Lecture Notes in Mathematics Edited by A. Dold and B Eckmann 729 Ergodic Theory Proceedings, Oberwolfach, Germany, June 11-17, 1978 Edited by M Denker and K Jacobs Springer-Verlag Berlin Heidelberg New York 1979Editors Manfred Denker Institut fur Mathematische Statistik und Wirtschaftsmathematik der Georg-August-Universitat Lotzestr. 13 D-3400 Gittingen Konrad Jacobs Universitat Erlangen-Nurmberg Mathematisches Institut Bismarckstr. 11/2 D-8520 Erlangen AMS Subject Classifications (1970): 28A65, 54H20 ISBN 3-540-09517-9 Springer-Verlag Berlin Heidelberg New York ISBN 0-387-09517-9 Springer-Verlag New York Heidelberg Berlin Library of Gongross Cataloging in Publication Data Main entry under ttle: Ergodic theory. (Lecture notes in mathematics : 728) Includes bibliographical references and index. 1. Ergodic theory--Congresses. 2: Topological dynarnies--Congresses. |. Denker. Manfred. 1944- I Jacobs. Konrad, 1928- I. Sones: Lecture notes in mathematics (Berlin); 728. ©A3.L28 no. 729 [OASIS] 5107.85 [518°42] 79-17368, This work 1s subject to copyright. All rights are reserved, whether the whole or part of the material 1s concerned, specifically those of translation. reprinting. re-use of illustrations, broadcasting, reproduction by photocopying 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 fec 1s payable to the publisher. the amount of the fee to be determined by agreement with the publisher. © by Springer-Verlag Berlin Heidelberg 1979 Printed in Germany Printing and binding: Beltz Offsetdruck, Hemsbach/Bergstr. 2141/3140-543210Ded cated to the memory of Rufus Bowen Oa7 | 973Introduction During the week of June 11-17, 1978, a conference on Eraodic Theory was held in the Mathematisches Forschungsinstitut at Oherwolfach, West Germany, in which mathematicians from Austria, Canada, France, Great Britain, India, Israel, Japan, Poland, United States, and West Germany participated. The main topics consisted of recent re- search, and significant results obtained during 1977 and 1978 were presented. This volume contains those of the results which will not be published elsewhere, The organizers would like to thank the MFO for their support of the conference, the Sprinaer-Verlaq for the publication of these articles, and in particular each of the parti- cipants for contributing to the success of the conference. Géttingen/Erlangen, 15th February, 1979 Manfred Denker Konrad JacobsContents List of participants List of talks Aaronson, J On the categories of ergodicity when the measure is infinite 1 Adler, R.L.: A selection of problems in topological dynamics a Akcoglu,M.A. Pointwise ergodic theorems in L, snaces 13 Alpern,S. Generic properties of measure preserving homeomorphisms 16 Auslander,J. On disjointness in topological dynamics and ergodic theory 28 Feldman,J. Reparametrization of probability preserving n=flows 34 Hamachi,T./ Pundamental homomorphism of normalizer Osikawa,M.: group of ergodic transformation 43 Helmberg,G Some remarks on ¢-independence of parti- tions and on topological Rochlin sets 58 Hofbauer, F Maximal measures for piecewise monotoni- cally increasing transformations on [0,1] 66 Le@rappier,F. A variational principle for the topological conditional entropy 78 LinM Weak mixing for semi-groups of Markov oper- ators without finite invariant measures 89 Lind,D.A.t Ergodic group automorphisms and specification 93 Marcus ,B./ Measures of maximal entropy for a class of Newhouse,S.: skew products 105 Marcus,B./ Petersen,K.: Balancing ergodic averages 126 Misiurewicz,™. Invariant measures for continuous transforma tions of [0,1] with zero topological entropy 144 Moulin Ollagnier,J./ Pinchon,D.: Dynamical systems of total orders 153 Newhouse, S see Marcus and Newhouse Osikawa,M.: see Hamachi and OsikawaParry ,W.: Parry,W.: Petersen,K.: Pinchon,D.: Schmidt ,K.: Schweiger, F. Thaler,M.+ vu An information obstruction to finite expected coding length The Lorenz attractor and a related population model see Marcus and Petersen see Moulin Ollagnier and Pinchon Unique ergodicity and related problems A modified Jacobi-Perron algorithm with explicitly given invariant measure Ergodic properties of real transformations 163 169 188 199 203Participants J.Aaronson (Rennes) R.Adler (Yorktown Heights) M.A.Akcoglu (Toronto) St.Alpern (London) J.Auslander (Maryland) A.Beck (Madison) A.Bellow (Evanston) J.R.Blum (Tucson) S.G.Dani (Bombay) M,Denker (Gdttingen) Y¥.Derrienic (Rennes) J.Feldman (Berkeley) H.Haller (Erlangen) T.Hamachi (Fukuoka) G.Helmberg (Innsbruck) M.Herman (Paris) F.Hofbauer (Wien) K.Jacobs (Erlangen) A.B.Katok (Paris) M.Keane (Rennes) G.Keller (Minster) U.Krengel (G8ttingen) W.krieger (Heidelberg) F.Ledrappier (Paris) M.Lin (Beer-Sheva) D.A.Lind (Seattle) B.Marcus (Chapel Hill) M.Misiurewicz (Warschau) J.Moulin Ollagnier (Paris) J.Neveu (Paris) R.Nirnberg (G8ttingen) W.Parry (Coventry) K.Petersen (Chapel Hill) D.Pinchon (Paris) B.Roider (Innsbruck) K.Schmidt (Coventry) F.Schweiger (Salzburg) C.Series (Cambridge) K.Sigmund (Wien) M,Srorodinsky (Tel-Aviv) L.Sucheston (Columbus) wW.Szlenk (Warschau) M.Thaler (Salzburg) J.