] is a maximal commutative subring of the rst Weyl algebra A.

6. G-simple subrings of strongly G-graded rings


In
this section we shall describe how simplicity of a strongly G-graded ring
R = gG Rg is related to G-simplicity of the subrings Z(Re ) and CR (Re ). If Re
is commutative, then Re = Z(Re ), and hence we have an action : G Aut(Re ).

Proposition 6.1. Let R = gG Rg be a strongly G-graded ring, where Re
is commutative. If R is a simple ring, then Re is a G-simple ring (with respect to
the canonical action).
Proof. Let J be an arbitrary nonzero G-invariant ideal of Re . Denote by JR
the right ideal of R generated by J. From the fact that J is a G-invariant ideal of
Re we conclude that JR is also a left ideal of R. Indeed, for g, h G and c J,
rh Rh , sg Rg we have sg c rh = g (c) sg rh JR. Furthermore, R is unital
and hence JR must be nonzero. The ring R is simple and therefore we conclude
that JR = R. In particular we see that Re JR. From the gradation we get
Re JRe J Re
and hence J = Re . This shows that Re is G-simple.

The preceding proposition is a generalization of Corollary 5.2. In the following


useful lemma, Re is not required to be commutative.

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

169
11


Lemma 6.2. Let R = gG Rg be a strongly G-graded ring and S a subring
of CR (Re ) satisfying the following three conditions:
(i) 1R S.
(ii) S is invariant under G (with respect to the canonical action).
(iii) S is G-simple (with respect to the canonical action).
Then I S = {0} for each proper ideal I of R.
Proof. Let S be a subring of CR (Re ) satisfying conditions (i)-(iii) of the
above, and I be an ideal of R such that IS = {0}. The set J = IS is an ideal of S.
ng (i) (i)
By (ii), for any x J and every g G, we have g (x) = i=1
ag x bg1 IS = J.
This shows that J is a G-invariant ideal of S. By assumption J is nonzero and hence
by (iii), J = S. In particular this shows that 1R J I, and hence R = I.

By observing that both CR (Re ) and Z(Re ) are subrings of CR (Re ) satisfying
conditions (i) and (ii) of Lemma 6.2 we obtain the following corollary.

Corollary 6.3. Let R = gG Rg be a strongly G-graded ring. If CR (Re )
(respectively Z(Re )) is a G-simple ring (with respect to the canonical action), then
I CR (Re ) = {0} (respectively I Z(Re ) = {0}) for each proper ideal I of R.
Recall from [5], that a ring R is said to be a PI-ring (abbreviation for polynomial
identity ring) if for some n Z>0 there exists some f Z x1 , x2 , . . . , xn , i.e. the
free polynomial ring over Z in n variables, such that f (a1 , a2 , . . . , an ) = 0 for each
(a1 , a2 , . . . , an ) Rn . Furthermore, a ring is said to be semiprime if {0} is a
semiprime ideal [8, Denition 10.8, Denition 10.15].

Corollary 6.4. Let R = gG Rg be a semiprime PI-ring which is strongly
G-graded. If either Z(Re ) or CR (Re ) is a G-simple ring (with respect to the canonical action), then R is a simple ring.
Proof. Let I be a nonzero ideal of R. It follows from [18, Theorem 2] that
I Z(R) = {0}. Clearly Z(R) Z(Re ) CR (Re ) and hence by Corollary 6.3 we
conclude that I = R.

As we shall see Theorem 6.6 requires Re not only to be commutative, but
maximal commutative in R. We begin by proving the following which applies to
the more general situation when Re is not necessarily maximal commutative in R.

Proposition 6.5. Let R = gG Rg be a strongly G-graded ring, where Re is
commutative. If CR (Re ) is a G-simple ring (with respect to the canonical action),
then R is a simple ring.
Proof. Let I be an arbitrary nonzero ideal of R. Since Re is commutative it
follows from Theorem 1.3 that I CR (Re ) = {0}. By Corollary 6.3 we conclude
that I = R and hence R is a simple ring.

By combining Proposition 6.1 and Proposition 6.5 we get the following theorem.

Theorem 6.6. Let R = gG Rg be a strongly G-graded ring. If Re is maximal commutative in R, then the following two assertions are equivalent:
(i) Re is a G-simple ring (with respect to the canonical action).
(ii) R is a simple ring.

170
12

JOHAN OINERT

As an immediate consequence of Theorem 6.6 we get the following corollary,


which can also be retrieved from Theorem 1.5 together with Proposition 4.2.

Corollary 6.7. Let R = gG Rg be a strongly G-graded ring where Re is
maximal commutative in R. If Re is a simple ring, then R is a simple ring.
The following remark shows that the rings considered in Corollary 6.7 are in
fact G-crossed products.
 Remark 6.8. Recall that a commutative and simple ring is a eld. If R =
gG Rg is a strongly G-graded ring and Re is a eld, then R is a G-crossed
product. Indeed, for each g G, we have Rg Rg1 = Rg1 Rg = Re . Hence, for an
arbitrary g G we may x some nonzero a Rg and by Lemma 4.1 choose some
nonzero b Rg1 such that ab = c Re \ {0}. This means that c is invertible in
Re and hence a is right invertible in R, with right inverse bc1 . The other half of
Lemma 4.1 may be used to show that a also has a left inverse. We conclude that for
each g G, Rg contains an invertible element and hence R is a G-crossed product.
One should note that Proposition 6.5 and Theorem 6.6 are more general than
Theorem 1.5 in the sense that Re is not required to be simple. On the other hand,
this does not come for free. We have to make an additional assumption on Re ,
namely that it be commutative.
Remark 6.9. Note that Theorem 6.6 especially applies to G-crossed products.

One may think that for a simple strongly graded ring R =
gG Rg where
Re is commutative and G-simple, this would imply that Re would be maximal
commutative in R. In general this is not true, as the following example shows.
Example 6.10. Consider the eld of complex numbers C = R  Z2 as a Z2 graded twisted group ring (see e.g. [12] for details). Clearly C is simple as is R.
Hence R is also Z2 -simple, but it is not maximal commutative in C.
The purpose of the following example is to present a strongly group graded
ring which is not a crossed product, and to identify a G-simple subring.
Example 6.11 (A strongly group graded, noncrossed product, matrix ring).
Let R = M3 (C) denote the ring of 3 3-matrices over C. By putting

C C 0
0 0 C
R0 = C C 0 and R1 = 0 0 C
0 0 C
C C 0
one may verify that this denes a strong Z2 -gradation on R. However, note that
R is not a crossed product with this grading since the homogeneous component R1
does not contain any invertible element of M3 (C)! A simple calculation yields


a 0 0

Z(R0 ) = 0 a 0  a, b C


0 0 b
and in fact one may verify that CR (R0 ) = Z(R0 ). In order to dene an action
: Z2 Aut(Z(R0 )) we need to make a decomposition of the identity matrix
I = 1R , in accordance with (2.1). Let Ei,j denote the 3 3-matrix which has a 1

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

171
13

in position (i, j) and zeros everywhere else. The decomposition in R0 is trivial, but
in R1 we may for example choose
I = E1,3 E3,1 + E2,3 E3,2 + E3,2 E2,3 .
From these decompositions we are now able to dene the map : Z2 Aut(Z(R0 )).
One easily sees that Z(R0 ) has two nontrivial ideals. By calculating, we get
1 (E1,1 + E2,2 ) = E3,3
and
1 (E3,3 ) = E1,1 + E2,2 .
From this we conclude that the two nontrivial ideals of Z(R0 ) are interchanged by
the map 1 , and therefore they are not invariant under the action of Z2 . This shows
that for our simple ring M3 (C), the subring Z(R0 ) = CR (R0 ) is in fact Z2 -simple.
Remark 6.12. Proposition 6.1 shows that in a simple strongly graded ring R
where Re is commutative, we automatically have that Re = Z(Re ) is G-simple. In
Proposition 6.5 we saw that for a strongly graded ring R where Re is commutative,
G-simplicity of CR (Re ) implies simplicity of R. After seeing Example 6.11 it is
tempting to think that the converse is also true (even for noncommutative Re ), i.e.
simplicity of R gives rise to G-simple subrings. The natural questions are:
(1) If R is strongly group graded and simple, is CR (Re ) necessarily G-simple?
(2) If R is strongly group graded and simple, is Z(Re ) necessarily G-simple?
Recall that the center of a simple ring is a eld. Thus, if G is the trivial group,
then the answers to both questions are clearly armative. Let us therefore consider
the case when G is an arbitrary nontrivial group. Note that if R is commutative,
then it is trivial to verify that the answers to both questions are armative. As we
have already mentioned, if Re is commutative then the answer to question no. 2
is armative. Furthermore, if Re is maximal commutative, then by Theorem 6.6
we conclude that the answer to question no. 1 is also armative. The case that
remains to be investigated is that of a noncommutative ring R where Re is not
maximal commutative (we may not even assume for it to be commutative) in R.
From Example 6.10 we learnt that simplicity of a strongly graded ring R does
not immediately imply maximal commutativity of the neutral component Re . However, for skew group rings there is in fact such an implication, as the following
theorem shows.
Theorem 6.13. Let R = Re  G be a skew group ring with Re commutative.
The following two assertions are equivalent:
(i) Re is a maximal commutative subring of R and Re is G-simple.
(ii) R is a simple ring.
Proof. By Theorem 6.6, (i) implies (ii). Suppose that (ii) holds. It follows
from Theorem 3.5 that Re is maximal commutative in R and by Proposition 6.1
we conclude that Re is G-simple. This concludes the proof.

It follows from [11, Corollary 10] that the assumptions made in [3, Corollary
2.1] force the coecient ring to be maximal commutative in the skew group ring.
By the assumptions made in [3, Theorem 2.2] the same conclusion follows by [3,
Proposition 2.2] together with [11, Corollary 6]. This shows that Theorem 6.13 is
a generalization of [3, Corollary 2.1] and [3, Theorem 2.2].

172
14

JOHAN OINERT

Remark 6.14. Note that, in Theorem 6.13, the implication from (i) to (ii)
holds in much greater generality. Indeed, it holds for any strongly graded ring.
A majority of the objects studied in [19, 20, 21] satisfy the conditions of
Theorem 6.13 and hence it applies. We shall show one such example.
Example 6.15 (Skew group algebras associated to dynamical systems). Let
h : X X be a bijection on a nonempty set X, and A CX an algebra of
: Z Aut(A)
functions, such that if f A then f h A and f h1 A. Let h
(n)

be dened by hn : f
f h
for f A and n Z. We now have a Z-crossed

system (with trivial h-cocycle)


and we may dene the skew group algebra A h Z.
For more details we refer to the papers [19, 20, 21], in which this construction has
been studied thoroughly.
By Theorem 6.13 we get the following corollary, since CX is commutative.
Corollary 6.16. Following Example 6.15, let Ah Z be the skew group algebra
associated to a dynamical system (X, h). The following assertions are equivalent:
(i) A h Z is a simple algebra.
(ii) A is a maximal commutative subalgebra of A h Z and A is Z-simple.
7. Application: Z-graded algebraic crossed products associated to
topological dynamical systems
Let (X, h) be a topological dynamical system, i.e. X is a compact Hausdor
space and h : X X is a homeomorphism. The algebra of complex-valued continuous functions on X, where addition and multiplication is dened pointwise, is
denoted by C(X). Dene a map
: Z Aut(C(X)),
h

n (f ) = f h(n) ,
h

f C(X)

and let C(X) h Z be the algebraic crossed product associated to our


dynamical
system. Recall that elements of C(X) h Z are written as formal sums nZ fn un ,
where all but a nite number of fn C(X), for n Z, are nonzero. The multiplication in C(X) h Z is dened as the bilinear extension of the rule
2

n (gm ) un+m
(fn un )(gm um ) = fn h
for n, m Z and fn , gm C(X). We now dene the following sets:


x X | h(n) (x) = x , n Z
Pern (h) =

Pern (h)
Per(h) =
nZ>0

Aper(h) = X \ Per(h)
Elements of Aper(h) are referred to as aperiodic points of the topological dynamical system (X, h). By Urysohns lemma, C(X) separates points of X and
hence by [19, Corollary 3.4] we get the following.
2In ring theory literature this would be referred to as a skew group algebra, but here we adopt
the terminology used in [19, 20, 21] which comes from the C -algebra literature. Note however,
that this is not a C -crossed product, but an algebraic crossed product.

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

173
15

Lemma 7.1. The commutant of C(X) in R = C(X) h Z is given by







n
fn un  supp(fn ) Per (h), fn C(X), n Z .
CR (C(X)) =
nZ

The topological dynamical system (X, h) is said to be topologically free if and


only if Aper(h) is dense in X. Using topological properties of our (completely
regular) space X together with the remarks made in [19], in particular [19, Theorem
3.5], one can prove the following.
Lemma 7.2. C(X) is maximal commutative in C(X) h Z if and only if (X, h)
is topologically free.
If I is an ideal of C(X) then we denote

supp(f )
supp(I) =
f I

where supp(f ) = {x X | f (x) = 0} for f C(X). Note that a subset S X is


Z-invariant if and only if h(S) = S.
Lemma 7.3. C(X) is Z-simple if and only if there are no nonempty proper
h-invariant closed subsets of X.
Proof. Suppose that C(X) is not Z-simple. Then there exists some proper
nonzero ideal I  C(X) such that supp(I) = is a proper h-invariant closed subset
of X. Conversely, suppose that there exists some nonempty proper h-invariant
closed subset S  X. Let B C(X) be set of functions which vanish outside S.
Clearly B is a proper nonzero Z-invariant ideal of C(X) and hence C(X) is not
Z-simple.

Definition 7.4. A topological dynamical system (X, h) is said to be minimal
if each orbit of the dynamical system is dense in X.
Note that a topological dynamical system (X, h) is minimal if and only if there
are no nonempty proper h-invariant closed subsets of X.
Remark 7.5. If X is innite and (X, h) is minimal, then (X, h) is automatically
free and in particular topologically free. Indeed, take an arbitrary x X and
suppose that it is perodic. By minimality, the orbit of x which by periodicity is
nite, must be dense in X. This is a contradiction, since X is Hausdor, and hence
each x X is aperiodic.
Theorem 7.6. If (X, h) is a topological dynamical system with X innite, then
the following assertions are equivalent:
(i) C(X) h Z is a simple algebra.
(ii) C(X) is maximal commutative in C(X) h Z and C(X) is Z-simple.
(iii) (X, h) is a minimal dynamical system.
Proof. (i) (ii): This follows from Theorem 6.13.
(iii) (ii): Let (X, h) be minimal. By Remark 7.5 (X, h) is topologically free
and by Lemma 7.2 this implies that C(X) is maximal commutative in C(X) h Z.
Furthermore, since (X, h) is minimal there is no nonempty proper h-invariant closed
subset of X and hence by Lemma 7.3 it follows that C(X) is Z-simple.
(ii) (iii): Suppose that (X, h) is not minimal. Then there exists some nonempty
proper h-invariant closed subset of X and by Lemma 7.3 C(X) is not Z-simple. 


JOHAN OINERT

174
16

For C -crossed product algebras associated to topological dynamical systems


the analogue of the above theorem, Theorem 1.2, is well-known (see e.g. [1], [17]
or [23, Theorem 4.3.3]).
Example 7.7 (Finite single orbit dynamical systems). Suppose that X =
{x, h(x), h(2) (x), . . . , h(p1) (x)} consists of a nite h-orbit of order p, where p
is a positive integer. One can then show that C(X) h Z
= Mp (C[t, t1 ]), i.e.
the skew group algebra associated to our dynamical system is isomorphic (as a Calgebra) to the algebra of p p-matrices over the ring of Laurent polynomials over
C. Indeed, let : C(X) h Z Mp (C[t, t1 ]) be the C-algebra morphism dened
by

f (x)
0
...
0
0

f h(x) . . .
0

(f ) = ..

..
.
.
.
.
.

.
.
.
0
0
. . . f h(p1) (x)
for f C(X), and

0 0 ... 0 t
1 0 ... 0 0

(u1 ) = 0 1 . . . 0 0 .
.. .. . .
.. ..
. .
. . .
0

0 ... 1 0

Calculating, one sees that






fn un =

nZ


n
 nZ fnp (x) t n
t
 nZ fnp+1 (h(x))
(2)
(x)) tn
nZ fnp+2 (h
..
.

nZ

f(n+1)p1 (h(p1) (x)) tn


n
...
 nZ f(n1)p+1 (x) t n
. . .  nZ f(n1)p+2 (h(x)) t
(2)
...
(x)) tn
nZ f(n1)p+3 (h
..
..
.
.

(p1)
...
f
(h
(x)) tn
nZ np

and by looking at the above matrix row by row, it is straightforward to verify that
is bijective (see [22, 24] for a similar isomorphism of C -algebras).
Clearly (X, h) is a minimal dynamical system and by Lemma 7.3 we conclude
that C(X) is Z-simple. However, each element of X is n-periodic and hence (X, h)
is not topologically free, which by Lemma 7.2 entails that C(X) is not maximal
commutative in R = C(X) h Z. The ring C[t, t1 ] is not simple (e.g. by Example
4.4) and via the isomorphism we conclude that C(X) h Z is never simple. From
extends to an action of Z on CR (C(X)).
Section 2.1 it is clear that the action h
Finally, by Proposition 6.5, we conclude that the commutant of C(X) is never
Z-simple for our nite single orbit dynamical system.
References
[1] R. J. Archbold and J. S. Spielberg, Topologically free actions and ideals in discrete C dynamical systems, Proceedings of the Edinburgh Mathematical Society 37 (1993), 119124.
[2] S. Caenepeel and F. Van Oystaeyen, Brauer groups and the cohomology of graded rings,
Monographs and Textbooks in Pure and Applied Mathematics, 121, Marcel Dekker, Inc.,
New York, 1988.

SIMPLE GROUP GRADED RINGS AND MAXIMAL COMMUTATIVITY

175
17

[3] K. Crow, Simple regular skew group rings, J. Algebra Appl. 4 (2005), no. 2, 127137.
[4] E. C. Dade, The equivalence of various generalizations of group rings and modules, Math.
Z. 181 (1982), no. 3, 335344.
[5] V. Drensky and E. Formanek, Polynomial identity rings, viii+200 pp., Advanced Courses in
Mathematics. CRM Barcelona, Birkh
auser Verlag, Basel, 2004.
[6] I. N. Herstein, Noncommutative Rings, xi+199 pp., The Carus Mathematical Monographs,
no. 15, The Mathematical Association of America, New York, 1968.
[7] G. Karpilovsky, The Algebraic Structure of Crossed Products, x+348 pp., North-Holland
Mathematics Studies, 142, Notas de Matem
atica, 118, North-Holland, Amsterdam, 1987.
[8] T.Y. Lam, A rst course in noncommutative rings, xvi+397 pp., Graduate Texts in Mathematics, 131. Springer-Verlag, New York, 1991.
[9] C. N
ast
asescu and F. Van Oystaeyen, Methods of Graded Rings, Lecture Notes in Mathematics, 1836, Springer-Verlag, Berlin, 2004.
[10] E. Nauwelaerts and F. Van Oystaeyen, Introducing Crystalline Graded Algebras, Algebr.
Represent. Theory 11 (2008), No. 2, 133148.

[11] J. Oinert
and S. D. Silvestrov, Commutativity and Ideals in Algebraic Crossed Products, J.
Gen. Lie. T. Appl. 2 (2008), no. 4, 287302.
, On a Correspondence Between Ideals and Commutativity in Algebraic Crossed Prod[12]
ucts, J. Gen. Lie. T. Appl. 2 (2008), No. 3, 216220.
[13]
, Crossed Product-Like and Pre-Crystalline Graded Rings, Chapter 24 in S. Silvestrov,
E. Paal, V. Abramov, A. Stolin (Eds.), Generalized Lie theory in Mathematics, Physics and
Beyond, pp. 281296, Springer-Verlag, Berlin, Heidelberg, 2009.
, Commutativity and Ideals in Pre-Crystalline Graded Rings. To appear in Acta Appl.
[14]
Math. (2009)

[15] J. Oinert,
S. Silvestrov, T. Theohari-Apostolidi and H. Vavatsoulas, Commutativity and
Ideals in Strongly Graded Rings. To appear in Acta Appl. Math. (2009)
[16] D. Olesen and G. K. Pedersen, Applications of the Connes spectrum to C -dynamical systems
II., J. Funct. Anal. 36 (1980), no. 1, 1832.
[17] S. C. Power, Simplicity of C -algebras of minimal dynamical systems, J. London Math. Soc.
(2) 18 (1978), no. 3, 534538.
[18] L. Rowen, Some Results on the Center of a Ring with Polynomial Identity, Bull. Amer.
Math. Soc. 79 (1973), no. 1, 219223.
[19] C. Svensson, S. Silvestrov and M. de Jeu, Dynamical Systems and Commutants in Crossed
Products, Internat. J. Math. 18 (2007), no. 4, 455471.
, Connections Between Dynamical Systems and Crossed Products of Banach Algebras
[20]
by Z, in Methods of Spectral Analysis in Mathematical Physics, Conference on Operator Theory, Analysis and Mathematical Physics (OTAMP) 2006, Lund, Sweden, Operator Theory:
Advances and Applications, Vol. 186, Janas, J., Kurasov, P., Laptev, A., Naboko, S. and
Stolz, G. (Eds.), pp. 391401, Birkh
auser, 2009.
, Dynamical systems associated to crossed products, Preprints in Mathematical
[21]
Sciences 2007:22, LUFTMA-5088-2007; Leiden Mathematical Institute report 2007-30;
arxiv:0707.1881. To appear in Acta Appl. Math. (2009)
[22] C. Svensson and J. Tomiyama, On the commutant of C(X) in C -crossed products by Z and
their representations, J. Funct. Anal. 256 (2009), 23672386.
[23] J. Tomiyama, Invitation to C -algebras and topological dynamics, World Scientic Advanced
Series in Dynamical Systems, 3. World Scientic Publishing Co., Singapore, 1987.
, C -algebras and topological dynamical systems, Rev. Math. Phys. 8 (1996), 741760.
[24]
[25] F. Van Oystaeyen, On Cliord Systems and Generalized Crossed Products, J. Algebra 87
(1984), 396415.
Centre for Mathematical Sciences, Lund University, P.O. Box 118, SE-22100 Lund,
Sweden
E-mail address: Johan.Oinert@math.lth.se

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALGEBRAS


WITH A FINITE WATATANI INDEX
HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

Abstract. We introduce notions of the Rohlin property and the approximate


representability for inclusions of unital C -algebras. We investigate a dual
relation between the Rohlin property and the approximate representability.
We prove that a number of classes of unital C -algebras are closed under
inclusions with the Rohlin property, including:
AF algebras, AI algebras, AT algebras, and related classes characterized
by direct limit decomposition using semiprojective building blocks.
C -algebras with stable rank one.
C -algebras with real rank zero.

1. Introduction
A. Kishimoto [11], R. Herman and V. Jones [3], [4] investigated a class of nite
group actions with what we presently call the Rohlin property. After that a number
of results for group actions of C -algebras with the Rohlin property were found in
the literature (see [12], [13] , [14], [7], [17]).
In [7], M. Izumi introduced the Rohlin property and the approximate representability for nite group actions. He proved that an action of a nite abelian
group has the Rohlin property if and only if its dual action is approximately representable. We extend the notions of the Rohlin property and the approximate
representability for inclusions of unital C -algebras with nite Watatani index in
the sense of [21]. We investigate a dual relation between the Rohlin property and
the approximate representability. We prove that an inclusion has the Rohlin property if and only if its dual inclusion is approximately representable. It contains that
an action of a nite group has the Rohlin property if and only if its dual action is
approximately representable as a nite dimensional C -Hopf algebra action. Note
that the dual action of an action of a non-commutative nite group is not an action of some group though it is an action of some nite dimensional commutative
C -Hopf algebra.
In [16], H. Osaka and N. C. Phillips proved that crossed products by nite group
actions with the Rohlin property preserve various properties of C -algebras. Since
an action of a nite group with the Rohlin property is an outer action by [17,
Remark 1.4 and Lemma 1.5] and the crossed product algebra A  G and the xed
point algebra A by an outer action of a nite group G are Morita equivalent,
we can immediately see that xed point algebras by nite group actions with the
Rohlin property also preserve various properties of C -algebras by [16]. We extend
2000 Mathematics Subject Classication. Primary 46L55; Secandary 46L35.
Research of the rst author partially supported by the JSPS grant for Scientic Research
No.20540220.
1

177

c
2009
American Mathematical Society

178
2

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

their results and prove that a number of classes of unital C -algebras are closed
under inclusions with the Rohlin property, including:
AF algebras, AI algebras, AT algebras, and related classes characterized by
direct limit decomposition using semiprojective building blocks.
C -algebras with stable rank one.
C -algebras with real rank zero.
This paper is organized as follows: In Section 2 we collect basic facts on Watatani
index theory for C*-algebras and nite group actions on C*-algebras with the Rohlin
property.
In section 3 we introduce the Rohlin property and the approximately representability for conditional expectations and deduce basic properties of conditional
expectations possessing it. We prove that if an inclusion has a conditional expectation with the Rohlin property, then it is the unique conditional expectation of its
inclusion. So the property that a conditional expectation has the Rohlin property
is actually a property of its inclusion.
In section 4 we introduce a notion of the Rohlin property for actions of nite
dimensional C -Hopf algebras. We construct examples of inclusions with the Rohlin
property which do not come from nite group actions with the Rohlin property
using nite dimensional C -Hopf algebras actions with the Rohlin property.
In section 5 we prove that inclusions with the Rohlin property preserve various
properties of C -algebras which generalize results of [16].
2. Preliminaries
In this section we collect notations and basic facts which will be used in this
paper.
2.1. Index theory for C*-algebras.
2.1.1. Watatani index for C -algebras. We introduce an index in terms of a quasibasis following Watatani [21].
Denition 2.1. Let A P be an inclusion of unital C*-algebras with a conditional
expectation E from A onto P .
(1) A quasi-basis for E is a nite set {(ui , vi )}ni=1 A A such that for every
a A,
n
n


a=
ui E (vi a) =
E (aui ) vi .
i=1

i=1

(2) When {(ui , vi )}ni=1 is a quasi-basis for E, we dene IndexE by


IndexE =

n


ui vi .

i=1

When there is no quasi-basis, we write IndexE = . IndexE is called the


Watatani index of E.
Remark 2.2. We give several remarks about the above denitions.
(1) IndexE does not depend on the choice of the quasi-basis in the above formula, and it is a central element of A [21, Proposition 1.2.8].

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

179
3

(2) Once we know that there exists a quasi-basis, we can choose one of the form
{(wi , wi )}m
i=1 , which shows that IndexE is a positive element [21, Lemma
2.1.6].
(3) By the above statements, if A is a simple C -algebra, then IndexE is a
positive scalar.
(4) Let {(ui , vi )}ni=1 be a quasi-basis for E. If A acts on a Hilbert space H
faithfully, then
can dene the map E 1 from P  B(H) to A B(H)
we
n
1
by E (x) = i=1 ui xvi for x in P  B(H). In fact, for any x P  B(H)
and a A
n

ui xvi a
E 1 (x)a =
=

i=1
n

i,j=1
n


ui xE(vi auj )vj


ui E(vi auj )xvj

i,j=1
n


auj xvj = aE 1 (x).

j=1

(5) If IndexE < , then E is faithful, that is, E(x x) = 0 implies x = 0 for
x A.
2.1.2. C -basic construction. In this subsection, we recall Watatanis notion of the
C -basic construction.
Let E : A P be a faithful conditional expectation. Then AP (= A) is a preHilbert module over P with a P -valued inner product
x, yP = E(x y), x, y AP .
We denote by EE and E the Hilbert P -module completion of A by the norm
1
x P = x, xP 2 for x in A and the natural inclusion map from A into EE . Then
EE is a Hilbert C -module over P . Since E is faithful, the inclusion map E from A
to EE is injective. Let LP (EE ) be the set of all (right) P -module homomorphisms
T : EE EE with an adjoint right P -module homomorphism T : EE EE such
that
T ,  = , T  , EE .
Then LP (EE ) is a C -algebra with the operator norm T = sup{ T : = 1}.
There is an injective -homomorphism : A LP (EE ) dened by
(a)E (x) = E (ax)
for x AP and a A, so that A can be viewed as a C -subalgebra of LP (EE ).
Note that the map eP : AP AP dened by
eP E (x) = E (E(x)), x AP
is bounded and thus it can be extended to a bounded linear operator, denoted by
eP again, on EE . Then eP LP (EE ) and eP = e2P = eP ; that is, eP is a projection
in LP (EE ). A projection eP is called the Jones projection of E.

180
4

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

The (reduced) C -basic construction is a C -subalgebra of LP (EE ) dened to be


Cr A, eP  = span{(x)eP (y) LP (EE ) : x, y A }



Remark 2.3. Watatani proved the following in [21]:


(1) IndexE is nite if and only if Cr A, eP  has the identity (equivalently
Cr A, eP  = LP (EE )) and there exists a constant c > 0 such that E(x x)
cx x for x A, i.e., x 2P c x 2 for x in A by [21, Proposition 2.1.5].
Since x x P for x in A, if IndexE is nite, then EE = A.
(2) If IndexE is nite, then each element z in Cr A, eP  has a form
z=

n


(xi )eP (yi )

i=1

for some xi and yi in A.

(3) Let Cmax


A, eP  be the unreduce C -basic construction dened in Denition 2.2.5 of [21], which has the certain universality (cf.(5)). If IndexE is

nite, then there is an isomorphism from Cr A, eP  onto Cmax


A, eP  ([21,

Proposition 2.2.9]). Therefore we can identify Cr A, eP  with Cmax


A, eP .

So we call it the C -basic construction and denote it by C A, eP . Moreover we identify (A) with A in C A, ep (= Cr A, eP ) and we denote
C A, ep  = {

n


xi eP yi : xi , yi A, n N}.

i=1

(4) If IndexE is nite, then IndexE is a central invertible element of A and


from C A, eP  onto A such
there is the dual conditional expectation E
that
P y) = (IndexE)1 xy for x, y A
E(xe
has a nite index and faithfulness.
by [21, Proposition 2.3.2]. Moreover, E
(5) Suppose that IndexE is nite and A acts on a Hilbert space H faithfully
and e is a projection on H such that eae = E(a)e for a A. If a map
P x xe B(H) is injective, then there exists an isomorphism from
the norm closure of a linear span of AeA to C A, eP  such that (e) = eP
and (a) = a for a A [21, Proposition 2.2.11].
The next lemma is very useful.
Lemma 2.4. Let A P be an inclusion of unital C*-algebras with a conditional
expectation E from A onto P . If IndexE is nite, then for each element z in the C basic construction C A, eP , there exists an element a in A such that zeP = aeP .
In fact,
P )eP .
zeP = (IndexE)E(ze
Proof. For each z C A,
such that z =
 eP  there areelements xi , yi A 

n
i=1 xi eP yi . Then zeP =
i xi eP yi eP =  i xi E(yi )eP , i.e., a =
i xi E(yi ) A.
P ) = (IndexE)1
On the other hand, E(ze
x
E(y
)
by
Remark
2.3
(3) and hence
i
i
i
P ).
we have a = (IndexE)E(xe
Denition 2.5. Let A P be a inclusion of unital C -algebras with a nite index
and let Q be a C -subalgebra of P . Q is said to be a tunnel construction for the
inclusion A P if A is the basic construction for the inclusion P Q.

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

181
5

In the factor case, a tunnel construction always exists for any index nite subfactor [10]. But an inclusion of C -algebras does not have a tunnel construction
in general. We shall give a necessary and sucient condition for an existence of a
tunnel construction in the next proposition.
Proposition 2.6. Let A P be an inclusion of unital C -algebras and E a
conditional expectation from A onto P with IndexE < . If there is a projection
e A such that E(e) = (IndexE)1 , then we have
eP e = Qe,

Q = P {e}

In particular, if e is a full projection, i.e., there are elements xi , yi of A such that



n
i=1 xi eyi = 1, then Q is a tunnel construction for A P such that e is the Jones
projection for P Q.
1

Proof. Let eP be the Jones projection for E. Then eP eeP = E(e)ep = (IndexE) eP .
We shall prove that eeP e = (IndexE)1 e. Put f = (IndexE)eeP e. Then it is easy
be the dual conditional expectation
to see that f is a projection and f e. Let E
f ) = 0. And hence we have f = e by the faithfulness of E
and
of E. Then E(e
eeP e = (IndexE)1 e.
Let F be a linear map on P dened by F (x) = (IndexE)E(exe). We shall prove
exe = F (x)e = eF (x) for x P . Since eeP e = (IndexE)1 e, we have for x P
eF (x)eP

= e(IndexE)E(exe)eP
= e(IndexE)eP (exe)eP
= (IndexE)(eeP e)xeeP = exeeP

and hence eF (x)eP = exeeP . Then using Remark 2.3 (3) we have
eF (x) =
=

(IndexE)E(eF
(x)eP )

(IndexE)E(exeeP ) = exe.

Moreover F (x)e = (eF (x )) = (ex e) = exe, and hence exe = F (x)e = eF (x).
Let Q be the C -subalgebra of P dened by Q = P {e} . We saw F (x) Q
for any x P . Conversely if x is an element of Q, then F (x) = (IndexE)E(exe) =
(IndexE)E(e)x = x. Therefore F is a conditional expectation from P onto Q and
eP e = Qe. If xe = 0 for some x Q, then x = (IndexE)E(e)x = (IndexE)E(xe) =
0 and hence a map Q x xe Qe is injective. By Remark 2.3 (5), the norm
closure of the linear span {xey : x, y P } is the basic construction for P Q. For
any a A
P ae)
(IndexE)E(e
P aeeP e)
= (IndexE)2 E(e

= (IndexE)2 E(E(ae)e
P e) = (IndexE)E(ae)e,

ae =

and hence Ae = P e. Similarly, we have eA = eP . If e is a full projection, then


A

= the linear span of {xey : x, y A}


= the linear span of {xey : x, y P }.

So A is the basic construction for the inclusion P Q with the Jones projection e.
It means that Q is a tunnel construction for A P .

182
6

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

2.2. Finite group actions on C -algebras with the Rohlin property. For a
C -algebra A, we set
c0 (A) = {(an ) l (N, A) : lim an = 0}
n

l (N, A)/c0 (A).

We identify A with the C -subalgebra of A consisting of the equivalence classes


of constant sequences and set
A = A A .
For an automorphism Aut(A), we denote by and the automorphisms
of A and A induced by , respectively.
Izumi dened the Rohlin property for a nite group action in [7, Denition 3.1]
as follows:
Denition 2.7. Let be an action of a nite group G on a unital C -algebra A.
is said to have the Rohlin property if there exists a partition of unity {eg }gG A
consisting of projections satisfying
(g ) (eh ) = egh

for g, h G.

We call {eg }gG Rohlin projections.


The next lemma is essentially contained in [17, Lemma 1.5]. But we give a short
proof of it for the self-contained.
Lemma 2.8. Let be an action of a nite group G on a unital C -algebra A. If
has the Rohlin property, then is an outer action.
Proof. Suppose that g is not the unit element of G. If g is an inner automorphism
Adu for some unitary element u in A, then (g ) (eh ) = ueh u = eh for h in G
since eh A = A A . Hence if g has the Rohlin property, then g is outer.
Let A P be an inclusion of unital C -algebras. For a conditional expectation E
from A onto P , we denote by E , the natural conditional expectation from A onto
P induced by E. If E has a nite index with a quasi-basis {(ui , vi )}ni=1 , then E
also has a nite index with a quasi-basis {(ui , vi )}ni=1 and Index(E ) = IndexE.
Proposition 2.9. Let be an action of a nite group G on a unital C -algebra A
and E the canonical conditional expectation from A onto the xed point algebra
P = A dened by
1 
E(x) =
g (x) for x A,
#G
gG

where #G is the order of G. Then has the Rohlin property if and only if there
1
1, where E is the conditional
is a projection e A such that E (e) = #G

expectation from A onto P induced by E.


Proof. Suppose that has the Rohlin property with a partition of unity {eg }gG
A consisting of projections satisfying
(g ) (eh ) = egh
Then
E (e1 ) =

for g, h G.

1 
1 
1
(g ) (e1 ) =
eg =
1,
#G
#G
#G
gG

gG

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

183
7

where e1 is the projection in the partition of unity {eg }gG which corresponds to
the unit element of G.
1
Conversely, suppose that there is a projection e A such that E (e) = #G
1.
Dene eg = (g ) (e) A for g G. Then

eg = #GE(e) = 1,
gG

i.e., {eg }gG A is a partition of unity. It is obvious that (g ) (eh ) = egh for
g, h G. Hence has the Rohlin property.
3. Conditional expectations of unital C -algebras with the Rohlin
property
Denition 3.1. A conditional expectation E of a unital C -algebra A with a
nite index is said to have the Rohlin property if there exists a projection e A
satisfying
E (e) = (IndexE)1 1
and a map A x xe is injective. We call e a Rohlin projection.
Proposition 3.2. Let G be a nite group, an action of G on a unital simple
C -algebra A, and E the canonical conditional expectation from A onto the xed
point algebra A . Then has the Rohlin property if and only if E has the Rohlin
property.
Proof. Suppose that has the Rohlin property. By [9, Theorem 4.1 and Remark
4.6], is saturated and IndexE is nite. The simplicity of A implies that the map
A x xe is injective. So we have that E has the Rohlin property by Proposition
2.9. Conversely, if E has the Rohlin property, then has the Rohlin property by
Proposition 2.9.
Denition 3.3. A conditional expectation E from a unital C -algebra A onto
P with a nite index is said to be approximately representable if there exists a
projection e P satisfying for any x A
exe = E(x)e
and a map P x xe is injective.
Proposition 3.4. Let A P be an inclusion of unital C -algebras and E a
conditional expectation from A onto P with a nite index. Let B be the basic
the dual conditional expectation of E from B onto
construction for A P and E
A. Then
is approximately representable;
(1) E has the Rohlin property if and only if E
has the Rohlin property.
(2) E is approximately representable if and only if E
Proof. (1): Let eP be the Jones projection for the inclusion A P . Suppose that
E has the Rohlin property with a Rohlin projection e A . Then
1

eP eeP = E (e)eP = (IndexE)

eP

in B .

Let f be an element in B dened by f = (IndexE)eeP e. It is easy to see that f


(e f ) = e (IndexE)eE(e
P )e = e e = 0,
is a projection and f e. Since E
1

we have f = e by the faithfulness of E , and hence eeP e = (IndexE) e. Since

184
8

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

is determined by E(xe
P y) = (IndexE)1 xy for x and y in A, we have for any
E
element z in B

eze = E(z)e,
and the map A x xe Ae for x A is injective by the denition of the Rohlin
is approximately representable.
property. Therefore E
is approximately representable with a projection
Conversely, suppose that E

for any z in B and a map A x xe Ae


e A satisfying that eze = E(z)e
for x A is injective. Then we have
P )e = (IndexE)1 e.
eeP e = E(e
Dene an element w in B by w = (IndexE)eP eeP . Then
(eP w)e = e(eP w)e = eeP e (IndexE)eeP (eeP e) = 0.
E
and the injectivity of the map x xe for x A, we have
By the faithfulness of E
w = eP , i.e., eP eeP = (IndexE)1 eP . Since E (e)eP = eP eeP = (IndexE)1 eP ,
1
we have E (e) = (IndexE) . Hence E has the Rohlin property.
(2): Suppose that E is approximately representable with a projection e P
satisfying that exe = E(x)e for any x in A. Let {(ui , vi )}ni=1 be a quasi-basis for
E. Dene an element f in B by
f=

n




ui eeP vi = (E )1 (eeP ) .

i=1

It is easy to see that f is a projection and commutes with elements of A by Remark 2.2 (4). f also commutes with eP . In fact, since eeP = eP e, we have
f eP

=
=

n

i=1
n


ui eeP vi eP
ui eE(vi )eP =

i=1

n


ui E(vi )eeP = eeP

i=1

and
eP f

=
=

n

i=1
n


eP ui eeP vi
eP ui eP evi =

i=1

n


eeP E(ui )vi = eeP .

i=1

Therefore f is an element in B  B = B since B is generated by A and eP .


By Remark 2.3 (5), there exists an isomorphim from B onto C A, e such that
(eP ) = e and (a) = a for a A. So we have
 n

n


ui (eP )vi =
ui eP vi = 1
(E )1 (e) =
i=1

i=1

and
(f ) =
E

n


P )vi
ui eE(e

i=1

= (IndexE)1 (E )1 (e) = (IndexE)1 1.

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

185
9

Suppose that xf = 0 for some x B. Let C A, eP , e be a C -algebra generated


by A, eP , and e in B . Let be an automorphism on C A, eP , e dened by
(a) = a for a A, (eP ) = e, and (e) = eP . Since (f ) = f and (x) = (x) for
((x)f ) =
x B, we have 0 = (xf ) = (x)f . Since (x) A , we have 0 = E
1
(f ) = (IndexE) (x). Hence x = 0, i.e., the map B x xf Bf is
(x)E
has the Rohlin property with a projection f B .
an injective map. Therefore E
has the Rohlin property with a Rohlin projection
Conversely, suppose that E
(f eP ). Using the fact
f B . Dene an element e in A by e = (IndexE)E

that f eP = eP f = (IndexE)E (f eP )eP by Lemma 2.4, we have


e2

(f eP )
(f eP )E
= (IndexE)2 E
(f eP )eP f )
((IndexE)E
= (IndexE)E
(f eP ) = e,
= (IndexE)E

and hence e is a projection in A . There exists an element x in P such that


f eP = xeP since
f eP = eP f eP eP B eP = (eP BeP )

= (P eP ) = P eP .

Then we have
(f eP ) = (IndexE)E
(xeP ) = (IndexE)xE(e
P ) = x.
e = (IndexE)E
Hence e P and f eP = eeP . Since e commutes with any element in P , we have
e P . For any a A
eae

(f eP )
(f eP )aE
= (IndexE)2 E
(f eP )af eP )
(E
= (IndexE)2 E
(f eP a)eP f )
((IndexE)E
= (IndexE)E
(f eP aeP f ) (by Lemma 2.4)
= (IndexE)E
(f E(a)eP f )
= (IndexE)E
(f eP ) = E(a)e.
= E(a)(IndexE)E

Suppose that xe = 0 for some x P . Since eeP = f ep , xf eP = (xeP )f = 0


and xeP = 0 by the injectivity of the map B y yf Bf . Hence x = 0 by
the injectivity of the map P x xeP P eP . Therefore E is approximately
representable.
Proposition 3.5. Let A P be an inclusion of unital C*-algebras and E a conditional expectation from A onto P with a nite index. If E is approximately
representable, then P  A P .
Proof. Let e be a projection in P such that exe = E(x)e for x in A. If x is an
element in P  A, then x also commutes with e. Hence xe = exe = E(x)e. Since
a map y P ye P e is injective, there is an isomorphism from the basic
construction C A, eP  onto C A, e by Remark 2.3 (5). Then
xeP
Therefore x = E(x) P .

= x(e) = (xe) = (exe)


= (E(x)e) = E(x)(e) = E(x)eP .

