Kamae, T. and Keane, M. Osaka J. Math. 34 (1997), 653-657 A SIMPLE PROOF OF THE RATIO ERGODIC THEOREM 1. Introduction TETURO KAMAE and MICHAEL KEANE (Received July 9, 1996) Throughout the development of ergodic theory, attention has been devoted by many authors, beginning with the classical struggles of Birkhoff and von Neumann, to proofs of different forms of ergodic theorems. Recenttly, a standard principle has begun to emerge (see [6], but also [4], [5]). The aim of this short note is to apply the principle to obtain a proof of HopΓs ratio ergodic theorem ([!]). Fundamental Lemma. Let (α n )n=o,ι, and (& n )n=o,ι, be sequences of nonnegative real numbers for which there exists a positive integer M such that for any n = 0,1, there exists an integer m with 1 < m < M satisfying that 0<i<m 0<ί<m Then for any integer N with N > M, Y^ > α \ V^ 3 / ^ n n > / > ^ / 0<n<AΓ 0<n<N-M Proof. The proof of the Fundamental Lemma is easy. By the assumption, we can take integers 0 = rao < mi < < ra& < TV with ra^+i mi < M (ί = 0, 1,, k - 1) and N - m k < M such that _j_ι a n > ra for any i = 0, 1, -,& 1. Then, by adding these inequalities, we have an - Gn - bn - 0<n<N 0<n<m k 0<n<m fc 0<n<7V-M We apply this Lemma to prove the Ratio Ergodic Theorem [8]. Let (Ω,/3, μ) be a σ-finite measure space and T : Ω > Ω be a measure preserving transformation.
654 T. KAMAE AND M. KEANE Let / and g be integrable functions on Ω such that (1) g(ω) > 0 and ^g(t n ω) = oo 00 n=q for all ω G Ω. Then we have the following theorem. Π Ratio Ergodic Theorem. The following limit exists for almost all ω e Ω : v / ' n^oo 5 r( Moreover, r is T-ίnvarίant and //*,./, (2) Jfdμ = Jrgdμ In the special case that μ is a finite measure, the above theorem is nothing but a consequence of the individual Ergodic theorem applied to / and g separately. 2. The proof of the Theorem We may and do assume without loss of generality that f(ω) > 0 for any ω e Ω. For any function h on Ω and a positive integer n, we denote h n (ω] := h(ώ) + h(tω) + + h(t n ~ l ω). Let f(ω) := Irnin-^oo n and / \ r /nm r(ω) :=lm n^to --^, where we admit -hoc as a value. Then, r and r are T-invariant, that is, f(ω) = r(tω) and r(ω) r(tω) for any ω G Ω, in virtue of (1). We fix ε > 0 and L > 0. For any ω G Ω, there exists a positive integer m such that
SIMPLE PROOF OF RATIO ERGODIC THEOREM 655 where we denote a~b := min{α, b}. Let v be the finite measure on Ω defined by dv(ω) g(ω}dμ(ω}. Then for any δ > 0, there exists a positive integer M such that v(ω 0 ) > υ(ω) - (5, where Let Ω 0 := {u; G Ω; there exists ra with 1 < m < M such that f m (ω)<(r(ωyl)g m (ω) (l-ε)}. Then, for a n = F(T n ω) and b n = (r(ωyl)g(t n ω) - (1 - ε) (n = 0,!, ), the assumption of the Fundamental Lemma holds since if T n ω ΩQ, then we can take the same m as in the definition of ΩQ, nothing that / < F as well as f Λ L is T- invariant, and if T n ω ΩQ, then we can take m = 1. Therefore, by the Fundamental Lemma, for any ω Ω and any positive integer TV > M, F N (ω] > (r(ωyl)g N (ω) (1 - ε), where we use the fact that r is T-invariant. We integrate the both sides in the above inequality by dμ(ω] and use the fact that μ as well as r is T-invariant. We have N JFdμ = JF N dμ > ί(r^l)g N (1 - έ)dμ = (N - M) ί (ΓL)gdμ (1 - ε). On the other hand, since fdμ> Fdμ- I Lgdμ > Fdμ - Lδ, J J JΏ.\Ώ. Ό J we have Γ N - M f J fdμ > ^ J(rL)gdμ (1 - ε) - Lδ. Here, letting TV»oo, 5 0, ε 0 and L > oc in this order, we have (3) J fdμ> jrgdμ. This also implies that rg is integrable and by (1), f(α ) < oo for almost all ω G Ω. Now, we prove equality in (3) by establishing the reverse inequality. Fix ε > 0. Then for any ω Ω, there exists a positive integer m such that
