Lecture Notes in

Mathematics
Edited by A. Dold and B. Eckmann
Series: Department of Mathematics, University of Maryland,
College Park
Adviser: L. Greenberg

458

Peter Waiters

Ergodic Theory-
Introductory Lectures
I. I I

Springer-Verlag
Berlin.Heidelberg New York 1975
Dr. Peter Waiters
Mathematics Institute
University of Warwick
Coventry/England

Library of Congress Cataloging in Publication Data

Walters, Peter, 1943-

Ergodic theory.

(Lecture notes in mathematics ; &58)

Bibliography: p.
Includes index.
1. Ergodic theory. I. Title. II. Series:
Lectures notes in mathematics (Berlin) ; /4.58.
QA3.L28 no. ~58 ~QA313~ 510'.8s r515'.&2~ 75-9853
ISBN 0-387-07163-6

AMS Subject Classifications (1970): 2 8 A 6 5

ISBN 3-540-07163-6 Springer-Verlag Berlin • Heidelberg • New York

ISBN 0-387-07163-6 Springer-Verlag New York • Heidelberg • Berlin

This work is subject to copyright. All rights are reserved, whether the whole
or part of the material is concerned, specifically those of translation,
reprinting, re-use of illustrations, broadcasting, reproduction by photo-
copying machine or similar means, and storage in data banks.
Under § 54 of the German Copyright Law where copies are made for other
than private use, a fee is payable to the publisher, the amount of the fee to
be determined by agreement with the publisher.
© by Springer-Verlag Berlin • Heidelberg 1975. Printed in Germany.

Offsetdruck: Julius Beltz, Hemsbach/Bergstr.

 Preface

These are notes of a o n e - s e m e s t e r i n t r o d u c t o r y course on Ergodic

Theory that I gave at the U n i v e r s i t y of M a r y l a n d in College Park

during the fall of 1970. I assumed the audience had no p r e v i o u s

k n o w l e d g e of E r g o d i c Theory. My aim was to present some of the basic

facts in m e a s u r e t h e o r e t i c Ergodic Theory and T o p o l o g i c a l Dynamics and

show how they are r e l a t e d so that the a u d i e n c e would have the founda-

tions to read the r e s e a r c h papers if they w i s h e d to pursue the subject

further.

At the b e g i n n i n g of Chapter 1 I give a list of examples of

measure-preserving transformations and at the end of each section of

Chapter 1 I i n v e s t i g a t e w h e t h e r these examples have the properties

d i s c u s s e d in that section. These examples were chosen because of

their varied p r o p e r t i e s and importance in the subject. Similarly in

C h a p t e r 5, on T o p o l o g i c a l Dynamics, a list of examples is given and

the properties d i s c u s s e d in that chapter are c o n s i d e r e d for these

examples.

I tried to deal with entropy as simply as possible. In the dis-

cussion of entropy I have inserted without proof some of the more

d i f f i c u l t theorems when I thought they wePe r e l e v a n t to the d i s c u s s i o ~

In p a r t i c u l a r I have discussed the recent deep results of D. S.

Ornstein on B e r n o u l l i automorphisms and K o l m o g o r o v automorphisms.

In the final chapter I have p r e s e n t e d the new t r e a t m e n t of topo-

logical entropy due to R. E. Bowen. One of the beauties of this

treatment is that t o p o l o g i c a l entropy can be defined for a u n i f o r m l y

continuous map of any metric space and that its value remains un-

changed under certain types of c o v e r i n g maps. This enables one to

give an elegant c a l c u l a t i o n of the t o p o l o g i c a l and (Haar) measure

 IV

theoretic entropies of affine t r a n s f o r m a t i o n s of f i n i t e - d i m e n s i o n a l

tori.

Since these notes have not been fully edited many r e f e r e n c e s are

m i s s i n g and it is likely that credit is often not given where it is

due. The theorems and definitions are n u m b e r e d independently, but a

corollary is given the same number as the t h e o r e m to which it is a

corollary.

Thanks are due to Victor Charles Stasio and S u e l l e n E s l i n g e r who

took notes of the course and also to A l l a n Jaworski for editing and

compiling the bibliography. Special thanks are due to Betty Vander-

slice for her superb typing.

--Peter Waiters
 Contents

Chapter O: Preliminaries
§i. Introduction 1
§2. Measure Theory 3
§3. Hilbert Spaces 8
§4. 'Haar Measure 9
§ 5 . Character Theory i0
§6. Endomorphisms of Tori 12
Chapter i: Measure-Preserving Transformations 16

§i. Examples 16
§2. Problems in Measure Theoretic Ergodic Theory 19
§3. Recurrence 20
§4. Ergodicity 21
§5. The Ergodic Theorem 29
§6. Mixing 37

Chapter 2: Isomorphism and Spectral !nvariants 51

§i. Isomorphism of Measure-Preserving Transformations 51

§2. Conjugacy of Measure-Preserving Transformations 53
§3. Spectral Isomorphism 54
§4. Spectral Invariants 57
§5. Examples 59

Chapter 3: Measure-Preserving Transformations 63

with Pure Point Spectrum

§i. Eigenfunctions 63
§2. Pure Point Spectrum 64
§3. Group Rotations 67

Chapter 4: Entropy 70
§i. Partitions and Subalgebras 70
§2. Entropy 72
§3. Conditional Entropy 76
§4. Properties of h(T,A) 80
§S. Properties of h(T) 83
§6. Examples 92
§7. How good an invariant is entropy? 96
§8. Bernoulli and Kolmogorov Automorphisms 98
§9. Pinsker Algebra i07
§lO. Sequence Entropy 108
§ll. Comments 109
§12. Non-invertible Transformations ii0
 Vl

Chapter 5: Topologica! Dynamics 112

§0. Introduction 112

§i. Minimality 113
§2. Topological Transitivity i17
§3. Topological Conjugacy and Discrete Spectrum 122
§4. Invariant Measures for Homeomorphisms 128

Chapter 6: Topological Entropy 140

§i. Definition by Open Covers 140

§2. Bowen's Definition 146
§3. Connections with Measure Theoretic Entropy 155
§4. Topological Entropy of Linear Maps
and Total Affines 160
§5. Expansive Homeomorphisms 168
§6. Examples 182

Bibliography 185

Index 197
 Chapter 0: Preliminaries

31. Introduction

G e n e r a l l y speaking, ergodic theory is the study of t r a n s f o r m a -

tions and flows from the point of view of r e c u r r e n c e properties,

m i x i n g properties, and o t h e r global, d y n a m i c a l properties c o n n e c t e d

with a s y m p t o t i c behavior. Abstractly,one has a space X and a

transformation T of X (or a family {Tt: t ~ ~} of t r a n s f o r m a -

tions of X) w i t h some structure on X w h i c h is p r e s e r v e d by T

(or by {Tt}). The nature of most of the w o r k so far can be cate-

gorized into one of the four f o l l o w i n g types:

(i) measure theoretic:

Here one deals with a measure space X and a measure pre-

serving t r a n s f o r m a t i o n T: X ~ X.

(2) topological:

Here X is a t o p o l o g i c a l space and T: X ~ X is a continuous

map.

(3) mixture of (i) and (2):

In this situation X is a t o p o l o g i c a l space equipped with a

measure m on its Borel sets while T: X ~ X is continuous and

preserves m.

(4) smooth:

One considers a smooth m a n i f o l d X and a smooth map

T: X ~ X.

We shall deal with some topics from (i), (2), and (3).

To see how this study arose consider, for example, a system of

k particles in 3-space m o v i n g under known forces. Suppose that the

phase of the system at a given time is c o m p l e t e l y determined by the

positions and the momenta of each of the k particles. Thus, at a

given time the system is determined by a point in 6k-dimensional

space. As time continues the phase of the system alters according to

the d i f f e r e n t i a l equations governing the motion, e.g., Hamilton's

equations

dqi ~H dPi ~H

dt ~Pi dt aqi

If we are given an initial condition and such equations can be unique-

ly solved then the c o r r e s p o n d i n g solution gives us the entire history

of the system, which is determined by a curve in phase space.

If x is a point in p h a s e space representing the system at a

time t0, let Tt(x) denote the point of phase space r e p r e s e n t i n g

the system at time t+t 0. From this we see that Tt is a transforma-

tion of phase space and, moreover, T O = id., Tt+ s = TtOT s. Thus

{Tt: t E ~} is a o n e - p a r a m e t e r group of transformations of phase

space. In dynamics one is interested in the asymptotic properties of

the family {Tt}. It seems reasonable to study the system at discrete

times t 0, 2t0, 3t 0 ..... i.e., study the family {T~ }~-i ' since we
0 -
expect the properties of {T t} to be reflected in those of {T~ }.
0
For this reason, as well as the fact that it is simpler, one

studies individual transformations and their iterates. One is par-

ticularly interested in the flow on an energy surface, which is some-

times smooth (hence considerations of type (4) arise), and sometimes

is not (one then investigates along the lines of (2) ). Measure

theory enters the picture via a theorem of Liouville which tells us

that if the forces are of a certain type one can choose coordinates

in phase space so that the usual 6k-dimensional measure in these

coordinates is preserved by each t r a n s f o r m a t i o n T t.

Around 1900 Gibbs suggested using the m e a s u r e - t h e o r e t i c approach

in mechanics because of the difficulty of solving the equations of

motion and also because this deterministic approach does not answer

several important questions in mechanics. In discovering statistical

mechanics Gibbs suggested looking at what happens to subsets of phase

space. For example, if A and B are subsets of phase space what

is the p r o b a b i l i t y that the system is in B at the time t given

that the system is in A at the time t0? Given that the system be-

gins in A at time t0 what is the average time the system spends

in B? Such questions motivate the type of study u n d e r t a k e n in

ergodic theory.

We now list some general references for the m a t e r i a l we shall

discuss:

For topics of types (I) and (3) mentioned above see Halmos [2],

Billingsley [i], Hopf [i], Jacobs [i] [2], Parry [3], Rohlin [3][4][5],

Friedman [i], Shields [2]. In addition to the Shields notes, further

details on the results of Ornstein described in Chapter 4 may be found

in a forthcoming book by Friedman and Ornstein [2].

For m a t e r i a l of type (2) see Gottschalk and Hedlund [I],

Nemytskii and Stepanov [I], and Ellis [i].

And m a t e r i a l of type (4) may be found in Avez and Arnold [i],

Smale [i], A b r a h a m [i], A b r a h a m and Robbin [i], Nitecki [i].

Khinchin [i] provides a good sketch of the foundations of ergodic

~heory. For extensive bibliographies see Jacobs [i][2]~ Gottschalk

[i], and Smale [i]. A recent survey is Mackey [i].

~2. Measure T h e o r y

General reference - Halmos [i].

We recall some fundamental notions from measure theory.

Let X be a set. A ~-al~ebra of subsets of X is a collection

8 of subsets of X satisfying:
(i) X E 8, (2) B E 8 = X\B 6 B,

(3) Bn E B, n > i = 0 Bn ~ B.
n=l
We t h e n call (X,B) a measurable space. A measure space is a

triDle (X,B,m) where X is a set, B is a q - a l g e b r a of subsets

of X, and m is a f u n c t i o n m: B ~ R+ satisfying

m( 0 B n) = ~ m ( B n)
n:l n:l

if {B n} is a p a i r w i s e disjoint sequence of e l e m e n t s of B. We say

that (X,B,m) is a p r o b a b i l i t y space, or a n o r m a l i z e d measure space,

if m(X) = i. We shall usually deal with such spaces.

A collection A of s u b s e t s of a set X is an a l s e b r a if:

(i) X E A, (2) A E A = X\A 6 A,

n
(3) A I .... ,A n ~ A = ~ A. E A.
i=l l

When one is t r y i n g to e q u i p a measurable space (X,B) with a

measure one usually knows what the m e a s u r e should be on an a l g e b r a

A & B, and so, one w o u l d like to k n o w w h e n this function defined

on A can be e x t e n d e d to a m e a s u r e on S. The following result

deals with this situation.

Hahn-Kolmogorov Extension Theorem:

Given a set X, an a l g e b r a A of s u b s e t s of X, let m: A ~ R+

be a f u n c t i o n satisfying

m(X) = i, m(~JA n) : ~ - m ( A n)
n n

whenever An E A V n, ~JA n E A, and the {A n } are disjoint. Then

n
there is a u n i q u e probability measure m defined on the q-algebra

generated by A such that m(A) = m(A) whenever A 6 A.

 A monotone class of subsets of X is a c o l l e c t i o n C of subsets

of X such that if E1 c E2 c ... belong to C then U En ~ C and

n
if F I • F 2 9 ... b e l o n g to C then N Fn ~ C.
n

Theorem:

If A is an a l g e b r a of subsets of X then the a - a l g e b r a gener-

ated by A equals the m o n o t o n e class g e n e r a t e d by A.

If (X,B,m) is a finite measure space, one can easily deduce

from this theorem that if A is an algebra generating the a - a l g e b r a B

then for any B E B and g > 0 there exists A E A with m(AAB) < s.

(In fact,

C = {B E B I V s > 0 ~ B EA with m(BAB g ) < ~}

is a monotone class and contains A.)

Notation: If A is an algebra we shall write a(A) for the a-alge-

bra generated by A.

Direct Products:

Let (Xi,~i,mi), i E Z be p r o b a b i l i t y spaces. Their direct

product

(X,B,m) = ~ (Xi,Bi,m i)

is defined as follows:

(a) X = ~ X.
i=_® !
(b) Let n1 < n2 < ... < nr be integers, and An. E Bn. i = l,...,r.
1 1
We define a m e a s u r a b l e rectangle to be a set of the form

{(xj) E X: Xn. E An. V i: l_<i_<r}.

1 1

Let A be the c o l l e c t i o n of all finite unions of m e a s u r a b l e rec-

tangles. A is an algebra: (i) and (3) are obvious; to show (2)

observe that
r
(xj) ( An. , 1 ~ i ~ r} = U {(xj) Ix E X \ An.} ( A
X \{ I Xni l i=l ni ni l

and that A is closed under finite intersections. Let B be the

o-algebra generated by A.

(c) Each element of A can be written as a disjoint finite union of

measurable rectangles so that we define

r
m({(xj) I Xn. EAn. , l_<i_<r}) : 7~ mn.(Ani )
l i:l 1

and then extend m to A in the obvious manner. The conditions of

the Hahn-Kolmogorov T h e o r e m can be shown to be satisfied, and thus we

can extend m to B. Hence, we obtain a probability space.

Measurable Functions:

A Borel subset of R is a member of the o-algebra generated by

the open sets.

Let (X,B,~) be a measure space; f: X ~ R is measurable if for

all c E R, f-l(c,®) 6 B, or equivalently f-l(D) E B V Borel sets

D c R.

A function f: X ~ C, f : fl + if2 is measurable if fl: X ~ R

and f2: X ~ R are measurable.

If X is a topological space and B is the o-algebra of Borel

subsets of X (the o-algebra generated by the open subsets of X)

then each continuous function f: X ~ C is measurable.

Intesration:
n
A simple function is a function of the form ~ aiXA. , where
i:l m
a i E R, the {Ai} are disjoint members of 8, and XA. denotes
l
the characteristic function of A i. Simple functions are measurable.

We define the integral for simple functions by:

f ( Z ai×Ai)dm = ~ aim(Ai ).
i=l i=l
 Suppose f: X ~ R is measurable and f e 0; then there exists

an increasing sequence of simple functions fn m f" For example we

could take

i-I if i-i < f(x) < ___i i : 1,...,n2 n

2n 2n 2n
fn(X) =
n if f(x) { n.

We define I f dm = lim I fndm and note that this definition is inde-

n~-
pendent of the chosen sequence {fn }.

Suppose f: X ~ R is measurable; then f = f+ - f- where

f+(x) : max {f(x),O} ~ 0

f-(x) : max {-f(x),0} e 0.

We say that f is integrable if f f+dm, f f-dm < -, and we define

ffdm=Jf m ff m
Now if f: X ~ C is measurable, f : fl + if2' f is integrable if

fl and f2 are integrable and we define

Observe that f is integrable if and only if Ifl is integrable.

Let LI(~) denote the class of all integrable functions (X,B,~) ~ C.

Lebesgue Dominated Conversence Theorem:

Suppose {fn } is a sequence of measurable functions, f n ~ f a.e.

and there exists an integrable function g such that Ifn(X) I ~ g(x)

a.e. for all n; then f and each f are integrable and

n

I fnd~ ~ I fd~.
§3. Hilbert Spaces

General r e f e r e n c e - Halmos [3].

H is a H i l b e r t space if it is a Banach inner product space,

i.e., (i) H is a vector space over the complex numbers C

(2) T h e r e is an inner product on H, i.e., a map

(.,.): H x H ~ C such that:

(a) (.,.) is b i l i n e a r

(b) (f,f) ~ 0 for all f in H

(c) (f,f) = 0 if and only if f = 0

(d) (f,g) : (g,f)'.

(3) llfll = (f,f)½ is a n o r m on H inducing a complete metric

t o p o l o g y on H.

Let (X,B,m) be a m e a s u r e space. Consider all m e a s u r a b l e func-

tions f: X ~ C such that f IfI2dm < ®; we define an e q u i v a l e n c e

r e l a t i o n on this set by saying that f ~ g if and only if f = g a.e.

The set of e q u i v a l e n c e classes relative to this r e l a t i o n forms a Hil-

bert space w h i c h we denote by L2(X,B,m) = L2(m), where the inner

product of two functions is given by

(f'g) : I f g d m -

Recall: The Schwarz Inequality

In any H i l b e r t space H

I (f,g)l ~ llfll'llgll for all f,g in H.

A unitary o p e r a t o r U on a Hilbert space H is an i s o m o r p h i s m

of H, i.e., U is a linear bijective map p r e s e r v i n g the inner

product ((Ux,Uy) = (x,y), x,y in H). It follows that U is con-

tinuous.
~4. Hear Measure

General reference - Pontrjagin [i].

Theorem:

Let G be a c o m p a c t topological group; then there exists a fi-

nite measure m defined o n the Borel subsets of G such that

m(xE) = m(E) for all x in G, and for a l l Borel sets E. We c a l l

such a measure a Hear measure.

E.g., let K = {z E C: Izl : i} and m denote normalized circu-

lar Lebesgue measure. Then m(aU) = m(U) for all sets U measurable

on K.

Theorem:

If m and ~ are both finite Haar measures on t h e compact

topological group G then m = c~ where c > 0. Thus there exists

o n l y one n o r m a l i z e d Haar measure on G.

Remarks:

(i) If U c G is a n o n - e m p t y open set t h e n it has positive Haar

measure. This is b e c a u s e

G : ~.~ g U : gl U U g2 U U ... U gn U
g~G

by compactness.

(2) In t h e H i l b e r t space L2(K,B,m) where m is t h e H e a r m e a s u r e on

the u n i t circle K, the functions

f(z) : z n, n E Z

f o r m an o r t h o n o r m a l basis.
 i0

~5. Character Theory

General r e f e r e n c e - P o n t r j a g i n [i].

Many of our examples w i l l be rotations, e n d o m o r p h i s m s or affine

transformations of compact groups. (We mean e n d o m o r p h i s m in the sense

of t o p o l o g i c a l groups, i.e., an abstract group a n d o m o r p h i s m w h i c h is

continuous.) In some proofs we will use the c h a r a c t e r theory of com-

pact abelian groups, w h i c h we summarize in this section. For those

not familiar w i t h c h a r a c t e r theory, proofs in the later sections in-

volving c h a r a c t e r s w i l l u s u a l l y be p r e c e d e d by the proof in a special

case where the group used is the unit circle and then classical

Fourier analysis will be used.

Let G be a locally compact abelian group with a countable topo-

logical base. Let G denote the c o l l e c t i o n of all continuous homo-

morphisms of G into the unit circle K. The members of G are the

characters of G. G is an abelian group under the o p e r a t i o n of

pointwise m u l t i p l i c a t i o n of functions. With the compact open topology

becomes a locally compact abelian group with a countable topologi-

cal base. We have the f o l l o w i n g results:

(I) G is compact - G is discrete.

(2) (G) is n a t u r a l l y i s o m o r p h i c (as a t o p o l o g i c a l group) to G, the

isomorphism b e i n g given by the map

a ~ a where ~(T) = 7(a) for all 7 E G.

(3) If G is compact then G is c o n n e c t e d ~ G is torsion free.

A

(4) G1 ×G 2 = G 1 ×G 2 where "×" denotes direct product.

So in some sense we can study compact abelian groups by studying dis-

crete countable groups.

Examples: (See §6 for proofs.)

(i) Let G = K = {z ~ C: Izl = i}. Each h o m o m o r p h i s m of K to it-

self is of the form z ~ zn (n E Z) so that G ~ Z.

 ii

(ii) Let G = Kn the n-torus. By (4) ~n ~ Z n and in fact each

m e m b e r of ~n has the form

Pl P2 Pn
Z ( Z l , Z 2 , . . . , z n) = z I z2 ...z n

where (pl,...,pn) E Z n.

(5) If H is a closed subgroup of G and H ~ G there exists a

T ~ G, 7 ~ i such that T(h) : i V h E H. (We shall write this

T(H) : i.)

(6) Let G be compact. The m e m b e r s of ~ are m u t u a l l y o r t h o g o n a l

members of L2(m), where m is Haar measure.

Proof. It suffices to show

T(x)dm(x) = 0 if T g I.

If a ~ g then since m is Haar m e a s u r e

Choosing a so t h a t 7(a) g 1 we h a v e I ~(x)dm(x) O. //

(7) If G is compact, the members of G form an o r t h o n o r m a l basis

for L2(m) where m is n o r m a l i z e d Haar measure.

This is part ~f the P e t e r - W e y l theorem and can be easily deduced

from the S t o n e - W e i e r s t r a s s theorem, which implies that finite linear

combinations of characters are dense in C(G) = space of complex-

valued continuous functions of G.

(8) If A: G ~ G is an e n d o m o r p h i s m we can define an e n d o m o r p h i s m

A: G ~ G by AT = T °A, T E G. It is easy to see that A is one-

to-one if and only if A is onto and A is onto if and only if

is one-to-one. Therefore A is an a u t o m o r p h i s m if and only if

is an automorphism.

Recall that for compact groups G, G is m e t r i c iff G has a

 12

countable topological base.

~6. Endomorphisms of T o r i

We shall view the n-torus in two w a y s : - multiplicatively as

Kn = K×K×...×K , and additively as Rn/z n where Rn is n - s p a c e and

n times

Zn is the subgroup of Rn consisting of points with integer coordi-

nates. A topological group isomorphism is g i v e n by K n ~ Rn/z n,

2~ix, 2,iXn) , ..
(e ,...,e ~ (x I . ,x n) + Z n.

Theorem:

(i) Every closed subgroup of K is e i t h e r K or is a f i n i t e cyclic

group consisting of all p-th roots of u n i t y for some integer p > 0.

(2) The only automorphisms of K are the identity and the m a p

-i
Z ~ Z

(3) The only homomorphisms of K are the maps

Cn: z ~ z n, n E Z.

(4) The only homomorphisms of Kn to K are maps of the form

( Z l , . . . , z n) ~ z m1"
I ...'z mn
n where m I ,...,m n 6 Z•

Proof: (i) Let H be a closed subgroup of K; if H is in-

finite it has a limit point so, V e > 0 3 a,b E H 9 d(a,b) < ~,

a ~ b. Then d(b-la,l) < ~ and t h e r e f o r e the elements of H are

e-dense in K V ~. Thus, H = K.

If H is f i n i t e and has p elements then ap = 1 V a E H. So

each element of H is a p - t h root of u n i t y , and since there are p

elements in H, H must consist of all the p-th roots of unity.

(2) Let 8: K ~ K be an a u t o m o r p h i s m . 8(i) = i. Since -i is the

only element of K of o r d e r 2 we h a v e 8(-1) = -i. Since i,-i

 13

are the o n l y e l e m e n t s of o r d e r 4 either 8(i) = i and 8(-i) = -i

or 8(i) = -i and 8(-i) = i. C o n s i d e r the first case. Since 8

maps intervals to i n t e r v a l s , the i n t e r v a l [l,i] from 1 to i is

either mapped to i t s e l f or to [i,l] (all i n t e r v a l s go a n t i c l o c k w i s e ) .

But since [i,-~] does not contain -i it c a n n o t be m a p p e d to [i,-~]

so 8[i~] = [l,-~l]. T h e o n l y e l e m e n t of o r d e r 8 in [l,-~l] is

e ~i/4 and so t h i s m u s t be f i x e d by 8. Therefore 811 ,e~i/4] =

i/4].
' e ( e 2 ~ i / 2 k ) = e 2~i/2 k
[l,e ~ By i n d u c t i o n one shows t h a t for

each k > 0. It f o l l o w s that e fixes all the 2k-th r o o t s of u n i t y

V k > 0 and h e n c e is the identity. In the s e c o n d case one shows

that 8(e 2"i/2k) = e -2~i/2k V k > 0 and h e n c e 8(z) = z -1, z E K.

(3) Let 8: K ~ K be an e n d o m o r p h i s m . If 8 is n o n - t r i v i a l , its

image, e(K), is a c l o s e d connected subgroup of K and so 8(K) = K

by (i). The k e r n e l Ker 8 is a c l o s e d subgroup of K so e i t h e r

Ker 8 = K or Ker 8 = Hp, the group of all p-th r o o t s of u n i t y , for

some p. The first case c o r r e s p o n d s to t r i v i a l 8. If Ker 8 = H

P
let ~p: K / H p ~ K be the isomorphism given by ~p(ZHp) = z p, and let

81: K/Hp ~ K be the isomorphism i n d u c e d by 8 ( 8 1 ( z H p) = 8(z) ).

-i
Then 81a p is an a u t o m o r p h i s m of K and by (2) e i t h e r 81aDl(z)-_ = z

V z E K or 81epl(z) = z -I V z E K. Hence either 8(z) = 81(zH p) =

81~pl(zP) = zp V z E K or %(z) = z -p V z ~ K.

(4) Let ¥i: K ~ Kn be d e f i n e d by Ti(z) = (l,l,...,l,z,l,...,l).

÷
i-th p l a c e
If 8: K n ~ K is a h o m o m o r p h i s m then 8o~i: K ~ K is an e n d o m o r p h i s m
m.
and so 8oTi(z) = z 1 for some mi E Z by (3). Hence

e ( Z l , . . . , z n) = e ( Z l ( Z l ) - T 2 ( z 2 ) ' . . . ' T n ( Z n ) )

= 8 y l ( Z l ) . e Z 2 ( z 2 ) . . . . . e T n ( Z n)

m I m2
= z I "z 2 - . . . ' z ~ n //
 14

Theorem:

(i) Every endomorphism A: K n ~ K n is of the form:

a! aln a ann
A(Zl,...,Zn) = (z I i "''''Zn ''''' zlnl '' ...z n )

where aij ( Z. In additive notation,

A<<I!) + zn) : [aij] < i i ) + zn.

n n

(2) A maps Kn onto Kn iff det[aij] ~ 0.

(3) A is an automorphism of Kn iff det[aij] = ±i.

Proof: (1) Let ~i: Kn ~ K be the projection to the i-th

coordinate. Then ~l.oA: K n + K is a homomorphism, so by (4) of the

previous theorem

z~il ai2 aln

~ioA(Zl,...,Zn ) = .z 2 "...'z n

where aij E Z.

(2) Assume det[aij] = 0. 3 ml,...,m n integers not all zero

mlA I +... + m n nA = 0 where Ai is the i-th row of A. Then each

ml m n = I.
point (~l,...,~n) of K n in the image of A satisfies el "''~n

Thus A(K n) ~ K n. Conversely if A(K n) ~ K n then the points of

A(K n) are annihilated by a nontrivial character of K n, say

mI mn
(Zl,...,z n) ~ z I "...'z n (this is by (5) of ~5). Then

mlA 1 + ... + mnA n = 0 and so det[a i j ] = 0.

(3) If A is an automorphism represented by a matrix [A] then A -I

is also an automorphism represented by a matrix [B], and since AA -I =

I = A-IA we have that [B] = [A] -I. Since [B] is an integer matrix,

det[A] = ±i. Conversely, if det[A] = ±l, [A] -I has integer entries

and if B is the endomorphism of Kn it defines we have AB = BA = I. //

 15

Notation:

If A is an e n d o m o r p h i s m of the n-torus, [A] will always de-

note the associated matrix and A will denote the linear transforma-

tion of Rn determined by [A]. So if ,: R n ~ Rn/z n is the n a t u r a l

projection (~(x) = x + Z n) w e h a v e nA = An.

Let A: Kn ~ K n be an e n d o m o r p h i s m . We n o w c o n s i d e r h o w the map

~: ~n + ~n (introduced in ~5) acts as a m a p of Zn when ~n is

identified with Zn by the isomorphism:

m2 m I m2 mn
n
T ~ when T ( Z l , Z 2 , . . . , z n) = z I .z 2 ....'z

One r e a d i l y checks that the endomorphism A: Zn ~ Zn is g i v e n by

n n

where [A] t denotes the transpose of the m a t r i x [A].

 Chapter i: Measure-Preservin~ Transformations

~I. Examples

Suppose (Xi,Bl,ml), (X2,B2,m 2) are probability spaces.

Definition i.i:

a) T: X 1 ~ X 2 is m e a s u r a b l e if T-I(B2 ) c B 1 (i.e.,

B 2 E B 2 = T-IB 2 E BI).

b) T: X 1 ~ X 2 is m e a s u r e - p r e s e r v i n g if T is measurable and

mI(T-I(B2)) = m2(B 2) V B 2 E B 2.

c) We say that T: X 1 4 X 2 is an invertible m e a s u r e - p r e s e r v i n g

transformation if T is measure-preserving, bijective, and T -I is

also measure-preserving.

Remarks:

(I) We should write T: (Xl,Bl,ml) + (X2,B2,m 2) since the measure-

preserving property depends on the B's and m's.

(2) If T: X 1 ~ X 2 and S: X 2 ~ X 3 are m e a s u r e - p r e s e r v i n g so is

SoT: X1 ~ X3 •

(3) Measure-preserving transformations are the structure preserving

maps (morphisms) between measure spaces.

(4) We shall be m a i n l y interested in the case (XI,BI,ml) = (X2,B2,m 2)

since we wish to study Tn (see §i, Ch. 0).

In practice it would be difficult to check, using Defs. i.!,

whether a given t r a n s f o r m a t i o n is m e a s u r e - p r e s e r v i n g or not since one

usually does not have explicit knowledge of all the members of B.

However we often do have explicit knowledge of an algebra A gener-

ating B (for example, when X is the unit interval A may be all

finite unions of intervals, and when X is a direct product space

 17

A m a y be the c o l l e c t i o n of all finite u n i o n s of m e a s u r a b l e rectan-

gles). The f o l l o w i n g result is t h e r e f o r e desirable in c h e c k i n g

whether transformations are m e a s u r e - p r e s e r v i n g or not.

Theorem i.i:

Suppose (XI,BI,ml), ( X 2 , 8 2 ~ m 2) are p r o b a b i l i t y spaces and

T: X 1 ~ X 2 is a map. Let A2 be an a l g e b r a w h i c h generates B 2.

If A 2 E A 2 = T - I ( A 2 ) E 81 and mI(T-I(A2)) = m 2 ( A 2) then T is

measure-preserving.

Proof: Let C 2 = {B E 82: T-I(B) E 81~ m I ( T - I ( B ) ) =m2(B)} ; we

w a n t to s h o w that C 2 = B 2. However A2 £ C2 and C2 is e a s i l y

seen to be a m o n o t o n e class~ so the r e s u l t follows since the ~ - a l g e b r a

generated by C2 is the m o n o t o n e class g e n e r a t e d by A 2. //

Examples of M e a s u r e - P r e s e r v i n g Transformations:

(i) I = identity on (X,B,m) is o b v i o u s l y m e a s u r e - p r e s e r v i n g .

(2) Let K = {z E C: Izl = i}~ 8 = Borel sets~ and m = H a a r measure.

Define T: K ~ K by T(z) = az where a is a fixed e l e m e n t of K.

T is m e a s u r e - p r e s e r v i n g since m is Haar m e a s u r e .

(3) The t r a n s f o r m a t i o n T(x) = ax defined on any c o m p a c t group G

(where a is a f i x e d e l e m e n t of G) p r e s e r v e s Haar measure.

(4) Any continuous endomorphism of a c o m p a c t g r o u p onto i t s e l f pre-

serves Hear measure.

Proof: Let A: G ~ G be a c o n t i n u o u s endomorphism~ and

m = Haar measure on G. Let ~(E) = m(A-I(E)). ~ is a Borel proba-

bility measure and

~(Ax.E) = m(A-I(Ax'E)) = m(x'A-iE) = ~(E).

Since A maps G onto G, ~ = m by the u n i q u e n e s s property of Haar

measure. //

For e x a m p l e ~ T(z) = z n preserves Haar measure on the unit circle.

 18

(5) Any affine t r a n s f o r m a t i o n of a compact group G p r e s e r v e s Haar

measure. An affine t r a n s f o r m a t i o n is a map of the form T(x) = a.A(x)

where a E G is fixed and A: G ~ G is a surjective endomorphism.

T is m e a s u r e - p r e s e r v i n g because it is the c o m p o s i t i o n of m e a s u r e -

preserving transformations. When A = I we have example (3) and

when a is the identity element of G we have example (4).

(6) Let Y = {0,...,k-l}, and give m e a s u r e Pi to i such that

k-I
~- Pi = i. We let X = ~ Y t o g e t h e r with the direct product
i=0 --
measure. Define T: X ~ X by:

T({xi}) : {yi}, where Yi : Xi+l"

T preserves the m e a s u r e of each m e a s u r a b l e r e c t a n g l e and thus it pre-

serves the m e a s u r e of sets w h i c h are finite d i s j o i n t unions of m e a s u r -

able rectangles. By T h e o r e m 1.1 T is m e a s u r e - p r e s e r v i n g . We call

T the t w o - s i d e d (p0,...,Pk_l)-Shift.

(7) Let Y be as above, X = ~ Y with the direct product measure.

0
Let T: X ~ X be defined by

(x0,xl,...) ~ (Xl,X2,...).

By an analogous a r g u m e n t to the one in example (6) we see that T is

measure-preserving. We call T the o n e - s i d e d (p0,...,Pk_l)-Shift.

(8) Let 12 be the unit square e q u i p p e d with Lebesgue measure and I

the unit interval with Lebesgue measure. Then p: 12 ~ I defined by

p(x,y) = x is m e a s u r e - p r e s e r v i n g .

Given any set X !, any p r o b a b i l i t y space (×2,B2,m2) and any

onto
map T: X 1 ~ X 2 we can choose a ~ - a l g e b r a B1 and a measure m1

on X1 to make T measure-preserving. In fact let B 1 = T-IB2 and

define mI by m I ( T - I B 2) = m2(B2).

Conversely, if (XI,BI,m I) is any p r o b a b i l i t y space, X 2 any set

 19

onto
and T: X 1 ~ X2 any map, then we can choose a c - a l g e b r a B2 and a

measure m2 on X2 so that T is m e a s u r e - p r e s e r v i n g . Put

B 2 = {B: B c X 2 and T-IB E B I}

and m2(B) : mI(T-IB) for B ( B 2.

~2. Problems in M e a s u r e T h e o r e t i c Er$odic T h e o r y

(a) E x t e r n a l Problems:

How do we apply m e a s u r e t h e o r e t i c ergodic theory to other

branches of m a t h e m a t i c s and physics? In these applications one has

a space X with some structure on it, and a map T of X which pre-

serves this structure. To apply the theory of measure p r e s e r v i n g

t r a n s f o r m a t i o n s one needs an invariant measure for T w h i c h acts

"nicely" with respect to the structure on X. For example, if X is

a t o p o l o g i c a l space we would like the measure to be a Borel measure

which is positive on n o n - e m p t y open sets.

Examples:

(I) H a m i l t o n i a n Mechanics: Here one has a o n e - p a r a m e t e r group of

d i f f e o m o r p h i s m s of a m a n i f o l d and there is a smooth measure on the

m a n i f o l d p r e s e r v e d by each diffeomorphism.

(2) N u m b e r Theory: To study continued fractions one studies

T: [0,i) * [0,i) given by:

f 0 if x : 0
T(x)
{l/x} if x ~ 0}

where the {y} denotes the f r a c t i o n a l part of y. T preserves the

Gauss m e a s u r e on [0,i) which is given by:

- 1 [ 1 dx, A ~ [0,i).
m(A) log 2 l+x
 20

(b) Internal Problems:

The main internal problem in measure theoretic ergodic theory is:

Given two measure preserving transformations when are they isomorphic?

(i.e., when can we consider them to be the same?) We look for invari-

ants. A property P is an invariant if when T1 has the property P

and T2 is isomorphic to TI then T2 has the property P. Invari-

ants give good negative answers, i.e., if T1 has property P and

T2 does not, then TI and T2 are not isomorphic. The invariants

we shall study are of two t y p e s : - - s p e c t r a l invariants and entropy.

Before discussing the notion of isomorphism we shall introduce

some general concepts such as recurrence and mixing.

~3. Recurrence

Theorem 1.2: (Poincar4 Recurrence Theorem)

Let T be a m e a s u r e - p r e s e r v i n g transformation of a p r o b a b i l i t y

space (X,S,m). Let E E B, m(E) > 0. Then almost all points of E

return infinitely often to E under positive iteration by T, (in

fact, we have the stronger result that: 3 F ¢ E, m(F) = m(E) ) if

x ~ F 3 integers 0 < nI < n2 ... ) Tni(x) E F V i).

Proof: For N ~ 0 let EN = 0 T-n(E). We have T-I(EN ) = EN+I,

n=N
E N ¢ EN_ 1 ¢ EN_ 2 ¢ ... ¢ E0, E c E0, and m(EN+ I) = m ( T - I ( E N ) ) = m ( E N)

since T is measure-preserving. Therefore, for each N, m(E N) =

m(E 0 ) and,

m(~ E N) = m(E0).
N=0

EN = 6 0 T-nE, which is the set of all points entering E

N=0 N:0 n=N
infinitely often under positive iteration by T. Moreover

F = E N (6 E N) consists of all points of E which enter E

N=0
 21

i n f i n i t e l y often under positive iterates of T. Since EN c E 0

N=0
and both sets have the same measure,

m(F) = m(E N ~ E N) = m(E N E 0) = m(E).

N=0

It remains to show a point of F returns to F infinitely often.

Let x E F, then 3 0 < nI < n2 ... 9 Tni(x) E E V i. Consider

Tn1(x). Tn1(x) E E and enters E infinitely often under positive

iterates~ namely, n2-nl, n3-nl, ... since Tni-n1(Tn1(x)) =Tni(x) E E;

thus Tn:(x) E F. Similarly one shows Tni(x) E F V i. //

Remark:

We do not need that T be m e a s u r e - p r e s e r v i n g in the hypothesis;

we need only assume that T is incompressible, i.e., if B E B,

T-IB c B then m(B) = m(T-IB).

~4. Ergodicity

Let (X,B,m) be a p r o b a b i l i t y space and T: X ~ X be a measure-

p r e s e r v i n g transformation. If T-IB = B for B E ~, then

T - I ( x \ B) = X \ B and we could study T in two separate parts,

namely TIB and TIx\B. If 0 < m(B) < 1 this has simplified the

study of T. We need a concept of i r r e d u c i b i l i t y for m e a s u r e -

p r e s e r v i n g transformations, such that if T has this i r r e d u c i b i l i t y

property then the study of T cannot be split into two parts as

above. E r g o d i c i t y is such a concept. Also, we would like some way

of splitting a m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n into ergodic parts

in a canonical way. This can be done in r e a s o n a b l y well behaved

measure spaces. (See Rohlin [3].)

Definition 1.2:

T: (X,B,m) ~ (X,B,m) is er$odic if for B E B, T-IB = B = m ( B ) = 0

or m(B) = i.
 22

Theorem 1.3:

The f o l l o w i n g are e q u i v a l e n t for m e a s u r e - p r e s e r v i n g T: X ~ X:

(i) T is ergodic.

(2) m(T-IB~B) = 0, B ~ B = m(B) = 0 or i.

(3) V A,B E B, m(A),m(B) > 0 3 n ~ 0 9 m ( T - n A 0 B) > 0.

Proof: (i) = (2). Suppose m(T-IBAB) = 0. Let

B = fi 0 T-iB ~ B. Then
n=0 i=n

B = (B UT-1B UT-2B U . . . ) n (T-1B UT-2B U . . . ) A (T-2B UT-3B U . . . ) n . . . ;

and t h e r e f o r e T-IB : B® and m(B ) : m(B). Hence, by (i),

m(B®) = 0 or i, and t h e r e f o r e m(B) = 0 or i.

(2) = (3). Let m(A) > 0, m(B) > 0 and s u p p o s e (3) is false,

i.e., V n > 0 m ( T - n A n B) = 0. Then

m(( 0 T-nA) 0 B) = 0.
n=l

Let A' = 0 T-nA. Then T - I A ' c A' and m(A') = m(T-IA'); so that
n=l
m ( T - I A ' A A ') = 0, and by (2) we h a v e m(A') = 0 or I. But T-IA cA'

and T is m e a s u r e - p r e s e r v i n g , so that m(A') = i. But this contra-

dicts the above fact t h a t m(A' A B) = 0.

(3) = (1). Suppose (I) is false, i.e., 3 B E 6, T-IB = B and

0 < m(B) < i. Then m ( T - n B N (XXB)) = 0 V n > 0 which contradicts

(3). II

A characterization of e r g o d i c i t y in terms of f u n c t i o n s is g i v e n

by the f o l l o w i n g results.
 23

Theorem 1.4:

Let T: (X,B,m) ~ (X,B,m) be m e a s u r e - p r e s e r v i n g ; then the

following are equivalent:

(i) T is e r g o d i c .

(2) Whenever f is m e a s u r a b l e and (foT)(x) = f(x) V x E X then

f is c o n s t a n t a.e.