186
10

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

Remark 3.6. If a conditional expectation E : A P is approximately representable, then E is the unique conditional expectation from A onto P by [21, Corollary 1.4.3]. In other words, the property that E is approximately representable is
actually a property of the inclusion A P . By Proposition 3.4, the property that
E has the Rohlin property is actually a property of the inclusion A P . So we
call A P an inclusion with the Rohlin property.
When an inclusion A P has a nite index, if P is simple, then A is a nite
direct sum of simple closed two-sided ideals by [6, Theorem 3.3]. Therefore the
above propositions immediately implies the following:
Corollary 3.7. Let A P be an inclusion of unital C*-algebras and E a conditional expectation from A onto P with a nite index. If E is approximately representable and P is simple, then A P is an irreducible inclusion, i.e., P  A
=C
and A is simple.
Proof. By Proposition 3.5, we have P  A P  P
= C by the simplicity of P .
On the other hand, A A P  A
= C and hence A is simple by [6, Theorem
3.3].
Corollary 3.8. Let E be a conditional expectation from a unital C*-algebra A
onto P with a nite index. If E has the Rohlin property and A is simple, then
A P is an irreducible inclusion and P is simple.
of E is approximately
Proof. By Proposition 3.4, the dual conditional expectation E
representable. Therefore the inclusion C A, eP  A is irreducible and C A, eP 
is simple by Corollary 3.7. Since A C A, eP  is isomorphic to P  A as linear
spaces (see the proof of Proposition 3.11 of [8]), A P is irreducible. The simplicity
of P comes from [21, Corollary 2.2.14].
4. Examples of finite dimensional C -Hopf algebra actions with the
Rohlin property
In this section we shall construct examples of inclusions of C -algebras with the
Rohlin property. It contains examples which do not come from nite group actions
with the Rohlin property.
Following [20], we state the denition of a nite dimensional C -Hopf algebra and
its basic properties. Throughout this section, H and H 0 denotes a nite dimensional
C -Hopf algebra and its dual C -Hopf algebra, respectively.
Denition 4.1. We say that a nite dimensional C -algebra H is a C -Hopf algebera if H has the following properties.
(1) There exist linear maps;
(a) comultiplication : H H H,
(b) counit  : H C,
(c) antipode S : H H,
and  are C -algebra homomorphisms and S is a *- preserving antimultiplicative involution. We have (1) = 1 1, (1) = 1 and
S(1) = 1, where 1 is the unit element in H.
(2) The following identities hold;
(a) ( id) = (id ) ,

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

187
11

(b) ( id) = (id ) = id, where C H and H C are identied


with H,
(c) (m (S id))((h)) = (h) = (m (id S))((h)) for any h H,
where m : H H H denotes the multiplication.
We use Sweedlers notation ([19]) as (h) = h(1) h(2) for h H which suppresses a possible summation when we write the comultiplications.
Let A be a unital C -algebra and H a nite dimensional C -Hopf algebra.
Denition 4.2. By an action of H on A, we mean a bilinear map (h, x) h x of
H A to A such that for h H and x, y A
(1)
(2)
(3)
(4)
(5)

h xy = (h(1) x)(h(2) y),


h 1 = (h)1,
1 x = x,
(h x) = S(h ) x .
h (l x) = hl x for h, l H, x A.

Let AH be the xed point algebra dened by


AH = {x A : h x = (h)x, h H}
Denition 4.3. If a map (h, x) h x of H A to A is an action of a nite
dimensional C -Hopf algebra H on a unital C -algebra A, then we dene the crossed
product A  H in the following way.
A  H is just A H as a vector space but its multiplication and its -operation
are dened as follows.
(x  h)(y  k)
(x  h)

= xh(1) y  h(2) k
=

h(1) x  h(2)

for x, y A and h, k H.
We generalize group actions with the Rohlin property in Denition 2.7 and dene
C -Hopf algebra actions with the Rohlin property as follows:
Let (h, x) h x of H A to A be an action of nite dimensional C -Hopf
algebra. We denote by h {(xn ) + c0 (A)} the natural action of H on A induced
by h x, i.e.,
h {(xn ) + c0 (A)} = (h xn ) + c0 (A).
Denition 4.4. Let (h, x) hx of H A to A be an action of a nite dimensional
C -Hopf algebra. The action (h, x) h x is said to have the Rohlin property if
there exists a unital injective C -homomorphism of H 0 into A satisfying


for h H and H 0 .
h () = (2) (h) (1)
We call a Rohlin homomorphism.
Remark 4.5. Let be an action of a nite group G on a unital C -algebra A. Let
H be a group C -algebra C (G) and H 0 a commutative C -algebra C(G). Then
we can consider H and H 0 as Hopf C -algebras and naturally extend the group
action to the Hopf action by g x = g (x) for g G and x A. Suppose that has
the Rohlin property with Rohlin projections {eg }gG as a group action. Dene a

188
12

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

unital C -homomorphism from H 0 to A by (g ) = eg1 for each g G. Since



(h1 ) = tG h1 t t1 , we have

h1 (2) (g)(h1 (1) ) =
t1 , g(h1 t )
tG

(h1 g1 ) = egh .

And hence g has the Rohlin property as a C -Hopf algebra action.


Conversely, suppose that g has the Rohlin property with a Rohlin homomorphism as a C -Hopf algebra action. Dene a partition of unity {eg }gG A
by eg = (g1 ). Then {eg }gG gives Rohlin projections.
Proposition 4.6. Let (h, x) h x of H A to A be an action of a nite
dimensional C -Hopf algebra H on a unital C -algebra A with the Rohlin property.
If A is a simple unital C -algebra, then the action (h, x) h x is saturated in the
sense of Denition 4.2 of [20].
Proof. Let H 0 be the dual C -Hopf algebra and a Rohlin homomorphism from
H 0 into A . Let P be the xed pointed algebra AH and B the crossed product
A  H.
There exists a minimal and central projection e H, called the distinguished
projection, such that he = (h)e for any h H. Then we have 1  e B is the
Jones projection for A P . We shall prove that 1  e is a full projection in B.
Let H 0 be the unique faithful normalized trace on H, called the Haar
trace, such that (h(1) )h(2) = (h)1 = h(1) (h(2) ) for any h H and let f be the
projection in A dened by f = ( ). Then for h H
f (1  h)f

= ( )(1  h)( )





= ( ) 1  h(1) ( ) 1  S(h(2) 1  h(3)



= ( ) h
1  h(2)
(1) ( )

 


= (2) h(1) (1) 1  h(2)

 

 

= (2) h(1) 0 (1)
1  h(2)




= ( ) 1  h(1) h(2)
=

(h) ( ) = (h) f.

Since f commutes with elements in A, we have f xf = E1 (x)f for x B, where E1


is the canonical conditional expectation from B onto A dened by E1 (ah) = (h)a
for a A and h H. By the simplicity of A, the map A x xf is injective and
hence f is the Jones projection for E1 . Therefore there are b1 , b2 , bl in B such
l
that i=1 bi f bi = 1. So we have
l


bi f (1  e)f bi = (e) =

i=1

1
,
n

where n is the dimension of H. If fk , k = 1, 2, are projections in A such that


f = (fk ) + c0 (A), then
lim

l

i=1

nbi fk (1  e)fk bi = 1.

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

189
13

Hence (1  e) is full in B. It implies that h is a saturated action by Theorem 4.3


of [20].
Corollary 4.7. Let (h, x) hx of H A to A be an action of a nite dimensional
C -Hopf algebra H on a unital C -algebra A with the Rohlin property and P the
xed point algebra AH . If A is simple, then the inclusion A P has a nite index
and the Rohlin property.
Proof. Let e be the distinguished projection in H. Then the canonical conditional
expectation E of A onto P is dened by E(x) = e x for x A. By the previous
proposition and Proposition 4.5 of [20], we have IndexE = dim(H). Let be the
Haar trace in H 0 and f the projection in A dened by f = ( ) as in the proof
of the previous proposition. Then we have
1
e f = e ( ) = (2) (e)((1) ) = (e) = ,
n
where n is the dimension of H. Hence we have E (f ) = n1 . So E has the Rohlin
property.
Example 4.8. For n N, we denote by Mn the n by n matrix algebra over C and
by Mn the uniformly hypernite (UHF) algebra of type n . Let H be a nite
dimensional C -Hopf algebra with dim(H) = n and H 0 the dual C -Hopf algebra.
We dene an action of H on H 0 by h = (2) (h)(1) . Then we identify H 0  H
with Mn . Let {(Ai , ij ) : i, j N, i j} be an inductive system, where the Ai are
Mni = Mn Mn Mn , and ij is the usual injective -homomorphism from



i times

Ai to Aj with ik = jk ij for i j k dened by ij (x) = x 1 1

1 .

(ji) times

We dene an action of H on Ak by

 k
k


(k)
xi =
h(i) xi
h
i=1

for h H and xi Mn ,

i=1

where h x = (1  h(1) )x(1  S(h(2) )) for x Mn and h H. Then for x =


k
i=1 xi Ak and l k we have

 k



h(l) kl (x) = h(l)
1
xi 1

i=1
(lk) times

 k

=
h(i) xi (h(k+1) ) (h(l) )

=

i=1
k

i=1


h(i) xi



1

1 = kl h(k) x

for h H.

(lk) times

So we can introduce an action of H on A = lim(Ai , i ) (= Mn ) so that

(i)
h i (x) = i (h x) for x Ai ,
where i is the inclusion map from Ai into A such that j (ij (x)) = i (x) for
x Ai and i j. Then h has the Rohlin property. To this end, we construct a

190
14

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

Rohlin homorphism with respect to h as follows: For H 0 , we set i = i ()


and () = (i ) + c0 (A) A , where i is the inclusion map from Mn into the i
th tensor factor of A = Mn dened by i (x) = i (1 1
1 x) for x Mn .

(i1) times

Then becomes a Rohlin homomorphism. In fact, for h H and H 0


h k

h k (1 1

1 )

(k1) times



= k h(k) (1 1 1 )


= k (h(1) ) (h(2) ) (h(k1) ) h(k)
= k (1 1 1 h )


= (2) (h)k 1 1 1 (1) = (2) (h)(1) k .
Therefore we have
h () = h ((k ) + c0 (A))
= (h k ) + c0 (A)




= (2) (h)(1) k + c0 (A) = (2) (h) (1) .
Remark 4.9. Let A be an simple unital C -algebra and P the xed point algebra
AH by an action of nite dimensional C -Hopf algebra H with the Rohlin property
and
P A B B1
the tower of iterating basic construction. Then the action is saturated by Proposition 4.6, the inclusion A P is irreducible and P is simple by Corollary 3.8.
Since the action is saturated, the crossed product A  H is isomorphic to the basic
construction B for A P and the index of the natural conditional expectation E
form A onto P equals dimC (H) (cf. [20]). For any h H, the element 1  h is in
the relative commutant P  (A  H). So we can see
(P  (A  H)) = 1  H =
H.
(P  B) =
The dual action of H 0 on A  H is also saturated since (A  H)  H 0 is simple (cf.
[9, Theorem 4.1 and Remark 4.6]). By the same argument as the above, we have
A B1 is isomorphic to H 0 . Therefore if an inclusion A P comes from a nite
group action with the Rohlin property, then A B1 is a commutative algebra. So,
if H is not cocommutative, then the inclusion A P does not come from a nite
group action.
5. Inclusions of C -algebras with the Rohlin property
In [16], Osaka and Phillips proved that crossed products by nite group actions
with the Rohlin property preserve various properties of C -algebras. In this section,
we extend their result and prove that inclusions with the Rohlin property preserve
various properties of C -algebras.
Lemma 5.1. Let A P be an inclusion of unital C*-algebras and E a conditional
expectation from A onto P with a nite index. If E has the Rohlin property with
a Rohlin projection e A , then for any x A there exists the unique element
y of P such that xe = ye.

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

191
15

Proof. Let eP be the Jones projection for the inclusion A P . By the proof of
Proposition 2.6, we have eeP e = (IndexE)1 e. Therefore for any element x in A
(eP xe)
xe = (IndexE)E
(eP xeeP e)
= (IndexE)2 E
(E (xe)eP e) = (IndexE)E (xe)e,
= (IndexE)2 E
is the dual conditional expectation for E. Put y = (IndexE)E (xe) P .
where E
Then we have xe = ye.
Suppose that ye = ze for y, z P . Then
z = (IndexE)E (e)z = (IndexE)E (ez) = (IndexE)E (ey) = y.
Therefore we obtain the uniqueness of y.
Remark 5.2. Let D be the C -subalgebra of P dened by D = {e} P . By
Proposition 2.6, we have eP e = De. Moreover, eA e = De by the above lemma.
Theorem 5.3. Let A P be an inclusion of separable unital C*-algebras and E
a conditional expectation from A onto P with a nite index. If A is an AF algebra
and E has the Rohlin property, then P is also an AF algebra.
Proof. We shall prove that for every nite set S P and every > 0, there is a nite
dimensional C -subalgebra Q of P such that every element of S is within of an
element of Q. Since A is an AF algebra and S P A, there is a nite dimensional
C -subalgebra R of A such that every element of S is within 4 of an element of

(r)
Mnr . Since e commutes with
R. Let {eij } be a system of matrix units of R
=
(r)

each element of A and a map A x xe Ae is injective, {eij e} is also a system


of matrix units of type R. Since Ae = eAe eA e, , there are elements pij of
(r)

D = {e} P such that eij e = pij e by Lemma 5.1 and Remark 5.2. By the
(r)

(k)

(r)

(r)

uniqueness of pij , {pij } is a system of matrix units of type R. For every (i, j, r), let

(pijk )
k=1 be a sequence of P such that pij = (pijk )k=1 +c0 (A). For every x S, let

(r) (r)
(r)
xR = i,j,r xij eij R, xij C such that x xR < 4 . Let x be the element

(r) (r)
of P dened by x = i,j,r xij pij . For each k N, we dene x
k as the element

(r) (r)

R
R
of P by xk = i,j,r xij pijk . Since xx = xex e = xex e = xx
by the injectivity of a map D x xe De, we have x x < 4 and hence
(r)

(r)

(r)

lim sup x x
k <
k

.
4

(r)

Since {pij } is a system of matrix units of type R, we have


(r) (s)

(r)

lim sup pijk pmnk rs jm pink = 0.


k

Choose > 0 according to [5, Lemma 1.10] for R and for /(2 dim(R)). There is
(r)
k0 N such that {pijk0 } is a set of approximate matrix units of type R within in P
and x x
k0 <

(r)

for every x in S. Then there is a set {fij } of exact matrix units



(r)
(r)
(r) (r)
of type R in P with fij pijk0 < /(2 dim(R)). Put x0 = i,j,r xij fij P .

192
16

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

Since x
k0 x0 < 2 , we have

x x0 x x
k0 + xk0 x0

+ = .

2 2

So we can choose a nite dimensional C -algebra Q generated by {fij } P .


Therefore P is an AF algebra.
(r)

Following [16], we introduce several notations to describe the local approximate


characterizations of the classes of direct limit algebras constructed using some common families of semiprojective building blocks.
Denition 5.4. Let C be a class of separable unital C -algebaras. Then C is nitely
saturated if the following closure conditions hold:
(1) If A C and B
= A, then B
 C.
(2) If A1 , A2 , . . . , An C then nk=1 Ak C.
(3) If A C and n N, then Mn (A) C.
(4) If A C and p A is a nonzero projection, then pAp C.
Moreover, the nite saturation of a class C is the smallest nitely saturated class
which contains C.
Denition 5.5. Let C be a class of separable unital C -algebras. We say that C
is exible if :
(1) For every A C, every n N, and every nonzero projection p Mn (A),
the corner pMn (A)p is semiprojective in the sense of Denition 14.1.3 of
[15], and is nitely generated.
(2) For every A C and every ideal I A, there is an increasing sequence
I0 I1 of ideals in A such that
n=0 In = I, and such that for every
n the C -algebra A/In is in the nite saturation of C.
Denition 5.6. Let C be a class of separable unital C -algebras. A unital local
C-algebra is a separable unital C -algebra such that for every nite set S A and
every > 0, there are a C -algebra B in the nite saturation of C and a unital
homomorphism : B A such that dist(a, (B)) < for all a S.
Theorem 5.7. Let C be any exible class of separable unital C -algebras. Let
A P be a nite index inclusion with the Rohlin property. If A is a unital local
C-algebra, then P is also a unital local C-algebra.
Proof. We shall prove that for every nite set S P and every > 0, there are a
C -algebra Q in the nite saturation of C and a unital homomorphism : Q P
such that S is within of an element of (Q).
Since A is a unital local C-algebra, for nite set S P A and 2 > 0, there is
a C -algebra Q in the nite saturation of C and a unital homomorphism : Q A
such that S is within 2 of an element of (Q). As in the proof of Theorem 5.3,
Ae = eAe eA e = De, where e is a Rohlin projection for the inclusion A P
and D = {e} P . By Lemma 5.1, we can dene a map : A D such that
ae = (a)e for a A. It is easy to see that is a unital injective homomorphism
and (x) = x for x P . So we can dene a unital homomorphism : Q P
by (q) = ((q)). Since x a = (x) (a) = x (a) for x S and
a A, we know that S is within 2 of an element of (Q).

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

193
17

For n N, let In be an ideal of l (N, P ) dened by

In = {(ak )
k=1 l (N, P ) : ak = 0 for k > n}.


Then {In } is an increasing chain of ideals in l (N, P ) and c0 (P ) = In . Since
C is exible, Q has the semiprojectivity. By the denition of the semiprojectivity
(Denition 14.1.3 in [15]), there exist n N and so that the diagram
l (N, P )/In


Q

commutes. For each k N, let k be a map from Q to P so that (q)

= (k (q))
k=1 +

In for q Q. By the above commutative diagram, we have (q) = (k (q))


k=1 +
c0 (P ) and k is a homomorphism for k > n. For x S, we can choose qx Q such
that x (qx ) < 2 . Then we have

lim sup x k (qx ) < .


2
k
Since S is a nite set, there exists k0 > n such that x k0 (qx ) < for every x
in S. Therefore P is a unital local C-algebra.
We have the following result.
Corollary 5.8. Let A P be an inclusion of separable unital C -algebras with
the Rohlin property.
(1) If A is a unital AI algebra, as dened in Example 2.2 in [16], then P is a
unital AI algebra.
(2) If A is a unital AT algebra, as dened in Example 2.3 in [16], then P is a
unital AT algebra.
(3) If A is a unital AD algebra, as dened in Example 2.4 in [16], then P is a
unital AD algebra.
(4) If A is a unital countable direct limit of one dimensional noncommutative
CW complexes (Denition 2.5 in [16]), then so is P .
Proof. Since the relevant classes are exible by [16], we may apply Theorem 5.7.
The notion of topological stable rank for a C -algebra A, denoted by tsr(A),
was introduced by Rieel, which generalized the concept of the dimension of a
topological space [18]. A unital C -algebra A has topological stable rank one if the
set of invertible elements of A is dense in A. We have the following result.
Theorem 5.9. Let A P be an inclusion of unital C -algebras with the Rohlin
property. If tsr(A) = 1, then tsr(P ) = 1.
Proof. We shall prove that for every element x of P and every > 0, there is an
invertible element y of P such that x y < .
So x x P and > 0. Since tsr(A) = 1 and x P A, there is an invertible
element a A such that x a < 2 . Let : A D = {e} P be the
injective homomorphism dened in the proof of Theorem 5.7, where e is the Rohlin
projection in A . Since (x) = x, we have x (a) < 2 . Let (an ) be a sequence
of elements in P so that (a) = (an ) + c0 (P ). Then we have

lim sup x an < .


2
n

194
18

HIROYUKI OSAKA , KAZUNORI KODAKA, AND TAMOTSU TERUYA

Since a is invertible element in A, so is (a) = (an ) + c0 (P ) in P . Therefore there


is k N such that ak is an invertible element in P and x ak < .
The theory of real rank for C -algebra, developed by Brown and Pedersen [2],
formally resembles the theory of topological stable rank, but there are important
dierences under the surface. On the other hand, the real rank zero property is one
of the most signicant properties that a C -algebra can have. A unital C -algebra
A has real rank zero if the set of invertible self-adjoint elements of A is dense in
the set of self-adjoint elements of A. We have, then, the following theorem. Since
its proof is very similar to the proof of Theorem 5.9, we omit it.
Theorem 5.10. Let A P be an inclusion of unital C -algebras with the Rohlin
property. If A has real rank zero, then P has real rank zero.
Acknowledgement The authors would like to thank the referee for his useful
comments and constructive suggestion.
References
[1] R. J. Blattner, M. Cohen and S. Montgomery, Crossed products and inner actions of Hopf
algebras, Trans. Amer. Math. Soc. 298 (1986), p. 671711.
[2] L. G. Brown and G. K. Pedersen, C -algebras of real rank zero, J. Funct. Anal. 99(1991), p.
131149.
[3] R. H. Herman and V. F. R. Jones, Period two automorphisms of UHF C -algebras, J. Funct.
Anal. 45(1982), p. 169176.
[4] R. H. Herman and V. F. R. Jones, Models of nite group actions, Math. Scand. 52(1983), p.
312320.
[5] J. G. Glimm, On a certain class of operator algebras, Trans. Amer. Math. Soc. 95 (1960), p.
318340.
[6] M. Izumi, Inclusions of simple C -algebras, J. reine angew. Math. 547 (2002), p. 97138.
[7] M. Izumi, Finite group actions on C -algebras with the Rohlin propertyI, Duke Math. J.
122(2004), p. 233280.
[8] J. A. Jeong, H. Osaka, N. C. Phillips and T. Teruya, Cancellation for inclusions of C algebras of nite depth, to appear in Indiana U. Math J. arXiv:0704.3645v1 [math.OA].
[9] J. A. Jeong and G. H. Park, Saturated actions by nite dimensional Hopf *-algebras on
C -algebras Intern. J. Math 19(2008), p. 125144.
[10] J. F. R. Jones, Index for subfactors, Inventiones Math. 72(1983), p. 125
[11] A. Kishimoto, On the xed point algebra of UHF algebra under a periodic automorphism of
product type , Publ. Res. Inst. Math. Sci. 13(1997/1998), p. 777791.
[12] A. Kishimoto, Automorphisms of AT algebras with Rohlin property, J. Operator Theory
40(1998) p. 277294.
[13] A. Kishimoto, Unbouded derivations in AT algebras, J. Funct. Anal. 160(1998), p. 270311.
[14] H. Nakamura, Aperiodic automorphisms of nuclear purely innite simple C -algebras, Ergodic Theory Dynam. Systems 20(2000), p. 17491765.
[15] T. A. Loring, Lifting Solutions to Perturbing Problems in C -algebras, Fields Institute Monographs no. 8, American Mathematical Society, Providence RI, 1997.
[16] H. Osaka and N. C. Phillips, Crossed products by nite group actions with the Rokhlin property, arXiv:math.OA/0704.3651.
[17] N. C. Phillips, The tracial Rokhlin property for actions of nite groups on C -algebras
arXiv:math.OA/0609782.
[18] M. A. Rieel, Dimension and stable rank in the K-theory of C*-algebras, Proc. London Math.
Soc. 46(1983), p. 301333.
[19] M. E. Sweedler, Hopf algebras, Benjamin, New York, 1969.
[20] W. Szyma
nski and C. Peligrad, Saturated actions of nite dimensional Hopf *-algebras on
C -algebras, Math. Scand. 75 (1994), p. 217239.
[21] Y. Watatani, Index for C -subalgebras, Mem. Amer. Math. Soc. 424, Amer. Math. Soc.,
Providence, R. I., (1990).

THE ROHLIN PROPERTY FOR INCLUSIONS OF C -ALBEBRAS

195
19

Department of Mathematical Sciences, Ritsumeikan University, Kusatsu, Shiga, 5202152 Japan


E-mail address: osaka@se.ritsumei.ac.jp
Department of Mathematical Sciences, Faculty of Science, Ryukyu University,, Nishiharacho, Okinawa 903-0213, Japan
E-mail address: kodaka@math.u-ryukyu.ac.jp
Department of Mathematical Sciences, Ritsumeikan University, Kusatsu, Shiga, 5202152 Japan
E-mail address: teruya@se.ritsumei.ac.jp

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

The C-envelope of a semicrossed product


and Nest Representations
Justin R. Peters
Abstract. Let X be compact Hausdor, and : X X a
continuous surjection. Let A be the semicrossed product algebra corresponding to the relation f U = U f or to the relation
U f = f U. Then the C -envelope of A is the crossed product of
a commutative C -algebra which contains C(X) as a subalgebra,
with respect to a homeomorphism which we construct. We also
show there aresuciently many nest representations.

1. Introduction
In [15] the notion of the semi-crossed product of a C -algebra with
respect to an endomorphism was introduced. This agreed with the notion of a nonselfadjoint or analytic crossed product introduced earlier
by McAsey and Muhly ([12]) in the case the endomorphism was an automorphism. Neither of those early papers dealt with the fundamental
question of describing the C -envelopes of the class of operator algebras
being considered.
That open question was breached in the paper [13], in which Muhly
and Solel described the C -envelope of a semicrossed product in terms
of C -correspondences, and indeed determined the C -envelopes of many
classes of nonselfadjoint operator algebras.
While it is not our intention to revisit the results of [13] in any
detail, we recall briey what was done. Given a C -algebra C and an
endomorphism of C one forms the semicrossed product A := C  Z+
as described in Section 3. First one views C as a C -correspondence E
by taking E = C as a right C module, and the left action given by the
2000 Mathematics Subject Classication. primary 46L70; secondary 37B99,
47L65.
c Mathematical
0000
(copyright Society
holder)
c
2009
American

1
197

2198

JUSTIN R. PETERS

endomorphism. One then identies the tensor algebra (also called the
analytic Toeplitz algebra) T+ (E) with the semicrossed product A. The
C -envelope of A is given by the Cuntz-Pimsner algebra O(E).
The question that motivated this paper was to nd the relation
between the C -envelopes of semicrossed products, and crossed products. Specically, when is the C -envelope of a semicrossed product a
crossed product? If the endomorphism of C is actually an automorphism, then the crossed product C  Z is a natural candidate for the
C -envelope, and indeed, as noted in [13], this is the case. In this paper
we answer that question in case the C -algebra C is commutative (and
unital). Indeed, it turns out that the C -envelope is always a crossed
product (cf Theorem 4).
For certain classes of nonselfadjoint operator algebras, nest representations play a fundamental role akin to that of the irreducible
representations in the theory of C -algebras. The notion of nest representation was introduced by Lamoureux ([10], [11]) in a context with
similarities to that here. We do not answer the basic question as to
whether nest representations suce for the kernel-hull topology; i.e.,
every closed ideal in a semicrossed product is the intersection of the
kernels of the nest representations containing it. What we do show is
that nest representations suce for the norm: the norm of an element
is the supremum of the norms of the isometric covariant nest representations (Theorem 2). The results on nest representation require some
results in topological dynamics, which may not be new, but we have not
seen in the literature. Other related results on topological dynamics
and crossed products are in [2],[3].
Semicrossed products can be dened by either of the relations
f U = U f or U f = f U.
The semicrossed products corresponding to these relations admit dierent representations, and are generally not isomorphic. (See discussion
in section 4.) Nevertheless, they have the same C -envelope. (Theorem 4)
The history of work in anaylytic crossed products and semicrosed
products goes back nearly forty years. While in this note we do not
review the literature of the subject, we mention the important paper
[6] in which the Jacobson radical of a semicrossed product is determined and necessary and sucient conditions for semi-simplicity of
the crossed product are obtained. We use this in Proposition 3 to show
that the simplicity of the C -envelope implies the semisimplicity of the
semicrossed product.

C -ENVELOPE AND NEST REPRESENTATIONS

199
3

In their paper [4], Davidson and Katsoulis view semicrossed products as an example of a more general class of Banach Algebras associated with dynamical systems which they call conjugacy algebras. They
have extracted fundamental properties needed to obtain, for instance,
the result that conjugacy of dynamical systems is equivalent to isomorphism of the conjugacy algebras. It would be worthwhile to extend the
results here to the broader context. In their recent work [5] they indicate that for multivariable dynamical systems which do not commute,
the C -envelope of the semicrossed product is not a crossed product
for n 2 noncommuting actions.
2. Dynamical Systems
In our context, X will denote a compact Hausdor space. By a
dynamical system we will simply mean a space X together with a mapping : X X. In this article, the map will always be a continuous
surjection.
Definition 1. Given a dynamical system (X, ) we will say (following the terminology of [16]) the dynamical system (Y, ) is an extension of (X, ) in case there is a continuous surjection p : Y X
such that the the diagram

Y Y

p
p

()

X X
commutes. The map p is called the extension map (of Y over X).
Notation. In case p is a homeomorphism, it is called a conjugacy.
Given a dynamical system (X, ) there is a canonical procedure for
producing an extension (Y, ) in which is a homeomorphism.
= {(x1 , x2 , . . . ) : xn X and xn = (xn+1 ), n = 1, 2, . . . }.
Let X
is a closed subset of the product Xn where Xn = X, n =
As X
n=1
is compact Hausdor. Dene a map : X
X
by
1, 2, . . . , so X
(x
1 , x2 , . . . ) = ((x1 ), x1 , x2 , . . . ).
This is continuous, and has an inverse given by
1 (x1 , x2 , . . . ) = (x2 , x3 , . . . ).
X by
Dene a continuous surjection p : X
p(x1 , x2 , . . . ) = x1 .

4200

JUSTIN R. PETERS

)
With the map p, the system (X,
is an extension of the dynamical
system (X, ) in which the dynamics of the extension is given by a
homeomorphism.
Definition 2. In the case of an extension in which the dynamics
is given by a homeomorphism, we will say the extension is a homeomorphism extension.
)
Notation. We will call the extension (X,
the canonical home
omorphism extension. If x
X, x = (x1 , x2 , . . . ), we will say that
(x1 , x2 , . . . ) are the coordinates of x.
Definition 3. Given a dynamical system (X, ), a homeomorphism extension (Y, ) is said to be minimal if, whenever (Z, ) has the
property that it is a homeomorphism extension of (X, ), and (Y, ) is
an extension of (Z, ) such that the composition of the extension maps
of Z over X with the extension map of Y over Z is the extension map
of Y over X, then (Y, ) and (Z, ) are conjugate.
Lemma 1. Let (X, ) be a dynamical system. Then the canonical
)
homeomorphism extension (X,
is minimal.
Proof. Suppose (Z, ) is a homeomorphism extension of (X, ), p :
Z and q : Z X are continuous surjections, and the diagram
X

X
X

p
p

Z Z

q
q

X X
over X,
commutes and the composition q p is the extension map of X
i.e., the projection onto the rst coordinate.

Observe that the canonical homeomorphism extension (Z,
) of
(Z, ) is in fact conjugate to (Z, ). Indeed, the map which sends
z Z  (z, 1 (z), 2 (z), . . . ) is a conjugacy. Thus it is enough to
)

show that (X,
is conjugate to (Z,
).

Dene a map r : Z X by
z := (z, 1 (z), 2 (z), . . . ) Z  x := (q(z), q( 1 (z), q( 2 (z)), . . . ).

C -ENVELOPE AND NEST REPRESENTATIONS

201
5

since
Observe that this maps into X,
(q( (n+1) (z))) = q(( (n+1) (z)))
= q( n (z))
Let x = (x1 , x2 , . . . ) be any element
Next we claim r maps onto X.
Let zn Z be any element such that q(zn ) = xn , n = 1, 2, . . . .
of X.
Let
zn := ( n1 (zn ), . . . , zn , 1 (zn ), . . . ).
n

zm ) agrees with
A subsequence of {zn } converges, say, to z. Since r(
x in the rst n coordinates for all m n, it follows that r(
z ) = x.

To show that r is one-to-one, dene a map p : X Z by


p(
x) = (p(
x), 1 p(
x), 2 p(
x) . . . ).
Z is surjective implies that p is
Note that the fact that p : X
Then
surjective. Let x = (x1 , x2 , x3 , . . . ) X.
x), 2 p(
x) . . . )
r p(
x) = r(p(
x), 1 p(
= (q p(
x), q 1 p(
x), q 2 p(
x), . . . )
= (x1 , q p 1 (
x), q p 2 (
x), . . . )
= (x1 , x2 , x3 , . . . ) = x.
where we have used the fact that q p is the projection onto the rst
coordinate of x. Since p is surjective and r p is injective, it follows
that r is injective, and hence r is a conjugacy.

Lemma 2. Let (X, ) be a dynamical system, and let (Y, ) be a
minimal homeomorphism extension. Then (Y, ) is conjugate to the
).
canonical homeomorphism extension, (X,

Proof. By assumption there is a continuous surjection p : Y X


such that the diagram

Y Y

q
q

X X
commutes.

6202

JUSTIN R. PETERS

Consider the diagram

Y Y

q
q

X
X

p
p

X X
over X (i.e., the
where p denotes the canonical extension map of X
projection onto the rst coordinate), and the map q is dened as follows:
For y Y, q(y) = (q(y), q 1 (y), q 2 (y), . . . ).
since (q (n+1) (y) = q (n+1) (y) =
Note that the image lies in X
q n (y).
Next, observe that p q(y) = q(y), so the extension property is
satised. Hence, by denition of minimality of the homeomorphism
extension (Y, ), the map q is a conjugacy.

Corollary 1. Let (X, ) be a dynamical system. Then there exists a minimal homeomorphism extension (Y, ) which is unique up
)
to conjugacy. In particular, the canonical extension (X,
is such a
homeomorphism extension.
If (X, ) is a dynamical system, then the map : C(X) C(X)
sending f  f , is a -endomorphism. is a -automorphism i
is a homeomorphism. We can dualize the preceding results as follows:
Corollary 2. Given a dynamical system (X, ), there is a min with automorphism
imal commutative C -algebra C(X)
admitting

an embedding : C(X)  C(X) such that


= . Furthermore,
this commutative C -algebra is unique up to isomorphism.
Proof. Consider the inductive limit

C(X) C(X) C(X) . . . .


The inductive limit is a C -algebra, C(Y ) containing C(X) as a subalgebra, and C(Y ) admits an automorphism, satisfying (f ) = (f )
for f C(X).

C -ENVELOPE AND NEST REPRESENTATIONS

203
7

)
But, with (X,
the minimal homeomorphism extension of (X, ),
we can consider the inductive limit
and viewing C(X)  C(X),
1 (C(X))
2 (C(X)) . . . .
C(X)
id

id

The two inductive limits are isometrically isomorphic, as we have


the commutative diagram

C(X)

n (C(X))

id

(n+1)

C(X)
(n+1) (C(X)))
and we have the relation
Thus, we may identify Y with X,
n (f ), f C(X), n Z+ ,

(n+1) (f ) =
hence
(f ) =
(f ), or
=
by .
if we denote the embedding of C(X) into C(X)

Definition 4. Given a dynamical system (X, ), a point x X
is periodic if, for some n N, n 1, n (x) = x. If n is the smallest
integer with this property, we say that x is periodic of period n. If x
is not periodic, we say x is aperiodic. If for some m N, m (x) is
periodic, then we say x is eventually periodic.
Remark 1. If is a homeomorphism, then a point is eventually
periodic i it is periodic; but if is a continuous surjection, it is possible
to have a point x which is aperiodic and eventually periodic.
)
Lemma 3. Let (X, ) be a dynamical system, and (X,
its mini is
mal homeomorphism extension. Then a point x = (x1 , x2 , . . . ) X
aperiodic i for any n N, xn = xm for at most nitely many m N,
and x is periodic i x is periodic.
Proof. x is periodic of period p i p (
x) = x, equivalently,
(x1 , x2 , . . . ) = (p (x1 ), p (x2 ), . . . )
= (x1 , . . . , xp , x1 , . . . , xp , . . . )
which uses the relation that p (xp+j ) = xj , j N.
This shows that if x is periodic, the coordinates of x form a periodic
sequence; the converse is also clear.


8204

JUSTIN R. PETERS

Definition 5.
(1) Recall a dynamical system (X, ) is topologically transitive if for any nonempty open set O X,
n

O = X.
n=0

(2) A dynamical system (X, ) is minimal if there is no proper,


closed subset Z X such that (Z) = Z.
(3) A point x in a dynamical system (X, ) is recurrent if there is
a subsequence {ni } of N such that ni (x) x.
Remark 2. There should be no confusion between the two distinct
uses of minimal.
)
Theorem 1. Let (X, ) be a dynamical system, and (X,
the
minimal homeomorphism extension.
is metrizable.
(1) X is metrizable i X
)
(2) (X, ) is topologically transitive i (X,
is topologically transitive.
(3) (X, ) has a dense set of periodic points i the same is true of
).
(X,

(4) (X, ) is a minimal dynamical system i the minimal homeomorphism extension has the same property.
(5) The recurrent points in X are dense i the recurrent points in
are dense.
X
Proof. (1) is routine.
)
(2) Let (X,
be topologically transitive, and
= O X. Then
1

so by assumption X
=

O := p (O) is nonempty in X,
n (O).
n=0
Let x X and x p1 (x). Then there exists n such that x
So y = n (
y ), y O.
x), and so p(
y ) = p(n (
x) = n p(
x) =
n (
n
n
n
y ) (O).
(x). Thus x (p(
For the other direction, by Corollary 1 we can assume, without loss
)
of generality, that (X,
is the canonical minimal homeomorphism
have the form
extension of (X, ). The basic open sets in X
=X
[O1 ON
O
n=N +1 Xn ]
for some N N, O1 , . . . ON open sets in X, and where Xn = X for all
n > N.
is nonempty, there is a point x ON such that j (x)
If O
ON j , j = 1, . . . , N 1. Hence by continuity of there is a neighborhood U ON such that j (U ) ON j , j = 1, . . . , N 1.

C -ENVELOPE AND NEST REPRESENTATIONS

205
9

be arbitrary, x = (x1 , . . . , xN , . . . ). By the topological


Let x X
transitivity of (X, ) we can nd n N such that n (xN ) U. Thus,
x) = (n (x1 ), . . . , n (xN ), . . . )
n (
N 1 (U ) . . . (U ) U
n=N +1 Xn

O1 ON 1 ON n=N +1 Xn
which nishes the proof.
so that n (
x) O
let x X and x p1 (x).
(3) If the periodic points are dense in X,
converging to x. By Lemma 3,
Then there is a sequence {
yn } X
p(
yn ) is periodic in X, and converges to x.
For the converse, note that if x X is periodic, say of period n, then
p(
there is a point x X,
x) = x with x periodic of period n. Indeed,
if x has orbit x, (x), . . . , n1 (x), then, setting xj = n+1j (x), j =
1, . . . , n, take x to be the point with coordinates
x = (x1 , x2 , . . . , xn , x1 , x2 , . . . , xn , . . . ).
is a basic open set in X,
we use the argument in (2) to nd an
If O
integer N and an open set U as in (2). Let y U be periodic, and set
is periodic and belongs to O.

x = N 1 (y). The point x X


)
(4) Assume (X,
is a minimal dynamical system, and Y X
a nonempty closed, -invariant subset. Then p1 (Y ) is a nonempty
Thus, Y = X, and so
so p1 (Y ) = X.
closed invariant subset of X,
(X, ) is minimal.
Conversely, assume (X, ) be a minimal dynamical system, and

let Y = {Yi }iI be a maximal chain of closed invariant subsets of X,


ordered by inclusion. Then
Y = iI Yi
is the minimal element of the chain, hence Y has no proper invariant subset. As Y
= , p(Y ) is a nonempty invariant subset of X, so
p(Y ) = X. Taking = |
Y , and q = p|Y , we have that (Y, ) is a
homeomorphism extension of (X, ). While we do not know a priori
that (Y, ) is a minimal homeomorphism extension of (X, ), if (Y, )
is not a minimal homeomorphism extension, there is an intermediate
extension (Z, ), as in the proof of Lemma 1. As is a minimal homeomorphism on Y , is a minimal homeomorphism on Z. It follows that
the minimal homeomorphism extension of (X, ) which lies between X
and Y is necessarily a minimal homeomorphism.
)
Since (X,
was the canonical minimal extension, and since any
two minimal extensions are conjugate, it follows that the dynamical
)
system (X,
is minimal.

206
10

JUSTIN R. PETERS

are dense, let U be any nonempty


(5) If the recurrent points in X
open set in X. Then there exists y p1 (U ) which is recurrent. But
then y := p(
y ) U is recurrent.
Now assume (X, ) has a dense set of recurrent. First we show that
if x X is recurrent, there is x p1 (x) which is recurrent. So, let
be such that p(
x) = x
x X be recurrent, and let x = (x1 , x2 , . . . ) X
(so x = x1 ). By the compactness of X and a standard diagonalization
argument, there is a subsequence {nj } of N and yi X, i = 1, 2, . . . ,
such that
lim nj (xi ) = yi , i = 1, 2, . . .
j

and y1 = x1 . Since (xi+1 ) = xi , the same relation holds for the yi ,


Since
and hence y := (x1 , y2 , y3 , . . . ) X.
lim nj (yi ) = lim nj i+1 (x1 ) = lim nj (xi ) = yi ,
j

i = 1, 2, . . . , this shows that y p1 (x) is recurrent.


X
be a basic open set, and let U be an open set in X
Now, let O
and N N be as in the proof of (2). Let xN U be recurrent; by the
above assertion we can nd x = (x1 , . . . , xN , . . . ) which is recurrent in

X, and by construction x lies in O.



3. Representations of Semicrossed Products
For the moment we will take an abstract approach: Let (X, ) be
a dynamical system, and consider the algebra generated by C(X) and
a symbol U , where U satises the relation
() f U = U (f ), f C(X).
Throughout this section we will assume this relation is satised. We
will consider the other relation in the next section.
The elements F of this algebra can be viewed as noncommutative
polynomials in U ,
F =

N


U n fn , fn C(X), N N.

n=0

Let us call this algebra A0 .


In [15] we formed the Banach Algebra 1 (A0 ) by providing a norm

to elements F as above as ||F ||1 = N
n=0 ||fn || and then completing A0
in this norm. On the other hand, we can dene the class of representations of A0 and complete A0 in the resulting norm. Either approach
yields the same semicrossed product.

C -ENVELOPE AND NEST REPRESENTATIONS

207
11

By a representation of A0 we will mean a homomorphism of A0 into


the bounded operators on a Hilbert space, which is a -representation
when restricted to C(X), viewed as a subalgebra of A0 , and such that
(U ) is an isometry.
Fix a point x X and, for convenience, set x1 = x, x2 = (x), x3 =
2 (x), . . . . Dene a representation x of A0 on 2 (N) by
x (f )(z1 , z2 , . . . ) = (f (x1 )z1 , f (x2 )z2 , . . . ),
2
with (zn )
n=1  (N) and

x (U )(z1 , z2 , . . . ) = (0, z1 , z2 , . . . ).
Observe this is a representation of A0 since
x (f U )(z1 , z2 , . . . ) = (0, f (x2 )z1 , f (x3 )z2 , . . . )
and
x (U f )(z1 , z2 , . . . ) = x (U )(f (x1 )z1 , f (x2 )z2 , . . . )
= (0, f (x1 )z1 , f (x2 )z2 , . . . )
= (0, f (x2 )z1 , f (x3 )z2 , . . . )
)
Let (X,
be the canonical homeomorphism extension. (cf denition 2 and Corollary 1.) We will consider A0 as embedded in A0 ,
and U , satisfying the same
where A0 is the algebra generated by C(X)
and set x = p(
X is the map
relation (). Let x X
x) where p : X
in diagram ().
f = f p, and for F = N U n fn , fn
For f C(X), let f C(X),
n=0

n fn . Observe that
C(X), let F = N
U
n=0
x (F ) = x (F ).
3.1. Nest Representations. For nonselfadjoint operator algebras,
the representations which can play the role of the primitive representations in the case of C -algebras are the nest representations. Recall,
a representation of an algebra A on a Hilbert space H is a nest
representations if the lattice of subspaces invariant under is linearly
ordered.
Let (X, ) be a dynamical system with a homeomorphism, and
x a point in X which is aperiodic.
Lemma 4. Let x X be aperiodic. Then weak closure of x (C(X))
is a masa in B(H).

