JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS ARTICLE NO.

204, 854 867 1996.

0472

A Construction of a Non-Measure-Preserving Endomorphism Using Quotient Relations and Automorphic FactorsU

Karma Dajani
Department of Mathematics, Uni ersiteit Utrecht, The Netherlands

and Jane Hawkins

Department of Mathematics, Uni ersity of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27599 Submitted by Dorothy Maharam Stone Received November 10, 1995

1. INTRODUCTION The main purpose of this paper is to construct ergodic, nonsingular, conservative n-to-one endomorphisms which preserve no equivalent finite measure. While such examples are known to exist w6, 9, 12, 19x, our examples are fundamentally different from the existing ones. It was noticed independently in w6, 19x that the two-sided type III Bernoulli shift of Hamachi w9x is the natural extension of an exact two-to-one endomorphism with no equivalent -finite invariant measure. Also in w12x it is shown that Cartesian products of finite measure-preserving exact with type III automorphisms yield examples of n-to-one endomorphisms with no equivalent invariant measure. In Section 5 we give an example of a two-to-one endomorphism whose ergodic nonsingular measure is neither exact nor is it equivalent to a product measure of an automorphism with an exact endomorphism.
* The authors were partially supported by NSF Grant DMS-910356. 854 0022-247Xr96 $18.00
Copyright 1996 by Academic Press, Inc. All rights of reproduction in any form reserved.

CONSTRUCTION OF AN ENDOMORPHISM

855

Our construction is based on the connections between the Radon Nikodym derivative of the original endomorphism and that of its maximal automorphic factor. Using a chain rule for factor maps, we prove a theorem giving a condition for a maximal automorphic factor which does not preserve any equivalent measure to force the endomorphism to have that same property Theorem 4.5 of this paper.. As an important step in proving that the hypotheses of Theorem 4.5 are satisfied, we characterize the maximal automorphic factor of an endomorphism as the action of a quotient relation of two natural relations present in every endomorphism. In 1989, Feldman, Sutherland, and Zimmer studied ergodic relations, subrelations, and quotient relations w8x. They are also discussed in an earlier paper by K. Schmidt w17x. In this paper, we compute the quotient relation of the orbit relation for a noninvertible ergodic countable-to-one endomorphism by a natural subrelation which is trivial precisely when the map is invertible. We show that we obtain the group of integers acting by the maximal automorphic factor of the endomorphism as the quotient. We use duality properties of the relations to characterize exactness of an endomorphism in Proposition 3.8; this extends an earlier result from w1x. Some results about measure theoretic properties of ergodic endomorphisms are proved in Section 2; the characterization of the maximal automorphic factor as an action of a quotient relation is presented in Section 3. The construction is done in Sections 4 and 5. This paper is the third in a study conducted by the authors on nonsingular endomorphisms, their factors, and their natural extensions cf. w5, 6x.. 2. PRELIMINARY MEASURE THEORETIC RESULTS ABOUT NONSINGULAR ENDOMORPHISMS Throughout this paper we will assume that X , B, . is a Lebesgue probability space and T : X X is a nonsingular conservative ergodic endomorphism which is surjective and countable-to-one almost everywhere. The assumption that X . - results in no loss of generality when T is n-to-one; this is shown in Lemma 2.4.. By a result of Rohlin w16x see also w20x., we can assume by replacing X by a measurable T-invariant subset of full measure if necessary that T is forward nonsingular as well, so that T satisfies that for all A g B, A. s 0 m Ty1A. s 0 m TA. s 0. We apply a well-known result of Rohlin w16x to obtain a measurable partition s A1 , A 2 , A 3 , . . . 4 of X into at most countably many pieces satisfying: i. A i . ) 0 for each i; ii. the restriction of T to each A i , which we will write as Ti , is one-to-one;

856 iii. property iv.

DAJANI AND HAWKINS

each A i is of maximal measure in X _D j - i A j with respect to ii.; T1 is one-to-one and onto X .

When we say that the endomorphism T is n-to-one, we mean that every partition s A1 , A 2 , A 3 , . . . 4 satisfying i. iv. contains precisely n atoms and that Ti is one-to-one and onto X for each i s 1, . . . , n. Equivalently, for -a.e. x g X , the set Ty1 x contains exactly n points. For each x g A i , let J T i x . s d Tird . x ., and for x g X , let J T x . s i J T i A i x .. This is the Jacobian function for T , defined by W. Parry w15x, and is independent of the choice of . Our nonsingularity assumptions imply that JT ) 0 a.e. In order to define the Radon Nikodym derivative of T , we consider the following identities holding a.e. see w10x or w18x.:

T x . '

d Ty1 d d d T

x. s

yg T y1 x

JT y . 1 .