(3) Whenever f is m e a s u r a b l e and (foT)(x) = f(x) a.e. then f is

constant a.e.

(4) Whenever f ~ L2(m) and (foT)(x) = f(x) V x E X then f is

constant a.e.

(5) Whenever f ~ L2(m) and (foT)(x) = f(x) a.e. then f is

constant a.e.

Proof: (i) = (3). Suppose f is m e a s u r a b l e and loT = f a.e.;

while T is ergodic. We can a s s u m e that f is r e a l - v a l u e d for if

f is c o m p l e x - v a l u e d we can consider the real and imaginary parts

separately. Define

X(k,n) = {x: k/2 n ~ f(x) < (k+l)/2 n} k ~ Z, n > 0.

We h a v e

T-iX(k,n)AX(k,n) c {x: (foT)(x) ~ f(x)}

and h e n c e m(T-iX(k,n)AX(k,n)) : 0 so that by (2) of T h e o r e m 1.3

m(X(k,n)) = 0 or i.

Fix n, then ~J X(k,n) : X which is a d i s j o i n t union; so for

kEZ
each n 3 unique kn ) m(X(kn,n)) = 1. Let Y = 5 X(kn,n).
n=l
Then m(Y) : i and f is c o n s t a n t on Y so that f is c o n s t a n t

a.e.

Trivially we have (3) = (2) = (4), (5) = (4), and (3) ~ (5).

So it r e m a i n s to show:

(4) = (!). Suppose T-IE = E, E E B. Then ×E E L2(m) and

 24

(×EOT)(x) = XE(X) V x E X so, by (4) ×E is constant a.e. Hence

P

m(E) = ] XEdm = 0 or i. //

Note:

A similar c h a r a c t e r i z a t i o n in terms of Ll(m) functions is true,

since in the last part of the proof XE is in Ll(m) as well as

L2(m). Also we could use real Ll(m) or L2(m) spaces.

Remark:

The following remark comes later (Theorem 5.5) but we preview it

now in order to analyze our examples.

Let X be a compact metric space and m a Borel probability

measure on X which gives positive measure to every non-empty open

set. If T: X ~ X is continuous and ergodic with respect to m then

m({x I {Tnx I n ~ 0 is dense}}) = i.

Proof: Let {Un}~= 1 be a base for the topology of X.

{Tax I n ~ 0} is dense in X = x E ~ 0 T-ku • Since

n=l k=0 n

T-I(0 T-kUn) c 0 T-ku and T is measure-preserving and ergodic

k=0 k=0 n

we have m( 0 T-ku ) = 0 or i. Since 0 T-ku is a non-empty

k=0 n k= 0 n

open set we have m( 0 T-kUn ) = i. The result follows. //

k:0

Note that this result is applicable when m is Hear measure on a

compact metric group and T is an affine transformation.

Examples:

We shall now see when the examples of §I are ergodic.

(i) I on (X,B,m) is ergodic iff all members of B have measure

0 or I.

(2) Consider T: K ~ K, T(z) = az. T is ergodic iff a is not a

root of unity.

Proof: Suppose a is a root of unity, then ap = 1 for some

 25

p # 0. Let f(z) = zP; then clearly f ~ constant a.e. and foT : f.

Conversely, suppose a is not a root of u n i t y and foT = f, f EL2(m).

Let f(z) : b n zn be its Fourier series. Then by above,

~ b anz n = ~ b zn and t h e r e f o r e b (a n - i) = 0. If n # 0 then

n n n

b : 0, and so f is c o n s t a n t a.e. //
n

(3) Let T(x) : ax on a c o m p a c t metric group G, then T is er-

n
godic iff {a }n~Z is dense in G. In particular, T ergodic =

G is abelian.

Proof: Suppose firstly that Tx = ax is ergodic. By t h e above

remark it f o l l o w s that {anx0} ~ is d e n s e for some x0 and so {an}~

n. n.
is d e n s e since if y E G 3 {n i} with a Ix 0 ~ yx 0 i.e., a I ~ y.
n
Conversely, suppose {a }n~Z is d e n s e in G. This implies g is

abelian. Let f E L2(m) and loT = f. By ( 7 ) of §5 o f C h a p t e r 0

f can be r e p r e s e n t e d as ~ biYi, where 7 i ~ G. Then

i
biYi(a)Yi(x) : ~- biYi(x) so that if b. ~ 0 then ¥i(a) : I
i i
and so 7i ~ i. Therefore only the constant term of the Fourier

series of f can be n o n - z e r o , i.e., f is c o n s t a n t a.e. //

(4) For an e n d o m o r p h i s m A of a c o m p a c t metric group G necessary

and sufficient conditions for e r g o d i c i t y are known. When G is

abelian, Halmos [4] and R o h l i n independently proved that A is

ergodic = whenever y o A n = y, n > 0, then 7 = i. Before proving

the general result we w i l l illustrate the p r o o f by showing that

2
A(z) = z on K is e r g o d i c .

Suppose foA = f, f E L2(m). We have that if f(z) = ~ an zn

then ~ anz2n = ~ an z n and therefore a n : a2n : a4n : ... So

if n ~ 0 we m u s t have an : 0 because ~ fan 12 < ®. Therefore

f is c o n s t a n t a.e.
 26

Proof of the ~eneral result: Suppose that w h e n e v e r yAn : ~ we

have y = i, and let foA = f with f E L2(m). Let f(x) have the

Fourier series [ any n where Yn E G and [ fan 12 < -. Then

~ anYn(AX) = ~anYn(X) , so that if Yn' Tn oA' Tn oA2' "'" are all

distinct their c o r r e s p o n d i n g c o e f f i c i e n t s are equal and therefore

zero. So if a n ~ 0, Yn (Ap) = Yn for some p • 0. Then Yn = 1 by

assumption and so f is c o n s t a n t a.e.

C o n v e r s e l y let A be ergodic and yAn = y, n • 0. If n is

the least such integer, f = T + T A + ... + y A n-I is invariant under A

and not a.e. c o n s t a n t (being the sum of o r t h o g o n a l functions), contra-

dicting the ergodicity. //

Consider now the case when G is the n-torus K n. The ergodicity

condition then becomes: A: K n * K n is ergodic ~ the matrix [A] has

no roots of u n i t y as eigenvalues.

Proof: Recall that, under the i d e n t i f i c a t i o n i n ~ Z n, if

m = Im~l E Zn then [A]t m = A(m). If A is not ergodic then

\m/
n
k
[A] m = m for some m ~ O and k > 0. Then [A]t has 1 as an

eigenvalue so [A] t has a k-th root of unity as an eigenvalue.

C o n v e r s e l y if [A] t has a k-th root of unity as an eigenvalue

then [A]~ has 1 as an e i g e n v a l u e so that [A]~ - I induces a

singular linear t r a n s f o r m a t i o n R n. Hence [A]~ - I induces a many-

to-one map of Zn into Zn and so there exists O ~ m E Zn with

k
[A] t m_ = m._ //

So, for example, all the e n d o m o r p h i s m s z ~ z k, Ikl • I, of K

are ergodic.

Chu [i] has c o n s i d e r e d the case w h e n G is nonabe!ian. He has

shown that a continuous e n d o m o r p h i s m of a compact group G onto it-

self is ergodic if and only if the induced map on the r e p r e s e n t a t i o n

 27

ring R(G) has no finite orbit except the constant functions. The

r e p r e s e n t a t i o n ring, R(G), is the ring generated by the coefficients

of all i r r e d u c i b l e unitary representations of G over the complex

field.

(5) For affine t r a n s f o r m a t i o n s of compact m e t r i c groups n e c e s s a r y and

sufficient conditions for ergodicity are known. The simplest case is

when G is a compact, connected, metric, abelian group. If Tx =

a.A(x) is an affine t r a n s f o r m a t i o n of the compact, connected, metric~

abelian group G then the following are equivalent:

(a) T is ergodic.

(b) (i) Whenever ToA k = T for k > 0 then ToA = ¥, and,

(it) the smallest closed subgroup c o n t a i n i n g a and BG (where

Bx = x-l'A(x) ) is G (i.e., [a,BG] = G).

(c) 3 x0 E G with {Tn(x0): n ~ 0} dense in G.

(d) m({x: {Tnx: n ~ 0 is dense}}) = i.

(Note that conditions (i) and (it) reduce to the conditions given in

(3) and (4) in the special cases. The e q u i v a l e n c e of (a) and (b) was

i n v e s t i g a t e d by Hahn, Hoare and Parry.)

Proof: First note that B is an e n d o m o r p h i s m of G and com-

mutes with A.

(b) = (a). Suppose (i) and (it) of (b) hold. If loT = f,

f ~ L2(m) let f = ~ biTi' Yi ~ ~" Then

i biYi(a)Ti(Ax) = ~i biYi(x)" (*)

So if 7 i, YioA, 7ioA 2 , .. . are all distinct then b 1• = 0 or else

Ibi 12 < - is violated. Hence, if bi ~ 0 then Ti °An = Yi for

some n > 0, and by (i) yoA = y. But then (*) implies Yi(a) = 1

and so Ti(x) = 1 V x ~ [a,BG] and by (it) T i = I. So f is con-

stant a.e.

(a) = (d). This follows by the remark above.

 28

(d) = (c) is t r i v i a l .

(c) = (b). It r e m a i n s to show that if 3 x0 ~ G with

{Tnx0 : n ~ 0} dense in G t h en c o n d i t i o n s (i) and (ii) of (b) hold.

Suppose T oAk = 7, k ~ !, y E G. Let Yl = yoB. Then Tl(Tkx) =

Yl(a.Aa.....Ak-la)Yl(Akx) = 7(a-IAka)Yl(X) = 71(x). Hence Yl

assumes o n l y the finite number of v a l u e s Yl(X0), Yl(TX0), ...,

Yl(Tk-lx0) on the d e n s e set {Tnx0 : n z 0} and h e n c e assumes only

these values on G. Since G is c o n n e c t e d T1 m u s t be c o n s t a n t ,

i.e., Tl = i. Hence yA : T and c o n d i t i o n (i) holds.

If [a,BG] # g 3 T ~ i, T E G, with y(a) = i, and T(Bx) =I.

Then y(Tx) : 7(x) and so T assumes o n l y the value y(x 0) on the

dense set {Tnx0 : n ~ G} and t h e r e f o r e T is a c o n s t a n t . Hence

y ~ i, a contradiction, and we have shown that c o n d i t i o n (c) i m p l i e s

(ii). //

Wh e n G is Kn the e q u i v a l e n c e of (a) and (b) b e c o m e s : T = a.A

is e r g o d i c iff

(i) the m a t r i x [A] has no p r o p e r roots of u n i t y (i.e.,

other than i) as e i g e n v a i u e s ,

and (ii) [ a , B K n] = K n.

This is e a s i l y proved by a m e t h o d s i m i l a r to the one used in (4)

for the e n d o m o r p h i s m case.

Conditions for e r g o d i c i t y of a f f i n e transformations of c o m p a c t

nonabelian groups m a y also be found in Chu [i].

(6) The 2-sided (p0,...,Pk_l)-shift is ergodic.

Proof: Let A = the a l g e b r a generated by finite unions of

measurable rectangles. Suppose T-IE = E, E ~ B. Let ~ > 0 be

given, and c h o o s e A ~ A ) m(EAA) < ~; thus

Im(E) - m ( A ) I = Im(E h A ) + m(E\A) - m ( A hE) - m(A\E)I

< m(E\A) + m(A\E) < s.

 29

Choose n so large that B = T-nA depends upon different coordinates

from A; so, m(B NA) = m(B)m(A) = m(A) 2.

m(EAB) = m(T-nE&T-nA) = m(E~A) <

and since Ea(A NB) c E&AUEbB w e have m ( E a ( A N B)) < 2s, hence

Im(E) - m ( A N B) I < 2~

and

Im(E) -m(E)2I ~ Im(E) - m ( A 0 B) I + Im(AN B) -m(E)21

< 2g + Im(A) 2 -m(E)21

2~ + m(A)Im(A) -m(E) I +m(E)Im(A) -m(E) I

< 4s

since m(A),m(E) ~ 1. Since ~ is arbitrary m(E) = m(E) 2 which im-

plies that m(E) = 0 or i. //

(7) By a similar argument, we see that the 1-sided (p0,...,Pk_l)-

shift is ergodic.

§5. The Er$odic Theorem

The first major result in ergodic theory was proved in 1931 by

G. D. Birkhoff [i].

Theorem 1.5: (Birkhoff Er$odic Theorem)

Suppose T: (X,B,m) ~ (X,~,m) is measure-preserving (where we

n-i
allow (X,B,m) to be o-finite) and f E Ll(m). Then ~! i~0
"= f(Ti(x))

converges a.e. to a function f* 6 Ll(m). Also, f*oT = f* a.e., and

if m(X) < ~, S f*dm = S fdm.

 3O

Note:

If T is ergodie then f* = a constant a.e. and if m(X) < -

then f, _ 1 ~ f dm a.e.
m(X) J

Motivation:

(i) Suppose T: (X,B,m) ~ (X,B,m) is measure-preserving and E E B.

For x ( X, we could ask with what frequency do the elements of the

set {x, T(x), T2(x), ...} lie in the set E?

Clearly Ti(x) ~ E iff XETi(x) = I, so the number of elements

n-i
of {x, T(x), ..., Tn-l(x)} in E is ~ XETk(x); and so the rela-
k=0
rive number of elements of {x, T(x), ..., Tn-l(x)} in E equals
n-i
1 n-l~_
~ XETk(x ) . If m(x) = 1 and T is ergodie then ~i ~ XETi(x )
=0
m(E) a.e. by the note; and thus the orbit of almost every point of X

enters the set E with asymptotic relative frequency m(E).

(ii) We define the time mean of f to be

n-i
i
lim ~ ~ f(Tl(x))
n~- i =0

and the phase or space mean of f to be

1 IX f(x)dm.
m(X)

The ergodic theorem implies these means are equal if T is ergodic.

(The converse is also true.) So, it is important to verify ergodicity

for transformations arising in physics. This application to time

means and space means is more realistic in the ease of a 1-parameter

flow {Tt} of measure-preserving transformations. The ergodic theo-

rem then asserts lim ~ f(Ttx)dt exists a.e. for f E L!(m) and
T~® 0

equals ~ 1 [
)X f dm in the emgodie case if the map (t,x) ~ Ttx .is
 31

measurable.

A__nnApplication to Number Theory

Borel's T h e o r e m on Normal Numbers:

Almost all numbers in [0,i) are normal to base 2, i.e.,

1 / the number of l's in the first n digits ~ 1

" ~ of the binary expansion of x E [0,i) J ~ ~ a.e.

Proof : Let T: [0,i) ~ [0,i) be defined by T(x) : 2x mod i.

We know that T preserves Lebesgue measure and is ergodic, by exam-

pie 4 at the end of §4.

aI a2
Suppose x = -- + + ... has a unique binary expansion. Then
2 V
T(x) = T + --~ + --~ + . . . . . + + ... Let f(x) :
2 V
X[½,1)(x). Then

1 iff ai+ I = i
f(Ti(x)) = f ( a i +a li + 2+2 7 + "'" ) = I
0 iff ai+ ! = 0

Hence, the number of l's in the first n digits of the dyadic ex-
n-i
pansion of x is ~- f(Ti(x)). Dividing both sides of this equality
i=0
by n and applying the ergodic theorem we see that

i n-li=0
[ f(Tlx)
• a. e .' I X [½,1)dm = ~i a.e.

(using the fact that the binary-rational points form a set of Lebesgue

measure zero). //

The ergodic theorem can be applied to give other number theoretic

results. Some are obtained in Billingsley [i] and Avez-Arnold [I].

We now consider some preliminaries to the proof of the ergodic

theorem.
 32

Definition 1.3:

Let T: (X,B,m) ~ (X,B,m) be m e a s u r e - p r e s e r v i n g . Define an

operator UT on c o m p l e x - v a l u e d functions on X by:

(UTf)(x) : f(T(x)).

We have U T L P ( m ) c LP(m) and, since T is m e a s u r e - p r e s e r v i n g

IIUTfllp = IIflIp. Let L~(m) denote the r e a l - v a l u e d LP(m) functions,

then U T L ~ ( m ) c L~(m).

To prove Birkhoff's theorem we need:

Theorem 1.6: (Maximal Ergodic Theorem)

Let U: L~(m) ~ L~(m) be a p o s i t i v e linear operator (i.e.,

f ~ 0 = Uf ~ 0) w h i c h has n o r m ~ i. Let N > 0 be an integer. De-

fine f0 = 0, fn = f + Uf + U2f + ... + un-lf, and F N = 0~n~N

max f n >- 0.

Then I fdm ~ 0.
{X:FN(X)>0}

Proof: (due to A. Garsia) Clearly F N £ L~(m). We have for

0 -< n -< N FN -~ fn so, UF N >_ Uf n by p o s i t i v i t y , and hence

UFN + f ~ fn+l" Therefore

UFN(X) + f(x) a max f (x)

l~n~N n

= max f (x) when FN(X) > 0

0ANON n

= FN(X).

Thus f ~ F N - UF N on A = {x: FN(X) > 0}, so

 33

IA f d m >- IA FNdm - IA UFNdm

= ;X FNdm - ;A UFNdm since FN : 0 on X\A.

>- ;X FNdm - IX UFN dm since FN t 0 = UF N e 0.

>_ 0 since HUH -< i. //

Remark:

The conditions of Theorem 1.5 hold if U = UT for measure-

preserving T.

Corollary 1.6:
1
Let T: X ~ X be measure-preserving. If g E LR(m) and

i n-i
B e = {x EX: sup ~ [ g(Tm(x)) > e}
n~l m=0

then

I gdm ~ em(B flA)

B NA a

if T-IA : A and m(A) < ®.

Proof: We first prove this result under the assumptions m(X) < ®

and A = X. Let f = g -a, then Be = 0 {x: FN(X) > 0} so that

N:0
f f d m > 0 by Theorem 1.6 and therefore I gdm ~ am(B ). In
B B =
e e
the general case, using TIA in the place of T we see that

f g dm em(A 0 B ). //
AnB a
Proof of Birkhoff's Theorem: It suffices to prove the theorem
n-i
for f ( L (m). Let f~(x) : lim ~i ~ f(Ti(x)) and f,(x), :
n i:0
1 n-I
lim ~ ~ f(Ti(x)). We have f*oT = f*, f,oT = f, because if
n i:0
 34

n-i
= i 2 f(Tlx)
• (n+l)
an(X) then --6-- an+l(X) - an (Tx) = f(x)
n" For real
i=0
numbers ~ < a, let

Ea, ~ : {x EX: f,(x) < ~, a <f*(x)}.

Then T-IEa,~ : Ea, ~ and

1 n-I
Ea, ~ N {x EX: sup ~ ~ f(Ti(x)) > a} : E .
hal i=0 a,~

We now prove that m(Ea, ~) < - so that we can apply Corollary 1.6.

Suppose a > O. Let C c Ea, ~ with m(C) < -. Then h = f-a× C

is integrable and by the maximal ergodic theorem

(f-a×c)dm ~ O. (H N defined analogously to the FN in the

O{x H (x)>O}
N:O
maximal ergodic theorem.) But C c ~J {x: H N ( X ) > O} so that
N=O

IX Ifldm > am(C). Therefore m(C) _< [i IX Ifldm for every subset

of Ea, ~ with finite measure and hence m(Ea, ~) < ~. If a < 0 then
< 0 so we can apply the above with -f and -~ replacing f and

a to get m(Ea, ~) < ®.

n-i
Let B = {x ~X: sup ~1 ~ f(Tlx) > a}. Then by Corollary 1.6:
n>_l i=0

; fdm : ; fdm >_ am(Ea,~ nBa) = am(Ea,~), i.e.,

Ea,~ Ea,~NBa

I fdm > am(E ) (*)

E - a,~

If we replace f,a,~ by -f,-~,-a respectively we get that

(-f)* : -f,, (-f), : -f* and

I f dm ~ ~m(E a ) (**)
E a,~ '~ "
 35

So, if = > ~ then m(E ,~) : 0, and since

{x: f,(x) < f*(x)} ~ ~<=

U E ='~ '

~,~ rational

we have m{x: f,(x) < f*(x)} : O i.e., f*(x) : f (x) a.e. Therefore
i n-i
~ f(Ti(x)) converges a.e.
i=0
To show f* E Li(m) we use the part of Fatou's Lemma that says
for non-negative i~tegrable functions gn lim I gndm < " implies
lim gn is integrable. Let

i ln_1
gn(X) : ~ ~- fCTl(x))
i:O
Then

I
gndm =
I I 1 n-i
~
~ i:0
f(Ti(x) ) dm _< Iftdm

so that lim I gndm < "' and by Fatou's Lemma lim gn = If* is
integrab!e. Hence f~ is integrable.

remains to show t h a t I f m-- I if mCX < - . ,et

D nk : {x EX: ~_k<f*(x) <k+l.n~ where k E Z, n >_ i. For each small

n n
> 0 we have D k n B( k_~) = Dk and by Corollary 1.6
n
I k n
fdm e ( ~ - s ) m ( D k) so that
n
Dk
fdm ~ ~ m(D ) . (***)
n
Dk

Then

f*dm _< k+In m(D ) _< ~ m(D ) + f dm (by (***)).

n n
Dk Dk

Summing over k we get that

 36

IX f*dm _< re(X)

n
+ IX fdm V n >_ I;

F
thus | f*dm ~ | fdm since m(X) < ®, Applying this to -f instead
JX
of f gives IX (-f)*dm E I - fdm i.e., -; f,dm ~ - ; fdm. Since
X X X
f, = f* a.e. we get that f f*dm a I fdm.
JX ~X
Hence, I f*dm : I fdm. //

Corollaries 1.5:

(i) Let (X,B,m) be a probability space and T: X ~ X measure-

preserving, then T is ergodic iff V A,B E B
n-i
i_ Z m(T-iA n B) ~ m(A)m(B).
n i=0

Proof: (=) Suppose T is ergodic. Putting f = ×A in Theo-

rem 1.5 gives ~i n~l

i=0 ×A(Ti(x) ) ~ m(A) a.e. Multiplying by ×B:

n-i
ni i__~
.= 0 ×A(Ti(x))×B ~ m(A)×B a.e.

By the dominated convergence theorem if we integrate we get

i n-i
~- m(T-iA n B) ~ m(A)m(B).
i=0

n-i
(=) Let T-1E : E, E E B. Let A : B : E. Then 1 m(E)

m(E) 2 so m(E) : m(E) 2, hence : 0 or i. //

(ii) __L
p Er$odic Theorem: (Von Neumann [i], [2])
Let i -< p < -. Let T be measure-preserving on the probability
space (X,B,m). If f E LP(m), 3 f* £ LP(m) ) f*oT : f* a.e.

n-i
and II~i ~.:0 f(Tix) f*(x) llp ~ 0.
 37

Proef: If g
is bounded and measurable then g E L p V p and
n-i
by the ergodic theorem we have that 1 [ g(Tix) * g*(x) a.e.
n i=O
Clearly g* E L'(m) and hence g* ( L P ( m ) . Also,

g(Tix) - g, (x)
n-i ip
I ~1 ~ ~ 0 a.e. and by the bounded convergence
i:O
n-i
theorem, II~1 ~"=0 g(Tix) - g*(x)ilp ~ 0 i.e., V ~ > 0 3 N(~,g)

if n > N(s,g) and k > 0

n-1 n+k-i
1 1
II ~-'=
±=~0 g(Tix) n+k i=0 g(Tix) II
P < •

n-i °
Let f E LP(m), and Mn(f)(x) = 1 i__~
0": f(TZx) We must show that

{Mn(f) } is a Cauchy sequence in LP(m). Note that IiMn(f)IIp _ IIflIp

Choose g E L'(m) 9 IIf -glI < g/4; then

P

llMnf - Mn+kf IIp ~ IiMnf -Mng[I p + IIMng- Mn+kgH p + l[Mn+kg-Mn+kfil p

_~ ~/4 + ~/2 + g/4 =

if n > N(s/2,g) and k > 0. We have f*oT = f* a.e. because

(n+l)(Mn+if)(x) _ (Mnf)(Tx) : f(x)

n n

§6. Mixing

We have seen that T is ergodic iff V A,B E B,

~N=.l m(T_iA NB) ~ m(A)m(B)

N
l=0• °

Definitions 1.4:

(i) T is weak-mixin $ if V A,B E B

N-I
1 ~ Im(T-iA NB) - m(A)m(B)I ~ 0.
 38

(ii) T is s t r o n g - m i x i n ~ if V A,B ~ B

m(T-NAnB) ~ m(A)m(B).

Note:

(i) T strong-mixing = T weak-mixing.

(ii) T weak-mixing = T ergodic.

This is so because if {an} is a sequence of real numbers t h e n

n-i 1
0 I " I n--
an ~ = ~ ~ fail ~ 0 = ~ ai ~ 0.
i:0 i-~-0":

(Put a = m ( T - n A N B) - m(A)m(B).)
n
(iii) An example of an ergodic T w h i c h is not w e a k - m i x i n g is given

by T(z) = az on K, where a is not a root of unity. (See the

end of this section for the proof.)

(iv) There are examples of w e a k - m i x i n g T which are not strong-

mixing. Kakutani has an example c o n s t r u c t e d by c o m b i n a t o r i a l methods,

and M a r u y a m a c o n s t r u c t e d an example using Gaussian processes. Chacon

and K a t o k - S t e p i n also have examples. Indeed, if (X,B,m) is a proba-

bility space, let ~(X) denote the c o l l e c t i o n of all invertible

measure-preserving transformations of (X,8,m). If we topologize

~(X) with the "weak" t o p o l o g y (see Halmos [2])~ the class of weak-

mixing t r a n s f o r m a t i o n s is of second c a t e g o r y while the class of strong

mixing t r a n s f o r m a t i o n s is of first category.

The following result shows it suffices to check the convergence

properties on an algebra g e n e r a t i n g B.

T h e o r e m 1.7:

If T: X ~ X is m e a s u r e - p r e s e r v i n g and A is an algebra

generating B then
 39

(i) T is ergodic iff 'V A,B E A

n-I
1 ~ m(T-iA A B ) ~ m(A)m(B),
n i=O

(ii) T is weak-mixing iff V A,B ~ A

1 n-1
~ Im(T-iAAB) - m(A)m(B) I ~ O, and
i=O

(iii) T is strong-mixing iff V A,B ~ A

m(T-nA AB) ~ m(A)m(B).

Proof: Let 6 > 0 and E,F E B. Choose Eo,F 0 E A with

m(EAE O) < 6, m(FAF O) < 6. Then

m((T-nE AF)A(T-nE0 0 £0 )) < 26

and therefore

]m(T-nE nF) - m(T-nE 0 0 F0) I < 26.

Hence
n-I
i 0K k1:~ m(T-kE nF - m(E)m(£) I

_<
i1 n-1
~ [m(T-kE NF) - m(T-kE0 NF0)] I
k:0

1 n-1
"+ ] ~- ~- m(T-kE 0 A F O) - m(Eo)m(F O) ]
k=O
1 n-1
+ I~ ~ m(Eo)m(F O) - m(Eo)m(F)I
k:O
n-1
1 ~
+ I~-k= ° mCEo)m(F ) _ m(E)m(F) I

1 n-i
-<26+ I ~ ~ m(T-kEo NF O) - m(Eo)m(F O)I + s + 6
k:O

for each n. (i) follows by the known behavior of the right hand side
 40

of this inequality.

To prove (ii) first show

Im(T-kE N F) - m(E)m(F) I ~ 4e + Im(T-kE0 N F 0) - m(E0)m(F0) I (*)

, 1 n-i
by leaving out ~ ~ " from the above inequalities, and then take
k:0
Cesaro averages of each side of (*).

(iii) follows immediately from the inequality (*). //

Theorem 1.8:

If {an} is a bounded sequence of real numbers then the follow-

ing are equivalent:

n-i
1 [ fail ~ 0.
(I) E i=0

(2) 3 J c Z +, J of density zero, i.e.,

cardinality (J n {O,l,...,n-l}) ) O,
n

such that lima = 0 provided n ~ J.

n n

(3) ~-
1 "= ~o
n-i
lail
2
-.,. o .

Proof: If M c Z+ let aM(n) denote the cardinality of

{0,1,...,n-l} N M.
1
(i) = (2). Let Jk : {n ~ Z+: lanl ~ ~} (k > 0). Then

Jl c J2 c . . . . Each Jk has density zero since ~ I~± fail

n i:0
1 1 (n). Therefore there exist integers 0 : £0 < £1 < £2 < "'"
K ~ ~Jk
such that for n { £k,

! i
eJk+l (n) < ~ •

Set J : U [Jk+l N [£U,Zk+l)].~ We now show that J has density

k:0
 41

zero. Since Jl c J2 c ..., if £k ~ n < £k+l we have

J n [0,n) : [J n [0,£k)] U [J n [£k,n)] c [Jkn [0,£k)] U [Jk+l n [O,n)],

and therefore

1 ~j(n) ~ ~[ajk(Zk) + ZJk+l(n)] ~ ~[aJk(n) + ~Jk+l(n)] < ~i + k-~l .

i
Hence ~ ~j(n) ~ 0 as n ~ -, i.e., J has density zero. If

1
n > £k and n ~ J then n ~ Jk+l and therefore lanl < ~-~. Hence

lim lanl = 0.
Jinx®

(2) ~ (I). Suppose lanl ~ K V n. Let s > 0. There exists

N such that n ~ Ns, n ~ J imply lanl < s~ and M such that

~j(n)
n ~ Me implies - - < s. Then n ~ max(N ,M e ) implies
n s

K i= 0 fail : n jn{o,1 ..... n-l} i~Jn{0,1 ..... n-l}

K
< ~ ~j(n) + s < (K+I)~.

(i) - (3). By the above it suffices to note that lim ..lan] = 0

Jinx-
iff lira fan 12 = 0. //
J~n~-

Corollary 1.8:

T is weak-mixing iff V A,B ( B 3 J(A,B) of density zero in

Z+ ) lim m(T-nA nB) = m(A)m(B) iff V A,B ( B

n ~
n{J(A,B)
1 n-1
Im(T-iA nB) - m(A)m(B)l 2 ~ 0.

Remark:

To say T is strong-mixing means Zhat any set B ( B as it

 42

moves under T becomes, asymptotically, independent of a fixed set

A ( B. T is weak mixing means B becomes independent of A if we

neglect a few instants of time. T is ergodic means B becomes inde-

pendent of A on the average.

The next result expresses the mixing concepts in functional form.

Recall that UT is defined on functions by UTf = foT.

Theorem 1.9:

Suppose (X,B,m) is a probability space and T: X ~ X is

measure-preserving. Then

(a) T is ergodic iff for all f,g ( L 2 ( m )

! n-i
I (UTf,g) ~ (f,l)(l,g)
n i=0

iff for all f (L2(m)

n-i
1 ~ (UTf, f ) ~ (f,l)(l,f).
i=O

(b) T is weak-mixing iff for all f,g (L2(m)

n-i
i E i(u f,g) - - o
n i:O

iff for all f (L2(m)

n-i
! ~ l(UTf,f) - (f,l)(l,f) I ~ 0
n i=0

iff for all f ~ L2(m)

n-i . i2
i (U~f,f ,l)(l,f) 0
i=O

(c) (i) T is strong-mixing iff

(2) for all f,g (L2(m), (U~f,g) + (f,l)(1,g)

iff (3) for all f (L2(m), (U~f,f) ~ (f,l)(l,f).

 43

Proof: Ca), Cb), and (c) are proved using similar methods. We

shall prove (c) to illustrate the ideas. Slight modification of this

proof will prove (a) and (b).

(2) = (i). This follows by putting f =XA, g =XB, for A,B E B.

n
(i) = (3). We easily get that for any A,B ( B, (UTXA,XB)
n
(XA,I)(I,XB). Fixing B, we get that (UTh,X B) ~ (h,l)(l,X B) for
any simple function h. Then, fixing h, we get that (U~h,h)

(h,l)(l,h). So (3) is true for all simple functions.

Suppose f (L2(m), and let e > 0. Choose a simple function h

9 llf -hll 2 < e, and choose N(e) so that n ~ N(s) implies

l(U~h,h) -(h,l)(l,h) I < e. Then if n ~ N(s)

- (f,1)Cl,f)l <

n
÷ ICUSh,f)- (U h,h) l + (h.1)Cl,h)l

+ I(h,1)(l,h) -(f,1)(l,h) I + l(f,l)(1,h) -(f,1)(l,f)

l(UT(f-h),f) ; + l(UTh,f-h) I

+ ~ + ICl,h) l l C h - f , 1 ) l + ICf,1)llCl,h-f~l

Ilf-hlI211fll2 + llf-hlI211hIl2 + ~ + llh}1211f-hll2 + IlflI211h-fll2

by the Schwartz inequality

~]Ifll2 + ~(llfll2 + ~ ) + ~ + (llfll2 + ~ ) ~ + ~IIfll2.

n
Therefore (UTf,f) ~ (f,l)(l,f) as n ~ ®.

(3) = (2). Let f E L2(m) and let Hf denote the smallest

(closed) subspace of L2(m) containing f and the constant functions

and satisfying UTH f c Hf.

= n
Ff {g ( L2(m): (UTf,g) ~ (f,l)(l,g)}

is a subspace of L2(m), is closed, contains f and the constant

 44

functions and is UT invariant so it contains Hf. If g ~ Hf then

(U f,g) : 0 for n ~ 0 and (1,g) : 0 and therefore Hf ¢ Ff.

Hence Ff = L2(m). //

Definition 1.5:

If T: X ~ X is measure-preserving, define T×T: XxX ~ XxX by

(T×T)(x,y) = (T(x),T(y)). This is a m e a s u r e - p r e s e r v i n g transformation

on (XxX,BxB,mxm) by T h e o r e m i.i since it is measure-preserving on

measurable rectangles and hence on finite disjoint unions of such

rectangles.

Theorem I.I0:

If T is a m e a s u r e - p r e s e r v i n g transformation on a probability

space X then the following are equivalent:

(i) T is weak-mixing.

(2) TxT is ergodic.

(3) TxT is weak-mixing.

Proof: (i) = (3). Let A,B ( B, C,D ( B. 3 Ji,J2 of density

zero such that

lim m ( T - n A n B) = m(A)m(B)
n~J1
n~

lim m(T-nC ND) = m(C)m(D).

nCJ2
n~

Then

lim (mxm){(T×T)-n(AxC) n (BxD)} = lim m(T-nA Q B)m(T-nC n D)

n ~J1 UJa n~J1 UJa
n~ n~

= m(A)m(B)m(C)m(D)

: (mxm)(AxC)(mxm)(CxD).

Thus the proper relationship holds for rectangles and hence for finite

disjoint unions of these rectangles. These we know form an algebra F

 45

which generates the ~-algebra B. By Corollary 1.8 we have

n-i
1 [ Im(T-iA N B) - m(A)m(S) I ~ 0 V A,B E F and the result follows
n i=0
by Theorem 1.7.

(3) = (2) is clear.

(2) = (1). Let A,B E B. We have that

n-i = 1 n-i
n~ i=0
~ m ( T - i A n B) n ~0"= (m×m)((TxT)-i(AxX) n (BxX))

(m×m)(A~X)(m×m)(B×X) by (2)

= m(A)m(B).

Also
n-i ~2 1 n-i
1 [
n i=0
(m(T-iA n B), = X (mxm)((TxT)-l(AxA) n (BxB))
i=0

(mxm)(AxA)(m×m)(BxB) by (2)
: m(A)2m(B) 2

Thus

n-i
i_ Z {m(T-iA QB) -m(A)m(B)}
n i=0

n-I
{m(T-iAN B) 2 - 2 m ( T - i A n B)m(A)m(B) +m(A)2m(B) 2}
n i=0

2m(A)2m(B) 2 - 2m(A)2m(B) 2 = 0.

Therefore T is weak-mixing by Corollary 1.8. //

Definition 1.6:

Let T: (X,B,m) ~ (X,B,m) be a measure-preserving transformation

on a probability space. We say that k is an eisenvalue of T,

(k E C) if 3 f ~ 0 E L2(m) ~ UTf = Xf in L2(m); i.e.,

f(Tx) = kf(x) a.e. We call f an eisenfunction corresponding to k.

 46

Remarks:

(i) If k is an e i g e n v a l u e of T then Ikl = 1 since

llfll2 = IIUTfll2 = (UTf,UT f) = (kf,kf) = Ikl211fll2.

(ii) k = 1 is always an eigenvalue c o r r e s p o n d i n g to any constant

eigenfunction.

D e f i n i t i o n 1.7:

We say that T: X ~ X has continuous s p e c t r u m if 1 is the only

eigenvalue of T and the only e i g e n f u n c t i o n s are the constants.

Observe that T h a s continuous s p e c t r u m iff k = 1 is the only

e i g e n v a l u e and T is ergodic.

We shall n e e d the f o l l o w i n g result from spectral theory to prove

the next theorem. The proof can be found in Halmos [3].

Spectral T h e o r e m for U n i t a r y Operators:

Suppose U is a u n i t a r y operator on a complex Hilbert space H.

Then V f E H, 3 a unique finite Borel measure ~f on K 9

(unf, f) = I knd~f (k) V n ~ Z.

JK

If T is an i n v e r t i b l e m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n then

UT is unitary, and if T has continuous s p e c t r u m then ~f has no

atoms for all f E L2(m) with (f,l) = 0.

T h e o r e m i.ii:

If T is an invertible m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n of a

p r o b a b i l i t y space then T is w e a k - m i x i n g iff T has continuous

spectrum.

Proof: (=). Suppose fT = kf a.e., f E L2(m). If k ~ 1

then i n t e g r a t i o n gives (f,l) = 0 and by the w e a k - m i x i n g p r o p e r t y

 ~7

n-i
i_Io
.: i - o
i.e.~
1 n-1
[ I(kif, f) I -* 0.
i:0

Since Ikll = i, this gives (f,f) = 0 and therefore f = 0 a.e.

If k = 1 then f = constant a.e. by the ergodicity of T.

(=). Suppose T has continuous spectrum. We show that if

f ~ L2(m) then
n-i
I ~ l(U~f,f) - (f,ll(l,f) l2 ~ O.
i=O

If f is constant a.e. this is true. Hence, all we need to show is

that (f,l) = 0 implies

1 n-i
~- I(U~f,f) I2 ~ 0.
i:O

By the spectral theorem it suffices to show that if ~f is a continu-

ous (non-atomic) measure on K then

1 nil if kid~f(k)12 -~ 0.
i:O

We have
i n~l (I kid~f(k)
%1
i:O
if id f( )12 : ~ i=0
I k_id~f(k))

1 n-1
I _id~f (~) )

- 1 nf 1 II (k~)id(~f×~f)(k,~) (by Fubini's Theorem)

- ~i:O
K×K

:
;f(inl i) ~ ~- (k~) d(~f×Zf)(k,~).
i=0
K×K
 48

If (k,T) is n o t in the diagonal of K×K then

n i=0 (kr~)

as n ~ -. Since ~f has no atoms (~f×~f)(diagonal) : 0 and t h e r e -

fore the integrand ~ 0 a.e. The integrand has m o d u l u s ~ i, so that

we can a p p l y the b o u n d e d convergence theorem to o b t a i n the result. //

We n o w investigate the m i x i n g properties of the examples men-

Tioned in 21.

Examples:

(i) I : identity on (X,B,m). I is e r g o d i c iff all the e l e m e n t s

of B have measure 0 or 1 iff I is s t r o n g - m i x i n g .

(2) T(z) = az on K. T is n e v e r weak-mixing since if f(z) = z

then f(Tz) = f(az) : af(z) and f ~ constant. (This has used the

trivial part of T h e o r e m i.ii.)

(3) No r o t a t i o n on a c o m p a c t group is w e a k - m i x i n g . We h a v e already

mentioned that if T is e r g o d i c then the group G is a b e l i a n ; and

then if Tx = ax and y is any character of g we have T(Tx) =

T(a)T(x), which shows that T does not have continuous spectrum.

(4) Endomorphisms of c o m p a c t metric groups are strong-mixing iff

weak-mixing iff e r g o d i c .

Proof : We shall give the proof when G is a b e l i a n . It suf-

fices to show that if the endomorphism A: G ~ G is e r g o d i c then A

n
is s t r o n g - m i x i n g . If y,5 ( G then (UAY,5) : 0 e v e n t u a l l y u n l e s s
n
7 : 5 m i. So a l w a y s (UA7,5) ~ ( y , l ) ( l , 5 ) . Fix 5 ( G. The col-

lection

n
H 6 : {f 6 L2(m): (UAf,6) ~ (f,l)(l,5) }

is a c l o s e d subspace of L2(m). (To c h e c k H5 is closed, suppose

 4g

fk E H and fk ~ f E L2(m). For 6 ~ 1 it is clear that H8 =

L2(m). So suppose (1,5) = 0. Then

l(UAf,6)i
n ~ n
I(uAf,5) (U~fk,6) I + l(UXfk,5)I
n
llf-fkll2116112 + l(UAfk,5) I (by the Schwarz inequality)

n
: IIf-fkIl2+ I(UAfk,6)l <

if n ~ N(s) where k is chosen so that llf-fkll2 < ~/2 and N(~)

is chosen so that n ~ N(s) implies l(U~fk,6) i < e/2.) Since H6

contains G, it is equal to L2(m). Fix f ~ L2(m) and consider

£f = {g E L2(m): (U~f,g) ~ (f,l)(l,g)}. Lf is a closed subspace of

L2(m), contains G by the above~ and so equals L2(m). Hence A is

strong-mixing. //

(5) An affine t r a n s f o r m a t i o n T = a.A on a compact metric abelian

group is strong-mixing iff it is weak-mixing iff A is ergodic.

Proof: We shall give the proof in the case when G is connected.

Let Bx = x-iA(x) and recall that T is ergodic iff

(i) 7 °Ak : Y k > 0 implies yeA = y, and

(ii) [a,BG] = G.