208
12

JUSTIN R. PETERS

Proof. Let {xn } be the sequence x1 = x, . . . , xj = j1 (x) for


j > 1. It is enough to show that the operator en belongs to the weak
closure, where en is the multiplication operator which is 1 in the nth
coordinate and zero elsewhere. We can nd fm C(X) satisfying

1 for j = n
fm (xj ) =
0 for j
= n, j m
and fm is real-valued, 0 fm 1. Indeed, this follows from the Tietze
Extension Theorem.
As x (fm ) en weakly, we have en in the weak closure of x (C(X),
and we are done.

Proposition 1. Let (X, ) be a dynamical system. If x X is
aperiodic, then x is a nest representation.
Proof. Since the weak closure of (C(X)) is a masa (Lemma 4),
the closed subspaces S of 2 (N) invariant under (C(X)) are the vectors
z 2 (N) which are supported on a given subset of N. If such a
subspace is also invariant under (U ) then it has the form
S = {
z 2 (N) : zn = 0 for n N }
for some N N. But then the subspaces S are nested.

To periodic points we can associate another class of nest represen be periodic of period N , so x = (x1 , x2 , . . . ) with
tations. Let x X
xi+N = xi for i N. Let : A B(2 (N )) by (f )(z1 , . . . , zN ) =
(f (x1 )z1 , . . . , f (xN )zN ) and (U )(z1 , . . . , zN ) = (zN , z1 , . . . , zN 1 ).
For C crossed products B := C(X)  Z where is a homeomorphism, we have the representations x and y, for x aperiodic, y
periodic, and T given as follows: x acts on 2 (Z), where x (U ) is
the bilateral shift (to the right), and
x (f )(n ) = (f ( n (x))n ) n Z, f C(X).
y, acts on the nite dimensional space 2 (p), where p is the period
of the orbit of y. y, (U ) is a cyclic permutation along the (nite) orbit
of y composed with multiplication by , and y, (f ) acts like x (f )
along the orbit of y. These representations correspond to the pure state
extensions of the states on C(X), f f (x) in the cases where x is
aperiodic or periodic, respectively, and so are irreducible. However, not
all irreducible representations of B need be of this form. Nevertheless,
Tomiyama has shown:

C -ENVELOPE AND NEST REPRESENTATIONS

209
13

Proposition 2. Every ideal of B is the intersection of those ideals


of the form ker(x ) and ker(y, ) (x aperiodic, y periodic, T)
which contain it.
This is Proposition 4.1 of [18].
Corollary 3. If (Y, ) is a dynamical system with a homeomorphism, then for F C(Y )  Z,
||F || = max{A, B}
where
A = sup{||x (F )|| : x aperiodic}
and
B = sup{||y, (F )|| : y periodic, T.
Proof. Denote by || || the crossed product norm, and by || ||
the norm dened in the statement of the corollary. Let I be the ideal
in C(Y )  Z of all F with ||F || = 0. Every ideal of the form ker(x )
(x aperiodic) and ker(y, ) (y periodic, T) contains I. Since the
zero ideal also has this property, it follows from Proposition 2 that
I = (0).

Lemma 5. Let (X, ) be a dynamical system with a homeomorphism, and let y X be periodic. For F C(X)  Z+ , we have
||y (F )|| supT ||y, (F )||.
Proof. Since any F C(X)  Z+ can be approximated by elements with nitely many nonzero Fourier coecients, we can assume
F has this property. Let y have period p, and we can assume
F =

kp


U n fn , fn C(X),

n=0

for some k Z .
Let = (1 , . . . , p ) Cp be any vector of norm 1, and x T.
For N N dene a vector 2 (N) of norm 1 by

= (1 , . . . , N p , 0, 0, . . . ) where i+jp = N j i / N ,
i = 1, . . . p, j = 0, . . . N 1.
+

Now, for k j < N,

< y (F ), ei+jp >= jk / N < y, (F ), epi >

where en resp. epn are standard basis vectors in 2 (N), resp., in Cp .


Thus, if N/k is large, it follows that ||y (F )|| is close to ||y, (F )||.
This proves the lemma.


210
14

JUSTIN R. PETERS

Lemma 6. Let (X, ) be a dynamical system with a homeomorphism, and F C(X)  Z+ . For any x X,
||x (F )|| = sup{||y (F )|| : y Orbit(x)}.
Proof. Given > 0, there is a vector 2 Z, = (n )nZ with
only nitely many n
= 0, and such that
||x (F )|| ||x (F )|| + .
Suppose n = 0 for n < N, for some N Z+ . Let y = N (x), and
dene a vector 2 (N) by: j = jN 1 , j = 1, 2, . . . . Then |||| = 1,
and
||y (F )|| = ||x (F )||.
The lemma now follows.

)
Corollary 4. Let (X, ) be a dynamical system, and (X,
a
X the continuous
minimal homeomorphism extension, with p : X
surjection for which the diagram () commutes. Let F C(X)  Z+ ,
Then
and let x X.
y ), for y Orbit(
x)}.
||x (F || = sup{||y (F )|| : y = p(
and 2 (N),
Proof. Observe that for any y X,
y(F ) = y (F ).
Now apply Lemma 6.

Definition 6. For a dynamical system (X, ) ( not necessarily


a homeomorphism), a periodic point y X and T, we dene y,
exactly like y, in the case where is a homeomorphism.
Remark 3. Since y, is irreducible, the same is true for y, , and
in particular y, is a nest representation.
Corollary 5. Let (X, ) be a dynamical system, F C(X) Z+ .
Then
||F || = max{A, B}
where
A = sup{||x (F )|| : x aperiodic}
and
B = sup{||y, (F )|| : y periodic, T.
Proof. Note the constant A is the same as in Corollary 3 , and
by Lemma 6 the constant B is the same as in Corolary 3. For y X
periodic and T, we have by Lemma 5
sup ||y, (F )|| = sup ||P iy, (F )|| ||y (F )|| ||y (F )||
T

C -ENVELOPE AND NEST REPRESENTATIONS

211
15

is periodic and p(
where y X
y ) = y.
By Cor. II.8 of [15], ||F || = supxX ||x (F )||. Thus, denoting the
sup{A, B} by ||F || , by Corollary 3 it follows that ||F || is the norm of
 Z. On the other hand,
F in the crossed product C(X)
||F || sup ||x (F )|| = ||F || sup ||x (F )||
xX

x
X

and the last term is dominated by the norm of F in the crossed product, since the norm there is given by the supremum over all covariant
representations.

Theorem 2. Let (X, ) be a dynamical system, F C(X)  Z+ .
Then
||F || = sup{||(F )|| : is an isometric covariant nest representation}.
Proof. Indeed, we have found a subclass of the isometric covariant
nest representations, namely the y, (y periodic, T) and x (x
aperiodic) which yield ||F ||.

From the above, we obtain the following
)
Theorem 3. Let (X, ) be a dynamical system, and (X,
its
minimal homeomorphism extension. Then the embedding of the semi  Z into the crossed product is
crossed product C(X)  Z+  C(X)
a completely isometric isomorphism.
Corollary 6. With notation as above, the semicrossed product
 Z+ is semisimple.
C(X)  Z+ is semisimple i C(X)
Proof. This follows from part (5) of Theorem 1 and the main
result of [6]. The latter proof is given in the metric space context,
but carries over to the compact Hausdor context. ([19] Chapter I
Proposition 2.18 p. 63 and 2.24 p.66.)

If the crossed product is a simple C -algebra, the crossed product is
necessarily the C -envelope. However, as we will now show, it is always
the case that the crossed product is the C -envelope, even if it is not
simple.
Lemma 7. Let (X, ) be a dynamical system, and C(X)  Z+ the
associated semicrossed product. Then the endomorphism of C(X),
(f ) = f , extends to an endomorphism, again denoted by , of the
semicrossed product.
Z. The element U , which
Proof. Embed C(X) Z+  C(X)
is an isometry in the semicrossed product, embeds to a unitary in

212
16

JUSTIN R. PETERS

the crossed product, and one can dene an automorphism


on the
crossed product, which extends the automorphism, also denoted by

of C(X),
(f ) = f .

 Z one has U f U = f .
This is as follows: in C(X)
For F
an element of the crossed product, dene
(F ) = U F U . Note that,
if {fn } are the Fourier coecients of F , then {fn }
are the Fourier
coecients of
(F ).
In particular, if F belongs to the semicrossed product, and so its
then the
Fourier coecients belong to the subalgebra C(X)  C(X),
Fourier coecients of (F
) also belong to the subalgebra C(X), since
for f C(X),
(f ) = (f ). Thus, if we denote this map of C(X) Z+
by , it is an endomorphism of the semicrossed product extending the
endomorphism of C(X).

Lemma 8. Let (X, ) be a dynamical system, and embed
 Z. Then
A := C(X)  Z+  C(X)
 Z is a dense subalgebra.

k (A) C(X)
k=0
Proof. It follows from the proof of Corollary 2 that, viewing C(X)
that
n (C(X)) is a dense subalgebra of
as embedded in C(X),
n=0

C(X).
 Z with
 Z and > 0, there is G C(X)
Given F C(X)
N
n
nitly many nonzero Fourier coecients, say G =
n=0 U gn , with
||F G|| < . By the rst paragraph, for each gn there is an hn in
with ||gn hn || <  . But if H =
the dense subalgebra of C(X)
N +1
N
n

k
U
h
,
we
have
||F

H||
<
2 ,
and
H

(A).
n
k=0
n=0

4. The C -envelope
In the introduction we mentioned that two relations can be used to
dene the semicrossed product:
(1)

fU = U f

or
(2)

Uf = f U

Up to this point we have been working exclusively with relation 1 One


observes that the the representations x do not satisfy relation 2 Thus,
the development of the theory obtained thus far is not directly applicable to semicrossed products based on relation 2. There is, however,
an analogous class of representations of representations, which we now
introduce.

C -ENVELOPE AND NEST REPRESENTATIONS

213
17

With (X, ) as before, choose x = x1 X and choose a backward orbit (x1 , x2 , x3 , . . . ) = O, by which we mean that (xn+1 ) =
xn , n 1. Then, with the Hilbert space H = 2 (N), dene the orbit
representation O by
O (U )(z1 , z2 , . . . ) = (0, z1 , z2 , . . . )
O (f )(z1 , z2 , . . . ) = (f (x1 )z1 , f (x2 )z2 , . . . )
one easily veries that the relation
O (U f ) = O (f U )
is satised.
With these representations O in place of the representations x
of section 3.1, all the results of that section carry over to semicrossed
product algebras dened by relation 2.
Remark 4. Given a dynamical system (X, ), the semicrossed algebra based on relation 1 is generally not isomorphic with the semicrossed product algebra based on relation 2. Indeed, even if is a
homeomorphism the two semicrossed products are non-isomorphic.
Suppose is a homeomorphism, and U, satisfy relation 1. Writing f as g 1 , and every function in C(X) can be so written, we
obtain
U g = g 1 U
which is relation 2 with 1 in place of . Similarly, relation 2 yields
relation 1 with 1 in place of . There is a theorem that two semicrossed products are isometrically isomorphic if and only if the dynamical systems which dene them are conjugate. ([17]) However, it
is known that a homeomorphism of a compact metric space is not, in
general, conjugate to its inverse. This shows that, even in the case of
homeomorphisms, semicrossed products dened by relation 1 are not
isomorphic with those dened by relation 2.
)
Theorem 4. Let (X, ) be a dynamical system, and (X,
its
+
minimal homeomorphism extension. Let C(X) Z be the semicrossed
product dened by either relation 1 or 2. Then the C -envelope of the
 Z.
semicrossed product is the crossed product C(X)
Proof. We give the proof assuming the semicrossed product is
dened by relation 1. However, the proof for relation 2 would be identical, except that in the element G below, the U n and gn would be
transposed.
By Theorem 3 the embedding
 Z
C(X)  Z+  C(X)

214
18

JUSTIN R. PETERS

is completely isometric. Suppose there is a C -algebra B, a completely isometric embedding : C(X)  Z+ B, and a surjective
 Z B.
C -homomorphism q : C(X)
If q is not an isomorphism, let 0
= F ker(q). Assume ||F || =
N
n
1. By Lemma 8 there is an element G =
n=0 U gn with gn

n
and such
(C(X)), viewing C(X) as a subalgebra of C(X),
n=0
1
+
that ||F G|| < 2 . In particular, there is m Z such that gn m
C(X), 0 n N .

m C(X)  Z+ and ||GU m || = ||G|| > 12 .
Now GU m = N
n=0 gn
On the other hand, since q(F U m ) = q(F )q(U m ) = 0 we have
1
||q(GU m )|| = ||q(GU m F U m )|| ||GU m F U m || ||G F || < .
2

This contradiction shows that ker(q) = (0), hence that C(X)  Z


is the C envelope of the semicrossed product.

Finally, we make use of the relation between properties of dynamical
systems and their extensions to obtain
Proposition 3. Let (X, ) be a dynamical system. If the C envelope of the semicrossed product is a simple C -algebra, then C(X)
Z+ is semi-simple.
Remark 5. The converse is false.

Proof. By Theorem 4, the C -envelope is a crossed product, C(X)

Z, where (X, )
is the (unique) minimal homeomorphism extension of
(X, ). As is well known (e.g. [14] Proposition 7.9.6), the crossed prod )
uct is simple if and only if the dynamical system (X,
is minimal; i.e.,
every point has a dense orbit. By Theorem 1, this is equivalent to the
condition that (X, ) is minimal. In particular, the system (X, ) is
recurrent; so by [6] it follows that the semicrossed product is semisimple.

References
[1] Wm. B. Arveson, Subalgebras of C -algebras, Acta Math 123, 1969, 141224.
[2] T. M. Carlsen and S. Silvestrov, On the Exel crossed product of topological
covering maps, arXiv:[math.OA]/0811.0056, to appear in Acta Appl. Math.
[3] T. M. Carlsen and S. Silvestrov, C -crossed products and shift spaces, Espo.
Math. 25, 2007, no. 4, 275307.
[4] K. Davidson and E. Katsoulis, Isomorphisms between Topological Conjugacy
Algebras, J. Reine Angew. Math. 621, 2008, 2951.
[5] K. Davidson and E. Katsoulis, Operator Algebras for Multivariable Dynamics,
arXiv.math/0701514v5, 2007.

C -ENVELOPE AND NEST REPRESENTATIONS

215
19

[6] A. Donsig, A. Katavolous, and A. Manousos, The Jacobson Radical for Analytic Crossed Products, J. Func. Anal. 187, 2001, 129145.
[7] R. Exel and A. Vershik, C -algebras of irreversible dynamical systems, Canad.
J. Math. 58, 2006, no. 1, 3963.
[8] M. Hamana, Injective Envelopes of Operator Systems, Publ. Res. Inst. Math.
Sci. 15, 1979,773785.
[9] E. Katsoulis and D. Kribs, Tensor Algebras of C -Correspondences and their
C -envelopes, J. Func. Anal. 234, 2006, no. 1, 226233.
[10] M. Lamoureux, Nest representations and dynamical systems, J. Func. Anal.
114, 1993, 345376.
[11] M. Lamoureux, Ideals in some continuos nonselfadjoint crossed product algebras, J. Func. Anal. 142, 1996, 221248.
[12] M. McAsey and P. S. Muhly, Representations of nonselfadjoint crossed products, Proc. London Math. Soc. Ser. 3 47, 1983, 128144.
[13] P. S. Muhly and B. Solel, Tensor Algebras over C -Correspondences: Representations, Dilations, and C -envelopes, J. Func. Anal. 158, 1998, 389457.
[14] G. Pedersen, C -Algebras and their Automorphism Groups, Academic Press,
London-New York-San Francisco, 1979.
[15] J. R. Peters, Semi-crossed products of C -algebras, J. Func. Anal. 59, 1984,
498534.
[16] T. Pennings and J. R. Peters, Dynamical Systems from Function Algebras,
Proc. Amer. Math. Soc. 105, 1989, 8086.
[17] S. Power, Classication of analytic crossed product algebras, Bull. London
Math. Soc. 24 1992, 368372.
[18] J. Tomiyama, The interplay between topological dynamics and theory of C algebras, II, Kiyoto Univ. RIMS, No., 2000, 171.
[19] J. de Vries, Elements of topological dynamics Mathematics and its Applications
257, Kluwer Academic Publishers Group, Dordrecht, 1993.
Department of Mathematics, Iowa State University, Ames, Iowa,
USA and
E-mail address: peters@iastate.edu

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Freeness of actions of nite groups on C*-algebras


N. Christopher Phillips
Abstract. We describe some of the forms of freeness of group actions on
noncommutative C*-algebras that have been used, with emphasis on actions
of nite groups. We give some indications of their strengths, weaknesses, applications, and relationships to each other. The properties discussed include
the Rokhlin property, K-theoretic freeness, the tracial Rokhlin property, pointwise outerness, saturation, hereditary saturation, and the requirement that the
strong Connes spectrum be the entire dual.

Recall that an action (g, x)  gx of a group G on a space X is free if whenever


g G \ {1} and x X, then gx = x. That is, every nontrivial group element acts
without xed points. So what is a free action on a C*-algebra?
There are several reasons for being interested in free actions on C*-algebras.
First, there is the general principle of noncommutative topology: one should nd
the C*-algebra analogs of useful concepts from topology. Free (or free and proper)
actions of locally compact groups on locally compact Hausdor spaces have a number of good properties, some of which are visible from topological considerations
and some of which become apparent only when one looks at crossed product C*algebras. We describe some of these in Section 1. Second, analogs of freeness,
particularly pointwise outerness and the Rokhlin property, have proved important
in von Neumann algebras, especially for the classication of group actions on von
Neumann algebras. Again, this fact suggests that one should see to what extent
the concepts and theorems carry over to C*-algebras. Third, the classication of
group actions on C*-algebras is intrinsically interesting. Experience both with the
commutative case and with von Neumann algebras suggests that free actions are
easier to understand and classify than general actions. (A free action of a nite
group on a path connected space corresponds to a nite covering space.) Fourth,
noncommutative analogs of freeness play an important role in questions about the
structure of crossed products. Freeness hypotheses are important for results on
both simplicity and classiability of crossed products.
It turns out that there are many versions of noncommutative freeness. They
vary enormously in strength, from saturation (or full Arveson spectrum) all the way
up to the Rokhlin property. The various conditions have dierent uses. The main
point of this article is to describe some of the forms of noncommutative freeness
2000 Mathematics Subject Classication. Primary 46L55; Secondary 46L35, 46L40.
Research partially supported by NSF grant DMS-0701076.
c
2009
American
c Mathematical
0000
(copyright Society
holder)

1
217

218
2

N. CHRISTOPHER PHILLIPS

that have been used, and give some indications of their strengths, weaknesses,
applications, and relationships to each other. To keep things simple, and to keep
the focus on freeness, we restrict whenever convenient to actions of nite groups. For
one thing, our knowledge is more complete in this case. Also, we would otherwise
have to deal with noncommutative properness; we discuss this issue briey below.
We give a rough summary of the dierent conditions and their uses, in approximate decreasing order of strength. The strongest is free action on the primitive
ideal space. Outside the class of type I C*-algebras, this condition seems too strong
for almost all purposes, and we accordingly say little about it. Next is the Rokhlin
property. This is the hypothesis in most theorems on classication of group actions.
When the group is nite, it also implies very strong structure preservation results
for crossed products. K-theoretic freeness is close to the Rokhlin property, at least
when the K-theory is suciently nontrivial. Unlike the Rokhlin property, it agrees
with freeness in the commutative case. The tracial Rokhlin property is weaker than
the Rokhlin property, and much more common; its main use is in classication theorems for crossed products. The main use of pointwise outerness, at least so far,
has been for proving simplicity of crossed products. Hereditary saturation and having full strong Connes spectrum are weaker conditions which give exactly what is
needed for crossed products by minimal actions to be simple. Unfortunately, they
are hard to verify. Saturation is the condition which makes the crossed product
naturally Morita equivalent to the xed point algebra.
For an action of a noncompact group, there is a big dierence between actions
that are merely free and those that are both free and proper. Recall that an action
of a locally compact group G on a locally compact space X is proper if for every
compact set K X, the set {g G : gK K = } is compact in G. Equivalently,
the map (g, x)  (x, gx) is a proper map, that is, inverse images of compact sets
are compact.
One of the good things about a free action of a compact Lie group on a locally
compact space X is that X is a principal G-bundle over the orbit space X/G. (See
Theorem 1.1.) As a special case, if G is nite then X is a covering space (not
necessarily connected) of X/G. As discussed after Theorem 1.1, this remains true
for noncompact G if the action is proper. It fails otherwise. One should compare
the action of Z on R by translation (free and proper) with the action of Z on the
circle S 1 generated by an irrational rotation (free but not proper). In the second
case, the orbit space is an uncountable set with the indiscrete topology, and the
quotient Borel space is not countably separated. However, free actions of this type
are very important. Here, for example, the crossed product is the well known
irrational rotation algebra.
There may be nearly as many versions of properness of actions on C*-algebras
as there are of freeness of actions on C*-algebras, but the subject has been less well
studied. We refer to the work of Rieel. (For example, see [66].) In this survey, we
simply avoid the issue. This is not to say that properness of actions on C*-algebras
is not important. Rather, it is a subject for a dierent paper. In most situations
in this paper in which the issue arises, we will consider only the analog of freeness
without properness.
Returning to freeness, many of our examples will involve simple C*-algebras,
since much of what has been done has involved simple C*-algebras. Indeed, for
one of our conditions, the tracial Rokhlin property, a satisfactory denition is so

FREENESS OF ACTIONS ON C*-ALGEBRAS

219
3

far known only in the simple case. For similar reasons, we have much less to say
about actions on purely innite C*-algebras than about actions on stably nite
C*-algebras. It seems possible (although proofs are still missing) that the dierences between some of our conditions disappear in the purely innite case. See the
discussion at the end of Section 4.
This paper is organized as follows. In Section 1, we recall a number of theorems
which characterize freeness of actions of nite or compact groups on compact or
locally compact spaces. Some involve C*-algebras, while others are purely in terms
of topology. These results suggest properties which might be expected of free actions
on C*-algebras. Several of them implicitly or explicitly motivate various denitions
of noncommutative freeness. In each of the remaining four sections, we discuss
a notion of noncommutative freeness, or a group of notions which seem to us to
be roughly comparable in strength (with one exception: some of the conditions
in Section 5 are much weaker than the others). See the further discussion at the
end of Section 1. The order of sections is roughly from strongest to weakest. The
comparisons are inexact partly because some versions of noncommutative freeness,
in their present form, are useful only for restricted classes of groups or C*-algebras,
and sometimes there are no interesting examples in the overlap. For example, we
dont know how to properly dene the tracial Rokhlin property for actions on C*algebras which are not simple, which makes it awkward to compare this property
with freeness of an action on a commutative C*-algebra.
As will become clear in the discussion, there are a number of directions in which
further work is needed.
We describe some standard notation. Throughout, groups will be at least
locally compact. All groups and spaces (except primitive ideal spaces and spaces
of irreducible representations of C*-algebras) will be Hausdor. For an action of a
group G on a locally compact space X, we let C (G, X) denote the transformation
group C*-algebra, and we let X/G denote the orbit space. For an action : G
Aut(A) of G on a C*-algebra A, written g  g , we denote the crossed product by
C (G, A, ). We further denote the xed point algebra
{a A : g (a) = a for all g G}
by A , or by AG if is understood. The action : G Aut(C0 (X)) coming from
an action of G on X is g (f )(x) = f (g 1 x). Note that in this case C0 (X)G can
be canonically identied with C0 (X/G). The restriction of : G Aut(A) to a
subgroup H G is |H , and the restriction to an invariant subalgebra B A is
() |B .
All ideals in C*-algebras are assumed closed and two sided. We will denote the
cyclic group Z/mZ by Zm . (No confusion with the m-adic integers should occur.)
For a Hilbert space H, we denote by L(H) and K(H) the algebras of bounded and
compact operators on H.
We would like to thank Dawn Archey, George Elliott, Akitaka Kishimoto, Hiroyuki Osaka, Cornel Pasnicu, Costel Peligrad, Marc Rieel, Masamichi Takesaki,
and Dana Williams for useful discussions and email correspondence.
1. The commutative case
In the main part of this section, we give some characterizations and properties
of free actions of nite (sometimes more general) groups on compact spaces. In

220
4

N. CHRISTOPHER PHILLIPS

some parts of the rest of this survey, we will concentrate on simple C*-algebras, so,
without proper interpretation, what one sees here may provide little guidance.
Theorem 1.1. Let G be a compact Lie group and let X be a locally compact
G-space. The action of G on X is free if and only if the map X X/G is the
projection map of a principal G-bundle.
A principal G-bundle is a locally trivial bundle with ber G, and where the
transition maps between trivializations are given by continuous maps to G, regarded
as acting on itself by translation.
This result is also true for actions of locally compact Lie groups which are free
and proper. See the theorem in Section 4.1 of [54]. The denition of properness
given there is dierent, but for locally compact X it is equivalent. See Condition (5)
in Theorem 1.2.9 of [54]; the notation is in the introduction of [54] and Denition
1.1.1 there.
Theorem 1.2. Let G be a nite group, and let X be a totally disconnected
compact G-space. The action of G on X is free if and only if X is equivariantly
homeomorphic to a G-space of the form G Y, where G acts on itself by translation
and acts trivially on Y.
Proof. We rst claim that for every x X, there is a compact open set
K X such that x K and the sets gK, for g G, are disjoint. To see this, for
each g G choose disjoint compact open sets Lg and Mg such that x Lg and
gx Mg . Then take

K=
(Lg g 1 Mg ).
gG\{1}

Since X is compact, we can now nd compact open sets K1 , K2 , . . . , Kn X


which cover X and such that, for each m, the sets gKm , for g G, are disjoint. Set


g(K1 K2 Km1 ) .
Lm = Km X \
gG

(This set may be empty.) One veries by induction


 on m that the sets gLj , for
g G and j = 1, 2, . . . , m, are disjoint and cover gG g(K1 K2 Km ). Set
Y = L1 L2 Ln . Then the sets gY, for g G, form a partition of X. The
conclusion follows.

Theorem 1.3. Let G be a compact group and let X be a locally compact Gspace. The action of G on X is free if and only if appropriate formulas (see Situation 2 of [65]) make a suitable completion of Cc (X) into a C0 (X/G)C (G, X)
Morita equivalence bimodule.
Proof. That freeness implies Morita equivalence is Situation 2 of [65]. (It
actually covers proper actions of locally compact but not necessarily compact
groups.) Both directions together follow from Proposition 7.1.12 and Theorem 7.2.6
of [58].

Theorem 1.4. Let G be a compact group and let X be a locally compact Gspace. The action of G on X is free if and only if for every g G \ {1} and every
g-invariant ideal I C(X), the action of g on C(X)/I is nontrivial.

FREENESS OF ACTIONS ON C*-ALGEBRAS

221
5

This is really just a restatement of the requirement that g have no xed points.
It is included for comparison with the conditions in Section 4.
Theorem 1.5. Let G be a compact group and let X be a locally compact Gspace. The action of G on X is free if and only if every ideal I C (G, X) has
the form C (G, U ) for some G-invariant open set U X.
Proof. Suppose the action of G on X is free. The conclusion follows (in fact,
in the more general case of a free and proper action of a locally compact group)
from Theorem 14 of [17] and its proof.
Suppose the action of G on X is not free. Our argument is very close to the
proof of Proposition 7.1.12 of [58]. Choose x X and g G\{1} such that gx = x.
Let H G be the subgroup given by H = {g G : gx = x}, and set S = Gx. Then
S is a closed subset of X which is equivariantly homeomorphic to G/H. So, using
Corollary 2.10 of [18] for the second isomorphism, C (G, X) has a quotient
C (G, S)
= C (G, G/H)
= K(L2 (G/H)) C (H).
Let : C (G, X) K(L2 (G/H)) C (H) be the composition of this isomorphism
with the quotient map.
Since H is a nontrivial compact group, C (H) is not simple. (For example,
consider the kernel of the map to C induced by the one dimensional
trivial repre

sentation.) Let J C (H) be a nontrivial ideal. Then 1 K(L2 (G/H)) J is
an ideal in C (G, X) which does not have the form C (G, U ) for any G-invariant
open set U X.


For the next characterization, recall the equivariant K-theory KG


(X) of a locally compact G-space X, introduced in [69]. (Also see Section 2.1 of [58].) It is a
module over the representation ring R(G) (see [68]), which can be thought of as the
equivariant K-theory of a point, or as the Grothendieck group of the abelian semigroup of equivalence classes of nite dimensional unitary representations of G, with
addition given by direct sum. The ring multiplication is tensor product. There is a
standard homomorphism R(G) Z which sends a representation to its dimension,
and its kernel is called the augmentation ideal and written I(G). (See the example
before Proposition 3.8 of [68].) We also need localization of rings and modules,
as discussed in Chapter 3 of [5]. Our notation follows part (1) of the example on
page 38 of [5].
The following two results are parts of Theorem 1.1.1 of [58]. They are essentially due to Atiyah and Segal (Proposition 4.3 of [6] and Proposition 4.1 of [69]).

Theorem 1.6 (Atiyah and Segal). Let G be a compact Lie group and let X be a
compact G-space. The action of G on X is free if and only if for every prime ideal
P in the representation ring R(G) which does not contain I(G), the localization

KG
(X)P is zero.
Theorem 1.7 (Atiyah and Segal). Let G be a compact Lie group and let X be
a compact G-space. The action of G on X is free if and only if the natural map

K (X/G) KG
(X) is an isomorphism.
Freeness is also related to C*-index theory. The C*-basic construction used in
the following result is found in Sections 2.1 and 2.2 of [73]; the two versions are the
same by Lemma 2.2.9 of [73].

222
6

N. CHRISTOPHER PHILLIPS

Theorem 1.8. Let G be a nite group, and let X be a compact G-space. Suppose the set of points x X with trivial stabilizer Gx is dense in X. Dene a
conditional expectation E : C(X) C(X)G by

1
f (gx).
E(f )(x) =
card(G)
gG

Then E has index-nite type in the sense of Watatani (see Denition 1.2.2 of [73],
and see Lemma 2.1.6 of [73] for the C*-algebra version) if and only if the action of
G on X is free. Moreover, in this case, the C*-basic construction gives an algebra
isomorphic to C (G, X).
Proof. See Propositions 2.8.1 and 2.8.2 of [73].

Some nontriviality condition on the action is necessary, since one must rule out
the trivial action and the action of G on its quotient G/H by a subgroup H.
Here is an example of preservation of structure associated with freeness.
Theorem 1.9. Let G be a nite group. If G acts freely on a topological manifold M, then M/G is a topological manifold. If M and the action are smooth, then
so is M/G.
The theorem has more force in the smooth case. For example, if G is a nite
cyclic group, acting on R2 by rotation, then R2 /G is a topological manifold, but is
not smooth in a neighborhood of the orbit of 0.
The situation for actions on noncommutative C*-algebras is much more complicated. There are at least six rough categories of conditions which, when properly
dened, at least approximately correspond to freeness in the commutative case. We
list them in descending order of strength.

Free action on the primitive ideal space.


The Rokhlin property and the closely related property of K-freeness.
The tracial Rokhlin property.
Outerness.
Hereditary saturation (full strong Connes spectrum), and the closely related property of full Connes spectrum.
Saturation.
At least outside the type I case, free action on the primitive ideal space seems
to be too strong a condition, as we hope to persuade you in this survey. One
section below is devoted to each of the others, except that we treat the last two
together. For each condition, we give denitions, say what it implies in the standard
examples discussed below, describe some applications, describe how it is related to
previously discussed conditions, and say something about permanence properties.
We also state some open problems.
We will discuss a number of examples in this survey, but we will use two
kinds of examples systematically. One kind is arbitrary actions on separable unital
type I C*-algebras. For them, most of our freeness conditions turn out to be
equivalent to freeness of the induced action on the primitive ideal space Prim(A)
of the algebra A. The other is product type actions on UHF algebras. We recall
these, and give convenient conventions, in the following denition. For simplicity,
we stick to actions of Z2 .

FREENESS OF ACTIONS ON C*-ALGEBRAS

223
7

Definition 1.10. For n Z>0 let d(n) and k(n) be integers with d(n) 2
and 0 k(n) 12 d(n). Choose projections pn , qn Md(n) such that pn + qn = 1
and rank(q
n ) = k(n). The associated product type action on the UHF algebra
A=
generated by the innite tensor
n=1 Md(n) is the action : Z2 Aut(A)

product automorphism of order 2 given by = n=1 Ad(pn qn ).
The conjugacy class of the action does not depend on the choice of the projections pn and qn . We need only consider 0 k(n) 12 d(n), since replacing any
particular k(n) by d(n) k(n) gives a conjugate action.
To give an idea of what the various conditions mean, let the notation be as in
Denition 1.10, and consider the following specic cases:
If d(n) = 2 and k(n) = 1 for all n, then has the Rokhlin property and
is K-theoretically free. See Examples 2.3 and 2.21.
If d(n) = 3 and k(n) = 1 for all n, then has the tracial Rokhlin property,
but does not have the Rokhlin property and is not K-theoretically free.
See Examples 2.21 and 3.3.
If d(n) = 2n and k(n) = 1 for all n, then is pointwise outer but does
not have the tracial Rokhlin property. See Examples 3.3 and 4.2.
There are no actions of this specic type which are hereditarily saturated
butnot pointwise
outer. However, the action of Z22 on M2 generated by

1 0
Ad 0 1 and Ad ( 01 10 ) is hereditarily saturated but not pointwise outer.
See Example 5.6.
If d(n) = 2 for all n, and k is given by k(n) = 0 for n 2 and k(1) = 1,
then is saturated but not hereditarily saturated. See Examples 5.3
and 5.5.
If d(n) = 2 and k(n) = 0 for all n, then is not saturated. See Example 5.3.
2. The Rokhlin property and K-theoretic freeness
We treat the Rokhlin property and K-theoretic freeness together because, in the
situations to which they apply well, they seem roughly comparable in strength. The
usefulness of both is limited to special classes of C*-algebras. The Rokhlin property
is not useful if there are too few projections, since no action of any nontrivial group
on a unital C*-algebra with no nontrivial projections can have the Rokhlin property.
Thus (Example 2.4 below), even if the nite group G acts freely on X, the action
of G on C(X) need not have the Rokhlin property.
The denition of K-theoretic freeness represents an attempt to turn Theorem 1.6 into a denition. The condition must be strengthened; see the discussion
before Denition 2.18 below. The usefulness of K-theoretic freeness depends on
the presence of nontrivial K-theory: the trivial action of any nite group on any
C*-algebra of the form O2 A satises the strongest possible form of K-theoretic
freeness. It is, of course, true that an action on a commutative unital C*-algebra is
K-theoretically free if and only if the action on the corresponding space is free. For
the product type action of Denition 1.10, K-theoretic freeness is equivalent to the
Rokhlin property, and this seems likely to be true for general product type actions
on UHF algebras, and perhaps more generally. (Example 2.5 limits how far this
idea can be taken.)
No known version of noncommutative freeness agrees both with K-theoretic
freeness in the presence of sucient K-theory and with the Rokhlin property on O2 .

224
8

N. CHRISTOPHER PHILLIPS

See Problem 2.28. However, for some applications one really needs the Rokhlin
property. See Example 2.26.
The Rokhlin property is as in Denition 3.1 of [25] (see below); we rst give
the equivalent form in Denition 1.1 of [59]. The property is, however, much older.
Early uses in C*-algebras (under a dierent name) can be found in [14], [21],
and [22]. The version for von Neumann algebras appeared even earlier, for example in [30]. It is a noncommutative generalization of the statement of the Rokhlin
Lemma in ergodic theory (for the case G = Z). The Rokhlin property can also be
considered to be modelled on Theorem 1.2. Note, though, that Condition (2) of the
denition is purely noncommutative in character, and is essential for the applications of the Rokhlin property. As we will see in Section 3, especially Example 3.12
and the following discussion, the Rokhlin property is quite rare.
Definition 2.1. Let A be a unital C*-algebra, and let : G Aut(A) be an
action of a nite group G on A. We say that has the Rokhlin property if for every
nite set S A and every > 0, there are mutually orthogonal projections eg A
for g G such that:
(1) g (eh ) egh < for all g, h G.
(2)
eg a aeg < for all g G and all a S.
(3)
gG eg = 1.
We call the (eg )gG a family of Rokhlin projections for , S, and .
Remark 2.2. The Rokhlin property can be neatly formulated in terms of central sequence algebras. Consider the C*-algebra of all bounded sequences in A
(bounded functions from Z>0 to A). It contains an ideal consisting of all sequences
vanishing at innity. Let A be the quotient. Let A be the relative commutant
in A of the embedded copy of A obtained as the image of the set of constant
sequences. The action : G Aut(A) induces an action : G Aut(A ).
(There is no continuity issue since G is discrete.) Denition 3.1
of [25] asks for
mutually orthogonal projections eg A for g G such that gG eg = 1 and
such that ( )g (eh ) = egh for all g, h G.
We make a few comments about what happens for more general groups. For
actions of Z, one does not consider families of projections indexed by Z (for which
the sum in Denition 2.1(3) would not make sense), but rather families indexed by
arbitrarily long nite intervals in Z. (The important points about intervals are that
they are Flner sets in the group and that they can tile the group.) Moreover, to
avoid K-theoretic obstructions, one must in general allow several orthogonal such
families indexed by intervals of dierent lengths. See the survey article [24] and
references there. For the application of the Rokhlin property to classication of
actions of Z, see the survey article [44]. For more general discrete groups, one
encounters further diculties with the choice of subsets of the group. The group
must certainly be amenable, and the results of [49] suggest that one may only be
able to require that the sum in Denition 2.1(3) be close to 1, necessarily in a
sense weaker than the norm topology. The resulting notion looks more like the
tracial Rokhlin property (Denition 3.1 below). In addition, if G is not discrete,
one must abandon projections. See [49] for actions of locally compact groups
on measure spaces, see [45] (especially the theorem in Section 6.1) for actions of
countable amenable groups on von Neumann algebras, see [42] for actions of Zd on
C*-algebras, see [35] for actions of R on C*-algebras, and see [23] for actions of

FREENESS OF ACTIONS ON C*-ALGEBRAS

225
9

compact groups on C*-algebras. Much of this, including more on the von Neumann
algebra versions (which came rst), is discussed in [24].
Example 2.3. Let : Z2 Aut(A) be as in Denition 1.10. Then has the
Rokhlin property if and only if k(n) = 12 d(n) for innitely many n Z>0 . This is
part of Proposition 2.4 of [61], but was known long before; see Lemma 1.6.1 of [14].
By combining tensor factors, we can write any such action as a product type
action so that k(n) = 12 d(n) for all n Z>0 .
It is trivial that if : G Aut(C(X)) has the Rokhlin property, then the action
of G on X is free. As we will see in Corollary 4.17 below, the Rokhlin property for
an action of a nite group on a unital type I C*-algebra A implies freeness of the
action on Prim(A). (One can also give a direct proof.) For totally disconnected X,
the converse is Theorem 1.2. In general, however, the converse is false. This is our
rst encounter with one of the main defects of the Rokhlin property, namely that
it is appropriate only for C*-algebras with a sucient supply of projections.
Example 2.4. Let X be the circle S 1 and let G = Z2 , with nontrivial element g.
Then g = generates a free action of G on X. The corresponding action of
G on C(X) does not have the Rokhlin property, because C(X) has no nontrivial
projections. Specically, if < 1 it is not possible to nd projections e1 , eg C(S 1 )
such that e1 + eg = 1 and g (e1 ) eg < .
Any free action on a connected compact manifold gives the same outcome.
Other easy examples are the action of Zn on the circle S 1 by rotation, and the
action of Z2 on the m-sphere S m via x  x.
However, there seem to be more subtle issues with the Rokhlin property than
merely lack of suciently many projections.
Example 2.5. In Example 4.1 of [61], there is an action of Z2 on a unital
C*-algebra A = lim An , obtained as the direct limit of actions on the An , such

that An
= C(S 2k , Ms(n) ) for suitable integers s(0) < s(1) < and a xed integer
k 1, and such that the action on An is the tensor product of the action on C(S 2k )
induced by x  x with an inner action on Ms(n) . These actions surely deserve
to be considered free. Therefore the direct limit action should also deserve to be
considered free. The direct limit algebra A is simple and AF, and in particular
is well supplied with projections by any reasonable standard. However, the direct
limit action does not have the Rokhlin property. In fact, its crossed product is
not AF, because the K0 -group of the crossed product has torsion isomorphic to Z2k .
See Proposition 4.2 of [61] for details. So the Rokhlin property would contradict
Theorem 2.6(2a) below.
We address this issue again in Problem 2.28 below.
Crossed products by actions with the Rokhlin property preserve many properties of C*-algebras. These can be thought of as being related to Theorem 1.9.
Theorem 2.6. Crossed products by actions of nite groups with the Rokhlin
property preserve the following classes of C*-algebras.
(1) Simple unital C*-algebras. (See Proposition 4.16 and Theorem 4.6 below.)
(2) Various classes of unital but not necessarily simple countable direct limit
C*-algebras using semiprojective building blocks, and in which the maps
of the direct system need not be injective:

226
10

N. CHRISTOPHER PHILLIPS

(a)
(b)
(c)
(d)

(3)
(4)

(5)
(6)
(7)
(8)

(9)

(10)

(11)

AF algebras. (See Theorem 2.2 of [59].)


AI algebras. (See Corollary 3.6(1) of [52].)
AT algebras. (See Corollary 3.6(2) of [52].)
Unital direct limits of one dimensional noncommutative CW complexes. (See Corollary 3.6(4) of [52].)
(e) Unital direct limits of Toeplitz algebras, a special case of the sort studied in [40] except not necessarily of real rank zero. (See Example 2.10
and Theorem 3.5 of [52].)
(f) Various other classes; see Section 2 and Theorem 3.5 of [52] for
details.
Simple unital AH algebras with slow dimension growth and real rank zero.
(See Theorem 3.10 of [52].)
D-absorbing separable unital C*-algebras for a strongly self-absorbing C*algebra D. (See Theorem 1.1(1) and Corollary 3.4(i) of [23]. See [23] for
the denition of a strongly self-absorbing C*-algebra.)
Unital C*-algebras with real rank zero. (See Proposition 4.1(1) of [52].)
Unital C*-algebras with stable rank one. (See Proposition 4.1(2) of [52].)
Separable nuclear unital C*-algebras whose quotients all satisfy the Universal Coecient Theorem. (See Proposition 3.7 of [52].)
Unital Kirchberg algebras (simple separable nuclear purely innite C*algebras) which satisfy the Universal Coecient Theorem. (See Corollary 3.11 of [52].)
Separable unital approximately divisible C*-algebras. (This is in Corollary 3.4(2) of [23], which also covers actions of compact groups; also see
Proposition 4.5 of [52].)
Unital C*-algebras with the ideal property and unital C*-algebras with the
projection property. (See [55]; also see [55] for the denitions of these
properties).
Simple unital C*-algebras whose K-theory:
(a) Is torsion free.
(b) Is a torsion group.
(c) Is zero.
(See the discussion below.)