1. 2.

T x . '

y1

Tx . s

T Tx .

The function T satisfies for every f g L1 X , B, .,

HX f Tx .

x . d x . s H f x . d x . .
X

3.

For any function satisfying 3. in place of T , we say that is Markovian for T and , and it was shown in w19x that T is the only Ty1 B measurable function which is Markovian for T and . Thus, if is a B measurable function which is Markovian for T and then g L1 X , B, . and E Ty1 B . s T . Here E h Ty1 B . denotes the conditional expectation of h g L1 X , B, . with respect to the sub--algebra Ty1 B. We call the function T the Radon Nikodym deri ati e of T . Similarly, T k x . s d rd Tyk .T k x . is the unique Tyk B measurable function which is Markovian for T k and . It is easily seen that k , x . ' T x . T Tx . T T ky 1 x . is a B measurable function which is Markovian for T k and so that E k , . Tyk B . s T k . From these observations we show that even though we do not have a chain rule we have a related identity which we call the Pseudo-Chain rule Proposition 2.3 below.. Before we can prove that, we generalize a result mentioned in w12x.

CONSTRUCTION OF AN ENDOMORPHISM

857

LEMMA 2.1. Assume that T is a nonsingular endomorphism which is n-to-one on X , B, ., with a -finite measure on B. Then for e ery k g , is -finite on Tyk B. Proof. We fix any k G 1. The endomorphism T k is m s n k-to-one so we can find a measurable partition s A1 , A 2 , A 3 , . . . , A m 4 such that the restriction of T k to each atom is one-to-one. Given any set B g Tyk B : B, since is -finite on B, we consider the set B l A1 s B1 and we . write it as B1 s D j1 s 1 B1, j1 with B1, j1 - . Now the B1, j1 s determine, via symmetric points, a unique countable partition of B l A 2 s B2 in fact of B l A p for every p s 1, . . . , m.. We write B2 s D j1 s 1 B2, j1 , where yk k B2, j1 s T2 T B1, j1 .; if B2, j1 . s for any j1 , we subdivide further if necessary since is -finite. We write B2 s D j1 s 1 D j 2 s 1 B2, j1 j 2 , with B2, j1 s D j 2s 1 B2, j1 j 2 and each B2, j1 j 2 of finite measure; we use this partition to refine the partition of B1 obtained previously. We proceed inductively, refining all previous partitions at each step, and this process stops after a finite number of steps, when we reach A m . Finally we write B s D j1 s 1 D j 2 s 1 D j m s 1 C j1 j 2 . . . j m , with C j1 j 2 . . . j m s D ps 1 B p , j1 j 2 . . . j m g yk T B and C j1 j 2 . . . j m . - . Remark 2.2. Lemma 2.1 is false when T is countable-to-one. A counterexample is X s with s Lebesgue measure, and Tx s tan x. Lebesgue measure is clearly -finite on with respect to the -algebra B of Borel sets, but it is not -finite with respect to Ty1 B. In w10x a generalization of the conditional expectation operator onto 1 Ty1 B, denoted E , is defined as a linear operator on the space of measurable functions on X when T is finite-to-one and is -finite. It is 1 . . defined by E h x s y g T y1 T x . h y .rJ T y .. T . Similarly, we define k . . for each k g , E h x s y g T yk T k x . h y .rJ T k y .. T k . Clearly for k . each B measurable h, the function E h is Tyk B measurable, and if 1 1 . h g L X , B, ., by Lemma 2.1 we have E h s E h Ty1 B .. PROPOSITION 2.3 Pseudo-Chain Rule.. Assume that T is a nonsingular endomorphism which is n-to-one on X , B, ., with a -finite measure on B. For each k g , for e ery i s 1, . . . , k y 1, and a. e. x g X we ha e
k T k s E T i . T ky i ( T i .

