On direct convergence and periodicity for transition probabilities of Markov chains in random environments (Q919712): Difference between revisions

Let \(\{\) P(\(\theta\)), \(\theta\in \Theta \}\) be a family of stochastic matrices acting on a countable set X, and let P(\(\theta\) ;x,y) denote the (x,y) element of P(\(\theta\)). Suppose \(\Theta\) is a measurable space and \(\Theta^*=\Theta^{{\mathbb{Z}}}\) inherits the product \(\sigma\)-algebra from \(\Theta\). Finally let \(\theta^*=\{\theta_ n\}\in \Theta^*\) denote a stationary sequence. A Markov chain in a random environment is a process \(\{X_ n:\) \(n\geq 0\}\) satisfying \[ P(X_{n+1}=y | \quad X_ n=x,\quad X_{n-1,...,}X_ 0;\theta^*)=P(\theta_ n;x,y)\quad a.s. \] Let \(P_ x\) denote the distribution of \(\{X_ n,T^ n\theta^*\}\) (T is the shift operator on \(\Theta^*)\) when \(X_ 0=x\), let \[ \delta_ n(x,y,\theta^*)=\| P_ x(\theta_ 0,...,\theta_{n-1})-P_ y(\theta_ 0,...,\theta_{n-1})\| \text{ and } \delta (x,y,\theta^*)=\lim_{n\to \infty}\delta_ n(x,y,\theta^*). \] Interest focusses on conditions delineating when \(\delta (x,y,\theta^*)=0\), the direct convergence mentioned in the title. Examples show that the usual decomposition into cyclic sub-classes of the chain \(\{(X_ n,T^ n\theta^*)\}\) is inadequate for this. However a new concept of `meeting', somewhat analogous to the familiar communication equivalence relation does the trick. The author uses this to formulate limit theorems for the frequency of times of positive probability of return to a set.