Further classes will appear in [55].


Many of the parts of Theorem 2.6 are special for actions of nite (or compact)
groups. For example, parts (2a) and (2b) fail for G = Z, because the crossed product
will have nontrivial K1 -group. We refer to Section 2.2 of [24] and the references
there for many positive results for actions of Z with the Rokhlin property.
The main ingredient for Theorem 2.6(11) is the following result of Izumi. Using
it, one can derive many other statements similar to the ones given.
Theorem 2.7 (Theorem 3.13 of [25]). Let A be a simple unital C*-algebra, let
G be a nite group, and let : G Aut(A) be an action with the Rokhlin property.
Then the inclusion AG A induces an injective map K (AG ) K (A).
The other ingredient is that the Rokhlin property implies K (C (G, A, ))
=
K (AG ). Proposition 5.15 below shows that the Rokhlin property implies hereditary saturation (Denition 5.4 below), and in particular implies saturation (Denition 5.2 below). Now combine Proposition 7.1.8 and Theorem 2.6.1 of [58].

FREENESS OF ACTIONS ON C*-ALGEBRAS

227
11

For some of the classes in Theorem 2.6, such as those in parts (3), (5), (6), (7),
and (10), it is expected that weaker conditions than the Rokhlin property should
suce. This is certainly true for part (1). See Theorem 4.6 and Corollary 5.12
below.
Theorem 2.6(4) holds for actions of second countable compact groups with
the Rokhlin property and, provided the strongly selfabsorbing C*-algebra is K1 injective, for actions of Z and R with the Rokhlin property. See Theorem 1.1(1)
of [23].
Theorem 2.6(11) brings back the point that the Rokhlin property is slightly
too strong. For example, if G is a nite group acting freely on a compact space X
such that K (X) is torsion free, it does not follow that K (C (G, X)) is torsion
free. The action of Z2 on S 2 generated by x  x is a counterexample. (The
K-theory of the crossed product is the same as for the real projective space RP 2 .)
Example 2.5 shows that this phenomenon can even occur for an action on a simple
AF algebra.
We also have the following result. It appears as Theorem 1.25 of [70], but had
been proved earlier as part of other work which is still in progress [55].
Theorem 2.8. Let : G Aut(A) be an action of a nite group on a unital
C*-algebra which has the Rokhlin property. Then every ideal J C (G, A, ) has
the form C (G, I, () |I ) for some G-invariant ideal I A.
For an application of these structural results, see [62]. This paper uses a crossed
product construction (following Connes von Neumann algebra construction [10])
to produce a simple separable exact C*-algebra A which is not isomorphic to its
opposite algebra, and which has a number of nice properties. The action has the
Rokhlin property, and this fact is crucial in the computation of K (A) and the
verication of a number of the properties of A.
The use of the Rokhlin property to obtain structural results for crossed products
by nite groups seems to be very recent (although it has a longer history for actions
of Z). Indeed, Theorem 2.6(2a) (crossed products by nite group actions with the
Rokhlin property preserve the class of AF algebras) could easily have been proved
long ago. It was actually proved only after the analogous statement for the tracial
Rokhlin property and C*-algebras with tracial rank zero, Theorem 3.7 (although
both theorems appear in the same paper). The main reason is that the tracial
analog seems to be much more useful for structure theory. See Section 3.
The main application so far of the Rokhlin property has been to the structure
and classication of group actions. This project began in von Neumann algebras.
As just one example, Jones proved (Corollary 5.3.7 of [30]) that every pointwise
outer action of a nite group G on the hypernite factor R of type II1 is conjugate to a particular model action of G. That is, up to conjugacy, there is only one
pointwise outer action of G on R. An essential step in the proof is showing that
pointwise outerness implies the von Neumann algebra analog of the Rokhlin property. Ocneanu [45] extended the result to actions of general countable amenable
groups. Classication of actions on C*-algebras with the Rokhlin property is the
main thrust of the papers [14], [21], and [22], and the more recent papers [25]
and [26].
We state four theorems from [26]. The rst two require the following model
action.

228
12

N. CHRISTOPHER PHILLIPS

Definition 2.9. Let G be a nite group, and let n = card(G). Let u : G Mn


be the image of the regular representation of G under some isomorphism L(l2 (G))
Mn . Let
g (a) = u(g)au(g) . Let DG be the UHF algebra
: G Aut(Mn ) be
G
G
D =
m=1 Mn , and let : G Aut(DG ) be the product type action g =
G
m=1 g .
For G = Z2 , the action G is obtained as in Denition 1.10 by taking d(n) = 2
and k(n) = 1 for all n.
Theorem 2.10 (Theorem 3.4 of [26]). Let A be a unital Kirchberg algebra (simple, separable, purely innite, and nuclear) which satises the Universal Coecient
Theorem. Let G be a nite group and let : G Aut(A) have the Rokhlin property. Suppose (g ) is the identity on K (A) for all g G. Then is conjugate to
the action g  idA G
g on A DG .
Theorem 2.11 (Theorem 3.5 of [26]). Theorem 2.10 remains true if instead
A is a simple separable unital C*-algebra with tracial rank zero which satises the
Universal Coecient Theorem.
Theorem 2.12 (Theorem 4.2 of [26]). Let A be a unital Kirchberg algebra
which satises the Universal Coecient Theorem. Let G be a nite group. Let
, : G Aut(A) have the Rokhlin property. Suppose (g ) = (g ) on K (A) for
all g G. Then there exists Aut(A) such that g = g 1 for all g G,
and such that is the identity on K (A).
Theorem 2.13 (Theorem 4.3 of [26]). Theorem 2.12 remains true if instead
A is a simple separable unital C*-algebra with tracial rank zero which satises the
Universal Coecient Theorem.
See [44] for results on classication of actions of Z with the Rokhlin property.
We should point out that some work has been done on classication of nite
and compact group actions without the Rokhlin property, assuming instead that
the action is compatible with a direct limit realization of the algebra, and also
that the actions on the algebras in the direct system have special forms. See, for
example, [19], [20], [34], and [71].
The Rokhlin property enjoys the following permanence properties. Proposition 2.14(1) is the analog of the fact that the restriction of a free action to a closed
subset of the space is still free, (2) is the analog of the fact that the restriction of
a free action to a subgroup is still free, (3) is the analog of the fact that a diagonal
action on a product is free if one of the factors is free, and (4) is the analog of the
fact that an equivariant inverse limit of free actions is free.
Proposition 2.14. Let A be a unital C*-algebra, let G be a nite group, and
let : G Aut(A) have the Rokhlin property. Then:
(1) If I A is a -invariant ideal, then the induced action of G on A/I
has the Rokhlin property.
(2) If H G is a subgroup, then |H has the Rokhlin property.
(3) If : G Aut(B) is any action of G on a unital C*-algebra, then the
tensor product action g  g g of G on Aut(A B), for any C* tensor
product on which it is dened, has the Rokhlin property.
In addition:

FREENESS OF ACTIONS ON C*-ALGEBRAS

229
13

(4) If A = lim An is a direct limit of a direct system of C*-algebras with

unital maps, and : G Aut(A) is an action obtained as the direct limit


of actions (n) : G Aut(An ), such that (n) has the Rokhlin property
for all n, then has the Rokhlin property.
Proof. For (1), let : A A/I be the quotient map. Let S A/I be nite
and let > 0. Choose a nite set T A such that (T ) contains S. Let (eg )gG
be a family of Rokhlin projections for , T, and . Then ((eg ))gG is a family of
Rokhlin projections for , S, and .
For (2), set n = card(G/H). Let S A be nite and let > 0. Choose a
system C of right coset representatives of H in G. Let (fg )gG be a family of
Rokhlin projections for , S, and /n. Then use the projections eh = cC fhc for
h H.
For (3), one rst checks that in Denition 2.1 it suces to x a subset R A
which generates A as a C*-algebra, and verify the condition of the denition only
for nite subsets S R. In this case, we take
R = {a b : a A and b B}.
We can further restrict to nite subsets of the form
Q = {a b : a S and b T },
for nite sets S A and T B. Set M = supbT b . Let (eg )gG be a family of
Rokhlin projections for , S, and /M. Then (eg 1)gG is a family of Rokhlin
projections for , Q, and .
To prove (4), rst use (1) to replace each An with its quotient by the kernel

of the map An A. Thus we may assume that
A=
n=0 An . As in the previous

part, we restrict to a generating set, here R = n=0 An . So let S R be nite and
let > 0. Choose n such that S An . Then a family of Rokhlin projections for
(n) , S, and is a family of Rokhlin projections for , S, and .

We cant reasonably talk about the Rokhlin property passing to invariant ideals
without a denition of the Rokhlin property for nonunital C*-algebras. We know of
no denition in the literature, but it seems reasonable to simply take the projections
to be in the multiplier algebra. An action of G on a C*-algebra A always extends
to an action g  M ()g on the multiplier algebra M (A). In general, one only gets
continuity of g  M ()g (a) in the strict topology, but this is irrelevant for nite
groups.
Definition 2.15. Let A be a not necessarily unital C*-algebra, and let : G
Aut(A) be an action of a nite group G on A. We say that has the multiplier
Rokhlin property if for every nite set S A and every > 0, there are mutually
orthogonal projections eg M (A) for g G such that:
(1) M ()g (eh ) egh < for all g, h G.
(2)
eg a aeg < for all g G and all a S.
(3)
gG eg = 1.
Proposition 2.16. Let A be a unital C*-algebra, and let : G Aut(A) be
an action of a nite group G on A. Let J A be a G-invariant ideal. Suppose
has the Rokhlin property. Then () |J has the multiplier Rokhlin property.

230
14

N. CHRISTOPHER PHILLIPS

Proof. Let F J be a nite set, and let > 0. Choose a family (eg )gG of
Rokhlin projections for , S, and . Let : A M (J) be the standard homomorphism obtained from the fact that J is an ideal in A. Then ((eg ))gG satises the
condition of Denition 2.15 for the given choices of F and .

Unfortunately, and this is related to the issues with the Rokhlin property already discussed, there is no theorem for extensions.
Example 2.17. Let G = Z2 . Let : G Aut(C(S 1 )) be as in Example 2.4.
(Recall that this action comes from the action on S 1 generated by  .) Let
J = C0 (S 1 \ {1}), which is an ideal in C(S 1 ). It is easily checked that () |J has
the multiplier Rokhlin property and that the induced action on C(S 1 )/J
= C2
has the Rokhlin property. As in Example 2.4, however, itself does not have the
Rokhlin property.
We now turn to K-theoretic freeness. As Theorem 1.6 shows, K-theory gives
a neat condition for an action of a compact Lie group on a compact space to be
free. The book [58] mostly describes the eort to turn this result into a denition
of freeness of actions on C*-algebras. Also see the survey [57]. This idea is useless
if there is no K-theory. For example, the trivial action on any C*-algebra of the
form O2 A is K-theoretically free. Moreover, unlike for actions on C(X), for C*algebras the condition of Theorem 1.6 passes neither to quotients by invariant ideals
nor to subgroups, so it is necessary to build these features into the noncommutative denition. (Example 4.1.7 of [58] implies that the property does not pass to
quotients by invariant ideals, although it is stated in terms of the ideal rather than
the quotient. Example 2.22 below gives an action of Z4 on a UHF algebra which
satises the condition of Theorem 1.6, but such that the restriction of the action to
the subgroup Z2 Z4 does not; the restriction is in fact inner.) Subject to these
caveats, the denition permits some interesting results, such as for C*-algebras of
type I and for product type actions on UHF algebras. It covers  on S 1 ,
unlike the Rokhlin property.
Equivariant K-theory for an action of a compact group G on a C*-algebra A,
denoted KG (A), is as in Chapter 2 of [58]. Like the equivariant K-theory of a space,
as discussed before Theorem 1.6, it is a module over the representation ring R(G),
and for a prime ideal P R(G), the localization KG (A)P is dened. The following
is from Denitions 4.1.1, 4.2.1, and 4.2.4, and the discussion after Denition 4.2.1,
in [58]. The term locally discrete K-theory refers to the I(G)-adic topology on
KG (A); see discussion after Denition 4.1.1 in [58].
Definition 2.18. Let A be a C*-algebra, and let : G Aut(A) be an action
of a nite group G on A. We say that has locally discrete K-theory if for every
prime ideal P in the representation ring R(G) which does not contain I(G), the
localization KG (A)P is zero. We say that is K-free if for every invariant ideal
I A, the induced action on A/I has locally discrete K-theory. We say that is
totally K-free if for every subgroup H G, the restricted action |H is K-free.
Other related conditions, including ones involving equivariant KK-theory, and
many more results and examples than can be discussed here, can be found in [58].
Also see the survey [57].
The denition generalizes reasonably to actions of compact Lie groups, but not
to actions of noncompact groups.

FREENESS OF ACTIONS ON C*-ALGEBRAS

231
15

The denition behaves well for type I C*-algebras:


Theorem 2.19 (Theorems 8.1.4 and 8.2.5 of [58]). Let A be a separable type I
C*-algebra, and let : G Aut(A) be an action of a compact Lie group G on A.
Then is totally K-free if and only if the induced action of G on Prim(A) is free.
In fact, one direction holds for any C*-algebra; Example 2.4 shows that the
corresponding statement for the Rokhlin property fails.
Theorem 2.20 (Theorem 4.3.8 of [58]). Let A be a C*-algebra, and let : G
Aut(A) be an action of a nite group G on A. Suppose that the induced action of
G on Prim(A) is free. Then is totally K-free.
For our standard product type action, total K-freeness is equivalent to the
Rokhlin property:
Example 2.21. Let : Z2 Aut(A) be as in Denition 1.10. Then the following are equivalent:
(1) is totally K-free.
(2) has locally discrete K-theory.
(3) I(Z2 )K0Z2 (A) = 0.
(4) k(n) = 12 d(n) for innitely many n Z>0 .
(5) has the Rokhlin property.
This follows from Theorems 9.2.4 and 9.2.6 of [58] and Proposition 2.4 of [61].
(These results also give a number of other equivalent conditions.)
In this example, the equivalence of Condition (2) with the rest is slightly misleading. The action of Z2 Z2 on M2 in Example 5.6 below is K-free but not
totally K-free (see Example 4.2.3 of [58]), and certainly does not have the Rokhlin
property. This kind of thing can happen even for cyclic groups:
Example 2.22. Example 9.3.9 of [58] gives an action of Z4 on the 2 UHF algebra (see Remark 9.3.10 of [58] for the identication of the algebra) which is a
direct limit of actions on nite dimensional C*-algebras, is K-free, but is not totally
K-free. It is observed there that the automorphism corresponding to the order two
element of Z4 is inner. Therefore Proposition 4.16 below implies that this action
does not have the Rokhlin property.
The crossed product is simple, in fact, isomorphic to the 2 UHF algebra, by
Remark 9.3.10 of [58].
The Rokhlin property implies total K-freeness in complete generality; in fact,
it implies that I(G)KG (A) = 0.
Proposition 2.23. Let A be a unital C*-algebra, let G be a nite group, and
let : G Aut(A) have the Rokhlin property. Then I(G)KG (A) = 0.
Proof. We start with some reductions. First, replacing A by C(S 1 ) A with
G acting trivially on S 1 , and using Proposition 2.14(3) and K0G (C(S 1 ) A)
=
K0G (A) K1G (A), we see that it suces to show that I(G)K0G (A) = 0. Next,
by Corollary 2.4.5 of [58], we need only consider elements of K0G (A) which are
represented by invariant projections in L(W )A for some nite dimensional unitary
representation space W of G. Again using Proposition 2.14(3), we may replace A
by L(W ) A, and thus consider the class [p] of an invariant projection in A.

232
16

N. CHRISTOPHER PHILLIPS

Next, we need only consider elements of I(G) which span I(G), so it suces to
prove that if V is a nite dimensional unitary representation space of G, with
representation g  ug , and if V0 is the same space with the trivial action of G,
then ([V ] [V0 ])[p] = 0 in K0 (G). Now let , 0 : A L(V0 V ) A be the
maps dened by 0 (a) = (1V0 0) a and (a) = (0 1V ) a. By Remark 2.4.6
of [58], it suces to nd a G-invariant element y L(V0 V ) A such that
y y = 0 (p) and yy = (p). Note that the action of G on L(V0 V ) A is given
by g  g = Ad(1 ug ) g .
+ 2n). Let (eg )gG be a family of Rokhlin
Set n = card(G) and = 1/(n2
projections for , {p}, and . Set w0 = hG uh eh , which is a unitary in L(V )A.
Since V0 = V as vector spaces, we can identify L(V0 V ) A with M2 (L(V ) A),
and dene w L(V0 V ) A by


0 w0
.
w=
w0 0
In matrix form, we have



0
1V p 0
0 w0

w0 (p)w =
0
0
w0 0
w0

w0
0

0
0

0
w0 (1V p)w0


.

Using orthogonality of the projections eg , we have


 


w0 (1V p)w0 1V p = w0 (1V p)w0 w0 w0 (1V p)



ugh1 eg peh ugh1 eg eh p

g,hG

peh eh p < n2 .

hG


Thus, w0 (p)w (p) < n . With xg =
hG ugh g (eh ), a calculation
shows that


0 xg
.
g (w) =
xg 0
Therefore

g (w) w = xg w0
g (eh ) egh < n.

hG

It follows that for g, h G, we have




(p)g (w)0 (p)h (w) (p) (p)
g (w) w + w0 (p)w (p) + h (w) w < (n2 + 2n) = 1.
Similarly,



0 (p)g (w) (p)h (w)0 (p) 0 (p) < 1.

Now set
b=

(p)g (w)0 (p),


n
gG

which is G-invariant. Using the estimates in the previous paragraph, we get


b b 0 (p) < 1 and

bb (p) < 1.


Therefore, with functional calculus evaluated in 0 (p) L(V0 V ) A 0 (p), the
G-invariant element y = b(b b)1/2 satises y y = 0 (p) and yy = (p).


FREENESS OF ACTIONS ON C*-ALGEBRAS

233
17

Corollary 2.24. Let A be a unital C*-algebra, let G be a nite group, and let
: G Aut(A) have the Rokhlin property. Then is totally K-free.
Proof. By Lemma 4.1.4 of [58], it suces to prove that for every subgroup
H G and every H-invariant ideal I A, the action of H on A/I induced
by |H has locally discrete K-theory. Now has the Rokhlin property by parts
(1) and (2) of Proposition 2.14. So I(H)KH (A/I) = 0 by Proposition 2.23. The
conclusion now follows from Proposition 4.1.3 of [58].

We return to the relationship between (total) K-freeness and the Rokhlin property after discussing permanence properties.
Proposition 2.25. Let A be a unital C*-algebra, let G be a nite group, and
let : G Aut(A) be an action of G on A.
(1) If I A is a -invariant ideal, then is totally K-free if and only if () |I
and the induced action of G on A/I are both totally K-free.
(2) If is totally K-free, and if H G is a subgroup, then |H is totally
K-free.
(3) If A = lim An is a direct limit of C*-algebras, and : G Aut(A) is an

action obtained as the direct limit of actions (n) : G Aut(An ), such


that (n) is totally K-free for all n, then is totally K-free.
Proof. Part (1) is built into the denition; see Proposition 4.2.6 of [58].
Part (2) is also built into the denition.
Part (3) was overlooked in [58]. For locally discrete K-theory, it is Lemma
4.2.14 of [58]. Let n : An A be the maps obtained from the direct limit realization of A. Let H G be a subgroup, and let J A be an H-invariant ideal.
Set Jn = 1
is an H-invariant ideal in An . Then J = lim Jn . For
n (J), which

 (n)
each n, the restriction |H |Jn has locally discrete K-theory. Therefore so does
|H |J .

One can also say a limited amount about actions on tensor products. See
Section 6.6 of [58].
We now come back to the relationship with the Rokhlin property.
Example 2.26. Let : Z2 A be the action of Example 2.5. The C*algebra A is a simple AF algebra. The construction of , together with Theorem 2.19 and Proposition 2.25(3), imply that is totally K-free. However, we saw
in Example 2.5 that does not have the Rokhlin property and that C (Z2 , A, )
is not AF. In particular, the crossed product of a simple AF algebra by a totally
K-free action need not be AF. In fact, parts (2a), (2b), and (2c) of Theorem 2.6 all
fail with total K-freeness in place of the Rokhlin property.
It seems to us that the fault is again with the Rokhlin property. The conclusion
of Proposition 2.23 stronger than ought to hold for a version of noncommutative
freeness. Indeed, there is a free action of a nite group G on a compact metric
space X such that I(G)K (C(X)) = 0. For example, the actions of G = Z2 on S 2n
and S 2n+1 generated by x  x give
0
0
KG
(S 2n )
(S 2n+1 )
= KG
= R(G)/I(G)n+1 ,

234
18

N. CHRISTOPHER PHILLIPS

as can be seen from Corollary 2.7.5 and the discussion after Corollary 2.7.6 in [4].
It then also follows (using the proof of Proposition 4.2 of [61]) that suitable choices
for A in Example 2.5 give I(G)K0G (A) = 0.
Problem 2.27. Suppose that A is a unital AF algebra, G is a nite group,
and : G Aut(A) is a totally K-free action which is a direct limit of actions on
nite dimensional C*-algebras. Does it follow that has the Rokhlin property?
One might want to assume that A is simple, or even that is a product type
action. For product type actions of Z2 , this is contained in Example 2.21.
It seems that the Rokhlin property and (total) K-freeness attempt to detect,
not quite successfully, a strong version of freeness of actions of nite groups on C*algebras, something which, to borrow a suggestion from Claude Schochet, might
be called a noncommutative covering space. The Rokhlin property is too strong,
even apart from the existence of projections, as is shown by Example 2.5, while
conditions involving K-theory are too weak when there is no K-theory. The following problem thus seems interesting, even though it is not clear what applications it
might have.
Problem 2.28. Find a well behaved version of freeness of nite group actions
on unital C*-algebras which agrees with total K-freeness for actions on AF algebras
and type I C*-algebras, and agrees with the Rokhlin property for actions on the
Cuntz algebra O2 .
One would hope for the following:
(1) The condition should pass to invariant ideals, to quotients by invariant
ideals, and to subgroups. It should also be preserved under extensions.
(2) The condition should be preserved when taking tensor products with arbitrary actions (with an arbitrary tensor norm such that the action extends
to the tensor product).
(3) The condition should be equivalent to total K-freeness for direct limit
actions on AH algebras.
(4) The condition should imply total K-freeness for general unital C*-algebras.
(5) The Rokhlin property should imply the condition for general unital C*algebras.
(6) For type I C*-algebras, the condition should be equivalent to free action
on the primitive ideal space.
(7) The condition should imply strong pointwise outerness (Denition 4.11
below) for arbitrary unital C*-algebras.
Example 2.26 shows that the freeness condition we are asking for should not
imply the Rokhlin property for actions on UHF algebras. It is not actually clear
that the right condition should imply the Rokhlin property for actions on O2 . As
evidence that a dierence between behavior on UHF algebras and on O2 should be
expected, consider the tensor ip A , the action of Z2 on A min A generated by
a b  b a. If A is a UHF algebra, then A never has the Rokhlin property ([53])
and is never K-free. (These statements follow easily from Example 2.21.) However,
O2 does have the Rokhlin property, by Example 5.2 of [25].

FREENESS OF ACTIONS ON C*-ALGEBRAS

235
19

3. The tracial Rokhlin property and outerness in factor representations


The tracial Rokhlin property is a weakening of the Rokhlin property, and which
is much more common. Unfortunately, for now we only know the right denition
of the tracial Rokhlin property for rather restricted classes of C*-algebras.
In retrospect, the tracial Rokhlin property could be motivated as follows. For
simple C*-algebras, one popular version of freeness of an action : G Aut(A)
is the requirement that g be outer for all g G \ {1}. (This condition is called
pointwise outerness in Denition 4.1, and it and its variants are the subject of Section 4.) Let R be the hypernite factor of type II1 . Then pointwise outer actions
of nite groups satisfy the von Neumann algebra analog of the Rokhlin property.
(Lemma 5.2.1 of [30] implies this statement.) One might then ask that an action
: G Aut(A) of a nite group G on a simple separable innite dimensional unital
C*-algebra A, with a unique tracial state , have the property that, in the weak
closure of the Gelfand-Naimark-Segal representation associated with (which is isomorphic to R), the action becomes outer. Under good conditions (see Theorem 3.5
below), this requirement is equivalent to the tracial Rokhlin property.
The actual motivation for the tracial Rokhlin property was, however, rather
dierent. It was introduced for the purpose of proving classication theorems for
crossed products. One should observe that the denition below is, very roughly,
related to the Rokhlin property in the same way that Lins notion of a tracially
AF C*-algebra is related to that of an AF algebra. (Tracially AF C*-algebras are
as in Denition 2.1 of [37]. The condition is equivalent to tracial rank zero as in
Denition 2.1 of [38]; the equivalence is Theorem 7.1(a) of [38].)
The usefulness of the tracial Rokhlin property comes from the combination
of two factors: it implies strong structural results for crossed products, and it is
common while the Rokhlin property is rare. In particular, the tracial Rokhlin
property played a key role in the solution of three open problems on the structure
of crossed products. See Theorems 3.14, 3.15, and 3.16 below. As we will see, the
actions involved do not have the Rokhlin property, while the next weaker freeness
condition, pointwise outerness (Denition 4.1), is not strong enough to make the
arguments work.
Definition 3.1. Let A be an innite dimensional simple unital C*-algebra,
and let : G Aut(A) be an action of a nite group G on A. We say that
has the tracial Rokhlin property if for every nite set F A, every > 0, and
every positive element x A with x = 1, there are nonzero mutually orthogonal
projections eg A for g G such that:
(1) g (eh ) egh < for all g, h G.
(2) eg a aeg
< for all g G and all a F.
(3) With e = gG eg , the projection 1e is Murray-von Neumann equivalent
to a projection in the hereditary subalgebra of A generated by x.
(4) With e as in (3), we have exe > 1 .
When A is nite, the last condition is redundant. (See Lemma 1.16 of [59].)
However, without it, the trivial action on O2 would have the tracial Rokhlin property. (It is, however, not clear that this condition is really the right extra condition
to impose.) Without the requirement that the algebra be innite dimensional, the

236
20

N. CHRISTOPHER PHILLIPS

trivial action on C would have the tracial Rokhlin property (except for the condition (4)), for the rather silly reason that the hereditary subalgebra in Condition (3)
cant be small.
Problem 3.2. Is there a reasonable formulation of the tracial Rokhlin property
in terms of central sequence algebras? We would like something analogous to the
formulation of the Rokhlin property given in Remark 2.2.
As for the Rokhlin property, the tracial Rokhlin property is only useful when
the algebra has a sucient supply of projections. The denition is only given for
simple C*-algebras, because we dont know the proper formulation of Condition (3)
without simplicity. We discuss these issues further below.
A version of this denition for actions of Z was given in [50]. The analog of
Condition (4) was omitted, and the algebra was required to be stably nite. A
slightly dierent version for Z, called the tracial cyclic Rokhlin property, appears
in Denition 2.4 of [39].
Since we require algebras with actions with the tracial Rokhlin property to
be simple, unital, and innite dimensional, they cant be type I. Thus, one of our
standard examples is irrelevant. For product type actions of Z2 , we have:
Example 3.3. Let : Z2 Aut(A) be as in Denition 1.10. Then the following
are equivalent:
(1) has the tracial Rokhlin property.
(2) If  is the action induced by on the type II1 factor obtained from the
trace via the Gelfand-Naimark Segal construction, then  is outer.
(3) For all N, we have


d(n) 2k(n)
= 0.
d(n)
n=N

(4) C (Z2 , A, ) has a unique tracial state.


See Proposition 2.5 of [61], where additional equivalent conditions are given.
In particular, by comparison with Examples 2.3 and 2.21, the choices d(n) = 3
and k(n) = 1 for all n give an action of Z2 which has the tracial Rokhlin property
but does not have the Rokhlin property and is not totally K-free.
On the other hand, the Rokhlin property implies the tracial Rokhlin property.
(This is trivially true for actions on all C*-algebras on which the tracial Rokhlin
property is dened.) Also, for actions as in Denition 1.10, locally discrete Ktheory implies the tracial Rokhlin property. This last statement is misleading,
since Example 2.22 and Proposition 4.3 below show that a K-free action on a
UHF algebra need not have the tracial Rokhlin property.
Problem 3.4. Let A be a simple separable unital tracially AF C*-algebra, let
G be a nite group, and let : G Aut(A) be totally K-free. Does it follow that
has the tracial Rokhlin property?
In particular, what happens for actions on simple unital AF algebras?
We now give the result promised in our initial discussion.
Theorem 3.5 (Theorem 5.5 of [11]). Let A be a simple separable unital C*algebra with tracial rank zero, and suppose that A has a unique tracial state . Let
: A L(H ) be the Gelfand-Naimark-Segal representation associated with . Let

FREENESS OF ACTIONS ON C*-ALGEBRAS

237
21

G be a nite group, and let : G Aut(A) be an action of G on A. Then has


the tracial Rokhlin property if and only if g is an outer automorphism of (A)
for every g G \ {1}.
The corresponding statement for actions of Z is also true (Theorem 2.18 of [51]).
Problem 3.6. Is there a related characterization of the tracial Rokhlin property for actions on simple separable unital C*-algebras with tracial rank zero which
have more than one tracial state?
Crossed products by actions with the tracial Rokhlin property cannot be expected to be as well behaved as those by actions with the Rokhlin property. Indeed,
Example 3.12(5) below shows that they do not preserve AF algebras or AI algebras, and Example 3.12(6) shows that they do not preserve AT algebras. The
tracial Rokhlin property does imply pointwise outerness (see Proposition 4.3 below), so that the crossed product of a simple unital C*-algebra by an action of a
nite group with the tracial Rokhlin property is again simple (Corollary 1.6 of [59]).
But the tracial Rokhlin property is much stronger than pointwise outerness. The
following theorems give classes of C*-algebras which are closed under formation
of crossed products by actions of nite groups with the tracial Rokhlin property.
Example 3.10 below shows that they do not hold for pointwise outer actions.
Theorem 3.7 (Theorem 2.6 of [59]). Let A be an innite dimensional simple
separable unital C*-algebra with tracial rank zero. Let : G Aut(A) be an action
of a nite group G on A which has the tracial Rokhlin property. Then C (G, A, )
has tracial rank zero.
It is shown in [39] that if A is an innite dimensional simple separable unital
C*-algebra with tracial rank zero and : Z Aut(A) is an action with the tracial
Rokhlin property and satisfying extra conditions (which hold in many interesting
examples), then C (Z, A, ) has tracial rank zero.
Theorem 3.8 ([53]). Let A be an innite dimensional simple separable unital
C*-algebra with tracial rank at most n, in the sense of Denition 2.1 of [38]. Let
: G Aut(A) be an action of a nite group G on A which has the tracial Rokhlin
property. Then C (G, A, ) has tracial rank at most n.
The following theorem combines several results from [2]. The analog for actions
of Z is in [50].
Theorem 3.9 ([2]). Let A be a stably nite innite dimensional simple separable unital C*-algebra with real rank zero and such that the order on projections over A is determined by traces. (See [2] for the denition of this condition.)
Let : G Aut(A) be an action of a nite group G on A which has the tracial
Rokhlin property. Then C (G, A, ) has real rank zero and the order on projections
over C (G, A, ) is determined by traces. If moreover A has stable rank one, then
so does C (G, A, ).
Example 3.10. Example 9 of [13] gives an example of a pointwise outer action (in the sense of Denition 4.1 below) of Z2 on a simple unital AF algebra A
such that C (Z2 , A, ) does not have real rank zero. This example shows that
Theorems 3.7, 3.8, and 3.9 fail for general outer actions.

238
22

N. CHRISTOPHER PHILLIPS

Example 3.11. The action : Z2 Aut(A) of Example 2.5 satises the tracial
Rokhlin property, by Proposition 4.2(3) of [61]. Since A is a simple AF algebra
and K0 (C (Z2 , A, )) has torsion, it follows that crossed products by actions of
nite groups with the tracial Rokhlin property do not preserve any of the classes
of AF algebras, AI algebras, or AT algebras.
In addition, if : G Aut(A) has the tracial Rokhlin property, and A is simple,
stably nite, and innite dimensional, then the restriction map from tracial states
on C (G, A, ) to -invariant tracial states on A is bijective. (See Proposition 5.7
of [11].) This is also false for general outer actions. (See Examples 3.3 and 4.2.)
The big advantage of the tracial Rokhlin property is that it is common (at least
on simple C*-algebras with many projections), while the Rokhlin property is rare.
Example 3.12. The following actions have the tracial Rokhlin property but not
the Rokhlin property. The citations are for proofs that the actions in question have
the tracial Rokhlin property; failure of the Rokhlin property is discussed afterwards.
(1) The action of Zn on a simple higher dimensional noncommutative torus
which multiplies one of the standard generators by exp(2i/n). (This is
Proposition 2.10 of [60].)
(2) The ip action of Z2 on a simple higher dimensional noncommutative
torus. (See Corollary 5.12 of [11].)
(3) The standard actions of Z3 , Z4 , and Z6 on an irrational rotation algebra.
(See Corollary 5.11 of [11].)
(4) For an arbitrary UHF algebra, many product type actions of Z2 . (See
Example 3.3.)

(5) Blackadars example [7] of an action of Z2 on
n=1 M2 such that the
crossed product is not AF. (See Proposition 3.4 of [61].)
(6) The actions of Example 2.5. (See Proposition 4.2 of [61].)
(7) Actions : Z2 Aut(A) similar to those of Example 2.5, as in Proposition 4.6 of [61]. Here, A is a simple AF algebra and K1 (C (Z2 , A, )) = 0.
(8) The tensor ip on A A for many stably nite simple approximately
divisible C*-algebras A. (See [53].)
As we will see, in many of these cases, in particular, in (1), (2), (3), (4) for odd
UHF algebras, (6), and many cases of (8), there does not exist any action of the
group on the C*-algebra which has the Rokhlin property.
There is one obvious obstruction to the Rokhlin property. Let A be a unital
C*-algebra. Let n Z>0 . Suppose, for simplicity, that the ordered group K0 (A)
has no nontrivial automorphisms which x [1], and that [1] is not of the form
n for any K0 (A). Then no group G with card(G) = n admits any action
: G Aut(A) with the Rokhlin property. Simply take < 1 in Denition 2.1 to
get g (e1 ) Murray-von Neumann equivalent to eg for all g G, and use triviality
of (g ) to get [g (e1 )] = [e1 ] in K0 (A). So one would get n[e1 ] = [1].
It is now immediate that no action of any nontrivial nite group on any irrational rotation algebra can have the Rokhlin property. Similarly, for any odd
m 3, no action of Z2 on the m UHF algebra can have the Rokhlin property.
For the same reason, no action of Z2 on any odd Cuntz algebra or on O has the
Rokhlin property.

FREENESS OF ACTIONS ON C*-ALGEBRAS

239
23

In fact, existence of an action of G with the Rokhlin property implies much


stronger restrictions on the K-theory (Theorem 3.2 of [26]), namely vanishing cohomology as Z[G]-modules for K (A) and certain subgroups. The following result
is a special case.
Proposition 3.13 ([26]). Let n Z>0 , let A be a unital C*-algebra, and let
: Zn Aut(A) be an action with the Rokhlin property which is trivial on K (A).
Then K (A) is uniquely n-divisible.
Proof. See the discussion after the proof of Theorem 3.2 of [26]

Proposition 3.13 rules out the actions in parts (2) and (6) of Example 3.12. It
also shows that if a UHF algebra admits an action of Z2 with the Rokhlin property,
then it must tensorially absorb the 2 UHF algebra. Even on UHF algebras which
satisfy this condition, Example 2.3 shows that most product type actions of Z2
do not have the Rokhlin property.
Theorem 2.7 contains a K-theoretic restriction of a dierent kind.
By contrast, there is no apparent K-theoretic obstruction to the Rokhlin property for actions of Z, and there is no action of Z which is known to have the tracial
Rokhlin property but known not to have the Rokhlin property.
The actions in parts (1), (3), and (2) of Example 3.12 play a key role, via
Theorem 3.7, in the proofs of the following recent solutions to open problems on
the structure of certain crossed product C*-algebras. In none of these proofs is
outerness of the action sucient (Example 3.10 shows that crossed products by
such actions do not necessarily preserve tracial rank zero), while on the other hand
the discussion above shows that none of the actions has the Rokhlin property.
Theorem 3.14 (Theorem 3.8 of [60]). Let be a nondegenerate skew symmetric real d d matrix, with d 2. Then the noncommutative d-torus A is a simple
AT algebra with real rank zero.
The relevance of actions of nite groups is that they allow reduction of the
general case to the case in which A can be written as an iterated crossed product
by actions of Z in such a way that all the intermediate crossed products are simple.
This case was solved by Kishimoto (Corollary 6.6 of [36]).
The action and the subgroups which appear in the following theorem are described, for example, in the introduction to [11].
Theorem 3.15 (Theorem 0.1 of [11]). Let R \ Q. Let A be the irrational
rotation algebra, and let : SL2 (Z) Aut(A ) be the standard action of SL2 (Z)
on A . Let F be any of the standard nite subgroups Z2 , Z3 , Z4 , Z6 SL2 (Z). Then
the crossed product C (F, A , |F ) is an AF algebra.
(The case F = Z2 was already known [8].)
Theorem 3.16 (Theorem 0.4 of [11]). Let A be the noncommutative d-torus
corresponding to a nondegenerate skew symmetric real d d matrix . Let : Z2
Aut(A ) denote the ip action. Then C (Z2 , A , ) and the xed point algebra
AZ2 are AF algebras.
The tracial Rokhlin property has the following permanence property:
Proposition 3.17 (Lemma 5.6 of [11]). Let A be an innite dimensional simple unital C*-algebra, and let : G Aut(A) be an action of a nite group G on

240
24

N. CHRISTOPHER PHILLIPS

A which has the tracial Rokhlin property. Let H be a subgroup of G. Then |H has
the tracial Rokhlin property.
Permanence properties involving ideals, quotients, and extensions dont make
sense, since the tracial Rokhlin property is (so far) dened only for actions on
simple C*-algebras. It seems plausible that a direct limit of actions with the tracial
Rokhlin property again has the tracial Rokhlin property, but nobody has checked
this.
Problem 3.18. Let G be a nite group, let A and B be innite dimensional
simple unital C*-algebras, let : G Aut(A) be an action with the tracial Rokhlin
property, and let : G Aut(B) be an arbitrary action. Does it follow that
min : G Aut(A min B) has the tracial Rokhlin property? (We use the
minimal tensor product to ensure simplicity.)
Lemma 3.9 of [59] is the very special case B = Mn and is inner. Proposition 4.3 of [39] gives a related result for actions of Z which have the tracial cyclic
Rokhlin property, Denition 2.4 of [39]. The assumptions are that A is simple,
unital, and has tracial rank zero, that B is simple, unital, and has tracial rank
at most one, that Aut(A) has the tracial cyclic Rokhlin property, and that
Aut(B) is arbitrary. The conclusion is that min Aut(A min B) has the
tracial cyclic Rokhlin property. The same proof gives the following result, pointed
out to us by Hiroyuki Osaka:
Proposition 3.19 (Osaka). Let G be a nite group, let A and B be innite
dimensional simple unital C*-algebras, let : G Aut(A) be an action with the
tracial Rokhlin property, and let : G Aut(B) be an arbitrary action. Suppose
A has tracial rank zero and B has tracial rank at most one. Then min : G
Aut(A min B) has the tracial Rokhlin property.

The key point is that the condition on 1 gG eg in Denition 3.1(3) can
be veried by using the values of tracial states on this element. A general positive
solution to Problem 3.18 requires relating hereditary subalgebras in A min B to
hereditary subalgebras in A, which might be dicult.
The tracial Rokhlin property, as given in Denition 3.1, suers from three major
defects: the algebra must be unital, it must have many projections, and it must be
simple.
Presumably the nonunital simple case can be handled by something like Definition 2.15. However, the correct analog of Condition (3) of Denition 3.1 is not
clear.
Archey [3] has made progress toward handling the simple unital case with few
projections. We refer to [3] for unexplained terminology in the following. Condition (3) is redundant, but under some conditions Condition (4) can be omitted,
leaving only the simpler Condition (3).
Definition 3.20 ([3]). Let A be an innite dimensional stably nite simple
unital C*-algebra, and let : G Aut(A) be an action of a nite group G on A.
We say that has the projection free tracial Rokhlin property if for every nite
set F A, every > 0, and every positive element x A with x = 1, there
are mutually orthogonal positive elements ag A for g G with ag = 1 for all
g G, such that:
(1) g (ah ) agh < for all g, h G.

FREENESS OF ACTIONS ON C*-ALGEBRAS

241
25

(2) ag c cag
< for all g G and all c F.
(3) With a = gG ag , we have (1 a) < for every tracial state on A.

(4) With a = gG ag , the element 1a is Cuntz subequivalent to an element
of the hereditary subalgebra of A generated by x.
For example, let Z be the Jiang-Su algebra [29]. Then Z Z has no nontrivial
projections. Archey shows [3] that the tensor ip on Z Z generates an action of
Z2 with the projection free tracial Rokhlin property. The other hypotheses of the
following theorem are also satised. Again, see [3] for unexplained terminology.
Theorem 3.21 ([3]). Let A be an innite dimensional stably nite simple unital
C*-algebra with stable rank one and with strict comparison of positive elements.
Further assume that every 2-quasitrace on A is a trace, and that A has only nitely
many extreme tracial states. Let : G Aut(A) be an action of a nite group G
on A which has the projection free tracial Rokhlin property. Then C (G, A, ) has
stable rank one.
One possible next step is to ask whether there is an analog of Theorem 3.5 using
the projection free tracial Rokhlin property, say for actions on simple separable
unital nuclear Z-stable C*-algebras with a unique tracial state. For that matter, one
might try using outerness in the Gelfand-Naimark-Segal representation associated
with a tracial state as a hypothesis for theorems on preservation of structure in
crossed products.
Finding the right denition for nonsimple C*-algebras seems to be the most
dicult problem. One guide is that a free action on a compact metric space should
presumably have the tracial Rokhlin property. There is work in progress for actions
of Z on quite special nonsimple C*-algebras.
4. Pointwise outerness
Pointwise outerness is easy to dene and, at least for discrete groups acting
on simple C*-algebras, has useful consequences. This and related conditions have
mostly been used to prove simplicity of crossed products C (G, A, ) when G is
discrete and A has no nontrivial G-invariant ideals, or, more generally, that every
ideal in C (G, A, ) is the crossed product by an invariant ideal of A. There are
also theorems on preservation of structure, for example for pure inniteness and
Property (SP).
Like K-theoretic freeness, pointwise outerness does not pass to invariant ideals
or their quotients. A useful condition for actions on nonsimple algebras must therefore be stronger. A number of strengthenings have been used. Recently introduced conditions include topological freeness [1], essential freeness of the action
on the space of unitary equivalence classes of irreducible representations [70], and
the Rokhlin* property [70]. In Denition 4.11 below, we give another possible
strengthening: requiring pointwise outerness for all actions of subgroups on invariant subquotients of the algebra. But we do not know how useful this condition
is.
Definition 4.1. An action : G Aut(A) is said to be pointwise outer if, for
g G \ {1}, the automorphism g is outer, that is, not of the form a  Ad(u)(a) =
uau for some unitary u in the multiplier algebra M (A) of A.
Such actions are often just called outer.