If A is ergodic then BG = G since the endomorphism B of

is one-to-one. Choose b E G so that B(b) = a. Put ¢(x) = bx;

then CT = A¢. ¢ preserves Haar measure m on G and hence induces

a unitary operator U¢ on L2(m). We then have that UT and UA

are isomorphic as Hilbert space operators by the induced conjugacy

UTOU ¢ : U¢oU A.

Now A is strong-mixing by (4), and hence UA satisfies the condi-

tions of part (c) of Theorem 1.9. But then UT satisfies these con-

ditions, and hence T is strong-mixing.

 50

Conversely if T is strong-mixing and A is not ergodic then

by (i) 7 °A = 7 for some T % i. But then

n
I(UTT,7)I = l(7(a)7(Aa)...T(An-la)7,7)l : I1711~ : 1

which does not converge to I(7,1)(1,7)I = 0, contradicting the

mixing of T. So if T is strong-mixing then A is ergodic. //

(6) The 2-sided (p0,...,Pk_l)-shift is strong-mixing. This is proven

by doing the easy verification on measurable rectangles, then on their

disjoint finite unions, and then applying Theorem 1.7.

(7) Similarly, the 1-sided (p0,...,Pk_l)-shift is strong-mixing.

 C h a p t e r 2: I s o m o r p h i s m and Spectral Invariants

§i. I s o m o r p h i s m of M e a s u r e - P r e s e r v i n g T r a n s f o r m a t i o n s

What should we mean by saying that two m e a s u r e - p r e s e r v i n g trans-

formations are the "same"? We must bear in m i n d that sets of measure

0 do not m a t t e r from the point of view of m e a s u r e theory.

Examples:

(I) Let T be the t r a n s f o r m a t i o n Tz = z 2 on the unit circle K

with Borel sets and H a a r measure, and let S be given by Sx = 2x

mod 1 on [0,1) w i t h Borel sets and Lebesgue measure. C o n s i d e r the

map ~: [0,i) ~ K defined by x ~ e 2~ix ~ is a b i j e c t i o n and pre-

serves m e a s u r e (check on finite unions of intervals @nd use T h e o r e m

i.i). Also ¢S = T¢. So, we want to regard T and S as the "same".

(2) Again, let S be the t r a n s f o r m a t i o n Sx : 2x mod 1 on [0,i)

with Borel sets and Lebesgue measure, and let T2: X ~ X be the

1-sided (½,½)-shift. Define ~: X ~ [0,i) by

aI a2 a3
~(al'a2'a3'''') = -+-2 V + V + "'"

is not o n e - t o - o n e only at points (al,a2,...) whose coordinates are

constant eventually. 4, though, is onto and ~T 2 = S~. Also ~ pre-

serves measure; we can check this out on dyadic intervals and then on

their finite d i s j o i n t unions and apply T h e o r e m i.i.

Suppose D is the set of points of the space X of the 1-sided

(½,½)-shift w h i c h have constant coordinates eventually. Then T21D =D

and so T21(X\D) = X\D. Let D2 consist of the dyadic r a t i o n a l s in

[0,i). Then S-ID2 = D2, so that S - I ( [ 0 , 1 ) \ D 2) = [0,1)\D 2.

 52

We see that the d i a g r a m

T2
X\D XkD
*I I ~ i:I onto

[0,1)\D2 S [0,1)\D 2

commutes.

We w o u l d like to c o n s i d e r these transformations as i s o m o r p h i c

since, after r e m o v i n g sets of m e a s u r e zero, we can t h r o w one to the

o t h e r by an i n v e r t i b l e measure-preserving transformation.

Definition 2.1:

Suppose ( X l , B l , m I) and (X2,B2,m 2) are p r o b a b i l i t y spaces to-

gether with m e a s u r e - p r e s e r v i n g transformations TI: X 1 ~ XI,

T2: X 2 ~ X 2. We say that T1 is i s o m o r p h i c to T2 if 3 M I E BI,

m l ( M I) = i, M 2 E B2, m 2 ( M 2) = i )

(i) TIMI c M1 ' T2M 2 C M 2, and

(ii) 3 an i n v e r t i b l e measure-preserving transformation

¢: M I ~ M 2 ) CTl(X) = T2¢(x) V x E M I.

We w r i t e T I ~- T 2. (In (ii) the set Mi (i = 1,2) is a s s u m e d to be

equipped with the G-algebra M i n B i = {M i n B I B E B i} and the re-

striction of the m e a s u r e mi to this G-algebra.)

Remarks :

(a) ~- is an e q u i v a l e n c e relation.

(b) T I ~- T 2 = T n~_
1 n
T2, V n> 0

(c) If TI and T2 are i n v e r t i b l e we can take MI,M 2 so that

TIM 1 = MI, T2M 2 = M2; we just take 5 TIMI, 5 T2M 2 as the new

sets.
 53

~2. Con~ugacy of M e a s u r e - P r e s e r v i n ~ Transformations

Although the notion of isomorphism, introduced above, is useful

in practice the following is m a t h e m a t i c a l l y more natural.

Given (X,B,m) we define an equivalence relation on B by

saying that A % B iff m(AAB) = 0. Let B denote the set of

equivalence classes. ~ is a Boolean o-algebra under the operations

induced from the usual operations on B. m induces a measure

on B. We call (B,m) a measure algebra. Note that for B E B , ×B

is a uniquely defined member of L2(m).

Suppose T: X ~ X is measure-preserving. If A % B then

T-IA % T-IB; so we have a map T-I: B ~ B which is defined by

T-I(B) = T~B. ~-i preserves unions, intersections, and complements,

and m(T-IB) = m(B).

Definition 2.2:

A map ~: (B2,m2) ~ (Bl,ml) of measure algebras is called an

isomorphism of measure algebras if # is a surjective bijection and

preserves complements and countable unions and

Definition 2.3:

We say that TI: X I ~ Xl, T2: X 2 ~ X 2 are conjugate if 3 a

measure algebra isomorphism #: (~2,m2) ~ (~!,ml) such that

-i -1
~T2 = T I ~.

Remarks:

(i) Conjugacy is an equivalence relation.

(2) T1 ~ T 2 = T1 and T2 are conjugate.

Just let ~ = ~-i, which is uniquely defined although is

not defined on the whole of X I.

 54

In Lebesgue spaces (i.e., probability spaces isomorphic to a sub-

interval of [0,i] with Lebesgue measure possibly together with

countably many atoms) con~ugacy implies isomorphism. In particular,

a compact separable metric space with a completed Borel measure is a

Lebesgue space.

~3. Spectral Isomorphism

Suppose T: X ~ X is a measure-preserving transformation on a

probability space (X,B,m). We have defined UT: L2(m) * L2(m) by

f ~ foT, and noted that V f,g 6 L2(m) we have (UTf,UTg) = (f,g).

Also, if T is one-to-one, UT is unitary. A spectral property

of T is a property of U T.

Definition 2.4:

Measure-preserving transformations T1 on (Xl,~l,ml), and T2

on (X2,B2,m 2) are spectrally isomorphic if 3 a linear operator

W: L2(m 2) ~ L2(m I) such that

(i) W is invertible

(ii) (Wf,Wg) = (f,g) V f,g ( L2(m2 )

(iii) UTIW = WUT2.

(The conditions (i), (ii) just say that W is an isomorphism of Hil-

bert spaces.)

Remarks:
(i) Spectral isomorphism is an equivalence relation.

(2) If #: (B2,m2) ~ (Bl,ml) is a measure algebra isomorphism then

induces an invertible linear map V: L2(m2 ) ~ L2(ml ), by

VXB = X~(B), with the properties:

(a) (Vf,Vg) = (f,g) V f,g ( L2(m2 ).

 55

(b) V, V -I map bounded functions to bounded functions.

(c) V is multiplicative on bounded functions.

Proof: V is defined as follows. Let B 2 E B2; then

V(XB 2) = X@(B2 ) which is unambiguous in L2(m2 ). We then extend V

to simple functions and then to L2(m2 ) functions. The properties

of @ guarantee this can be done. (a), (b), and (c) are proved by

checking first for characteristic functions, then for simple functions,

and then extending to the whole of L2(m2 ). //

(3) If T1 and T2 are conjugate then they are spectrally isomor-

phic.

Proof: Suppose @: (B2,m2) ~ (Bl~ml) isan isomorphism of

measure algebras such that = Tia,. Let V be defined as in
remark (2). It remains to check that

VUT2 = UTIV.

First, on characteristic functions

UTIV(X~2) : UTI(X@~2) : X~iI@~2 : × @~2-1 ~2 : V(× ~2 1 ~2 ) : VUT2(XB2)"

Therefore UTIV and VUT2 agree on characteristic functions and

hence on linear combinations of characteristic functions. By their

continuity we have UTIV = VUT2. //

The following tells us when spectral isomorphism implies conjugacy.

Theorem 2.1:

An invertible isometry V: L2(m2 ) ~ L2(ml ) is induced by an iso-

morphism of measure algebras (in the same sense of remark (2)) if both

V and V -I take bounded functions to bounded functions and V(fg) =

V(f)V(g) whenever f and g are bounded and in L2(m2 ).
 56

Proof: Let B 2 { B2. We have ×2B2 = ×B2 so that

V(X~2) = V(XB2)V(XB2) = V(XB2),

and we see that V(×B2) takes 1 and 0 as its only values. Thus

3 B 1 E B1 such that V(XB2) = XBI a.e. We define @: (B2,m2)

(Bl,ml) by @(#2 ) = #i" This is unambiguous since if m2(B2AA 2) = 0

then IIXB2 - XA211 = 0 so that IIV(XB2) V(XA2)II = 0. Clearly, V

is induced by @ in the sense of remark 2.

We now show that @ is an isomorphism of measure algebras.

First, • is invertible by doing the above for V -I. Also,

~2(B2 ) = m2(B 2) = XB2XB2 dm 2 = (XB2,XB2 )

= (V×B2,V×B2)~
~ = (×~(B2),X~(B2)). = ~I(~B2 ).

It remains to show that # preserves complements and countable

unions. First note that since V is norm-preserving and maps charac-
teristic functions to characteristic functions, V(1) = i.

Since ×#2 + X~2\#2 =

1 in L2(m2 ) applying V to both sides

gives X@B 2 + X @ ( X ~ B 2 ) = 1 so XI\@B2 = {(X2\B2 ). Therefore @

preserves complements.

Suppose B,C ~ ~2" Then

X#u 9 = xB + ×c - ×#nc = ×B + Xc - ×B×c.

Taking V of both sides we get:

X{(BUC) = X{(B) + X{(C) - ×{(B)x~(C) = ×{(B)U{(C)"

Thus ~(B UC) = ~(B) U ~(C) and hence, (by induction) preserves all
 57

finite unions.
Let BI,B2 , .... ~n'''" ( ~2' then

X ~ X a.e.
U
n ~i
i;1
iU__-lBi

and also in L2(m2) by the bounded convergence theorem. Since V is

an isometry it is continuous, so,

V< X n > ~ V<Xi_Y.l > = X (0 ] in L2(ml )

iUl~ i Bi • ~i=l B l;

On the other hand,

X : X

i=i

by the above and so converges to X in L2(ml). Therefore

0@B.
i= 1 ~l

~(iU=iBi) : iUl~Bi • //

Corollary 2.1:
If TI: X 1 ~ XI, T2: X 2 ~ X 2 are measure-preserving and if

UT1V = rUT2 for V: L2(m

2). ~ L2(ml ) satisfying the conditions of

Theorem 2.1, then T1 and T2 are conjugate.

§4. Spectral Invariants

Definition 2.5:
A property P of measure-preserving transformations is a
isomorphism
conjugacy invariant if the following holds:
spectral
 58

isomorphic
Given T1 has P and T2 is conjugate to TI,
spectrally isomorphic

then T2 has property P.

Note:

A spectral invariant is a conjugacy invariant, and a conjugacy

invariant is an isomorphism invariant.

Theorem 2.2:

The following are spectral invariants of measure-preserving

transformations:

(i) Ergodicity

(ii) Weak-mixing

(iii) Strong-mixing.

Proof: (i) T is ergodic iff {f ~ L2(m): UTf = f} is a one-

dimensional subspace.

(ii) T is weak-mixing iff 1 is the only eigenvalue and T

is ergodic.

(iii) Suppose WUT2 = UTIW and T1 is strong-mixing. We have

to show that

(UT2h,k) ~ (h,t)(1,k) V h, k E L2(m2 ).

This is true if h is constant or if k is constant, so assume

(h,l) = 0 = (k,l). Since T1 is ergodic then T2 is ergodic by (i)

and since W sends the invariant functions for T2 onto those for

TI, W maps the subspace of constants in L2(m2 ) onto the subspace

of constants in L2(ml ). So (Wh,l) = 0 = (l,Wk). Since W pre-

serves the inner product~

(UT2h,k) = (WUT2h,Wk) = (UT1Wh,~rk) ~ 0

since TI is strong-mixing. Therefore T2 is strong-mixing. //

 59

~5. Examples

Recall that T1 is isomorphic to T2

= TI is conjugate to T2

= T1 is spectrally isomorphic to T2

and the converse of the first implication holds in all "decent"

measure spaces.

(I) Consider Ti,T2: K ~ K given by Tl(Z) = alz, T2(z) = a2z

where aI is a root of unity and a2 is not a root of unity. Ti is

not ergodic while T2 is ergodic. Hence they cannot be spectrally

isomorphic.

(2) Let T(z) = az where a is not a root of unity. We know that

T is ergodic but not weak-mixing. Consider A: T 2 ~ T 2 defined by

A(z,w) = (zw,z). Since none of the eigenvalues of the matrix i 0

are roots of unity, A is weak-mixing. Hence T and A are not

spectrally isomorphic.

(3) Let at least two of the numbers {pl,P2,...~pn} be non-zero, and

n
Pi = i. Let the same be true for the numbers {ql,...,qm}. We
i=l
claim that the 2-sided (pl,...,pn)-shift and the 2-sided (ql,...,qm)-

shift are spectrally isomorphic but not n e c e s s a r i l y conjugate. A con-

sideration of entropy shows that they need not be conjugate. (See

Chapter 4.)

Consider the special case of the (½,½)-shift T, with

X = ~ {-i,i}. A basis for L2({-I,I}) consists of the constant

function 1 and the identity map sending

(-i) ~ (-i), ! ~ i.

Moreover, L2(T[xi ) is the tensor product of the spaces L2(Xi ) so

that we have an orthonormal basis for L2(X) consisting of all

 60

functions X ~ C of the form:

g0({Xn }) : 1

and, for nI < n2 < ... < n r

gnl, ... ,nr({Xn } ) : xnl'Xn2 ". "''Xnr

Note that

UTgnl, ... ,nr({Xn}) = ( g n l , . . . , n r °T) ({Xn})

: xnl+l'Xn2+l'...'Xnr+l : gnl+!,n2+l,...,nr+l({Xn}),

that is,

U T g n l , . . . , n r : g n l + l , . . . , n r + 1.

So we have an o r t h o n o r m a l basis for L2(X) of the form;

n n
f0 ---I, {UTfl}n( Z , {UTf2}n( Z , ...

Diagramatically, the basis has the form

f0 ~ i

-2 -i 2
"''' UT f l ' UT f l ' f l ' UTfl' UTfl' "'"
(*)
-2 -I 2
"''' UT f2' UT f2' f2' UTf2' UTf2' "'"
• . . : :

Definition 2.6:

An invertible measure-preserving transformation T: X ~ X has

countable Lebesgue spectrum if there exists an o r t h o n o r m a l basis for

L2(X) of the form:

n
f0 ~ i, {UTfj} j ~ i, n E Z,

i.e., a basis as in (*) above•

 61

Remarks:

(i) Any two transformations with countable Lebesgue spectrum are

spectrally isomorphic.

Proof: If T: X ~ X, S: Y ~ Y have bases:

n n(Z
f0 ~ i, {UTfj}j(Z+ for L2(X)

n n(Z
g0 ~ i, {Usgj}j(Z+ for L2(y)

n n
we define W: L2(y) ~ L2(X) by go ~ f0' Usgj ~ UTfj and extend by
linearity. Thus WU S : UTW and S and T are spectrally isomorphic.

(2) Using a similar method to the one used above for the (½,½)-shift

one can show that if at least two of {pl,p2,...,pn } are non-zero

then the 2-sided {pl,P2,...,pn}-shift has countable Lebesgue spectrum.

Theorem 2 . 3 :

If T has countable Lebesgue spectrum it is strong-mixing.

n
Proof: Let {f0,UTfm : n (Z, m >0 } be the basis. Then, clear-
!y, as p ~ -

(UTOUTfm,UTf q) ~ (U fm,l)(l,U fq) V k,n ( Z, m,q ~ 0.

Fix k and q and consider

Hk, q = {f ~ L2(m): (U f,U fq) ~ (f,l)(l,UTfq) }.

Hk, q is a closed subspace of L2(m) (c.f. proof in example (4) §6

n
Ch. i) and contains the basis {f0,UTfm : n (Z, m >0} and hence is
equal to L2(m). Fix f (L2(m) and let Lf = {g ( L2(m):

(U~f,g) ~ (f,l)(l,g)}. Lf is a closed subspace of L2(m), contains

the basis by The above, and therefore is equal to the entirety of

L2(m). Hence

(U~f,g) ~ (f,l)(l,g) V f,g ( L 2 ( m ) . //

 62

Suppose A: G ~ G is an ergodic automorphism of a compact

abelian metric group. Then the automorphism A: G * G has no finite

orbits except for the orbit of the identity. (This is what the er-

godicity of A says.) Since the characters form a basis for L2(m)

we can conclude that A has countable Lebesgue spectrum if we can

show there are infinitely many distinct orbits of A. This is proved

in Halmos [2].

In Chapter 4 we shall consider a whole class of transformations

with countable Lebesgue spectrum.

 Chapter 3: Measure-Preservin$ Transformations

with Pure Point Spectrum

In this chapter we study a class of m e a s u r e - p r e s e r v i n g transfor-

mations for which the conjugacy problem is solved and for which spec-

tral isomorphism implies conjugacy.

~i. Eisenfunctions

Suppose T is an ergodic m e a s u r e - p r e s e r v i n g transformation of a

probability space (X,B,m). Suppose k is an eigenvalue correspond-

ing to the e i g e n f u n c t i o n f, i.e., f ~ 0, f E L2(m), UTf =

kf E L2(m) ((foT)(x) = kf(x) a.e.). Then

(i) Ikl : 1 and Ifl is a constant a.e.

We have f(T(x))f(T(x)) = k[f(x)f(x) a.e. Integrating both

sides we get that llfll2 : IkI211fll2. Therefore Ikl = i. Also

If(T(x))l = Ikllf(x) I a.e. = If(x)I a.e. Thus Ifl is a T-invariant

function and, since T is ergodic, Ifl = a constant a.e.

(2) Eigenfunctions corresponding to different eigenvalues are

orthogonal.

Suppose k ~ ~, UTf = kf, UTg = ~g. Then

(f,g) = (UTf,UTg) = (kf,~g) = k~(f,g)

and k~ ~ 1 implies (f,g) = 0.

(3) If foT = kf, goT = kg then f = c.g where c is some con-

stant.

By (i) g ~ 0, so (f/g)oT = f/g which must be constant since

T is ergodic.

So (2) and (3) show that eigenspaces are 1-dimensional and

 64

mutually orthogonal.

(4) The eigenvalues of T form a subgroup of K.

If loT = kf, got = ~g then (f~)oT = k~f~.

By (2) if L2(m) is separable then the group of eigenvalues is

countable.

§2. Pure Point Spectrum

Definition 3.1:

An ergodic m e a s u r e - p r e s e r v i n g transformation T on a probabil-

ity space (X,B,m) has pure point spectrum (discrete spectrum) if

there exists an orthomormal basis for L2(m) which consists of eigen-

functions of T.

The following theorem shows that the eigenvalues determine com-

pletely whether two such transformations are conjugate or not.

Theorem 3.1: (Discrete Spectrum Theorem - Halmos and Von N e u m a n n

[i], 1942)

The following are equivalent for ergodic m e a s u r e - p r e s e r v i n g

transformations T1 and T2 each having pure point spectrum:

(i) T! and T2 are spectrally isomorphic.

(2) T1 and T2 have the same eigenvalues.

(3) T1 and T2 are conjugate.

Proof: (i) = (2) is trivial.

(3) = (i) is always true (see §3 of Chapter 2).

(2) = (I). For each eigenvalue k, choose fk ~ L2(ml )'

gk ~ L2(m2 ) such that

UT!f k : kfk, UT2g k : kg k

and

tf~L : Ig~i : i.
 65

We define W: L2(m 2) ~ L2(ml ) by W(gk) : fk and e x t e n d i n g by

lineariTy. We r e a d i l y see that W is a b i j e c t i v e isometry; m o r e o v e r

WUT2 = U T I W by c h e c k i n g this on the gk"

(2) = (3). To prove this we need the f o l l o w i n g result:

T h e o r e m 3.2:

Let H be a d i s c r e t e abelian group and K a divisible subgroup

of H (i.e., V k E K and V n > 0 3 a E K ) a n = k). Then

there exists a h o m o m o r p h i s m 4: H ~ K such that 41K = identity

(i.e., K is an a l g e b r a i c retract of H).

Proof: Let R consist of all retracts onto K from supergroups

of K in H, i.e., R consists of all pairs (M,~) where H~ M~ K

and ~: M ~ K is a h o m o m o r p h i s m such that 4I K = identity. R is

n o n - e m p t y as (K,id K) E R. We order R by extension, i.e.,

(MI,4 !) < (M2,42) if M1 ~ M2 and 421M 1 = 41 . This is a partial

ordering and every l i n e a r l y ordered subset has an u p p e r bound. So,

by Zorn's Lemma there exists a maximal element, say (L,p), of R.

We c l a i m That L = H. Suppose not, then c o n s i d e r g E H\L and

let M be The group generated by g and L.

Case i: If no p o w e r of g lies in L then every element of M can

be u n i q u e l y w r i t t e n in the form gla where a E L, i E Z. We define

~: M ~ K by %(gia) = p(a). We can easily check That % is a homo-

m o r p h i s m and that ~I K = id K. This then c o n t r a d i c t s the m a x i m a l i t y of

(L,p).

Case 2: Let n be The least positive integer such that gn E L.

Each element of M can be uniquely written as gla, where a E L,

0 ~ i ~ n-l. Since K is divisible, let g0 E K be such that

n " i
p(gn) = go" Then ~(gla) = g0p(a) defines a h o m o m o r p h i s m of M

into K such That ~I K = id k. Again, we have c o n t r a d i c t e d the

m a x i m a l i t y of (L,p).
 66

Thus it follows that L = H. //

We now prove that (2) = (3). Let A denote the group of eigen-

values of T 1 = the group of eigenvalues of T 2. Fix k E A. Let

fk E L2(m I) be chosen so that ]fkl = i, UTIf k = kf k and observe

that {fk: k E A} is a basis for L2(ml ). Also, choose gk E L2(m2 )

so that IgkI : i, UT2g k = kg k and observe that {gk: k E A} is a

basis for L2(m2).

UTifk~ = k~fk~ V k,~ E A

and also

UTi(fk.f ) = fk(T).f~(T) = (k~)(fk.f).

By (3) of §i there exists a constant r(k,~) E K such that

fk(x)f~(x) = r(k,~)fk (x) a.e.

Let H denote the collection of all functions X + K. Clearly,

H is an abelian group under pointwise multiplication. Moreover, K

is a subgroup of H if we identify constant functions with their

values.
By the previous Theorem 3.2 there exists a homomorphism ~: H ~ K

such that~ elK = id K. Let f* = ¢(fk)fk; then If~l = i, UTf ~ = kf~

and {f~: k ~ A} is a basis for L2(ml ). Also,

k ~ = ¢(fk)~(f~)fkf~
f'f* ¢(fkf~)fkf~

: ¢(r(k,~))¢(fk )fk r(k,~)

= r(k,~)¢(fk )r(k,~)fk~

: fek~"

Thus for all intents and purposes we can assume that fkf~ : fk~ and

gkg~ : gk~"
 67

Define W: L2(m2) ~ L2(ml ) by W(gk) = fk and extend by

linearity. W is bijective~ linear and preserves the inner product.

Also, WUT2 = UTIW. If we can show that W is multiplicative then

W is necessarily induced by an isomorphism of measure algebras (by

Corollary 2.1) and hence T1 and T2 are conjugate. But,

W(gkg~) : W(gk~) = fk~ : fkf~ : W(gk)W(g~)"

Let h,k be bounded functions in L2(m2 ). If we fix g~ and let a

!
finite linear combination of g~ s converge to h in L2(m2) we ob-

tain that W(hg ) = W(h)W(g ). Then if we let a finite linear combi-

nation of g ' s converge to k in L2(m2 ) we get that W(hk) =

W(h)W(k). It follows from this that W maps bounded functions to

bounded functions since this is also true for bounded h and any

k E L2(m2 ), and then W(h) is bounded since W(h)f E L2(m I) for

all f in L2(ml ) . //

Corollary 3.3:

If T is an invertible ergodic transformation with pure point

spectrum then T and T -I are conjugate.

Proof: They have the same eigenvalues. //

§3. Group Rotations

Example:

Let T: K ~ K be defined by T(z) : az where a is not a root

of unity. We know that T is ergodio. Let fn: K ~ C be defined

by fn(Z) : zn where n E Z.

f (Tz) : f (az) : anz n : anf (z).

n n n

Thus fn is an eigenfunction with eigenvalue an . Since the {fn }

form a basis for L2(K) we see that T is ergodic and has pure
 68

point spectrum.

These ideas carry over to rotations on compact abelian groups.

Theorem 3.3:

Let T, (T(g) = ag) be an ergodic r o t a t i o n of a compact abelian

group G. Then T has pure point spectrum. The e i g e n f u n c t i o n s of T

all consist of constant m u l t i p l e s of characters, and the e i g e n v a l u e s

are {y(a): 7 E G}.

Proof: Let y ( G. Then

zCTg) : yCag) : y ( a ) y ( g ) .

Therefore each c h a r a c t e r is an e i g e n f u n c t i o n and so T has pure point

spectrum. If there is a n o t h e r eigenvalue besides the members of

{y(a): y ( G} then the c o r r e s p o n d i n g e i g e n f u n c t i o n would be orthogo-

nal to all members of G, by (4) of §i, and so is zero. Hence

{y(a): 7 E G} consists of all the eigenvalues of T and the only

eigenfunctions are constant m u l t i p l e s of characters, using (3) of

§I. //

Theorem 3.4: ( R e p r e s e n t a t i o n Theorem)

An ergodic m e a s u r e - p r e s e r v i n g transformation T with pure point

spectrum is conjugate to an ergodic rotation on some compact abelian

group.

Proof: Let A = the group of all eigenvalues for T and give

the discrete topology. If L2(m) is separable then A is countable. In

the other case we shall need to use the character theory of groups

without a countable basis. Let G = A, the character group of A.

G is compact and abelian. The map ~: A ~ K given by ~(k) = k is

a h o m o m o r p h i s m of the d i s c r e t e group A and hence, by (2) of §5 of

Chapter 0, 3 a ( G so that ~(k) : k(a) k ~ A (where we write k

when we wish to c o n s i d e r "k" as a h o m o m o r p h i s m of G to K).

 69

Define S: G ~ G by S(g) : ag. We claim that S is ergodic.

Suppose fS : f, f E L2(G), f : ~bjkj, kj E A. Then the above

gives us that

~-bjkj(a)kj(g) ~ ~-bjkj(g)

so, bjkj(a)~ : bj. But ~]k'(a) : k.3 and therefore b.k.]] = bj. If

bj 0 then n e c e s s a r i l y kj = i. Thus kj(g) = 1 for all g ~ G,

and we get That f : a constant a.e. We know then that S is er-

godic, and by The previous theorem has pure point spectrum.

Again by The previous theorem the eigenvalues of S :

{T(a): T E G} = {a(k): k E A} = {k: k E A} : A. So, S and T have

the same eigenvalues and both have pure point spectrum. Hence the

Discrete Spectrum T h e o r e m tells us that they are conjugate. //

Theorem 3.5: (Existence Theorem)

Every subgroup A of K is The group of eigenvalues of an er-

godic m e a s u r e - p r e s e r v i n g Transformation with pure point spectrum.

Proof: The desired transformation is the rotation S con-

structed in The proof of Theorem 3.4. //

The conjugacy problem for ergodic m e a s u r e - p r e s e r v i n g Transforma-

tions with pure point spectrum is completely solved. We have some

very simple invarianTs, namely the eigenvalues, which determine when

two such Transformations are conjugate. Also there are some simple

examples, namely group rotations, such That each ergodic measure-

preserving Transformation with pure point spectrum is conjugate to

one of These examples. So each conjugacy class of ergodic measure-

preserving transformations with pure point spectrum is characterized

by a subgroup of K, and each subgroup of K corresponds To a con-

jugacy class.
 Chapter 4: Entropy

We are searching for c o n j u g a c y and/or i s o m o r p h i s m invariants.

In 1958 K o l m o g o r o v [I] i n t r o d u c e d The concept of entropy into ergodic

Theory, and this has been The most successful invarianT so far. For

example, in 1943 it was known That The t w o - s i d e d ( i / 2 , 1 / 2 ) - s h i f t and

the T w o - s i d e d ( i / 3 , 1 / 3 ) - s h i f t both have countable Lebesgue s p e c t r u m

and hence are s p e c T r a l l y isomorphic; but it was not known w h e t h e r they

were conjugate. This was r e s o l v e d in 1958 when K o l m o g o r o v showed That

They had entropies log 2 and log 3 r e s p e c t i v e l y and hence are not

conjugate. Von Neumann had had the same idea c o n s i d e r a b l y earlier,

but he was unable to prove That entropy was a c o n j u g a c y invariant.

The notion of entropy now used is slightly d i f f e r e n t from that used

by K o l m o g o r o v - The i m p r o v e m e n t was made by Sinai [I] in 1959.

§l. P a r t i t i o n s and Subal~ebras

Throughout, (X,B,m) will denote a p r o b a b i l i t y space.

D e f i n i t i o n 4.1:

A p a r t i t i o n of (X,B,m) is a disjoint c o l l e c t i o n of elements

of B whose union is X.

We shall be interested in finite partitions. They will be de-

noted by Greek letters, e.g., ~ : {AI,...,Ak}.

If ~ is a finite p a r t i t i o n of (X,B,m) then The c o l l e c t i o n of

all elements of B w h i c h are unions of elements of ~ is a finite

s u b - a - a l g e b r a of B. We denote it by A(~). Conversely, if C is a

finite s u b - a - a l g e b r a of B, say C = {Ci: i = l,...,n}, then the non-

empty sets of the form B I N . .. n B n where B l• = C.! or X \C i form

a finite p a r t i t i o n of (X,B,m). We denote it by ~(C). Thus we have

 71

a one-to-one correspondence between finite partitions and f i n i t e

sub-a-algebras of B.

Definition 4.2:

Suppose ~ and ~ are two finite partitions. ~ ~ ~ means that

each e l e m e n t of ~ is a u n i o n of e l e m e n t s of ~.

Note: ~ ~ n " A(~) ~ A(~)

Definition ~.3:

Let ~ = {AI,...,An} , ~ = {CI,...,Ck}. Then

v N : {A i n Cj: 1 ~ i ~ n , 1 ~ j ~k}.

If A and C are finite sub-a-algebras of B then A v C de-

notes the s m a l l e s t sub-~-algebra of B containing A and C.

Note: ~(A v C) = ~(A) v ~(C)

A(~ v n) = A(~) v A(~).

Suppose T: × ~ X is a m e a s u r e - p r e s e r v i n g transformation. If

~ = (AI,...,Am}, then by T-n~ we mean {T-nAI,...,T-nAm } and by

T-n(A) we m e a n {T-nA: A ~ A} (n ~ 0).

Note: if n ~ 0
~ (T-nA) : T-n~ (A)
t

A(T-n~) = T-nA(~)

T-n(AvC) = T-hA v T-nc

T-n(~ v~) = T-n~ v T-n~

_< ~ = T-n~ _< T-n~

A c_ C = T-nA c_ T-nc.
 72

D e f i n i t i o n 4.4:

If D and £ are (not n e c e s s a r i l y finite) s u b - G - a l g e b r a s of B,

~hen we write ~ ~ E if V D E D 3 E ~ E such That m(DAE) : 0

and V E E E 3 D E D such that m(DAE) = 0.

If D and E are finite, D ~ E, and if E(D) =

{ D I , . . . , D p , D p + I , . . . , D q} where m ( D i) > 0 for 1 ~ i ~ p and

m(D i) = 0 for p+l ~ i E q, then ~(E) = { E I , . . . , E p , E p + I , . . . , E s}

where m ( E i A D i) = 0 for 1 ~ i E p and m(E i) = 0 for p+l ~ i ~ s.

~ C means V D ~ D 3 C E C such that m(DAC) = 0.

92. Entropy

All logarithms are to base 2 and 0.1og 0 : 0.

Let A ~ B be finite. Let ~(A) : {AI,...,@~}. Then

k
H(A) : H(~CA)) : - [ m ( A i) log mCAi),
i=l

is called the entropy of A (or of ~(A) ). (This means that if

AI~...,A k denote The outcomes of an experiment then H(A) measures

The u n c e r t a i n t y removed (or i n f o r m a t i o n gained) by p e r f o r m i n g The

experiment. H(A) is a m e a s u r e of The u n c e r t a i n t y about which A.

l
a general point of X will b e l o n g To.)

Remarks:

(i) If A = {X,¢} then H(A) = 0. Here A represents the outcomes

of a "certain" e x p e r i m e n t so there is no u n c e r t a i n t y about the out-

come.

(2) If ~(A) = {AI,...,~} where m(A i) = i/k V i then

k
1 1
H(A) : - [ ~ log ~ : log k.
i=l
 73

Thus, we gain a lot of information if k is large. (Since all the

members of ~(A) have equal measure there is much uncertainty about

which Ai a point will belong to.)

(3) H(A) ~ 0.

(4) If A ~ C then H(A) : H(C).

Suppose T: X ~ X is measure-preserving.

If A is a finite sub-a-algebra of B we define

h(T,A) : lim ! H ( A v T-IA v ... v T-(n-!)A)

n

n-i
= lim ~1 H ( V T-iA),
n-~® i=0

which we call the entropy of T with respect to A. (Later (in

Corollary 4.4) we will show that the above limit always exists.)

(This means that if we think of an application of T as a passage of

n-i
one day of time, then V T-iA represents the combined experiment
i:0
of performing the original experiment represented by A on n con-

secutive days. h(T,A) is then the average information per day that

one gets from performing the original experiment daily forever.)

Remarks:

(5) h(T,A) a 0.
n-i n-i
(6) The elements of 5( V T-iA) = V ~(T-iA) are all the sets of
i:0 i:0
n-!
the form T-iA where ~(A) : {AI,...,Ak}.
i=0 mi

We define h(T) = sup h(T,A) where the supremum is taken over

all finite sub-o-algebras A contained in B and call this the

entropy of T. (h(T) is the m a x i m u m average information per day

obtainable by performing a finite experiment.)

 74

Remarks:

(7) h(T) ~ 0. h(T) could be +-.

(8) h(id x) : 0. If h(T) : 0 then h(T,A) : 0 for every finite A,

n-i
which implies that V T-iA does not change much as n ~ -.
i:0

Theorem 4.1:
Entropy is a conjugacy invariant and hence an isomorphism invari-

ant.

Proof: Let TI: X1 ~ X l , T2: X2 ~ X2 be m e a s u r e - p r e s e r v i n g

and let @: (B2,m2) ~ (Bl,ml) be an isomorphism of measure algebras

such that @T~I : T~I@. Let A2 be finite, A 2 c B2, and ~(A 2) :

{A!,.~.,Ar}. Choose B i ( BI such that ~i : ~(~i ) and so that

: {BI,...,Br} forms a partition of (Xl,Bl,ml). Let A 1 : A(~).
n-i
Now i=0N T?lBqi~ (where the qi ( {l,...,r}) has the same measure
n-i
as ~ T2iAqi since
i=0

iAq ) i00 iAql ) : i00 T i i@(Aql ) : n-i

n Ti iBqi : n-I
n T -i
1B .
i:0~ i ": . . . . : ~ ~ " i:0 i:0~, qi

n-i n-i
Thus, H( V TiiA I) = H( V T2iA 2) which implies that h(TI,A I)
i:0 i:0
h(T2,A 2) which in turn implies h(T I) >_ h(T2). By symmetry we then

get that h(T I) : h(T2). //

Theorem 4.2:
The function ~: [0,-) ~ R defined by:

0 if x = 0
~(x) :
x.log x if x ~ 0

is convex, i.e., ¢(=x+~y) e =¢(x) + ~¢(y) if x,y E [0,-),

~+~ = i.
 75

By i n d u c t i o n

k k

i=l i=l

k
if x. ( [0,'), ~. ~ 0, (~. : 1.
1 l i= 1 1

Proof:
~'(x) = log e + log x

¢"(X) = iog e > 0 on (0,-).

x

Suppose y > x; by the m e a n value theorem

~(y) ~(~x+~y) = ~'(z)~(y-x)

where ax+~y < z < y and

~(~x+~y) - ~(x) = ~'(w)~(y-x)

where x < w < ~x+~y.

Since ~" > 0 ~'(z) ~ ~'(w), thus

~(~(y)- ~(~x+~y)) : ~'(z)~(y-x)

e ~'(w)~(y-x) : c(~(x+~y) -~(x)).

Therefore ~(ex+~y) ~ e~(x) + ~(y) if x,y > 0, and h e n c e also if

x,y ~ 0 by continuity of ~. //

Corollary 4.2:

If ~ : {A I ..... Ak} then H(~) ~ log k.

Proof: Put ~l• = i/k and x.l : m ( A i) I ~ i ~ k. Then

H(~) ~ log k. //

Combined with remark (2) this corollary shows that among all t h e

partitions of X into k sets, the largest entropy is o b t a i n e d when

all the sets h a v e equal measure. This fits in w i t h our intuitive in-

terpretation of e n t r o p y .
 76

§3. Conditional Entropy

Let A,C ~ B be finite.

~(A) = {A 1 ..... ~ } , ~(C) = {C 1 ..... Cp}.

We define the entropy of A given ~ to be

k m(A i n C o ) m(A i N C o )
H(A/C) = - i m(Cj) [ log
j=l i=l m(Cj) m(Cj)

= - [ m(A i N Cj) log m ( A i N Cj) >_ 0

i,j m(Cj)

omitting the j-terms when m(Cj) = 0.

So to get H(A/C) one considers Cj as a measure space with

normalized measure m(')/m(Cj) and calculates the entropy of the

partition of C. induced by ~(A) (this gives
3
k m(A i N Cj) log m(A i N Cj)
) and then averages the answer taking
i=l m(Cj) m(Cj)

into adcount the size of Cj. (H(A/C) measures the average infor-

mation obtained from performing the experiment associated with A

given the outcome of the experiment associated with C.)

Let N denote the G-field {~,X}. Then H(A/N) = H(A). (Since

N represents the outcome of the trivial experiment one gains nothing

from knowledge of it.)

Remarks:

(i) H(A/C) a 0.

(2) If A ~ ~ then H(A/C) = H(~/C).

(3) If C ~ ~ then H(A/C) = H(A/~).

 77

Theorem 4.3:

If A,C,D are finite subalgebras of B then:

(i) H(A vC/g) = H(A/~) + H(C/A v p)

(ii) H(A v C) = H(A) + H(C/A)

(iii) A g C = H(A/D) ~ H(C/~)

(iv) A ~ C = H(A) ~ H(C)

(v) C ~ ~ ~ H(A/C) ~ H(A/~)

(vi) H(A) ~ H(A/N)

(vii) H(A v c/~) ~ H(A/~) + H(C/D)

(viii) H(A re) ~ H(A) + H(C).

(ix) If T is m e a s u r e - p r e s e r v i n g then:

H(T-IA/T-Ic) : H(A/C) and

(x) H(T-IA) = H(A).

(The reader should think of the intuitive meaning of each statement.

This enables one to remember these results easily.)

Proof: Let ~(A) = {Ai} , ~(C) = {Cj}, ~(~) = {D k} and assume,

without loss of generality, that all sets have strictly positive

measure (since if ~(A) = {AI,...,A k} with m(A i) > 0 1 ~ i ~ r

and m(A i) = 0 r ~ i 5 k we can replace ~(A) by

{AI,...,Ar_I,A r U Ar+ 1 U ... U Ak} ).

m(A i n cj N n k)
(i) H ( A v C/O) = - i,--,k
[j m(A i N Cj N D k) log
m(D k )

m(A iN Cj n D k) m(A iN C. N D k) m ( A iN D k)
But : ] unless m(A i N D k) = 0
m(Dk) m(AiN Dk) m(Dk)
and then the left hand side is zero and we need not consider it; and

therefore
 78

m(A i N D k)
H(A v c/9) = - [_ m(A iA Cj N D k) log
i,j,k m(D k)

- [ m(A i N Cj n D k) log m(Ain C~ n D k)

i,j~k m(A iN D k)

m(A i n D k)
= - [ m(A iA D k) log + H(C/A v 9)
i,k m(D k)

= H(A/9) + H(CIA v P).

(ii) PuZ 9 = N = {~,X}.

(iii) By (i)

H(CIg) = H(A v C/P) = H(AIP) + H(C/A v P) ~ H(AIg).

(iv) Put 9 = N in (iii).

m(D k N Cj) m(A iN D k)
(v) Fix i,j and let Zk = , xk : .. Then
m(Cj) m(D k)
by Theorem 4.2

(~
k
m(D k n c$)
m(Cj)
m(A i N D k) h
m(D k) /
~ ~
k
m(D k N C~)
m(Cj)
/m(A i N D k)
)
but since C ~ P the left hand side

/ m(A i A C~) > m(A iN C$)

m(Cj)
log
m(A iN C$)
m(Cj)
\ m(Cj) =

Multiply both sides by m(C.)