If X . - , and T is countable to one, then

T k s E T i Tyk B . T ky i ( T i .
Proof. We fix any k g , and choose any i s 1, . . . , k y 1. It is easily shown that T i T ky i ( T i is a B measurable Markovian function for

858

DAJANI AND HAWKINS

T k and . By w19, Example 1.3x, T k is the unique Tyk B measurable Markovian. Furthermore, since T ky i ( T i is Tyk B measurable we have k k E T i T ky i ( T i . s E T i . T ky i ( T i. The result follows if we show k . that if is Markovian for T k and , then E is also Markovian. Since k . . E x s y g T yk T k x . y .rJ T k y .. T k x ., we have for any f g L1 X , B, .,

HX f x . d x . s HX f ( T
since is Markovian; s
nk

x . x . d x .

js 1

H f ( T k Tjy k x . Tjy k x .
X
k

1 JT k Tjy k x .

d x .

by a change of variables and since Tjyk X 4jn s 1 forms a disjoint partition of X;

nk

js 1

H f x.
X

Tjyk x .
JT k Tjyk x .

d x .

HX f x .

y g T yk x .

y.
JT k y .

d x .

and since T k is Markovian,

HX f ( T

x.

y g T yk T k x .

y.
JT k y .

T k x . d x . .

The following lemma shows that the assumption that is a probability measure does not result in any loss of generality for n-to-one maps. LEMMA 2.4. If T is a nonsingular n-to-one endomorphism on X , B, X . with X a -finite measure, then there exists a finite measure ; X such that T s h ( Trh. X T for some Ty1 B measurable function h. Proof. By definition w11x, a measure s h d X is cohomologous to X if T s h ( Trh. X T . Then is cohomologous to X if and only if h is Ty1 B-measurable w11x, and is finite if and only if h is integrable. Therefore it suffices to find a positive h g L1 X , Ty1 B, X .; such an h exists since X is -finite on Ty 1 B by Lemma 2.1.

CONSTRUCTION OF AN ENDOMORPHISM

859

3. THE QUOTIENT RELATION R T r ST AND THE MAXIMAL AUTOMORPHIC FACTOR FOR COUNTABLE-TO-ONE MAPS T For each countable-to-one nonsingular endomorphism T , we define two amenable equivalence relations R T and ST : R T cf. w4, 10x.. The relation R T : X = X is defined as follows: x , y . g R T if and only if there exist m, n G 1 such that T n x s T m y. We also associate a subrelation ST : R T : X = X by x , y . g ST if and only if T n x s T n y for some n G 1. When the endomorphism T is clearly understood, we write R and S for R T and ST . For x g X , let R x . ' y g X : x , y . g R4 and S x . ' y g X : x , y . g S 4 . One can verify that S x . s D n G 0 Tyn T n x w10x. Similarly we define for each set A g B, R A. ' y : x , y . g R for some x g A4 , and we say that R is nonsingular if A. s 0 m R A.. s 0 for all A g B. We say R is ergodic with respect to . if R A. s A A. s 0 or X _ A. s 0. We have identical definitions for the subrelation S. If T is one-to-one, then S is trivial each equivalence class consists of exactly one point. and R is the usual equivalence relation associated to orbits. In w10x the following connections were proved to exist between the map T and its associated relations R and S. 1. 2. 3. 4. T is nonsingular m R is nonsingular. T is ergodic m R is ergodic. T is nonsingular S is nonsingular and the converse is false.. T is exact m S is ergodic.

Also, it is easily checked that: 5. For all n g , S T n x . s T n Sx . for a.e. x g X . 6. S is a subrelation of R of infinite index if and only if T is not invertible. We apply a technique defined by Feldman, Sutherland, and Zimmer to define the quotient relation R T rST in a measurable way w8x. In general the quotient of a countable ergodic relation by a subrelation can only be described as a groupoid, but in our case we obtain a genuine group, , acting naturally by T on a factor space of X . By convention T 0 x s x. We first define the choice maps of a relation and subrelation, following w8x. Let s A1 , A 2 , A 3 , . . . 4 denote a partition chosen to satisfy condik 4 tions i. iv. of Section 1. For each k g , let k s A1 , Ak 2 , . . . denote k ky 1 yi the partition of X defined by s E is 0 T ; it satisfies conditions k i. iv. for T k , so that the restriction of T k to A1 , written as T1k , is

860

DAJANI AND HAWKINS

one-to-one and onto X . For each n g , we define n : X X by

n x s

T nx
y1 T1k

if n G 0 x if n s yk , k g .