242
26

N. CHRISTOPHER PHILLIPS

An inner action : G Aut(A) is one for which there exists a homomorphism


g  ug , from G to the unitary group of M (A), such that g = Ad(ug ) for all
g G. (If G is not discrete, one should impose a suitable continuity condition.)
There exist (see Example 5.6 below) actions of nite groups which are pointwise
inner (so that each g has the form Ad(ug )) but not inner (it is not possible to
choose g  ug to be a group homomorphism).
In the literature, a single automorphism is often called aperiodic if it generates
a pointwise outer action of Z.
Example 4.2. Let : Z2 Aut(A) be as in Denition 1.10. Then the following
are equivalent:
(1) is pointwise outer.
(2) C (Z2 , A, ) is simple.
(3) For innitely many n, we have k(n) = 0.
See Proposition 2.6 of [61], where additional equivalent conditions are given.
By comparison with Example 3.3, the choices d(n) = 2n and k(n) = 1 for all n
give a pointwise outer action of Z2 which does not have the tracial Rokhlin property.
Moreover, becomes inner on the double commutant of the Gelfand-Naimark-Segal
representation from the unique tracial state on A. However, for examples of this
type, it follows that the tracial Rokhlin property implies pointwise outerness. This
is true in general.
Proposition 4.3 (Lemma 1.5 of [59]). Let A be an innite dimensional simple
unital C*-algebra, and let : G Aut(A) be an action of a nite group G on A.
If has the tracial Rokhlin property, then is pointwise outer.
For type I C*-algebras, see Theorem 4.12 and the discussion before Denition 4.11 below.
The main application so far of pointwise outerness has been to proofs of simplicity of reduced crossed products. This application is valid for general discrete

groups. The next theorem is due to Kishimoto. The expression ()
is as in Denition 5.9 below, and requiring that it be nontrivial is a strong outerness condition.
The second result is a corollary of the rst.
For the statement of this and several later results, the following denition is
convenient. It generalizes a standard denition for actions on topological spaces.
Definition 4.4. An action : G Aut(A) is minimal if A has no -invariant
ideals other than {0} and A.
Theorem 4.5 (Theorem 3.1 of [32]). Let : G Aut(A) be a minimal action
 g ) (with g being regarded
of a discrete group G on a C*-algebra A. Suppose that (
as an action of Z) is nontrivial for every g G \ {1}. Then the reduced crossed
product Cr (G, A, ) is simple.
Theorem 4.6 (Part of Theorem 3.1 of [32]). Let : G Aut(A) be an action
of a discrete group G on a simple C*-algebra A. Suppose that is pointwise outer.
Then Cr (G, A, ) is simple.
We note the following generalization, which is the corollary after Theorem 1
 is the space of unitary equivalence classes
in [1]. In this theorem and the next, A
of irreducible representations of the C*-algebra A, with the hull-kernel topology.

FREENESS OF ACTIONS ON C*-ALGEBRAS

243
27

Theorem 4.7 ([1]). Let : G Aut(A) be a minimal action of a discrete


group G on a C*-algebra A. Suppose that is topologically free, that is, for every
nite set F G \ {1}, the set


 : gx = x for all g F
xA
 Then C (G, A, ) is simple.
is dense in A.
r
In the following result of Sierakowski, minimality is not required, and the conclusion is accordingly that all ideals in the crossed product are crossed products of
ideals in the original algebra. The proof involves applying Theorem 4.7 to invariant
quotients. Exact actions are as in Denition 1.2 of [70]. In particular, every action
of an exact group is exact. Example 4.13 below shows that topological freeness
does not suce in this theorem.
Theorem 4.8 (Theorem 1.16 of [70]). Let : G Aut(A) be an exact action
 is
of a discrete group G on a C*-algebra A. Suppose that the action of G on A

essentially free, that is, for every G-invariant closed subset X A, the subset


x X : gx = x for all g G \ {1}
is dense in X. Then every ideal J C (G, A, ) has the form C (G, I, () |I ) for
some G-invariant ideal I A.
In Theorem 2.5 of [70], the same conclusion is obtained using the Rokhlin*
property (Denition 2.1 of [70]) in place of essential freeness. The Rokhlin* property is a weaker hypothesis, by Theorem 2.10 of [70]. For nite groups, the Rokhlin
property implies the Rokhlin* property, but the Rokhlin* property is much weaker
than the Rokhlin property, involving projections in the second dual of quotients of
the algebra.
 and
It is built into the denitions of both essential freeness of the action on A
the Rokhlin* property that these properties pass to quotients by invariant ideals.
Pointwise outerness of an action on a simple C*-algebra also implies that the
crossed product preserves pure inniteness and Property (SP) (every nonzero hereditary subalgebra contains a nonzero projection). The following two results are
special case of Corollaries 4.4 and 4.3 of [27]. (For the rst, one also needs Theorem 4.5.)
Theorem 4.9 ([27]). Let : G Aut(A) be a pointwise outer action of a
discrete group G on a unital purely innite simple C*-algebra A. Then C (G, A, )
is purely innite simple.
Theorem 4.10 ([27]). Let : G Aut(A) be a pointwise outer action of
a discrete group G on a unital simple C*-algebra A with Property (SP). Then
C (G, A, ) has Property (SP).
In the general statement of Corollary 4.4 of [27], it is allowed that
N = {g G : g is inner}
be nite instead of necessarily trivial. The conclusion is then that C (G, A, ) is
purely innite but not necessarily simple. In Corollary 4.3 of [27], dealing with
Property (SP), if G is nite then can be arbitrary.
Denition 4.1 is suitable only for actions on simple C*-algebras, since one can
always take the direct sum of an outer action on one C*-algebra and the trivial

244
28

N. CHRISTOPHER PHILLIPS

action on another. Although, to our knowledge, nothing has been done with it, the
most obvious way to rule out such actions seems to be the following denition. It
is motivated by Theorem 1.4 and adapted from [55].
Definition 4.11. An action : G Aut(A) is said to be strongly pointwise
outer if, for every g G \ {1} and any two g -invariant ideals I J A with
I = J, the automorphism of J/I induced by g is outer.
Theorem 4.12. Let : G Aut(A) be an action of a compact metrizable
group G on a type I C*-algebra A. Then is strongly pointwise outer if and only
if induces a free action on Prim(A).
Proof. Suppose the action of G on Prim(A) is not free. We prove that
is not strongly pointwise outer. Corollary 8.1.2 of [58] provides a G-invariant
composition series (I ) , for some ordinal , such that each composition factor
I+1 /I has Hausdor primitive ideal space. Choose g G \ {1}, < , and
P Prim(I+1 /I ) such that gP = P. Set B = I+1 /I . Then g descends to an
automorphism of B/P. Since B has type I and B/P is simple, there is a Hilbert
space H such that B/P
= K(H). Because all automorphisms of K(H) are inner,
we have contradicted strong pointwise outerness.
Now suppose the action on Prim(A) is free. Let g G \ {1} and let I J A
be g -invariant ideals with I = J. Then Prim(J/I) is a nonempty g-invariant subset
of Prim(A), and the automorphism of J/I induced by g is therefore nontrivial on
Prim(J/I). Thus, this automorphism cant be inner.

For any C*-algebra A, free action of the group on Prim(A) clearly implies the
conditions used in Theorems 4.7 and 4.8. Hence, by Theorem 2.10 of [70], this
condition implies the Rokhlin* property.
The following example shows that in Denition 4.11 it is not enough to assume
that the action on every g -invariant ideal is outer, or that the action on the
quotient by every g -invariant ideal is outer, or even both. The action fails to have
 However, it is topologically
the Rokhlin* property and is not essentially free on A.
free in the sense used in Theorem 4.7. Topological freeness is therefore not enough
to obtain the conclusion of Theorem 4.8.
Example 4.13. We construct an action : G Aut(A), with G = Z2 and in
which A is a separable type I C*-algebra, with the following properties:
(1) For every g G \ {1} and any g -invariant ideal I A with I = A, the
automorphism of A/I induced by g is outer.
(2) For every g G \ {1} and any g -invariant ideal I A with I = {0}, the
automorphism of I induced by g is outer.
(3) is not strongly pointwise outer.
(4) The action of G on Prim(A) is not free.
(5) is topologically free in the sense used in Theorem 4.7.
 is not essentially free in the sense used in Theo(6) The action of G on A
rem 4.8.
(7) does not have the Rokhlin* property.
(8) There is an ideal in C (G, A, ) not of the form C (G, I, () |I ) for any
G-invariant ideal I A.
Let K = K(H) be the algebra of compact operators on a separable innite
dimensional Hilbert space H. Let e1 , e2 L(H) be two innite rank projections

FREENESS OF ACTIONS ON C*-ALGEBRAS

245
29

such that e1 + e2 = 1, and let u L(H) be a unitary such that ue1 u = e2


and u2 = 1. Set B0 = K + Ce1 + Ce2 L(H), and set B = (K B0 )+ , the
unitization of K B0 . Let 0 : Z2 Aut(B0 ) be the action generated by Ad(u).
(Note that uB0 u = B0 even though u  B0 .) Then let : Z2 Aut(B) be the
action (idK 0 )+ . Finally set A = C (Z2 , B, ), and let : Z2 Aut(A) be the

dual action = .
It follows from Takai duality (7.9.3 of [56]) and the fact that crossed products
preserve exact sequences (Lemma 2.8.2 of [58]) that the map I  C (Z2 , I, |I )
denes a bijection between the -invariant ideals of B and the -invariant ideals
of A. The only nontrivial -invariant ideals of B are
I =KK

and

J = K B0 .

Accordingly, the only nontrivial -invariant ideals of A are






and M = C Z2 , K B0 , () |KB0 .
L = C Z2 , K K, () |KK
The action on K K is inner, so L
= (K K) (K K), and exchanges
the two summands. Since every nonzero invariant ideal in A contains L, it follows
that the action on every such ideal is outer. This proves (2).
The action on B/J
= C is trivial, so A/M
= C C, and exchanges the
two summands. Since every invariant ideal in A, other than A itself, is contained
in M, it follows that the action on the quotient by every such ideal is outer. This
proves (1).
However,
 the induced action on Prim(A) is not free, and the subquotient
M/L
= C Z2 , K K, (where exchanges the two summands) is a nonzero
invariant subquotient of A isomorphic to K M2 on which the action is inner.
Thus, we have (3) and (4).
Since B has ideals which are not -invariant (such as K (K + Ce1 )), Takai
duality implies that C (Z2 , A, ) has ideals which are not
-invariant. Such ideals
are not crossed products of invariant ideals in A. This proves (8).
 = Prim(A). We calculate Prim(A).
We prove (5). Since A has type I, we have A
As we saw above, we can write L = L1 L2 for ideals L1 , L2 L which are
exchanged by . Both L1 and L2 are easily seen to be primitive. The ideal L is
itself primitive, since (as we saw above) M/L
= K M2 , and is a xed point. The
isomorphism A/M
= C C shows that M is not primitive, but gives two more
primitive ideals P1 , P2 A, such that A/P1
= A/P2
= C. These are exchanged
by . Thus
Prim(A) = {L1 , L2 , L, P1 , P2 }.
The closed sets are
,

{P1 },

{L1 , L, P1 , P2 },

{P2 },

{P1 , P2 },

{L2 , L, P1 , P2 },

{L, P1 , P2 },
and

Prim(A).

For the nontrivial group element g, we thus have




 : gx = x = {L1 , L2 , P1 , P2 },
xA
which is dense in Prim(A).
The action on Prim(A) is not essentially free, because {L, P1 , P2 } is a closed
set in which the points not xed by g are not dense. This is (6). The statement (7)
follows from (8) and Theorem 2.5 of [70].

246
30

N. CHRISTOPHER PHILLIPS

The following example shows that it is not enough to consider only subquotients
invariant under the entire group, even when the algebra is commutative. In this
example, the action is not topologically free.
Example
 4.14. Let G be a nite group, and let H G be a nontrivial subgroup
such that gG gHg 1 = {1}. For example, take G to be the symmetric group S3
and take H to be one of its two element subgroups. Let G act on X = G/H by
translation, and let be the corresponding action on A = C(G/H). In the example
using an order two subgroup of S3 , the action of S3 on X is just the usual action
of S3 on a three element set by permutations. 
The stabilizer of gH X is gHg 1 . Since gG gHg 1 = {1}, every element
of G \ {1} acts nontrivially on X, so that is pointwise outer. Since is minimal,
is pointwise outer on J/I for every pair of G-invariant ideals I J with I = J.
However, one easily checks that is not strongly pointwise outer. (This also follows
from Theorem 4.12, or from Theorem 1.4.)
Corollary 2.10 of [18] implies that C (G, G/H)
= K(L2 (G/H))C (H). Even
though the action is minimal, this crossed product is not simple.
The analog of Theorem 4.6 would be a positive solution to the following problem.
Problem 4.15. Let : G Aut(A) be a strongly pointwise outer action of a
countable discrete group. Does it follow that every ideal J Cr (G, A, ) has the
form Cr (G, I, () |I ) for some G-invariant ideal I A?
Note that the desired conclusion fails in Examples 4.13 and 4.14.
As far as we can tell, this problem is still open, even when G = Z and is
minimal. (If G = Z and one assumes there are no nontrivial g -invariant ideals for
all g G \ {0}, then the desired conclusion holds. One substitutes Theorem 2.1
of [33] for Lemma 1.1 of [32] in the reasoning leading to the proof of Theorem 3.1
of [32].)
The converse is false. Theorem 4.7 covers some actions on algebras of the
form C(X) by (necessarily nonabelian) groups which are not free, and the crossed
product in Example 5.6 below is simple even though all the automorphisms are
inner.
The extra hypotheses in Theorems 4.5, 4.7, and 4.8, as well as the Rokhlin*
property of [70] and the notion of proper outerness used by Elliott [12], can be
thought of as ways of getting around the failure of outerness to have good permanence properties (as shown by Examples 4.13 and 4.14). The question is whether
these diculties are solved by asking for strong pointwise outerness.
Like the tracial Rokhlin property, the Rokhlin property implies strong pointwise
outerness:
Proposition 4.16. Let A be a unital C*-algebra, let G be a nite group, and
let : G Aut(A) have the Rokhlin property. Then is strongly pointwise outer.
Proof. Let g G \ {1}. Using Proposition 2.14(2), we may assume that
G is generated by g. Let I J A be g -invariant ideals with I = J. Using
Proposition 2.14(1), we may assume that I = 0.
Suppose g |J is inner, and let u M (J) be a unitary such that g (a) = uau
for all a J. Choose a G-invariant element a J such that a = 1. Set n =
card(G) and = 1/(15n2 ). Let (eg )gG be a family of Rokhlin projections in A

FREENESS OF ACTIONS ON C*-ALGEBRAS

247
31


for , {a, au }, and . (Note that au J A.) Write a =
hG eh a. Since
h (e1 ) eh < and h (a) = a for h G, we get h (e1 a) eh a < for all
h G. Thus

1 = a
eh a <
( h (e1 a) + ) = n( e1 a + ).
hG

hG

So e1 a > .
Using at the rst step g (e1 a) = ue1 au (since e1 a J) and also uau = a
(since g (a) = a), we now get
1
n

eg a e1 a eg a g (e1 a) + ue1 au uau e1 + ae1 e1 a


eg g (e1 ) + e1 (au ) (au )e1 + ae1 e1 a < 3.
1
Therefore, using 2n
,

2
1
n1 < e1 a 2 = a e1 e1 a
4n2
a e1 eg a + a e1 eg a e1 a < a e1 eg a + 3.

Since e1 eg = 0, it follows that


1
1
< 3 = 2 .
4n2
5n
This contradiction shows that g |J is in fact not inner.

Corollary 4.17. Let A be a unital type I C*-algebra, let G be a nite group,


and let : G Aut(A) have the Rokhlin property. Then the induced action of G
on Prim(A) is free.
Proof. Apply Theorem 4.12 and Proposition 4.16.

We give a short discussion of the permanence properties of strong pointwise


outerness. That strong pointwise outerness passes to ideals and quotients is built
into the denition; Example 4.13 shows it was necessary to do so. That strong
pointwise outerness passes to actions of subgroups is also built into the denition;
Example 4.14 shows that this was also necessary. We have not investigated whether
the direct limit of strongly pointwise outer actions is strongly pointwise outer. We
have also not investigated whether a tensor product of a strongly pointwise outer
action with another action is again strongly pointwise outer, although a very special
(and easy to prove) case of this appears in Lemma 4.20 below.
We have said little about purely innite simple C*-algebras in this survey.
There is some evidence that some of our freeness conditions collapse for such algebras, or at least for Kirchberg algebras (separable nuclear unital purely innite
simple C*-algebras). Theorem 1 of [43] shows that for actions of Z on unital Kirchberg algebras, pointwise outerness implies the Rokhlin property. The examples
below show that nothing this strong can be true for actions of nite groups, but it
is possible that, say, pointwise outerness implies the tracial Rokhlin property. See
Problem 4.24 below.
We need several lemmas for the proofs of properties of some of our examples.
Lemma 4.18. Let A be a unital C*-algebra with trivial center, let n Z>0 , and
let : Zn Aut(A) be an action such that each automorphism g , for g Zn , is
inner. Then is an inner action, that is, there is a homomorphism g  zg from
Zn to the unitary group U (A) of A such that g = Ad(zg ) for g Zn .

248
32

N. CHRISTOPHER PHILLIPS

Proof. Write the elements of Zn as 0, 1, . . . , n 1. Since 1 is inner, there is a


unitary v A such that 1 (a) = vav for all a A. In particular, 1 (v) = vvv = v.
Therefore 1n (a) = v n av n for all a A. Since 1n = idA and the center of A is
trivial, it follows that there is C such that v n = 1. Choose C such that
n = . Then z = 1 v is a unitary in A with z n = 1 such that k (a) = z k az k
for k = 0, 1, . . . , n 1 and all a A.

We dont know any counterexamples without the hypothesis that A have trivial
center, although we assume they exist.
Lemma 4.19. Let A be a simple unital C*-algebra, let p be a prime, and let
: Zp Aut(A) be an action of Zp on A. If C (Zp , A, ) is simple, then is
pointwise outer.
Proof. If is not pointwise outer, then, since p is prime, g is inner for all
g G. So Lemma 4.18 implies that is inner. In this case, the crossed product is

isomorphic to C (Zp ) A, which is not simple.
Example 2.22 shows that this lemma fails if p is not a prime. The action there
is not pointwise outer but the crossed product is simple.
Lemma 4.20. Let A be a simple unital C*-algebra, let p be a prime, and let
: Zp Aut(A) be a pointwise outer action of Zp on A. Let B be any simple unital
C*-algebra. Then the action g  ( min idB )g = g min idB of Zp on A min B
is pointwise outer.
Proof. By Theorem 4.6, the algebra C (Zp , A, ) is simple. Therefore so is
C (Zp , A, ) min B
= C (Zp , A min B, min idB ).
So Lemma 4.19 implies that min idB is pointwise outer.

The following three examples show that, for actions of Z2 on unital Kirchberg
algebras which satisfy the Universal Coecient Theorem, pointwise outerness, Kfreeness, and the Rokhlin property are all distinct, even in situations in which the
K-theory is suciently nontrivial that K-freeness should be useful.
Example 4.21. In Denition 1.10, take d(m) = 3 and k(m) = 1 for all m.
Let : Z2 Aut(A) be the resulting action. Then B = O A is a unital
Kirchberg algebra which satises the Universal Coecient Theorem. The action
is pointwise outer by Example 4.2. So = idO is pointwise outer by
Lemma 4.20. The K
unneth formula (Theorem 4.1 of [67]) implies
 1  that a  1 a
induces an isomorphism K (A) K (B). Since K0 (A)
Z
=
3 , Proposition 3.13
shows that does not have the Rokhlin property.
We further check that does not have locally discrete K-theory. There is an
obvious isomorphism : O C (Z2 , A, ) C (Z2 , B, ), which is equivariant for
the dual actions, using the trivial action on O . The K
unneth formula implies that
a  (1 a) denes an isomorphism from K (C (Z2 , A, )) to K (C (Z2 , B, ))
which is equivariant for the dual actions. Theorem 2.6.1 and Proposition 2.7.10
of [58] now imply that KZ2 (A)
= KZ2 (B) as R(Z2 )-modules. Example 2.21 shows
that does not have locally discrete K-theory. So neither does .
Example 4.22. Let : Z2 Aut(A) be as in Example 2.5. Then B = O A
is a unital Kirchberg algebra which satises the Universal Coecient Theorem. The

FREENESS OF ACTIONS ON C*-ALGEBRAS

249
33

action has the tracial Rokhlin property (Example 3.12(6)), so is pointwise outer
by Proposition 4.3. The proof of Proposition 4.2 of [61] shows that K0 (A) has
a summand isomorphic to the K0 -group of an odd UHF algebra. So does not
have the Rokhlin property for the same reason as in Example 4.21. However, in this
example, is totally K-free. It suces to check that has locally discrete K-theory.
For this, use Example 2.26 and the same argument as in Example 4.21. (Total Kfreeness has signicant content because the K-theory of B and of C (Z2 , B, ) is
highly nontrivial.)
Example 4.23. Let : Z2 Aut(O2 ) be as in Example 5.8 of [25]. The nontrivial group element acts on the standard generators s1 and s2 of O2 by s1  s1
and s2  s2 . Then is pointwise outer by the theorem in [41]. In Example 5.8
of [25], it is shown that the xed point algebra O2 is isomorphic to O4 . Thus

K0 O2 K0 (O2 ) is not injective. So Theorem 2.7 implies that does not have
the Rokhlin property. Thus, pointwise outerness does not imply the Rokhlin property even in the absence of obstructions of the type that appear in Proposition 3.13
and the preceding discussion.
It does, however, seem reasonable to hope for a positive solution to the following
problem. It is not clear that our denition of the tracial Rokhlin property is right
for actions on unital purely innite simple C*-algebras, and some modication may
be needed.
Problem 4.24. Let A be a unital Kirchberg algebra, let G be a nite group
and let : G Aut(A) be a pointwise outer action. Does it follow that has the
tracial Rokhlin property?
5. Full Connes spectrum
The noncommutative generalization of Theorem 1.3 is saturation. We will,
however, see that saturation is very weak, and we will primarily consider a stronger
condition, hereditary saturation, which can also be expressed in terms of the strong
Connes spectrum. Saturation says that the xed point algebra is essentially the
same as the crossed product. Hereditary saturation turns out to be exactly the
condition needed to ensure that every ideal in the crossed product is the crossed
product of an invariant ideal in the original algebra.
The following lemma is needed as preparation for the denition of saturation.
Lemma 5.1 (Proposition 7.1.3 of [58]). Let : G Aut(A) be an action of a
compact group on a C*-algebra A. Then the following denitions make a suitable
completion of A into an AG C (G, A, ) bimodule, which is almost a Morita equivalence bimodule in the sense of the denition on page 287 of [64] (the only missing
condition is that the range of , C (G,A,) need not be dense):
(1)
(2)
(3)
(4)

a x = ax for a AG and x A.
1
xf = G g1
 (xf (g))d(g) for x A and f L (G, A, ).
x, yAG = G g (xy ) d(g) for x, y A.
x, yC (G,A,) is the function g  x g (y) for x, y A.

The following denition is originally due to Rieel. A version for proper actions
of not necessarily compact groups appears in Denition 1.6 of [66]. Saturation has
been generalized to actions of nite dimensional Hopf *-algebras ([72], [28]).

250
34

N. CHRISTOPHER PHILLIPS

Definition 5.2 (Denition 7.1.4 of [58]). Let : G Aut(A) be an action


of a compact group G on a C*-algebra A. Then is said to be saturated if the
bimodule of Lemma 5.1 is a Morita equivalence bimodule, that is, the range of
, C (G,A,) is dense in C (G, A, ).
For an action of a compact group on C0 (X), it follows from Proposition 7.1.12
and Theorem 7.2.6 of [58] that saturation is equivalent to freeness of the action of G
on X. Saturation is a very weak form of freeness, since even inner actions on simple
C*-algebras can be saturated. See Proposition 7.2.1 of [58]. For
this re
 example,
0
sult shows that the action : Z2 Aut(M2 ) generated by Ad 10 1
is saturated.
Also see Example 5.3. In particular, crossed products of simple C*-algebras by saturated actions can be far from simple. (With as above, C (Z2 , M2 , )
= C C.)
Thus, the analog of Theorem 1.4 fails. It does ensure, by denition, that AG is
strongly Morita equivalent to C (G, A, ), which is a noncommutative analog of
Theorem 1.3. In particular, at least in the separable case, it ensures that the map
K (AG ) K (C (G, A, )) is an isomorphism. This statement is a noncommutative analog of the conclusion of Theorem 1.7.
Example 5.3. Let : Z2 Aut(A) be as in Denition 1.10. Then the following
are equivalent:
(1) is saturated.
(2) For some n, we have k(n) = 0.
(3) is nontrivial.
Obviously saturation implies the other two conditions, and the other two conditions are equivalent. So assume (2). Choose n0 such that k(n0 ) =
0. For n n0 ,
one can use Proposition 7.2.1 of [58] to see that the action of Z2 on nk=1 Md(k) is
saturated. Proposition 7.1.13 of [58] now implies that is saturated.
Saturation is far too weak for most purposes. For something more useful, one
must pass to hereditary saturation.
Definition 5.4 (Denition 7.2.2 of [58]). Let : G Aut(A) be an action of
a compact group G on a C*-algebra A. The action is said to be hereditarily saturated
if for every nonzero G-invariant hereditary subalgebra B A, the restricted action
() |B is saturated.
For our standard product type actions, we have the same behavior as for pointwise outerness:
Example 5.5. Let : Z2 Aut(A) be as in Denition 1.10. Then the following
are equivalent:
(1) is hereditarily saturated.
(2) C (Z2 , A, ) is simple.
(3) For innitely many n, we have k(n) = 0.
To see this, use Theorem 5.10(2) below, Corollary 5.12 below, and Example 4.2.
At least for noncyclic groups, the behavior of hereditary saturation is not as
close to that of pointwise outerness as Example 5.5 suggests.
Example 5.6 (Example 4.2.3 of [58]). Let A = M2 , let G = (Z/2Z)2 with
generators g1 and g2 , and set




1 0
0 1
and g2 = Ad
.
g1 = Ad
0 1
1 0

FREENESS OF ACTIONS ON C*-ALGEBRAS

251
35

These generate an action of G such that g is inner for all g G, but such that
there is no homomorphism g  ug U (A) with g = Ad(ug ) for all g G. The
crossed product C (G, M2 , ) is simple.
Lemma 4.18 shows that an analog of Example 5.6 is not possible if G is cyclic
and A is simple.
For actions on C*-algebras of type I, we have the following two results. Both
are contained in Theorem 8.3.7 of [58].
Theorem 5.7 ([58]). Let : G Aut(A) be an action of a compact Lie
group G on a type I C*-algebra A. If G acts freely on Prim(A), then is hereditarily
saturated.
Theorem 5.8 ([58]). Let : G Aut(A) be an action of a nite cyclic group G
on a type I C*-algebra A. Then is hereditarily saturated if and only if G acts freely
on Prim(A).
Example 5.6 shows that the converse of Theorem 5.7 is false.
Hereditary saturation is closely related to the strong Connes spectrum. Parts
(2) and (3) of the following denition are a special case of denitions of Kishimoto.
See the beginning of Section 2 of [31]. The denitions in [31] are given for the case
of an arbitrary locally compact abelian group, and are more complicated because,
without compactness, the eigenspaces in part (1) are usually too small to be useful.
One must use approximate eigenspaces instead. We refer to [31] for the denition
in that case, but we make some comments below about what happens in the locally
compact case.
Definition 5.9 ([31]). Let : G Aut(A) be an action of a compact abelian
group G on a C*-algebra A.
 the Pontryagin dual of G, we let A A be the eigenspace
(1) For G,
A = {a A : g (a) = (g)a for all g G}.

(2) The strong (Arveson) spectrum Sp()
of is



 : A AA = A .
Sp()
= G


(3) The strong Connes spectrum ()
of is the intersection over all nonzero


 () |B .
G-invariant hereditary subalgebras B A of Sp
The strong Arveson spectrum is a modication of a much older notion called
the (Arveson) spectrum Sp(), dened for actions of compact groups by the using
the condition A = {0} instead of A AA = A. Thus, the strong Arveson spectrum
is smaller. The Connes spectrum () is then as in Denition 5.9(3), but using
the Arveson spectrum instead of the strong Arveson spectrum. The Connes spectrum was introduced by Connes (Section 2.2 of [9]) for actions on von Neumann
algebras. The main early work for C*-algebras was done by Olesen and Pedersen.
See [46], [47], and [48]. Also see Sections 8.1 and 8.88.11 of [56], where a third
version, the Borchers spectrum, is also treated. We briey discuss the signicance
of the dierence after Theorem 5.14. The analog of the strong Connes spectrum for
von Neumann algebras gives the same thing as the Connes spectrum (Remark 2.4

of [31]). Some cases in which ()
= () are discussed at the end of Section 3
of [31].

252
36

N. CHRISTOPHER PHILLIPS

A version of the strong Arveson spectrum for actions of compact nonabelian


groups is given in Denition 1.1(b) of [16], and a version of the strong Connes
spectrum is given in Denition 1.2(b) of [16]. The values of both are subsets of the
 of unitary equivalence classes of irreducible representations of G.
space G
The relevance here is the following theorem, which follows from the discussion
after Lemma 3.1 of [16]. For abelian groups, the rst part is essentially originally
due to Rieel. See Theorem 7.1.15 of [58] and the comment after its proof, and
Theorem 7.2.7 of [58].
Theorem 5.10 ([16]). Let : G Aut(A) be an action of a compact group G
on a C*-algebra A. Then:


(1) is saturated if and only if Sp()
= G.


(2) is hereditarily saturated if and only if ()
= G.
Hereditary saturation, equivalently full strong Connes spectrum, is exactly the
condition needed for every ideal in the crossed product to be the crossed product
by an invariant ideal. Combining Theorem 5.10(2) with Theorem 3.3 of [16], we
get:
Theorem 5.11 ([16]). Let : G Aut(A) be an action of a compact group G
on a C*-algebra A. Then the following are equivalent:


(1) ()
= G.
(2) is hereditarily saturated.
(3) Every ideal J C (G, A, ) has the form C (G, I, () |I ) for some Ginvariant ideal I A.
Corollary 5.12. Let : G Aut(A) be an action of a compact group G on
a C*-algebra A. Then C (G, A, ) is simple if and only if is minimal (Deni
 if and only if is minimal and hereditarily saturated.
tion 4.4) and ()
= G,
See [15] for more on the Connes spectrum for actions of compact nonabelian
groups, including ways in which their behavior is both like and unlike that for
actions of compact abelian groups.
A further generalization to actions of suitable Hopf C*-algebras might be possible. As mentioned above, saturated actions of nite dimensional Hopf *-algebras
have been studied ([72], [28]). As far as we know, however, hereditarily saturated
actions have not been considered.
One gets similar results for abelian but not necessarily compact groups. These
are due to Kishimoto [31].
Theorem 5.13 (Lemma 3.4 of [31]). Let : G Aut(A) be an action of a
locally compact abelian group G on a C*-algebra A. Then



:
()
= G
 (I) I for all ideals I C (G, A, ) .
The following is a consequence of Theorem 5.13 and Takai duality (7.9.3 of [56]):
Theorem 5.14 ([31]). Let : G Aut(A) be an action of an abelian group G
on a C*-algebra A. Then the following are equivalent:


(1) ()
= G.
(2) Every ideal J C (G, A, ) has the form C (G, I, () |I ) for some Ginvariant ideal I A.

FREENESS OF ACTIONS ON C*-ALGEBRAS

253
37

In particular, it follows (Theorem 3.5 of [31]) that C (G, A, ) is simple if and




only if is minimal (Denition 4.4) and ()
= G.
The corresponding results using the Connes spectrum for an action of an abelian
group are that


:
() = G
 (I) I = for all ideals I C (G, A, )
(Proposition 8.11.8 of [56]), and that C (G, A, ) is prime if and only if A is G
prime (any two nonzero G-invariant ideals have nonzero intersection) and () = G
(Theorem 8.11.10 of [56]). We also mention Corollary 8.9.10 of [56]: an automorphism of a simple C*-algebra is inner if and only if the Connes spectrum of the
action of Z that it generates is {1}. We have chosen to emphasize the strong Connes
spectrum because of Theorems 5.11 and 5.14.
The Rokhlin property and the tracial Rokhlin property imply hereditary saturation:
Proposition 5.15. Let A be a unital C*-algebra, let G be a nite group, and
let : G Aut(A) have the Rokhlin property. Then is hereditarily saturated and


()
= G.
Proof. Combine Theorem 2.8 and Theorem 5.11.

The result probably also holds when G is a second countable compact group,

and the statement about ()
probably holds when G = Z and when G = R.
Proposition 5.16. Let A be an innite dimensional simple unital C*-algebra,
let G be a nite group, and let : G Aut(A) have the tracial Rokhlin property.


Then is hereditarily saturated and ()
= G.
Proof. Combine Proposition 4.3, Theorem 4.6, and Corollary 5.12.

For the relationship with strong pointwise outerness, the following easy to get
results are all we know. However, strong pointwise outerness ought to imply hereditary saturation in much greater generality.
Proposition 5.17. Let A be a simple unital C*-algebra, let p be a prime, and
let : Zp Aut(A) be an action of Zp on A. Then is hereditarily saturated if
and only if is pointwise outer.
Proof. The crossed product C (Zp , A, ) is simple if and only if is pointwise
outer, by Lemma 4.19 and Theorem 4.6. Also, C (Zp , A, ) is simple if and only if
is hereditarily saturated, by Corollary 5.12.

Proposition 5.18. Let : G Aut(A) be a pointwise outer action of a nite
group G on a simple C*-algebra A. Then is hereditarily saturated.
Proof. Combine Theorem 4.6 and Corollary 5.12.

Hereditary saturation has the following permanence properties.


Proposition 5.19. Let A be a unital C*-algebra, let G be a compact group,
and let : G Aut(A) be an action of G on A.
(1) If I A is a -invariant ideal, then is hereditarily saturated if and
only if () |I and the induced action of G on A/I are both hereditarily
saturated.

254
38

N. CHRISTOPHER PHILLIPS

(2) If A = lim An is a direct limit of C*-algebras, and : G Aut(A) is

an action obtained as the direct limit of actions (n) : G Aut(An ), such


that (n) is hereditarily saturated for all n, then is hereditarily saturated.
Proof. Part (1) is Proposition 7.2.3 of [58].
Part (2) was overlooked in [58]. For saturation, it is Proposition 7.1.13 of [58].
The rest of the proof follows the same argument as for the proof of Proposition 2.25(3).

One can see from Example 5.6 that hereditary saturation does not pass to
subgroups, since the nontrivial subgroups in that case act via inner actions. Example 2.22 shows (see Remark 9.3.10 of [58]) that saturation does not even pass to
subgroups of cyclic groups.
Problem 5.20. Which actions of nite groups have the property that their
restrictions to all subgroups are hereditarily saturated? Are such actions necessarily
strongly pointwise outer?
As far as we know, nobody has looked at this. Proposition 5.17 might be taken
as evidence in favor of the second part.
Lemma 4.20 and Proposition 5.17 imply a very special case of hereditary saturation of the tensor product of a hereditarily saturated action and an arbitrary
action. The general result, however, is false.
Example 5.21. Adopt the notation of Example 5.6. It follows from Corollary 5.12 that the action is hereditarily saturated. Let B = C2 , and dene
: G Aut(B) by g1 = idB and g2 (1 , 2 ) = (2 , 1 ) for 1 , 2 C. Then
AB
= M2 M2 , and g2 g2 interchanges the summands, so is minimal.
However, C (G, AB, ) has vector space dimension card(G)dim(AB) = 32,
and there is no simple C*-algebra of this dimension. Thus C (G, A B, ) is
not simple, so is not hereditarily saturated, by Corollary 5.12.
References
[1] R. J. Archbold and J. S. Spielberg, Topologically free actions and ideals in discrete C*dynamical systems, Proc. Edinburgh Math. Soc. (2) 37(1994), 119124.
[2] D. Archey, Crossed product C*-algebras by nite group actions with the tracial Rokhlin property, Rocky Mtn. J. Math., to appear.
[3] D. Archey, Crossed product C*-algebras by nite group actions with the projection free tracial
Rokhlin property, preprint (arXiv: 0902.3324 [math.OA]).
[4] M. F. Atiyah, K-Theory, W. A. Benjamin, New York, Amsterdam, 1967.
[5] M. F. Atiyah and I. G. Macdonald, Introduction to Commutative Algebra, Addison-Wesley,
Reading MA, Menlo Park CA, London, Amsterdam, Don Mills ON, Sydney, 1969.
[6] M. F. Atiyah and G. B. Segal, Equivariant K-theory and completion, J. Dierential Geometry
3(1969), 118.
[7] B. Blackadar, Symmetries of the CAR algebra, Ann. Math. (2) 131(1990), 589623.
[8] O. Bratteli and A. Kishimoto, Noncommutative spheres. III. Irrational rotations, Commun.
Math. Physics 147(1992), 605624.

[9] A. Connes, Une classication des facteurs de type III , Ann. Sci. Ecole
Norm. Sup. (4) 6(1973),
133252.
[10] A. Connes, Sur la classication des facteurs de type II , C. R. Acad. Sci. Paris S
er. A
281(1975), 1315.
[11] S. Echterho, W. L
uck, N. C. Phillips, and S. Walters, The structure of crossed products of irrational rotation algebras by nite subgroups of SL2 (Z), preprint (arXiv: math.OA/0609784).

FREENESS OF ACTIONS ON C*-ALGEBRAS

255
39

[12] G. A. Elliott, Some simple C*-algebras constructed as crossed products with discrete outer
automorphism groups, Publ. RIMS Kyoto Univ. 16(1980), 299311.
[13] G. A. Elliott, A classication of certain simple C*-algebras, pages 373385 in: Quantum and
Non-Commutative Analysis, H. Araki etc. (eds.), Kluwer, Dordrecht, 1993.
[14] T. Fack and O. Marechal, Sur la classication des sym
etries des C*-alg`
ebres UHF , Canad.
J. Math. 31(1979), 496523.
[15] E. C. Gootman and A. J. Lazar, Spectra for innite tensor product type actions of compact
groups, Canad. J. Math. 48(1996), 330342.
[16] E. C. Gootman, A. J. Lazar, and C. Peligrad, Spectra for compact group actions, J. Operator
Theory 31(1994), 381399.
[17] P. Green, C*-algebras of transformation groups with smooth orbit space, Pacic J. Math.
72(1977), 7197.
[18] P. Green, The structure of imprimitivity algebras, J. Funct. Anal. 36(1980), 88104.
[19] D. Handelman and W. Rossmann, Product type actions of nite and compact groups, Indiana
Univ. Math. J. 33(1984), 479509.
[20] D. Handelman and W. Rossmann, Actions of compact groups on AF C*-algebras, Illinois J.
Math. 29(1985), 5195.
[21] R. H. Herman and V. F. R. Jones, Period two automorphisms of UHF C*-algebras, J. Funct.
Anal. 45(1982), 169176.
[22] R. H. Herman and V. F. R. Jones, Models of nite group actions, Math. Scand. 52(1983),
312320.
[23] I. Hirshberg and W. Winter, Rokhlin actions and self-absorbing C*-algebras, Pacic J. Math.
233(2007), 125143.
[24] M. Izumi, The Rohlin property for automorphisms of C*-algebras, pages 191206 in: Mathematical Physics in Mathematics and Physics (Siena, 2000), Fields Inst. Commun. vol. 30,
Amer. Math. Soc., Providence RI, 2001.
[25] M. Izumi, Finite group actions on C*-algebras with the Rohlin property. I , Duke Math. J.
122(2004), 233280.
[26] M. Izumi, Finite group actions on C*-algebras with the Rohlin property. II , Adv. Math.
184(2004), 119160.
[27] J. A. Jeong and H. Osaka, Extremally rich C*-crossed products and the cancellation property,
J. Austral. Math. Soc. (Series A) 64(1998), 285301.
[28] J. A. Jeong and G. H. Park, Saturated actions by nite-dimensional Hopf *-algebras on
C*-algebras, International J. Math. 19(2008), 125144.
[29] X. Jiang and H. Su, On a simple unital projectionless C*-algebra, Amer. J. Math. 121(1999),
359413.
[30] V. F. R. Jones, Actions of nite groups on the hypernite type II1 factor , Mem. Amer. Math.
Soc. 28(1990), no. 237.
[31] A. Kishimoto, Simple crossed products of C*-algebras by locally compact abelian groups,
Yokohama Math. J. 28(1980), 6985.
[32] A. Kishimoto, Outer automorphisms and reduced crossed products of simple C*-algebras,
Commun. Math. Phys. 81(1981), 429435.
[33] A. Kishimoto, Freely acting automorphisms of C*-algebras, Yokohama Math. J. 30(1982),
3947.
[34] A. Kishimoto, Actions of nite groups on certain inductive limit C*-algebras, International
J. Math. 1(1990), 267292.
[35] A. Kishimoto, A Rohlin property for one-parameter automorphism groups, Comm. Math.
Phys. 179(1996), 599622.
[36] A. Kishimoto, Unbounded derivations in AT algebras, J. Funct. Anal. 160(1998), 270311.
[37] H. Lin, Tracially AF C*-algebras, Trans. Amer. Math. Soc. 353(2001), 693722.
[38] H. Lin, The tracial topological rank of C*-algebras, Proc. London Math. Soc. 83(2001), 199
234.
[39] H. Lin and H. Osaka, The Rokhlin property and the tracial topological rank , J. Funct. Anal.
218(2005), 475494.
[40] H. Lin and H. Su, Classication of direct limits of generalized Toeplitz algebras, Pacic J.
Math. 181(1997), 89140.
[41] K. Matsumoto and J. Tomiyama, Outer automorphisms on Cuntz algebras, Bull. London
Math. Soc. 25(1993), 6466.