-P.Thouvenot (Paris) X.M.Wilkinson (Nottingham)List of talks Aaronson, J Adler,R. Akcoglu,M.+ Alpern,S. Auslander,J.: Bellow,A.: Blum,R. Dani,S.G.: Feldman,J.: Hamachi, T.: Helmberg,G Hofbauer ,F Katok,A.B. Katok,A.B.: Keane M. Keller,G.: Ledrappier,F.: Lin,M Lind,D.A.: Mare 1Bet Misdurewicz,4.: Moulin Ollagnier, Neveu,J. Niirnberg,R.: Parry,W.: Petersen,K.: Schmidt ,K.: Schweiger ,P-: Series,C.: Smorodinsky ).+ Sucheston,L.: About transformations preserving infinite measures Topological entropy and equivalence of dynamical systems Pointwise ergodic theorems Generic properties of measure preserving horeomor- phisms Disjointness in ergodic theorv and topolodical dyna~ mics Another look at a.s.converaence Pointwise ergodic summahility methods on LCA aroups Invariant measures of horospherical flows Reparametri zation of prohability-preserving n-flows Fundamental homomorphism of normalizer orcur of ex- go@ic non-sinaular transformation On e-independence and topological Yohlin-sets Das maximale Mas fiir cie Transformation Bix > px + a(moe 1) Lyapunov exponents, entropy and invariant foliation in smooth ergodic theory Generalized rotation numbers for Anosov Flows Bernoulli schemes of the same entropy are finitarily isomorphic Piecewise ronotonic functions and exactness A variational principle for topological conditional entropy Weak mixing Specification for commact aroup automorphisms Topological entropy of some skew products Invariant measures for continuous transformations on [0,1] with zero topological entropy J.: A new proof of F.rélner's result: Countable amenable groups have an ameanina filter On the filling schema ane a simple proof of the Chacon-Ornstein thcorem. Constructions of strictly eraodic systems which are not loosely Bernoulli Generic properties of endororphisms Balancing ergodic averages Unique ergodicity The "jump transformation" and its applications Foliations and ergodic equivalence relations Bernoulli factors that span a transformation Operator ergodic theorems for superadditive processesxil SzlenkW.: An example of a ¢ynamical system with positive topological sequence entropy and zero metric sequence entropy Thaler,M.: Ergodische Figenschaften von reellen Transformatio- nen Wilkinson, K.: Stopping times and towersON THE CATEGORIES OF ERGODTCITY WHEN THE MEASURE TS INFINITE Jon AARONSON Let (X,B,y) denote the real line cquiped with Lebesgue mea- sure, and let G denote the group (under composition) of automorphisms (invertible, measure preserving transformations) of (X,B,1) endowed with the topology of convergence 1 ou 1 Gel Ad Tay + u(t, Ae TA) +0 VAc $= (Ae B: uA) < #} It was shown in [g]! that the ergodic automorphisms are re- sidual in G Here we prove category theorems Cor some stronger versions of ergodicity. Let Te G be ergodic. Then T is conservative, and, by Hop£'s ergodic theorem ({E], p.49) net met, 4 Y ccm*xa/ Eo g(T*x) + fy fdv/fy edu acc. WygeL fg #0 K50 k=0 1 The results in [5] actually apply to the semigroup of endomorphisms of (X,B,u). The proofs of all the results in that paper quoted here can easily be adapted to G (see the chapter on weak topology in [a].Note that lopf's theorem gives no information on the "abso- n-1 lute" asymptotic growth of che sums corks) . (€ > 0). In the stronger versions of ergodicty considered here, such information is (defined to he) available. Definition (a) Let TeG and in, sm ce. he will say that fm* is T if theve are constants d,~ such that - means thitt lere (and hence all) proba- bility measures Pos: PC - et 0 Fi + 0. (b) We will say that T if N is good for T. 1€ Te G has a good sequence .ny , then T is ergodic and the sequence “ly: satisfying (1) is defined uniquely up to asynp- totic equality. We will call this sequence the resun, subsequence oF a {ng Cor the ve ture Peet Tie (ot Ny. Definitions 1 (a) and (b) are applicable to automorphisms of a Linite measure space where they hoth coincide with ergodicity. Thus we only consider automorphisms of (X,B,u), where these definitions do not coincide with ergodicity (see “A]). We prove Theorem 1: The automorphisms with good sequences are residual in 6 and Theorem 2: The homogencous automorphisms are dense but meagre in G.We will need the 1, ket), Suppose Tc G is aperiodic (i.e. «(Ix Yn > 1) then (1) is dense in G where C(T) is a measurcable, invertible, measure miltiplying map. The existence of homogencous automorphisms was established by theorem 4.4 of [A?. Any homogencous automorphism is clearly ergodic and hence aperiodic. Also, if Ty i homogeneous, then so is every member of C(T,). Thus, hy the conjugacy Lemma, the komegencous cr dense eat Proof of theorem 1 het T) © G be homogencous with return sequence a, - As mentioned