The maps n are called choice maps, and are defined in this way to satisfy the following easily verified properties. LEMMA 3.1. 1. For e ery n g , each n is measurable. 2. For each n g , for e ery x , y . g S, n x ., n y .. g S. 3. For each x g X , R x . s D .. ns y S n x . 4. n ( m x ., nq m x .. g S for all n, m g . We now consider the ergodic decomposition of with respect to the relation S. We obtain a Lebesgue space Y , F , . and a canonical system of measures y 4y g Y on X such that for any A g B, A. s HY y A. d y . w16x. Let : X , B, . Y , F , . be the canonical projection; then y1F : B is the -algebra of S-invariant subsets. We show that the choice maps n4 on X induce automorphisms n on Y , by defining for each n g and y g Y , n x . ' n x .. PROPOSITION 3.2. For each n g , the map n is an automorphism of Y . Proof. We assume first that n g so that n s T n ; then by definition n x . s T n x .. It suffices to show that Tyny 1F s y 1F mod 0.. Let A g Tyny1F ; then A s Tyn B for some B g y1F and SB s B. We write A s Tyn B s Ty nSB s STy n B s SA g y 1F by applying 5. above. The reverse containment is shown by taking B g y1F ; then B s Tyn T n B s Tyn T nSB s Tyn ST n B g Tyny1F. The proposition is trivially true for n s 0 since 0 s Id X . For n g , we now prove the result for yn . We have shown that n is an automory1 y1 phism so n exists and is an automorphism. We claim that n x. s y1 . . . yn x a.e., hence y n s n . By Lemma 3.1 4 , x , n ( y n x g S so that x . s n ( y n x . s n y n x .. for a.e. x. The following result is immediate from Lemma 3.1 and the proof of Proposition 3.2. COROLLARY 3.3. 1. For e ery m, n g , n ( m s m q n ; 2. 0 s Id. Therefore we will write 1 s and n s ( ( n times. ' n ; i.e., the family n4n g defines a -action on the space Y , F , .. The group action generated by is the factor action on Y of the semigroup action generated by T on X .

CONSTRUCTION OF AN ENDOMORPHISM

861

We analyze more closely the partition of X given by s S x .: x g X 4 . In w10x it was shown that there is a one-to-one correspondence between S-invariant sets and sets in the tail field F n G 0 Tyn B. From this we see that the measurable hull of is X s F n G 0 Tyn , where is the point partition, and X generates the tail -algebra F n G 0 Tyn B. We denote by XX the factor space associated with the tail -algebra and by X the associated factor measure. It is clear that every automorphic factor of X , B, ; T . is contained in the tail XX , F n G 0 Tyn B, X ; T X ., which is isomorphic via the identity map. to the space Y obtained above; therefore we have just shown the following. PROPOSITION 3.4. The factor Y , F , ; . is the maximal automorphic factor of X , B, ; T .. Ergodicity of the relation S is the same as exactness of the endomorphism T w10x, so we have the following corollary. We remark that the measures y are not necessarily nonsingular for T . COROLLARY 3.5. For a.e. y g Y , the measure y is tail tri ial for T on X , B .; that is, for e ery A g F n G 0 Tyn B, y A. s 0 or 1. Remark 3.6. 1. If T s f = g and X , B, . s X 1 = X 2 , B1 = B2 , 1 = 2 . with f an automorphism of X 1 , B1 , 1 . and g an exact endomorphism of X 2 , B2 , 2 ., then Proposition 3.4 implies that Y , F , . , X 1 , B1 , 1 .. Furthermore, the factor endomorphism induced by T on Y is f. 2. Since for all n g , y1 n y . s T ny 1 y . for a.e. y g Y , we could define n s ( T n ( y 1 instead of using the choice function of w8x. DEFINITION 3.7 w8, 17x. For a nonsingular conservative countable-to-one endomorphism T , we define the quotient relation R T rST to be the group , endowed with the action generated by on Y , F .. The next proposition is a reflection of some duality properties of relations and subrelations discussed by Feldman, Sutherland, and Zimmer w8, Proposition 1.5x. The only if direction of Proposition 3.8 is also stated in w2x. By , M , . we will denote the Borel space of the integers with the discrete topology and counting measure. We give a short ergodic theoretical proof instead of using w8x. PROPOSITION 3.8. Suppose T is any countable-to-one ergodic nonsingular endomorphism on X , B, .. Then T is exact if and only if the product map : X = X = gi en by T x , m. s Tx , m q 1. is ergodic with respect T to s = .