256
40

N. CHRISTOPHER PHILLIPS

[42] H. Nakamura, The Rohlin property for Z2 -actions on UHF algebras, J. Math. Soc. Japan
51(1999), 583612
[43] H. Nakamura, Aperiodic automorphisms of nuclear purely innite simple C*-algebras, Ergod.
Th. Dynam. Sys. 20(2000), 17491765.
[44] H. Nakamura, Aperiodic automorphisms of certain simple C*-algebras, pages 145157 in:
Operator Algebras and Applications (Adv. Stud. Pure Math. vol. 38), Math. Soc. Japan,
Tokyo, 2004.
[45] A. Ocneanu, Actions of Discrete Amenable Groups on von Neumann Algebras, SpringerVerlag Lecture Notes in Math. no. 1138, Springer-Verlag, Berlin, 1985.
[46] D. Olesen and G. K. Pedersen, Applications of the Connes spectrum to C*-dynamical systems,
J. Funct. Anal. 30(1978), 179197.
[47] D. Olesen and G. K. Pedersen, Applications of the Connes spectrum to C*-dynamical systems, II , J. Funct. Anal. 36(1980), 1832.
[48] D. Olesen and G. K. Pedersen, Applications of the Connes spectrum to C*-dynamical systems, III , J. Funct. Anal. 45(1982), 357390.
[49] D. S. Ornstein and B. Weiss, Entropy and isomorphism theorems for actions of amenable
groups, J. Analyse Math. 48(1987), 1141.
[50] H. Osaka and N. C. Phillips, Stable and real rank for crossed products by automorphisms
with the tracial Rokhlin property, Ergod. Th. Dynam. Sys. 26(2006), 15791621.
[51] H. Osaka and N. C. Phillips, Furstenberg transformations on irrational rotation algebras,
Ergod. Th. Dynam. Sys. 26(2006), 16231651.
[52] H. Osaka and N. C. Phillips, Crossed products by nite group actions with the Rokhlin property, preprint (arXiv: math.OA/0609782).
[53] H. Osaka and N. C. Phillips, Crossed products of simple C*-algebras with tracial rank one by
actions with the tracial Rokhlin property, in preparation.
[54] R. S. Palais, On the existence of slices for actions of non-compact Lie groups, Ann. of
Math. (2) 73(1961), 295323.
[55] C. Pasnicu and N. C. Phillips, Permanence properties for crossed products by nite groups,
in preparation.
[56] G. K. Pedersen, C*-Algebras and their Automorphism Groups, Academic Press, London,
New York, San Francisco, 1979.
[57] N. C. Phillips, K-theoretic freeness of nite group actions on C*-algebras, pages 227243 in:
Group Actions on Rings, S. Montgomery (ed.), Contemporary Mathematics vol. 43, Amer.
Math. Soc., Providence RI, 1985.
[58] N. C. Phillips, Equivariant K-Theory and Freeness of Group Actions on C*-Algebras,
Springer-Verlag Lecture Notes in Math. no. 1274, Springer-Verlag, Berlin, Heidelberg, New
York, London, Paris, Tokyo, 1987.
[59] N. C. Phillips, The tracial Rokhlin property for actions of nite groups on C*-algebras,
preprint (arXiv: arXiv:0704.3651v3 [math.OA]).
[60] N. C. Phillips, Every simple higher dimensional noncommutative torus is an AT algebra,
preprint (arXiv: math.OA/0609783).
[61] N. C. Phillips, Finite cyclic group actions with the tracial Rokhlin property, Trans. Amer.
Math. Soc., to appear.
[62] N. C. Phillips and M. G. Viola, A simple separable exact C*-algebra not anti-isomorphic to
itself , in preparation.
[63] M. A. Rieel, Unitary representations of group extensions; an algebraic approach to the
theory of Mackey and Blattner , pages 4382 in: Studies in Analysis, Advances in Math.
Suppl. Stud., vol. 4, Academic Press, New York, London, 1979.
[64] M. A. Rieel, Morita equivalence for operator algebras, pages 285298 in: Operator Algebras
and Applications, (R. V. Kadison (ed.)), Proceedings of Symposia in Pure Mathematics
38(1982), part 1.
[65] M. A. Rieel, Applications of strong Morita equivalence to transformation group C*-algebras,
pages 299310 in: Operator Algebras and Applications, (R. V. Kadison (ed.)), Proceedings
of Symposia in Pure Mathematics 38(1982), part 1.
[66] M. A. Rieel, Proper actions of groups on C*-algebras, pages 141182 in: Mappings of
operator algebras (Philadelphia, PA, 1988), Progress in Mathematics vol. 84, Birkh
auser
Boston, Boston MA, 1990.

FREENESS OF ACTIONS ON C*-ALGEBRAS

257
41

[67] C. Schochet, Topological methods for C*-algebras II: geometric resolutions and the K
unneth
formula, Pacic J. Math. 98(1982), 443458.

[68] G. Segal, The representation ring of a compact Lie group, Inst. Hautes Etudes
Sci. Publ.
Math. 34(1968), 113128.

[69] G. Segal, Equivariant K-theory, Inst. Hautes Etudes


Sci. Publ. Math. 34(1968), 129151.
[70] A. Sierakowski, The ideal structure of reduced crossed products, preprint (arXiv: 0804.3772v1
[math.OA]).
[71] H. Su, K-theoretic classication for certain inductive limit actions on real rank zero C*algebras, Trans. Amer. Math. Soc. 348(1996), 41994230.
[72] W. Szyma
nski and C. Peligrad, Saturated actions of nite-dimensional Hopf *-algebras on
C*-algebras, Math. Scand. 75(1994), 217239.
[73] Y. Watatani, Index for C*-subalgebras, Mem. Amer. Math. Soc. 83(1990), no. 424.
Department of Mathematics, University of Oregon, Eugene OR 97403-1222, USA.
E-mail address: ncp@darkwing.uoregon.edu

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Examples of masas in C*-algebras


Jean Renault
Abstract. This paper illustrates the notion of a Cartan subalgebra in a C*algebra through a number of examples and counterexamples. Some of these
examples have a geometrical avour and are related to orbifolds and nonHausdor manifolds.

.
1. Cartan subalgebras of C -algebras
I recall in this section some denitions and results of [R3], to which I refer the
reader for a more complete exposition.
Definition 1.1. We shall say that an abelian sub-C -algebra B of a C -algebra
A is a Cartan subalgebra if
(1) B contains an approximate unit of A;
(2) B is maximal abelian;
(3) B is regular;
(4) there exists a faithful conditional expectation P of A onto B.
The main result of [R3] is a C -algebraic version of Feldman-Moores well
known theorem [FM2] on Cartan subalgebras in von Neumann algebras. This
theorem establishes an equivalence of categories between twisted countable standard measured equivalence relations and Cartan subalgebras in von Neumann algebras on separable Hilbert spaces. A notable dierence in the topological case is
that equivalence relations (also known as principal groupoids) have to be replaced
by topologically principal groupoids. The denition of a topologically principal
groupoid is related to the denition of a topologically free action, as given in [To,
Denition 2.1].
Definition 1.2. We say that a groupoid G on a topological space X (this
means that G has X = G(0) as its unit space) is topologically principal if the set of
points of X with trivial isotropy is dense.
All our examples of topologically principal groupoids will be groupoids of germs.
Suppose that an inverse semi-group acts on a topological space X by partial
homeomorphisms (i.e. homeomorphisms of an open subset of X onto another open
subset). Then the set G of germs of these homeomorphisms form a groupoid: let
1991 Mathematics Subject Classication. Primary 37D35; Secondary 46L85.
Key words and phrases. C*-algebras, masas, Cartan subalgebras, groupoids.
1
259

c
2009
American Mathematical Society

260
2

JEAN
JeanRENAULT
Renault

us write [g(x), g, x] the germ of g at a point x in the domain of g; the groupoid


operations are simply
[x, g, y][y, h, z] = [x, gh, z]

[x, g, y]1 = [y, g 1 , x].

We identify X with G(0) through the map x  [x, id, x], where id is the identity
map of X. The topology of germs turn G into an etale topological groupoid (etale
means that the range and source maps are local homeomorphisms). Such a groupoid
is not necessarily Hausdor, even if the unit space X is Hausdor, as we usually
assume. If is countable and G(0) is a Baire space, then G is topologically principal.
Conversely, let G be an etale topological groupoid on a topological space X. Then,
the inverse semi-group of its open bisections acts on X by partial homeomorphisms.
The corresponding groupoid of germs is a quotient of G. One says that G is eective
if this quotient map is injective. If G is Hausdor and topologically principal, then
it is eective.
Let me also recall the construction of the reduced C -algebra of a Hausdor
locally compact groupoid G equipped with a left Haar system = {x }. When
G is etale, one uses the counting measures on the bers Gx = r 1 (x) as a Haar
system. The following operations turn the space Cc (G) of compactly supported
complex-valued continuous functions on G into an involutive algebra:

f g() = f (  )g( 1 )ds() (  );
f () = f ( 1 ).
For each x G(0) , one denes the representation x of Cc (G) on the Hilbert space
L2 (Gx , x ), where Gx = s1 (x) and x = (x )1 , by x (f ) = f . One denes
the reduced norm f r = sup x (f ). The reduced C*-algebra Cr (G) is the
completion of Cc (G) for the reduced norm. We shall need a slight generalization of
the above construction. A twist over a groupoid G is a groupoid extension
TX E G
where T is the circle group, X is a space and, at the level of the unit spaces, the maps
X E (0) G(0) are identication maps. In the topological setting, we require
the maps to be continuous and the identication maps to be homeomorphisms. We
replace the complex-valued functions by the sections of the associated complex line
bundle. Essentially the same formulas as above provide the C*-algebra Cr (G, E).
Theorem 1.3. [R3] Let (G, E) be a twist with G etale, second countable locally compact Hausdor and topologically principal. Then C0 (G(0) ) is a Cartan
subalgebra of Cr (G, E).
Conversely, let B be a Cartan sub-algebra of a separable C*-algebra A. Then,
there exists a twist (G, E) with G etale, second countable locally compact Hausdor and topologically principal and an isomorphism of Cr (G, E) onto A carrying
C0 (G(0) ) onto B.
This theorem extends a theorem of Kumjian [Ku2] who deals with the principal
case and introduces the stronger notion of a diagonal. It requires the property of
unique extension of states, which is studied in [G, ABG].
Definition 1.4. One says that a sub-C*-algebra B of a C*-algebra A has the
unique extension property if all pure states of B extend uniquely to pure states of A.

EXAMPLES OF MASAS
IN C*-ALGEBRAS
Examples

261
3

A Cartan subalgebra which has the unique extension property is called a diagonal
(or a diagonal subalgebra).
Theorem 1.5. [Ku2, R3] Let B be a Cartan sub-algebra of a separable C*algebra A. Let (G, E) be the associated twist. Then, G is principal if and only if B
has the unique extension property.
2. Examples
2.1. Two non-conjugate Cartan subalgebras. Here are examples of C algebras containing at least two non-conjugate diagonal subalgebras (which are not
even isomorphic as algebras). I owe the rst one to A. Kumjian.
1. Let T be the circle and let n be an integer not smaller than 2. Let Dn be the
subalgebra of diagonal matrices in the algebra of matrices Mn (C). The C -algebra
C(T) Mn (C) obviously contains C(T) Dn = C(T)n as a diagonal subalgebra.
However, as shown for example in [Ku1, Example 3(iii)], C(T) Mn (C) can be
realized as the crossed product C -algebra C(T) Zn , of the action of Zn = Z/nZ
on the circle T by the rotation of angle 2/n. Therefore, it also contains C(T) as a
diagonal subalgebra. Both corresponding equivalence relations are equivalent: they
have the same quotient space T. The rst one is given by the trivial covering map
from T {1, . . . , n} onto T while the second is given by the covering map z z n
from T onto T.
2. Let : G H be a continuous homomorphism of locally compact abelian
groups G, H. Then G acts continuously on H by left multiplication and we can
G

form the crossed product C*-algebra GC0 (H). By dualizing, we get : H

and the crossed product C*-algebra H C0 (G). The Fourier transform gives an
isomorphism of these C*-algebras.
If G is discrete and is one-to-one, C0 (H) is a diagonal subalgebra. Similarly,
is another diagonal subalgebra. Both
is discrete and is one-to-one, C0 (G)
if H
conditions happen simultaneously if G is discrete, H is compact, is one-to-one
are not
and has dense range. There are such examples where C0 (H) and C0 (G)
isomorphic.
Example 2.1. G = Z2 , H = R/Z, (m, n) = m + n + Z where (1, , ) are
linearly independent over Q.
Remark 2.2. Let us mention a few more examples of non-conjugate diagonal
subalgebras. C. Phillips gives in [Ph] a collection of examples of non-isomorphic
diagonal subalgebras in simple C*-algebras. They arise from minimal dieomorphisms. The isomorphism of their C*-algebras is established through the classication theory. Cantor minimal systems also provide examples of C*-algebras possessing uncountably many non-conjugate diagonal subalgebras (see [R3, Section
6.1]).
2.2. Variations on the cross. The cross consists of the graph of the functions y = x and y = x on the domain [1, 1]. There are several ways to deal with
the singular point (0, 0). We shall present three of them.
The C*-algebra A = C([0, 1])M2 (C) has the obvious diagonal B = C([0, 1])
D2 . It is instructive to look at the pairs (Ai , Bi = Ai B), where Ai is one of the
following subalgebras of A.

262
4

JEAN
JeanRENAULT
Renault


A1 = {f A : f (0) =

A2 = {f A : f (0) =

A3 = {f A : f (0) =

A4 =

{f A : f (0) =

a
b

b
a


}

a
a

a
a


}

a
0

0
a


}

a
0

0
b


}

2.2.1. A groupoid of germs. In the rst example, B1 is a Cartan subalgebra of


A1 which does not have the unique extension property. Indeed, the states f  ab
both extend the pure state f  a of B1 . The pair (A1 , B1 ) can be realized as
(C (G), C([1, 1])), where G is the groupoid of the action of the group Z/2Z on
[1, 1] by the map T x = x. This groupoid can also be described as the groupoid
of germs of the pseudogroup generated by T :
G = {(x, 1, x), x [1, 1]}.
It is topologically principal but not principal: the isotropy is trivial at x = 0 and
the isotropy group at x = 0 is Z/2Z. It is an elementary example of an orbifold.
The C -algebra A1 is a CCR algebra with non-Hausdor spectrum. I owe to A.
Kumjian the observation that A1 does have a diagonal. Indeed, it is isomorphic to
A4 , which has B4 as a diagonal subalgebra.
2.2.2. A branched covering. The subalgebra B2 is maximal abelian in A2 . However, it does not satisfy the conditions (1) and (4) of the denition of a Cartan subalgebra. Its main defect is to be contained in the ideal f (0) = 0. The C -algebra
A2 can be realized as the C*-algebra of a non-etale principal groupoid, namely
the equivalence relation R associated to the previous groupoid G. Endowed with
the product topology of [1, 1] [1, 1], it is a proper groupoid. It has the Haar
system:

f dx = f (x, x) + f (x, x).
The quotient map, which can be realized as the map x |x| from [1, 1] to [0, 1] is
an elementary example of a branched covering (see [Fo]). The construction given
here appears in [De]. The C*-algebra A2 = C (R, ) does not contain C([1, 1])
as a subalgebra because the diagonal of R is not open; however, it contains B2 =
C([1, 1] \ {0}). The quotient map G R gives the inclusion A2 A1 . Note also
that A1 is the unitization of A2 . It can be readily checked or deduced from [MRW]
that A2 is a continuous trace C*-algebra.
2.2.3. An etale equivalence relation. The subalgebra B3 is a diagonal subalgebra in A3 . It is realized by the same equivalence relation R as above, but endowed with a ner topology which makes it etale. Following M. Molberg [Mo],
we consider the topology generated by the product topology and the diagonal
{(x, x), x [0, 1]}. Then R is etale but no longer proper.

263
5

EXAMPLES OF MASAS
IN C*-ALGEBRAS
Examples

3. Orbifolds and non-Hausdor manifolds


Moerdijk and Pronk have introduced [MP] the notion of an orbifold groupoid
(here, we only need the topological structure of the spaces, not their dierential
structure):
Definition 3.1. An orbifold groupoid is a proper, eective, etale, second countable, locally compact and Hausdor groupoid.
They arise in the following related situations:
orbifolds;
foliated manifolds for which all the leaves are compact with nite holonomy.
The explicit constructions involve some choices but provide equivalent groupoids
(see [MM]). It seems appropriate to dene a (topological) orbifold as an equivalence
class of proper (eective) groupoids. The example 2.2.1 of the previous section is
an elementary example of an orbifold groupoid. The general case keeps some of the
features of this elementary example: if G is an orbifold groupoid, its C -algebra
is a CCR algebra which admits C0 (G(0) ) as a Cartan subalgebra. There is an
intriguing link between the orbifold groupoid G = [1, 1]Z/2Z of example 2.2.1
and the non-Hausdor manifold obtained as the quotient of [0, 1] {0, 1} by the
equivalence relation which identies the two copies of (0, 1]:
R = {(z, (i, j)) [0, 1] ({0, 1} {0, 1}) : i = j

if

z = 0}.

This etale equivalence relation gives the diagonal B4 A4 and we have seen that
A1 and A4 are isomorphic C -algebras. The isotropy has been eliminated at the
expense of non-Hausdorness.
Non-Hausdor manifolds are a rich source of examples of CCR algebras. Let
us dene a locally compact space as topological space Y such that every point has
a compact Hausdor neighborhood. Such a space is T1 . Let us dene a desingularization of Y as a surjective local homeomorphism : X Y , where X is a
Hausdor locally compact space. Then, the graph R of the equivalence relation
(x) = (x ) on X, endowed with the product topology of X X is an etale equivalence relation. It is proper if and only if Y is Hausdor. The C -algebra C (R)
is CCR and its spectrum is homeomorphic to Y (see for example [OC] for these
facts). This provides a convenient way to construct CCR algebras with arbitrary
locally compact spectrum. Let Y be a topological space which is T1 . One says that
two points of Y are separated if they have disjoint neighborhoods and that a point
y Y is Hausdor if it is separated from any other point. The set of Hausdor
points of the spectrum of a separable CCR algebra is a dense G . In many cases
(for example, when the spectrum is compact (not necessarily Hausdor) or in the
case of the C -algebra of a connected and simply connected nilpotent Lie group),
the interior of this set is dense. However, in [Di], Dixmier gives an example of a
separable CCR algebra such that the set of Hausdor points of its spectrum has
an empty interior. Here is an easy construction of a similar algebra inspired by
Dixmiers example (I do not know whether the algebras are the same) and also by
the example A4 . Let Z be a Hausdor locally compact space and let {zn , n N}
be a countable subset of Z. Then
R = {(z, (i, j)) Z (N N) : i = j

if

z {zi , zj }}

264
6

JEAN
JeanRENAULT
Renault

is an open subgroupoid of Z (NN). Therefore, it is an etale equivalence relation


over X = Z N. The quotient space Y = X/R can be described as the disjoint
union of Z and N. The quotient map sends (z, i) X to z if z = zi and to i if
z = zi . Open subsets of Z, where a nite number of zi s have been replaced by i
form a base for the quotient topology. The space Y is locally compact. If none of
the zi s are isolated, the set of Hausdor points is Z \ {zi , i N}. It has an empty
interior if {zi , i N} is dense in Z. As said before, the C -algebra C (R) is CCR
and has Y as its spectrum.
4. The non-Hausdor case
According to Theorem 1.3, Cartan subalgebras in C -algebras are intimately related to topologically principal etale groupoids. These groupoids arise as groupoids
of germs. However, the Hausdorness condition required in the theorem is a severe restriction. It is still possible to dene the reduced C -algebra Cr (G) (and
Cr (G, E)) when G is an etale locally compact non-Hausdor groupoid. By deni
tion, an element of Cc (G) is a function of the form f = ni=1 fi , where fi Cc (Ui )
for some open Hausdor subset Ui G and fi is its extension by zero to G. Then,
the denitions are just as above. A function f in Cc (G) is not necessarily continuous on G. In particular, its restriction to G(0) is not necessarily continuous.
Thus, the existence of a conditional expectation onto C0 (G(0) ) is problematic. The
subalgebra C0 (G(0) ) may also fail to be maximal abelian. Shortly after the workshop, R. Exel gave me an example of a non-Hausdor groupoid of germs G such
that the subalgebra C0 (G(0) ) is not maximal abelian (see [Ex]). I then realized
that this example is related to an earlier example of G. Skandalis which appears in
[R2] and which I reproduce below. Skandalis purpose is dierent: it shows that
the C -algebra of a minimal foliation is not necessarily simple when the holonomy
groupoid is non-Hausdor. However, both pathologies are based on the same fact.
Let g1 , g2 be homeomorphisms of the circle T having for xed points set respectively the oriented arcs [a, b] and [b, a], where a, b are distinct points of T. Since
these homeomorphisms commute, they dene an action of Z2 on T such that (m, n)
acts as g1m g2n . Let [g(x), g, x] denote the germ of a homeomorphism g at x T.
Let G be the groupoid of germs of the g1m g2n s. The only points which have non
trivial isotropy are a and b. The isotropy subgroups at a and b are isomorphic to
Z2 . By construction, S(m, n) = {[g1m g2n (x), g1m g2n , x] : x T} is an open bisection
of G. The function
f = 1S(0,0) 1S(1,0) 1S(0,1) + 1S(1,1)
belongs to Cc (G), hence to Cr (G). It vanishes outside the nite set
{[a, (m, n), a], [b, (m, n), b], m, n = 0, 1}
but takes the values 1 on this set. Since this function has its support contained
in the isotropy group bundle G , it commutes with every element of C(T), however
it does not belong to C(T).
Acknowledgements. I thank the participants of the workshop for stimulating
discussions and comments, in particular R. Archbold and R. Exel. I also benetted
greatly from the help of A. Kumjian.

EXAMPLES OF MASAS
IN C*-ALGEBRAS
Examples

265
7

References
[ABG] R. Archbold, J. Bunce and K. Gregson, Extensions of states of C -algebras II, Proc. Royal
Soc. Edinburgh, 92 A (1982), 113122.
[De] V. Deaconu, Groupoid constructions and C -algebras, PhD Thesis, University of Iowa, 1995.
[Di] J. Dixmier, Points s
epar
es dans le spectre dune C -alg`
ebre, Acta Sci. Math. Szeged 22
(1961), 115128.
[Ex] R. Exel, A non-Hausdor etale groupoid, preprint, arXiv: 0812.4087v2 [math.OA].
[FM1] J. Feldman and C. Moore, Ergodic equivalence relations, cohomologies, von Neumann
algebras, I, Trans. Amer. Math. Soc. 234 (1977), 289324.
[FM2] J. Feldman and C. Moore, Ergodic equivalence relations, cohomologies, von Neumann
algebras, II, Trans. Amer. Math. Soc. 234 (1977), 325359.
[Fo] R.H. Fox: Covering spaces with singularities, Algebraic geometry and topology, Princeton
(1957), 243-257.
[G] K.D. Gregson, Extensions of pure states of C -algebras, Ph. D. thesis, University of Aberdeen
(1986).
[Ku1] A. Kumjian, Preliminary algebras arising from local homeomorphisms, Math. Scand. 52
(1983), 269273.
[Ku2] A. Kumjian, On C -diagonals, Can. J. Math., Vol. XXXVIII,4 (1986), 9691008.
[MM] I. Moerdijk and J. Mr
cun, Introduction to Foliations and Lie Groupoids, Cambridge University Press, Cambridge, UK, 2003.
[MP] I. Moerdijk and D. Pronk, Orbifolds, sheaves and groupoids, K-Theory 12 (1997), 3-21.
[Mo] M. Molberg, AF-equivalence relations, Math. Scand. 99 (2006), no. 2, 247256.
[MRW] P. Muhly, J. Renault and D. Williams, Continuous-trace groupoid C*-algebras, III, Trans.
Amer. Math. Soc. 348 (1996), 36213641.
[OC] L. Orlo Clark, Classifying the type of principal groupoid C -algebras, J. Operator Theory
57 (2007), no. 2, 251266.
[Ph] N. C. Phillips, Examples of dierent minimal dieomorphisms giving the same C*-algebras,
Israel J. Math. 160(2007), 189217.
[R1] J. Renault, A groupoid approach to C -algebras, Lecture Notes in Mathematics, Vol. 793
Springer-Verlag Berlin, Heidelberg, New York, 1980.
[R2] J. Renault, The ideal structure of groupoid crossed product C*-algebras (avec un appendice
par G. Skandalis), J. Operator Theory 25 (1991), 3-36.
[R3] J. Renault, Cartan subalgebras in C -algebras, Irish Math. Soc. Bulletin 61 (2008), 2963.
[To] J. Tomiyama, The interplay between topological dynamics and theory of C -algebras, Lecture
Notes Series, 2, Global Anal. Research Center, Seoul 1992.
dOrl
D
epartment de Math
ematiques, Universite
eans, BP 6759, 45067 Orl
eans Cedex
2, France
E-mail address: Jean.Renault@univ-orleans.fr

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

A Denition of Compact C -Quantum Groupoids


Thomas Timmermann
Abstract. We propose a denition of compact quantum groupoids in the
setting of C -algebras and associate to every such quantum groupoid a fundamental unitary. These two notions are based on a new approach to relative
tensor products of Hilbert modules and to ber products of C -algebras. Using the fundamental unitary, we associate to every compact quantum groupoid
a dual Hopf C -bimodule and a measurable quantum groupoid in the sense
of Enock and Lesieur. Examples related to compact groupoids, r-discrete
groupoids, and center-valued traces are outlined.

1. Introduction
1.1. Overview. In the setting of von Neumann algebras, measurable quantum
groupoids in particular compact ones were studied by Enock and Lesieur [6,
5, 8], building on Vallins Hopf-von Neumann bimodules and pseudo-multiplicative
unitaries [20, 21] and Haagerups operator-valued weights.
In this article, we propose a denition of compact quantum groupoids in the
setting of C -algebras briey called compact C -quantum groupoids building on the notion of a Hopf-C -bimodule and a C -pseudo-multiplicative unitary
[15, 16, 17]. To each compact C -quantum groupoid, we associate a regular C pseudo-multiplicative unitary, a dual Hopf C -bimodule, and a measurable quantum groupoid. To illustrate the general theory, we outline several examples of
compact C -quantum groupoids: the C -algebra of continuous functions on a compact groupoid, the reduced C -algebra of an etale groupoid with compact base,
and principal compact C -quantum groupoids.
Further results on the dual Hopf C -bimodule of a compact quantum groupoid
and a detailed discussion of the examples listed above can be found in [19]. An article on the general framework of Hopf C -bimodules and C -pseudo-multiplicative
unitaries is in preparation [15].
1.2. Plan. This article is organized as follows. The denition a compact quantum groupoid in the setting of C -algebras and the necessary preliminaries are introduced in Sections 24. Recall that a measured compact groupoid consists of a
2000 Mathematics Subject Classication. Primary 46L55; Secondary 16W30, 46L08, 57T05.
supported by the SFB 478 Geometrische Strukturen in der Mathematik of the DFG
(Deutsche Forschungsgemeinschaft).
c Mathematical
0000
(copyright Society
holder)
c
2009
American

1
267

268
2

THOMAS TIMMERMANN

base space G0 , a total space G, range and source maps r, s : G G0 , a multiplication Gs r G G, a left and a right Haar system, and a quasi-invariant measure
on G0 . Roughly, the corresponding ingredients of a compact C -quantum groupoid
are unital C -algebras B and A, representations r, s : B popq A, a comultiplication
: A A A, a left and a right Haar weight , : A B popq , and a KMS-state
on B, subject to several axioms. We introduce these ingredients in several steps.
First, we focus on the tuple pB, A, r, , s, q, which can be considered as a compact C -quantum graph, and review some related constructions (Section 2). Next,
we construct the ber product A A and the underlying relative tensor product
of Hilbert modules [15, 16, 17] (Section 3). Finally, we give the denition of a
compact C -quantum groupoid and establish rst properties like uniqueness of the
Haar weights up to scaling (Section 4).
In Sections 57, we study further properties of compact C -quantum groupoids
and give some examples. First, we associate to every compact C -quantum groupoid
a fundamental unitary and, using that unitary, a dual Hopf C -bimodule and a
measurable quantum groupoid (Section 5). The fundamental unitary generalizes
the multiplicative unitaries of Baaj and Skandalis [1] and can be considered as a
particular pseudo-multiplicative unitary in the sense of Vallin [21]. Second, we
sketch examples of compact C -quantum groupoids related to center-valued traces
on C -algebras and to compact or etale groupoids (Section 6, 7).
1.3. Preliminaries. Let us x some general notation and terminology.
Given a subset Y of a normed space X, we denote by rY s X the closed linear
span of Y . Given a Hilbert space H and a subset X LpHq, we denote by X 1 the
commutant of X. Given a C -algebra A and a C -subalgebra B M pAq, we denote
by A X B 1 the relative commutant ta P A | ab ba for all b P Bu. All sesquilinear
maps like inner products of Hilbert spaces are assumed to be conjugate-linear in
the rst component and linear in the second one.
We shall make extensive use of (right) Hilbert C -modules [7].
Let A and B be C -algebras. Given Hilbert C -modules E and F over B,
we denote the space of all adjointable operators from E to F by LB pE, F q. Let
E and F be C -modules over A and B, respectively, and let : A LB pF q be a
-homomorphism. Recall that the internal tensor product E b F is a Hilbert C module over B [7, 4] and the closed linear span of elements b , where P E and
P F are arbitrary, and x b | 1 b 1 y x|px| 1 yq 1 y and p b qb b b
for all , 1 P E, , 1 P F , b P B. We denote the internal tensor product by
 and drop the index if the representation is understood; thus, for example,
E  F E  F E b F .
We also dene a ipped internal tensor product F E as follows. We equip
the algebraic tensor product F d E with the structure maps x d | 1 d 1 y :
x|px| 1 yq 1 y, p d qb : b d , form the separated completion, and obtain a
Hilbert C -B-module F E which is the closed linear span of elements ,
where P E and P F are arbitrary, and x | 1  1 y x|px| 1 yq 1 y and
p qb b  for all , 1 P E, , 1 P F , b P B. As above, we drop the index
and simply write  instead of  if the representation is understood.
Evidently, the usual and the ipped internal tensor product are related by a unitary

: F  E
E  F ,   .
Finally, let E1 , E2 be Hilbert C -modules over A, let F1 , F2 be Hilbert C modules over B with completely positive maps i : A LB pFi q (i 1, 2), and

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

269
3

let S P LA pE1 , E2 q, T P LB pF1 , F2 q such that T 1 paq 2 paqT for all a P A.


Then there exists a unique operator S  T P LB pE1  F1 , E2  F2 q such that
pS  T qp  q S  T for all P E1 , P F1 , and pS  T q S  T [3,
Proposition 1.34]
2. Compact C -quantum graphs
The rst ingredient in the denition of a compact C -quantum groupoid is a
compact C -quantum graph with a coinvolution. Roughly, the latter consists of a
C -algebra B with a faithful KMS-state, a C -algebra A, two compatible module
structures consisting of representations B, B op A and conditional expectations
A B, B op , and a -antiautomorphism of A that intertwines these module structures. Thinking of the underlying graph of a groupoid, these objects correspond
to the space of units with a quasi-invariant measure, the total space of arrows, the
range and the source map, the left and the right Haar weight, and the inversion of
the groupoid.
Before we can dene compact C -quantum graphs and coinvolutions, we have
to recall KMS-states on C -algebras, introduce module structures on C -algebras
with respect to such states, and present an associated GNS-Rieel-construction.
2.1. KMS-states on C -algebras and associated GNS-constructions.
Let be a faithful KMS-state on a C -algebra B [11, 8.12]. We denote by
the modular automorphism group, by H the GNS-space, by : B H the
GNS-map, by p1B q the cyclic vector, and by J : H H the modular
conjugation associated to . Recall that
(2.1)

J pbq pi{2
pbq q for all b P Dompi{2
q.

We omit explicit mentioning of the GNS-representation : B LpH q and identify B with pBq; thus, pbq b for all b P B.
We denote by B op the opposite C -algebra of B, which coincides with B as
a Banach space with involution but has the reversed multiplication, by B B op ,
b bop , the canonical antiisomorphism, and by op : B op C, bop pbq, the
opposite state of . Using formula (2.1), one easily veries that op is a KMS-state,
op
op

that the modular automorphism group is given by t pbop q t


pbqop for
all b P B, t P R, and that one can always choose the GNS-space and GNS-map for
op such that Hop H and op pbop q J pb q for all b P B. Then op ,

Jop J , op pbq J pbq J for all b P B, and for all b P Dompi{2


q, x P B,

op pbop q pi{2
pbqq,

bop pxq pxi{2


pbqq.

For later application, we note the extension to von Neumann algebras: The
state
on N : B 2 LpH q given by y x |y y is the unique normal extension
of , it is faithful because is cyclic for op pB op q N 1 , and the Hilbert space
H and the map : N H , y y , form a GNS-representation for
.
2.2. Module structures and associated Rieel constructions. We shall
use the following kind of module structures on C -algebras relative to KMS-states:
Definition 2.1. Let be a faithful KMS-state on a unital C -algebra B. A module structure on a unital C -algebra A consists of a unital embedding r : B A
and a faithful completely positive map : A B such that r : A rpBq is

270
4

THOMAS TIMMERMANN

a unital conditional expectation, : is a KMS-state, and t r r t ,


t t for all t P R.
Given a module structure as above, we can form a GNS-Rieel-construction:
Lemma 2.2. Let be a faithful KMS-state on a unital C -algebra B, let pr, q
be a -module structure on a unital C -algebra A, and put : .
(1) There exists a unique isometry : H H such that pbq prpbqq
for all b P B.
(2) J J , b rpbq , paq ppaqq, a paq for all
b P B, a P A.
(3) There exists a op -module structure pr op , op q on Aop such that r op pbop q
rpbqop and op paop q paqop for all b P B, a P A. For all b P B,
op pbop q op pr op pbop qq.
(4) Put N : B 2 LpH q, M : A2 LpH q. Then r extends uniquely to a
normal embedding r : N M , and extends uniquely to a faithful normal
y rpyq ,
completely positive map : M N . Moreover,
,

x pxq , px
rpyqq pxq
rpyq for all x P M , y P N .
Proof. (1)(3) The proof of assertion (1) is straightforward, and J J

because Dompi{2
q is dense in B and because J pbq pi{2
pbq q

prpi{2 pbq qq pi{2 prpbqq q J pbq for all b P Dompi{2 q. The proof of
the remaining assertions is routine.
(4) Since rpbqaop aop b for all a P A, b P B and rAop H s H ,
r is continuous with respect to the -weak topologies on LpH q and LpH q and
extends uniquely to r : M N as claimed. The map is uniquely determined by

qpxq
for
pxq
x for all x P M , and pxq x |x y x | x y p

all x P M . Since is faithful, so are , and necessarily also . The proof of the
remaining assertions is routine again.

2.3. Compact C -quantum graphs. We need the following simple variant
of a Radon-Nikodym derivative for KMS-states:
Lemma 2.3. Let A be a unital C -algebra with a faithful KMS-state and a
positive invertible element such that pq 1 and t pq for all t P R.
(1) The state on A given by paq p 1{2 a 1{2 q for all a P A is a faithful
KMS-state and t Adit t t Adit for all t P R.
(2) The map : A H , a pa 1{2 q, is a GNS-map for , and the
associated modular conjugation J is equal to J .

1, pq
(3) If P A is another positive invertible element satisfying pq
t


for all t P R, and , then .
Definition 2.4. A compact C -quantum graph is a tuple G pB, , A, r, , s, ,
q that consists of
(1) a unital C -algebra B with a faithful KMS-state and a unital C -algebra
A,
(2) a -module structure pr, q and a op -module structure ps, q on A, respectively, such that rpBq and spB op q commute,
(3) a positive, invertible, -invariant element P A X rpBq1 X spB op q1 satisfying pq 1 and op p q .

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

271
5

Given such a compact C -quantum graph, we put : , 1 : op and


denote by , : H H the isometries dened in Lemma 2.2.
Till the end of this section, let G be a compact C -quantum graph as above.
Since prpbqqcop prpbqspcop qq pspcop qrpbqq cop prpbqq and pspbop qqc
cpspbop qq for all b, c P B, we can dene completely positive maps
(2.2)

: r : B ZpB op q

and

: : s : B op ZpBq.

We identify ZpBq and ZpB op q with B X B op LpH q in the natural way.


Clearly, r r and 1 s op s op . The compositions
s : and 1 r op are related to op and , respectively, as follows.
Lemma 2.5.
(1) pq P B and p 1 q P B op are positive, invertible, cenop
tral, invariant with respect to and , respectively, and ppqq
1 op pp 1 qq.
(2) 1 r op pq and s : op
.
p 1 q
Proof. (1) We only prove the assertions concerning pq. Since is positive
and invertible, there exists an 0 such that 1A , and since is positive, we
can conclude pq p1A q 1B . Therefore, pq is positive and invertible. It
is central because bpq prpbqq prpbqq pqb for all b P B, and invariant
under because t ppqq pt pqq pq for all t P R.
(2) The rst relation holds because 1 prpbqq pp 1{2 rpbq 1{2 qq pbpqq
ppq1{2 bpq1{2 q for all b P B. The second relation follows similarly.

2.4. Coinvolutions. The unitary antipode of a compact C -quantum groupoid will be a coinvolution of the underlying compact C -quantum graph.
Definition 2.6. A coinvolution for G is a -antiautomorphism R : A A
satisfying R R idA and Rprpbqq spbop q, pRpaqq paqop for all b P B, a P A.
Lemma 2.7. Let R be a coinvolution for G.
1

(1) Rpq 1 , pq p 1 qop , R 1 , t R R t


for all t P R.
: op
(2) pbq pb q for all b P B.
(3) There exists a unique antiunitary I : H H , 1 paq pRpaq q,
and I paq pRpa 1{2 q q, Ia I Rpaq for all a P A, I 2 idH ,
I J , IJ J I.

Proof. (1) The last equation follows from the fact that R is an antiautomorphism and that R 1 . Lemma 2.3 (3) implies that the element 1 Rp 1 q
1 1
is equal to because p 1 q 1 p 1 q p1q 1, t p 1 q Rpt
p qq
Rp 1 q 1 and paq 1 paq pRpaqq 1 p 1{2 Rpaq 1{2 q p 11{2 a 11{2 q
for all a P A. Finally, pq p RqpRp 1 qqop p 1 qop .
(2) p sqpbop q p R R sqpbop q p rqpbqop for all b P B.
(3) The formula for I denes an antiunitary since pRpaqRpaq q pRqpa aq
1
pa aq for all a P A. The rst two equations given in (3) follow immediately. The
1
q,
remaining equations follow from the fact that for all a P A, b P B, c P Dompi{2
I 2 paq pRpRpa 1{2 q 1{2 q q pa 1{2 1{2 q paq,
I J pb q I op pbop q I 1 pspbop qq prpbq q pb q,
1

J I 1 paq pi{2
pRpaq q q pRpi{2
paq q q IJ 1 1 paq.

272
6

THOMAS TIMMERMANN

3. The relative tensor product and the ber product


Fundamental to our denition of a compact C -quantum groupoid are C modules and C -algebras over KMS-states, the relative tensor product of such C modules, and the ber product of such C -algebras. The ber product is needed to
dene the target of the comultiplication, and the relative tensor product is needed to
dene this ber product and the domain and the range of the fundamental unitary.
For proofs and further details, we refer to [15, 16, 17].
3.1. C -modules and C -algebras over KMS-states. We adopt the framework of C -modules and C -algebras over C -bases [15, 16, 17], but restrict to
C -bases associated to KMS-states. A C -base is a triple pH, B, B : q consisting of
a Hilbert space H and two commuting nondegenerate C -algebras B, B : LpHq.
Let be a faithful KMS-state on a C -algebra B. Then pH , B, B op q is a C -base,
where H is the GNS-space for and B and B op act on H Hop via the GNSrepresentations. Thus, we can reformulate the theory developed in [17] for concrete
KMS-states instead of general C -bases.
Definition 3.1. A C --module is a pair H pH, q, where H is a Hilbert
space and LpH , Hq is a closed subspace satisfying rH s H, rBs ,
and r s B LpH q. A morphism between C --modules H and K is an
operator T P LpH, Kq satisfying T , T . We denote the set of all such
morphisms by LpH , K q.
Lemma 3.2. Let H be a C --module.
(1) is a Hilbert C -B-module with inner product p, 1 q 1 .
(2) There exist isomorphisms  H H,  , and H  H,
 .
(3) There exists a nondegenerate representation : B op LpHq such that
pbop qpq bop for all b P B, P , P H .
(4) Let K be a C --module and T P LpH , K q. Then T pbop q pbop qT
for all b P B, and left multiplication by T denes an operator in LB p, q,
again denoted by T .

Let 1 , . . . , n be faithful KMS-states on C -algebras B1 , . . . , Bn .
Definition 3.3. A C -p1 , . . . , n q-module is a tuple pH, 1 , . . . , n q, where
H is a Hilbert space and pH, i q is a C -i -module for each i 1, . . . , n such
that ri pBiop qj s j whenever i j. In the case n 2, we abbreviate

H : pH, , q. The set of morphisms betweenC -p1 , . . . , n q-modules H


n
pH, 1 , . . . , n q, K pK, 1 , . . . , n q is LpH, Kq : i1 LpHi , Ki q LpH, Kq.
Remark 3.4. If pH, 1 , . . . , n q is a C -p1 , . . . , n q-module, then i pBiop q
LpHj q and, in particular, ri pBiop q, j pBjop qs 0 whenever i j.
Definition 3.5. A C -p1 , . . . , n q-algebra consists of a C -p1 , . . . , n q-module pH, 1 , . . . , n q and a nondegenerate C -algebra A LpHq such that i pBiop qA
is contained in A for each i 1, . . . , n. In the cases n 1, 2, we abbrevi,
ate A
: p H , Aq. A morphism of C -p1 , . . . , n q-algebras
H : pH , Aq, AH
A ppH, 1 , . . . , n q, Aq and C ppK, 1 , . . . , n q, Cq is a nondegenerate -homomorphism : A M pCq such that rI,i i s i for each i 1, . . . , n, where
I,i : tT P LpHi , Ki q | T a paqT for all a P Au. We denote the set of all such
morphisms by MorpA, Cq.

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

273
7

Remark 3.6. If is a morphism between C --algebras A


H and CK , then
op
op
p pb qq pb q for all b P B [16, Lemma 2.2].

3.2. The C -module of a compact C -quantum graph. For every compact C -quantum graph, the GNS-Rieel-construction in Lemma 2.2 yields a C module as follows. Let be a faithful KMS-state on a unital C -algebra B again.
Lemma 3.7. Let pr, q be a -module structure on a unital C -algebra A. Put
: , H : H ,
p : rA s, : rAop s, where is as in Lemma 2.2.
(1) p H is a C -p, op q-module and p r op , r.
(2) AH is a C -op -algebra.