above, (Ig) is dense in G. Let not k dee ee Then, C(T)) ¢ Q@, and every member of CL has good sequences. We nrove theorem I by by showing that CL is a G, set in G Let P be a probability measure. [t is not hard to show that sets of the form « {6} are closed in G where is a dense subcollection of ¥ (i.e. nel +0), and o,0) = Ff 1, o TK, the a keo 4is Inis proof is therefore completed by establishing that g a It is not hara to check that tv) + fA) ave. Wor 1) from which the inclusion @e=(@! follows immediately. Now suppose T (2! . From the ahove representation of CQ! we see that Jot oT = 2 ae. Yoo d Thus, T is conservative, and, llopf's theorem(for conserva- tive but possibly non-ergodic transformations) yiclds. na1 oi ' TRI /e, 0) 9) hCG) acc. VF € je where hy(f) o T= hi (f) and fy h(t) du = fy fd. By (2), we now have : nt * 1 } fe +nce) ace. vet! where ny, 70 f WC) de = wal) fytae A : Since {A} is dense in F: nC) = J and ce CeProof of theorem 2 As remarked before, the horogeneous automorphisms are dense in G. We show that the non-homogeneous automorphisms are residual in G. Let TEG be homogeneous with return sequence (a, }- Then, by theorem 4.1 of (A), a, is regularly varying with index n 1 and, in particular, a,/ fw > . Hence : forno BE Fu (B)= 1 and no subsequence does = +0 ae Thus, if we fix an BEY, u (B) and tet © (Ey) denote those automorphisms T€G satisfying the convergence (3) along some subsequence n, , then & consists entirely of non-homogeneous automorphisms. It is not hard to see that : ey a f c. f\ U ures : ply > Frac 172) m=10 n=m ‘Keo where P~ uand P(X = , so€ is a Ggin G. Na if TéG@were ergodic, then, by Hopf's thoremthe convergence (3) would be satisfied by T for every BEF, and hence by every member of C(T) ¥Y BEY along some subsequence ne Thus, for ergodic TEG : C(r1< GB . So, by the conjugacy Lomma, the existence of an ergodic Té G would imply the density (and hence residuality) of G in G. To complete this proof, here is an ergodic Markov shift in © : vy (nt) 44 ns 0). Then n=O is a (bounded) Kaluza sequence (i.e ,,,/ u, * ) and hence }, a renewal sequence —null-recurrent since u,* 0 and Eu, =. Thus ("C"), there is an irrednc ible, null-recurrent stochastic matrix Q =£q) i,J€ with tac property that (n) : q eu Tad Mn t Since Q is irreducible and null-recurrent, the two- sided shift of Q: T, is an ergodic automorphism of an infinite measure space isomorphic to (X, B, ") and may therefore be regarded as an (crgodic) clement of G. It remains to show that T,€ ©: By construction, 1 A€ Ff, 4(A) > O such that on GAT, MY =u, vaHence 1/4 > 0, s0 Jon >= such that = 0 a.e. on A. (In this case, we may take n, = k) The set on which this convergence takes place is T,-invaciant, and, containing A # @, must by the ergodicity of be X . The Hopf ergodic theorem now establishes that T, © @- Do REFERENCE: [A] J. AARONSON : On the pointwise ergodic behaviour of transforma tions preserving = measures. To appear in Israel Journal of Maths. [c] K.L. CHUNG : Markoy Chains with stationary transition probabili- ties. Springer 104 Heidelberg (1960) (e] E. HOPF : Ergodentheorie. Chelsea (1948). fy} P. HALMOS : Lectures on ergodic theory. Chelsea (1956) [Ka] T. KALUZA : Uber die Koeffizienten reziproker Potenzreihen Math. Z. 28 p. 161-170 (1928). [kr] K. KRICKEBERG : Mischende Transformationen auf Mannigfaltigkei- ten unendlichen Masses : Z. Wahrsch. yerw. Geb. 7 (1966 = 161-181. [s] _U. SACHDEVA : On category of mixing in » measure spaces. Math. Systems theory, 5 (1971), p-319-330. Jon Aaronson Laboratoire des Probabilites, Universite de Rennes Avenue du General Leclerc F-35031 Rennes cedexA Selection of Problems in ropological Dynamics Roy L. Adler Mathematical Sciences Department IBM Thoras J. Watson Research Center Yorktown leights, New York Several results of ergodic theory strongly suggest analogues in topological dynamics. In this talk we shall indicate some unsolved ones which are related to the notion of topological entropy. We denote a general compact dynamical metric space and } a honeomorphism of X onto itself. Let h(X,9) denote its systen by (X,2) where X is a compact topological entropy. Two dynamical systems (x,t) , (Y.