] and sum over i and j to give

m(A i N Cj) m(A i n D k) m(Ain D k)

m(A i N Cj) log ~ ~ m(D kN Cj) log
i,j m(Cj) i,j,k m(D k) m(D k)

m(A i N D k) m(A i N D k)
= ~- m(D k) log
i,k m(D k) m(D k)
 79

or -H(AIC) ~ -H(AIO).

Therefore H(AIO) ~ H(AIC).

(vi) Put C = N in (v).

(vii) Use (i) and (v).

(viii) Set D = N in (vii).

(ix), (x) Clear from definitions. //

Theorem 4.4:

If {an}n~ 1 satisfies a n ~ 0, an+ m ~ a n + am V n~m, then

lim an/n exists and equals inf an/n.

n~® n

Proof: Fix m > 0. For each j > 0 j = km + n where 0-< n < n~

Then

_a.l -
an+km < _an
_ +
akm ~
an
- - +
ka m =
a n
- - +
a m
_ _ .

j n + km km km km km km m

aj a __ a. a
As j ~- then k ~- so lim < m and therefore lim --$- < i n f m
j m j m
a a. a. a.
But inf __m_m ~ l i m _ l so t h a t lim ._l exists and equals i n f _l. //
m j 3 j

Corollary 4.4:
n-i
If A c B then lim 1 H( V T-iA) exists.
n
n~® i=O

(See also the remark after Theorem 4.6.)

n-I
Proof: Let a = H( V T-iA) a O.
n i=O
 80

n+m-i
an+ m : H( V T -iA)
i:0

n-i n+m-i T_iA ) by (viii)

H( ~/ T-iA) + H( V
i:0 i:n of T h e o r e m 4.3.

m-i
= a + H( V T -iA) by (x) in T h e o r e m 4.3.
n i=O

: a + a
n m

We then apply T h e o r e m 4.4. //

§4. Properties of h(T,A)

n-I
Recall that h(T,A) : lim ~1 H( V T-iA).
n-~- i: 0

Theorem 4.5:

Suppose A,C are subalgebras of B and T is measure-

preserving. Then

(i) h(T,A) ~ H(A).

(2) h(T,A vC) ~ h(T,A) + h(T,C).

(3) A £ C = h(T,A) ~ h(T,C).

(4) h(T,A) ~ h(T,C) + H(A/C).

(5) If T is invertible and m ~ 1 then

m
h(T,A) : h(T, V TiA).
i:-m

Proof:

n-i n-i
(l) 1 H( V T-iA) -< i ~ H(T-iA) by (viii) of Theorem 4.3.
n i=0 n i=0

n-i
= ! T H(A) by (x) of T h e o r e m 4.3.
n i=0

= H(A).
 81

n-I n-1 n-i •

(2) H( V T-i(A VC)) : H( V T -iA v V T -IC)
i:0 i:0 i:0

n-i n-i
-< H( V T -iA) + H( V T-ic) by (viii) of Theorem 4.3.
i:0 i:0

(3) If A ~ C then

n-i n-i
V T -iA c_ V T -iC V n >_ i
i:0 i:0

so one uses (iv) of Theorem 4.3.

n-i n-i n-i

(4) H( V T -iA) -<H(( V T-iA) v ( V T-ic) ) by (iv) of Theorem 4.3
i:0 i:0 i:0

n-i n-i n-I

= H( V T -IC) + H(( V T-IA)/( V T-Ic)) by (ii) of Theorem 4.3.
i=0 i=0 i=0

But, by (vii) of Theorem 4.3

n-i n-! n-i n-I
H((V T-iA)/( V T-it))-< ~ H ( T - i A / ( V T-Jc))
i=0 i:0 i=0 j=0

n-i
-< Z H(T-iA/T-iC) by (v) of Theorem 4.3
i=0

= nH(A/C) by (ix) of Theorem 4.3.

n-i n-I
Thus, H( V T-iA) -< H( V T -iC) + nH(A/C).
i:0 i:0

k-1 m •
(5) h(T , TiA) = lim ~1 H( V T-j ( V TiA))
-m k -~- j=0 i = -m
m+k-i
1 H(
= lim ~ V T -iA)
k ~- -m
 82

m+k-1 m m+k-i
H( V T-iA) = H( V T-iA v V T-iA)
-m -m m

m+k-i m m+k-i
: H( V T-iA) + H(( g T-iA)/( V T-iA))
m -m m

by (ii) of Theorem 4.3

k-i m m+k-i
: H(V T-iA) + H(( V T-iA)/( V T-iA))
i=0 -m m

by (x) of Theorem 4.3.

We want to show that

m m+k-1
! H ( ( V T-iA)/( V T-iA)) ~ 0 as k ~ -
k -m m

m re+k-1 m
But, ~I H(( V T-iA)/( V T-iA)) _< kl H( V T-iA) by (vi) of Theo-
-m m -m

rem 4.3 and so the result follows. //

Theorem 4.6:

If A ~ B is finite, and T is measure-preserving then

n
h(T,A) : lim H(A/( V T-iA)).
n~ i:l

Proof: The limit exists since the right hand side is non-

increasing in n by virtue of (v) of Theorem 4.3. We show by in-

duction that for n a 2

n-i n-i j
H( V T-iA) = H(A) + [ H(A/( k/ T-iA)).
i=0 j=l i=l

This is true for n : 2 by (ii) of Theorem 4.3. Assume that this

equality is true for n. We shall prove that it holds for n+l.

 83

n n n
H( V T-iA) : H ( V T-iA) + H ( A / ( V T-iA))
i:0 i:l i:l

by (ii) of Theorem 4.3

n-1 n
= H( V T -iA) + H(A/( ~/ T-iA))
i=0 i=l
by (x) of Theorem 4.3

n j
: H(A) + ~ H(A/( V T-iA)). (*)
j=l i=l

Dividing the above by n and taking the limit the result follows,

using the fact that the Cesaro limit of a convergent sequence is the

ordinary limit. //

Remark:
n-i
1 H(V T -iA) decreases to h(T,A).
n i:0

n-1 n
Proof: By (~) H( V T-iA) >_ nH(A/( V T -iA)) using (v) of
i=0 i=l
Theorem 4.3. Hence

n-i n n-i
n[H( V T-iA) + H(A/( V T-iA))] -< (n+l)H( V T-iA)
i:0 i=l i=0

n n-1 •
ioeo nH( V T-iA) <_ (n+l)H( ~/ T-1A)°
i:0 i=0

§5. Properties of h(T)

Theorem 4.7:
(i) For m > 0, h(T m) : mh(T).

(2) If T is invertible then h(T m) = Imlh(T) V m E Z.

Proof: We first show that

m-i
h(T TM , V T -iA) = mh(T,A).
i:0
 84

This follows since

k-i • m-i m km-i

lim I H ( V T-m3(v T-iA)) = lim ~-~ H( V T-iA)
k~- j :0 i:0 k~- i:0

: mh(T,A) .

m-i
Thus, mh(T) : m • sup h(T,A) : sup h(T m , V T -iA)
A finite A i:0

sup h(Tm,C) : h(Tm).

C

m-i
Also, h(Tm, A) -< h( Tm , V T-iA) : mh(T,A) by (3) of Theorem 4.5 and
i:0
so, h(T m) _< mh(T). The result follows from these two inequalities.

(2) It suffices to show that h(T -I) = h(T); and all we need to show

is that h(T-I,A) = h(T,A) for all finite A. But

n-i n-i
H( V TiA) = H(T-(n-l) V T iA) by (x) of Theorem 4.3
i:0 i=0

n-i
: H( V T-JA). //
j:O

Theorem 4.8:

Let (X,B,m) be a probability space and B0 be an algebra such

that the ~ - a l g e b r a generated by B0 (denoted by ~(B 0) ) satisfies

~(B 0) ~ B. Let C be a finite subalgebra of B. Then for every

> 0, there exists a finite algebra 9, ~ ~ SO such that

H(P/C) ÷ H(C/P) < ~.

Proof: Let ~(C) = {CI,...,Cr} and assume without loss of

generality that each Ci has positive measure. We can do this since

if CI,...,C s have positive measure and Cs+I,...,C r have zero

measure where 1 z s ~ r then ~(C') : {CI,...,Cs_I,C s U ... U Cr}

is such that H(C'/~) = H(C/~) and H(~/C') = H(~/C).

Since ¢(x) = x.log x is continuous and ~(0) = 0, ¢(i) = 0,

 85

3 0 < 50 < i such that -¢(x) < e/2r if 0 <_ x _< 50 or

1-60 _< x_< i.

We first show that if we choose a partition ~(D) = {DI,...,Dr}
m(C i)
) m(CiAD i) _< min 6 - 6 V i, then H(C/D) < e/2. For, if
l_<i<r 0 2
m(C i)
(D) satisfies the above inequality, then m(C i) < m(D i) +6 < m(D i)+ 2

m(C i )
which implies ~ < m(D i) and hence,

m(D i) - m(D i n c i) ~ m(DiAC i) < 6 < 50m(Di).

m(D i n c i)
Thus ~ I- 6
0"
m(D i )

Therefore, if j ~ i then m(Co n D i)

50 and hence,
m(D i )

(m(C~ n n i)
H(C/D) =- ~i m(Di)~j ¢\ m ( D . ) )
l

[ m(D i) ~ ~12r : ~/2.

i j

5
So m(CiADi) < O min m(C i) implies H(C/D) < s/2. If
2 l~i~r
5o
m(CiAD i) ~ -~- min m(C i) then this still holds and also m(Ci6D i) <
l~i~r
3
m(Ci)/4 which implies m(D i) > ~ m(C i) and hence

60 4 60
m(CiAD i) < -4-" 7 min m(Di) < T min m(Di).
l~i~r l~i~r

Therefore H(D/C) < s/2 also.

So it suffices to show we can choose D i 6 B0 with

5o
m(CiAD i) _< -~- min m(C i) = ~.
l_<i<r
 86

Choose k > 0 9 k(r-l)[l + r ( r - 1 ) ] < a; for each i, choose

Bi E B0 ~ m(CiAB i) < k. If i ~ j then B i N B j c (BiACi) U (BjACj)

so that m(Bi n B j) < 2k. Let N = U (BiNBj). We have

r-i
m(N) < r(r-l)k. Set D. = B.\N for 1 _< i < r and D = X ~ ~J D..
l l r i=l 1
{DI,...,D r } is a p a r t i t i o n of X and each D i £ B 0. If i < r then

DiAC i c (BiACi) U N and so,

m ( D i A C i) < k[l + r ( r - l ) ] < o.

r-I
However DrAC r c ~J (DiAC i) and therefore
i:l

m ( D r A C r) < (r-l)k[l +r(r-l)] < a.

So the theorem is proved. //

(If {A n } is a sequence of subalgebras of B then V An de-

n
notes the s u b - a l g e b r a of B generated by t h e A .)
n

Corollary 4.8.

If {A n } is an increasing sequence of finite algebras and

C ~ V An then H(C/A n) ~ 0 as n ~ -.
n

Proof: Let B0 = 0 A ; B0 is an algebra and C ~ ~(B 0) by

m=l m
hypothesis. By T h e o r e m 4.8 V ~ > 0 3 finite ~ 80 ~ H(C/P )< ~.

But ~ ! Am0 for some m0 since Ds is finite, so, if m >_ m 0

H(C/A m) ~ H(C/Am0) ~ H(C/P ) < ~.

Thus, H(C/A m) tends to zero as m ~ ®. //

For entropy to be useful we require methods of calculating its

value. The following is one of the main tools for calculating entropy.
 87

Theorem 4.9: (Kolmo~orov-Sinai Theorem)

Let T: (X,B,m) ~ (X,B,m) be an invertible m e a s u r e - p r e s e r v i n g

transformation and let A be a finite subalgebra of B 9

TnA o= B. Then h(T) = h(T,A).

n=-~

Proof: Let C ~ B be finite. We want to show that

h(T,C) e h(T,A).

m m
h(T,C) ~ h(T, V TiA) + H(C/ V TiA)
i=-m i=-m

by (4) of Theorem 4.5,

m
: h(T,A) + H(C/ V TiA)
i: -m

by (5) of Theorem 4.5.

Let A m : i:-m
V T iA. It suffices to show that H(C/A m) goes to zero

as m ~ -. This follows by Corollary 4.8. //

A similar result holds when T is not n e c e s s a r i l y invertible:-

Theorem 4.10:

If T: X * X is a measure-preserving transformation (but not

necessarily invertible) and if A is a finite algebra contained in

with V T i o_ B then h(T) : h(T,A).

i:0
m-i
Proof: This is similar to the previous theorem; use V T-iA
i=0
m
in the place of V T !A, and the formula
i=-m
m-i
h(T,A) : h(T, V T-iA)
i:0

(the proof of which is similar to (5) of Theorem 4.5). //

The following is sometimes useful in showing transformations

 88

have zero e n t r o p y . We shall use it l a t e r to show a rotation of The

unit circle has zero entropy.

Corollary 4.10:

If T is invertible and T-iA o= B for some finite A then

i:0
h(T) = 0.

Proof: By T h e o r e m 4.10

h(T) = h(T,A)

n
: l i m H(A/ v T-iA) by T h e o r e m 4.6.
n~- i=l

n
But T-iA ~ T-IB : B. Let A = V T-iA ; then A1 c A 2 c ...
i:l n i:l

and A o B By C o r o l l a r y 4 8 H ( A / A n) ~ 0 and h(T) : 0 //

n:l n : " " "

Remarks:

Entropy can be d e f i n e d for any countable partition of (X,B,m)

as follows: If ~ = {A!,A2,...} then

H(~) = - ~ m(Ai)log m ( A i)
i

(which m a y be infinite).

A countable partition E of X is c a l l e d a generator for an

invertible measure-preserving transformation T if

TnA(~) ~ B.

As in T h e o r e m 4.9, one can p r o v e that if { is a g e n e r a t o r and-

H(~) < - Then h(T) = h(T,~).

The b a s i c theorem on existence of g e n e r a t o r s ~ with H(~) <

was given by R o h l i n in 1963. To state it we need the following defi-

nitions: We s a y that (X,B,m) is c o u n t a b l y generated if t h e r e exists

a countable collection {B } of e l e m e n t s of B such That The

n
 89

~ - a l g e b r a g e n e r a t e d by the {B n} ~ B. (X,B,m) is c o m p l e t e if every

subset of a set of m e a s u m e zero is measurable. Suppose T: X ~ X is

a measure-preserving transformation, and X is c o u n t a b l y g e n e r a t e d

and complete. We say that T is aperiodic if

m( U {x ~ X: Tn(x) =x}) = 0.
nEZ
n~0

(The c o u n t a b l y g e n e r a t e d and completeness c o n d i t i o n s ensure that this

set is measurable.) Note that T ergodic implies T is a p e r i o d i c

(unless the space is finite).

T h e o r e m 4.11: (Rohlin [2], IS63)

Suppose (X,B,m) is a countably generated c o m p l e t e n o n - a t o m i c

p r o b a b i l i t y space and T: X ~ X is an invertible m e a s u r e - p r e s e r v i n g

transformation. Then T has a generator ~ with H(~) < - iff

h(T) < ® and T is aperiodic.

Thus, if T is ergodic and h(T) < - then T has a g e n e r a t o r

with H(~) < ®.

R e c e n t l y Krieger [i] proved:

Theorem 4.12:

If (X,B,m) is c o u n t a b l y generated and T is an invertible

ergodic m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n such that h(T) < ® then

T has a finite g e n e r a t o r

= {A 1 .... ,An}.

In fact ~ may be taken so that e h(T) S n ~ e h(T) + i.

Hence finite g e n e r a t o r s exist in the most i n t e r e s t i n g cases al-

though they may be d i f f i c u l t to find.

We now prove some more results that are useful for c o m p u t a t i o n

of entropy.
 90

Theorem 4.13:

If B0 is an a l g e b r a and ~(B O) ~ B then

h(T) : sup h(T,A)

where the s u p r e m u m is t a k e n over all finite subalgebras A of B0 .

Proof: Let 5 > 0. Let C ~ B be finite. By T h e o r e m 4.8 there

exists a finite D ~ B0 such that

5

H ( C / ~ 5 ) < 5.

Thus, h(T,C) ~ h ( T , ~ 5) + H(C/D5 ) by (4) of

Theorem 4.5
h ( T , P 5 ) + 5.

Therefore, h(T,C) ~ sup h(T,D) + ~,

D~B o
O finite

and thus h(T) ~ sup h(T,D).

DgB 0
D finite

The opposite i n e q u a l i t y is obvious. //

Theorem 4.14:

Let An be finite s u b a l g e b r a s of B such that A 1 ~ A 2 ~ ...

and A ~ B. Then h(T) = lim h(T,An).

n=l n - n~®

Proof: We note that h(T,A n) is an increasing sequence by (3)

of T h e o r e m 4.5. B0 = 0 A is an algebra and ~(B 0) ~ B. By Theo-

n= 1 n
rem 4.13 h(T) = sup h(T,C). If C ~ B0 is finite then C ~ An0
C~B 0
C finite
for some nO. Thus

h(T,C) ~ h(T,Ano).
 gl

which implies h(T) ~ lim h(T,A n)

and hence h(T) : lim h(T,An). //

n~

Theorem 4.15:

h(TI×T 2) = h(T I) + h(T2).

Proof: Let TI: (XI,BI,m I) ~ (XI,BI,ml) and

T2: (X2,B2,m 2) ~ (X2,B2,m2).

If A I g BI, A 2 g B2 are finite then AIXA 2 is finite;

~(AIxA 2) : {AIXA2: A I ~ ~(AI), A 2 ~ ~(A2)}.

Let TO : U AlXk 2
AI~BI,A2~B2
finite

= the algebra of finite unions

of measurable rectangles.

Thus, a(T 0) = BIXB 2 by definition of BI×B 2, and by Theorem 4.13,

h(TI×T 2) = sup h(TI×T2,C).

C~/o
C finite

But if C is finite, C & T0 then C ~ AI×A 2 for some finite

A 1 ~ B I, A 2 & B 2. Hence

h(TlXT 2 ) : sup h(TIXT2,AI×A2 ) •

AIgB 1
A2~B 2
AI,A 2 finite
 92

n-i
H( V (TIXT2)-i(AIXA2))
i:0
n-1 n-i
: (cV T iAl X V T iA2 )
i:0 i:0

: - ~ (ml×m2)(Ck×Dj)'log (mlxm2)(CkXDj)

n-i
where the Ck are the members of ~( V TIiAI),
i:0
n-i
and the D4J are the members of ~( ~/ T2iA2 )
i:0

: _ [ ml(Ck)m2(Dj)'log(ml(Ck)m2(Dj))

: - 5- ml(Ck)m2(Dj)'[log ml(C k) + log m2(Dj)]

: _ ~ ml(Ck)'log ml(C k) - ~ m 2 ( D j ) ' l o g m2(D j)

n-i n-1
: H( V TliA I) + H( V T2iA2).
i:0 i:0

Thus h(TIXT2,AIXA2 ) : h(TI,A I) + h(T2,A 2)

SO~ h(TlXT 2) : h(T I) + h(T2). //

~6. Examples

We shall now calculate the entropy of our examples.

(I) If I: (X,B,m) ~ (X,8,m) is the identity, then h(I) = 0. This

is because h(I,A) = lim ~ H(A) = 0. Also, if T p = I for some p ~ 0

n
then h(T) = 0. This follows since 0 = h(T p) = p'h(T) by Theo-

rem 4.7. Hence any measure-preserving transformation of a finite

space has zero entropy.

(2) Let T: K ~ K be T(z) = az.

Case I: Suppose {an: n ~ Z} is not dense, i.e., a is a root of

 93

unity. Thus ap = 1 f o r some p # 0 ; and TP(z) :aPz : z so

h(T) = 0 by e x a m p l e (I).

Case 2: Suppose {an: n ~ Z} is dense in K. Then {an: n < 0} is

dense in K. Let ~ = {AI,A2} where

A 1 : upper half circle [i,-I)

A 2 = lower half circle [-I,i)

For n > 0 T-n~ consists of s e m i - c i r c l e s beginning at a -n and

-a -n Since {a-n: n > 0} is dense any s e m i - c i r c l e belongs to

T-nA(~). Hence any arc b e l o n g s to V T-nA(~). Thus,

n:0 n:0

B = V T-nA(~) and so, h(T) = 0 by C o r o l l a r y 4.10. //

n:0

(3) Any rotation of a c o m p a c t m e t r i c abelian g r o u p has entropy zero.

Proof: (a) Suppose X : K n, the n - t o r u s , and T ( Z l , . . . , z n) :

(alZl,...,anZn). Then T = TI×T2x...×Tn where Ti: K ~ K is d e f i n e d

by Ti(z) = aiz. By e x a m p l e (2) h(T i) = 0 for all i so by T h e o -

r e m 4.15

n
h(T) : ~ h(T i) : 0.
i=l

(b) General Case. Let T: G ~ G be T(x) : ax. Let G : {71,T2,...}.

Let H n = Ker 71 N ... N Ker 7 n. Hn is a c l o s e d subgroup of G and

(G/~HHn) is the g r o u p generated by {Tl,...,Tn}. Thus

i
(G/H) : finite group × Z n,
n
i
so G/H n = F n × K n

i
where Fn is a f i n i t e g r o u p and K n is a f i n i t e - d i m e n s i o n a l torus.

T induces a map Tn: G/H n ~ G/H n by Tn(gH n) = agH n . Tn is a

 94

rotation on G/Hn, so that it can be w r i t t e n Tn : T n,l × Tn,2 where

i
Tn, 1 is a rotation of Fn and Tn, 2 is a rotation of K n. Thus,

h(T n) : h(Tn, I) + h(Tn, 2) : 0

by example (i) and case (a) of this proof.

Note that Hn ~ {e} so that V A(G/H n) ~ B where A(G/H n)

n
denotes the a-algebra consisting of those elements of B that are

unions of cosets of H n.

Therefore, if B0 = 0 A(G/H n) then by Theorem 4.13

n=l

h(T) : sup h(T,C).

C~B 0
C finite

However, if C ~ B0 is finite then C E A(G/H n) for some n and so,

h(T,C) E h(T n) = 0. Thus, h(T) = 0. //

Corollary:

Any ergodic t r a n s f o r m a t i o n with pure point spectrum has zero

entropy.

This follows from Theorem 3.4. (Actually we have shown the re-

sult only when L2(m) is separable since the above calculation was

for a metric group G.)

(4) Endomorphisms of compact groups:

If A is an e n d o m o r p h i s m of the n-torus Kn onto Kn we shall

show in Chapter 6 that h(A) = ~ loglk) where the summation is over

all eigenvalues of the matrix [A] with absolute value greater than

one.

One can write down a complicated formula for the entropy of an

endomorphism of a general compact metric abelian group. See Juzvinskii

[l].
 95

(5) Affine transformations:

We shall show in Theorem 6.10 that when T = a.A acts on Kn

then h(T) = h(A).

(6) The two-sided (p0,...,Pk_l)-shift has entropy

- ~ Pi" log Pi"

i=0

Let X = ~ {0,1,...,k-l} and T be the shift. Let Ai =

{{Xk}l x 0 = i}, 05i~ k-l. Then ~ = {A 0 .... ,Ak_l} is a partition

of X. Let A = A(K). By definition of B,

TiA = B.

By the Kolmogorov-Sinai Theorem (4.9),

h(T) = lim i H(A vT-IA v ... v T-(n-I)A).

n

A typical element of ~(A v T - I A v ... vT-(n-l)A) is

At0 N T-IAil n ... N T-(n-I)A.

in- I

= {{Xn}: x0 =i0, X l = il .... ,Xn_l = in_l }

which has measure • .. • • . Thus,

Pl0"Pil • Pln_l

H(A vT-IA v... vT-(n-1)A)

= - ~- (Pi0'''''Pi n-i ) "l°g(Pi0'''" "Pi n-i )

k-i
l c ""Pi
io=...=in_l=O Pio n-1 )[log pl0. + ... + log Pin_l]

k-i
= - n [ Pi "l°g Pi"
i=0
 96

k-i
Therefore, h(T) : h(T,A) : - [ p-.log~ Pi' //
i:0

Corollary :

The 2-sided (i/2~i/2)-shift has entropy log 2; the 2-sided

(I/3,1/3,1/3)-shift has entropy log 3. Thus these transformations

cannot be conjugate.

(7) The 1-slded (P0''''~Pk-I)-shift has entropy

k-i
- 5- Pi'l°g Pi"
i:0

The proof is very similar to the one in example (6) but Theorem 4.10

is used instead of Theorem 4.9.

An example of a t r a n s f o r m a t i o n with infinite entropy is the

following:

(8) Let I = (0,i] with Borel sets and Lebesgue measure. Let

X = ~ I with product measure and let T be The shift on X. Then

h(T) = ®.

To see this let An i : {x }: i-i < x0 _< ~ n > 0 1 < i< n .

, j n - -

Then m(An, i) : I/n and ~n : {An,l,... ,An,n} is a partition of X.

Hence h(T,~ n) = log n by the same argument as used in example (8)

(using the independence of ~n'T-l~n'" "''T-k~n )" Therefore,

h(T) >- log n for each n, and so h(T) = ®. //

~7. Ho___ww$ood an invariant is entropy?

Definition 4.5:

An invariant P for an equivalence relation is a complete in-

variant if whenever T and S both have the property P then T

and S are equivalent.

 97

Entropy is, in general, far from complete.

(a) An example of two ergodic measure-preserving transformations with

equal entropy which are not conjugate.

Let T: K ~ K be defined by T(z) = az, a ~ K, where {a n } is

dense in K, and let S: K ~ K be defined by S(z) = bz, b ~ K,

where {b n} is dense in K. T and S are ergodic and h ( T ) = 0=h(S)

by example (2) of the previous section. If we choose a,b so that

{an}i® ~ {bn}~, then T and S are not conjugate (in fact, they

are not even spectrally equivalent) by Theorem 3.1. //

(b) An example (due to Anzai) of two ergodic and speetrally

equivalent measure-preserving transformations with equal entropy which

are not conjugate.

Let T: K 2 ~ K 2 and S: K 2 ~ K 2 be defined by

T(z,w) = (az,zPw) S(z,w) = (az,zqw)

where {an}~ is dense in K and p,q are non-zero integers. Ob-

serve that T and S are affine transformations, T and S are

ergodic, and h(T) = h(S) = 0 by example (5) of §6. By considering

the characters of K one can easily show that L2(m) has a basis of

the form {gn: n ~ O} U {U ~fq : j ~ Z, q ~ O} where gn(T) = a n gn"

Similarly L2(m) has a basis {gn } O {U~hq: j E Z, q > 0}. One then

defines a unitary operator W: L2(m) ~ L2(m) by W(gn) = gn and

W(U~fq) = U~hq and extending. Clearly WU T = UsW showing T and S

are spectrally isomorphic.

However if p # ±q T and S are not conjugate. As mentioned

before conjugacy and isomorphism coincide for m e a s u r e - p r e s e r v i n g

transformations of K equipped with completed Haar measure m. We

shall show T and S are not isomorphic. Suppose ST = S~ and

$(z,w) = (f(z,w),g(z,w)). f and g are only defined almost everywhere

 g8

but this will not affect our argument as we shall c o n s i d e r t h e m as

members of L2(m). We have f(T) = af and g(T) = fqg. Since f is

an e i g e n f u n c t i o n with e i g e n v a l u e a, by remark (3) of ~i of Chapter 3,

f(z,w) = c'z for some c E K. The second e q u a t i o n then becomes

g(T(z,w)) = cqzqg(z,w). If one now expresses g as a Fourier series~

then it is s t r a i g h t f o r w a r d to show that g(z,w) = kznw m where k E K,

pm = q for some m E Z and a n : c q. So ¢(z,w) = (cz,kznw TM) is an

affine t r a n s f o r m a t i o n and for ¢ to be an i n v e r t i b l e measure-preserving

t r a n s f o r m a t i o n one needs m : ±I, i.e., p : ±q. //

H o w e v e r we can c o n s i d e r the p r o b l e m of c o m p l e t e n e s s of entropy

for certain c o l l e c t i o n s of m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n s , and

this we do in the next section.

§8. Bernoulli and K o l m o g o r o v A u t o m o r p h i s m s

(As general r e f e r e n c e s for this section see Shields [2] and Friedman
and Ornstein [2].)

D e f i n i t i o n 4.6:

Let (Y,F,~) be a p r o b a b i l i t y space. Let

(X,B,m) : IT (Y,F,~)

and let T: X ~ X be the shift

T({Yn}) = {Xn} where Xn = Yn+l n E Z.

T is an i n v e r t i b l e m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n and is called

the Bernoulli a u t o m o r p h i s m w i t h state space (Y,F,~).

Examples of Bernoulli automorphisms:

(i) the 2-sided (p0,...,Pk_l)-shift. Here Y = {0,1,...,k-l}.

(2) the example (8) of ~6. Here y = (0,i].

(3) If T is a Bernoulli a u t o m o r p h i s m so is T 2.
 99

(4) If T1 and T2 are Bernoulli a u t o m o r p h i s m s so is TlXT 2.

Remark:

If T is a B e r n o u l l i a u t o m o r p h i s m then h(T) < - iff 3 a

countable p a r t i t i o n ~ on (Y,F,~) ~ H(~) < - and A(~) = F. In

this case h(T) = H(F).

We shall call a p r o b a b i l i t y space a Lebesgue space if the iden-

tity map of it is i s o m o r p h i c (as a m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n )

to the identity map on a p r o b a b i l i t y space consisting of a subinterval

of [0,1] with L e b e s g u e m e a s u r a b l e sets and Lebesgue measure t o g e t h e r

with some atoms.

R e c e n t l y e n t r o p y has been shown to be a complete invariant for

the class of B e r n o u l l i automorphisms:

T h e o r e m 4.18: (Ornstein [i] and [2])

Two B e r n o u l l i a u t o m o r p h i s m s (whose state spaces are Lebesgue

spaces) with the same entropy are conjugate; i.e., entropy is a com-

plete invariant for the o o n j u g a c y of Bernoulli automorphisms.

(Isomorphism follows from c o n j u g a c y by the a s s u m p t i o n on the

state spaces.)

Certain special cases had been w o r k e d out earlier by M e s h a l k i n

[i] and by Blum and Hanson [i]. This result reduces the conjugacy

p r o b l e m for B e r n o u l l i a u t o m o r p h i s m s to their state spaces, since the

entropy depends only on the state space. It is possible, for example,

for a Bernoulli a u t o m o r p h i s m with a state space of two points to be

conjugate to a B e r n o u l l i a u t o m o r p h i s m with a countably infinite state

space.

Note:

Given any x > 0 one can find n > 0 and {pl,P2,...,pn} ,

n n
Pi ~ 0 i ~ Pi = ! such that - ~ Pi "!°g Pi = x. Hence V x > 0
=0 i=0
 i00

3 a Bernoulli a u t o m o r p h i s m w i t h entropy x.

C o r o l l a r y 4.15:

(i) Every Bernoulli a u t o m o r p h i s m has an n-th root.

(ii) Every Bernoulli a u t o m o r p h i s m can be written as a direct product

of two B e r n o u l l i automorphisms.

Proof: (i) Let T be a Bernoulli a u t o m o r p h i s m and n > 0. Let

1
S be a Bernoulli a u t o m o r p h i s m with h(S) = ~ h(T). Then Sn is a

Bernoulli a u t o m o r p h i s m with entropy h(T)~ and therefore Sn and T

are conjugate.

(ii) Let T be a B e r n o u l l i automorphism. Let S be B e r n o u l l i w i t h

h(S) = %.h(T). Then h(S×S) = h(T) and since S×S is Bernoulli,

S×S is conjugate to T. //

Ornstein has proved m a n y deep results about Bernoulli automor-

phisms~ some of which we now summarize:

T h e o r e m 4.17: (Ornstein [3] etc.)

(i) Every root of a B e r n o u l l i a u t o m o r p h i s m is a Bernoulli automor-

phism. (S is an n-th root if S n = T.)

(ii) Let T be a B e r n o u l l i automorphism. If F is a s u b - ~ - a l g e b r a

of B with TF = F then T r e s t r i c t e d to the measure space

(X,F,mlF) is a Bernoulli a u t o m o r p h i s m (i.e., a factor of a B e r n o u l l i

a u t o m o r p h i s m is a Bernoulli automorphism).

(iii) If Fn 2 B is an i n c r e a s i n g sequence of ~-algebras with

TFn : Fn and T r e s t r i c t e d to (X,Fn,mIF) is a Bernoulli automor-

n
phism V n > 0, then T is a Bernoulli a u t o m o r p h i s m (i.e., an in-

verse limit of Bernoulli a u t o m o r p h i s m s is a Bernoulli automorphism).

The following class of t r a n s f o r m a t i o n s were introduced by

K o l m o g o r o v in 1958 by a n a l o g y with r e g u l a r stochastic processes.

 I01

Definition 4.7:

An invertible measure-preserving transformation T of a proba-

biliZy space (X,B,m) is a Kolmogorov automorphism (K-automorphism)

if 3 a sub-a-algebra K of B such that:

(i) K ~ TK.

(ii) V TnK ~ B.
n:0

(iii) 6 T-nf o N = {X,¢}

n=0

(If A,C are a-algebras A ~ C will mean V A E A 3 C ~ C with

m(AAC) = 0. If BI,B 2 E B then B1 ~ B 2 means m ( B l \ B 2) = 0 and

B1 ~ B 2 means m(BIAB 2) = 0.)

Theorem 4.18:

Every Bernoulli automorphism is a Kolmogorov automorphism.

Proof: Let the s t a t e space for T be (Y~F,#). If F E F, let

}= < { x n} (X: x0 ~ F ~ ( B. Let G = {F: F (F}, which is called the
0
time-0 a-alsebra. Let K = V TiG- We now verify that K satis-

fies the conditions for a Kolmogorov automorphism.

0 • 1 •
(i) K = V TIG c V TIG = TK.
i=-® i=-®

n
(ii) ¢ TnK = V V TIG = ¢ TiG = B by definition of B.
n=0 n=0 i= . . . .

(iii) We have to show 6 T-nK ~ N = {X,¢}. Fix A E 6 T -nK =

0 0
~ -n
V T iG. Let B ( ¢ TkG, j ~ Z. Since A ~ V TiG, A and
n:0 -- k=j i<j

B are independent, and therefore m ( A n B) = m(A)m(B). The collection

of all sets B for which m ( A N B) = m(A)m(B) is a monotone class,

 102

and, by the above, contains 0 V TkG. Therefore V B ~ B,

j:-- k:j
m ( A N B) = m(A)m(B). Put B = A, then m(A) : m(A) 2 w h i c h implies

m(A) : 0 or i. Hence

T-nK o N /!
n=O

It was an open p r o b l e m from 1958 to 1969 as to w h e t h e r the con-

verse of T h e o r e m 4.18 was true, i.e., w h e t h e r a K o l m o g o r o v automor-

phism is c o n j u g a t e to a Bernoulli automorphism. This was shown to be

false by Ornstein.

T h e o r e m 4.19: (0rnstein [6])

There is an example of a K o l m o g o r o v a u t o m o r p h i s m T with the

following properties:

(i) T is not c o n j u g a t e to a Bernoulli automorphism.

(ii) T does not have a square root.

(iii) T is not c o n j u g a t e to T -I.

Corollary 4.19:

E n t r o p y is not a complete invariant for the class of K o l m o g o r o v

automorphisms.

Proof: Let T be the example of Ornstein. By T h e o r e m 4.9

h(T) ~ 0. Choose a Bernoulli a u t o m o r p h i s m S with h(S) = h(T).

S and T are not isomorphic by (i).

Remarks:

(i) Property (iii) of T contrasts with the b e h a v i o r of ergodic

t r a n s f o r m a t i o n s w i t h pure point spectrum. (See C o r o l l a r y 3.3.)

(2) Ornstein's example is defined by induction and so is fairly com-

p l i c a t e d to describe. It is t h e r e f o r e important to check w h e t h e r the

more "natural" e x a m p l e s of Kolmogorov a u t o m o r p h i s m s are B e r n o u l l i

 103

a u t o m o r p h i s m s or not. We c o n s i d e r some of these at the end of this

section.

(3) Sinai has proved that if T is an ergodic invertible m e a s u r e -

p r e s e r v i n g t r a n s f o r m a t i o n of (X,B,m), which we assume is i s o m o r p h i c

to the unit interval w i t h Lebesgue measure, and h(T) > 0 and if S

is a Bernoulli a u t o m o r p h i s m with h(S) ~ h(T) then there exists a

measure-preserving transformation ¢ such that CT = S¢, i.e., S

is a factor of T.

The next t h e o r e m shows that K o l m o g o r o v a u t o m o r p h i s m s and Ber-

noulli a u t o m o r p h i s m s are s p e c t r a l l y the same:

T h e o r e m 4.20: (Rohlin)

A Kolmogorov automorphism T of a probability space (X,B,m),

where L2(m) is separable, has c o u n t a b l e Lebesgue spectrum.

Proof: Assume B ~ {X,¢} = N or else the result is trivial.

We have (i) K ~ TK (ii) V TnK =

o B (iii) ~T-nK =
o N. We split

the proof into three parts:

a) We first show that K has no atoms, i.e., if C E K and m(C) > 0

then 3 D E K with D c C and m(D) < m(C).

Suppose C is an atom of K with m(C) > 0. Then TC is an

atom of TK and since K ~ TK either TC ~ C or m(C O TC) = 0. If

TC ~ C then TC ~ C since both sets have the same measure so that

C ~ ~T-nK and t h e r e f o r e m(C) = i. Hence B ~ {X,¢}, a contradic-

tion. On the other hand, suppose m ( T C O C) = 0. Then either for some

k • 0 Tkc ~ C (and we use the above proof to get a contradiction)

or m(Tkc N C) = 0 V k • 0 and then C U TC U T2C U ... has in-

finite measure, a contradiction.

b) Let H = {f E L2(m): f is K-measurable}. Then UTH c H. Let

m
H : V • UTH. From UTnH : ~ U_V U~+IH (n,m • O) we conclude
-n
 104

that L2(m) = ~)U~V @ C where C is the subspace of constants. It

- m

suffices to show V is infinite-dimensional since if {fl,f2,f3,...}

is a basis for V, then nf m". n ~Z

{f0 ~ i, U T m>0} is a basis for
L2(m).

c) We now show V is infinite-dimensional. Since TK ~ K (we are

assuming B ~ N) we know V ~ {0}. Let g E V, g ~ 0 and then

G = {x: g(x) ~ 0} satisfies m(G) • 0. Since G E K and using a)

we know XGH = {XGf: f E H} is infinite-dimensional. ×GH = V' @ ×GUT H

where V' c V so either V' is infinite-dimensional (and hence V

is) or ×GUTH is infinite-dimensional. In this case there is a

linearly independent sequence of functions {×GUTfn} where the fn

are bounded functions in H. Then {gUTf n} are linearly independent

in H. It suffices to show these functions are in V. But if f ~ H

then

(gUTfn,UTf) = (g,UT(ffn)) = 0

so gUTf n ~ V. //

Corollary 4.20:

A K-automorphism is strong mixing.

Proof: By Theorem 2.3. //

Kolmogorov automorphisms are connected to entropy by the follow-

ing result (half of which was proved by Pinsker).

Theorem 4.21: (Rohlin g Sinai [i])

Let (X,B,m) be isomorphic to [0,i] with Lebesgue measure.

Let T: X ~ X be invertible and measure-preserving. Then T is a

K-automorphism iff h(T,A) > 0 V finite A ~ N; i.e., T has

completely positive entropy.

 105

Remark:

This shows that K-automorphisms are "the opposites" of transfor-

mations with zero entropy (since h(T,A) = 0 V A in the zero

entropy case).

Examples:

(1) Group Autemorphisms. Rohlin proved that any ergodio a u t o m o r p h i s m

of a compact abelian metric group is a K-automorphism and later

Yusinskii proved the theorem in the non-abelian case. Katznelson Ill

has shown that ergodic automorphisms of finite-dimensional tort are

conjugate to Bernoulli automorphisms. Chu [1] and Lind [1] have in-

dependently extended Katznelson's results to the (countably) infinite-

dimensional torus. Katznelson and Weiss [2] have also solved the

case where the dual group is the discrete rationals but w h e t h e r an

ergodic a u t o m o r p h i s m of a general compact abelian metric group is

Bernoulli is not yet known.

(2) Markov Chains. Consider a two-sided Markov chain with transition

matrix [Pij]. The shift T on the space of sequences of integers

becomes a measure-preserving transformation for the Markov measure de-

fined by [Pij] and an initial vector [pl,...,pk ] satisfying

[Pl .... 'Pk][Pij ] : [P! .... 'Pk ]" (See Billingsley [i].) It is known

that T is ergodic iff the chain is irreducible (i.e., V pairs of

(n)
states i,j 3 n > 0 with Pij > O) and T is strong mixing iff

the chain is irreducible and aperiodic (i.e., V states i

(n)
g.c.d. {n: Pit > O} : I). Friedman and Ornstein [i] have shown

that if T is strong mixing then it is conjugate to a Bernoulli auto-

morphism. Therefore, from the point of view of ergodic theory mixing

Markov chains are the same as Bernoulli automorphisms, i.e., we can

represent the space as a direct product measure space so that T

becomes the shift on the new space.

 106

(3) One can generalize the notion of a f i n i t e - d i m e n s i o n a l torus to

obtain another kind of h o m o g e n e o u s space:--namely a nilmanifold. Let

N be a c o n n e c t e d , simple connected, ni!potent Lie group and D a

discrete subgroup of N so that the quotient space N/D is compact.

N/D is c a l l e d a nilmanifold. When N = Rn and D = Zn we get an

n-torus. The Haar measure on N determines a normalized Borel

measure on N/D. If A: N ~ N is a ( c o n t i n u o u s ) automorphism with

AD = D then this induces a map A: N / D ~ N/D, which we c a l l an

automorphism of N/D. A always preserves the m e a s u r e m. Parry has

investigated the ergodic theory of such m a p s and has shown that if A

is e r g o d i c then A is a K - a u t o m o r p h i s m . A subclass of the automor-

phisms of N/D, the A n o s o v ones, are k n o w n to be conjugate to Ber-

noulli automorphisms, but it has not yet been proved that the others

are.