(3) aop i{2


paq for all a P Dompi{2
q X rpBq1 .
(4) A ` pA X rpBq1 qop LpHp q and Aop ` pA X rpBq1 q LpH q.
Proof. (1) Lemma 2.2 implies that Hp is a C --module and H a C -module. The equations for p and follow from the fact that by Lemma
2.2, p pbop qa a bop arpbqop rpbqop a and pbqaop aop b
aop rpbq rpbqaop for all b P B, a P A. Hence, rp pB op qs rrpBqop Aop s
and r pBqp
s rrpBqA s
p, so p H is a C -p, op q-module.
(2) By (1), r pBqAs rrpBqAs A.
(3) Since t prpBqq rpBq for all t P R, restricts to a one-parameter group

of automorphisms of AXrpBq1 ; in particular, a P Dompi{2


qXrpBq1 is dense in AX
1
op

rpBq . Now, the claim follows from the fact that a pbq prpbqi{2
paqq

pi{2
paqrpbqq i{2
paq pbq for all a P rpBq1 X Dompi{2
q and b P B.
(4) We only prove the rst inclusion, the second one follows similarly. Clearly,
rAp
s
p, and by (3), rpA X rpBq1 qop
ps rApA X rpBq1 qop s rA s
p.

op

For a compact C -quantum graph, Lemmas 2.7 and 3.7 imply:


Proposition 3.8. Let G pB, , A, r, , s, , q be a compact C -quantum
graph. Put : , 1 : op and
(3.1)

H : H ,

p : rA s,

: rAop s,

p : rA s,

: rAop s.

p q is a C -p, op , op , q-module.
(1) pH,
p, , ,
(2) p r op , r, p sop , s.

op
(3) A,
H is a C -p , q-algebra.
(4) Let R be a coinvolution for G and dene I : H H by 1 paq

p
pRpaq q. Then I J
p and IJ .


3.3. The relative tensor product of C -modules. The relative tensor


product of C -modules over KMS-states is a C -algebraic analogue of the relative
tensor product of Hilbert spaces over a von Neumann algebra. We summarize the
denition and main properties; for proofs and further details, see [15, 16, 17].
Let be a faithful KMS-state on a C -algebra B and let H , K be a C -module and a C -op -module, respectively. The relative tensor product of H and
K is the Hilbert space H b K :  H  . It is spanned by elements   ,

where P , P H , P , and the inner product is given by x  | 1  1  1 y


x| 1 1 1 y x| 1 1 1 y for all , 1 P , , 1 P H , , 1 P .

274
8

THOMAS TIMMERMANN

Obviously, there exists a ip isomorphism


    .

: H b K K b
H,
op

The isomorphisms  H H,  , and H  K,  ,


of Lemma 3.2 induce the following isomorphisms, which we use without further
notice:
H  H b K  K,

    .

Using these isomorphisms, we dene the following tensor products of operators:


S b T : S  T P Lp  Kq LpH b Kq for all S P LpH q, T P pBq1 ,

S b T : S  T P LpH q LpH b Kq for all S P pB op q1 , T P LpK q.

Note that S  T S  id T S  T for all S P LpH q, T P LpK q.


For each P , P , there exist bounded linear operators
|y1 : K H b K,  ,

x|1 : |y1 : 1  p 1 q,

|y2 : H H b K,  ,

x|2 : |y2 :  1 p 1 q.

We put |y1 : t|y1 | P u and similarly dene x|1 , |y2 , x|2 .


Assume that H pH, 1 , . . . , m , q is a C -p1 , . . . , m , q-module and that
K pK, , 1 , . . . , n q is a C -pop , 1 , . . . , n q-module, where 1 , . . . , m , 1 , . . . , n
are faithful KMS-states on C -algebras A1 , . . . , Am , C1 , . . . , Cn . We put
i : r|y2 i s LpHi , H b Kq,

j : r|y1 j s LpHj , H b Kq

for all i 1, . . . , m, j 1, . . . , n. Then pH b K, 1 , . . . , m , 1 , . . . , n q

is a C -p1 , . . . , m , 1 , . . . , n q-module, called the relative tensor product of H and


K and denoted by H b K. For all i 1, . . . , m, a P Ai and j 1, . . . , n, c P Cj ,

op

pi q pa q i paop q b id,

pj q pcop q id b j pcop q.

The relative tensor product has nice categorical properties:


is a C -p1 , . . . , m , q-module, K

pH,

m , q
Bifunctoriality. If H
1, . . . ,

op

pK, , 1 , . . . , n q a C -p , 1 , . . . , n q-module, and S P LpH, Hq, T P LpK, Kq,

then there exists a unique operator S b T P LpH b K, H b Kq satisfying

pS b T qp   q S   T

for all P , P H , P .

We need the following straightforward analogue not mentioned in [17]:


be C --modules, K , H
C -op -modules, and
Lemma 3.9. Let H , K

J: K K
antiunitaries such that IJ and JJ .
I : H H,
such
b K
(1) There exists a unique antiunitary I b J : H b K H
J

op

that pI b Jqp   q IJ  J  JJ for all P , P H , P .


J

(2) pI b Jq|y1 |IJ y1 J and pI b Jq|y2 |JJ y2 I for all P ,


J

P .

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

275
9

(3) pI b JqpS b T q pISI b

JT J qpI b Jq for all S P LpH q, T P


op

LpK q.

Unitality. If we embed B, B into LpH q via the GNS-representations, then


U : pH , B op , Bq is a C -pop , q-module and the maps
op

H b B op H H,   bop bop ,

H B b K K, b   b,

are isomorphisms of C -p1 , . . . , m , q-modules and C -pop , 1 , . . . , n q-modules


H b U H and U b K K, respectively, natural in H and K.

Associativity. Assume that , 1 , . . . , l are faithful KMS-states on some C pK, , 1 , . . . , n , q is a C -pop , 1 , . . . , n , q-module and
algebras and that K
L pL, , 1 , . . . , l q a C -p op , 1 , . . . , l q-module. Then the isomorphisms
pH b Kq b L  K   H b pK b Lq

bL
are isomorphisms of C -p1 , . . . , m , 1 , . . . , n , 1 , . . . , l q-modules pH b Kq

b Lq. From now on, we identify the Hilbert spaces above and denote them
H b pK

by H b K b L.

3.4. The ber product of C -algebras. The ber product of C -algebras


over KMS-states is an analogue of the ber product of von Neumann algebras. We
summarize the denition and main properties; for proofs and further details, see
[15, 16, 17].
Let be a faithful KMS-state on a C -algebra B, let AH be a C --algebra,

and let CK
be a C -op -algebra. The ber product of AH and CK
is the C -algebra
A C : tx P LpH b Kq | x|y1 , x |y1 r|y1 Cs and x|y2 , x |y2 r|y2 Asu.

If A and C are unital, so is A C, but otherwise, A C may be degenerate.

Clearly, conjugation by the ip : H b K K b


H yields an isomorphism
op

Ad : A C C op A.

Assume that A pH, Aq is a C -p1 , . . . , m , q-algebra and C pK, Cq a


C -pop , 1 , . . . , n q-algebra, where 1 , . . . , m , 1 , . . . , n are faithful KMS-states
on some C -algebras and H pH, 1 , . . . , m , q, K pK, , 1 , . . . , n q. If A C

is nondegenerate, then pH b K, A Cq is a C -p1 , . . . , m , 1 , . . . , n q-algebra,

called the ber product of A and C and denoted by A C.

Aq
is a C -p1 , . . . , m , q-algebra and
Assume furthermore that A pH,

op
K
pH,


Cq
is a C -p , 1 , . . . , n q-algebra, where H
1, . . . ,
m , q,
C pK,

pK, , 1 , . . . , n q. Then for each P MorpA, Aq, P MorpC, Cq, there exists a
such that for all x P A C, S P
unique morphism P MorpA C, A Cq

q, T P LpK , K
q satisfying Sa paqS, T c pcqT for all a P A, c P C,
LpH , H

p qpxqpS b T q pS b T qx.

We shall need the following simple construction:

276
10

THOMAS TIMMERMANN

, C C -op -algebras, and


Lemma 3.10. Let AH , CK
be C --algebras, AH

K
and
R : A Aop , S : C C op -homomorphisms. Assume that I : H H

J : K K are antiunitaries such that IJ , JJ and Rpaq I a I,


Spcq J c J for all a P A, c P C. Then there exists a -homomorphism
op such that pR Sqpxq pI b Jq x pI b Jq
R S : A C pA op Cq

for all x P A C. This -homomorphism does not depend on the choice of I, J.

Proof. Evidently, the formula denes a -homomorphism R S. The deJ

nition does not depend on the choice of J because we have x|1 pR Sqpxq| 1 y1
J

J xIJ |1 x |I 1 J y1 J SpxI 1 J |1 x|IJ y1 q for all x P A C by Lemma 3.9 (2),

and a similar argument shows that it does not depend on the choice of I.

Unfortunately, the ber product need not be associative, but in our applications, it will only appear as the target of a comultiplication whose coassociativity
will compensate the non-associativity of the ber product.
4. Compact C -quantum groupoids
A compact C -quantum groupoid consists of a compact C -quantum graph
with a coinvolution and a comultiplication satisfying several relations, most importantly, left- and right-invariance of the Haar weights and a strong invariance
condition relating the coinvolution to the Haar weights and the comultiplication.
Before we give the precise denition, we recall the underlying notion of a Hopf
C -bimodule and the left- and right-invariance conditions; afterwards, we prove
some elementary properties of compact C -quantum groupoids.
4.1. Hopf C -bimodules over KMS-states. Let be a faithful KMS-state
on a C -algebra B.
Definition 4.1 ([17]). A comultiplication on a C -pop , q-algebra A,
H is a
,
,
,
morphism P MorpAH , AH AH q that makes the following diagram commute:

/ A A

id


A pA Aq

_


A A

id

/ pA Aq A  


/ LpH H Hq.

A Hopf C -bimodule over is a C -pop , q-algebra with a comultiplication.

Let pA,
H , q be a Hopf C -bimodule over . A bounded left Haar weight for
,
pAH , q is a non-zero completely positive contraction : A B satisfying
(1) pa pbqq paqb for all a P A, b P B, and
(2) px|1 paq| 1 y1 q ppaqq 1 for all a P A and , 1 P .
A bounded right Haar weight for pA,
H , q is a non-zero completely positive contraction : A B op satisfying

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

277
11

(1) pa pbop qq paqbop for all a P A, b P B, and


(2) px|2 paq| 1 y2 q ppaqq 1 for all a P A and , 1 P .

Remarks 4.2. Let pA,


H , q be a Hopf C -bimodule over .
op
op
(1) p pb q pcqq pcq b pb q for all b, c P B by Remark 3.6.

Let : A B be a completely positive contraction.


(2) If condition (1) above holds, then : A pBq is a conditional
expectation.
(3) If condition (2) above holds and rx|1 pAq|y1 s A, then also (1) holds
because px|1 paq| 1 y1 pbqq px|1 paq| 1 by1 q ppaqq 1 b
px|1 paq| 1 y1 qb for all a P A, b P B, , 1 P .
Similar remarks apply to conditions (1) and (2).
4.2. Denition of compact C -quantum groupoids. Given a compact
C -quantum graph pB, , A, r, , s, , q with coinvolution R, we use the notation
of Proposition 3.8, put : , 1 : op , J : J J 1 ,
H : H ,

p : rA s,
: rAop s,
p : rA s,
: rAop s,
and dene an antiunitary I : H H by I 1 paq pRpaq q for all a P A and a
-antihomomorphism R R : A A A op A by x pI b Iq x pI b Iq

(see Lemma 3.10).


Definition 4.3. A compact C -quantum groupoid is a compact C -quantum
graph pB, , A, r, , s, , q with a coinvolution R and a comultiplication for A,
H
such that
(1) rpAq|y1 s r|y1 As rpAq| y1 As and rpAq|y2 s r|y2 As
rpAq| y2 As;
(2) is a bounded left and a bounded right Haar weight for pA,
H , q;
(3) Rpx |1 paqpdop b 1q| y1 q x |1 paop b 1qpdq| y1 for all a, d P A.

Let pB, , A, r, , s, , , R, q be a compact C -quantum groupoid.


Lemma 4.4. ta P A X rpBq1 | paq 1 b au spB op q and ta P A X spB op q1 |

paq a b 1u rpBq.

Proof. We only prove the rst equation. Clearly, the right hand side is contained in the left hand side. Conversely, if a P A X rpBq1 and paq 1 b a, then

a x |1 paq| y1 sppaqq by right-invariance of .

4.3. The conditional expectation onto the C -algebra of orbits. We


study the maps : r : B ZpB op q, : : s : B op ZpBq introduced in
(2.2). As before, we identify ZpB op q and ZpBq with B X B op LpH q.
Proposition 4.5.
(1) and : are conditional expectations onto a C subalgebra of B X B op , and pbq : pbop q for all b P B. In particular,
s : op 1 r.
(2) s r and : .
(3) t t for all t P R.

(4) pbi{2
pdqq pdi{2
pdqq for all b, d P Dompi{2
q.

278
12

THOMAS TIMMERMANN

Lemma 4.6. Let b, c, e P B, d P Dompi{2


q and x rpbqspcop q, y rpdqspeop q.

Then x |1 pxqpy op b 1q| y1 rp pbi{2 pdqqqrpeqspcopq.

pdqq pbi{2
pdqq, so
Proof. By Lemma 3.7, rpbqrpdqop rpbi{2

x |1 pxqpy op b 1q| y1 x |1 prpbqy op b spcop qq| y1

p rpbqrpdqop speop qop qspcop q


rp rpbqrpdqop eqspcop q

pdqqqrpeqspcopq.
rp pbi{2

Proof of Proposition 4.5. (1), (2) Left- and right-invariance of , imply


psppaqqq sppaqq x |2 paq| y2
x |1 paq| y1 rppaqq prppaqqq
for all a P A. Therefore, : and : : p rq r .
Next, s r because sp pbqq spprpbqqq x |1 prpbqq| y1 rp pbqq for
all b P B by Lemma 4.6. In particular, for all b, c, d P B,
pbq pcq pdq pbqprpcqq pdq psp pbqqrpcqsp pdqqq
prp pbqc pdqqq p pbqc pdqq.
op

(3), (4) Let t P R. Then t p pBqq pBq because t t


r
1

t r t r r t . Therefore, restricts to the modular


automorphism group for the trace : | pBq , which is id pBq , and hence t

q. By Lemma 4.6 and Denition 4.3 (3),


t t . Let b, d P Dompi{2

pdqqq x |1 prpbqqprpdqop b 1q| y1


rp pbi{2

op
Rpx |1 prpdqqprpbq b 1q| y1 q

sp pdi{2
pbqqq.

Since s r and r is injective, we can conclude pbi{2


pdqq pdi{2
pbqq. 

4.4. The modular element. The modular element of a compact C -quantum


groupoid can be described in terms of the element : pq p 1 q P B X B op
(see Lemmas 2.5, 2.7) as follows.
Proposition 4.7. rpqspq1 and pq b .

Proof. By Lemma 2.5 (1), : rpqspq1 is positive, invertible, and invariant


with respect to . Moreover, 1 paq p1{2 a1{2 q for all a P A because
1 pspq1{2 aspq1{2 q op p 1{2 paq 1{2 q p s qpaq
p 1 r qpaq p 1{2 paq 1{2 q prpq1{2 arpq1{2 q
and pq
for all a P A by Proposition 4.5 and Lemma 2.5. By Lemma 2.3, ,
1
1
1
rpq b spq rpq p q b pqspq b because P B X B op . 

An important consequence of the preceding result is that for every compact


C -quantum groupoid, there exists a faithful invariant KMS-state on the basis:
Corollary 4.8. p qop .

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

279
13

Proof. We get p 1{2 paq 1{2 q prpq1{2arpq1{2 q 1 pspq1{2 aspq1{2 q



p 1{2 paq 1{2 q for all a P A.
op

Therefore, we could in principle restrict to compact C -quantum groupoids


with trivial modular element 1A .
The KMS-state can be factorized into a state on the commutative C algebra pBq ZpBq and a perturbation of as follows. We dene maps
1 : B pBq, b p 1{2 b 1{2 q, | pBq : pBq C, b p 1{2 b 1{2 q.
Note that p 1 q 1 because pq prpqspq1q p 1 q.
Proposition 4.9. 1 .
Proof. By Propositions 4.5 and 4.7, pbq prpbqq 1 p 1{2 rpbq 1{2 q
p 1{2 prp 1{2 b 1{2 qq 1{2 q p 1 qpbq for all b P B.

op

4.5. Uniqueness of the Haar weights. The Haar weights of a compact C quantum groupoid are not unique but can be rescaled by elements of B as follows.
For every positive P B op , the map spq : A B given by a pspq1{2 aspq1{2 q
is a bounded left Haar weight for pA,
H , q because
spq px|1 paq| 1 y1 q px|1 p1 b spq1{2 qpaqp1 b spq1{2 q| 1 y1 q

px|1 pspq

1{2

asp

1{2

qq| y1 q spq paq 1

for all a P A, , 1 P . Similarly, for every positive P B, the map rpq : A B op


given by a prpq1{2 arpq1{2 q is a bounded right Haar weight for pA,
H , q.
s, ,
q
be a compact C -quantum
Theorem 4.10.
(1) Let pB, , A, r, ,

graph and a bounded left Haar weight for pA,


H , q. Then ,
1 1

where p q .
s, ,
q
be a compact C -quantum graph and a bounded
(2) Let pB, , A, r, ,
q
1 .
right Haar weight for pA, , q. Then , where p
H

op

1 :
Proof. We only prove (1), the proof of (2) is similar. Put : ,
1

, : p q. Let a P A. Then

(4.1)

|1 paq| y1 q prppaqqq.
psppaqqq
px

We apply to the left hand side and nd, using Lemma 2.5 (2),
1{2 aspq
1{2 q p 1{2 spq
1{2 aspq
1{2 1{2 q.
ppaqq 1 pspq
psppaqqq op

ppaqq

Next, we apply to the right hand side of (4.1) and nd 1 prppaqqq


1{2
1{2
1
1{2
1{2
prpq arpq q. Since rpqspq , we can conclude pspq aspq q paq
for all a P A and in particular

pb paqq
prpbq aq pspq1{2 rpbq aspq1{2 q pb pspq1{2 aspq1{2 qq

for all b P B, a P A. Since is faithful, paq


pspq1{2 aspq1{2 q for all a P A.

280
14

THOMAS TIMMERMANN

5. The fundamental unitary


In the theory of locally compact quantum groups, a fundamental r
ole is played
by the associated multiplicative unitaries, whose theory was developed by Baaj,
Skandalis [1] and Woronowicz [22]. We shall associate to every compact C quantum groupoid a C -pseudo-multiplicative unitary [17] that can be considered
as a generalized multiplicative unitary. This unitary will be used to prove that
the coinvolution of a compact C -quantum groupoid reverses the comultiplication,
to construct a generalized Pontrjagin dual of the compact C -quantum groupoid
in form of a Hopf C -bimodule, and to associate to every compact C -quantum
groupoid a measured quantum groupoid in the sense of Enock and Lesieur [4, 8].
5.1. C -pseudo-multiplicative unitaries. The notion of a C -pseudo-multiplicative unitary extends the notion of a multiplicative unitary [1], of a continuous
eld of multiplicative unitaries [2] and of a pseudo-multiplicative unitary on C modules [9, 18], and is closely related to pseudo-multiplicative unitaries on Hilbert
spaces [21]; see [17, Section 4.1]. We give the precise denition and the main
properties; for proofs and details, see [15, 17]. Let be a faithful KMS-state on a
C -algebra B.
Definition 5.1. A C -pseudo-multiplicative unitary over consists of a C p , q and a unitary V : H p b H H b H such that
p , , op q-module pH, ,
op
op

V p q ,

V pp q p ,

p ,
p
V pp q

V p q

and the following diagram commutes:


V b id
op

H p b
H p b
H
op
op

/ H b H p b
H
op

id bV

/ H b H b H,

id b V

V bid

op


H p b
pH

b Hq
op

pH p b
Hq b H
op

id b

23

op


H p b
H

b
H
op
op

V b id
op

/ H b H
`

b H
p

op

where 23 is given by pH q

pH p q , p q p q.
p
p , q, V q be a C -pseudo-multiplicative unitary. We abbreviate the
Let ppH, ,
operators V b
id and V b id by V12 , the operators id bV and id bV by V23 , and
op

qV12 23 by V13 . Thus, the indices indicate those positions in a relative


pid b
op

tensor product where the operator acts like V . We put

p 1 LpHq.
p q : x|2 V |y2 LpHq,
ApV q : x|1 V |y
ApV
p q LpH q, ApV q LpH p q, so that we can dene
The assumptions on V imply ApV

p LpH p b Hq,
pV : A

op

aqV,
p
a V pid b p

V : A LpH b Hq,

a V pap b
idqV .
op

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

281
15

p , q, V q is regular if rx|1 V |y2 s r s, and wellDefinition 5.2. ppH, ,


p
p V q and pApV q, , V q are Hopf C -bimodules over op and
p q, ,
behaved if pApV
H
H
, respectively.
We cite the following result [17, Theorem 4.14]:
Theorem 5.3. Every regular C -pseudo-multiplicative unitary is well-behaved.
5.2. The fundamental unitary of a compact C -quantum groupoid.
Throughout this section, let pB, , A, r, , s, , , R, q be a compact C -quantum
groupoid. We use the same notation as in the preceding section.
Theorem 5.4. There exists a unique regular C -pseudo-multiplicative unitary
p , q, V q over such that V |a y1 paq| y1 for all a P A.
ppH, ,
Uniqueness is evident. The proof of existence proceeds in several steps.
Proposition 5.5.

(1) There exists a unitary V : H p b


H H b H
op

such that V |a y1 paq| y1 for all a P A.


(2) V pa  dop q paqp  dop q for all a, d P A.
p ,
p V pp
(3) V pp q p , V pp q
pq
p, V pp
q
p .
Proof. (1) Let a, a1 P A, , 1 P H. Since is a right-invariant,
xpaqp  q|pa1 qp  1 qypH b Hq x|x |1 pa a1 q| y1 1 y

x| p a a1 q 1 y
xa  |a1  1 ypH p b Hq .
op

Therefore, there exists an isometry V : H p b


H H b H satisfying V |a y1
op

paq| y1 for all a P A. The relation rpAq|y2 s r|y2 As implies V pp q


rV |A y1 s rpAq| y1 s rpAq|y2 s r|y2 A s p , whence V is
surjective and unitary.
(2) By Proposition 4.7, we have for all a, d P A
V pa  dop q V pa 1{2  dop q
V pa 1{2  dop 1 q
pa 1{2 qp  dop 1{2 q
paqp 1{2 b 1{2 qp 1  dop 1{2 q paqp  dop q.

(3) The rst relation was already proven above. Since rpAq| y1 As r|y1 As,
p
p rV |A y1 A s rpAq| y1 A s r|y1 A s p and similarly
V p q
p
V p
pq
p. Finally, by (2), for all b P B and a, d P A
V |aop y2 d b V pdrpbq  aop q pdrpbqqp  aop q
pdqprpbq  aop q pdq|aop y2 b
and hence rV |y2
ps rV |Aop y2 A s rpAq|Aop y2 s rpAq|y2 s
p .

r|y2 A s
Condition (3) of Denition 4.3 yields the following inversion formula for V :

282
16

THOMAS TIMMERMANN

Theorem 5.6. V pJ b IqV pJ b Iq.


J

Proof. Put V : pJ b IqV pJ b Iq. Then for all a, b, c, d P A


J

xa  bop 1 |V pcop 1 dop qy xpaqp  bop 1 q|cop  dop y


x  1 |pa qpcop b 1qp  pbop q dop qy

x 1 |x |1 pa qpc

op

op
b 1q| y1 pdb q y,

xa  bop 1 |V pcop 1  dop qy xpa qop  Ibop 1 |V pc 1  Idop J qy


x  Ibop 1 |paop b 1qpc qp  Idop qy

x  I

|paop

b 1qpc

qp

 Ipbop q dop qy

xI 1 |x |1 paop b 1qpc q| y1 Ipdb qop y

x 1 |Ix |1 paop b 1qpc q| y1 Ipdb qop y.

Now, the claim follows from condition (3) in Denition 4.3.

Proof of Theorem 5.4. By Lemma 3.9 and Propositions 3.8, 5.5, multiplication by pJ p b IqV pJ p b Iq acts on subspaces of LpH , H p b
Hq as follows:
op
J

pJ p b Iq

pJ p b Iq

V
p 1 s J
p
r|y2 s r|y2 s
r|y
r|y1 IJ s r|y1 s,

pJ p b Iq

pJ p b Iq

r|y2 s r|y2
ps r|y2
ps r|y2 J
pJ s r|y2 s,
pJ p b Iq

pJ p b Iq

p r|y1
p
p 1 s
p 1
r|y
ps r|y
ps r|y1 I
pJ s r|y1 s.
p .
p
Theorem 5.6 implies V p q , V p q , V pp q
Next, we prove that V23 V12 V12 V13 V23 . Let a, d P A, P H and p2q
p idq pid q , 13 paq 23 ppaq
b
idq23 . Then
p
op

V23 V12 pa  d  q V23 ppaqp b


idqp  d  q
op

p2q

paqp  pdqp  qq

V12 13 paqp  pdqp  qq


V12 V13 pa  pdqp  qq V12 V13 V23 pa  d  q.
Finally, V is regular since by Theorem 5.6, Lemma 3.9 and Proposition 3.8,
rx|1 V |y2 s rx|1 pJ p b IqV pJ p b Iq|y2 s
J

p 1 V |y2 Js
rIx|
rIx |1 pAq|y2 Js
rIx |1 |y2 AJs rIJ J AJs r s.

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

283
17

p , q, V q
By Theorem 5.3, the regular C -pseudo-multiplicative unitary ppH, ,
p
p q, ,
p V q.
yields two Hopf C -bimodules pApV q, , V q and pApV
H

Proposition 5.7. pApV

q,
H , V

pA,
H , q.

p 1 s rx|1 pAq| y1 s rAx|1 | y1 s


Proof. We have ApV q rx|1 V |y

op

rA p qs rAspB qs A, V paq V pap b



1qV paq for all a P A.
op

p q,p ,
p V q will be discussed in the next subsection.
The Hopf C -bimodule pApV
H
Our rst application of the unitary V will be to prove that the coinvolution
reverses the comultiplication. We need the following lemma:
p
Lemma 5.8.
(1) px|1 V | 1 y1 q x|1 V12 V13 | 1 y1 for all P , 1 P .
1
1
1
p
(2) Rpx|1 V | y1 q xJ J |1 V |JJ y1 for all P , P .
p px|1 V | 1 y1 q V ppx|1 V | 1 y1 q p b 1qV
Proof. (1) For all P , 1 P ,
op

1
| y1 x|1 V12 V13 | 1 y1 ; see also [17, Lemma 4.13].
x|1 V23 V12 V23
p Rpx|1 V | 1 y1 q
(2) Lemma 3.9 and Theorem 5.6 imply that for all P , 1 P ,
1

1
Ix |1 V |y1 I xJ J |1 pJ b Iq V pJ b Iq |JJ y1 xJ J |1 V |JJ y1 . 
J

Theorem 5.9. pR Rq Ad R.
J

p By Lemma 5.8,
Proof. Let P and 1 P .
pAd pR Rq qpx|1 V | 1 y1 q pAd pR Rqqpx|1 V12 V13 | 1 y1 q
J


V12 |y1 pI b Iqq.
Ad ppI b Iq x 1 |1 V13
J

By Lemma 3.9 (2), we can rewrite this expression in the form



Ad pxJ 1 J |1 pJ p b pI b IqqV13
V12 pJ b I b Iq |JJ y1 q.
J

By Lemmas 3.9, 5.8 and Theorem 5.6 this expression is equal to


Ad pxJ 1 J |1 V13 V12 |JJ y1 q xJ 1 J |1 V12 V13 |JJ y1
pxJ 1 J |1 V |JJ y1 q pRpx|1 V | 1 y1 qq.

Remarks 5.10.
(1) One can prove the existence of a regular C -pseudomultiplicative unitary ppH, , ,
pq, W q satisfying W |a y2 paq| y2
for all a P A and express this unitary in terms of V as follows: W
pI p b IqV pI b Iq; see [19, Theorem 5.10].
J

(2) Using Theorem 5.9, one can prove the following


analogue of condition(3)
`
in Denition 4.3; see [19, Lemma 4.7]: R x |2 paqp1 b dop q| y2

x|2 p1 b aop qpdq| y2 for all a, d P A. If we would replace the former

condition by the latter, we could develop the same theory using W instead
of V and nally conclude that also the former condition holds.

284
18

THOMAS TIMMERMANN

p q,p ,
pV q
5.3. The dual Hopf C -bimodule. The Hopf C -bimodule pApV
H
p
obtained from ppH, , , q, V q can be considered as the generalized Pontrjagin dual
p : ApV
p q.
of our initial compact C -quantum groupoid. Let us describe A
Proposition 5.11.
(1) For each a P A, there exists an operator paq P
LpHq such that paq pdq px |2 pdq|aop y2 q for all d P A, and
paq JpRpaqqJ.
(2) xxop |2 V |y op y2 pyx q for all x, y P A.
p rpAqs.
(3) A
p: A
p A,
p p
(4) There exists a -antiautomorphism R
a Jp
a J.
p is the closed linear span of all operators of the form
Proof. By denition, A
xxop |2 V |y op y2 , where x, y P A. Let x, y, d P A and put a yx . Then
xxop |2 V |y op y2 d xxop |2 V pd  y op q
xxop |2 pdqp  y op q px |2 pdq|aop y2 q.
This calculation proves the existence of the operators paq for all a P A and that
p rpAqs. By Theorem 5.6, Lemma 3.9 and Proposition 3.8,
A
paq pxxop |2 V |y op y2 q
xy op |2 pJ b IqV pJ b Iq|xop y2
J

JxIy op J |2 V |Ixop J y2 J
JxRpy qop |2 V |Rpx qop y2 J JpRpxq RpyqqJ JpRpaqqJ.

Remarks 5.12.

p is a coin(1) Using Theorem 5.6, one can show that R

` ,p
p V in the sense that it reverses the comultiplication:
p ,
volution of A
H
p Ad pR
p p Rq
pV R
p
p V ; see [19, Corollary 7.6].

` ,p
op
p , LpKqB,B , given
(2) Put K : H . There exists a morphism p
P Mor A
K
H

` ,p
p ,
p V in the
by paq paq for all a P A, which is a counit for A
H
sense that the maps
pV : A
p LpKqB op A
p LpK B op b Hq LpHq,
pp
op idq
op
op

pV : A
pA
p p B LpKq LpH p b B Kq LpHq
q
pid op p
op
op

are equal to the embedding A LpHq; see [19, Proposition 7.7].


5.4. The passage to measurable quantum groupoids. The compact C quantum groupoid pB, , A, r, , s, , , R, q can be completed to a measurable
quantum groupoid in the sense of Enock and Lesieur [4, 8] as follows.
Put N : B 2 LpH q, N op : pB op q2 N 1 LpH q, M : A2 LpHq
s, the normal extensions of , r, , s, ; see Lemma 2.2.
and denote by
, r, ,
The comultiplication V extends uniquely to a normal -homomorphism
: M LpH b Hq via x V px p b 1qV . To obtain a Hopf-von Neumann

op

we need to identify pM

bimodule [20] pN, M, r, s, q,


q with a subalgebra of the
ber product M s
rM [13] which acts on the relative tensor product H s b rH [14].

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

285
19

Let us recall the denition of H s b rH and of M srM . Put

DprH;
q : t P H | DC 0@y P N : }
r pyq} C}y }u.
q if and only if the map N H,
Thus, an element P H belongs to DprH;
y rpyq, extends to an operator R pq P LpH , Hq. Clearly, R pq R p 1 q P
N 1 for all , 1 P DprH;
q. The space H s b rH is the separated completion of

q with respect to the sesquilinear form


the algebraic tensor product H d DprH;
dened by x d | 1 d 1 y x|
spR pq R p 1 qq 1 y for all , 1 P H, , 1 P
DprH;
q. We denote the image of an element d in H s b rH by s b r. Clearly,

q DprH;
q, and for each x, y P M 1 , there exists a well-dened oprpN q1 DprH;
erator xs b ry LpH s b rHq such that pxs b ryqp s b rq x s b ry for all P H,

q. Now, M s
rM pM 1 s b rM 1 q1 LpH s b rHq.
P DprH;

Lemma 5.13.
(1) aop P DprH;
q, R paop q aop P for all a P A.
(2) There exist inverse isomorphisms

H b H H 
H s b rH
 H H b H

such that p  a q s b ra , p s b rq  R pq for all


op

op

q, P H .
P H, a P A, P , P DprH,
Proof. (1) For all a P A, y P N , rpyqaop aop rpyq aop y .
(2) and are well-dened inverse isometries because
}  aop }2 x| p paop q aop qy
x|
spR paop q R paop qqy } s b raop }2 ,

} s b r} x|
spR pq R pqqy

x| R pq R pqy
x|R pq p qR pqy }  R pq}2 ,
p qp  aop q  R paop q  aop  aop


for all , a, , , as above.

as a map M
We identify H b H with H s b rH via , and consider

LpH s b rHq.

,

,
Theorem 5.14. pN, M, r, s, ,
q is a measurable quantum groupoid.
Proof. First, the relation pAq A A pA1 b idq X pid b A1 q [17,

and the fact that


Lemma 3.8] implies pM
q M srM , and the denition of

is a Hopf-von
V is a C -pseudo-multiplicative unitary imply that pN, M, r, s, q
Neumann bimodule.
Second, one has to check that and are left- and right-invariant, respectively.
This follows from the fact that these maps are normal extensions of and , which
are left- and right-invariant, respectively.

286
20

THOMAS TIMMERMANN

Finally, one has to check that the modular automorphism groups of



1
op
1

commute, but this follows from the fact that 1{2 .



and
6. Principal compact C -quantum groupoids
Principal compact C -quantum groupoids are particularly simple examples of
compact C -quantum groupoids. We give the denition and discuss some of the
main properties. For proofs and further details, see [19].
Recall that a compact groupoid G is principal if the map G G0 G0 given
by x prpxq, spxqq is injective or, equivalently, if CpGq rr pCpG0 qqs pCpG0 qqs.
This condition can be carried over to compact C -quantum groupoids as follows:
Definition 6.1. A compact C -quantum groupoid pB, , A, r, , s, , , R, q
is principal if A rrpBqspB op qs.
To simplify the following discussion, we only consider the case where 1A ,
which is not a serious restriction; see Corollary 4.8.
Essentially, a principal compact C -quantum groupoid is completely determined by the conditional expectation : B pBq ZpBq introduced in Subsection 4.3. The rst result in this direction is the following proposition:
Proposition 6.2. Let pB, , A, r, , s, , 1A q be a compact C -quantum graph
such that A rrpBqspB op qs. Then the following two conditions are equivalent:
(1) There exist R, such that pB, , A, r, , s, , 1A , R, q is a compact C quantum groupoid.
(2) pbq : pbop q for all b P B, : B pBq is a conditional expectation,

, r s , and pbi{2
pdqq pdi{2
pbqq for all b, d P

Dompi{2 q.

To every compact groupoid G, we can associate a principal compact groupoid
whose total space is tprpxq, spxqq | x P Gu. Likewise, we can associate to every
compact C -quantum groupoid a principal one:
Corollary 6.3. Let pB, , A, r, , s, , 1A , R, q be a compact C -quantum
R| . Then there exists
groupoid and put A rrpBqspB op qs, |A , |A , R
A

1 , R,

q
is a principal
a unique -homomorphism such that pB, , A, r, , s, ,
A

compact C -quantum groupoid.



A principal compact C -quantum groupoid can be reconstructed from the conditional expectation as follows. Assume that
C is a commutative unital C -algebra with a faithful state ,
B is a unital C -algebra with a -module structure p, q such that pCq
ZpBq.
We put : , identify C with pCq via , dene an isometry : H H as
in Lemma 3.7, and put : rB s LpH , H q, op : rB op s LpH , H q.
Theorem 6.4. There exists a unique principal compact C -quantum groupoid
pB, , A, r, , s, , 1A , R, q such that A B op b B op LppH q op b pH qq and

rpbq b op b 1op , pb op b cop q b pcq, spcop q 1 op b cop , pb op b cop q

pbqcop for all b, c P B.

Essentially, every principal compact C -quantum groupoid is of the form above:

A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

287
21

Proposition 6.5. Let pB, , A, r, , s, , 1A , R, q be a principal compact C quantum groupoid.


(1) C : pBq is a commutative unital C -algebra, : |C is a faithful state
on C, pid, q is a -module structure on B, and .
Denote by : H H the isometry c c . Put : rB s, op : rB op s.
(2) There exists a unitary : H pH q op b pH q such that for all b, c P B,

prpbqop spcop qop q bop   c , prpbqspcop q q b   cop .


(3) Ad restricts to an isomorphism A B op b B op such that rpbqspcop q

b op b cop for all b, c P B.

7. Examples related to groupoids


Prototypical examples of compact C -quantum groupoids are the function algebra of a compact groupoid and the reduced groupoid C -algebra of an r-discrete
groupoid with compact space of units. We outline these examples; for proofs and
further details, see [19]. For some background on groupoids, see [10, 12].
Let G be a locally compact, Hausdor, second countable groupoid with unit
space G0 , range and source maps rG , sG : G G0 , left Haar system , and associ1
ated right Haar system 1 . For each u P G0 , put Gu : rG
puq, Gu : s1
G puq. Let
1
0
G be a probability measure on G with full support and dene measures G , G
on G such that for all f P Cc pGq,


1
f dG
f pxq du pxq dG puq,
f dG

f pxqd1
u pxq dG puq.
G

G0

Gu

G0

Gu

We impose the following assumptions:


(1) the space of units G0 is compact;
1
are equivalent;
(2) G is quasi-invariant in the sense that G and G
1
(3) the Radon-Nikodym derivative D dG {dG is continuous.
To equip the function algebra and the reduced groupoid C -algebra of G with
the structure of Hopf C -bimodules, we use a C -pseudo-multiplicative unitary
naturally associated to G [15, 17]. This unitary is constructed as follows.
Denote by the trace on CpG0 q given by f G0 f dG , put H : L2 pG, G q,
and dene Hilbert C -CpG0 q-modules L2 pG, q, L2 pG, 1 q to be the completions
of the pre-C -module Cc pGq, where for all , 1 P Cc pGq, u P G0 , f P CpG0 q, x P G,

1
x |ypuq
1 pxqpxqdu pxq,
pf qpxq pxqf prG pxqq in case of L2 pG, q,
u
G
1
2
1
x |ypuq
1 pxqpxqd1
q.
u pxq, pf qpxq pxqf psG pxqq in case of L pG,
Gu

There exist embeddings j : L2 pG, q LpH , Hq and j : L2 pG, 1 q LpH , Hq


such that pjpqqpxq pxqprG pxqq and pjpqqpxq pxqDpxq1{2psG pxqq for all
P Cc pGq, P CpG0 q, x P G, and with : jpL2 pG, qq and : jpL2 pG, 1 qq,
the tuple pH, , , q is a C -pop , , op q-module.
The relative tensor products H b
H and H b H can be described as follows.
op

2
2
on G2s,r : tpx, yq P G G | sG pxq rG pyqu and r,r
on
Dene measures s,r

288
22

THOMAS TIMMERMANN

G2r,r : tpx, yq P GG | rG pxq rG pyqu such that for all f P Cc pG2s,r q, g P Cc pG2r,r q


2
f ds,r
:
f px, yq dsG pxq pyq du pxq dG puq,
G2s,r

G0

2
g dr,r
:

G2r,r

G0

Gu

GsG pxq

Gu

gpx, yq du pyq du pxq dG puq.

Gu

Then there exist isomorphisms


2
2
2
: H b
H L pGs,r , s,r q,
op

2
: H b H L2 pG2r,r , r,r
q

such that for all , P Cc pGq, P Cc pG q, px, yq P

G2s,r ,

px1 , y 1 q P G2r,r ,

pjpq   jpqqpx, yq pxqDpxq1{2 psG pxqqpyq,


pjpq   jpqqpx1 , y 1 q px1 qprG px1 qqpy 1 q.
2
2
2
2
2
2
We identify H b
H with L pGs,r , s,r q and H b H with L pGr,r , r,r q via , .
op

Theorem 7.1. There exists a unique regular C -pseudo-multiplicative unitary


ppH, , , q, V q such that pV qpx, yq px, x1 yq for all P Cc pG2s,r q, px, yq P
G2r,r , P Cc pGq, z P G.

, p
p
The Hopf C -bimodules pApV q,
H , V q and pApV qH , V q can be described as
follows. Embed C0 pGq into LpHq via the representation given by multiplication
operators, and denote by Cr pGq the reduced groupoid C -algebra of G, that is, the
closed linear span of all operators Lpgq P LpHq, where g P Cc pGq and

gpzqDpzq1{2 f pz 1 xqdrG pxq pzq for all f P Cc pGq, x P G.


pLpgqf qpxq
GrG pxq

(1) ApV q Cr pGq and

gpzqDpzq1{2 pz 1 x, z 1 yqdrG pxq pzq


pV pLpgqqqpx, yq

Proposition 7.2.

GrG pxq

for all g P Cc pGq, P Cc pG2r,r q, px, yq P G2r,r .


p q C0 pGq and p
p V pf qqpx, yq f pxyqpx, yq for all f P C0 pGq,
(2) ApV
2
P Cc pGs,r q, px, yq P G2s,r .

Using the preceding result, it is not dicult to prove the following theorems:
Theorem 7.3. If G is compact, there exists a compact C -quantum groupoid
p V , Rq such that rpf q r pf q, spf q s pf q,
pCpG0 q, , CpGq, r, , s, , D1,
G
G

gpyqdu pyq,
ppgqqpuq
gpyqd1
ppgqqpuq
u pyq
Gu

and pRpgqqpxq gpx

Gu
0

q for all f P CpG q, g P CpGq, u P G0 , x P G.

Theorem 7.4. Let G be r-discrete and let be the family of counting measures.
Embed CpG0 q into Cc pGq by extending functions outside of G0 by 0. Then there
exists a compact C -quantum groupoid pCpG0 q, , Cr pGq, , , , , Lp1G0 q, R, V q
such that pf q Lpf q, pLpgqq g|G0 , RpLpgqq Lpg : q for all f P CpG0 q,
g P Cc pGq, where g : pxq gpx1 q for all x P G.