¥) are called topologically conjugate, (X,+) = (8) , if there exists a homeomorphism 8 of X onto ¥ such that af = 2. Lt follows easily from the definitions that = is an equivalence relation and topological entropy an invariant. We shall be concerned mainly with the following special models of dynamic systems: full symbolic shifes (s“,») where § is a finite set of symbols with the discrete topology, S” the space of doubly infinite sequences of elements of $ with the product topology, and the shift transformation: subshifts (x,7) where X is a closed s-invariant eubsct of s¢; subshifes of finite type where Ko SF CAG Cyeeel,) and CyyeesC, are a finite nunber of finite cylinder sets; topological Markov shifes where X= (1) = Gee Co gitgiXyreed? x8 ty mL, rez} and T= Ce.) is a zero~one transition matrix (a subshift of we %ntl Finite type is always topologically conjugate to a topological Markov shift usually where over a different symbol set, and conversely); and toral automorpht, x= ez" and ¢ is given dy a unimodular matrix. I. The Conjugacy Probler conjugacy and constructing conjugating maps between specified systems? For sub- Is there an algorithm for deternfning topological shifts of finite type, R. Willfams [Wi] has ericd but not yet successfully to reduce the problem to one of solving diophantine equations. For toral automorphisms the program has been slightly more successful. Reduction to a diophantine problem was accomplished in [AP] and [Ar]. ‘wo toral automorphisms are topologically conjugate Af and only if their corresponding unimodular matrices are conjugate elements in cL(n,Z). Here we are in a special case of the general problem of determining con- jugate elements in a finitely presented group which is the second of Delin's three fundamental decision problems [MKS, p. 24, p. 168], the first being the word prob- lem. For GL(n,2) it has only been settted for n= 1,2, For ne 2 it is a prob- Jem of solving cuadratic equations in two variables and connected with Pell's equation.IL. A Weaker Isomorphism. Greater success can be achieved with the program attempted in I by using a slightly weaker notion of equivalence which we call almost topological conjugacy. In terms of this relation we can prove a topological analogue to Ornstein's isomorphism theorem for Markov shifts [0]. The precise measure theoretic version is to be found in [ASS]. Isomorphism Theorem for Topologi logical Markov shifts are almost topologically conjugate (with the conjugating maps al Markov Shifts [AM]. Two irreducible topo~ constructed by an algorithn) if and only if they have the same period and topologi- cal entropy. In order to give a definition of almost topological conjugacy we need two other notions which ve shall leave unspecified for the moment. We need a condition of indecomposability restricting the class of dynamical system: and a notion of Regligibiltty for sets where maps fai} to be one-to-one. For indecomposable dynari- cal systems we say (Y,t) is an almost conjugate extension of (X,4) or (X,3) an almost conjugate factor of (¥,t) if there exists a negligible invariant set NX and a continuous boundedly-finite-to-one map * of Y onto X which satisfies 1, gr = ty and maps Y. one-to-one onte X: tems are said to Two dynamical sy be almost topologically conjugate , (%,¢) ~ (¥,2), tf they have a common Jugate extension. 1t turns out we have more than one version of the relation ~ de- pending on how we specify indecomposability and negligibility. First vers restrict dynamical systems to those supported by probability measure which is positive on open sets and take as negligible sets those which have measure zero with respect to all such measures. Second version: we restrict dynamical systems to topologically transitive nonwandering ones and take as negligible sets those containing points whose future orbit or past orbit hits some open set a finite number of times. Third version: we take the same class of dynamical systems and choose in each the set of nendoubly transitive points as a single universal negligible set. 1n the last two versions the negligible sets are of first category besides being of measure zero with respect to any ergodically supporting measure if such exist. In each version ~ is an equivalence relation and topological entropy an invariant, although these facts are no longer such easy consequences. The proof with respect to version one was given in [AN]. The de~ tails shoul4 be carried out for the others. In any case they all lead to the sane isomorphism theorem for topological Markov shitts with essentially the same proof. One might ask which version is best. All three equivalence relations are topo~ logical in character, but the first uses the concept of measure. ‘The second yields the strongest relation, while the third seems the most appealing. There may well be others, but one must be careful. For example, if we were to call two topo~ logically transitive nonwandering systems equivalent if they are topologically conjugate with respect to the subspace of doubly transitive points, then we get the10 same isonorphism theorem for subshifts of finite type; but for general subshifts topo logical entropy will not even be an invarlant. TIT. oral automorphisms. R. Bowen has proved that hyperbolic toral automor- phisms are almost conjugate (all versions) actors of aperiodic topological Markov shifts (B-1, B-2). Thus we have the following corollary to the isomorphism theorem. Corollary. Iwo hyperbolic tora] automorphisms are almost topologically con Jugate tf and only if they have the same eutropy. What about nonhyperbolic toral automorphisms? Using