Abstract: | Asymptotic properties of the waiting timeW
k
(x,y) until an initial segment of lengthk of a sample pathx of an ergodic finite-alphabet process is seen in an independently chosen sample pathy are discussed. Wyner and Ziv have shown that for irreducible Markov chains, (1/k) logW
k
(x,y) converges in probability to the entropyH of the process. In this paper, almost sure convergence toH is established for the somewhat larger class of functions of irreducible Markov chains and convergence in probability toH is established for weak Bernoulli processes. A stationary coding of an i.i.d. process is constructed for which there is a subsequencek(n) such that (1/k(n) logW
k(n
)(x,y), converges in probability to +. Positive and negative results for the case when only approximate matches are required are also obtained.Partially supported by NSF Grant DMS-9024240.From 9/1 to 12/1 earch year: Mathematies Institute, POB 127, 1364 Budapest, Hungary. 011-361-1-177-175. At other times: Department of Mathematics, University of Toledo, Toledo, Ohio 43606, (419) 537-2069. |