Nonuniqueness in \(g\)-functions. (English) Zbl 0786.60043

Let \((X_ n)\) be a one-sided shift with a finite state space \(A\). The conditional expectation \(P(X_ 0=a_ 0 \mid X_{-1}=a_{-1},\;X_{- 2}=a_{-2},\dots)\) gives a function \(f(a_ 0 \mid a_{-1},a_{- 2},\dots)\) with \[ 0 \leq f(a_ 0 \mid a_{-1},a_{-2},\dots) \leq 1, \quad \sum_{a \in A} f(a \mid a_{-1},a_{-2},\dots)=1. \] Such functions are called \(g\)-functions; they were introduced by W. Doeblin and R. Fortet [Bull. Soc. Math. Fr. 65, 132-148 (1937; Zbl 0018.03303)]. To a \(g\)-function there might exist a corresponding stationary process, i.e. a (shift-)invariant measure on \(A^ N\). M. Keane [Invent. Math. 16, 309-324 (1972; Zbl 0241.28014); for more results see also H. Berbee, Probab. Theory Relat. Fields 76, 243-253 (1987; Zbl 0611.60059)] showed that this is the case for continuous \(g\)-measures and gave conditions for which the measure is strongly mixing and unique. B. Petit [C. R. Acad. Sci., Paris, Sér. A 280, 17-20 (1975; Zbl 0301.28012)] showed that all differentiable \(g\)-functions \(f\) with \[ \varepsilon \leq f \leq 1-\varepsilon \tag{*} \] for some \(0<\varepsilon<1/2\) have unique measures which are weakly Bernoulli. S. Kalikow [Isr. J. Math. 71, No. 1, 33-54 (1990; Zbl 0711.60041)] noticed that the continuity of a \(g\)-function \(f\) is equivalent to uniform convergence of the martingales \(P(X_ 0=a_ 0 \mid X_{- 1},X_{-2},\dots,X_{-n})\), \(n=1,2,\dots,\) \(a_ 0 \in A\), where \((X_ i)\) is the corresponding stationary process. The martingales converge uniformly if and only if \((X_ i)\) is a random Markov chain, i.e. there exist r.v. \(N_ i\) with values in \(\mathbb{N}\) such that \((X_ i,N_ i)\) is a stationary process, \(N_ 0\) is independent of \((a_ i,N_ i)_{i<0}\) and \[ P(X_ 0=a_ 0 \mid X_{-1}=a_{-1},\dots,X_{-n}=a_{-n},N_ 0=n)=P(X_ 0=a_ 0 \mid (X_ i)_{i<0}= (a_ i)_{i<0}, N_ 0=n) \tag{**} \] for all \(n\), \((a_ i)_{i<0}\).
From the proof of Theorem 7 of the Kalikow’s paper one can derive that if a continuous \(g\)-function \(f\) satisfies (*) and (**) with \(EN_ 0<\infty\), then the corresponding invariant measure is determined uniquely. The paper under review gives an example of a continuous \(g\)- function which satisfies (*) but has two measures.
Reviewer: D.Volný (Praha)


60G10 Stationary stochastic processes
Full Text: DOI


[1] M. Aizenman, J. Chayes, L. Chayes and C. Newman,Discontinuity of the magnetization in one-dimensional 1/|x| 2 Ising and Potts models, J. Stat. Physics50 (1988), 1–40. · Zbl 1084.82514
[2] H. Berbee,Chains with infinite connections: uniqueness and Markov representation, Probability Theory and Related Fields76 (1987), 243–253. · Zbl 0611.60059
[3] W. Doeblin and R. Fortet,Sur les chaines a liaisons completes, Bull. Soc. Math. Fr.65 (1937), 132–148. · Zbl 0018.03303
[4] T. Harris,On chains of infinite order, Pacific Journal Math.5 (1955), 707–724. · Zbl 0066.11402
[5] S. Kalikow,Random Markov processes and uniform martingales, Israel Journal of Mathematics71 (1990), 33–54. · Zbl 0711.60041
[6] S. Kalikow, Y. Katznelson and B. Weiss,Finitarily deterministic generators for zero entropy systems, Israel J. Math.79, (1992), 33–45. · Zbl 0768.60074
[7] M. Keane,Strongly mixing g-measures, Invent. Math.16, (1972), 309–324. · Zbl 0241.28014
[8] B. Petit,Schemas de Bernoulli et g-mesures, C. R. Acad. Sci. Paris, Ser. A280 (1975), 17–20. · Zbl 0301.28012
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. It attempts to reflect the references listed in the original paper as accurately as possible without claiming the completeness or perfect precision of the matching.