the notion of specification D. Lind [L-2] has shown that nonhyperbolic toral automorphisms are not almost con- Jugate factors of subshifts of Finite type; so the Isomorghism theorem of [AM] cannot be applied. Perhaps che answer lies in some kind of generalization dealing with skew products of topological Markov shifts and rotattons on the circle. This leads to the following question. Given a noahyperbolic tora] automorphism, does there exist a topological Markov shift with the sane entropy? Constructing transi~ tion matrices T with specified entropies (Perron characteristic values) is a difficult problem and not much ix known. However one can find transition matrices which have entropies equal to those of hyperbolic toral autororpl isms by using Bowen's result. Thus we are Jed to an algebraic number theory question. Given a nonhyperbolic tora] automorphism, docs there exist a hyperbolic one with the same entropy? Next what about a converse to Sowen's results? When is an aperiodic copological Narkov shift an almost conlugate extension of @ toral automorphism? If this could be proved for en infinite number of nxn transition matrices of the o1...0 ooL form Te a1 L10...0 whose entropy approaches 0 as a +* , it would solve the problem of the entropy infimum of toral automorphisms discussed in [L-1]. Le goes without vaying that we can ask similar questions for natural extensions of toral endonorpiisms and other group automorphisms. IV. Nonglgebraic Entropy. So far we have only dealt with systems whose en- troples are logarithns of algebraic nunbers, Can we get {sonorphism results for systens whose entropfes are logarithms of transcendental numbers? Is there a wider class of subshifts beyond sofic ones [CP,Ke] tor which topological entropy ts a complete almost topological coujugucy invariant? 1 believe a promising line cens defined {n terms of renewal blocks such would be to investigate symbolic sy ag those associated with the g-traustormation. (P]." ‘actors and Extensions. What are the analogues to Ornstein's criteria for determining when a process is isomorphic to a Bernoulli shift? In some sense the work of Bowen [B-1, B-2] on diffeomorphivms fits this description. Are there criteria for determining when one subshift of finite type is an almost conjugate factor of another? What can be said about the class of shifes which have no nontrivial almost conjugate factors? A factor is trivial if it is topologically conjugate to the extension in question. Does the full 2-shift have nontrivial almost conjugate factors? VI. n-point Extensions. We say (¥,%) is an almost n-point_extension of (X,9) if there exists a negligible invariant set N=X and a con finite-to-one map * of Y onto X which satisfies $1 = xy and maps Y~1—'N n-to-one onto X-N. Two n-point extensions (¥,¢), (4st) of (X,2) are called equivalent if they have a common almost conjugate extension (Wt,:) such that the factor maps down to (X,;) comme. Tn [AM] we prove all 1-point extensions are equivalent. However the problem becomes more interesting for n 23. What we are after is the analogue to Rudolph's theorem [R] which classifies k-point extensions of Bernoulli shifts by an invariant which iy a certain algebraic structure In the symmetric group on k-points. Rudolph's invariant will be one here also, but is it complete? This result could then be used to study the problem of n-point extensions over different systems. An n-point extension (Y,») of (X,¢) is said to be ) af yy) and (W,=), have a equivalent to an n-point extension (Wjz) of (Z common almost conjugate extension which iv an n-point extension of a common almost conjugate extension of (X,2) and (Z,r) and the factor maps in the diagram commute VII. Other Group Actions. Are there isomorphism theorems for other group actions such as flows or 7 etions? As a first step for Z*-actions one should examine the system of Narkley and Paul [NP]. What is che relation between topological entropy and VIII. Kriege’ the number of symbols necded to represent a dynamical system? Is the analogue to Theorer Krieger's theorem [K] on generators true? Namely -~ given A(T),3), does there where the number of symbols in S is exist a subshift of finite type in 6,6 the next larger integer to ey there may be @ vay to use the circles of Williams (See [AMJ) to get such a result.12 [aN] R. L. Adler and B. Marcus, Topological entropy and equivalence of dynamical systems (preprint), to appear in Nenoirs amer. Nath. Soc. (aP] RL. Adler and R, Palais, Homeororphic conjugacy of automorphisms on the torus, Proc. of Amer. Math. Soc. 16 (1965) 1222-1225. [ass] RL. Adler, P. Shields, and M, Smorodinsky, Irreduclble Markov Shifts, Ann. of Math. Star. 43 (1972) 1027-1029. [ar] D. Z, Arov, Topological sinititude of automorphisms and translations of compact commutative groups, Uspelil Nat. Nauk 18 (1963) 133-138. [B-1] R. Bowen, Markov partitions and minimal sets for axiom