Rates of convergence for everywhere-positive Markov chains (Q1347202)

The following situation is considered: For a Markov chain with general state space \(({\mathcal X}; {\mathcal B})\) given by its transition kernel \(P(x;A)\) it is assumed that there exist an invariant probability measure \(\pi\) and a \(\sigma\)-finite reference measure \(\lambda\) being mutually absolutely continuous with respect to \(\pi\). Then it is shown that under the usual positivity condition on \(P(x;A)\) and an appropriate compactness condition related to the associated transition operator \(P\) for each initial distribution \(\nu\) with \(\nu P^ m \ll \pi, m\) some positive integer, the distributions \(\nu P^ n\) converge in total variation norm geometrically fast to \(\pi\). This generalizes a result of Schervish and Carlin (1992). Replacing the compactness condition by some condition on the density \(d \nu P^ n/d \pi\) sharpens the result in such a manner that a quantitative estimate of the convergence rate is stated.