The simplest examples are as follows: Let

N = 1 : x,y,z E R .
0

N satisfies the above conditions with the operation of m a t r i x multi-

plication and the natural topology from R 3. Let

D = 1 n : m,n,p 6 Z .
0 !

Then N/D is a n i l m a n i f o l d . The automorphism

I(
i x z i 2x+y z+x +xy+
0 1 y -~ 0 1 x+y
0 0 1 0 0 1

of N induces an e r g o d i c automorphism of N/D, and it is u n k n o w n if

this is B e r n o u l l i .
 107

Pinsker A l s e b r a

Let T be a m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n of a p r o b a b i l i t y

space (X,B,m) w h i c h is i s o m o r p h i c to [0,i] with Lebesgue measure.

Let

P(T) = V {A: A c B, A finite, h(T,A) : 0}.

This is called the P i n s k e r o-alGebra.

One can show that T-Ip(T) = P(T). One can also prove that if A

is finite then A c P(T) iff h(T,A) = 0. Thus, P(T) is the maxi-

mum ~ - a l g e b r a such that T r e s t r i c t e d to (X,P(T),mIp(T)) has zero

entropy. Note that P(T) = B iff h(T) = 0 and P(T) = N iff T

is a K - a u t o m o r p h i s m (by T h e o r e m 4.21).

Theorem 4.22: (Rohlin)

If T is an i n v e r t i b l e m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n with

h(T) > 0 then UT has countable Lebesgue spectrum in

L2(B) ~ L2(p(T)).

This reduces the study of the spectrum of invertible m e a s u r e -

preserving t r a n s f o r m a t i o n s to those with zero entropy.

Corollary 4.22:

T r a n s f o r m a t i o n s with pure point spectrum have zero entropy.

Proof: L2(m) cannot have a subspace on which UT has countable

Lebesgue spectrum.

The types of s p e c t r u m that can occur for zero entropy transfor-

mations are unknown. There are examples of zero entropy t r a n s f o r m a -

tions with countable Lebesgue s p e c t r u m (from G a u s s i a n processes and

horocycle flows).

In the space of i n v e r t i b l e m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n s of

(X~B,m) with the weak topology, the set of t r a n s f o r m a t i o n s of zero

 108

entropy forms a dense G6 .

P i n s k e r [i] c o n j e c t u r e d that any ergodic m e a s u r e - p r e s e r v i n g

t r a n s f o r m a t i o n could be w r i t t e n as a direct p r o d u c t of one with zero

entropy and a K - a u t o m o r p h i s m . However, (ii) of T h e o r e m 4.19 allows

us to obtain a c o u n t e r e x a m p l e for if T: X ~ X is the example of

0rnstein w i t h no square root then the t r a n s f o r m a t i o n S of the space

{0} × X U {i} × X given by S(0,x) = (l,x). S(l,x) = (0,Tx) provides

a counterexample to the Pinsker conjecture. This example is not

mixing (since S2 is not ergodic), but 0 r n s t e i n has c o n s t r u c t e d a

mixing t r a n s f o r m a t i o n w h i c h violates the Pinsker conjecture.

Sequence E n t r o p y

(See: Kushnirenko [i].)

Let (X,B,m) be a p r o b a b i l i t y space isomorphic to [0,i] with

Lebesgue measure. Let T: X ~ X be an i n v e r t i b l e m e a s u r e - p r e s e r v i n g

transformation.

Let F : {tl,t2,...} be a sequence of integers. Let A be a

finite algebra A ¢ B.

Define hF(T,A) = lim sup ~1 H(TtlA v ... v T t nA)

n ~

and define hF(T) = sup hF(T,A).

A finite

It is easily shown that hF(T) is a c o n j u g a c y invariant for each F.

Entropy and spectral p r o p e r t i e s are c o n n e c t e d by the following:

T h e o r e m 4.23: (Kushnirenko [i])

T has d i s c r e t e s p e c t r u m iff hr(T) = 0 V F.

One can also show that suPFhr(T) : ® or log k, for some

k > 0, k ~ Z; moreover, those T with suPFhF(T) : log k are easy

to describe.
 10g

Problem:

If T has q u a s i - d i s c r e t e s p e c t r u m (see Hahn and Parry [i]), what

sequences give hp(T) > 07

hp(T) has been c a l c u l a t e d except in the cases when T has zero

entropy and P has "large gaps" hp(T) will only give new informa-

tion when h(T) = 0. (See: Newton [i].)

~ii. Comments

E n t r o p y was i n t r o d u c e d as a c o n j u g a c y invariant for m e a s u r e -

p r e s e r v i n g transformations. It was soon r e a l i z e d that entropy was

more than just an a s s i g n m e n t of a n u m b e r to each transformation.

Kclmogorov automorphisms and t r a n s f o r m a t i o n s with zero entropy

have r e c e i v e d the most treatment. They are "opposites" from the point

of view of entropy. K o l m o g o r o v a u t o m o r p h i s m s are important for appli-

cations as it seems that the most i n t e r e s t i n g smooth systems are

K o l m o g o r o v and even Bernoulli.

By T h e o r e m 4.22 we know that the spectral theory of invertible

measure-preserving transformations reduces to that for the zero en-

tropy case. The f o l l o w i n g is still an open problem:

If h(T) = 0 what kind of s p e c t r u m can UT have?

For t r a n s f o r m a t i o n s with zero entropy the i s o m o r p h i s m problem is

only solved for ergodic t r a n s f o r m a t i o n s with discrete spectrum, to-

tally ergodic t r a n s f o r m a t i o n s with q u a s i - d i s c r e t e s p e c t r u m and some

other special cases. Sequence entropy may play a role in the isomor-

phism p r o b l e m for zero entropy transformations.

We note again that in the weak t o p o l o g y on the set of all in-

vertible m e a s u r e - p r e s e r v i n g t r a n s f o r m a t i o n s on a given space (X,B~m),

the set of t r a n s f o r m a t i o n s with zero entropy is a dense G5 (count-

able i n t e r s e c t i o n of open sets) and the set of w e a k mixing transfor-

mations is also a dense G 0, Since the set of strong mixing

 Ii0

transformations is a set of first category it follows that "most"

transformations are weak mixing, have zero entropy, but are not strong

mixing.

The main p r o b l e m to consider for Kolmogorov automorphisms seems

to be to find more examples of Kolmogorov automorphisms that are not

conjugate Bernoulli automorphisms. One should first check this fact

for all the usual ways of constructing new transformations from old

ones (e.g., is a weak mixing group extension of a Bernoulli automor-

phism a Bernoulli automorphism?). Then one might hope to find a new

invariant that may be complete for Kolmogorov automorphisms.

912. Non-invertible Transformations

Suppose T: (X,B,m) ~ (X,B,m) is m e a s u r e - p r e s e r v i n g but not

necessarily invertible; assume that (X,B~m) is isomorphic to [0,I]

with Lebesgue measure. Note that

B ~ T-IB ~ T-2B ~ ...

Let B : A T-nB; so, T-1B : S , and TI(X,B ,m) is invertible.

n:0
One can show that UT has countable Lebesgue spectrum in

L2(B) ~ L2(B ) where countable Lebesgue spectrum in this situation

means there is a basis of the form

{Unfm: n > 0 and m > 0 } .

This reduces the study of spectral properties of measure-

preserving transformations to those which are invertible (in fact,

invertlble ones with zero entropy by Theorem 4.22).

One can also show that P(T) ~ S (i.e., if h(T) = 0 then T

is invertible modulo sets of measure zero; more precisely T -I is a

measure algebra isomorphism).

 Ill

The analogous c o n c e p t to a K - a u t o m o r p h i s m is an exact endomor-

phism.

D e f i n i t i o n 4.8:

T: X ~ X is an exact e n d o m o r p h i s m if

A T-nB ° N ; i.e , B ° N
n:0

So exact e n d o m o r p h i s m s are as far from being invertible as

possible. Examples of exact endomorphisms are the o n e - s i d e d Bernoulli

shifts. Exact e n d o m o r p h i s m s are strong mixing (by the above remarks

about s p e c t r u m and a p r o o f like that of Theorem 2.3).

It was c o n j e c t u r e d that every ergodic m e a s u r e - p r e s e r v i n g trans-

formation is a product of an exact e n d o m o r p h i s m and an invertible

measure-preserving transformation. This is not so (Parry and Welters).

Also, o n e - s i d e d Bernoulli shifts with the same entropy are not

n e c e s s a r i l y conjugate since an m-to-i map cannot be c o n j u g a t e to an

n-to-I map if m ~ n. So entropy is far from complete for exact en-

domorphisms. Parry and Walters (1971) c o n s t r u c t e d two exact endomor-

phisms S,T with s-nB = T-nB V n > 0, S2 = T 2 (= h(S) = h(T))

but with S and T not conjugate. The m e t h o d used involved the

Jacobian of an e n d o m o r p h i s m , a concept which was introduced in Parry

[3]. (It is not known if there are two K - a u t o m o r p h i s m s S,T with

S2 = T 2 but with S and T not conjugate.) Also, exact endomor-

phisms need not be c o n j u g a t e to one-sided Bernoulli shifts; in fact a

one-sided M a r k o v chain which is exact need not be conjugate to a one-

sided Bernoulli shift.

 C h a p t e r 5: Topological Dynamics

§0. Introduction

In m e a s u r e t h e o r e t i c ergodic theory one studies the asymptotic

properties of m e a s u r e - p r e s e r v i n g transformations. In t o p o l o g i c a l

dynamics one studies the asymptotic properties of continuous maps.

Theorem 5.0:

Let X be a compact H a u s d o r f f space. The f o l l o w i n g are equiva-

lent:

(i) X is metrizable.

(2) X has a countable base.

(3) C(X) (the space of all c o m p l e x - v a l u e d continuous functions on X)

has a countable dense subset.

Proof: See Kelley [i].

We shall c o n s i d e r compact metric spaces X and h o m e o m o r p h i s m s

T: X ~ X. C(X) is a Banach algebra with

F1fll : sup If(x)J.

x6X

The map UT: C(X) ~ C(X)~ defined by (UTf)(x) : f(Tx) is clearly a

multiplicative linear isometry of C(X) onto C(X)~ i.e.~ UT is a

Banach algebra automorphism.

Remarks:

Compactness is a "finiteness" condition and corresponds to the

assumption of a finite measure in the measure t h e c r e t i c work. The

assumption of m e t r i z a b i l i t y is not n e e d e d for m a n y of the results but

it often shortens proofs and most applications are for metric spaces.
 I13

We assume that T is a homeomorphism, rather than a continuous map,

for simplicity.

Typical examples that we shall study are:

Examples:

(i) I on any X.

(ii) Tx = ax where X is a compact metric group. (On such a group

there exists a left invariant metric d, i.e.,

d(bx,by) = d(x,y) V b,x,y 6 X ).

(iii) an a u t o m o r p h i s m of a compact metric group.

(iv) an affine t r a n s f o r m a t i o n Tx = a.A(x) where A is an auto-

m o r p h i s m of a compact group G and a 6 G.

(v) Let Y = {0,1,...,k-l} with the discrete topology. Let

X = ~ Y with the product topology. A metric on X is given by:

Ixn-ynJ
d({Xn},{Yn}) = n =-®
i 2 jnj The shift T ( T{x n} = {yn } with

Yn = Xn+l ) is a h o m e o m o r p h i s m of X. Note That here we have a

special case of (iii) since X is a compact group under the operation

{x n} + {yn } = {(Xn+Yn)mod(k)} ,

and T is an a u t o m o r p h i s m of X. d is an invariant metric on X.

§i. Minimality

X will denote a compact metric space a n d T: X ~ X a homeomor-

phism. We would like to find a concept of "irreducible piece" to play

the role e r g o d i c i t y played for m e a s u r e - p r e s e r v i n g transformations.

Definition 5.1:

T is m i n i m a l if V x £ X {Tnx: n E Z} is dense in X.

OT(X) = {Tnx: n E Z} is called the T-orbiT of x.

 114

Theorem 5.1:

T is m i n i m a l iff TE = E and E closed = E = ¢ or X.

Proof: Suppose T is m i n i m a l and suppose E is closed, E @

and TE = E. Choose x E E. Then OT(X) c E by the T - i n v a r i a n c e

of E, and X = OT(X) so E = X i.e., E = X. Conversely, V

x E X, 0T(X) is a c l o s e d T-invariant non-empty set, and hence is

all of X. //

Definition 5.2:

A closed subset E of X which is T - i n v a r i a n t is called a

minimal set w i t h r e s p e c t to T: X ~ X if TIE is minimal.

Theorem 5.2:

Any h o m e o m o r p h i s m T: X ~ X has a minimal set.

Proof: Let E denote the c o l l e c t i o n of all closed n o n - e m p t y

T-invariant subsets of X. Clearly E ~ ¢ since X belongs to E.

E is a p a r t i a l l y ordered set under inclusion. Every linearly ordered

subset of E has a least element (the intersection of the e l e m e n t s of

the chain. The least element is n o n - e m p t y by Cantor's intersection

property.) Thus, by Zorn's Lemma, E has a minimum element. This

element is a m i n i m a l set for T. //

Remark:

Ergodicity has the p r o p e r ti e s :

(i) An ergodic transformation is indecomposable.

(ii) Every m e a s u r e - p r e s e r v i n g transformation on a decent space can be

decomposed into ergodic pieces in a nice way.

By its d e f i n i t i o n , a minimal transformation is indecomposable.

We know that each h o m e o m o r p h i s m T: X ~ X has a minimal set. Pbwever,

in general, one cannot partition X into T - i n v a r i a n t closed sets E

such that X = U~ E a, TE a = E a V a and TIE a is m i n i m a l (although

 115

we can in some important cases). If T has such a d e c o m p o s i t i o n it

is sometimes called semi-simple. An example of a t r a n s f o r m a t i o n not

a d m i t t i n g such a d e c o m p o s i t i o n is an ergodic a u t o m o r p h i s m of a compact

metric group. This will be clear from the next section.

Definition 5.3:

A point x E X is a periodic point of T if Tnx = x for some

n ~ 0. The least such positive n with this p r o p e r t y is called the

period of x under T.

T h e o r e m 5.3:

Let T: X ~ X be a minimal homeomorphism. Then:

(i) T has no n o n c o n s t a n t invariant continuous functions.

(2) If X is not finite T has no p e r i o d i c points.

Proof: (i) fT(x) = f(x) implies fTn(x) = f(x) V n E Z, and

so f is c o n s t a n t on the dense subset OT(X) of X. Thus f is

constant on X.

(2) If Tnx = x~ n ~ 0 then {x,Tx~...,Tn-lx} is a closed

T-invariant set and by the m i n i m a l i t y c o n d i t i o n it is the whole

space X. //

Remarks:

(i) If T has no n o n c o n s t a n t T - i n v a r i a n t functions then T need

not be minimal. As an example of this~ let A be an ergodic automor-

phism of a compact metric group G # {e}. Then A is not minimal

since A(e) = e. But A satisfies p r o p e r t y (i)~ since if fA(x) =

f(x)~ f continuous~ then by ergodicity~ f = c o n s t a n t a.e. and,

since H a a r m e a s u r e is positive on open sets f is constant every-

where.

(ii) The fact that a minimal h o m e o m o r p h i s m of a n o n - f i n i t e space has

no p e m i o d i c points rules out many important examp!es~ such as ergodic

automorphisms of compact metric groups.

 116

We n o w check whether the examples mentioned in §0 are m i n i m a l or

not.

Examples:

(i) I is m i n i m a l iff X = a single point.

(ii) Let G be a c o m p a c t metric group and T(x) = ax. T is m i n i -

mal iff {an: n E Z} is d e n s e in X.

Proof: (=) OT(e) = {an: n ~ Z}.

(=) Let x E X. We m u s t show that OT(X) = X. Let y E X.

ni ~ yx -I i.e.,
3 ni 9 a

n.
l
a -x ~ y
II

n.
T l(x) ~ y.

Therefore OT(X) is d e n s e in X. //

(iii) An automorphism of a c o m p a c t metric group G is m i n i m a l iff

G = {e}. The proof is t r i v i a l .

(iv) For a f f i n e transformations of c o m p a c t metric groups necessary

and sufficient conditions for m i n i m a l i t y are known. For e x a m p l e , if

G is also abelian and connected then T = a.A is m i n i m a l iff

BnG : {e} and [a,BG] : G

n:O

where B is the endomorphism of G defined by B(x) : x-l'A(x) and

[a,BG] denotes the smallest closed subgroup of G containing a and

BG. This was proved by H o a r e and Parry [i].

(v) The shift on k symbols is m i n i m a l iff k = 0. This is seen

from (iii) above.

 117

~2. Topolo$ical Transitivity

Definition 5.4:

T: X ~ X is t o p o l o g i c a l l y transitive if 3 x0 ~ X 9 OT(X 0)

is dense in X.

Note:

T minimal = T t o p o l o g i c a l l y transitive.

T h e o r e m 5.4:

The f o l l o w i n g are equivalent:

(i) T is t o p o l o g i c a l l y transitive.

(2) TE = E, E closed, E ~ X = E is nowhere dense (or, e q u i v a l e n t -

ly, if TU = U, U open, U # ¢ then U is dense).

(9) If U,V are n o n e m p t y open sets then 3 n ~ Z

Tn(u) n V ~ 4.

(4) {x E X: 0T(X) ~ X} is a set of first category.

Proof: (i) = (2). Suppose 0T(X 0) : X and let E ~ 4, E

closed, TE = E, E # X. Suppose U ~ E is open, U ~ 4. Then

3 p 9 TP(x 0) E U & E so that OT(X 0) ~ E and X = E, a contra-

diction. Therefore E has no interior.

(2) = (3). Suppose U,V ~ ¢ are open sets. Then 0 Tnu is

an open T - i n v a r i a n t set; so, it is n e c e s s a r i l y dense by condition (2).

Thus 0 Tnu N V ~ ¢.

(3) = (4). Let UI,U2,...,Un,... be a c o u n t a b l e base for X.

Then OT(X) ~ X

3 n ~ 0T(X) N Un : 4

3 n ) Tm(x) £ X\U n V m ~ Z
 118

3 n with x ~ 6 Tm(XkUn )

n:l m:-®

It suffices to show 6 Tm(XkUn ) is a nowhere dense set V n. Its

complement is 0 Tm(Un) which is c l e a r l y dense by c o n d i t i o n (3).

Hence the result follows.

(4) = (1). This is clear since a compact metric space is of

second category. //

The following theorem gives many examples of t o p o l o g i c a l l y tran-

sitive homeomorphisms.

Theorem 5.5:

Let X be a compact metric space and T: X ~ X a homeomorphism,

m a Borel probability measure on X giving positive measure to every

non-empty open set. If T is ergodic with respect to m, then

m{x E X: OT(X) = X} = i. In particular, T is t o p o l o g i c a l l y transi-

tive.

Proof: Let UI,U2,... be a countable base for the topology.

By the previous proof

n:l k=-- n

The closed set "%'[

] Tk(X\Un ) is T-invariant, so by ergodicity has

measure 0 or i. But Un is contained in complement of this set

and m(U n) > 0, since Un is open. Therefore

m( 6 Tk(X\Un)) : 0

and so m{x: 0T(X) ~ X} = 0. Hence m{x: 0T(X) = X} = I. //

 119

Corollary 5.5:

Let G be a compact metric group and T: G ~ G an affine trans-

formation. T is ergodic with respect to Haar measure ~ T is topo-

iogicaily transitive.

Proof: (=) This is obvious as Haar measure is positive on

non-empty open sets.

(=) This proof is like the last part of the proof in example (5)

of ~4 Chapter I, which deals with the case when G is connected and

abelian. //

Theorem 5.6:

If T is topologically transitive then T has no nonconstant

continuous invariant functions.

Proof: Suppose f E C(X), fT(x) = f(x). If OT(X 0) = X then f

is constant on 0T(X0), a dense set~ and hence is constant on X. //

Remarks:

(i) If all the T-invariant continuous functions are constant then T

need not be topologically transitive. The following is an example to

illustrate this:

Let X = K2x{0} U K 2 x { l ~ ( e , 0 )
(e~l)

i.e., two copies of the two-torus joined at the identity. Let

A: K 2 ~ K 2 be an ergodic a u t o m o r p h i s m and define T: X ~ X by

T(x,0) = (Ax,0), T(x,l) = (Ax,l).

Then T is not topologically transitive since T preserves K2x{0}

and K2x{l}. However, each continuous T-invariant function is con-

stant since it must be constant on both K2x{0} and K2x{l}, be-

cause A is ergodic, and these two constants must be the same because

they must agree at the point (e,0) e (e,l). //

 120

(2) T Can be t o p o l o g i c a l l y t r a n s i t i v e and have a dense set of peri-

odic points. To illustrate this we prove that any a u t o m o r p h i s m A

of K2 has a dense set of p e r i o d i c points.

Fix n > 0. C o n s i d e r the finite subgroup of K2 c o n s i s t i n g of

points of the form (Wl,W 2) where w E = w n2 = i. These are all the

elements of K2 of group order n. Since A is an a u t o m o r p h i s m it

preserves this finite s u b g r o u p and hence, each m e m b e r of this subgroup

is a p e r i o d i c point for A. If we now vary n we obtain a dense set.

This proof can o b v i o u s l y be e x t e n d e d to an a u t o m o r p h i s m of K n,

n > i. //

(3) Topologically transitive homeomorphisms enjoy some of the proper-

ties of m i n i m a l h o m e o m o r p h i s m s and also allow other i n t e r e s t i n g things

to occur; e.g., a dense set of p e r i o d i c points. (2) and (3) of Theo-

rem 5.4 show that t o p o l o g i c a l t r a n s i t i v i t y is (in some sense) a topo-

logical a n a l o g u e of ergodioity. Also, t o p o l o g i c a l l y transitive

homeomorphisms are " i n d e c o m p o s a b l e " ; i.e., we cannot write

X : ~J E , TE a = Ea and Ea closed

when T is t o p o l o g i c a l l y transitive. So it seems that t o p o l o g i c a l l y

transitive h o m e o m o r p h i s m s are better building blocks than m i n i m a l

homeomorphisms. If T has a d e c o m p o s i t i o n into minimal pieces then

each piece is also t o p o l o g i c a l l y transitive. So, the best thing to

do is to try to get a d e c o m p o s i t i o n of T into t o p o l o g i c a l l y transi-

tive pieces, and then see if these pieces are also minimal.

A distal h o m e o m o r p h i s m (i.e., x ~ y = H 5 > 0 )

d(Tn(x),Tn(y)) > 5 V n E Z) can be d e c o m p o s e d into minimal pieces

(Ellis [i]). An A x i o m A* h o m e o m o r p h i s m can be d e c o m p o s e d into topo-

logically transitive pieces (Smale [i]). But, not all h o m e o m o r p h i s m s

can be d e c o m p o s e d into t o p o l o g i c a l l y transitive pieces; e.g., see the

example in r e m a r k (i) above.

 121

The f o l l o w i n g gives a sufficient but not n e c e s s a r y c o n d i t i o n for

a t o p o l o g i c a l l y t r a n s i t i v e h o m e o m o r p h i s m to be minimal.

Theorem 5.7:

If X is a compact m e t r i z a b l e space, T: X ~ X a topologically

transitive homeomorphism, and if there exists a metric on X making

T an isometry, then T is minimal.

Proof: Suppose such a metric is d, i.e., d(Tx,Ty) = d(x,y).

Let OT(X 0) = X and c o n s i d e r x E X. We want to show that OT(X) = X.

Let y E X and let • > 0. There exist n,m E Z such that

d(x,Tm(x0)) < Z , d(y,Tn(x0)) <

SO, d(y,Tn-m(x)) ~ d(y,Tn(x0)) + d(Tn(x0),Tn-m(x))

= d(y,Tn(x0 )) + d ( T m ( x o ) , x )

< 2~.

Therefore OT(X) = X. //

We now check our examples for topological transitivity.

Examples :

(i) I is t o p o l o g i c a l l y transitive iff X = one point.

(it) T(x) = ax is t o p o l o g i c a l l y transitive iff T is m i n i m a l iff

T is ergodic iff {an: n ~ Z} is dense in X.

Proof: Suppose OT(X 0) : X, i.e., the closure of the set

{anx0: n ~ Z} equals X. There exist {n i} such that

n. n.
1 i.e. a l
a .x 0 ~ y.x 0 , ~ y.

So, {an: n ~ Z} is dense in X. (Another proof would be to apply

Theorem 5.7 or C o r o l l a r y 5.5.) //

(iii) An a u t o m o r p h i s m A of a compact metric group is t o p o l o g i c a l l y

 122

transitive iff A is ergodic. (See C o r o l l a r y 5.5.)

(iv) An affine t r a n s f o r m a t i o n T of a compact metric group X is

t o p o l o g i c a l l y t r a n s i t i v e iff T is ergodic. (See Corollary 5.5.)

(v) The shift on k symbols is t o p o l o g i c a l l y transitive. Consider

X = ~ {0,1,...,k-l}, T = shift. We know T is an a u t o m o r p h i s m of

the compact m e t r i c group X. The H a a r measure on X is the measure

given by the weights i/k,...,i/k. (To check this, fix x E X and

show, by c h e c k i n g on r e c t a n g l e s and u s i n g T h e o r e m i.i, that this

measure is invariant under t r a n s l a t i o n by x.) T is ergodic with

respect to Haar measure, and therefore T is t o p o l o g i c a l l y transitive

by C o r o l l a r y 5.5.

~3. Topological Conjugacy and Discrete S~ectrum

When should we c o n s i d e r two h o m e o m o r p h i s m s of compact spaces to

be the "same" from a d y n a m i c a l point of view? The following seems

the most suitable:

Definition 5.5:

Let T: X ~ X, S: Y ~ Y be h o m e o m o r p h i s m s of compact spaces.

T is t o p o l o g i c a l l y conjugate to S if there exists a h o m e o m o r p h i s m

¢: X ~ Y such that ~T = S¢.

Notes:

(i) This is an e q u i v a l e n c e relation.

(2) If T and S are t o p o l o g i c a l l y c o n j u g a t e then T is m i n i m a l

iff S is m i n i m a l and T is t o p o l o g i c a l l y t r a n s i t i v e iff S is

t o p o l o g i c a l l y transitive.

D e f i n i t i o n 5.6:

Let X be a compact metric space, T: X ~ X a homeomorphism,

f a c o m p l e x - v a l u e d c o n t i n u o u s f u n c t i o n on X. We say that f is an
 123

eigenfunction for T if 3 k E C )

f(Tx) = kf(x) V x E X, and f ~ 0.

We t h e n c a l l k the c o r r e s p o n d i n g eigenvalue for f.

Remarks:

Suppose T is t o p o l o g i c a l l y transitive.

(i) f(Tx) = kf(x), f E C(X) = Ikl = 1 and If(x) I = c o n s t a n t .

Proof: f(Tx) = kf(x) = If(Tx) I = Ikllf(x) I. Therefore,

sup If(Tx) i : i~Isup If(x)l

xEX x{X

and since T is o n t o = sup If(x) I .

xEX

Therefore Ikl = I. Hence, If(Tx) l = If(x) I and by T h e o r e m 5.6

If(x) l = c o n s t a n t . //

(2) If fT = kf, gT = kg, f,g (C(X) then f = eonstant.g.

Proof: By (i), Ig(x) l ~ 0 V x ( X since g ~ 0. Therefore

(f/g)(Tx) = (f/g)(x) = f/g = c o n s t a n t by T h e o r e m 5.6. //

(3) Eigenfunctions corresponding to d i s t i n c t eigenvalues are l i n e a r -

ly i n d e p e n d e n t in C(X).

Proof: Let fn(TX) = knfn(X) where {kn} are all d i s t i n c t for

n = l,...,k. Suppose V x ( X,

alf!(x) + a 2 f 2 (x) + ... + ~ k f k ( x ) = 0

where the a. E C for i = l,...,k.

l

By a p p l y i n g the above equation to T1x i n s t e a d of x, we get

a l ~ f l ( X ) ÷ a2~2f2i (×) ÷ ... ÷ a k ~ f k ( ~ ) : 0 ~ x ~ X.

Hence
 (
124

ii
I 1 1

kI k2 "'" kk)(a2f2.(x)) =
• ..

k-i k-i
kI k2 . . .

All the k.l S are distinct so the matrix is nonsingular. Therefore

V x E X,

i.e., aifi(x) : 0 V x E X, i = 1,...,k, i.e., a i = O,

;
i = l,...,k since f. ~ O. Hence, the f. s are linearly inde-
l 1
pendent in C(X). //

(4) The eigenvalues form a subgroup of the circle group K.

Under our assumptions T has only countably many eigenvalues.

To check there are only countably many eigenvalues it suffices to show

that if h: X ~ K is an e i g e n f u n c t i o n corresponding to an eigenvalue

~ 1 then Ilh- i]I > 1/4. For then two eigenfunctions, with values

in K~ corresponding to different eigenvalues will be greater than

distance 1/4 apart and, since C(X) has a countable dense set,

there can only be countably many eigenvalues. So let h(Tx) = ~h(x),

i. Choose x0 ~ X and p so that ~Ph(x O) is in the left-hand

half of the unit circle. Then

]lh -iii : sup IIh(x) -llI

xEX

>- Nh(mPx 0) -iN : li~Ph(x 0 ) - l N > 1/4. II

Definition 5.7:

Let T: X ~ X be a h o m e o m o r p h i s m of the compact metric space X.

 125

We say that T has topological discrete spectrum if the smallest

closed linear subspace of C(X) containing the eigenfunctions of T

is C(X), i.e., the eigenfunctions span C(X).

Note:
When T is topologically transitive and has topological discrete

spectrum, 3 fn: X ~ X, n : 1,2,... linearly independent, spanning

C(X), such that fnT(X) = knfn(X). The following is a representation

theorem for topologically transitive homeomorphisms with topological

discrete spectrum.

Theorem 5.8: (Halmos and von Neumann [i])

The following are equivalent for a homeomorphism T of a compact

metric space X:

(i) T is topologically transitive and is an isometry for some metric

on X.

(2) T is topologically conjugate to a minimal rotation on a compact

abelian metric group.

(3) T is minimal and has topological discrete spectrum.

(4) T is topologically transitive and has topological discrete spec-

trum.

Proof: (I) = (2). Let d be the isometry metric for T. Sup-

pose OT(X 0) = X. Define a multiplication * in OT(X 0) by

TnxO~Tmxo = Tn+mxo. We have

d(Tnx0,Tmx0,TPx0,Tqx0 ) : d(Tn+mx0,TP+qxG)

d(Tn+mx0,TP+mx0) + d(TP+mx0,TP+qx0)

: d(Tnx0,TPx 0) + d(Tmx0,Tqx0).

Hence, the map *: OT(X0)×OT(X 0) ~ OT(X 0) is uniformly continuous and

therefore can be extended uniquely to a continuous map ,: X×X ~ X.
 126

Also, d(T-nx0,T-mx0 ) = d(Tmx0,Tnx0 ) and so,

0T(X0) inverse ~ 0T(X0)

is u n i f o r m l y continuous and can be u n i q u e l y extended to a c o n t i n u o u s

map of X. Thus we get that X is a t o p o l o g i c a l group and is also

abelian since it has a dense abe!ian subgroup {Tnx0: n ( Z}. Since

T ( T n x 0) = Tn+ix0 = Tx0,Tnx 0 we have Tx = T x 0 , x and so T is the

rotation by Tx 0 .

(2) = (3). If T is a m i n i m a l rotation on a c o m p a c t abelian

group G then each character of G is an e i g e n f u n c t i o n . Let A be

the c o l l e c t i o n of all finite linear combinations of c h a r a c t e r s . Then

A is a s u b a l g e b r a of C(X), contains the c o n s t a n t s , is closed under

complex conjugation, and s e p a r a t e s points. Applying the Stone-

Weierstrass Theorem we see that ~ = C(X).

(3) = (4) is trivial.

(4) = (i). We can choose eigenfunctions f : X ~ K, n ~ i,

n
with fn(T) = knf n and where the fn are l i n e a r l y independent and

span C(X). Define a metric on X by:

d(x,y) = f Ifn(X) - fn(Y)l

n=l 2n

Iknfn(X) knfn(Y) I
Then d(Tx,Ty) = ~ = d(x,y).
n=l 2n

It r e m a i n s to c h e c k that d gives the t o p o l o g y on X. If d(Xn,X) -~0

then for all n ~ i, as m ~ -,

1 ifn(Xm ) _ fn(X)i ~ d(Xm,X) ~ 0.

2n

Thus, V n ~ i, fn(Xm) ~ fn(X) as m ~® and s i n c e {fn } separates

 127

points, xm ~ x as m ~ -. Conversely, suppose Xm ~ x. Let ~ > 0,

and c h o o s e N such t h a t

~. 2 <
n=N+l 2n 2

By the c o n t i n u i t y of the f u n c t i o n s fl'''''fN 3 M ) m>M =

Ifi(Xm) - fi(x) I < ~/2 i = I,...,N. If m > M then

1
d(Xm'X) : i~--I V Ifi(Xm) - fi(x) l

N
-< L. ~ ' - + - -<
i=l 21 2 2

i.e., d(Xm,X) ~ 0. //

Remark:

If Tx = ax is a m i n i m a l rotation of a c o m p a c t m e t r i c abelian

group G it is s t r a i g h t f o r w a r d to show that the set of e i g e n v a l u e s

of T is {T(a): Y ~ G} and e v e r y eigenfunction is a c o n s t a n t

multiple of a c h a r a c t e r . In fact, this f o l l o w s from Theorem 3.3 since

each c o n t i n u o u s eigenfunction is an L 2 - e i g e n f u n c t i o n .

We h a v e the f o l l o w i n g isomorphism theorem.

Theorem 5.9: (Topological Discrete S p e c t r u m Theorem)

Two m i n i m a l homeomorphisms of c o m p a c t m e t r i c spaces both having

topological discrete s p e c t r u m are t o p o l o g i c a l l y conjugate iff they

h a v e the same e i g e n v a l u e s .

Proof: (i) The p r o o f is a l o n g the lines of the p r o o f of T h e o -

r e m 3.1, but instead of u s i n g T h e o r e m 2.1 we use the B a n a c h - S t o n e

Theorem. This says that if X,Y are c o m p a c t spaces, ~: C(Y) ~ C(X)

is a b i j e c t i v e linear isometry, and ~(f.g) = ~(f)¢(g), t h e n there

exists a homeomorphism ~: X ~ Y such that ~(f)(x) = f(~(x)).

 128

(2) This theorem can also be p r o v e d using T h e o r e m 5.8 and char-

a c t e r theory. By T h e o r e m 5.8 we can suppose T is a m i n i m a l rotation

of a compact abelian group G, Tx = ax, and S is a minimal rota-

tion of a compact abelian group H, Sy = by. We are assuming

{y(a): 7 ~ G} = {8(b): 5 E H}. Define a map e: H ~ G by 8(5)(a) =

5(b). This is w e l l - d e f i n e d and a bijection. Moreover, 8 is easily

checked to be a group automorphism and hence induces an a u t o m o r p h i s m

C: G ~ H. It is easy to show that CT = SC. //

Remark:

Thus the t h e o r y of t o p o l o g i c a l l y transitive homeomorphisms with

topological discrete spectrum is e n t i r e l y analogous to that of ergodic

measure-preserving transformations with pure point spectrum.

§4. Invariant Measures for H o m e o m o r p h i s m s

In this section we c o n s i d e r some c o n n e c t i o n s between the topo-

logical and m e a s u r e theoretic systems. We first prove some results

about Borel m e a s u r e s including the fact that a Borel measure on a

metric space is d e t e r m i n e d by how it integrates continuous functions.

By a Borel m e a s u r e on X is meant a measure defined on the Borel sub-

sets of X, (i.e., the smallest o-algebra containing the closed sets).

Theorem 5.10:

A Borel probability measure m on a metric space X is r e g u l a r

(i.e., if B denotes the Borel sets then V B ~ S and V ~ > 0 3

an open set U and a c l o s e d set C with C B ~ U and

m(U \ C ) < ~).

Proof: (The proof does not require X to be m e t r i c but that

each closed set be a GS.) Let R be the c o l l e c t i o n of all sets such

that the r e g u l a r i t y condition holds, i.e., R = {A 6 B: V s > 0 3

open U , closed C with C c A ¢

_ _ U and m(U \C ) < a}. We show
 129

that R is a q-algebra. Let A E R; we show that XXA E R. Let

s > 0. 3 open Us, closed Cs with Cs g A g U s ) m(Us\C ~) < s.

Thus, XNU g X\A g X \ C and (XkC s) \ (\\Us) : Us\Cs, so

m((XkC ) \ (XkUs)) : m(Us\C s) < s.

Therefore XkA E R.

We now show R is closed under countable unions. Let

AI,A2,... ~ R and let A : 0 A.. Let s > 0 be given. There

i:l l
exist open Us,n, closed Cs, n such that Cs, n A n ~ U s ,n and

m(Us,n\ Cs, n) < ~/3 n. Let U s : n:l

0 U c,n (which is open)

c¢ = 0 C n' and choose k such that m(C \ b Cs, n) < ~/2.

n= 1 s, ~ n:l
k
Let C = LJ Cs,n (which is closed). We have C~ _c A c_ Us. Also,
n=l

m(Us\C s) ~ m(Us\C s) + m(Cs\C e)

~ m(Us, n \ Cs, n) + m(Cs\C s)

n:l

S S
--< 2-- + = S
n:l 3 n 2

Therefore R is a ~-algebra.

To complete the proof we show that R contains all the closed

subsets of X. Let C be a closed set and s > 0. Define

U n = {x E X: d(C,x) < i/n}. This is an open set, U 12 U 2 2 ... 2 U n 2 ...

and A U. = C. Choose k such that m(Uk\C) < s and let C = C

i=l i s
and U s = U k. This shows C E R. //
 130

Corollary 5.10:

For a B o r e l probability measure m on a m e t r i c space X we h a v e

that for a B e r e l set B

m(B) : sup m(C) and m(B) : inf m(U).

C closed U open
C & B U ~ B

Theorem 5.11:

Let m,~ be two Borel probability measures on the m e t r i c

space X. If IX f d m : IX f d~ V f E C(X) then m = ~.

Proof: By the above corollary it s u f f i c e s to show that m(C) =

~(C) for all closed sets C g X. Suppose C is c l o s e d and let

> O. By the regularity of m there exists an o p e n set U, U 2 C

such that m(UkC) < e.

Define f: X ~ R by

f(x) = I 0 if x ~ U

[ d(x,X\U)
d ( x , X \ U ) + d(x,C)
if x E U.

f is w e l l - d e f i n e d since the denominator is not zero. Also f is

continuous, f = 0 on XkU, f : 1 on C, and 0 ~ f(x) ~ 1 V

x ( X. Then,

~(C) ~ I f d~ = I f dm ~ m(U) < m(C) + s.

JX ~x

Therefore ~(C) < m(C) + s V ~ > 0, so ~(C) ~ m(C). By s y m m e t r y

we get that m(C) ~ ~(C). //

Theorem 5.12: (Riesz Representation Theorem)

Let X be a compact metric space and J: C(X) ~ C a continuous

linear map such that J is a p o s i t i v e operator (i.e., if f ~ 0 then

J(f) ~ 0) and J(1) = i. Then there exists a Borel probability

 131

measure ~ on X such that

J(f) : f d~

for all f in C(X).

Proof: See Halmos [i], p. 247. //

The next theorem expresses the fact that the unit ball in the
dual space of C(X) is weakly compact.

Theorem 5.13:

If {~n } is a sequence of Borel probability measures on a com-

pact metric space X, then there is a subsequence {~n } which con-

verges in the weak topology, i.e., 3 a Borel probability measure

on X such that

IX f d~n"d ~~ IX
l f

for all f in C(X).

Proof: We write ~(f) = ; f d~ when f (C(X) and ~ is a

Borel measure. Choose fl,f2,.., dense in C(X). Consider the se-

quence of complex numbers {~n(fl)}. This is bounded by llflll~ and

so has a convergent subsequence, say { ~ l ) ( f l ) }. Consider
{ (1)(f2)} ; this is bounded and so has a convergent subsequence
~n
{~2)(f2)}. Notice that {~2)(fl)} also converges. We proceed in

, (i)
this manner, and for each i ~ i, construct a subsequence i~n } of

{bn} such that {~n(i)~; ~ i~~ n(i-l),~ & "'" ~ {~n(!)} g {~n} and so that

{~i)(f)} converges for f : fl,f2,...,f i. Consider the diagonal

{~n)}. The sequence {~n)(fi)} converges for all i; thus

(n)
~n (f)} converges for all f (C(X) (by an easy approximation
 132

argument). Let J(f) : lim ~n)(f).r Clearly J: C(X) ~ C is linear

n~

and bounded, as IJ(f)l ~ llfll. Also J(1) = I, and if f ~ 0 then

J(f) a 0. By the Riesz Theorem, there exists a Borel p r o b a b i l i t y

measure on X such that J(f) = f f d~ for all f E C(X), i.e.,

IX f a~n(n)
.
~ IX f d~. //

Corollary 5.13:

The space of Borel p r o b a b i l i t y m e a s u r e s on a compact metric

space X is itself a compact metric space under the weak topology.

Proof: Let fl,f2,.., be dense in C(X). Define

i:l 2ilifill

D is a metric on the space of Borel p r o b a b i l i t y m e a s u r e s which gives

rise to the weak topology. The compactness follows from the previous

theorem. //

Theorem 5.14: (Krylov and B o g o l i o u b o v [i])

If T is a h o m e o m o r p h i s m of a compact metric space X then

there exists a Borel p r o b a b i l i t y measure on X w h i c h is preserved

by T.