A diffeonorphisns, amer. J. Math. 92 (1970) 907-918, [8-2] R. Bowen, Markov partitions for axion A diffeonorphisms, Amer. 92 (1970) 725-747. Math . (er) (Kk) W. Krfeger, On entropy and generators of measure-preserving transformations, Trans. Aner. Math. Soc. 199 (1970) 453-464. [L-1] D. A, Lind, Ergodic automorphisms of the infinite torus are Bernoulli, Israel J. Math. 17 (1974) 162-168, [1-2] D. A. Lind, Ergodic group automorphisms and spectfication, (this conference) [KS] W. Magnus, A. Karass, and 0. Solftar, Combinatorial Group Theory, Pure and Appl. Math. 13, Interscience New York (1966). [xP] Nv G. Markley and M. E. Paul, Matrix subshifts for 2 Univ. of Nd. Zechnical Report Tr 78-33 (1978). symbolic dynamica, [0] D. S. Ornstein, Ergodic Theory Randorness and Dynamical Systems, Yale Math. Monographs 5, New Haven and London, Yale Univ. Press (1974). it) Parry, On the 8-expansion of real numbers, Acta Nath. Acad. Sci. Hungar. 11 (1960) 401-416. [R] D, Rudolph, Counting the relative finite factors of a Bernoulli shift. Israel Journal of Math. 30 (1978) 255-263, (Wi] 8. F. Willlams, Classification of shifts of finite type, Ann. of Math. 98 (1973) 120-153. Errata, Ann. of Math. 99 (1974) 380-381. [Wel B. Weiss, Subshifts of finite type and sofie systems, Monatsh. Math. 77 (1973) 462-474.Pointwise Ergodic Theorems in L,, Spaces Mustafa A. Akcoglu Let 1, = L,(X,F,u) be the usual Banach Spaces associated with a mea~ sure space (X,f,u), Let T: 1, ~ 2, be a contraction (i.e, a linear operator with norm not more than 1) and let Ay, k = 1, denote the po- a pk-1 ot lynomial Ay (z) = Yio) 2 By the pointwise ergodic theorem in 1, we mean the existence of (7) ave. Lim A (TE for all £ © Ly This has been investigated by many authors, starting with the Hopf's generalization [8] of the classical Birkhoff theorem [4] to an opera~ tor theoretic setting. The situation for a positive contraction (me < 1%) now seems to be well understood: if 1< p< = then (*) exists [1], but it does not have to exist if p= 1 [6] or if p The general case of a not necessarily positive contraction is, however, essentially open, although there are some partial results in this di~ rection, the most important ones being the following theorems. Theorem 1 (Dunford-Schwartz [7]}: IT£ T is simultaneously a contrac- tion of all L, spaces, 1 p< =, then (*) exists, for all lepse. Theorem 2 (Bellow [3]): If 1 0, there exists a unitary operator T and a sequence of integers k, such that aA, (TD) - < z ' req(T) ~ Spe, forall ned. - We define T= J7,A,?, where the complex numbers %, will be chosen as follows, always satis~ Proof: Let P, be the projection on »,14 fying the conditions that ‘}, = 1 and 2, #1. Let %, be arbi- trary. Choose an integer k, such that ty Oy) <£, and a number ny > 0 such that TA, (A) = 1) <6, wheneler 2-1! < ny. TE yy ky, ny are chosen for! 1 ~ n then choose , such that Paarl! < ny for all - Then choose k,,, > k, such that A Oa) < Ener FOF 1 4+ nel and choose ,,, > 0 such that iARTTT (AD = 1] < epg, whenever !)-1' 0, there is an f in G satisfying deg £0) B f(a), h(x)) < © ae. The above result was recently established ([3 ], Topological Gonjugacy Lema) for the case where h is a honeonorphism with a fixed point, The present version is now an extension (rather than analog ) of Halmos' Conjugacy Lemma (8, p. 77], since te has the sane hypotheses and asserts che density of (e/g f: £€6} with respect co a finer topology (compact-open vs. weak). The paper is organized as foliows. In section 2 we prove a purely measure theoretic result (Theorem 1) which we had originally viewed as a generalization of Rohlin's Tower Theorem. At the conference, U. Krengel observed that Theorem 1 may also be seen as "almost" a special case of Theorem 1.1 of [6 ]. (it would be a special case if we restricted g to be ergodic rather than aperiodic.) However, rather than relying on the deep results of [6 ], we have decided to retain our original proof. In section 3, we show how Theorem 1 may be used to deduce the Conjugacy Lemma (and almost conversely). In section 4 we first consider homeonorphisns, We define the NBD condition, prove a property of NBD spaces (Theorem 2), and prove our main result (Theorem 3). Generalized Rohlin Towers. The previous (restricted) proof of the Conjugacy Lemma used the following well known result. Rohlin Tower Theorem (see [8], p- 71 for a proof): Let g in G be aperiodic. Then for every positive < and positive integer M, there is a (M-Rohlin) ‘ M partition $= {S,}) with gS, 5,,, for i and u(Sy) <2. (In such a situation we define the "top" of the partition S, denoted 1(S), by T(S) = Sy. fon Tt is now clear that the fixed point assumption in the Conjugacy Lenma of [3] was required because of the possibility that u(g Sy fl Sy) >0 when Rohlin's Theorem was appealed to. To generalize this result to partitions {5,}where ule 5, 1 8,) is to be zero for certain pairs i,j specified in advance, we first must introduce sone definitions. An nn matrix B consisting of 0's and 1's is said to be aperiodic if BY has all positive entries (written BY >0) for some positive integer N. Two nn matrices are called equivalent if their corresponding entries have the sane sign. A probability distribution (pj, .