Random stationary processes (Q1201166)

Consider a stationary process \(X=\{X_ n\}^ \infty_{n=-\infty}\) with finite state space \(A\) (the ``alphabet''). Treat \(X\) as an \(A^ \infty\)- valued function, so that its distribution constitutes a stationary probability measure \(\nu\) on \(A^ \infty\), the elements of which are called ``words''. The authors discuss an inductive, Kolmogorov-style construction of such a measure \(\nu\) out of a sequence \(\{\nu_ n\}^ \infty_{n=1}\), where \(\nu_ n\) is a probability on \(A^ n\) which is to be the distribution of any \(n\) successive \(X_ i\)'s. Although \(\nu_ n\) is constrained to be compatible with \(\nu_ 1,\dots,\nu_{n-1}\), there are parameters which can be freely chosen in the determination of \(\nu_ n\). The authors decide to randomize these parameters in the most natural way they can think of, and the result is a canonical probability \(\hat\mu\) on the space \({\mathcal M}(A^ \infty)\) of probability measures on \(A\). No \(\sigma\)-algebra of subsets of \({\mathcal M}(A^ \infty)\) is mentioned, nor do the authors clarify how, in precise detail, \(\hat\mu\) is then converted into a canonical probability \(\mu\) on the space \({\mathcal M}(A^ \infty)\times A^ \infty\). Therefore, the precise content of their main results remains a mystery to this reviewer. However, the authors state: ``It is shown that with \([\mu]\) probability 1, the random stationary process has zero entropy and gives positive probability to every periodic infinite word''.