Proof: Fix x E X. For f E C(X) and n a 0, define

n-i
Jn(f) = ~1 i__~
.= 0 f(Ti(x))

J : C(X) ~ C satisfies the conditions of the Riesz R e p r e s e n t a t i o n

n
Theorem (no~e that IJn(f)l ~ IIfll), so there exists a Borel proba-

bility measure ~n on X such that

 133

Jn(f) = IX f d~ n for all f ~ C(X).

By Theorem 5.13 there exists a subsequence {~n. } and a Borel proba-

]
bility measure ~ on X such that

Jnj(f) = IX f d~n3 + IX f d~ for all f E C(X).

Since IJnj(f°T) - Jnj (f)I : n~ IfTnJ(x) - f(x)I

I_..
-< "'211fll "~ 0 as j ~-
n.
3

we have IX foT d~ = IX f d ~

i.e., IX f d~T-i : IX f d~ V f E C(X).

So by the uniqueness Theorem 5.11 for Borel measures we have that

~(T-IB) = ~(B) for all Borel sets B. //

Theorem 5.15:

Let T be a homeomorphism of a compact metric space X, and

let MT denote the collection of all T-invariant Borel probability

measures on X (by Theorem 5.14, M T # ¢). Then

(i) MT is closed in the weak topology,

(2) MT is a convex set, and
(3) if m ~ MT then m is an extreme point of MT iff m is
ergodic with respect to T.

Proof: (I) Suppose {~n} c M T converges to ~ in the weak

topology. Then
 134

I fT d~n ~ I fT d~
H

Ifd~n -~ [fd~

so that ~ is T-invariant.

(2) is obvious.

(3) Suppose m 6 MT, m not ergodic. There exists a Borel

set E such that T-IE : E a.e. and 0 < m(E) < i. Define measures

mI and m2 by

mlCB
). _ m(B N E) and m2tB
~ . , : m ( B N (X\E))
m(E) m(X\E)

for B a Borel set. Note that mI and m2 are in M T, m I ~ m2,

and

m(B) : m(E)ml(B) + (i- m(E))m2(B),

so that m is not an extreme point of M T.

Conversely, suppose m 6 MT is ergodic, and

m : zm I + (l-~)m 2

where ml,m 2 ~ MT, 0 ~ z ~ 1. We must show m I : m 2. m I << m

(m I is absolutely continuous with respect to m) so that the Radon-
dmi(x)
Nikodym derivative dml/dm exists, (i.e., ml(E) = dm(x) ).
dm

: i
dm
So, ml(E) : mI(T-IE) - - 1 dm(x)
T_IE dm

dml(
: IE d--~- T-Iy) dm(T-ly)

dm 1
= IE d-~--(T-ly) dm(y).
 135

dml 1 dml
Therefore d--~--(T-y) : d--~--(y) a.e.(m) (by uniqueness of the Radon-

Nikodym derivative). But, m is ergodic, so that dmlldm =

constant = k a.e.(m). Therefore

1 : ml(X) : IX k dm : k.m(X) : k.

Since k : i, mI = m and therefore m 2 : m : m 1. //

Definition 5.8:

T is u n i q u e l y er$odic if there is only one T - i n v a r i a n t Borel

p r o b a b i l i t y m e a s u r e on X, i.e., M T = one point.

Remark:

T is u n i q u e l y ergodic with respect to m implies that T is

ergodic with respect to m.

Unique e r g o d i c i t y is connected to m i n i m a l i t y by:

Theorem 5.16:

Suppose T is u n i q u e l y ergodic and m is its unique invariant

measure. T is m i n i m a l iff m(U) > O for all n o n e m p t y open sets U.

Proof: Suppose T is minimal. If U is open, U ~ ¢ then

X = 0 Tn(U) ; so if m(U) = 0 then m(X) = 0, a contradiction.

Conversely, suppose m(U) > 0 for all open n o n e m p t y U. Suppose

also that T is not minimal, i.e., there exists a closed set K such

that TK = K, K ~ X. TIK has an invariant Borel p r o b a b i l i t y

measure ~K on K by T h e o r e m 5.14. Define ~ on X by ~(B) =

~ K ( K n B) for all Borel sets B. Then ~ E MT and ~ ~ m because

m(X\K) > 0 as X\K is n o n e m p t y and open while ~(X\K) = 0. This

contradicts the unique e r g o d i c i t y of T. //

The f o l l o w i n g results formulate unique e r g o d i c i t y in terms of

 136

ergodic averages.

Theorem 5.17:

The following are equivalent:

n-i
1
(I) V f £ C(X), n i=0 f(Tix) converges uniformly to a constant.

1 n~l
(2) V f E C(X), ~ i:0 f(Tlx) converges pointwise on X to a

constant.

(3) 3 m ( MT ~ V f (C(X) and V x ( X,

1 n-i
~ .
f(Tlx) -~ I f dm.
n i=0

(4) T is u n i q u e l y ergodic.

Proof: (i) = (2) holds trivially.

(2) = (3). Define k: C(X) ~ C by

n-i
k(f) = lim 1 ~ fTi(x).
n-~- n i:O

Observe that k is a linear operator and is continuous since

n-i
1
1 ~ Z fT1(x) I -< Ifl •
i:O

Also k(1) = I and f >_ 0 = k(f) >_ 0. Thus by the Riesz Representa-

tion Theorem there exists a Borel p r o b a b i l i t y measure m such that

k(f) = ; f dm. But k(fT) = k(f) and so, I fT dm = I f dm, i.e.,

I f dmT-i = I f dm which implies that mT-i = m by 5.ii, so that

m ( MT •
(3) = (4). Suppose that v ( M T. We have

n-i
1 [ fTl(x) ~ f* V x,
i=0

where f* : I f dm. Integrating with respect to ~, and using the

 137

bounded convergence theorem we get that

I f d~ : I fe d~ : fe : I f dm V f 6 C(X).

Hence ~ : m by 5.11. Therefore T is uniquely ergodic.

n-i
(4) = (1). If ~1 ~~-':0 fTi(x) converges uniformly to a constant

then this constant must be J f dm, where m is the unique T - ~ v ~ i ~ t

measure. Suppose (i) is false. Then 3 g E C(X), 3 e > 0 9 V N

3 n > N and 3 xn ~ X 9

n-i
i ~0 1i~
.:
= gTi(xn ) - I g dm I -> e

n-i
1
Define Jn: C(X) ~ C by Jn(f) : ~ ~ fT1(Xn ). J satisfies the
i:0 n
conditions of the Riesz R e p r e s e n t a t i o n Theorem. Hence, Jn(f) = I f dlxn
for some Borel probability measure ~n" Moreover, there exists a sub-

sequence {~n. } such that

l

Jni(f) : I f d~n"d ~ lI f

for all f E C(X) and for some Borel probability measure ~ (by
Theorem 5.13). Then

ni
IJni(fT) - Jni(f)l : n~i IfT (Xn')1 - f(Xn')ll

2JJfll
_~ n. ~ O,
l

so %hat I fT d~ : I f d~.

Hence, ~ E M T. But, II g d ~ - I g dml ~_ e so that ~ ~ m contra-

dicting the uniqueness of m. //

We now see which of our examples are uniquely ergodic.

 138

Examples:

(i) I is u n i q u e l y ergodic iff X : one point, since M T = all

Borel p r o b a b i l i t y measures.

(ii) T(x) = ax on a compact group is uniquely ergodic iff T is

minimal.

Proof: (=) follows from T h e o r e m 5.16 and the fact that Haar

measure is p o s i t i v e on o p e n sets.

(=) T is m i n i m a l ,, {a n } is dense in G. Therefore G is

abelian. If 1 ~ Y ( G then

n-i n-i
1 [ y(Tix): ~1 [ y(ai)T(x) _ y(x) (y(a) n -i) (] as n ~ ~

i=O i:O n TCa) - 1

(note that T(a) X i). So (2) of T h e o r e m 5.17 holds when f is a

character and the c o n d i t i o n (2) will hold for each f E C(X) by

approximation, since finite linear combinations of c h a r a c t e r s are

dense in C(X). //

(iii) An a u t o m o r p h i s m of a compact group G is u n i q u e l y ergodic iff

G = {e}, since Hear m e a s u r e is p r e s e r v e d and so is the point measure

c o n c e n t r a t e d at e.

(iv) An affine t r a n s f o r m a t i o n of a compact connected abelian metric

group is u n i q u e l y e r g o d i c iff it is minimal.

Proof: (=) follows by Theorem 5.16.

(=) follows, as in example (ii), by checking that (2) of Theo-

rem 5.17 holds. This was done by Hahn and Parry [i]. //

(v) The B e r n o u l l i shift on k symbols is u n i q u e l y ergodic iff

k = i. This is by example (iii).

An e x c e l l e n t survey of unique e r g o d i c i t y and related topics can

be found in J. C. Oxtoby [!].

 139

Recent results of Jewett [i] and Krieger [i] imply that any

ergodic invertible measure-preserving transformation of a Lebesgue

space is isomorphic in the sense of Chapter 2 to a uniquely ergodic

system. This indicates a certain lack of measure-theoretic import

for the concept of unique ergodicity. Hahn and Katznelson [I] have

found uniquely ergodic transformations in shift spaces with arbi-

trarily large measure-theoretic entropies.

 Chapter 6: Topological Entropy

Adler, Konheim, and M c A n d r e w [I] introduced topological entropy

as an i n v a r i a n t of t o p o l o g i c a l conjugacy and also as an a n a l o g u e of

measure theoretic entropy.

91. Definition by O p e n Covers

All logarithms are to base 2. Let X be a c o m p a c t topological

space. We shall be interested in o p e n covers of X which we d e n o t e

by e,~, . . . .

Definition 6.i:

If ~,~ are open covers of X their ~oin ~ v ~ is g i v e n by:

Definition 6.2:

An o p e n cover ~ is a r e f i n e m e n t of an o p e n cover a, written

< ~, if e v e r y member of ~ is a s u b s e t of a m e m b e r of ~. In

particular, ~ < ~ v ~, ~ < ~ v ~.

Definition 6.3:

If e is an o p e n cover of X and T: X ~ X is c o n t i n u o u s then

T-l~ = {T-I(A): A E a} is an open cover of X.

Note:

T-l(~ v ~) : T - l ( a ) v T-l(~) and < ~ = T-la < T-l~.

Definition 8.4:

If ~ is an o p e n cover of X let N(~) = the number of sets in

a finite subcover of = with smallest cardinality. We d e f i n e the

entropy of ~ by: H(~) = log N(~).

 141

Remarks:

(i) H(e) ~ 0.

(2) H(e) = 0 iff N(a) = 1 iff X ~ ~. H(e) is small m e a n s that

t h e r e are a few sets in a which cover X. H(e) is large means that

some part of X is c o v e r e d by a large n u m b e r of sets in e and not

by a s m a l l n u m b e r .

(3) ~ < 6 ~ H(~) ~ H(6).

Proof: Let {BI,...,BN(6)} be a s u b c o v e r of 6 with minimal

cardinality. V i 3 Ai E e ) A i ~ B i. So, {AI,...,AN(6)}

covers X and is a s u b c o v e r of ~. Thus N(a) ~ N(~). //

(4) H(e v 6) ~ H(~) + H(6).

Proof: Let {AI,...~AN(~)} be a s u b c o v e r of ~ of m i n i m a l car-

dinality, and {BI,...,BN(6)} be a s u b c o v e r of ~ of m i n i m a l cardi-

nality. Then

{A i n Bj: 1 ~ i ~ N(e), 1 ~ j ~ N(~)}

is a s u b c o v e r of e v 6 so, N (e v 6) ~ N(e)N(~). //

(5) If T: X ~ X is a c o n t i n u o u s m a p then H(T-I~) ~ H(~). If T

is also surjective then H(T-Ia) : H(=).

Proof: If {AI,...,AN(e)} is a s u b c o v e r of ~ of m i n i m a l car-

dinality then {T-IAI,...,T-IAN(g)} is a s u b c o v e r of T-le, so

N(T-I~) ~ N(=). If T is onto, and {T-IA!,...,T-IAN(T-Ie)} is a

subcover of T-la of m i n i m a l cardinality then {AI,...,AN(T-Ia)}

also c o v e r s X, so N(e) ~ N(T-I=). //

Theorem 6.1:

If = is an o p e n c o v e r of X and T: X ~ X is c o n t i n u o u s , then

lim ~ H(~ v T-I~ v ... v T - ( n - l ) ~ ) exists.

n
n~
 142

Proof: Recall that if we set

a = H(s v T-I~ v ... v T-(n-l)~)

n

then by T h e o r e m 4.4 it suffices to show that:

> 0, and < a + a V m,n.

an - an+m - n m

By (i), a n ~ O, and

an+ m = H(~ v T-I~ v ... v T-(n+m-!)s)

H(s v T-is v ... v T-(n-l)s)

+ H(T-n~ v ... v T-(n+m-l)~) by (4)

: a + -'-M(T-n~ v ... v T - '(n + m - l ) ~ )

n

: a + ---'H(T-n(s v ... v T-(m-l)s))

n

a n + H(s v ... v T-(m-l)s) by (5)

: a
n
+ a.
m
II

Definition 6.5:

If ~ is an open cover of X and T: X ~ X is a c o n t i n u o u s

map then the entropy of T relative to s is g i v e n by:

h(T,s) : lim ~ H(s v T-is v ... v T-(n-l)~).

n
n~

Remarks:

(6) h(T,s) ~ 0 by (I).

(7) s < ~ = h(T,a) _< h ( T , ~ ) .

n-i n-i
Proof: s < ~ = V T-is < V T-i~, so by (3) we have that
i:0 i:0
n-i n-i
H( V T-is) _< H( V T-i~). Hence h(T,s) _< h ( T , ~ ) . //
i=0 i=0
 143

Note that if ~ is a f i n i t e subcover of = then a < ~ so t h e n

h(T,a) ~ h(T,~).

(8) h(T,a) E H(=).

Proof: By (4) we h a v e

H(a v T-Ia v ... v T-(n-l)a)

n-i
~ H(T-I~)
i=0

n.H(=) by (5). //

Definition 6.5:

If T: X ~ X is c o n t i n u o u s , the topological entropy of T is

given by:

h(T) = sup h(T,=)

where = ranges over all o p e n covers of X.

Remarks:

(8) h(T) ~ 0.

(10) In the definition of h(T) one can t a k e the supremum over

finite covers of X. This follows from (7).

(ii) h(I) = 0 where I is the identity map of X.

The next result shows that topological entropy is an invariant

of t o p o l o g i c a l conjugacy.

Theorem 6.2:

If X1,X 2 are compact spaces and Ti: Xi ~ Xi are continuous

for i = 1,2, and are topologically conjugate, then they have the

same entropy.

Proof: Suppose ~: X 1 ~ X 2 is a h o m e o m o r p h i s m such that

~T 1 = T2~. Let = be an open cover of X 2. Then,

 144

h(T2,s) = lim ~i H(s v T21s v . . . v T2(n-l)e)

n

: l i m ~1 H(¢-l(a V T21a V ... v T2(n-1)s)) by (5)

n

: lim .i
r H(¢-I v Til¢-la v ... v Ti(n-l)¢-la)
n

: h(Tl,¢-la).

By taking suprema the result follows. //

Adler, Konheim, and Mclhndrew proved several results about h(T).

In the next section we give a definition of topological entropy for

any uniformly continuous map of a metric space (not n e c e s s a r i l y com-

pact). This d e f i n i t i o n can also be given for uniform spaces. The

definition will reduce to the previous definition in the compact case.

We shall prove the properties of h(T) with this new definition.

However, one result we would like to note is the following:

Theorem 6.3:

If T: X ~ X is a h o m e o m o r p h i s m of a compact space X, then

h(T) : h(T-l).

Proof:

h(T,a) : lim ~ H(a v T-Is v ... v T-(n-1)s)

n
n

: lim ~ H(Tn-I(= v T-Is v ... v T-(n-l)s))

n
n

: lim ~ H(s v Tav ... v Tn-la)

n
n

: h(T-l,s). //

Adler, Konheim, and McAndrew could not answer the following

questions which they stated as conjectures in their paper.

 145

Notation:

From now on the m e a s u r e t h e o r e t i c entropy of a t r a n s f o r m a t i o n T

preserving a measure m will be w r i t t e n hm(T).

Con$ecture i:

Let X be compact and m a regular Borel measure on X. If

T: X ~ X is a h o m e o m o r p h i s m p r e s e r v i n g m then

hm(T) ~ h(T).

[This has been p r o v e d by Goodwyn [i]. T need only be continuous. We

shall give a p r o o f w h e n X is a f i n i t e - d i m e n s i o n a l torus. (See Theo-

rem 6.9.]

Conjecture 2:

Let X be a compact metric space and T: X ~ X a homeomorphism.

(By T h e o r e m s 5.i0 and 5.14 we know that MT, the set of Borel meas-

ures on X invariant u n d e r T , is nonempty.) Then h(T) = sup hm(T).

m~M T
[Partial c o n t r i b u t i o n s were made by Goodwyn Ill and Dinaburg [i], but

Goodman [i] finally proved it. We can drop the condition of X being

metric and allow T to be only continuous, provided we define MT to

be all T - i n v a r i a n t r e g u l a r Borel p r o b a b i l i t y measures on X.]

Conjecture 3:

If X is a compact metric group and T is an a u t o m o r p h i s m of

then h(T) = hm(T) where m is Haar measure.

[This was shown by Berg [I] and g e n e r a l i z e d by Bowen [4]: T can be

an affine t r a n s f o r m a t i o n and need not n e c e s s a r i l y be invertible. We

shall prove this. (See T h e o r e m 6.10.)]

Conjecture 4:

Let {T t } be a o n e - p a r a m e t e r group of h o m e o m o r p h i s m s of a com-

pact space X. Then h(T t) = Itlh(Tl).

 146

[This was proved by B o w e n when X is m e t r i c . ]

Conjecture 5:

Let X,Y be compact spaces. Let {Tx: x ~ X} be a f a m i l y of

homeomorphisms of Y so t h a t

T(x,y) = (X,Tx(Y))

is a c o n t i n u o u s map of X×Y. Then

h(T) = sup h(Tx).

xEX

[This was proved in the case where X and Y are m e t r i c by Bowen.]

§2. Bowen's Definition

If (X,d) is a m e t r i c space and x E X then B g (x) will denote

the o p e n ball centered at x and of r a d i u s s. UC(X,d) will denote

the collection of all u n i f o r m l y continuous maps ~: X ~ X.

Let T E UC(X,d); n E Z, n > 0; and s > 0. If K ~ X, a

set F ~ X is s a i d to (n,s)-span K with respect to T if V x E K

3 y E F such that

max d(T i x , T 1 "y) ~ E.

0Ei~n-I

For K compact, let rn(~,K) be the smallest cardinality of any

(n,s)-spanning set for K with respect to T. We show later (Theo-

rem 6.4) that r (~,K) < ®.

n
1
Set rT(~,K) = lim sup ~ log r n ( ~ , K ) .
n

A set E c X is (n,e)-separated with respect to T if, whenever

x,y E E, x ~'y then

max d(Tix,Tiy) > e.

0~i~n-i
 147

For K compact, let Sn(S,K) denote the largest cardinality of

any ( n , s ) - s e p a r a t e d subset of K with respect to T. We later show

(Theorem 6.4) that Sn(S,K) is finite.

1
Set ST(S,K) : lim sup ~ log Sn(S,K).

We define h(T,K) : lim rT(s,K) : lim ST(S,K). These limits

s~O s~O
exist and are equal by T h e o r e m 6.4. We then define

hd(T) : sup h(T,K).

K compact

Remarks:

(i) This d e f i n i t i o n can also be given in the context of u n i f o r m

spaces.

(2) hd(T) measures the amount of expansion in T (for the metric

d). For rn(s,K) and Sn(e,K) to increase as n increases we need

some e x p a n s i o n for T.

(3) The ideas for this d e f i n i t i o n come from the work of Kolmogorov

on the size of a metric space. If (X,p) is a metric space then a

subset F is said to e-span X if V x £ X 3 y ~ F with

p(x,y) ~ s, and a subset E is said to be e-separated if whenever

y,z ~ E, y # z, then p(y,z) > s. The e-entropy of (X,p) is then

the l o g a r i t h m of the m i n i m u m n u m b e r of elements of an e-spanning set

and the s - 9 a p a c i t y is the l o g a r i t h m of the m a x i m u m number of elements

in an e-separated set. So in the above d e f i n i t i o n s we are considering

the metric spaces (K,d n) where dn is the r e s t r i c t i o n to the com-

pact set K of the metric

Pn(X,y) : max d(Tix,Tiy).

0~i~n-i

Then h(T,K) : lim iim sup ! [e-entropy of (K,Pn)]. (It follows from
S~O n ~
 148

the proof of the next t h e o r e m that to define h(T,K) it suffices to

consider s p a n n i n g sets for K which are subsets of K.) ST(g,K) is

the average e - c a p a c i t y of the spaces (K,d n) and h(T,K) is the

limit of ST(~,K) as ~ ~ 0.

Theorem 6.4:

Suppose K is compact. Then

(i) rn(g,K) ~ Sn(~,K) ~ rn(~/2,K) < - and

(ii) if eI < z 2 then rT(Sl,K) £ rT(~2,K) and

ST(~I,K) ~ ST(¢2,K).

Proof: (i). We first show r (g,K) < -. There exists a 6 > 0

n
such that d(x,y) < 5 implies

max d(Tix,Tiy) < ~.

0~i~n-I

Then r (~,K) is less than the number of 5-balls needed to cover K

n
and hence is finite.

We shall now prove Sn(Z,K) ~ rn(~/2,K). Suppose E ~ K is an

( n , ~ ) - s e p a r a t e d set and that F (n,~/2)-spans K. Define ~: E ~ F

by choosing for each x E E some point ~(x) E F with

max d(Ti~(x),Ti(x)) ~ ~/2. If ~(x) = ~(y) then

0~i~n-I

max d(Tix,Tiy) ! s/2 + z/2 =

0~i~n-i

so that x = y. Hence ~ is one-to-one and the c a r d i n a l i t y of E is

less than or equal to the c a r d i n a l i t y of F. Therefore Sn(Z,K)

rn(~/2,K).

Finally we show rn(z,K) ~ Sn(~,K). Let E be an ( n , z ) - s e p a r a t e d

subset of K of m a x i m u m cardinality. We claim that E (n,~)-spans

K, since if not 3 x E K
 149

max d(Tix,Tiy)< > z V y CE.

0~i~n-i

Then E U {x} is an (n,z)-separated subset of K, contradicting the

choice of E.

(ii) is obvious. //

Hence the d e f i n i t i o n of h(T,K) makes sense.

Remarks:

(I) hd(T) depends on d.

(2) If K ~ K 1 U ... D K m are all compact then

h(T,K) ~ max h(T,Ki).

l~i~m

Proof: Certainly, Sn(g,K) ~ S n ( ~ , K I) + ... + Sn(S,Km). Fix

> 0. V n choose Kin(e)

S n ( S , K i (~)) : max Sn(e,Kj).

n j

Then Sn(S,K) ~ m . S n ( S , K i (s)) and so,

n

log Sn(S,K) e log m + log Sn(S,K i (~)).

n

Choose n.] ~ - such that

1 1
log Sn (e,K) ~ lim sup ~ log Sn(e,K)
] ]

and so that Ki (e) does not depend on j (i.e., Ki (g) : K(s)

n.] n.]

V j). Thus, ST(~,K) ~ ST(S,K(~)).

Choose ~ ~ 0 so that K(~ ) is c o n s t a n t ( = Ki0 , say). Thus,

h(T,K) ~ h(T,Ki0) ~ max h(T,Kj). //

J
 150

(3) V 6 > 0, in order to compute hd(T) it suffices to take the

supremum of h(T,K) over compact sets of diameter less than 6.

This is true by (2).

(4) If X is compact, hd(T) = h(T,X).

Proof: By (2), if K c X, K compact, then

h(T,K) ~ h(T,X). //

Definition 6.7:

Two metrics d and d' on X are uniformly equivalent if

id.: (X,d) ~ (X,d') and

id.: (X,d') ~ (X,d)

are bo~h uniformly continuous.

In this case, T 6 UC(X,d) iff T E UC(X,d').

Theorem 6.5:

If d and d' are uniformly equivalent and T 6 UC(X,d) then

hd(T) = hd,(T).

Proof: Let ~i > 0. Choose ~2 > 0 )

d'(x,y) < s2 = d(x,y) < ~i

and choose ~3 > 0

d(x,y) < g3 = d'(x,y) < ~2"

Let K be compact. Then

rn(el,K,d) ~ rn(e2,K,d') and

rn(~2,K,d') ~ rn(e3,K,d).

Hence, rT(Sl,K,d) ~ rT(e2,K,d') ~ rT(e3,K,d).

 151

If z I ~ 0, then g2 ~ 0, and ~3 ~ 0 so we h a v e

hd(T,K) = hd,(T,K). //

Remark:

If X is c o m p a c t and if d and d' are equivalent metrics then

they are u n i f o r m l y equivalent. Also, each continuous map T: X ~ X

is u n i f o r m l y continuous.

Theorem 6.8: (Lebessue Coverin$ Lemma)

If (X,d) is a c o m p a c t metric space and ~ is a f i n i t e open

cover of X then there exists a 5 > 0 such that each subset of X

of d i a m e t e r ~ 5 lies in some m e m b e r of =.

Proof: Let = = {AI,...,Ap}. Assume the theorem is false. Then

for all n there exists Bn X such that d i a m ( B n) ~ i/n and Bn

is not contained in any A i. Choose x n ~ Bn and select a subse-

quence {Xni} which converges, say X n i ~ x. Suppose x ~ A.3 ( a.

Let a = d(x,X\Aj) > 0. Choose ni such t h a t n i > 2/a and

d(Xn.,X) < a/2. Then if y ( Bn.

l l

i + ~ a.
d(y,x) _< d(y,Xni) + d(Xni ,x) ~ n i 2 <

So y ~ Aj. Hence Bn. c_ A 4

J' a contradiction. //
l

Theorem 6.7:

When X is c o m p a c t , Bowen's definition of e n t r o p y coincides with

the open cover definition.

Proof: For the d u r a t i o n of this proof let h*(T,a) and h*(T)

denote the numbers that occur in the open cover definition.

Let ~ = { A I , . . . , A p} be an o p e n cover of X. We shall show that

h*(T,~) ~ h(T). Let 5 be a L e b e s g u e number for ~. Let F be a

 152

(n,~/2)-spanning set for X of m i n i m u m cardinality. For z E F

choose A i 0 ( z ) , . . . , A in_l (z) in ~ so that Bs/2(Tkz) ! Aik(Z). Let

C(z) m Al0. (z) N T - I A i I ( Z ) N ... n T-(n-l)A.In_l(Z),

which is a m e m b e r of s v T-is v ... v T-(n-l)s.

We have X = U C(z) since if x ( X 3 z E F

zEF

max d(TZx,Tlz) 5 6/2

0~i~n-i

and hence x ~ T-k(Bs/2(Tkz)) g T-kAik(Z) , 0 ~ k ~ n-l; so x E C(z).

Hence N(s v T-is v ... v T-(n-l)e) ~ IFI = r n ( 8 / 2 , X ) , and

h*(T,a) ~ ~T(5/2,X) ~ h(T,X) : h(T)

since X is compact. Therefore h*(T) E h(T).

To prove the converse let 6 > 0 be given. Choose an open cover

s = {AI,...,A p} of X such that d i a m ( A i) < 8 for all i. Let E

be an ( n , 6 ) - s e p a r a t e d subset of X with maximal cardinality. Two

members of E cannot belong to the same element of

s v T-Is v ... v T - ( n - ! ) s since if

n-i
x,y E (-] T-JA. x,y ~ E
j :0 ~j

then max d(TJx,TJy) < 8 and so x : y.

0~j~n-i

So, N(~ v T-I~ v ... v T - ( n - l ) e ) ~ IEI : Sn(5,X).

Therefore h*(T) ~ h*(T,~) ~ ST(5,X).

Letting 8 ~ 0 we have h*(T) ~ h(T,X) : h(T). //

Notes:

(i) If we had set up the definitions using uniformities we would get

the above for compact Hausdorff spaces.

(2) Since a maximal separated set is spanning we get by the first

part of the proof of Theorem 8.7 that Sn(8/2,X)

N(a v T-l= v ... v T-(n-l)=) where 8 is a Lebesgue number for =.

Theorem 6.8:

(i) If T E UC(X,d) and m > 0 then h(T m) = m'h(T).

(2) Let T i ~ UC(Xi,d i) i = 1,2. Define a metric on XI×X 2 by

d((Xl,X2),(yl,Y2)) = max {dl(Xl,Yl),d2(x2,Y2) }. Then

hd(Tl×T 2) ~ hdl(T 1) + hd2(T2).

If X1 and X2 are compact then equality holds.

Proof: (i). Since r (~,K,T TM) ~ r (~,K,T) we have

n mn

I m
log rn(z,K,Tm) ~ ~-~ log rmn(~,K,T)

and therefore hd(T TM) S m.hd(T).

Since T is uniformly continuous, V e > 0 3 8 > 0 9

d(x,y) < 8 = max d(TJx,T]y) < ~.

0~j~m-i

So an (n,8)-spanning set for K with respect to T TM is also an

(nm,~)-spanning set for K with respect to T. Hence, rn(8,K,Tm)

rmn(~,K,T) so, m.~T(~,K) ~ rTm(5,K). Thus,

m'hd(T,K) ~ hd(Tm,K).

(2). Let Ki ~ Xi be compact, i = 1,2. If Fi is an (n,~)-

spanning set for Ki with respect to Ti then Fi×F 2 is an

 154

(n,e)-spanning set for KIXK 2 with respect to TIXT 2. Hence,

rn(e,KlXK2,TlXT2) ~ rn(e,Ki,Tl)'rn(s,K2,T 2)

which implies

rTlXT2(e,KlXK2) -< rTl-(~,KI) + rT2-(~,K2) "

Therefore

hd(TI×T2,KI×K2 ) -< hdl(Ti,K I) +hd2(T2,K2).

Let ~i: XI×X2 ~ Xi i : 1,2 be the projection map. If

K ~ XIXX 2 is compact then K I : ~I(K) and K 2 : ~2(K) are compact
and K & KIXK 2. Hence

hd(TlxT2,K) ~ hd(TI×T2,KIXK2).

Therefore

hd(TlXT 2) = sup hd(TlXT2,K)

K~XIXX 2
compact

= sup hd(TIXT2,KI×K2 )
KIgX I
K2~X 2
cpt.

sup hdl(Ti,K I) + sup hd2(T2,K 2)

KI~X I K2~X 2
cpt. cpt.

: hdl(T I) + hd2(T2).

Now suppose XI and X2 are compact. Let ~i be an open cover of

Xi and have Lebesgue number 5i (i = 1,2). If Si is a maximal
(n,Si/2)-separated set for Xi with respect to Ti then SIXS 2 is
an (n,6)-separated set for XIXX 2 with respect to TIXT 2 where
 155

8 = min(81/2,62/2). Therefore

Sn(5,Xl×X2 ) ~ Sn(81/2,Xl)'Sn(82/2,X 2)

N(~ 1 v Tila I v ... v T i ( n - l ) a l ) . N ( a 2 v T21a 2 V ... v T 2 ( n - l ) a 2 )

by note 2 above. Hence

1
h(Tl×T 2) _> lim sup ~ log Sn(6,Xl×X2 )
n~-

lim 1 log N(a I v T I I ~ 1 v ... v T i ( n - l ) ~ I)

n~-

lim ~1 log N(~ 2 v T21a 2 v ... v T2(n-1)= 2 )

n~

= h(Tl,a I) + h(T2,~2).

Since al,~2 were arbitrary we get h(TI×T 2) a h(T I) + h(T2). //

Remarks:

(i) If T is a homeomorphism and T { UC(X,d), T -I ( U C ( X , d ) then

hd(T) ~ hd(T -I) in general. We shall show later that if T: R ~ R

is defined by T(x) = 2x then h(T) = log 2 while h(T -I) = 0 using

the usual metric on R. (Note that T has expansion but T -I does

not.) However, on compact spaces h(T) = h(T -I) (Theorem 8.3). This

is because on a compact space T -1 has "as much expansion" as does T.

(2) Equality probably holds in (2) for non-compact X1 and X2 but

I do not know a proof.

§3. Connections with Measure Theoretic Entropy

In this section we shall prove conjecture 3 (assuming conjec-

ture 1 is true) and we shall prove conjecture 1 when X is a

 156

finite-dimensional torus.

Theorem 6.9: (Goodwyn)

Let X be a compact space and T: X ~ X continuous. If m is

a T-invariant regular Borel probability measure on X, then

h (T) ~ h(T).
m

We shall prove this theorem when X is a finite-dimensional

torus since the proof is easier in this case.

Proof: Let X = K k, T: K k ~ K k be any continuous map, and

let m belany T-invariant Borel p r o b a b i l i t y measure on K k. We wish

to show that h (T) ~ h(T).

m
Consider Kk as Rk/z k with metric

d(x+zk,y+Z k) = inf fix - y +vll x~y ( R k

v(Z k

where II'll denotes the usual Euclidean norm.

Fix an integer q ~ 0. Consider a d e c o m p o s i t i o n of the unit

k-cube in Rk into all sets of the form

Pl Pl +I P2 P2 +I
(x I ..... Xk): 2-~ ~ x ! < 2q ' 2-~ ~ x 2 < - - 2 q '

Pk Pk +I
where
2"--~"~ xk < 2q

0 ~ Pi < 2q for i : l,...,k }.

This induces a partition of the torus Kk which we denote by

~q = {A l,...,A2kq}. Any ball in Kk of radius 2 q+21 intersects at

most 2k members of ~q. Let Y = {CI,...,C s} be a cover of Kk by

open balls of radius 2--~+

2. Let 25 be a Lebesgue number for Y.

For all x ( K k, fix some C(x) E Y with B6(x) _c C(x). Let F be

 157

an (n,5)-spanning set for Kk with respect to T of minimal cardi-

n-i
nality. Let x E ~ T-JA i , Ai. E ~q. Choose y E F with
j:0 j ]

max d(TJx,TJy) ~ 5,
Osj~n-i

and hence, T3x E C(T]y). Thus T]x E Ai. n C(TJy). Hence if

]

n-i
U n : {(i 0 .... ,in_l): N T-JAi. @ ¢}
j=0 ]

then IUnl < 2knIF I : 2knr (5,Kk). So, using Corollary 4 . 2

-- n

Hm(~ q v T - l ~ q v... v T-(n-l)~q) ~ log IUnl ~ n'log 2k + log rn(5,Kk).

1
Thus, hm(T,~ q) _< log 2k + lim ~ log rn(8,Kk)
n

= log 2 k + rT(6,K k)

_< log 2 k + h(T) = k + h(T).

(Note that we are taking logarithms to base 2.)

But, A(~q) 2 B as q ~ -. So, by Theorem 4.14

hm(T) = lim hm(T,~ q) _< k + h(T).

q~®

But this holds for any continuous T, so, in particular for Tn

n> 0. If n > 0

hm(T) = ~1 hm(Tn ) _< ~i [k + h(Tn)] = _k + h(T)

n

so, by letting n ~ ® we get the desired result. //

 158

Theorem 6.i0: (Bowen)

Let X be a compact metric group and T: X ~ X, T : a'A an

affine transformation. If m denotes Haar measure on X then

hm(T) = h(T) = hm(A) = h(A).

Proof: By the previous theorem hm(T) _< h(T), and so it remains

to prove that h(T) _< hm(T). Suppose d is a left invariant metric

on X. Let Be(y) = {x: d(x,y) < s} and

n-I
D (x,e,T) : ~ T-kB (Tkx).
n k:0 e

By induction we shall show that

T-kB (Tkx) : x.(A-kBe(e)).

It is true for k : 0 by the invariance of the metric d. Assuming

it holds for k we prove it for k+l.

T-(k+l)B (Tk+ix) = T-l(T-kB (Tk(Tx)))

g

= T-l(Tx.A-kB $ (e))

: x.(A-(k+l)B (e)).
g

n-i
Hence, Dn(X,s,T) = x • ~ A-kB (e) : x'D (e,e,A).
k=0 n

Also, m(Dn(X,e,T)) : m(Dn(e,e,A)).

Let > 0. Let ~ = {AI,...,A n} be a partition of X into Borel

n-i n-i
sets of diameter < e. If x ~ ~ T-kA. then N T-kA" c_
k=0 lk k=0 ik
n-1
X.Dn(e,s,A) , since if y ~ ~ T-kA. then Tk(x),Tk(y) ~ Aik V k,
k=0 lk

and hence y ~ T-kBe(Tkx) V k, i.e., y ~ Dn(X,~,T) = X.Dn(e,s,A).

 159

n-i
Thus, m(k=0
~ T-kAik) -< m(Dn(e'e'A)) and taking logs we see that

n
T m(N T-kAik) log m ( N T - k A i k )
i0,. • • ,in_l--i

n
m(~ T-kA. ) log m(Dn(e,e,A))
i0,...,in_l:l ik

= log m(Dn(e,s,A)).

1
Thus, hm(T) >_ hm(T,~) : lim ~ H(~ v ... vT-(n-l)~)
n

1
>- lim sup [- ~ log m(Dn(e,~,A))] .
n

Hence, since ~ was arbitrary we obtain that

1
hm(T) ~ lim lim sup [- ~ log m(Dn(e,s,A))] .
~0 n

(The limit clearly exists.) Consider now an (n,g)-separated set with

respect to T, E N X, having maximal cardinality. Then

U D (x,g/2,T) = U x.D (e,~/2,A)

xEE n x~E n

is a disjoint union because of the choice of E. Therefore

Sn(S,X).m(Dn(e,~/2,A)) e i

and so s (~,X)
n
m(Dn(e,¢/2,A))

1
Therefore ST(S,X) _< !im sup [- n log m(Dn(e,~/2,A))]
n

and letting ~ ~ 0 we see that

 160

1
h(T) : hd(T,X) ~ lim lim sup [- ~ log m ( D n ( e , z / 2 , A ) ) ]

hm(T).

1
Thus, hm(T) = h(T) = lim lim sup [- ~ log m(Dn(e,z~A))].
~0 n

We can replace T by A here since the right hand side is independ-

ent of a. //

Note:

The formula

1
h(T) = lim lim sup [- ~ log m(Dn(e,~,A) )]
~0 n

illustrates that T measures "the amount of expansion" in T.

§4. T o p o l o $ i c a ! Entropy o f Linear Maps and Total Affines

Our aim in this section is to compute the t o p o l o g i c a l entropy

(and hence by T h e o r e m 6.10 the measure t h e o r e t i c entropy) of affine

transformations of f i n i t e - d i m e n s i o n a l tori. Recall (see §5 of Chap-

ter 0) that we can view the n-torus Kn either m u l t i p l i c a t i v e l y as

KxKx...xK (n factors) or a d d i t i v e l y as Rn/z n. Each e n d o m o r p h i s m A

of Kn onto Kn is given, in the a d d i t i v e notation, by

A(x + Z n) = [A].x + Z n x ~ R n,

where [A] is an nxn n o n s i n g u l a r m a t r i x with integer entries. [A]

determines a linear t r a n s f o r m a t i o n A of Rn and nA = An where

~: R n ~ K n is the natural p r o j e c t i o n given by ,(x) = x + Z n.

Let If'If denote the usual Euclidean n o r m on R n. We define a

metric d on Rn/z n by
 161

d(x+zn,y+Z n) : inf llx - Y +vll x,y ( R n.

v(Z n

d is left and right invariant and, for every x ( Rn n maps the

ball of radius 1/4 about x in Rn isometrically onto the ball of

radius 1/4 about ,(x) in Rn/z n •

The next theorem deals with such a situation and asserts that

hd(A) = h~(A) in this case, where d denotes the metric on Rn in-

duced by the Euclidean norm II'II. (Since llAx-Ayll ~ llAll.llx-yll we

know A E UC(Rn,d).) This will reduce the problem of calculating the

entropy of A to that of calculating the entropy of A.

Theorem 6.11:

Let (X,d),(X,d) be metric spaces and ~: X ~ X a continuous

surjection such that there exists 6 > 0 with

~IBs(~): Bs(x) ~ B6(~(~))

an isometric surjection for all x ( X. If T (UC(X,d) and

(UC(X,d) satisfy ~T = T~ then

ha(T) = h~(T).

Proof: If K is compact in X of diameter < 6 then n(K) is

compact in X of diameter < 6. Every compact subset of X of di-

ameter < 6 is of this form. Let ~ > 0 be such that ~ < 5 and

if d(x,Y) < ~ then d(Tx,Ty) < 8.

Suppose E & K is an (n,g)-separating set with respect to T.

We first prove that n(E) is an (n,s)-separating subset of n(K)

with respect to T. To prove this, let x ~ y belong to E. Then

~(x) # "(9). Let i0 be chosen so t h a t d(Tix,T19) ~ s if i ~ i0

• i0+i
and d(Tl0+Ix,T y) > ~. (This can be done since E is an (n,s)-

separating set with respect to T.) By our choice of s,

 162

i0+l ia+l " ~i0+l

d(T x,T ~ 9) < 6 and so Tl0+l 9 E Bs(T x) which is mapped
i0+l
isometrically onto B6(T ~(x)). So,

i0+l i0+l • .
d(T ~(x),T .(y)) : d(T10+Ix,T10+ly) > e.

Thus ~(E) is (n,s)-separated with respect to T. Therefore,

Sn(S,K,T) ~ Sn(~,~(K),T).

To prove the converse inequality, suppose E is an (n,e)-

separated subset of .(K) g X with respect to T, where K is com-

pact and of diameter < 5. Let E = .-I(E) n K. Then E is an (n,s)-

separated set with respect to T since if ~(~i~,~ig) E s where

x,y { E then d(Ti~(x),Ti~(y)) ~ ~. Hence,

s (~,~(~),T)
n _< s
n (~,~,~).