862

DAJANI AND HAWKINS

Proof. . Suppose that T is not exact. Then there exists a set A, 0 - A. - 1, such that A s Tyn T nA. for all n g . We construct a of the form A s D j g T j A = j4. Clearly nontrivial invariant set for T . A ) 0 and the same is true for its complement, and it is easy to see y1 A . s A. that T is not ergodic. Then . We now suppose that T is exact, but that T yi , say F i G 0 T B is trivial, and there exists a nontrivial invariant set for T yi yi . B M B M B B g = ; in particular, B g F i G 0 T = s F iG 0T = M. Then B is of the form X = C with C g M , but invariance of C implies C s . The contradiction implies the result.

is orbit equivalent i.e., isomorphic as an amenable Remark 3.9. T equivalence relation w4x. to a skew product over the relation R T by a -valued index cocycle w8, Definition 1.4x; in this setting the cocycle appears as the constant cocycle 1 for the endomorphism T . Clearly the commutes with the action of the integers on X = transformation T given by k x , m. ' x , m q k ., k g , which is generated by the automorphism 1 ' . Taking the ergodic decomposition of X = with re gives a Lebesgue space Y , F , . on which is well-defined and spect to T generates an integer action i.e., is invertible.. By the proof of Proposition 3.8 and the discussion above or the duality results in w8x. we see that the automorphism on Y , F , . is isomorphic to the maximal automorphic factor of T on X , B, ..
4. TYPE III MAXIMAL AUTOMORPHISMS AND ENDOMORPHISMS Throughout this section we assume that the endomorphism T on X , B, . is n-to-one and is -finite. Using the notation above, we denote by on Y , F , . the maximal automorphic factor of T . Letting s y1 as before, the function d rd denotes the Radon Nikodym derivative of the measure with respect to . The following identity follows from w3x and is an immediate consequence of Lemma 4.1 below.: for a.e. y g Y , d d

y . s E T

nG 0

F Tyn B y . ' E

nG 0

F Ty n B x .

for any x g X such that x . s y. The next three results are known identities about factors and conditional expectation operators cf. w3x..

CONSTRUCTION OF AN ENDOMORPHISM

863

LEMMA 4.1. For e ery measurable function h with h ( T g L1 X , ., and a.e. y g Y , we ha e E h ( T

nG 0

F Tyn B y . s E

h T

nG 0

F Ty n B

( y . d rd . y .

s ET y1 h

nG 0

F Tyn B

( y. .

We recall that the measure on X , B . disintegrates over the factor Y , F , . by B . s HY y B . d y ., and for any h g L1 X , B, ., the function E h F n G 0 Tyn B . y . s Hy1 y h x . d y x .. LEMMA 4.2. The disintegration of satisfies y1 y Ty1 ; y for a.e. y g Y. Proof. For any set B g B, we apply Lemma 4.1 to the function h s B and evaluate the equality at y1 y. This gives, for a.e. y,

y1 y Ty 1 B .

d y1 d

y. s H

Xl y1 y

B x .

d Ty1 d

x . d y x . .

Since both d y1 rd . y . and d Ty1 rd . x . are assumed to be strictly positive functions, and B G 0 everywhere, the result follows immediately. We have the following chain rule for nonsingular endomorphisms decomposed over a factor. P ROPOSITION 4.3. For a.e. x g X , d T y 1 r d . x . s d y1 rd . x . d y1 x .Ty1 rd x . . x .. In addition,

T x . s

d d

x.

d x . d x . Ty 1

(T x . .

Using the terminology introduced by Krieger for integer actions and originating with von Neumann factors., we define an ergodic automorphism or endomorphism T on X , B, . to be type III if it admits no invariant -finite measure ; w14x. In w10x, T .y1 is defined to be a 1 . conditional coboundary for T if T .y1 s h ( T .rE h for some measurable function h, and it is proved there that T admits an equivalent -finite invariant measure if and only if T .y1 is a conditional coboundary. A conditional coboundary is a coboundary if and only if h is Ty1 B measurable w11x. It was shown earlier that an ergodic automorphism T is type III if and only if T is not a coboundary cf. w14x..