A DEFINITION OF COMPACT C -QUANTUM GROUPOIDS

289
23

References
[1] Saad Baaj and Georges Skandalis, Unitaires multiplicatifs et dualit
e pour les produits crois
es

ebres, Ann. Sci. Ecole


Norm. Sup. (4) 26 (1993), no. 4, 425488. MR MR1235438
de C -alg`
(94e:46127)

[2] Etienne
Blanchard, D
eformations de C -alg`
ebres de Hopf, Bull. Soc. Math. France 124
(1996), no. 1, 141215. MR 97f:46092
[3] Siegfried Echterho, S. Kaliszewski, John Quigg, and Iain Raeburn, A categorical approach
to imprimitivity theorems for C -dynamical systems, Mem. Amer. Math. Soc. 180 (2006),
no. 850, viii+169. MR MR2203930
[4] Michel Enock, Continuous elds of C*-algebras and Lesieurs Quantum Groupoids, preprint,
September 2006, arXiv:OA/0605512.
, Quantum groupoids of compact type, J. Inst. Math. Jussieu 4 (2005), no. 1, 29133.
[5]
MR MR2115070 (2006c:46059)
[6] Michel Enock and Jean-Michel Vallin, Inclusions of von Neumann algebras, and quantum
groupoids, J. Funct. Anal. 172 (2000), no. 2, 249300. MR 2001h:46115
[7] E. C. Lance, Hilbert C -modules. a toolkit for operator algebraists, London Mathematical Society Lecture Note Series, vol. 210, Cambridge University Press, Cambridge, 1995.
MR MR1325694 (96k:46100)
[8] Franck Lesieur, Measured quantum groupoids, M
emoires de la SMF 109 (2007).
[9] Moto Ouchi, Pseudo-multiplicative unitaries on Hilbert C -modules, Far East J. Math. Sci.
(FJMS), Special Volume, Part II (2001), 229249. MR 2002i:46060
[10] Alan L. T. Paterson, Groupoids, inverse semigroups, and their operator algebras, Progress
in Mathematics, vol. 170, Birkh
auser Boston Inc., Boston, MA, 1999. MR MR1724106
(2001a:22003)
[11] Gert K. Pedersen, C -algebras and their automorphism groups, London Mathematical Society Monographs, vol. 14, Academic Press Inc. [Harcourt Brace Jovanovich Publishers],
London, 1979. MR MR548006 (81e:46037)
[12] Jean Renault, A groupoid approach to C -algebras, Lecture Notes in Mathematics, vol. 793,
Springer, Berlin, 1980. MR MR584266 (82h:46075)
[13] Jean-Luc Sauvageot, Produits tensoriels de Z-modules et applications, Operator algebras
and their connections with topology and ergodic theory (Busteni, 1983), Lecture Notes in
Mathematics, vol. 1132, Springer, Berlin, 1985, pp. 468485. MR MR799587 (87d:46067)
[14] M. Takesaki, Theory of operator algebras. II, Encyclopaedia of Mathematical Sciences, vol.
125, Springer-Verlag, Berlin, 2003. MR MR1943006 (2004g:46079)
[15] Thomas Timmermann, C -pseudo-multiplicative unitaries and Hopf C bimodules, in preparation.
, C -pseudo-Kac systems and and duality for coactions of concrete Hopf C [16]
bimodules, Tech. Report 484, SFB 478, WW-University Muenster, October 2007,
arXiv:0709.4617.
, C -pseudo-multiplicative unitaries, Tech. Report 481, SFB 478, WW-University
[17]
Muenster, September 2007, arXiv:0709.2995.
, Pseudo-multiplicative unitaries on C -modules and Hopf C -families, J. Noncomm.
[18]
Geom. 1 (2007), 497542.
[19]
, Compact C -quantum groupoids, Tech. Report 503, SFB 478, WW-University Muenster, October 2008, arXiv:0810.3771.
[20] Jean-Michel Vallin, Bimodules de Hopf et poids op
eratoriels de Haar, J. Operator Theory 35
(1996), no. 1, 3965. MR 97f:46095
, Unitaire pseudo-multiplicatif associ
e `
a un groupode. Applications `
a la
[21]
moyennabilit
e, J. Operator Theory 44 (2000), no. 2, 347368. MR 2002c:22004
[22] S. L. Woronowicz, From multiplicative unitaries to quantum groups, Internat. J. Math. 7
(1996), no. 1, 127149. MR MR1369908 (96k:46136)
t Mu
nster, Einsteinstr.
Fachbereich Mathematik und Informatik, W.W.-Universita
nster, Germany
62, 48149 Mu
E-mail address: timmermt@math.uni-muenster.de

This page intentionally left blank

Contemporary Mathematics
Volume 503, 2009

Complex dynamical systems and associated C -algebras


Yasuo Watatani
Abstract. We overview our study on the interplay between complex dynamical systems and operator algebras. Iteration of a rational function R gives
. Since the Riemann
a complex dynamical system on the Riemann sphere C
is decomposed to the union of the Julia set JR and Fatou set FR , we
sphere C
OR (JR ) and OR (FR ) by considering R as
associate three C -algebras OR (C),
JR and FR respectively. We show how properties of R
dynamical systems on C,
as complex dynamical systems are related with the structure of the associated
C -algebras and their K-groups. One of our aims is to analyze the singularity
structure of the branched points in terms of operator algebras. For example, we
show that the extreme KMS states are parameterized by the branched points.
We also associate C -algebras with algebraic correspondences and study their
relations.
This is based on joint works with M. Izumi and T. Kajiwara.

1. Introduction
P (z)
with relatively prime polyLet R be a rational function of the form R(z) = Q(z)
nomials P and Q. The degree of R is denoted by N = deg R := max{deg P, deg Q}.
C
on
We regard a rational function R as a N -fold branched covering map R : C
= C {}. The sequence (Rn )n of iterations of R gives a
the Riemann sphere C
The Fatou set FR of R is the maximal
complex analytic dynamical system on C.
n

open subset of C on which (R )n is equicontinuous (or a normal family), and the


The Fatou set FR is a
Julia set JR of R is the complement of the Fatou set in C.
stable part and the Julia set JR is an unstable part.
We also study group actions. We identify the group of linear fractional transformations (Mobius transformations)

az + b
(ad bc = 1, a, b, c, d C),
cz + d
with P SL(2, C). Then a discrere subgroup of P SL(2, C) is called a Kleinian
such that is a normal
group. The ordinary set of is the open set of z C
family on some neibourhood of z, and the limit set of is the complement of
R(z) =

1991 Mathematics Subject Classication. 46L55, 46L08, 37F05, 37F10.


Key words and phrases. rational function, complex dynamical system, branched point, Julia
set, purely innite C -algebra, Hilbert C -bimodule.
c
c 2009
American Mathematical Society
XXXX

1
291

292
2

YASUO WATATANI

The ordinary set is a stable part and the limit set is


the ordinary set in C.
an unstable part.
Recall that Sullivan discovered a dictionary between the theory of complex
analytic iteration and the theory of Kleinian groups in [46]. Sullivans dictionary
shows a strong analogy between the limit set of a Kleinian group and the
Julia set JR of a rational function R. We associate C -algebras with these two
classical objects and study the interplay between complex dynamical systems and
associated C -algebras. We shall also conrm Sullivans dictionary in the level of
C -algebras.
For any group action : Homeo() on a compact Hausdor space , we
can associate a C -algebra C() as the crossed product. The interplay between
topological dynamical systems of homeomorphisms and the associated C -crossed
products have been fruitful. See a book [47] by Tomiyama, for example. In the
case of the complex dynamical system given by iteratitons of a single Mobius transformation, the relation with type I C -algebras and representations were studied
in Silvestrov-Tomiyama [44] and Silvestrov-Wallin [45]. For a regular self-covering
map : , Deaconu [10] associate an r-discrete groupoid and study their
groupoid C -algebra C (, ) of Renault [42]. The same equivalence relation was
considered in Arzumanian-Vershik [2]. Moreover for a branched covering : ,
Deaconu-Muhly [11] also introduced a C -algebra C (, ) as the C -algebra of
the r-discrete groupoid by avoiding the branched points. In particular, they study
In order to capture information of
rational functions R on the Riemann sphere C.

the branched points, in [28] we introduced slightly dierent C -algebras OR (C),


OR (JR ) and OR (FR ) associated with a rational function R acting on the Riemann
the Julia set JR and the Fatou set FR of R respectively. These algebras are
sphere C,
consturcted as Cuntz-Pimsner algebras of C -bimodules (or C -correspondences)
using the graph of R including branched points. For example, the algebra OR (JR ) is
generated by C(JR ) and {Sf ; f C(graph R|JR )}. We regard the algebra OR (JR )
as a certain analog of the crossed product C( )  of C( ) by a boundary
action of a Kleinian group . In fact the crossed product is generated by C( )
and {g : g }. Moreover commutation relations (a)Sf = Sf a for a C(JR )
and g (a)g = g a for a C( ) are similar, where (a)(x) = a(R(x)) and
g (a)(x) = a(g 1 x).
Recall that C. Delaroche [1] and Laca-Spielberg [32] showed that a certain
boundary action of a Kleinian group on the limit set yields a simple nuclear purely
innite C -algebra as groupoid C -algebra or crossed product. Dutkay-Jorgensen
[12] study a spectral theory on Hilbert spaces built on general nite-to-one maps.
Carlsen-Silvestrov [9] study C -algebras associated with regular covering maps and
show that topological freeness is equivalent to several conditions including maximal
abelianess.
Our C -algebras are also related with C -algebras of irreversible dynamical
systems of Exel [13], Exel-Vershik [14] and, C -algebras associated with subshifts
of Matsumoto [34], graph C -algebras of Kumjian-Pask-Raeburn-Renault [25], and
their generalizations for topological relations of Brenken [4], topological graphs
of Katsura [22], [23], and topological quivers of Muhly-Solel [35] and of MuhlyTomforde [36]. Some of our C -algebras are isomorphic to C -algebras associated
with self-similar sets [29] and Mauldin-Williams graphs [20].

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

293
3

Table 1. a dictionary
Kreinian group
C -algebra
Rational function R
Julia set JR
limit set
OR (JR )
unstable
purely innite
unstable, self-similar
Fatou set FR
ordinary set
OR (FR )
no wandering domains thm.
Ahlfors niteness thm.
?
the number of generators
?
N = deg R
deg R 2
non elementary
not C(T)
orbit structure
orbit structure
?
invarinat measure +
KMS state
invariant measure +
branched points (singularity)
?
extreme KMS states
?
K-groups
?
?
?
gauge action
Blaschke product
Fuchsian group
?
Hausdor dimension
Hausdor dimension
?
We show an incomplete dictionary between complex dynamical systems and
C -algebras.
Let R be a nite Blaschke product of degree at least two with R(0) = 0.
Then there exists a relation between the associated composition operator CR on
the Hardy space and the C -algebra OR (JR ) as in Hamada-Watatani [18]. It is
interesting to know the case of general rational functions.
2. Branched points
Let R be a rational function with N = deg R. We always assume that a
rational function R is not a constant function. If deg R = 1 and the Julia set
JR is not empty, then JR is, in fact, one point. Therefore the coecient algebra
A = C(JR ) = C and the Hilbert C -bimodule is just X = C(graphR|JR ) = C.
Thus the classical Toeplitz algebra appears in this case and the associated C algebra OR (JR ) becomes C(T). Therefore throught the paper we assume that
deg R 2. Then the Julia set JR is an uncountable perfect set. Therefore the
coecient algebra A = C(JR ) is innite dimensional.
Recall that a branched point (or critical point) of R is a point z0 at which R
is not locally one to one. It is a zero of R or a pole of R of order two or higher.
The image w0 = R(z0 ) is called a branch value (or critical value) of R. Using
appropriate local charts, if R(z) = w0 + c(z z0 )n + (higher terms) with n 1 and
c = 0 on some neighborhood of z0 , then the integer n = e(z0 ) = eR (z0 ) is called the
branch index of R at z0 . Thus e(z0 ) 2 if z0 is a branched point, and e(z0 ) = 1 if
z0 is not. Therefore R is an e(z0 ) : 1 map in a punctured neighborhood of z0 . By
the Riemann-Hurwitz formula, there exist 2N 2 branched points counted with
multiplicity, that is,

(e(z) 1) = 2 deg R 2.

zC

we have
Furthermore for each w C,

zR1 (w)

e(z) = deg R.

294
4

YASUO WATATANI

Let BR be the set of branched points of R and R(BR ) be the set of the branch values
\ R(BR ) is a N : 1 regular
\ R1 (R(BR )) C
of R. Then the restriction R : C
\ R(BR ) has an
covering, where N = deg R. This means that any point y C
open neighborhood V such that R1 (V ) has N connected components U1 , . . . , UN
and the restriction R|Uk : Uk V is a homeomorphism for k = 1, . . . , N . Thus
R has N analytic local cross sections Sk = (R|Uk )1 . But if y is in R(BR ), then
there exist no such open neighborhood V . This fact causes many diculties to
analyze the associated C -algebra, since we include the branched points to construct
the C -correspondence. Even if we will construct the associated groupoid, the rdiscreteness is not satised in general. This is the reason why we associate our
C -algebras by Cuntz-Pimsner construction. One of our aims is to analyze the
singularity structure of the branched points in terms of operator algebras.
3. Construction of the associated C -algebras
Since a rational function R of degree at least two is not a homeomporphism, we
can not use crossed product construction. We replace crossed pruduct construction
by Cuntz-Pimsner construction to obtain the associated C -algebra.
We recall Cuntz-Pimsner algebras [39]. Let A be a C -algebra and X be a
Hilbert right A-module. We denote by L(X) be the algebra of the adjointable
bounded operators on X. For , X, the rank one operator , is dened by
, () = (|) for X. The closure of the linear span of rank one operators is
denoted by K(X).
A family (ui )iI in X is called a basis, (or a tight frame more precisely as in
[15]) of X if

x=
ui (ui |x)A for any x X,
iI

where the sum is taken as unconditional norm convergence, see [26]. Furthermore
(ui )iI is called a nite basis if (ui )iI is a nite set. If A has a unit and X has
a nite basis, then X is algebraically nitely generated and projective over A and
K(X) = L(X).
We say that X is a Hilbert C -bimodule (or C -correspondence) over A if X
is a Hilbert right A- module with a homomorphism : A L(X). In this note,
n
we assume that X is full and is injective. Let F (X) =
be the full
n=0 X
0
Fock module of X with the convention X = A. For x X, the creation operator
Tx L(F (X)) is dened by
Tx (a) = xa

and Tx (x1 xn ) = x x1 xn .

We dene iF (X) : A L(F (X)) by


iF (X) (a)(b) = ab

and iF (X) (a)(x1 xn ) = ((a)x1 ) xn

for a, b A. The Cuntz-Toeplitz algebra TX is the C -subalgebra of L(F (X)) generated by iF (X) (a) with a A and Tx with x X. Let jK : K(X) TX be the homomorphism dened by jK (x,y ) = Tx Ty . We consider the ideal IX := 1 (K(X)) of
A. Let JX be the ideal of TX generated by {iF (X) (a)(jK )(a); a IX }. Then the
Cuntz-Pimsner algebra OX is the the quotient TX /JX . Let : TX OX be the
quotient map. Put Sx = (Tx ) and i(a) = (iF (X) (a)). Let iK : K(X) OX be
the homomorphism dened by iK (x,y ) = Sx Sy . Then ((jK )(a)) = (iK )(a)
for a IX . We note that the Cuntz-Pimsner algebra OX is the universal C -algebra

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

295
5

generated by i(a) with a A and Sx with x X satisfying that i(a)Sx = S(a)x ,


Sx i(a) = Sxa , Sx Sy = i((x|y)A ) for a A, x, y X and i(a) = (iK )(a) for
a IX . We usually identify i(a) with a in A. If X has a nite basis {u1 , u2 , . . . , un },
then the Cuntz-Pimsner algebra OX is the universal C -algebra generated by i(a)
with a A and Sx with x X satisfying that i(a)Sx = S(a)x , Sx i(a) = Sxa ,
Sx Sy = i((x|y)A ) for a A, x, y X and Su1 Su1 + Su2 Su2 + + Sun Sun = I. If
X has a countable
nbasis {u1 , u2 , . . . }, then the last condition should be replaced by
i(a) = limn k=1 i(a)Suk Suk under the operator norm convergence for any a
n
(a) = limn k=1 (a)uk ,uk
IX . Since (a) K(X), we automatically have 
n
under the operator norm convergence, because ( k=1 (a)uk ,uk )n is an approximately units for K(X).
There exists an action : R Aut OX with t (S ) = eit S , which is called
the gauge action. Since we assume that : A L(X) is isometric, there is an
embedding n : L(X n ) L(X n+1 ) with n (T ) = T idX for T L(X n ) with
the convention 0 = : A L(X). We denote by FX the C -algebra generated
by all K(X n ), n 0 in the inductive limit algebra lim L(X n ). Let Fn be the

C -subalgebra of FX generated by K(X k ), k = 0, 1, . . . , n, with the convention


F0 = A = K(X 0 ). Then FX = lim Fn .

and X = C(graph R) be the set of continuous functions on C

Let A = C(C)
2
; y = R(x)} is the graph of
and graph R respectively, where graph R = {(x, y) C
R. Then X is an A-A bimodule by
(a b)(x, y) = a(x)(x, y)b(y),

a, b A, X.

We dene an A-valued inner product ( | )A on X by




(|)A (y) =
e(x)(x, y)(x, y), , X, y C.
xR1 (y)

Thanks to the branch index e(x), the inner product above gives a continuous function and X is a full Hilbert bimodule over A without completion. The left action
of A is unital and faithful.
Since the Julia set JR is completely invariant under R, i.e., R(JR ) = JR =
R1 (JR ), we can consider the restriction R|JR : JR JR , which will be often
denoted by the same letter R. Let graph R|JR = {(x, y) JR JR ; y = R(x)} be
the graph of the restriction map R|JR and X(JR ) = C(graph R|JR ). In the same
way as above, X(JR ) is a full Hilbert bimodule over C(JR ). Since the Fatou set FR
is also completely invariant, X(FR ) := C0 (graph R|FR ) is a full Hilbert bimodule
over C0 (FR ).
is dened as the Cuntz-Pimsner algebra of the
Denition The C -algebra OR (C)
When the Julia set JR is not
Hilbert bimodule X = C(graph R) over A = C(C).
empty (for example deg R 2), we dene the C -algebra OR (JR ) as the CuntzPimsner algebra of the Hilbert bimodule X = C(graph R|JR ) over A = C(JR ).
When the Fatou set FR is not empty, the C -algebra OR (FR ) is dened similarly.
4. Ideal and simplicity
is decomposed
Let R be a rational function. Then the Riemann sphere C
into two parts FR and JR . The stable part FR is called the Fatou set of R and
the unstable part JR is called the Julia set. We shall consider the corresponding

296
6

YASUO WATATANI

which was rst pointed out


decomposition for the associated C -algebra OR (C),

by Deaconu and Muhly [11] in the case of their construction. Let B = C(C)
and consider the ideal I = {b B; b|JR = 0} of B, so I
= C0 (FR ) and B/I
=
C(JR ). We consider a submodule and quotient module of a Hilbert bimodule
Y = C(graph R) over B. The right Hilbert I-module Y I := {f b Y ; f Y, b I}
is also described as YI := {f Y ; (f |g)B I for all g Y }. Since (f |f )B I
means that

(f |f )B (y) =
e(x)|f (x, y)|2 = 0
xR1 (y)

for all y JR , we have


Y I = {f Y ; f (x, y) = 0 for all (x, y) graph R|JR }.
because JR is complete invariant.
Invariant ideals for bimodules are introduced in Kajiwara-Pinzari-Watatani [27]
and developed by Fowler-Muhly-Raeburn [17] in general case. In our situation I is
a Y -invariant ideal of B, i.e., (I)Y Y I. In fact, the condition is equivalent to
that (f |(a)g)B I for any a I and f, g Y , and it is easily checked as

e(x)f (x, y)a(x)g(x, y) = 0
(f |(a)g)B (y) =
xR1 (y)
1

for y JR , because x R (JR ) = JR and so a(x) = 0 for a I.


Therefore Y /Y I is naturally a Hilbert bimodule over B/I
= C(JR ). We can
identify Y /Y I with a bimodule X = C(graph R|JR ) over A = C(JR ). Applying a
result of Fowler-Muhly-Raeburn [17, Corollary 3.3], we have a natural decomposi on the Riemann sphere C.

tion of the C -algebra OR (C)


I = {b
Theorem 4.1 ([28]). Let R be a rational function, B = C(C),
B; b|JR = 0}
= C0 (FR ) and A = C(JR )
= B/I. Then the ideal I(I) generated by I
is Morita equivalent to OR (FR ) and we have the following exact sequence:
in OR (C)
OR (JR ) 0
0 I(I) OR (C)
The Julia set JR is an unstable part and have a certain self-similarity property.
We can translate this property by the language of C -algebra.
Theorem 4.2 ([28]). Let R be a rational function with deg R 2. Then the
C -algebra OR (JR ) associated with R on the Julia set JR is simple and purely
innite.
Remark Let R be a rational function with deg R 2. Then the C -algebra
OR (JR ) is separable and nuclear, and satises the Universal Coecient Theorem.
Therefore the isomorphism class of OR (JR ) is completely determined by the Kgroups together with the class of the unit by the classication theorem of KirchbergPhillips [24], [40]. The only simplicity of OR (JR ) can be obtained by a result of
Schweizer [43].
5. Some examples and computation of their K-groups
If rational functions R1 and R2 are topologically conjugate, then their C and OR (C)
are isomorphic. Therefore their K-groups Ki (OR (C))

algebras OR1 (C)


2
1

and Ki (OR2 (C)) are isomorphic. Similarly Ki (OR1 (JR1 )) and Ki (OR2 (JR2 )) are
isomorphic. Moreover if rational functions R1 and R2 are topologically conjugate,

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

297
7

then the gauge actions are also conjugate. Therefore the K-groups of the xed
point algebras are also topologically conjugate invariant. We investigate what kind
of information of complex dynamical systems is described by the K-theory.
We calculate the K-groups by the following six-term exact sequence due to
Pimsner [39].
K0 (IX )


id[X]

K0 (OR (C))

K0 (C(C))

0


K1 (OR (C))

K1 (C)

id[X]

K1 (IX )

on the Riemann sphere


We show that the K-groups of the C -algebra OR1 (C)
depend only on the number # BR of the branched points of R.
Theorem 5.1. Let R be a rational function with deg R 2. Then

K0 (OR (C))
=Z

BR +1


and K1 (OR (C))
= Z.

Remark. But we have no general formulae for Ki (OR (JR )) at the present moment.

=
Example. Let R(z) = z n for n 2. Then BR = {0, }. Therefore K0 (OR (C))

Z.
The
Julia
set
J
is
the
unit
circle
T.
The
map

:
Z3 and K1 (OR (C))
=
R
C(T) L(X) can be identied with the n-times around embedding. The Hilbert
bimodule XR over A = C(T) is isomorphic to An as a right A-module. In fact,
let ui (z, w) = 1n z i1 for i = 1, . . . , n. Then (ui |uj )A = i,j I and {u1 , . . . , un }
is a basis of XR . Hence Si := Sui , (i = 1, ..., n) are generators of the Cuntz
algebra On . We see that (z ui )(z, R(z)) = ui+1 (z, R(z)) for i = 1, ..., n 1 and
(z un )(z, R(z)) = z n = (u1 z)(z, R(z)). The left multiplication by z is a unitary
and denoted by U . Therefore C -algebra OR (JR ) associated with the complex
dynamical system on the Julia set JR = T is the universal C -algebra generated by
a unitary U and n isometries S1 , ..., Sn satisfying S1 S1 + +Sn Sn = I , U Si = Si+1
for i = 1, ..., n 1 and U Sn = S1 U . We have K0 (OR (JR )) = Z Z/(n 1)Z
and K1 (OR (JR )) = Z. The xed point algebra OR (JR ) by the gauge action
is a Bunce-Deddence algebra of type n . Therefore K0 (OR (JR ) )
= Z[ n1 ] and

K1 (OR (JR ) ) = Z.
We remark that Yamashita studies circle correspondences C -algebras and constructs Rieel like concrete projections of OR (JR ) in [48].
Example. Let R(z) = z 2 2. Then the Julia set JR is the interval [2, 2] and
it contains a critical point 0. Since IX = {a C([2, 2]); a(0) = 0}, K0 (IX ) =
K1 (IX ) = 0. Applying the above six term exact sequence with K0 (A) = Z and
K1 (A) = 0, we have K0 (OR (JR )) = Z and K1 (OR (JR )) = 0. Since the identity
I of OR (JR ) represents the generator of K0 (OR (JR )) = Z, the algebra OR (JR ) is
isomorphic to the Cuntz algebra O . (JR , R) is topologically conjugate to a tent
map ([0, 1], h) dened by

2x,
0 x 12 ,
h(x) =
1
2x + 2,
2 x 1.
Example. (quadratic polynomial) Let Pc (z) = z 2 +c. If c is not in the Mandelbrot
set M := {c C; Pcn (0) is bounded }, then OPc (JPc ) is isomorphic to the Cuntz

298
8

YASUO WATATANI

algebra O2 . In fact (JPc , Pc ) is topologically conjugate to the full two shift. If c is


2
in the interior of the main cardioid C = { z2 z4 C; |z| < 1}, then the Julia set
is homeomorphic to the unit circle S 1 and Pc on the Julia set JPc is topologically
conjugate to P0 (z) = z 2 on S 1 by the structural stability of hyperbolic maps Pc
for c in the interior of the main cardioid. Hence We have K0 (OPc (JPc )) = Z and
K1 (OPc (JPc )) = Z.
Example. Let R be a rational function with N = deg R 2. Let z0 be a
(super)attracting xed point of R. If all of the critical points of R lie in the
immediate attracting basin of z0 , then OR is isomorphic to the Cuntz algebra ON .
In fact (JR , R) is topologically conjugate to the full N-shift by [3, Theorem 9.8.1.
2
]. For example, If R(z) = 2z z1 , then OR (JR )
= O2 .
Example. Tchebychev polynomials Tn are dened by cos nz = Tn (cos z). For
example, T1 (z) = z, T2 (z) = 2z 2 1. Then the Julia set JTn is the interval
[1, 1] and JTn contains n 1 critical points for n 2. Since K0 (A) = Z,
K1 (A) = 0, K0 (IX ) = 0 and K1 (IX ) = Zn2 , we have K0 (OTn (JTn )) = Zn1
and K1 (OTn (JTn )) = 0. Recall that, if the Julia set of a polynomial P of degree
n 2 is the interval [1, 1], then P = Tn or P = Tn , see [3, page 11].
Example. We consider a rational function R by Lattes such that the Julia set is
(z 2 +1)2

the entire Riemann sphere. Let R(z) = 4z(z


2 1) . Then JR = C contains six critical
points. Thus K0 (OR (JR )) = Z7 and K1 (OR (JR )) = Z.
6. Branched points and KMS states
We study the gauge action on the C -algebra . If rational functions P and Q
are topological conjugate, then the associated bimodules are isomorphic. Therefore
) and (OQ (C),
) are conjugate in the sense
the C -dynamical systems (OP (C),
OQ (C)
such that t = t for
that there exists an isomorphism : OP (C)
any t. Hence the structure of the KMS states for the gauge action is an invariant
for complex dynamical systems of rational functions up to topological conjugacy.
The structure of the KMS states reects that of the singular points of R. We can

completely classify the KMS states for the gauge action of OR (C).
If R has no
exceptional points, then the gauge action has a phase transition at = log deg R in
the following sense: In the region 0 < log deg R, no KMS-state exists. A unique
KMS-state exists at = log deg R, which is of type III1/ deg R and corresponds to
the Lyubich measure. The extreme -KMS states at > log deg R are parameterized by the branched points of R and are factor states of type I. If R has exceptional
points, then there appear additional -KMS states for 0 < log deg R parameterized by exceptional points. We can recover the degree of R, the number of
branched points, the number of exceptional points from the structure of the KMS
states. The orbits of exceptional points are distinguished by 0-KMS states. Here
we dene a 0-KMS state to be an -invariant tracial state.
For example, let P (z) = z 3 , Q(z) = z 3 + 1 and R(z) = (1/18)z 3 (1/2)z 2 + 6.
We denote by EP (resp. BP ) the set of exceptional points (resp. branched points)
of P . Then EP = BP = {0, }, EQ = {}, BQ = {0, }, ER = {} and BR =
), (OQ (C),
)
{0, 6, }. Our results show that C -dynamical systems (OP (C),

and (OR (C), ) are not conjugate to each other.

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

299
9

We recall that Olsen-Pedersen [38] showed that a -KMS state for the gauge
action of the Cuntz algebra On exists if and only if = log n and that log n-KMS
state is unique. Since then, several authors have discussed KMS states for the
gauge action. See the references in [19]. Thanks to a result due to Laca-Neshveyev
[31], we can describe the structure of the KMS states in terms of a certain PerronFrobenius type operator studied in [41] and [31].
We need to describe the exceptional points and the Lyubich measure to state
is an exceptional point for R if the backward orbit
our theorem. A point z in C

O (z) of z is nite. We denote by ER the set of exceptional points. It is known


that ER is a subset of FR BR . A rational function R of degree at least two has
at most two exceptional points by the Riemann-Hurwitz formula. The reader is
referred to [3, section 4.1] for basic properties of exceptional points,
We recall the (so called) Lyubich measure L studied by Brolin [5], Freire
Lopes-M
ane [16] and Lyubich [33]. Let x be the Dirac measure on x for x C.
\ ER and each n N, we dene a probability measure y on the
For any y C
n
by
Riemann sphere C
yn =

N n e(x)e(R(x)) . . . e(Rn1 (x))x .

xRn (y)

The sequence (yn )n converges weakly to a measure L , which is called the Lyubich
\ ER . The support
measure. The measure L is independent of the choice of y C
of L is the Julia set JR and L is the unique invariant measure of maximal entropy.
Now we can state a classication theorem of the extreme KMS states as in [19].
A direct and simple proof using concrete construction of bases is explained in this
proceedings by Kajiwara [21].
Theorem 6.1 ([19]). Let R be a rational function with N = degR 2.
Then the set
Consider the gauge action on the associated C -algebra OR (C).
ex(K ()) of extreme -KMS states of are completely classied as follows:
(Case 1.) Suppose that R has two exceptional points ER = {w1 , w2 } such that
R(w1 ) = w1 and R(w2 ) = w2 . Then the branched points B(R) = {w1 , w2 }.
(1) If = 0, then the set of extreme -invariant traces consists of two points
,w1 and ,w2 such that the restriction ,wi |C(C)
is the Dirac measure
wi for i = 1, 2 .
(2) If 0 < < log N , then ex(K ()) consists of two points ,w1 and ,w2
such that the restriction ,wi |C(C)
is the Dirac measure wi for i = 1, 2 .
(3) If = log N , then ex(K ()) consists of three points L , w1 and w2 ,
where the restriction L |C(C)
is the Lyubich measure and the restriction
wi |C(C)
is the Dirac measure wi for i = 1, 2.
(4) If log N < , then ex(K ()) consists of two points ,w1 and ,w2 such
that the restriction ,wi |C(C)
is the Dirac measure wi for i = 1, 2 .
(Case 2.) Suppose that R has two exceptional points ER = {w1 , w2 } such that
R(w1 ) = w2 and R(w2 ) = w1 . Then B(R) = {w1 , w2 }.
(1) If = 0, then the set of extreme -invariant traces consists of one point
1
1
w1 ,w2 such that the restriction w1 ,w2 |C(C)
corresponds to 2 w1 + 2 w2 .

300
10

YASUO WATATANI

(2) If 0 < < log N , then ex(K ()) consists of two points ,1 and ,2
correspond to the measures
such that the restriction to C(C)
,1 =

1
e
e
1
,2
+
and

w +
w .
w
w
1
2
e + 1
e + 1
e + 1 1 e + 1 2

(3) If = log N , then ex(K ()) consists of three points L , ,1 and ,2 ,


correspond to the Lyubich measure L , ,1
where the restriction to C(C)
,2
and .
(4) If log N < , then ex(K ()) consists of two points ,1 and ,2 , where
correspond to the ,1 and ,2 .
the restriction to C(C)
(Case 3.) Suppose that R has one exceptional point, i.e., E(R) = {w}.
(1) If = 0, then the set of extreme -invariant traces consists of one point
,w such that the restriction ,w |C(C)
is the Dirac measure w .
(2) If 0 < < log N , then there exists a unique -KMS state ,w and its
restriction ,w |C(C)
corresponds to the Dirac measure w .
(3) If = log N , then ex(K ()) consists of two points L and w , where the
correspond to the Lyubich measure L and the Dirac
restriction to C(C)
measure w .
(4) If log N < , then ex(K ()) is parameterized by the branched points
B(R). For w B(R), the corresponding extreme -KMS state ,w is
determined by its restriction ,w |C(C)
. The corresponding measure ,w
is given by
on C
,w = m,w


1
ek

k=0

z ,

zRk (w)

where m,w is the normalized constant.


(Case 4.) Suppose that R has no exceptional points.
(1) If = 0, then there exists no -invariant traces.
(2) If 0 < < log N , then there exist no -KMS states.
(3) If = log N , then there exists a unique -KMS state and its restriction
|C(C)
is the Lyubich measure.
(4) If log N < , then ex(K ()) is parameterized by the branched points
B(R) as in case 3.
In any case the extreme -KMS state L corresponding to the Lyubich measure
is a factor state of type III1/N . All the other extreme -KMS states are factor
states of type I.
L

Since Theorem 6.1. is the complete classication, we can recapture some information of sigularity of the original complex dynamical system from the associated
) as follows:
C -algebra with the gauge action (OR (C),
Theorem 6.2 ([19]). Let P and Q be rational functions of degree at least two.
) and (OQ (C),
)
Consider the gauge action . If C -dynamical system (OP (C),
are conjugate, then the covering degree deg P = deg Q, the number of branched
points # B(P ) =# B(Q), the number of exceptional points # EP =# EQ and the
orbit structures of the exceptional points under P and Q are the same.

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

301
11

Remark. Let us consider the case of the C -algebra OR (JR ) on the Julia set JR .
Since there exist no exceptional points on the Julia set JR , only (Case 4) occurs
and a similar classication of the KMS state is possible.

7. Backward orbit structure of branched points


We study the orbit structure of branched points under iteration. We can count
the number of n-th backward orbits starting at branched points in terms of associated C -algebras with gauge actions. We use a Perron-Frobenius type operator to
show it. The proof depends on the fact that extreme KMS states are parameterized
by the branched points.
Denition. Let R be a rational function with N = degR 2. We denote by bn (z)
We dene the
the number of the n-th backward orbit Rn (z) starting at z C.

associated sequence b(z) := (bn (z))n=0 . If the backward orbit n=1 Rn (z) has no
intersection with the set BR of the branched points, then
b(z) = (1, N, N 2 , N 3 , . . . , N n , . . . ).
In general the sequence b(z) measure the existance of branched points in the backn
ward orbit
(z) starting at z.
n=1 R
Theorem 7.1. Let Q and R be rational functions with the degrees at least
OR (C)
such that
two. Suppose that there exists an isomorphism h : OQ (C)
1
R = hQ h , where Q and R are the associated guage actions. Then their
backward orbit structures starting at the branched points are same, that is,
{b(z) | z BQ } = {b(z) | z BR }
Example. Let R(z) = z 2 . Then BR = {0, }. We calculate that
b(0) = (1, 1, 1, 1, 1, . . . ), b() = (1, 1, 1, 1, 1, . . . )
Example. Let R(z) = z 2 + 1. Then BR = {0, }. We have that
b(0) = (1, 2, 4, 8, 1, . . . , 2n , . . . ),

b() = (1, 1, 1, 1, 1, . . . )

Example. Let R(z) = z 2 1. Then BR = {0, } and R(0) = 1, R(1) = 0. We


have that
b(0) = (1, 2, 3, 6, 11, . . . ), b() = (1, 1, 1, 1, 1, . . . )
Example. There exists a constant c with R(z) = z 2 + c such that R3 (0) = 0 and
R(0) = 0, R2 (0) = 0. Then BR = {0, }. We have that
b(0) = (1, 2, 4, 7, 14, . . . ), b() = (1, 1, 1, 1, 1, . . . )
Remark. The theorem above shows that we can distiguish these examples of
complex dynamical systems in terms of C -algebras.

302
12

YASUO WATATANI

8. Algebraic correspondences
Sullivan discovered a dictionary between the theory of complex analytic iteration and the theory of Kleinian groups in [46]. Sullivans dictionary shows a strong
analogy between the limit set of a Kleinian group and the Julia set JR of a
rational function R. Therefore it is natural to generalize both Kleinian groups and
rational maps. In fact there exist such objects called algebraic correspondences or
holomorphic correspondences. Many works on algebraic correspondences has been
done, for example, in Bullet [6], Bullet-Penrose [7], [8] and M
unzner-Rasch [37].
Let p(z, w) be a polynomial in two variables. Then the solution of the algebraic
equation p(z, w) = 0 is called an algebraic correspondence. We regard it as the
graph of the multivalued function z  w dened implicitly by p(z, w) = 0. Note
that there exist two dierent ways to compactify the algebraic curve p(z, w) = 0.
The standard construction in algebraic geometry is to consider the zeroes of a homogeneous polynomial P (z, w, u) in the complex projective plane CP 2 . But we
choose the second way after [7] and introduce four variables z1 , z2 , w1 , w2 and a
polynomial
z1 w1
),
p(z1 , z2 , w1 , w2 ) = z2m w2n p( ,
z2 w2
which is separately homogeneous in z1 , z2 and in w1 , w2 . We identify the Riemann
with the complex projective line CP 1 . We denote by [z1 , z2 ] an element
sphere C
1
of CP . Then the algebraic correspondence Cp of p(z, w) on the Riemann sphere is
C
dened by
a compact subset of C
C
| p(z1 , z2 , w1 , w2 ) = 0}.
Cp := {([z1 , z2 ], [w1 , w2 ]) C
To simplify notation, we write it by
C
| p(z, w) = 0}
Cp = {(z, w) C
for short if no confusion can arise.
P (z)
be a rational function with polynomials P (z),
For example, let R(z) = Q(z)
Q(z). Put p(z, w) = Q(z)wP (z). Then the algebraic correspondence Cp of p(z, w)
C
| w = R(z), z C}
of
on the Riemann sphere is exactly the graph {(z, w) C
R.
Therefore we regard the algebraic correspondence Cp of a general polynomial
p(z, w) as the graph of the algebraic function z  w implicitly dened by the
equation p(z, w) = 0. Then the iteration of the algebraic function is described
naturally by a sequence z1 , z2 , z3 , . . . satisfying (zk , zk+1 ) Cp for k = 1, 2, 3, . . . .
We dene the path space Pn of length n by
n+1 | (zi , zi+1 ) Cp , i = 1, . . . , n }.
Pn = { (z1 , z2 , . . . , zn+1 ) C
Any non-zero polynomial p(z, w) C[z, w] has a unique factorization into irreducible polynomials:
p(z, w) = g1 (z, w)n1 gp (z, w)np
where each gi (z, w) is irreducible and gi and gj (i = j) are prime each other.
We assume that any polynomial p(z, w) we consider is reduced, that is, the above
powers ni = 1 for any i. We also assume that any gi (z, w) is not a function only in
z or w. In particular the degree m in z and the degree n in w of p(z, w) are both
greater than or equal to one.

COMPLEX DYNAMICAL SYSTEMS AND ASSOCIATED C -ALGEBRAS

303
13

Denition. As in the case of rational functions, we introduce the C -algebra Op (C)


C
| p(z, w) = 0} as
associated with an algebraic correspondence Cp = {(z, w) C

the Cuntz-Pimsner algebra of the Hilbert C -bimodule Xp = C(Cp ) over A = C(C)


in [30].
is said to be p-invariant if the following conditions are
A closed subset J in C

satised: For z, w C,
(i)z J and p(z, w) = 0 implies w J,
(ii)w J and p(z, w) = 0 implies z J.
Under the condition, we can dene Cp (J) = { (z, w) J J | p(z, w) = 0 }, A =
C(J), Xp (J) = C(Cp (J)) similarly. Then Xp (J) is a full Hilbert C -bimodule
(C -correspondence) over A = C(J) and the left action is unital and faithful.
We also introduce the C -algebra Op (J) as the Cuntz-Pimsner algebra of the
Hilbert C -bimodule Xp (J) = C(Cp (J)) over A = C(J).
Denition. Let p(z, w) be a non-zero polynomial in two variables and J a p For any subset U of J and a natural number n, we dene a
invariant subset of C.
(n)
subset U
of J by
U (n) = { w J | (z1 , z2 , . . . , zn , w) Pn

for some

z1 U, z2 . . . zn J }.

Then p is said to be expansive on J if for any nonempty open set U J in J with


the relative topology there exists a natural number n such that U (n) = J.
Denition. Let N be a natural number. We dene the set GP(N ) of N -generalized
periodic points by
GP(N ) ={w J | z J m, n

0 m = n N, (z, z2 , z3 , . . . , zn , w) Pn ,

(z, u2 , u3 , . . . , um , w) Pm }.
P (z)
be a rational function with polynomials P (z), Q(z).
For example, let R(z) = Q(z)
Put p(z, w) = Q(z)w P (z). Then
n

GP(N ) = N
n=1 {w C | R (w) = w}.

Denition. A polynomial p in two variables is said to be free on J if for any


P (z)
be a
natural number N , GP(N ) is a nite set. For example, let R(z) = Q(z)
rational function with polynomials P (z), Q(z). Put p(z, w) = Q(z)w P (z). If
deg R 2, then p is free on any p-invariant set J.
We shall give a sucient condition for simplicity of the associated C -algebras.
Theorem 8.1 ([30]). Let p(z, w) be a reduced non-zero polynomial in two variables with a unique factorization into irreducible polynomials:
p(z, w) = g1 (z, w) gp (z, w),
where each gi (z, w) is irreducible and gi and gj (i = j) are prime to each other. We
assume that any gi (z, w) is not a function only in z or w. Let J be a p-invariant
set with no isolated points. Suppose that p is expansive and free on J. Then the
associated C -algebra Op (J) is simple and purely innite.
Example. Let p(z, w) = (w z m1 )(w z m2 ) . . . (w z mr ) such that m1 , . . . , mr
are all dierent, where r is the number of irreducible components. Then J := T is

304
14

YASUO WATATANI

a p-invariant set. Let b = # B(p) be the number of the branched points. Then we
have
K0 (Op (T)) = Zb , and K1 (Op (T)) = Z/(r 1)Z.
If m1 , m2 , . . . , mr are relatively prime, then p is expansive and free on J = T and
the associated C -algebra Op (T) is simple and purely innite.
References
[1] C. Anatharaman-Delaroche, Purely innite C -algebras arising from dynamical systems, Bull.
Soc. Math. France 125(1997), 199-225
[2] V. A. Arzumanian and A. M. Vershik, Star algebras associated with endomorphisms, Operator
Algebras and Group Representations,vol. I, Pitman, 1984, pp. 17-27.
[3] A. F. Beardon, Iteration of rational functions GTM 132, 1991, Springer New York.
[4] B. Brenken, C -algebras associated with topological relations, J. Ramanujan Math. Soc., 19
(2004), 35-55.
[5] H. Brolin, Invariant sets under iteration of rational functions, Ark. Mat. 6 (1965), 103-144.
[6] S. Bullett, Dynamics of quadratic correspondences, Nonlinearity 1 (1988), 27-50.
[7] S. Bullett and C. Penrose, A gallery of iterated correspondences, Experimental Mathematics
3 (1994), 85-105.
[8] S. Bullett and C. Penrose, Regular and limit sets for holomorphic correspondences, Fundamenta Mathematica 167 (2001), 111-171.
[9] T. M. Carlsen and S. Silvestrov, On the Exel crossed product of topological covering maps,
arxiv:[math.OA]/0811.0056, to appear in Acta Appl. Math.
[10] V. Deaconu, Groupoids associated with endomorphism, Trans. Amer. Math. Soc., 347 (1995),
1799-1786.
[11] V. Deaconu and M. Muhly, C -algebras associated with branched coverings, Proc. AMS. 129
(2001), 1077-1086.
[12] D. E. Dutkay and P. E. T. Jorgensen, Hilbert spaces built on a similarity and on dynamical
renormalization, J. Math. Phys., 47 (2006), 053504, 1-20.
[13] R. Exel, A new look at the crossed-product of a C -algebra by an endomorphism, Ergodic
Theory Dynam. Systems, 23 (2003), 1733-1750.
[14] R. Exel and A. Vershik, C -algebras of irreducible dynamical systems, Canad. J. Math. 58
(2006), 39-63.
[15] M. Frank and D. Larson, Frames in Hilbert C -modules and C -algebras, J