-.) p,) is said to be consistent with an nn 0-1 matrix B if it is invariant under sone stochastic matrix (yy) equivalent to B. We can now state the main result of this section and two corollaries, the second of which shows why we consider these results as generaliza~ tions of the Tower Theorem. Theoren 1: Let (py, ..., p,) be a consistent probability distribution for the nn aperiodic 0-1 matrix B. Then for any aperiodic transformations g in G, there is a partition P= {P,}0 fof X wich KR Sey dey es ms and ug PLP) = 0 for 444 with by = Corollary 1: Let (p;))) 4, 1 = 1, s+) my be an aperiodic stochastic matrix (aixing Markov chain). Then for any aperiodic g in G there is a partition . a ey with 4G PY HP) = Py IPL) Corollary 2: For any K* 2, let nj, «-+, ny be relative prime integers and = 1. Then for Jet ay) «+15 dy be positive numbers such that nay + +--+ nedy any aperiodic g in G there exist sets ¢ Ly sey K with HQ) = ay K and such that X= UU g3(Q,)_ is a partition (into K stacks of heights t=1 5-0 Ry eee me ‘The proof of Theorem | is based on three lenmas, for which we need the following notation. Fix g, 0, 5, (py, » p,) satisfying the hypotheses of Theoren 1, and fix N with BN > 0. All partitions considered (except Roblin partitions) will have n elenents, and for these we define a (complete) metric a GR = Fu, AR), where 4 denotes symmetric difference, For any inl partition R we define W(R) = {x in Kix €R, and g(x) €R, for 1,j with i by 5 70} + Im this notation, the last assertion of Theorem 1 is that 4u(W(P)) = 0.19 It is useful to have a description of the set D(S), consisting of all probability distributions consistent with B, as a subset of WR". We shall only need to know that D(B) is the inner (or relative interior or algebraic interior) set of a convex polyhedron. To describe this set, let I denote the power set of (1, ++) a} and define 9: T—->T by § € 9(A) ff by, = 1 for sone 1 be the subalgebra of I” given by AET, tf by, = 1 and 5 € (A) imply i € A. D(B) can now be written (the proof is not hard) as the LEA, Let 1. set of all probability distributions y = (yj) ++) ¥,) satisfying Quay Roy 2 yy for ACT, and 1€A 4 € 8Ca) (2.2) a < BR ee : 1€A 4 © 8a) At this point fix > 0 to be half the distance from (pj, +++5 P,) to the "boundary" of D(B), that is, the set D(B) - p(s). For distances such as 2 1_ 2 My this in R", we will always use the metric |y! - y’ + For any partition R, let 6 R denote its distribution (uR, UR.) and let “ee S(R/Y) denote its relative distribution on the subset Y, 1a) (UR, NY), +++, UCR, 1¥)), We will also use the notation 5 V to denote an n-tuple (distribution) when no partition V has been mentioned. In particular we now define 6 P= (py, +++ Pq), although @ partition P with this distribution renains to be found. Finally, for any distribution 6V, ler 5° denote its orthogonal Euclidean projection onto the affine subspace containing D(B), the subspace defined by (2.1). M Lemna 1: Let {5,¥uj be a Rohlin partition for g, and let (ays ++ 4,)= 5 Q be in DG). Then there is a partition R with BR = (ays vee a.) and WOR) ¢ TCS). Proof: By definition of D(B), (qj, ---s q,) 48 invariant under a stochastic matrix py) equivalent to 5, We define R recursively on $1,525 then separately on Sy. On S, define R so that 4(R/S,) = 5Q.20 been defined on Sy, K =M-2, define R on Spy, dy requiring KHL 8(R/B(Sy Ry) = (Pyys Pyyy ser» Pgg)+ This recursively defines R on X - Sys On Sy define R so that S(R/Sy) =8 Q. This definition ensures that 5R = 6Q, and the fact that 0 whenever b,, = 0 guarantees that W(R) < T(S). Pay ra Lemma 2: Let f >0 and let R be any partition of X. Then there is a Rohlin partition $ with H(T(S)) < B and a partition Q satisfying W(Q) © 1(S) and Laer’ < 20-4) nQW(R)). Rohiin's Theorem gives us the s= {S.J with u(t(s)) < 8. The Broo! Ht algorithm for defining Q consists of coding the R-columns of S$. Consider a typical column whose base is the set of x in S, such that g/(x) €R, for dl 50, ..., M2. We code the sequence (iy, ---, ty.) into another sequence (gy -++y typ) satisfying 3) ba ghoe yy ye for j 0 Let jg denote the least j with b 4 = 0. Since pete yi BN 0 we may £111 in the blanks in the sequence (tystyy sos ty wrarsssearoty 9) so that the resulting sequence will have no transitions corresponding to 0's of B. Apply the sane process beginning at the first place J, (j; = jgtm) where another illegal" transition occurs. When the coded sequence (ij, «.., ty ,) satisfying (2,3) 4s found, define Q by setting g/(x) €Q,1 for $= 0, «++, M-2, and i X€S,, The fact that 1, satisfies (2.3) ensures that W(Q) ¢ T(S). The fact that in going fron {4,} to (4,} every illegal transition resulted in no nore than N-1 changes guarantees that [/@-R!’ = 2(N-1) u(W(R))« Unfortunately, § Q may be even further from (p,, « p,) = 6P than 5 R is. To correct this,in the next lenma we "average" Q with another partition L 1 ee projection onto D(B) is 6 P. Lema di Let $= {s,J{, be a Roblin partition. Let Q be a partition with [8 "Q- 6 P| < % and W(Q) © T(S). Then there is a partition U of X witha wD) C1(s), 6'U = 5P and — |U-Q! < (2) [8%Q-6 P|. Proof: (At first reading it is helpful to assume that dim D(B) = n-1, so that * * 6" = 6 for ali partitions.) First observe that 6 °(Q) € D(B), because |e"(@q) - 8 P< 2X. Define 0 by [5" Q- SP| = 6% and define 6 L by 5L=OP- 1/8 (6 Q-5 2). A simple calculation shows that i * e Celt OOo a OL By convexity, this implies that 6 L lies in the affine subspace containing D(B), defined by equations (2.1), Furthermore [5 L- 5 P| = 4, so that ‘SL € D(B). Let Gy k= 1 K, be the colums of the Roblin Tower {s,}%- ee Ry wee Ry ons 0} al defined by partitioning S, into Q- (M- 1) names. Divide column C, into two k it and C2 with u(C2) = Ou(C,). Similarly divide rhe exceptional set x ad Oy acy) y+ Sintlarly Pt Sy into two sets Cl and C2 so that 6(Q/CL) = 5(Q/C2). Finally, partition 7 Boot te az. hi wh ede S42 1,2, observe that es columns ¢, 7 y by defining we ko and u(%) = 78, and thar 5(Q/¥,) = 5(Q/¥,) = 5Q. We define the partition U_ separately on yt and vy, on yi, set U=Q. Define U on y using, Lemma 1 (or more technically, its proof) so that 6(U/Y,) = 6L, Since Q=U on vi, and u(y!) = 1/16, we have 2 Te ‘lu-all = < 20=(/) Io'Q-8 Furthermore, 5 U = u(y aqu/y!) + uci?) squ/y?) 1 eo: fee te Taking projections, we have 54 4 Boe ue kot stu ese Proof of Theorem 1: Choose fy, k = 1,2, with §,< 4/2N and22 z F< =, Using Lemma 1 define a partition Ro with § Rt = 6 P and ket u((R})) < Ry. We now recursively construct a sequence {R*}"_, satisfying Bye A (2.3) UR) < fy (2.4) Koe5p, and (2.5) Bea cc pe where C= 4N/M42N. Clearly satisfies (2.3) and (2.4). Suppose that partitions R“ satisfying (2.3) and (2.4) have been found for k< j. We then “ construct satisfying (2.3), (2.4) and (2.5) as follows. Apply Lenma 2 with rem) ond pe B, to find a Rohiin partition S$ with u(T(S))< 8; and a partition Q satisfying W(Q) © T(S) and far Hy < 20eayuanRd}) < 20-1) Fy < Oe Since s*xit asp, we have sp! os [sq - srt) < qerdt < Consequently Q satisfies the hypotheses of Lemma 3 with respect to S, and * therefore there is a partition U with w(t) < 1(s), SU tug < (af) Is* Q- SP! < Gn Byal\ + Setting RJ =u, it is immediate that R! satisfies (2.3) and (2.4). To check (2.5) we estimate ved - Rit qo Rh We now observe that (2.5) implies that the R* form a Cauchy sequence with respect to the partition metric | -; , Since this metric is complete RE Bi —» 0, since both u(W(+)) and there is a partition P with Exo) are continuous with respect to the partition metric, it follows that ww) = 0 and 45" 3=5P, where 5P still is shorthand for the distribution (Py) sees PA) and is not necessarily the distribution of P, which is not yet defined. But since u(W(S)) = 0 it follows that ? belongs to D(B) by defining23 = ue 8, 1,)/uE,). For elenents of D(B), Pay 5 B= 5P = (pj, s++) py)- The partition P= P satisfies all the requirenents of the Theoren, 3. Gonjugacy Lemma. In this section we prove the Conjugacy Lemma using Theoren 1. We also show that for "non-critical" matrices B, we may prove Theorem 1 using the Conjugacy Lemma. Thus, Theorem 1 and the Conjugacy Lenma are in the sane relationship as Rohlin's Tower Theorem and Halnos' Conjugacy Lenma. Conjugacy Lemma: Let g and h be in G, with g aperiodic. Then for any € >0 there is an f in G satisfying d(f'g £0), hG@)) < € 4 ae. Proof: Let Q= {Q,2}.; be a measurabJe partition of X satisfying both 4th Q)< €/2 and 4(Q,) < €/2 for t= 1, «44, where d in this context denotes set diameter, Define an n%n 0-1 matrix B by bys 1 iF HQ, Q; #0 (bar denotes closure). We claim that 3 is apertodic and that * Q belongs to D(3). To see this let f and => be as defined prior to Lemma 1, and observe that GB. z a3 bQ, for A€D y Lea deea 9 Suppose equality holds in (3.1) for some A€T other than $ and {1, e+e, nhs ® Then the set Y= ( eu u is the disjoint union of two 4 © (a) u LEA nonempty closed sets and =(Y) = 1, Tt follows that X-Y is an open set with teasure 0 and therefore empty (by assumption on 4), or X=¥, But then X is not connected, contradicting the assumption on X, Therefore strict inequality holds in (3.1) for all nonempty proper subsets A. This means, according to (2.1) and (2.2), that © Q € D(B), and that @, 0, n}}. Choose an integer N sufficiently large so that 1/N is less than the minimum difference between the left and right hand sides of (3.1) for A €T - Repeated to application of (3.1) shows that : 2@j) 510 for all A in Py je ow