656 T. KAMAE AND M. KEANE Let v be the finite measure on Ω defined by dv(ω] f(ω)dμ(ω). Then for any 6 > 0, there exists a positive integer M such that ι/(ω 0 ) > z'(ω) δ, where Ωo := {ω E Ω; (1 < 3 m < M) [/ m ( ω ) < (r(ω) + e)0m(")]}- Let 0 ω Ω 0. Then, for b n = F(T n ω) and a n = (r(ω) + ε)g(t n ω] (n = 0, 1, - ), the assumption of the Fundamental Lemma holds since if T n ω G ΩO, then we can take the same ra as in the definition of Ω 0, nothing that / > F, and if T n ω ^ Ω 0, then we can take m = 1. Therefore, by the Fundamental Lemma, it holds that for any ω G Ω and any positive integer TV > M, F N (ω) < (r(ω) + ε)g N (ω), where we use the fact that r is T-invariant. We integrate both sides in the above inequality by dμ(ω) and use the fact that μ as well as r is T-invariant. We have (N -M} I Fdμ = I F N dμ < (r + ε)g N dμ = N (r + ε)gdμ. On the other hand, since we have ffdμ < I Fdμ + / fdμ < ί Fdμ + «, J J 7Ω\Ω 0 ^ Here, letting TV > oo, ί J, 0 and ε 0 in this order, we have (4) ίfdμ< ίrgdμ. By (1),(3), (4) and the trivial inequality that r(α ) > r(ω) for any ω G Ω, it follows that r(ω) = r(ω) < oo for almost all ω G Ω. Hence, r(ω) exists and r(ω) = f(ω) = r(ω) for almost all ω G Ω. The equality (2) also follows, the proof is complete. REMARK. The basic idea used here originates from [4], which is developed in [5] so that a new type of proofs of Birkhoίf type ergodic theorems without
SIMPLE PROOF OF RATIO ERGODIC THEOREM 657 using the maximal ergodic lemma is created. The article [9] has escaped attention until now for obvious reasons, being published in an engineering journal. It seems to be an early instance of an application of the basic idea, but is still cluttered with unnecessary details, as is the development in [5]. The clearest and simplest exposition is, in our opinion, contained in [7] and based on [6]. Other relevant references include [2], [3] and [10], and there may well be earlier ones which we have not yet found. References [1] E. Hopf: Ergodentheorίe, Ergebnisse der Mathematik, Springer-Verlag, Berlin, 5 (1937). [2] R.L. Jones: New proofs for the maximal ergodic theorem and the Hardy-Lίttlewood maximal theorem, Proc. 87 (1983), 681-684. [3] R.L. Jones: Ornsteίn's Llog+ L theorem, Proc. 91 (1984), 262-264. [4] T. Kamae: A simple proof of the ergodic theorem using nonstandard analysis, Israel J. Math. 42-4 (1982), 284-290. [5] Y. Katznelson and B. Weiss: A simple proof of some ergodic theorems, Israel J Math. 42-4 (1982), 291-296. [6] M. Keane: Ergodic Theory, Symbolic Dynamics and Hyperbolic Space, Chapter 2, T. Bedford et al.(ed), Oxford Science Publications, (1991). [7] M. Keane: The Essence of the Law of Large Numbers, in Algorithms, Fractals and Dynamics, Plenum Press, N.Y. (1995), 125-129. [8] K. Petersen: Ergodic Theory, Cambridge Studies in Advanced Mathematics 2, Cambridge University Press, (1983). [9] P.C. Shields: The ergodic and entropy theorems revisited, IEEE Transactions on Information Theory, 33 (1987), 263-266. [10] J.M. Steele: Kingman's subaddίtive ergodic theorem, Ann. Inst. H. Poincare Probab. Statist. 25 (1989), 93-98. T. Kamae Department of Mathematics Osaka City University Sugimoto 3-3-138 Sumiyoshi-ku, Osaka 558, Japan M. Keane Centrum vor Wiskunde en Informatica Postbus 94079 1090GB Amsterdam The Netherlands