Drift conditions and invariant measures for Markov chains. (Q1879539): Difference between revisions

Let \(\{\Phi _ {n}\}_ {n\geq 0}\) be a Markov chain on a measurable space \((X,\mathfrak B)\) with a countably generated \(\sigma \)-algebra \(\mathfrak B\), let \(P\) be its transition probability. Assume that there exist a set \(C\subseteq X\), \(b\in \mathbb R\), and a function \(V: X\to \left [0,\infty \right ]\) finite for at least one \(\bar x\in X\) such that the drift condition \(PV(x) \leq V(x) - 1 + b\mathbf 1_ {C} (x)\), \(x\in X\), is satisfied. It is known that under various additional hypotheses on the set \(C\) and the chain \(\{\Phi _ {n}\}\) the drift condition is sufficient for the existence of an invariant measure for \(P\). Another theorem of this type is proven, requiring no irreducibility or continuity assumptions. Namely, besides the drift condition suppose that \(\lim _ {n\to \infty }\sup _ {y\in C} P(y,B_ {n}) = 0\) whenever \(B_ {n}\in \mathfrak B\), \(B_ {n}\searrow \emptyset \). Then there exist finitely many orthogonal invariant probability measures \(\pi _ 1,\dots , \pi _ {N}\) such that \[ \lim _{n\to \infty }\frac 1{n}\sum _ {m=1}^ {n} P^ {m}(x,A) = \sum ^ {N}_ {i=1} L_ {i}(x)\pi _ {i}(A) \] for every \(A\in \mathfrak B\), every \(x\in X\) with \(V(x)<\infty \) and some constants \(L_ {i}(x)\geq 0\) satisfying \(\sum ^ {N}_ {i=1} L_ {i}(x) = 1\). The set \(\{y\in X;\; V(y)<\infty \}\) admits a decomposition \(\{V<\infty \} = \bigcup ^ {N}_ {i=1} H_ {i} \cup E\), where \(H_ 1,\dots ,H_ {N}\) are disjoint maximal Harris sets and \(\mathbf P_ {x}\{\tau <\infty \} = 1\) for all \(x\in E\), \(\tau \) being the first hitting time of the set \(\bigcup ^ {N}_ 1 H_ {i}\). If \(\{V<\infty \} = X\), then there exists a unique invariant probability measure \(\pi \) for \(P\) if and only if \(\{\Phi _ {n}\}\) is \(\psi \)-irreducible. In the \(\psi \)-irreducible case the measures \(n^ {-1} \sum ^ {n}_ {m=1} P^ {m}(x,\cdot )\) converge to \(\pi \) as \(n\to \infty \) in the total variation norm for each \(x\in X\).