Therefore Sn(e,K,T) : Sn(S,.(K),T)

and hence h~(T,K) = hd(T,.(K)).

d

By remark (3) of §2

h~(T) : hd(T). //
d

Corollary 6.11:

If A: K n ~ K n is an endomorphism then hd(A) = h~(A) where

is the linear map of Rn covering A, d is the metric on Rn de-

termined from the Euclidean norm and d is any metric on K n.

We shall now proceed towards calculating the entropy of a linear

map of Rn .
 163

Theorem 6.12:

Suppose A: Rp ~ Rp is a l i n e a r map, and p a metric determined

by a n o r m on R p. Then:

(i) hp(A) ~ log Idet A I if det A ~ 0, and

(ii) if all the e i g e n v a l u e s of A have the same a b s o l u t e value

then

h (A) : m a x {0,p'log ~}.

P

Proof: All n o r m s on Rp are uniformly equivalent, so, by T h e o -

r e m 6.5 we can a s s u m e that p is the m e t r i c given by the E u c l i d e a n

norm. Obviously, A (Uc(RP,p) as

p(Ax,Ay) = IIAx-AyJI-< llAlllIx-yll : IIAIIp(x,Y).

(i). Let m denote Lebesgue measure on R p. Then

m(A(E)) = Idet AI.m(E)

for all Borel sets E _c R p. Let K c Rp be c o m p a c t and m(K) > 0.

If F (n,~)-spans K then K _c U Dn(X,~,A) : [J [x + D n ( 0 , ~ , A ) ]

x(F x(F
n-i
where D (x,s,A) = ~ A - i B (Aix) (as in the p r o o f of T h e o r e m 6.10).
n i=0 s

Thus, m(K) ~ m(Dn(0,s,A)).rn(s,K) : m(Dn(0,s,A)).rn(s,K)

re(K)
i.e., r (~,K)
n m(Dn(0,s,A))

Therefore ~A(S,K) ~ lim 1 [log m(K) - log m ( D n ( 0 , s , A ) ) ]

n

= lim [- ~i log m(D n (0,s,A))].

n

m ( B g (0))
But, m(D (0,s,A)) ~ m(A-(n-l)B (0)) =
n s idet A i n - i
 16~

1
so t h a t ~A(~,K) >_ l i m ~ [log Idet AI n-I - l o g m ( B (0))]
8
n

: log Idet AI.

Therefore hp(A) a h (A,K) £ log Idet A I .

(ii). In this case,

Idet A I : Iproduct of e i g e n v a l u e s I = ~P.

So, by (i) h (A) ~ p log • and then, clearly,

P

h (A) ~ m a x {0 , p log ~}.

P

We now have to show the o p p o s i t e inequality. Assume first that

IIAII > i. Let K be a c o m p a c t subset of Rp of d i a m e t e r < 1/2.

Choose b ~ Rp such that 0 E b + K ~ K b. The d i a m e t e r of K b < 1/2

P
so that Kb c I1 where

P
I 1 - { ( x I ..... Xp) E RP: Ixil _ 1 V i}.

For 8 such that 0 < 8 < 1 let

F(8) m {(n18 ..... npS): n i E Z, Ini8 I < 2}.

Observe that IF(8)I _< (5/8) p, and 3 c > 0 ~ V y ~ I~, 3

x E F(8) 9 p(x,y) < c8. F(8) is an (n,NAllncs)-spanning set for

Kb with respect to A, since if y ~ Kb 3 x (F(8)

p(Aix,Aiy) ~ llAillp(x,y) ~ llAIlic6 ~ IIAllncs,

for 0 ~ i ~ n-l.

But then, F(8) - b is an (n,llAllncs)-spanning set w i t h r e s p e c t to A

g
for K. Let s > 0, and set 8 - - - < I, for n sufficiently
llAlln.c
large. Thus
 165

for sufficiently large n. Also,

1
rA(~,K) : lim ~ log rn(¢,K)
n

lim ~ [log 5 + n log IIAII + log c - log ~]

n

: p log lIAll.

So, h (A) _< p log IIAII

P

and h e n c e h (A) ~ max {0 , p log IIAII} if IIAII > i.

P

If NAIl ~ 1 then h (A) : 0 since a (l,s)-spanning set is an (n,e)-

P
spanning set. Thus, in all cases

h (A) e max {0 , p log IIAII}.

p

However, h (A) : ~ h (A n ) for n > 0,

p n p

1 max
_< -- {0 , p log llAnll}
n

: max {0 , p log IIAnIIl/n}.

But, llAnllI/n ~ the spectral radius of A, which here is p r e c i s e l y ~.

Therefore

h (A) _< m a x {0 , p log ~}. //

P

Remark:

If A: Rp ~ Rp is l i n e a r and the m e t r i c p is d e t e r m i n e d by a
1
norm then hp(A) = lim lim sup [- ~ log m ( D n ( 0 , ~ , A ) ) ] where m is
8~0 n~=
 166

n-I
Lebesgue measure on Rp and D (0,s,A) :~ A-IB (0).
n i:0

Proof: We can suppose p is determined by the Euclidean norm.

In the proof of (i) of Theorem 6.12 we showed

rA(S,K) Z lim sup [- E1 log m(D n (0,s,A))] and hence

n~

1
hp(A) e lim lim sup [- ~ log m(Dn(0,~,A))]. Let K be the p-cube
~0 n~® q
in Rp with center 0 and side length 2q. If E is an (n,s)-

separated subset of Kq then U Dn(X,g/2,A) is a disjoint union

xEE

and ~J Dn(X,~/2,A) : U x + D (0,s/2,A) c Kq+2 . Hence

xEE xEE n -

Sn(g/2,Kq)-m(Dn(0,s/2,A)) ~ (q+2~) p

and hence 1
SA(S/2,Kq) -< lim sup [- ~ m(Dn(0,~/2,A))].
n-~

Therefore hp(T) : sup hp(Kq) ~ lim lim sup [- ~1 m(Dn(0,~/2,A)) ] //

q e~0 n~ =

Theorem 6.13:

Suppose A: R p ~ R p is linear and p is a metric coming from a

norm. Then

h (A) : E log ;~il

P i~i1>l

where kl,...,k p are the eigenvalues of A.

Proof: By the Jordan D e c o m p o s i t i o n Theorem (Jordan Canonical

Form), we can write Rp as a direct sum of subspaces

R p : E 1 @ ... @ E k

where A(E i) g E i for i : l,...,k and A i = AIE" has all its

l
 167

e i g e n v a ! u e s with the same n o r m ~i" Thus

A = A I @ ... • A k

k
and hp(A) ~ [ h (A i)
i=l P

by use of T h e o r e m 6.8 and (since 6.8 is stated in terms of a specific

metric) the fact that all norms on Rp are equivalent. By Theo-

rem 6.12

k
hp(A) ~ i=l~ max {0 , dim E..l log ~i }

= ~ (dim E. • log :.)

•>I l l
l

: K log J~iJ"
J~i1>l

We can suppose p is d e t e r m i n e d by the E u c l i d e a n norm. By the

1
above remark we have hp(A) = lim lim sup [- ~ log m(Dn(0,s,A))]. Write
s~O n~
Rp as a direct sum of two subspaces Rp = F 1 @ F 2 so that AF i ~ F i

(i = 1,2) and A i = AIF I has eigenvalues with absolute value greater

than one and A2 : AIF 2 has eigenvalues with absolute value less than

or equal to one. Since D n ( 0 , ~ , A ) _c Bs(0 ) n A-(n-l)Bs(0) we have

m(Dn(0,~,A)) ~ csldet A1 (n-l)] for some c independent of n.

i
Therefore lim sup [- : m(D n (0 's,A)] > log - ]det All = ~ log Ikil.
n~- Ixil>1
Therefore

h@(A) ~ X log Ikil. //

i×iI>l
 168

T h e o r e m 6.14:

Suppose T: K p ~ K p is an affine t r a n s f o r m a t i o n , Tx : a.A(x)

where a ~ Kp and A is a surjective e n d o m o r p h i s m of Kp . If m

is Haar measure, then

h(T) : hm(T) : hm(A) = h(A) : log i~il,

l~il-l
the ki'S being the eigenvalues of the m a t r i x [A] which repre-
sents A.

Proof: We know by T h e o r e m 6.10 that

h(T) = hm(T) = hm(A) = h(A)

and by C o r o l l a r y 6.!1 that h(A) = h(A), where A denotes the cover-

ing linear map of A. h(A) is c a l c u l a t e d in T h e o r e m 8.13. //

Note:

We have given a full proof of this result when the space is a

f i n i t e - d i m e n s i o n a l torus (since we proved T h e o r e m 6.9 only in this

case). The above proof is due to Bowen. This formula for the entropy

of an a u t o m o r p h i s m was first stated by Sinai [I].

95. Expansive H o m e o m o r p h i s m s

As an analogue of the m e a s u r e t h e o r e t i c concept of a generator,

one could make the following definition:

Let (X,d) be a compact metric space, and T: X ~ X a homeo-

morphism.

D e f i n i t i o n 6.8:

A finite open cover u of X is a s e n e r a t o r (weak generator)

for T if for every bisequence {An } of members of u,

 169

T-n2 is at most one point

n
n_-_~

5 T-nA n is at most one point ).

These c o n c e p t s are due to Keynes and Robertson [i].

Theorem 6.15:

T has a g e n e r a t o r iff T has a weak generator.

Proof: (=) is trivial.

(-). Let ~ be a weak g e n e r a t o r for T,

: {BI,...,Bs} ,

and let 6 be a Lebesgue number for 6" Let a be a finite open

cover by sets Ai having diam(A i) ~ 5. So if Ai is a bisequence

n
in a then V n 3 Jn ) Ain ~ Bjn. Hence,

T-n~i c_ 6 T-nBjn
-~ n -~

which is either empty or a single point. So a is a generator. //

The f o l l o w i n g shows that a generator d e t e r m i n e s the topology

on X.

Theorem 6.16:

Let a be a g e n e r a t o r for T. Then V ~ > O 3 N > 0 ) each

N
set in V T-na has diameter < ~. Conversely, V N > 0 3 ~ > 0
-N
such that d(x,y) < s implies
N
x,y ~ ~ T-nA
-N n

for some A_N,...,A N E a.

 170

Proof: Suppose the first part of the t h e o r e m does not hold.

3 ~ > 0 ) V j • 0 3 xj,yj, d(xj,yj) • ~ and 3 Aj, i ~ ~,

-j ~ i ~ j with xj,yj E A
i:-j
T-IA. ..
],l
We can s u p p o s e that x. ~ x~
]
yj ~ y since X is c o m p a c t , and h e n c e x ~ y. Consider the sets

Aj, 0. Infinitely m a n y of t h e m c o i n c i d e since e is finite. Thus

xj,yj E A0, say, for i n f i n i t e l y many j and h e n c e x,y E A0"

Similarly, for e a c h n, infinitely many A. coincide and we ob-

3,n
rain An ~ ~ with x,y ~ T-nAn . Thus,

x,y E 5 T -n~
n

contradicting the f a c t that e is a g e n e r a t o r .

To p r o v e the c o n v e r s e let N • 0 be given. Let 8 • 0 be a

Lebesgue number for 5. Choose e > 0 such that d(x;y) < ~ implies
• Tiy) < 8
d(Tlx, for -N ~ i ~ N. Hence if d(x,y) < ~ and Ill N N

then Tlx,Tly E Ai for some A i E e. Hence

N
x,y ~ ~ T-IA.. //
-N 1

The~analogue of the K o l m o g o r o v - S i n a i Theorem is:

Theorem 6.17:

If a is a g e n e r a t o r for T then

h(T) : h(T,e).

Proof: Let ~ be any o pen cover. Let 5 be a L e b e s g u e number

N
for ~. Choose N • 0 so that each m e m b e r of V T-na has d i a m e t e r
N -N
< 8. Then ~ < V T-n~, and so,
-N
 171

N
h(T,~) _< h ( T , V T-no)
-N

n-1 N
: lim 1H(V T-i( V T-na))
k-~® i:0 -N

N+k-I
= lim 1 H( V T-no)
k~® -N

2N+k-i
= lim 1 H( V T-na)

2N+k-i
= lim 2N+k-i 1 H( V T-no)
k 2N+k-i

: h(T,a).

Therefore, h(T,~) ~ h(T,a) for all open covers 8" Hence

h(T) = h(T,a). //

Remark:

The same result holds for weak generators.

Generators are connected with the notion of expansive h o m e o m o r -

phism, which was studied long ago.

Definition 6.10:

A homeomorphism T: X ~ X is expansive if 3 8 > 0 9 if

x ~ y then 3 n ( Z ) d(Tnx,Tny) ~ 5. We call 5 an expansive

constant for T.

Remark:

A n o t h e r way to define an expansive h o m e o m o r p h i s m is as follows.

Consider X×X with T×T acting on it. Define a metric D on X×X

by D((u,v),(x,y)) = max {d(u,x),d(v,y)}. Then T is expansive

3 8 > 0 such that if (x,y) is not an element of the diagonal~

 172

some power of T×T takes (x,y) out of the 5 - n e i g h b o r h o o d of the

diagonal.

The f o l l o w i n g t h e o r e m is due to Reddy, and Keynes and Robertson.

T h e o r e m 6.18:

T is e x p a n s i v e iff T has a g e n e r a t o r iff T has a w e a k gener-

ator.

Proof: By T h e o r e m 6.15 it suffices to show T is e x p a n s i v e

iff T has a generator.

(=) Let 6 be an expansive constant for T and ~ a finite

cover by open balls of radius 6/2. Suppose x~y E A T-nAn where

A n E =. Then, d(Tnx,Tny) ~ 5 V n, so, by a s s u m p t i o n x = y.

Therefore = is a generator.

(=) Conversely, suppose = is a generator. Let 5 be a Le-

besgue n u m b e r for =. If d(Tnx,Tny) ~ 6 V n then V n 3 A E

n

Tnx,Tny ~ A and so,

n

x,y ~ 6 T-nA
n

which is at most one point. Hence x : y and T is expansive. //

Corollary 6.18:

(i) Expansiveness is independent of the metric (however, the expan-

sive c o n s t a n t does change).

(2) T is expansive iff Tk is expansive, k ~ 0.

(3) Expansiveness is a t o p o l o g i c a l c o n j u g a c y invariant.

Proof: (I) This is trivial, since having a g e n e r a t o r has

nothing to do with the metric.

(2). If = is a generator for T then

a vT-I= v ... v T - ( k - l ) ~
 173

is a g e n e r a t o r for T k. If e is a generator for Tk then e is

also a g e n e r a t o r for T.

(3) is trivial. //

The n e x t result shows how to find measure t h e o r e t i c generators

for e x p a n s i v e homeomorphisms.

Theorem 6.19:

Let T be e x p a n s i v e with constant 6. If

= {Cl,...,Cs}

is a p a r t i t i o n of X into Borel sets of d i a m e t e r < 6, then ~ is a

measure t h e o r e t i c g e n e r a t o r for any T-invariant Borel p r o b a b i l i t y

measure.

Proof: Let Ci be a bisequence of members of ~. If

n
Qm

x,y E ~ T-nCi then Tnx,Tny ~ Ci for all n, and hence

-~ n n
d(Tnx,Tny) < 8 V n. By e x p a n s i v e n e s s x = y. Thus ~ T-nCi = ¢
--~ n

or : one point. Hence

V TnA(~) : 8. //

Thus, for e x p a n s i v e h o m e o m o r p h i s m s there are many m e a s u r e theo-

retic generators.

Examples:

(i) Isometries are never expansive except on finite spaces.

(2) Let A be an a u t o m o r p h i s m of the n-torus, and [A] the corre-

sponding matrix. Then A is expansive iff [A] has no eigenvalues

of modulus i.

Sketch of proof: One first shows that A is expansive iff the

linear map A of Rn that covers A is expansive. Then show that

 174

is expansive iff the complexification of A is expansive. Then

one shows that the complexification of A is expansive iff the trans-

formation given by the Jordan normal form is expansive. Lastly, one

shows that the normal form is expansive iff there are no eigenvalues

of modulus i.

(Note: By Theorem 6.19, any partition of Kn into sufficiently small

n-rectangles is a measure theoretic generator for an expansive auto-

morphism of Kn.)

(3) The two-sided shift on k symbols is expansive.

Proof (i): Let the state space be {0,1,...,k-l}. Let

A i = <{Xn}: x 0 : i}, i : 0,1,...,k-l. Then A 0 U A 1 U ... U Ak_ 1 : X

and each Ai is open. ~ = {A0,...,Ak_ I} is a generator for the

shift since if x E 5T-nAi where the Ai E ~ then

-~ n n

x = (...,i_2,i_l,i0,il,i2,...).

We then use Theorem 8.18. //

Proof (2): Let d be the metric given by:

IXn - ynl
d({Xn}, {Yn})
n=--
~ 2 Inj

Suppose {x n} ~ {yn }. Then for some no, Xn0 ~ Yn0 and

d(Tn0 {Xn},T n0 {Yn }) = _~ 1 iXn+n0

n=-- 2 Inl Yn+n01

IXn0 - Yn0 I ~ 1.

Thus 1 is an expansive constant. //

Remarks:

(a) If T: X ~ X is expansive and Y is a closed subset of X with

 175

TY = Y then TIy is expansive.

(b) If T!: X I ~ XI, T2: X 2 + X 2 are expansive then so is

TI×T2: XI×X 2 ~ XI×X 2. Any finite direct product of e x p a n s i v e homeo-

morphisms is expansive~ but infinite products are not.

(c) If T: X ~ X is expansive and S: Y ~ Y is a h o m e o m o r p h i s m

with
T
X ,X

Y ,Y
S

commutative for ~: X * Y a continuous surjection then S need not

be expansive. S is expansive if ~ is a k-to-one covering map. So

expansiveness is not p r e s e r v e d under the operation of taking factors.

Example: (Parry and Walters)

C o n s i d e r the 2-torus K2; identify (z,w) with (z,w). The

map ~ is a 2-to-one covering map except at four points:

(i,i), (-i,I), (-i,-i), and (I,-i).

Let A: K 2 ~ K 2 be an automorphism. The q u o t i e n t space is S2 and

A induces a h o m e o m o r p h i s m on S2 which can be shown to be non-

expansive.

A n o t h e r example can be c o n s t r u c t e d as follows:

Let Tz = az be a minimal r o t a t i o n of K. We shall r e p r e s e n t T

as a factor of a subset of the t w o - s i d e d shift on two symbols. Con-

sider the cover of K by the closed intervals (arcs) between -i and

i on K. Call one of them A0 and the other A I.

If z ~ Kk{an,-an: n E Z} we can uniquely associate

Al

a member of ~ {0,1} to z by z ~ {an}i. if

Tnz ~ A a Let A denote the subset of ~ {0,1}

n -®
 176

arising in this way. The m a p

¢: A ~ K \ { a n , - a n : n E Z}

defined above satisfies ¢S(x) = T¢(x), x E A where S denotes the

shift, and if we can show ¢ is u n i f o r m l y continuous then ¢ extends

uniquely to a c o n t i n u o u s map ,: A ~ K with ~S = T~. Suppose ~ > 0

is given. Choose N > 0 so t h a t { l , a i l ~ a ±2 ,...,a ±N} is ¢ / 2 - d e n s e

in K. Suppose {bn} and {Cn} are two members of A such t h a t

bn = cn for Inl ~ N. We t h e n have to s h o w d(¢({bn}),¢({Cn}) ) < ~.

Let x = ¢({bn}) and y = ¢({Cn}). The assumption bn = Cn, Inl _< N

means that anx and any belong to the same element of the cover for

Inl E N. If y : -x then this clearly cannot happen. So s u p p o s e the

counter-clockwise distance from y to x is s m a l l e r than the clock-

wise distance. For some n with Inl ~ N anx is in the o p e n

interval of l e n g t h e starting at 1 and going counter-clockwise.

Hence any must also be in the u p p e r half of the c i r c l e and by the

assumption about the relative positions of x and y, any must be

between 1 and anx. Hence d(anx,any) < a and so d(x,y) < ~.

We shall now show that every expansive homeomorphism is a f a c t o r

of a s u b s e t of a t w o - s i d e d shift.

Theorem 6.20:

Let T: X ~ X be an e x p a n s i v e homeomorphism. Then 3 an inte-

ger k > 0, a closed subset ~ of

Xk = ~{0,1,...,k-l}

such t h a t ~ = 8, where a is the shift on Xk, and a continuous

surjection ~: ~ ~ X such that

,a(y) : T~(y) y E 8.
 177

Proof: The proof will resemble that of the p r e c e d i n g example.

Let 6 be an e x p a n s i v e constant for T. Choose a cover = =

{A0,...,Ak_ I} by closed sets such that diam(A i) < 6 V i and so

that the Ai i n t e r s e c t only in their boundaries. Let D denote the

union of the b o u n d a r i e s of the A i. Then D = 0 TnD is a first

c a t e g o r y set and so X\D is dense in X. For each x E X\D we can

assign, uniquely, a m e m b e r of Xk by x ~ {an}i. iff Tnx ~ A a

n
Let A denote the c o l l e c t i o n of all sequences arising in this way.

If ¢: A ~ X k D is the map defined above then ¢~(y) : T¢(y) V y ~A

and if we can show ¢ is u n i f o r m l y continuous it will then follow

that ¢ can be u n i q u e l y extended to a continuous map ~: A ~ X such

that ,~(y) = T~(y) V y ~ A.

Let z > 0 be given. Choose N > 0 so that each m e m b e r of

N
V Tne has d i a m e t e r less than ~, which can be done by T h e o r e m 6.16,
-N
since we can enlarge each Ai to an open set to o b t a i n a g e n e r a t o r

(remembering d i a m ( A i) < 6). If {an},{b n} E A and an = b n for

N
Inl ~ N then ¢({an}), ({bn}) are in the same m e m b e r of V Tn~
-N
and so d(¢{an},¢{bn}) < ~. Hence ¢ is u n i f o r m l y continuous. //

The p e r i o d i c points of T are associated with a g e n e r a t o r as

follows:

T h e o r e m 6.21:

Let T: X ~ X be expansive and let ~ be a g e n e r a t o r (or a weak

generator). Then Tkx = x iff

x = A Tk'i(Ao n TA1 n . . . n Tk-lAk_l )

where A. E e j = O,...,k-l.
]
 178

Proof: Suppose Tkx = x. Since avT= v ... v T k - l = is a c o v e r

of X, x E A 0 N TA 1 0 ... N T k - I A k _ l for some c o l l e c t i o n of A.'s

3
in =. Thus

x E T k ' i ( A 0 N T A 1 Q ... N T k - I A k _ l ) V i,

and this implies

x Efi T k ' i ( A 0 A TA 1 A . . . n Tk-lAk_l )

which is at m o s t one point. Therefore,

x = fi Tk'i(A0 N TA1 N . . . n T k - l A k _ l ).

The c o n v e r s e is t r i v i a l . //

The f o l l o w i n g gives an e s t i m a t e on the n u m b e r of f i x e d points

of Tn .

Corollary 6.21:

If T is e x p a n s i v e and a is a g e n e r a t o r for T with M mem-

bers, then

Nn(T) : I{x: T n x : x} I ~ M n (n > 0).

Topological entropy is c o n n e c t e d to p e r i o d i c points by

Theorem 6.22:

If T is e x p a n s i v e , then

h(T) ~ ~(T) = lim ! log N (T).

n~
n n

n-I
Proof: Let a be a g e n e r a t o r for T. Each e l e m e n t of V T-la
0
contains at m o s t one p o i n t f i x e d by T n, since if Tnx = x, Tny = y
 179

n-I °
and x,y ~ ~ T-IA. then
i=0 3i

x,y ¢ 6
k:--
T-nk(Aj0 n T-IAjl n ... n T-(n-I)A.
]n-i
)

which is at most one point. Therefore x : y. Thus,

N (T) ~ N(e vT-la v... vT-(n-l)a)

n

which implies that

1 log Nn(T) ~ ~1 H(a vT-l~ v ... vT-(n-l)~)

1
SO~ lim ~ log Nn(T) _< h(T,~) : h(T).
n-~

Therefore ~(T) ~ h(T). //

Consider our examples:

(!) Let A be an expansive automorphism of the torus Km.

Nn(A) = l{x: Anx = x}l = IKernel (An-I)l

= Idet ([A] n - I) I (by the proposition below)

: 1 7F (~?- :71
i

Where the ki are the eigenvalues of the matrix [A]. So,

~(A) : lim i ~- (log Ik~ - ll).

n 1
n~- k.
l

If Ikil > 1 then

= 1 n
1 iog iiXn - i I K [log Ikil + log Ii - k:nl]

log Ikil + 0 : log Ikil.

 180

If Ikil < 1 then

1 ×n-ll ~ 0.

So, ~(A) = ~ log Ikil = h(A) by T h e o r e m 6.14.

lql>l

Therefore, in this case we have equality.

(2) If T is the t w o - s i d e d shift on k symbols then

N (T) = kn and ~(T) = log k = h(T),

n

so that here too we have equality.

In these examples e(T) and h(T) coincide. However, this is

not true for all e x p a n s i v e homeomorphisms as there are examples of

minimal expansive homeomorphisms (e(T) = 0 since m i n i m a l i t y implies

there are no p e r i o d i c points when the space is infinite) with positive

topological entropy (due to Furstenberg).

Problem:

If T is e x p a n s i v e and has a dense set of p e r i o d i c points then

is it true that e(T) = h(T)?

Bowen has shown this to be true u n d e r the s t r o n g e r assumption

that T is an a x i o m A* h o m e o m o r p h i s m .

In example (1) we used the following:

Proposition:

If B: K n ~ K n is an e n d o m o r p h i s m of Kn onto Kn with corre-

sponding matrix [B], (so det [B] ~ 0) then S is a Idet [B]I-to-

one map.

(We used the case B = An - I where [A] has no roots of unity

as e i g e n v a l u e s . )
 181

Proof: [B] is an invertible matrix since det [B] ~ 0. Thus,

we can write [B] = EIE2...E n where the Ei are elementary matrices.

Since [B] has integer entries, the Ei have rational entries. Each

Ei is one of the following forms:

(i) I with two rows interchanged,

(2) I with one row multiplied by c E Q, or

(3) I with the j-th row replaced by j-th row + c(k-th row), c E Q.

For each Ei, choose ei E Z such that e.E.l

l has integer entries.

In case (i), eiE i induces an leiln-to-one map of K n, i.e.., an

leilnldet Ell-to-one map. In case (2), eiE i induces an lei!nlcl-to -

one map of K n, i.e., an le ilnldet E il-tO-one map. In c a s e (3),

eiE i induces an le ilnldet E il-to-one map. Let Ci be the endomor-

phism of Kn determined by eiE i. Let B' be the endomorphism de-

termined by el...er[B]. Then B' : ClO ... o C r. If B' is a

b'-to-one map and B is a b-to-one map then since B' = C l O ... o Cr

r
b' : ~[ leilnldet Ell
i=l

r r
: ~[ let In -~ Idet Ell
i=l i=l

r
: ~- leiln}det [B]I.
i=l

r
But, also b' : -~ let In" b
i=l

since [B'] = el...er[B] , so

b = ldet [B]I. //

Consider ~(T) : lim ! log N (T).

n~- n n

e(T) is connected with the ~-function which was introduced (for

 182

diffeomorphisms) by Artin and Mazur [1]:

If T: X ~ X is a homeomorphism such that N (T) < - for all

n
n > 0 then we set

~T(Z) : exp ( A ~! znNn (T) ] , z E C.

Note that

~(T) = log radius of convergence of ~T

(Artin and Mazur showed that "most" diffeomorphisms have a ~-function

with positive radius of convergence, and Smale suggested that the

~-funotion might be a rational function of z for "most" diffeomor-

phisms. Manning [i] has shown this to be the case for axiom A diffeo-

morphisms but results of Simon [i] have answered Smale's conjecture

negatively (for K3).)

It is known that there are no expansive homeomorphisms of S I,

but not known if there are any on S 2. It seems reasonable to ask

whether a compact metric space admitting an expansive homeomorphism

is finite-dimensional.

95. Examples

We consider the topological entropy of some examples.

(i) Isometrics have zero entropy. This is clear from Bowen's defi-

nition. Hence rotations on compact metric groups and all topologi-

cally transitive homeomorphisms with topological discrete spectrum

have zero entropy (Theorem 5.8).

(2) The two-sided shift on k symbols has entropy log k. This is

proved by considering the obvious generator.

(3) Any homeomorphism of K has zero topological entropy.

 183

Proof: Let T: K ~ K be a homeomorphism. T maps intervals to

intervals as the intervals are the connected subsets. Suppose the

circle has length i.

Choose s > 0 such that

d(x,y) ~ s = d(T-Ix,T-ly) ~ 1/4.

Consider spanning sets for K with respect to T. Clearly,

rl(S,K) ~ [I/¢] + l, where [.] denotes the least integer function.

We estimate rn(s,K).

Suppose we have an (n-l,z)-spanning set F of minimal cardi-

nality rn_l(S,K). Consider the points of Tn-IF and the intervals

they determine. Add points to this set so that the new intervals

have length < 5. We have added at most [~] + 1 points. Let

F' = F U T-(n-l)(these new points).

We claim that F' is an (n,s)-spanning set for K. Let x ~ K. Then

3 y ~ F

max •
d(Tlx ,Tiy) ~ s.
O~i~n-2

If d(Tn-lx,Tn-ly) ~ s then our claim is proved. If there is no

y E F with both these properties, choose a y E F

max d(Tlx,Tly) ~ s.
0~i~n-2

Consider the interval between Tn-lx and Tn-ly which is mapped by

T -I to the e-interval [Tn-2x,Tn-2y]. Choose a point Tn-lz, z E F'

inside the chosen interval [Tn-lx,Tn-ly] and with d(Tn-lx,Tn-lz) ~s.

Then Tn-2z lies in the s-interval [Tn-2x,Tn-2y] and so,

d(Tn-2z,Tn-2x) ~ ~. The e-interval [Tn-2x,Tn-2y] is mapped by T -I

to an interval of length ~ 1/4, and hence to the e-interval

 184

[Tn-3x,Tn-3y]. So, since Tn-3z is in this interval,

d(Tn-3x~Tn-3z) ~ ~. Similarly, by induction

d(Tlx,Tlz) ~ ~ V i, 0 ~ i ~ n-l.

Thus, F' is an ( n , ~ ) - s p a n n i n g set for K. So,

rn(~,K) ~ rn_l(~,K) + [1/k] + 1

n([i/~] + i).

1
Therefore, rT(~,K) : lim ~ log rn(~,K) : 0,

so, h(T) = 0. //

Corollary:

Any h o m e o m o r p h i s m of [0,i] has zero t o p o l o g i c a l entropy.

Proof: T: [0,i] ~ [0,i] has either T(0) = 0 and T(1) = 1

or T(0) = 1 and T(1) = 0. In both cases T2 induces a h o m e o m o r -

phism of K. //

(4) If T: M m ~ M m is a d i f f e r e n t i a b l e map of an m - d i m e n s i o n a l

Riemannian m a n i f o l d Mm with Riemannian metric ll'II, then

hp(T) ~ max {0 , m log sup lldTxll}

xEM

where dTx: M x ~ MT(x) is the derivative of T at x and p is the

metric on M d e t e r m i n e d by the Riemannian metric. This has been

proved by several people.

--~--
 Bibliography

R. Abraham:

[i] Foundations of Mechanics, Benjamin, 1967.

R. Abraham and J. Robbin:

[i] Mappings and Flows, Benjamin, 1967.

R. L. Adler:

[i] Skew products of Bernoulli shifts and rotations, Israel

Journal of Mathematics, vol. 12, pp. 215-222, 1972.

R. L. Adler, A. G. Konheim and M. H. McAndrew:

[I] Topological entropy, Transactions of the American Mathema-

tical Society, vol. 114, pp. 309-319, 1965.

R. L. Adler and P. C. Shields:

[i] Skew products of Bernoulli shifts with rotations, Israel

Journal of Mathematics, vol. 12, pp. 215-222, 1972.

R. L. Adler and B. Weiss:

[i] Similarity of Automorphisms of the Torus, Memoirs of the

American Mathematical Society, no. 98, 1970.

H. Anzai:

[I] Ergodic skew-product transformations on the torus, Osaka

Mathematics Journal, vol. 3, pp. 83-99, 1951.

M. Artin and B. Mazur:

[I] On periodi c points, Annals of Mathematics, vol. 81, pp. 82-

99, 1965.

L. Auslander, L. Green, and F. Hahn:

[i] Flows on Homogeneous Spaces, Annals of Mathematics Studies,

no. 53, 1963.
 186

A. Avez and V. I. Arnold:

[i] Er~odic Problems in Classical Mechanics, Benjamin, 1968.

K. Berg:
[i] Convolution of invariant measures, maximal entropy,
Mathematical Systems Theory, vol. 3, pp. 146-150, 1969.

P. Billingsley:

[i] Er$odic Theory and Information, Wiley, 1965.

G. D. Birkhoff:

[i] Proof of the ergodie theorem, Proceedings of the National

Academy of Sciences USA, vol. 17, pp. 656-660, 1931.

J. R. Blum and D. L. Hanson:

/i] On the isomorphism problem for Bernoulli schemes, Bulletin

of the American Mathematical Society, vol. 69, pp. 221-223,
1963.

R. Bowen :

[i] Topological entropy and axiom A, Proceedings of the Summer

Institute on Global Analysis, Berkeley, California, pp. 23-
42, 1 9 6 8 .

[2] Markov partitions for axiom A diffeomorphisms, American

Journal of Mathematics, vol. 92, pp. 725-747, 1970.

[3] Markov partitions and minimal sets for axiom A diffeomor-

phisms, American Journal of Mathematics, vo!. 92, pp. 907-
918, 1970.

[4] Entropy for group endomorphisms and homogeneous spaces,

Transactions of the American Mathematical Society, vol. 153,
pp. 401-414, 1971.

[5] Periodic points, measures and axiom A, Transactions of the

American Mathematical Society, vol. 154, pp. 377-397, 1971.

[6] Periodic orbits for hyperbolic flows, American Journal of

Mathematics, vol. 94, pp. 1-30, 1972.

[7] Entropy-expansive maps, Transactions of the American Mathe-

matical Society, vol. 164, pp. 323-331, 1972.

[8] The equidistribution of closed geodesics, ~ e r i c a n Journal

of Mathematics, vol. 94, pp. 413-423, 1972.

[9] One-dimensional hyperbolic sets for flows, Journal of Dif-

ferential Equations, vol. 12, pp. 173-179, 1972.
 187

/i0] Symbolic dynamics for hyperbolic flows, American Journal

of Mathematics, vol. 45, pp. 429-460, 1973.

[ii] Topological entropy for noncompact sets, Transactions of

the American Mathematical Society, vol. 184, pp. 125-136,
1973.
[12] Entropy versus homology for certain diffeomorphisms,
Topology, vol. 13, pp. 61-67, 1974.

R. Bowen and P. Walters:

[i] Expansive one-parameter flows, Journal of Differential

Equations, vol. 12, pp. 180-193, 1972.

R. V. Chacon:
[I] Change of velocity in flows, Journal of Mathematics and
Mechanics, vol. 16, pp. 417-431, 1966.

Hsin- Chu:

[i] Some results on affine transformations of compact groups,

to appear.

J. P. Conze:
[1] Entropie des flots et des transformations affine sur les
espaces homogenes, Compte Rendu, vol. 270, pp. 547-548,
1970.

E. I. Dinaberg:
[i] The relation between topological entropy and metric entropy,
Doklady Akademii Nauk SSSR, vol. 190, 1970 (Russian),
Soviet Mathematics, vol. ii, pp. 13-16, 1970 (English).
[2] A connection between various entropy characterizations of
dynamical systems, Izvestija Akademii Nauk SSSR, Serija
Matematiceskaja, vol. 35, pp. 324-368, 1971 (Russian).

R. Ellis:
[I] Lectures on Topological Dynamics, Benjamin, 1969.

N. A. Friedman:
[i] Introduction Er~odic Theory, Van Nostrand, 1970.

[2] Bernoulli shifts induce Bernoulli shifts, Advances in

Mathematics, vol. i0, pp. 29-48, 1973.
 188

N. A. Friedman and D. S. Ornstein:

[i] An isomorphism of weak Bernoulli Transformations, Advances

in Mathematics, vol. 5, pp. 365-394, 1970.

[2] Entropy and the Isomorphism Problem, to be published by

Springer.

[3] Ergodic transformations induce mixing transformations,

Advances in Mathematics, vol. i0, pp. 147-163, 1973.

H. Furstenberg:

[i] Strict ergodicity and transformations of the torus,

American Journal of Mathematics, vol. 83, pp. 573-601, 1961.

[2] Disjointness in ergodic theory, Mathematical Systems Theory,

vol. i, pp. 1-50, 1967.

T. N. T. Goodman:

[i] Relating topological entropy and measure entropy, Bulletin

of the London Mathematical Society, vol. 3, pp. 176-180.
1971.
[2] Maximal measures for expansive homeomorphisms, Journal of
the London Mathematical Society, vol. 5, pp. 439-444, 1972.

L. W. Goodwyn:

[i] Topological entropy bounds measure-theoretic entropy,

Proceedings of the American Mathematical Society, vol. 23,
pp. 679-688, 1969.

[2] A characterization of symbolic cascades in terms of expan-

siveness and topological entropy, Mathematical Systems
Theory, vol. 4, pp. 157-157, 1970.

[3] Some counterexamples in topological entropy, Topology,

vol. Ii, pp. 377-385, 1972.

[4] Comparing topological entropy with measure theoretic

entropy, American Journal of Mathematics, vol. 74, pp. 366-
388, 1972.

W. H. Gottschalk:
[i] Bibliography for Dynamical Topology, fourth edition,
Wesleyan University, 1969.

W. H. Gottschalk and G. A. Hedlund:

[I] Topological Dynamics, American Mathematical Society Col-
loquium Publications, 1955.
 189

C. Grillenberger:
[i] Constructions of strictly ergodic systems I, II, Zeit-
schrift f~r Wahrscheinlichkeitstheorie, vol. 25, pp. 323-
334, 1973.

F. Hahn:

[I] On affine transformations of compact abelian groups, Ameri-

can Journal of Mathematics, vol. 85, pp. 428-446, 1963.

F. Hahn and Y. Katznelson:

[i] On the entropy of uniquely ergodic transformations, Trans-

actions of the American Mathematical Society, vol. 126,
pp. 3 3 5 - 3 6 0 , 1967.

F. Hahn and W. Parry=

[i] Minimal dynamical systems with quasi-discrete spectrum,
Journal of the London Mathematical Society, vol. 40,
pp. 309-323, 1965.

[2] Some characteristic properties of dynamical systems with

quasi-discrete spectrum, Mathematical Systems Theory,
vol. 2, pp. 179-190, 1968.

P. R. Halmos:
[i] Measure Theory, Van Nostrand, 1950.

[2] L_ectures on Ergodic Theory, Chelsea, 1953.

[3] Introduction to Hilbert Space and the Theory of Spectral
Multiplicity, Chelsea, 1957.
[4] On automorphisms of compact groups, Bulletin of the Ameri-
can Mathematical Society, vol. 49, pp. 619-624, 1943.

P. R. Halmos and J. Von Neumann:

[i] Operator methods in classical mechanics, II, Annals of
Mathematics, vol. 43, pp. 235-247, 1942.

H. Hoare and W. Parry:

[i] Afflne transformations with quasi-discrete spectrum, I,

Journal of the London Mathematical Society, vol. 41,
p p . 8 8 - 9 6 , 1966.
 190

[2] Affine transformations with quasi-discrete spectrum, II,

Journal O f the London Mathematical Society, vol. 41,
pp. 529-530, 1966.

[3] Semi-groups of affine transformations, Quarterly Journal

of Mathematics, Oxford, vol. 17, pp. 106-111, 1966.

E. Hopf:

[I] Er~odentheorie, Chelsea, 1937.

K. Jacobs:

[i] Neue Methode und Er~ebnisse der Er~odentheorie, Springer,

1960.

[2] Lecture Notes on Er~odic Theory, Aarhus University, 1962-3.

R. Jewett:

[I] The prevalence of uniquely ergodic systems, Journal of

Mathematics and Mechanics, pp. 717-729, 1978.

S. A. Juzvinskii:

[i] Calculation of the entropy of a group endomorphism,

Sibirskii Matematiceskii Zurnal, vol. 8, pp. 230-239, 1967
(Russian); Siberian Mathematical Journal, vol. 8, pp. 172-
178, 1967 (English).

[2] Metric properties of endomorphisms Of compact groups,

I zvestija Akademii Nauk SSSR, Serija Matemasiceskaja,
vol. 29, pp. 1295-1328, 1965 (Russian); American Mathemati-
cal Society Translations, Series 2, vol. 66, pp. 53-98,
1968 (English).

S. Kakutani:
[i] Induced measure preserving transformations, Proceedings of
the Imperial Academy of Tokyo, vol. 19, pp. 635-641, 1943.
[2] Examples of ergodic measure preserving transformations
which are weakly mixing but not strongly mixing, Recent
Advances in Topological Dynamics, Springer Lecture Notes,
no. 318, pp. 143-149, 1973.

Y. Katznelson:
[i] Ergodic automorphisms of T n are Bernoulli shifts, Israel
Journal of Mathematics, vol. 10, pp. 186-195, 1971.
 191

Y. Katznelson and B. Weiss:

[iJ Commuting measure-preservlng transformations, Israel Jour-

nal of Mathematics, vol. 12, pp. 161-173, 1972.

[2] Ergodic automorphisms of the solenoid are Bernoulli,

unpublished.

J. L. Kelley:

[i] General Topology, Van Nostrand, 1955.

H. B. Keynes and J. B. Robertson:

[i] Generators for topological entropy and expansiveness,

Mathematical Systems Theory, vol. 3, pp. 51-59, 1969.

A. I. Khinchin:

[i] Mathematical Foundations of Statistical Mechanics, Dover,

1949.

