Chapman-Kolmogorov equation for non-Markovian shift-invariant measures (Q1345615)

This article investigates the shift-invariant measures \(\mu\) on the space \(\Omega\equiv \{0,1, \dots, k - 1\}^{\mathbf Z}\) of doubly-infinite sequences of \(k\) symbols, which satisfy the additional condition (the `Chapman-Kolmogorov equation') that the family of \(k \times k\) matrices \(\Pi_ n\) defined for \(n \geq 0\) by \[ (\Pi_ n)_{i,j} = \mu \bigl( \{\omega \in \Omega : \omega_ n = j, \omega_ 0 = i \} \bigr)/ \mu \bigl( \{\omega : \omega_ 0 = i \} \bigr) \] form a semigroup: \(\Pi_{n + m} = \Pi_ n \Pi_ m\), \(\forall n,m \geq 0\). Of course, the shift- invariant measures given by the laws of stationary Markov chains with state-space \(\{0, \dots, k - 1\}\) satisfy this property. But within the set \(C_ \Pi\) of all shift-invariant measures with a fixed matrix \(\Pi_ i \equiv \Pi\) and satisfying the Chapman-Kolmogorov equation, the authors construct a dense class \(S_ \Pi\). They further show that (a) \(S_ \Pi\) contains non-Markov measures whenever \(\Pi\) is strictly positive, (b) for aperiodic irreducible \(\Pi\), the non-Markovian measures of \(S_ \Pi\) are ergodic and not weakly mixing and may have zero entropy, (c) the Markov measure is the entropy maximizer within \(C_ \Pi\), and (d) the \(C_ \Pi\) measures with infinite memory form a \(G_ \delta\) dense set. The authors also discuss the implications of their constructions for the spectral properties of dynamical systems: first, all of the \(C_ \Pi\) measures for aperiodic irreducible \(\Pi\) have (shift operators with) a Lebesgue spectral component, and second, an example of \textit{J. Mathew} and \textit{H. G. Nadkarni} [Bull. Lond. Math. Soc. 16, 402-406 (1984; Zbl 0515.28010)] of a process with Lebesgue spectral component of multiplicity 2 is shown to satisfy the Chapman-Kolmogorov equation for a simple partition of its measure-space into two cells.