864

DAJANI AND HAWKINS

LEMMA 4.4. If T x . s d rd . x . for a.e. x g X , then T is a T-coboundary on X if and only if d rd . x . is a -coboundary on Y for . If both are coboundaries, the transfer function is F n G 0 Tyn B-measurable. Proof. . It suffices to show that if T is measurable with respect to y1F : B, and is a coboundary, then the transfer function is y1F measurable as well. We suppose that T s h ( T .rh; i.e., h x . s h ( T .rT . x .. We show that for any x g X , and any y g ST x ., h x . s h y .. We first suppose that Tx s Ty. Then hTx . s hTy ., and T x . s T y . by assumption; therefore h x . s h y .. Suppose we have shown that if T k x s T k y for any k F n, then h x . s h y .. We now assume that T nq 1 x s T nq 1 w. Then T n Tx . s T n Tw ., so hTx . s hTw ., and we have h x . s h ( T .rT . x . s h ( T .r T . w . s h w .. By induction, we have that h is constant on ST x . and by ergodicity of the relation ST , it follows that h is constant x-a.e. Therefore, the transfer function depends only on x as claimed. . This direction is immediate since x . s Tx .. THEOREM 4.5. Suppose that T is a conser ati e ergodic n-to-one nonsingular endomorphism of X , B, . with maximal automorphic factor , and d yrd y Ty1 ' 1 for a.e. y. Then is of type III if and only if T is of type III. Proof. . Suppose T admits a -finite equivalent invariant measure X ; . We can assume without loss of generality by Lemma 1.4 that is recurrent see w18x or w19x for definitions. so that T s h ( T .rh for h a Ty1 B-measurable and B measurable . function. By Proposition 4.3, d d d x . d x . Ty 1

T x . s

x.

(T x .

x . for a.e. x. h d It follows from Lemma 4.4 that d rd . x . is a coboundary, which is a contradiction. . If admits a -finite equivalent invariant measure then d r d . x . is a coboundary. Then Proposition 4.3 and Lemma 4.4 imply that T cannot be of type III. which implies that

x. s

h(T

CONSTRUCTION OF AN ENDOMORPHISM

865

5. AN EXAMPLE OF AN ENDOMORPHISM WITH NO EQUIVALENT INVARIANT MEASURE We construct a two-to-one endomorphism T on the product space 4 4 X s Y = Yq' ns y 0, 1 n = n s 0 0, 1 n and a Borel measure on X . with the following properties: 1 T is the product of the two-sided shift with the one-sided shift; 2. T is nonsingular, conservative, and ergodic with respect to ; 3. T admits no -finite invariant measure equivalent to ; 4. T is neither exact with respect to , nor is the product measure of an exact with an automorphic measure. We denote by B = Bq the product Borel -algebra on Y = Yq. We define the measure on Y , B . to be the type III two-sided Bernoulli shift measure constructed by Hamachi in w9x. Then the measure on Y = Yq will be of the form C . s HY y C l y1 y . d y .. We specify the measures y on Yq, Bq . according to the following algorithm. We fix any g 0, 1.. We define two measures 0 and 1 on the space 0, 14 by 0 0. s 0 1. s 1r2, and 1 0. s 1r1 q ., 1 1. s r1 q .. For each y g Y , we define y to be the infinite product measure given by y s yi . is 0 . That is, we consider y s . . . , yy 1 , y 0 , y 1 , . . . , yn , . . . and the i th factor in the measure y is j if and only if yi s j. Each y will be an infinite product of factors of two different measures on Yq and, by results of Kakutani w13x, for a.e. y g Y , y will be singular with respect to the shift . We denote the invertible shift on Y by ; i.e., y . i s yiq 1. It follows y iq 1 yi that y s s is 0 is 1 . We now restrict our attention to a single fiber y4 = Yq with y a measure on D, and we compute y y 1. Suppose C g D is any cylinder set; then we can write C s z g Yq : z 0 s i 0 , z1 s i1 , . . . , z n s i n 4 , and

y1 C . s z : z 0 s 0, z1 s i 0 , . . . , z nq 1 s i n 4
j z : z 0 s 1, z1 s i 0 , . . . , z nq 1 s i n 4 . Therefore y y1 C . s y 0 0 . y 1 i 0 . y 2 i1 . y nq 1 i n . q y 0 1 . y 1 i 0 . y 2 i1 . y nq 1 i n .
n