A. N. Kolmogorov:

[i] A new metric invariant of transient dynamical systems and

automorphisms of Lebesgue spaces, Doklady Akademii Nauk
SSSR, vol. 119, pp. 861-864, 1958.

[2] On the entropy per time unit as a metric invariant of

automorphisms, Doklady Akademii Nauk SSSR, vol. 124,
pp. 754-755, 1959.

W. Krieger:

[1] On entropy and generators of measure-preserving transfor-

mations, Transactions of the American Mathematical Society,
vol. 149, pp. 453-464, 1970.

[2] On unique ergodicity, Proceedings of the Sixth Berkeley

Symposium on Mathematical Statistics and Probability, 1970.

N. Kryloff and N. Bogoliouboff:

[i] La th~orie g~n~rale de la mesure dans son application

l'~tude des systemes dynamique de la m~canique non lin~air%
Annals of Mathematics, vol. 38, pp. 65-113, 1937.

A. G. Kushnirenko:

[I] Metric invariants of entropy type, Uspehi Matematiceskih

Nauk, vol. 22, no. 5, pp. 57-65, 1967 (Russian), Russian
Mathematical Surveys, vol. 22, no. 5, pp. 53-61, 1967
(English).
 192

D. A. Lind:

[i] Ergodic automorphisms of the infinite torus are Bernoulli,

to appear.

G. W. Mackey:

[i] Ergodic Theory and its significance for statistical

mechanics and probability theory, Advances in Mathematics,
vol. 12, pp. 178-288, 1974.

A. Manning:

[i] Axiom A diffeomorphisms have rational zeta functions,

Bulletin of the London Mathematical Society, vol. 3,
pp. 215-220, 1971.

L. Markus:

[I] Lectures in Differentiable Dynamics, American Mathematical

Society Regional Conference Series, no. 3, 1971.

R. McCabe and P. C. Shields:

[i] A class of Markov shifts which are Bernoulli shifts,

Advances in Mathematics, vol. 6, pp. 823-328, 1971.

L. D. Meshalkin:

[i] A case of isomorphism of Bernoulli schemes, Doklady

Akademii Nauk SSSR, vol. 128, pp. 41-44, 1959.

V. Nemytskii and V. Stepanov:

[i] Qualitative Theory of Differential Equations, Princeton,

1960.

J. yon Neumann:

[I] Proof of the quasi-ergodic hypothesis, Proceedings of the

National Academy of Science USA, vol. 18, pp. 263-266, 1932.

[2] Zur Operatorenmethode in der klassichen Mechanik, Annals

of Mathematics, vol. 33, pp. 587-642, 1932.

D. Newton:
[I] On sequence entropy I, II, Mathematical Systems Theory,
vol. 4, pp. 119-128, 1970.
 193

Z. Nitecki:

[I] Differentiable Dynamics, M.I.T. Press, 1971.

D. S. Ornstein:

[i] Bernoulli shifts with the same entropy are isomorphic,

Advances in Mathematics, vol. 4, pp. 337-352, 1970.

[2] Two Bernoull i shifts with infinite entropy are isomorphic,

Advances in Mathematics, vol. 5, pp. 339-348, 1970.

[3] Factors of Bernoulli, shifts are Bernoulli shifts, Advances

in Mathematics, vol. 5, pp. 34@-364, 1970.

[4] Imbedding Bernoulli shifts in flows, contributions to

ergodic theory and probability, Lecture Notes in Mathema-
tics, Springer Berlin, pp. 178-218, 1970.

[5] Some new results in the Kolmogorov-Sinai theory of entropy

and ergodic theory, Bulletin of the American Mathematical
Society, vol. 77, pp. 878-890, 1971.

[6] An example of a Kolmogorov automorphism that is not a

Bernoulli shift, Advance{ in Mathematics, vol. 10, pp. 49-
62, 1973.

[7] The isomorphism theorem for Bernoulli flows, Advances in

Mathematics, vol. 10, pp. 124-142, 1973.

[8] A K-automorphism with no square root and Pinsker's conjec-

ture, Advances in Mathematics, vol. 10, pp. 89-102, 1973.

[9] A mixing transformation for which Pinsker's conjecture

fails, Advances in Mathematics, vol. 10, pp. 103-123, 1973.

D. S. Ornstein and P. C. Shields:

[i] An uncountable family of K-automorphisms, Advances in

Mathematics, vol. i0, pp. 83-88, 1973.

[2] Mixing Markov shifts of kernel type are Bernoulli, Advances

in Mathematics, vol. 10, pp. 143-148, 1973.

D. S. Ornstein and B. Weiss:

[i] Geodesic flows are Bernoullian, Israel Journal of Mathema-

tics, vol. 14, pp. 184-197, 1973.

J. C. Oxtoby:
[i] Ergodic sets, Bulletin of the American Mathematical Societ~
vol. 58, pp. 116-136, 1952.
 194

W. Parry:

[i] Symbolic dynamics and transformations of the unit interval,

Transactions of the American Mathematical Society, vol. 122,
pp. 368-378, 1966.

[2] Compact abelian group extensions of discrete dynamical

systems, Z. Wahrscheinlichkeitstheorie, vol. 13, pp. 95-
113, 1969.
[3] Entropy and Generators in Er~odic Theory, Benjamin, 1969.

[4] Ergodic properties of affine transformations and flows on

nilmanifolds, American Journal of Mathematics, vol. 91,
pp. 757-771, 1969.

[5] Spectral analysis of G-extensions of dynamical systems,

Topology, vol. 9, pp. 217-224, 1970.

[6] Dynamical systems on nilmanifolds, Bulletin of the London

Mathematical Society, vol. 2, pp. 37-40, 1970.

M. S. Pinsker:
[I] Dynamical systems with completely positive and zero
entropy, Doklady Akad. Nauk SSSR, vo!. 133, pp. 1025-1026,
1960 (Russian), Soviet Mathematics Doklady, vol. i,
pp. 937-938, 1960 (English).

L. S. Pontrjagin:
[I] Topological Groups, Gordon and Breach, 1968.

W. Reddy:

[1] The existence of expansive homeomorphisms on manifolds,

Duke Mathematical Journal, vol. 32, pp. 627-632, 1965.

V. A. Rohlin:
[i] A general measure-preserving transformation is not mixing,
Doklady Akademii Nauk, vol. 60, pp. 349-351, 1948.

[2] Generators in ergodic theory, Vestnik Leningradskogo

Universiteta, vol. 18, no. i, pp. 28-32, 1963.

[3] Selected topics in the metric theory of dynamical systems,

American Mathematical Society Translations, Series 2,
vol. 49, pp. 171-240, 1966.
[4] Metric properties of endomorphisms of compact commutative
groups, Izvestija Akademii Nauk, Serija Matematiceskaja,
vol. 28, pp. 867-874, 1964 (Russian); American Mathematical
Society Translations, Series 2, vol. 64, pp. 244-252,
1967 (English).
 195

[5] Lectures on the entropy theory of transformations with

invariant measure, Uspehi Matematiceskih Nauk, vol. 22,
no. 5, pp. 3-66, 1967 (Russian); Russian Mathematical
Surveys, vol. 22, no. 5, pp. 1-52, 1967 (English).

[6] Metric properties of endomorphisms of compact commutative

groups, I zvestija Akademii Nauk, Serija Matematiceskaja,
vol. 13, pp. 329-340, 1949 (Russian).

V. A. Rohlin and Ja. G. Sinai:

[i] Construction and properties of invariant measurable parti-

tions, Dok!ady Akademii Nauk SSSR, vol. 141, pp. 1038-i041,
1961 (Russian); Soviet Mathematics, vol. 2, pp. 1611-1614,
1961 (English).

D. Ruelle:

[i] Statistical mechanics on a compact set with Z 9 action

satisfying expansiveness and specification, Transactions
of the American Mathematical Society, vol. 185, pp. 237-
252, 1973.

P. C. Shields:

[i] Bernoulli shifts are determined by their factor algebras,

Proceedings of the American Mathematical Society, vol. 41,
pp. 331-332, 1973.

[2] The Theory of Bernoulli Shifts, University of Chigago

Lecture Notes, 1974.

C. P. Simon:

[1] Instability in Diffr(T 3) and the nongenericity of rational

zeta functions, Transactions of the American Mathematical
Society, vol. 174, pp. 217-242, 1972.

Ja. G. Sinai:

[I] On the concept of entropy of a dynamical system, Doklady

Akademii Nauk SSSR, vol. 124, pp. 758-771, 1959.

[2] On flows with finite entropy, Doklady Akademii Nauk SSSR,

vol. 125, pp. 1200-1202, 1959.

[3] On a weak isomorphism of transformations with an invariant,

Doklady Akademii Nauk SSSR, vol. 147, pp. 797-800, 1962
(Russian), Soviet Mathematics Doklady, vol. 3, pp. 1725-
1729, 1962 (English).

[4] On a weak isomorphism of transformations with an invariant

measure, Matematiceskii Sbornik, voi. 63, pp. 23-42,
1964 (Russian).
 196

[5] Classical dynamical systems with countably-multiple

Lebesgue spectrum, II, Izvestija Akademii Nauk SSSR, Serija
Matematiceskaja, vol. 30, pp. 15-68, 1956 (Russian);
American Mathematical Society Translations, Series 2, vol.
58, pp. 34-88, 1968 (English).

[6] Construction of Markov partitioning, Funkcional'nyi

Analiz i ego Prilo~enija, vol. 2, pp. 70-80, 1968.

[7] Markov partitions and u-diffeomorphisms, Functional Analy-

sis and Applications, vol. 2, pp. 61-82, 1968 (English);
Funkcional'nyi Analiz i ego Prilo~enija, vol. 2, pp. 64-
89, 1968 (Russian).

S. Sma!e:
[i] Differentiable dynamical systems, Bulletin of the American
Mathematical Society, vol. 73, pp. 747-817, 1967.

M. Smordinsky:
[i] On Ornstein's isomorphism theorem for Bernoulli shifts,
Advances in Mathematics, vo!. i0, pp. 1-9, 1973.
[2] Er~odic Theory~ Entropy, Springer Lecture Notes, no. 214,
1971.

B. Weiss:
[i] The isomorphism problem in ergodic theory, Bulletin of the
American Mathematical Society, vol. 78, pp. 668-684, 1972.
 Index

affine transformation 18
aperiodic 89

Bernoulli shift 18, 98

character theory lO
complete invariant 96
completely positive entropy 104
conditional entropy 76
conjugate transformations 53
continuous spectrum 46

direct product of probability spaces 5

direct product of transformations 44
discrete spectrum~ measure theoretic 64
topological 125
dist~l 120

eigenfunction, measure theoretic 45~ 63

topological 123
eigenvalue, measure theoretic 45
topological 123
endomorphisms of tori 12
entropy 72
entropy of affine transformations of tori 168
of Bernoulli shifts 95
of rotations 93
ergodicity 21
ergodic theorem, Birkoff 29
maximal 32
LP 36
exact endomorphism lll
expansive homeomorphism 171

generator, measure-theoretic 88
topological 168

•Haar measure
Hahn-Kolmogorov Extension Theorem
Hilbert spaces

induced operator on L p 32
invariant 57
invariant measures for homeomorphisms 128
isomorphism 52

Kolmogorov automorphism 101

Kolmogorov - Sinai Theorem 87
Krylov and Bogolioubov Theorem 132
 198

Lebesgue Covering Lemma 151

Lebesgue spectrum 60

Markov chain 105

measure algebra 53
measure preserving transformation 16
measure theory 3
minimal homeomorphism I13
minimal set 114
mixing 37

nilmanifold 106
non-invertible transformation 110
normal number 31

orbit 113

partitions 70
periodic point 115
Pinsker algebra 107
Poinca~e Recurrence Theorem 20
p u r e p o i n t spectrum 64

recurrence 20
refinement of open covers 140
Riesz Representation Theorem 130
rotations on groups 67

semi-simple homeomorphism 115

separated set 146
sequence entropy 108
a-algebra 3
spanning set 146
spectral isomorphism 54
spectral theorem 46
sub-u-algebras 70

topological conjugacy 122

topological entropy 143, 147
connection with measure theoretic entropy 155
topological entropy of affine transformations of tori 168
of Bernoulli shifts 182
of homeomorphisms of the circle 182
topological transitivity 117

uniformly equivalent metrics 150

uniquely ergodic 135
VoL 277: Seminaire Banach. Edit& par C. HouzeL VII, 229 pages.
I g72. DM 22,-
VoL 278: H, Jacquet, Automorphic Forms on GL(2) Part IL XIII, 142
pages, 1972. DM 18,-
VoL 279: R, Bott, S. Gitter and I. M. James, Lectures on Algebraic and
Differential Topology. V, 174 pages. 1972. DM 20,-
VoI. 280: Conference on the Theory of Ordinary and Partial Differen-
t=al Equations. Edited by W. N. Everitt and B. D. Sleeman. XV, 367
pages, 1972. DM 2g,-
VoL 281: Coherence in Categories. Edited by S. Mac Lane. VII, 235
pages. 1972. DM 22,-
Vol. 282: W. Klingenberg und P. Flaschel, Riemannsche Hilbertman-
nigfaitigkeiten, Periodische Geodatische. VII, 211 Seiten. 1972.
DM 22,~
Vol. 283: L. Illusie, Complexe Cotangent et Deformations II. VII, 304
pages. 1972. DM 27,-
Vol. 284: P. A. Meyer, Martingales and Stochastic Integrals I. Vl, 89
pages. 1972. DM 18,-
Vol. 285: P. de la Harpe, C~assical Banach-Lie Algebras and Banach-
Lie Groups of Operators in Hilbert Space. III, 160 pages. 1972.
DM 18,-
Vol. 286: S. Murakami, On Automorphisms of Siegel Domains. V,
95 pages, lg72.DM 18,-
Vol. 287: Hyperfunctions and Pseudo-Differential Equations. Edited
by H. I{omatsu. VII, 629 pages. 1973. DM 40,-
VoI. 288: Groupes de Monodromie en Geometrie Algebrique. (SGA 7
I). Ding~ par A. Grothendieck. IX, 523 pages, 1972. DM 55,-
Vol. 289: B, Fuglede, Finely Harmonic Functions. III, 188. 1972.
DM 20,-
Vol. 290: D. B. Zagier, Equivariant Pontrjagin Classes and Applica-
tions to Orbit Spaces. IX, 130 pages. 1972. DM 18,-
VoL 291 : P. Orlik, Seifert Manifolds, VIII, 155 pages. 1972. DM 18,-
Vol. 292: W. D, Wallis, A. P. Street and J. S. Wallis, Combinatorics:
Room Squares, Sum-Free Sets, Hadamard Matrices. V, 508 pages.
1972. DM 55,-
Vol. 293: R. A. DeVore, The Approximation of Continuous Functions
by Positive Linear Operators. VIII, 289 pages, 1972, DM 27,-
Vol, 294: Stability of Stochastic Dynamical Systems. Edited by R. F.
Curtain. IX, 332 pages, 1972. DM 29,-
Vol. 295: C. Dellacherie, Ensembles Analytiques Capacites. Mesures
de Hausdorff. XII. 123 pages. 1972. DM 18,-
Vol. 296: Probability and Information Theory II. Edited by M. Behara,
K. Krickeberg and J. Wolfowitz, V, 223 pages. 1973. DM 22,-
Vol. 297: J. Garnett, Analytic Capacity and Measure. IV, 138 pages.
1972. DM 18,-
VoL 298: Proceedings of the Second Conference on Compact Trans-
formation Groups. Part 1. Xlll, 453 pages. 1972. DM 35,-
Vol. 29 g: Proceedings of the Second Conference on Compact Trans-
formation Groups. Part 2. XlV, 327 pages. 1972. DM 2g,-
Vol. 300: P. Eymard, Moyennes Invariantes et Representations Uni-
taires. II, 113 pages. 1072. DM t8,-
Vol. 301: F, Pittnauer, Vorlesungen 0ber asymptofische Reihen. VI,
186 Seiten. 1972. DM 18,-
Vol. 302: M. Demazure, Lectures on p-Divisible Groups. V, 98 pages.
1972. DM 18,-
Vol. 303; Graph Theory and Applications. Edited by Y. Alavi, D. R. Lick
and A. T. White. IX, 329 pages. 1972. DM 26,-
Vot. 304: A. K. Bousfield and D. M. Kan, Homotopy Limits, Comple-
tions and Localizafions. V, 348 pages. 1972. DM 29,-
Vol. 305: Theorie des Topos et Cohomologie Elsie des Schemas.
Tome 3. (SGA 4). Dirige par M. Artin, A. Grothendieck et J. L. Verdier.
Vl, 640 pages. 1973. DM 55,-
Vol. 306: H. Luckhardt, Extensional G6del Functional Interpretation.
VI, 161 pages. 1973. DM 20,-
VoL 307: J. L. Bretagnolle, S. D. Chatterji et P..-A. Meyer, Ecole d'et~
de Probabilit~s: Processus Stochastiques, VI, 198 pages. 1973.
DM22,-
Vol. 308: D. Knutson, Z-Rings and the Representation Theory of the
Symmetric Group, IV, 203 pages. 1973. OM 22,-
VoI. 309: O. H. Sattinger, Topics in Stability and Bifurcation Theory.
VI, 190 pages. 1973. DM 20,-
Vol. 310: B. Iversen, Generic Local Structure of the Morphisms in
Commutative Algebra. IV, 108 pages. 1973. DM 18,-
Vol. 311 : Conference on CommutativeAIgebra. Edited by J.W. Brewer
and E. A. Rutter. VII, 251 pages. 1973. OM 24,-
Vol. 312: Symposium on Ordinary Differential Equations. Edited by
W. A. Harris, Jr. and Y. Sibuya. VIII, 204 pages. 1973. DM 22,-
VoL 313: K. J~rgens and J. Weidmann, Spectral Properties of Hamil-
tonian Operators. III, 140 pages. 1973. DM 18,-
Vol. 314: M. Deuring, Lectures on the Theory of Algebraic Functions
of OneVariable.VI, 151 pages. 1973. DM 18,-
Vol. 315: K. Bichteler, Integration Theory (with Special Attention to
Vector Measures). VI, 357 pages. 1973. DM 29,-
VoI. 316: Symposium on Non-Wen-Posed Problems and Logarithmic
Convexity. Edited by R. J. Knops. V, 176 pages. 1973. DM 20,-
VoL 317: Seminaire Bourbaki - vol. 1971/72. Exposes 400-417. IV,
361 pages. 1973. DM 2g,-
Vol. 318: Recent Advances in Topological Dynamics. Edited by A.
Beck. VIII, 285 pages. 1973. DM 27,-
VoL 319: Conference on Group Theory. Edited by R, W, Gatterdam
and K. W. Weston. V. 188 pages. 1973. DM 20,-
Vol. 320: Modular Functions of One Variable I. Edited by W. Kuyk. V,
196 pages. 1973. DM 20,-
VoL 321: Seminaire de Probabilit~s VII. Edite par P. A. Meyer. V|, 322
pages. 1973. DM 29,-
Vol. 322: Nonlinear Problems in the Physical Sciences and Biology.
Edited by I. Stakgold, D. D. Joseph and D. H, Sattinger. VIII, 3 5 7
pages. 1973. DM 29,-
Vol. 323: J. L Lions. Perturbations Singulieres dans les Problemes
aux Limites et en Contr6te Optimal. XII, 645 pages. 19?3. DM 46,-
Vo/. 324: K. Kreith, Oscillation Theory. VI, 109 pages. 1973. DM 18,-
Vol. 325: C.-C. Chou, La Transformation de Fourier Complexe et
L'Equation de Convolution. IX, 137 pages. 1933. DM 16,-
VoL 326: A. Robert, Elliptic Curves. VIII, 264 pages. 1973. DM 24,-
Vol. 327 : F_.Matlis, One-Dimensional Cohen-Macaulay Rings. XlI,
157 pages. 1973. DM 20,-
Vol. 328: J. R. Buchi and D. Siefkes, The Monadic Second Order
Theory of All Countable Ordinals, Vt, 217 pages. 1973. DM 22,-
Vol. 329: W. Trebels. Multipliers for (C, =)-Bounded Fourier Expan-
sions in Banach Spaces and Approximation Theory. VII, 103 pages.
1973. DM 18,-
Vol. 330: Proceedings of the Second Japan-USSR Symposium on
Probability Theory. Edited by G. Maruyama and Yu. V. Prokhorov.
VI, 650 pages. 1973. DM 40,-
VoL 331: Summer School on Topological Vector Spaces. Edited by
L. Waelbroeck. VI, 226 pages. 1973. DM 22,-
Vol. 332: Seminaire Pierre Letong (Analyse) Annee 1971-1972. V, 131
pages. 1973. DM 18,-
Vol. 333: Numerische, insbesondere spproximationstheoretische Be-
handlung yon Funktionalgleichungen. Herausgegeben yon R. An-
sorge und W. TOrnig. Vl, 296 Seiten. 1973. DM 2?,-
VoL 334: F. Schweiger, The Metrical Theory of Jacobi-Perron Algo-
rithm.V, 111 pages. 1973. OM 18,-
Vol. 335: H. Huck, R. Roitzsch, U. Simon, W. Vortisch, R. Walden, B.
Wegner und W. Wendland, Beweismethoden der Differentialgeome-
trie im Groaen. IX, 169 Seiten. 1973, DM 20,-
Vol. 336: L'Anaiyse Harmonique dans le Domaine Complexe. Edite
par E. J. Akutowicz. VIII, 169 pages. 1973. OM 20,-
Vot. 337: Cambridge Summer School in Mathematical Logic. Edited
by A, R. D. Mathias and H. Rogers. IX, 660 pages. 1973. DM 46,-
Vol: 338: J. Lindenstrauss and L. Tzafriri, Classical Banach Spaces.
IX, 243 pages. 1973. DM 24,-
Vol. 339: G. Kempf, F. Knudsen, D. Mumford and B. Saint-Dons t,
Toroidal Embeddings I. VIII, 209 pages. 1973. DM 22,-
Vol. 340: Groupes de Monodromie en G~om~trie Algebrique.
(SGA ? II). Par P. Deligne et N. Katz. X, 438 pages. 1973. DM 44,-
Vol. 341: Algebraic K-Theory I," Higher K-Theories. Edited by
H. 8ass. XV, 335 pages. 1973. DM 29,-
Vol. 342: Algebraic K-Theory II, "Classical = Algebraic K-Theory,
and Connections with Arithmetic. Edited by H. Bass. XV, 627
pages. 1973. DM 40,-
Vol. 343: Algebraic K-Theory III, Hermitian K-Theory and Geo-
metric Applications. Edited by H. Bass. XM, 572 pages. 1973.
DM 40,-
Vol. 344: A. S. Troelstra (Editor), Metamathematical Investigation
ot tntuitionistic Arithmetic and Analysis. XVll, 485 pages. 1973.
DM 36,-
VoL 345: Proceedings of a Conference on Operator Theory.
Edited by P. A. Fillmore. Vl, 228 pages. ]973. DM 22,-
Vol. 346: Futbik et al., Spectral Analysis of Nonlinear Operatore.
II, 287 pages. 1973. DM 26,-
VoL 34?: J. M. Boardman and R. M. Vogt, Homotopy Invariant
Algebraic Structures on Topological Spaces. X, 257 pages.
1973. DM 24,-
Vol. 348: A. M. Mathai and R. K. Saxena, Generalized Hyper-
geometric Functions with Applications in Statistics and Physical
Sciences. VII, 314 pages. 1973. DM 26,-
VoL 349: Modular Functions of One Variable IL Edited by W. Kuyk
and P. Deligne. V, 598 pages. 1973. DM 38,-
Vol. 350: Modular Functions of One Variable 111.Edited by W.
Kuyk and J.-P. Serre. V, 350 pages. ~973. DM 26,-
Vol, 351 : H. Tachikawa, Quasi-Frobenius Rings and Generaliza-
tions. Xl, 172 pages. 1973. DM 20,-
Vol. 352: .L D. Fay, Theta Functions on Riemann Surfaces. V,
137 pages. 1973. DM 18,-
Voi. 353: Proceedings of the Conference. on Orders, Group
Rings and Related Topics, Organized by J. S. Hsia, M. L. Madan
and T. G. Ralley. X, 224 pages. 1973. DM 22,-
Vol. 354: K. J. Devlin, Aspects of Constructibility. XII, 240 pages.
1973. DM 24,-
Vol. 355: M, Sign, A Theory of Semigroup Valued Measures.
V, 140 pages, 1973. DM 18,-
Vol. 356; W. L J. van der Kallen, Infinitesimally Central-Exten-
sions of Chevatley Groups. VII, 147 pages. 1973. DM.18,-
VoL 357: W. Borho, P. Gabriel und R. Rentschler, Primideale
in Einh•llenden aufl6sbarer Lie-Algebren. V, 182 Seiten. 1973.
DM 20,-
Vol. 356: F. L. Williams, Tensor Products of Principal Series
Representations. VI, 132 pages. 1973. DM 18,-
Vol. 359: U. Stammbach, Homology in Group Theory. VIII, 183
pages. 1973. DM 20,-
Vol. 360: W..I. Padgett and R. L. Taylor, Laws of Large Numbers
for Normed Linear Spaces and Certain Frechet Spaces. Vl,
111 pages. 1973. DM 18,-
VoL 361 : J. W. Schutz, Foundations of Special Relativity: Kine-
matic Axioms for Minkowski Space Time. XX, 314 pages. 1973.
DM 26,-
VoI. 362: Proceedings of the Conference on Numerical Solution
of Ordinary Differential Equations. Edited by D. Bettis. VIII,
490 pages. 1974. DM 34,-
VoI. 363: Conference on the Numerical Solution of Differential
Equations. Edited by G. A. Watson. IX, 221 pages. 1974. DM 20,-
VoL 364: Proceedings on Infinite Dimensional Holomorphy.
Edited by T. L. Hayden and T. J. Suffridge. VII, 212 pages. 1974.
DM 20,-
Vol. 365: R. P. Gilbert, Constructive Methods for Elliptic Equa-
tions. VII, 397 pages. 1974. DM 26,-
Vol. 366: R. Steinberg, Conjugacy Classes in Algebraic Groups
(Notes by V. V. Deodhar). Vl, 159 pages. 1974. DM 18,-
Vol. 367: K. Langmann und W. LLitkebohmert, Cousinverteilun-
gen und Fortsetzungssiitze. VI, 151 Seiten. 1974. DM 16,-
VoL 368: R. J. Milgram, Unstable Homotopy from the Stable
Point of View. V, 109 pages. 1974. DM 16,-
Vol. 369: Victoria Symposium on Nonstandard Analysis. Edited
by A. Hurd and P. Loeb. XVltl, 339 pages. 1974. DM 26,-
Vol. 370: B. Mazur and W, Messing, Universal Extensions and
One Dimensional Crystalline Cohomology. VII, 134 pages. 1974.
DM 16,-
Vol. 371 : V. Poenaru, Analyse Differentiel(e. V, 228 pages. 1974.
DM 20,-
Vol. 372: Proceedings of the Second International Conference
on the Theory of Groups 1973. Edited by M. F. Newman. VII,
740 pages. 1974. DM 48,-
Vot. 373: A. E. R. Woodcock and T. Poston, A Geometrical Study
of the Elementary Catastrophes. V, 25?" pages. 1974. DM 22,-
Vot. 374: S Yamamuro, Differential Calculus in Topological
Linear Spaces. IV, 179 pages. 1974. DM 18,-
Vol. 375: Topology Conference 1973. Edited by R. F. Dickman
Jr. and P. Fletcher. X, 283 pages. 1974. DM 24,-
Vol. 376: D, B. Osteyee and I. J. Good, Information, Weight of
Evidence, the Singularity between Probabdity Measures and
Stgnal Detection. Xl, 156 pages. 1974. DM 16.-
Vol. 377: A. M. Fink, Almost Periodic Differential Equations.
VIII, 336 pages. 1974. DM 26,-
VoL 378: TOPO 72 - General Topology and its Applications.
Proceedings 1972. Edited by R. AID, R. W. Heath andJ. Nagata.
XlV, 661 pages. 1974. DM 50,-
VoL 379: A. Badrikian et S. Chevet, Mesures Cylindriques,
Espaces de Wiener et Fonctlons Aleatoires Gausslennes. X,
383 pages. 1974. DM 32,-
VoL 380: M. Petrich, Rings and Semigroups. VIII, 182 pages,
1974. DM 16,-
VoL 381 : S/aminaire de Probabilites VIII. Edite par P. A. Meyer.
IX, 354 pages. 19?4. DM 32,-
~/ol. 382: J. H. van Lint, Combinatorial Theory Seminar Eind-
hoven University of Technology. Vl, 131 pages. 1974. DM 18,-
Vol. 383: Seminaire Bourbaki - vol. 1972173. Exposes 418-435
IV, 334 pages. 1974. DM 30,-
Vol. 384: Functional Analysis and Apphcations, Proceedings
1972. Edited by L. Nachbin. V, 270 pages. 1974. DM 22,-
Vol. 385: J. Douglas Jr. and T. Dupont, Collocation Methods for
Parabolic Equations in a Single Space Variable (Based on C'-
Piecewrse-Potynomial Spaces). V, 147 pages. 1974. DM 16,-
VoL 386: .I. Tits, Buildings of Spherical Type and Finite BN-
Pairs. IX, 299 pages. 1974. DM 24,-
Vol. 367: C. P. Bruter, Elements de la Theorie des Matro'(des.
V, t38 pages. 1974. DM 18,-
Vot. 388: R. L. Lipsman, Group Representations. X, 166 pages.
1974. DM 20,-
Vol. 389: M-A. Knus et M. Ojanguren, Theorie de la Descente
et Algebres d' Azumaya. IV, 163 pages. 1974. DM 20,-
Vol. 390: P. A. Meyer, P. Prlouret et F. Spitzer, Ecote d'Ete de
Probabdptes de Saint-Flour Ill - 1973. Edite par A. Badrikian
et P.-L Hennequin. VIII, 189 pages. 1974. DM 20,-
Vol. 391 : J. Gray, Formal Category Theory: Adjointness for 2-
Categories. XII, 282 pages. 1974. DM 24,-
VoI. 392: Geometrie Differentielle, Colloque, Santiago de
Compostela, Espagne 1972. Edite par E. Vidal. Vl, 225 pages.
1974. DM 20,-
Vol. 393 : G. Wassermann, Stability of Unfoldings. IX, 164 pages.
1974. DM 20,-
VoL 394 : W. M. Patterson 3rd. tterative Methods for the Solution
of a Linear Operator Equation in Hilbert Space - A Survey.
III, 183 pages. 1974. DM 20,-
Vol. 395: Numerische Behandlung nichtlinearer Integrodifferen-
tial- und Differentialgleichungen. Tagung 1973. Herausgegeben
yon R. Ansorge und W. T6rnig. VII, 313 Seiten. 1974. DM 28,-
Vol. 396 : K. H Hofmann, M. Mislove and A. Stralka, The Pontry-
agm Duality of Compact O-Dimensional Semilattices and its
Applications. XVl, 122 pages. 1974. DM 18,-
Vol. 397 : T. Yamada, The Schur Subgroup of the Brauer Group.
V, 159 pages. 1974. DM 18,-
VoL 398: Theories de I'lnformation, Actes des Rencontres de
Marseille-Luminy, 1973. Edite par .L Kampe de Feriet et C. Picard.
XlI, 201 pages. 1974. DM 23.-
Vol. 399: Functional Analysis and its Applications, Proceedings
1973. Edited by H. G. Garnir, K. R, Unni and J. H. Williamson.
XVII, 589 pages. 1974. DM 44,-
VoL 400: A Crash Course on Kleinian Groups - San Francisco
1974. Edited by L Bers and L Kra. VII, 130 pages. 1974. DM 18,-
VoL 401: F. Atiyah, Elliptic Operators and Compact Groups.
V, 93 pages. 1974. DM 18,-
VoL 402: M. Waldschmidt, Nombres Transcendants. VIII, 277
pages. 1974. DM 25,-
VoI. 403: Combinatorial Mathematics - Proceedings 1972. Edited
by D. A. Holton. VIII, 148 pages. 1974, DM 18,-
Vol. 404: Theorie du Potentiel et Analyse Harmonique. Editd par
J. Faraut. V, 245 pages. 1974, DM 25,-
Vol. 405: K. Devlin and H. Johnsbr&ten, The Souslin Problem.
VIII, 132 pages. 1974. DM 18,-
VoL 406: Graphs and Combinatorics - Proceedings 1973. Edited
by R. A. Bari and F. Harary. VIII, 355 pages. 1974. DM 30,-
Vol. 407: P, Berthelot, Cohomologie Cristalline des Sch@masde
Caracteristiclue 10> o. VIII. 598 pages. 1974. DM 44,-
Vol. 408: J, Wermer, Potential Theory, VIII, 146 pages. 1974.
DM 18,-
VoL 409: Fonctions de Plusieurs Variables Complexes, S@minaire
Franqois Norguet 1970-1973. XIII, 612 pages. 1974. DM 47,-
Vol. 410: S~minaire Pierre Lelong (Analyse) Annie 1972-1973.
VI, 181 pages. 1974. DM 18,-
VoL 411: Hypergraph Seminar. Ohio State University, 1972.
Edited by C. Berge and D. Ray-ChaudhurL IX, 287 pages. 1974.
DM 28,-
VoL 412: Classification of Algebraic Varieties and Compact
Complex Manifolds. Proceedings 1974. Edited by H. Popp. V,
333 pages. 1974. DM 30,-
Vol. 413: M. Bruneau, Variation Totals d'une Fonction. XIV, 332
pages. 1974. DM 30,-
Vol. 414: T. Kambayashi, M. Miyanishi and M. Takeuchi, Uni-
potent Algebraic Groups. VL 165 pages. 1974. DM 20,-
Vol. 415: Ordinary and Partial Differential Equations, Proceedings
of the Conference held at Dundee, 1974. XVII, 447 pages. 1974.
DM 37,-
Vol. 416: M. E. Taylor, Pseudo Differential Operators. IV, 155
pages. 1974. DM 18,-
VoI. 417: H. H. Keller, Differential Calculus in Locally Convex
Spaces. XVI, 131 pages. 1974. DM 18.-
VoL 418: Localization in Group Theory and Homotopy Theory
and Related Topics Battelle Seattle 1974 Seminar. Edited by
P. J. Hilton. VI, 171 pages. 1974. DM 20,-
Vol. 419: Topics in Analysis - Proceedings 1970. Edited by
O. E. Lehto, I. S. Louhivaara, and R. H. Nevanlinna, Xlll, 391 pages.
1974. DM 35,-
Vol. 420: Category Seminar. Proceedings, Sydney Category
Theory Seminar 1972173. Edited by G. M. Kelly. VI, 375 pages.
1974. DM 32,-
VoL 421: V. Po~naru, Groupes Discrets. VI, 216 pages. 1974.
DM 23,-
VoL 422: J.-M. Lemaire, Algebres Connexes et Homologie des
Espaces de Lacets. XIV, 133 pages. 1974. DM 23,-
Vol. 423: S. S. Abhyankar and A. M. Sathaye, Geometric Theory
of Algebraic Space Curves, XIV, 302 pages. 1974. DM 28,-
Vol. 424: L. Weiss and J. Wolfowitz, Maximum Probability
Estimators and Related Topics, V, 106 pages. 1974. DM 18,-
Vol. 425: P. R. Chernoff and J. E. Marsden, Properties of Infin,te
Dimens, onal Ham,ltonian Systems. IV, 160 pages. 1974. DM 20.-
VoI. 426: M. L. Silverstein, Symmetric Markov Processes. IX,
287 pages. 1974. DM 28,-
VoI. 427: H. Omori, Infinite Dimensional Lie Transformation
Groups. XlI, 149 pages. 1974. DM 18,-
Vol. 428: Algebraic and Geometrical Methods in Topology, Pro*
ceedings 1973. Edited by L. F. McAuley. Xl, 280 pages. 1974.
DM 28,-
Vol. 429: L Cohn, Analytic Theory of the Harish-Chandra C-
Function, IV, 154 pages. 1974. DM 18,-
Vol. 430: Constructive and Computational Methods for Differen-
tial and Integral Equations. Symposium 1974. Edited by D. L
Colton and R. P. Gilbert. VII, 476 pages. 1974. DM 37,-
Vol. 431 : Sdminaire Bourbaki - voL 1973174. Expos@s436-452.
IV, 347 pages. 1975. DM 30,-
Vol. 432: R. P. Pflug, Holomorphiogebiete, pseudokonvexe Ge-
biete und das Levi-Problem. VI, 210 Seiten. 1975. DM 23,-
VoL 433: W. G. Fads, Self-Adjoint Operators. VII, lt5 pages,
1975. DM 18,-
Vol. 434: Po Brenner, V. Thomde, and L B. Wahlbin, Besov
Spaces and Applications to Difference Methods for Initial Value
Problems. II, 154 pages. 1975. DM 18,-
VoL 435: C. F. Dunkl and D. E. Ramirez, Representations of Com-
mutative Semitopological Semigroups` VI, 181 pages. 1975.
DM 20,-
Vol. 436: L Auslander and R.Tolimieri, Abetian HarmonicAnalysis,
Theta Functions and Function Algebras on a Nilmanifold. V, 99
pages. 1975. DM 18,-
Vol. 43?: D. W. Masser, Elliptic Functions and Transcendence.
XIV, 143 pages. 1975. DM 18,-
Vol. 438: Geometric Topology, Proceedings 1974. Edited by
L C. Glaser and T. B. Rushing. X, 459 pages. 1975. DM 37,-
Vol. 439: K. Ueno, Classification Theory of Algebraic Varieties
and Compact Complex Spaces. XIX, 278 pages. 1975 DM 28,-
VoL 440: R. K. Getoor, Markov Processes: Ray Processes and
Right Processes. V, 118 pages. 1975. DM 18,-
VoL 441 : N. Jaccbscn, PI-Algebras` An Introduction. V, 115 pages.
1975. DM 18.-
Vol. 442: C. H. Wilcox, Scattering Theory for the d'Alembert
Equation in Exterior Domains. III, 184 pages. 1975, DM 20,-
Vol. 443: M. Lazard, Commutative Formal Groups. II, 236 pages.
1975. DM 23,-
VoL 444: F. van Oystaeyen, Prime Spectra in Non-Commutative
Algebra. V, 128 pages. 1975. DM 18,-
Vot. 446: Partial Differential Equations and Related Topics,
Proceedings 1974. Edited by J. A. Goldstein. IV, 389 pages.
1975. DM 32,-
Vol. 44?: S. Toledo, Tableau Systems for First Order Number
Theory and Certain Higher Order Theories, III, 339 pages. 1975.
DM 30,-
Vol. 448: Spectral Theory and Differential Equations. Proceedings
1974. Edited by W. N. Everitt. XII, 321 pages. 1975. DM 30,-
Vol. 449: Hyperfunctions and Theoretical Physics. Proceedings
1973. Edited by F. Pham. IV, 218 pages. 1975. DM 23,-
Vol. 450: Algebra and Logic, Proceedings 1974, Edited by J. N.
Crossley. VIII, 307 pages. 1975. DM 28,-
Vol. 452: Combinatorial Mathematics 111. Proceedings 1974.
Edited by Anne Penfold Street and W. D. Wallis` IX, 233 pages.
1975. DM 23,-
Vol. 453: Logic Colloquium. Symposium on Logic Held at Boston,
1972-73. Edited by R. Parikh. IV, 251 pages. 1975, DM 25,-
Vol, 457: Fractional Calculus and Its Applications. Proceedings
of the International Conference Held at the University of New
Haven, June 1974. Edited by B. Ross. VI, 381 pages. 1975. DM 32,-
Vol. 458: P. Waiters, Ergodic Theory - Introductory Lectures.
VI, 198 pages. 1975. DM 23,-
This series aims to report new developments in mathematical research
and teaching - quickly, informally and at a high level. The type of material
considered for publication includes:
1. Preliminary drafts of original papers and monographs
2. Lectures on a new field, or presenting a new angle on a classical field
3. Seminar work-outs
4. Reports of meetings, provided they are
a) of exceptional interest and
b) devoted to a single topic.
Texts which are out of print but still in demand may also be considered
if they fall within these categories.
The timeliness of a manuscript is more important than its form, which
may be unfinished or tentative. Thus, in some instances, proofs may be
merely outlined and results presented which have been or will later be
published elsewhere. If possible, a subject index should be included.
Publication of Lecture Notes is intended as a service to the international
mathematical community, in that a commercial publisher, Springer-Ver-
lag, can offer a wider distribution to documents which would otherwise
have a restricted readership. Once published and copyrighted, they can
be documented in the scientific literature.

Manuscripts
Manuscripts should comprise not less than 100 pages.
They are reproduced by a photographic process and therefore must be typed with extreme care. Symbols
not on the typewriter should be inserted by hand in indelible black ink. Corrections to the typescript
should be made by pasting the amended text over the old one, or by obliterating errors with white cor-
recting fluid. Authors receive 75 free copies and are free to use the material in other publications. The
typescript is reduced slightly in size during reproduction; best results will not be obtained unless the
text on any one page is kept within the overall limit of 18 x 26.5 cm (7 x 10V2 inches). The publishers
will be pleased to supply on request special stati~)nerywith the typing area outlined.
Manuscripts in English, German or French should be sent to Prof. A. Dold, Mathematisches Institut der
Universit~t Heidelberg, 69 Heidelberg/Germany, Tiergartenstra6e, Prof. B. Eckmann, EidgenOssische
Technische Hochschule, CH-8006 ZL~rich/Switzerland, or directly to Springer-Verlag Heidelberg.

Springer-Verlag, D-1000 Berlin 33, Heidelberger Platz 3

Springer-Verlag, D-6900 Heidelberg 1, Neuenheimer LandstraBe 28-30
Springer-Verlag, 175 Fifth Avenue, New York, NY 10010/USA

ISBN 3-540-07163-6
ISBN 0-387-07163-6