scientific article; zbMATH DE number 27013

Let \((X_ n)\), \(n\geq 0\), be a Markov chain on the state space \((E,{\mathcal E})\) with transition probabilities \(P_ n(x,\cdot)\). If \((X_ n)\) is Harris irreducible, then there exists a unique invariant measure \(\pi\) on \((E,{\mathcal E})\) such that \(\lim_{n\to\infty}\| P_ n(x,\cdot)- \pi\|=0\). In the absence of Harris irreducibility, the chain is, in general, not ergodic in the sense of convergence in the variation norm. The author studies conditions under which the chain taking values in a perfectly normal topological space (or in a metric space) is ergodic in the sense of weak convergence of measures. These ergodicity conditions are of three types and are roughly: (a) existence of a compact set \(V\) which the chain hits almost surely, such that for each \(n\in\mathbb{N}\), the family \(\{P_ n(x,\cdot):\;x\in V\}\) is tight; (b) positivity condition: for each non-empty open \(G\) and \(s\in\mathbb{N}\), there are \(k=k(G,s)\) and \(p=p(G,s)\) such that \(\inf_{x\in V}P_{k+j}(x,G)\geq p\), \(j=0,1,\dots,s\); (c) certain continuity condition in \(x\) on the one-step kernel \(P(x,\cdot)\), allowing for some discontinuities. It is shown that under these conditions there exists a unique invariant measure \(\pi\), charging every non-empty open set, such that \(P_ n(x,\cdot)\) converges weakly to \(\pi\) as \(n\to\infty\). As an intermediate step in proving the result, the author studies convergence of \(Q_ n(x,\cdot)=(1/n)\sum_{k=1}^ n P_ k(x,\cdot)\) under somewhat weaker conditions. Several examples of non-Harris chains are given illustrating introduced conditions.

Recommendations

Cited in

(16)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3983583)