y i j . s y C . .
jq 1

js 0

866

DAJANI AND HAWKINS

Since an infinite product measure is completely determined by its values on cylinder sets, we see that the two measures are equal. Consequently, d y y1 rd y s 1. To check that the endomorphism defined in this way is conservative, it is enough to see that d = .yn rd . y, z . s d yn rd . y . for all n G 0, so T n y, z . s T n, y, z .. s d nr n . y . s d . y . for all n G 0. Hamachi computes that ns 0 d rd w x a.e. 9, p. 279 , which gives the result. By Theorem 4.5, T admits no -finite equivalent invariant measure. To prove that T is ergodic with respect to , we apply Proposition 3.4 and use the fact that all T invariant sets must be in Y , B . since R T rST is just . The ergodicity of with respect to gives the result. The fact that is not equivalent to any product measure also follows from Proposition 3.4 and the uniqueness of the maximal automorphic factor; details of this argument are given in w5x.

REFERENCES
1. J. Aaronson, Ergodic theory for inner functions of the upper half plane, Ann. Inst. H. Poincare Sect. B 14 1978., 233 253. 2. J. Aaronson, M. Lin, and B. Weiss, Mixing properties of ) Markov operators and ergodic transformations and ergodicity of Cartesian products, Israel J . Math. 33 1979., 198 224. 3. R. Butler and K. Schmidt, An information cocycle for groups of non-singular transformations, Z. Wahrsch. Verw. Gebiete 69 1985. 347 360. 4. A. Connes, J. Feldman, and B. Weiss, An amenable equivalence relation is generated by a single transformation, Ergodic Theory Dynamical Systems 1 1981., 431 450. 5. K. Dajani and J. Hawkins, Rohlin factors, product factors, and joinings for n-to-one endomorphisms, Indiana Uni . Math. J. 42 1993., 237 258. 6. K. Dajani and J. Hawkins, Examples of endomorphisms and their natural extensions, Proc. Amer. Math. Soc. 120 1994., 1211 1217. 7. S. Eigen and C. Silva, n-Stack structures, in Measure and Measurable Dynamics, pp. 131 140, Contemp. Math., Vol. 40, Amer. Math. Soc., Providence, 1989. 8. J. Feldman, C. Sutherland, and R. Zimmer, Subrelations of ergodic equivalence relations, Ergodic Theory Dynamical Systems 9 1989., 239 269. 9. T. Hamachi, On a Bernoulli shift with non-identical factor measures, Ergodic Theory Dynamical Systems 1 1981., 273 284. 10. J. Hawkins, Amenable relations for endomorphisms, Trans. Amer. Math. Soc. 343 1994., 169 191. 11. J. Hawkins and C. Silva, Remarks on recurrence and orbit equivalence of nonsingular endomorphisms, in Dyn. Sys. Proc. Univ. Md., pp. 281 290, Lecture Notes in Math., Vol. 1342, Springer-Verlag, New YorkrBerlin, 1988. 12. J. Hawkins and C. Silva, Noninvertible transformations admitting no absolutely continuous -finite invariant measure, Proc. Amer. Math. Soc. 11 1991., 455 463. 13. S. Kakutani, On equivalence of infinite product measures, Ann. of Math. 49 1948., 214 224. 14. W. Krieger, On the Araki-Woods asymptotic ratio set and nonsingular transformations of a measure space, in Lecture Notes in Math., Vol. 160, pp. 158 177, Springer-Verlag, New YorkrBerlin, 1970.

CONSTRUCTION OF AN ENDOMORPHISM

867

15. W. Parry, Entropy and Generators in Ergodic Theory, Benjamin, Elmsford, New York, 1969. 16. V. Rohlin, On the fundamental ideas of measure theory, in Amer. Math. Soc. Transl., Vol. 71, pp. 1 54, Amer. Math. Soc., Providence, 1952. 17. K. Schmidt, Strong ergodicity and quotients of equivalence relations, Proc. Centre Math. Anal. Austral. Nat. Uni . 16 1988., 300 311. 18. C. Silva, On -recurrent nonsingular endomorphisms, Israel J . Math. 61 1988., 1 13. 19. C. Silva and P. Thieullen, The subadditive ergodic theorem and recurrence properties of Markovian transformations, J. Math. Anal. Appl. 154 1991., 83 99. 20. P. Walters, Roots of n:1 measure-preserving transformations, J. London Math. Soc. 44 1969., 7 14.