Polynomial ergodicity of Markov transition kernels. (Q2574534): Difference between revisions

Let \(\varPhi \) be a time-homogeneous Markov chain on a countably generated measure space \((X,\mathfrak B)\) with a \(\varphi \)-irreducible and aperiodic transition kernel \(P\). Let \(f\geq 1\) be a measurable function on \(X\) and \(r=(r(n))_ {n\geq 1}\) a polynomial sequence, that is, \(\liminf r(n)(n+1)^ {-\beta }>0\), \(\limsup r(n)(n+1) ^ {-\beta }<\infty \) for some \(\beta >0\). The kernel \(P\) is called \((f,r)\)-ergodic, if there exists a unique invariant measure \(\pi \) for \(P\), \(\pi (f)<\infty \) and \[ \lim _ {n\to \infty } r(n)\sup _ {| g| \leq f}\Bigl | \mathbf E_ {x} g(\varPhi _ {n})-\pi (g)\Bigr | = 0 \tag{1} \] for all \(x\) is a full and absorbing set. \textit{P.\ Tuominen} and \textit{R.\ L.\ Tweedie} [Adv.\ Appl.\ Probab.\ 26, 775-798 (1994; Zbl 0803.60061)] showed that (1) holds provided that there exists a petite set \(C\in \mathfrak B\) such that \[ \sup _ {x\in C} \mathbf E_ {x}\left [ \sum ^ {\tau _ {C}-1}_ {k=0} r(k)f(\varPhi _ {k})\right ] <\infty , \tag{2} \] where \(\tau _ {C}\) is the first hitting time of the set \(C\). In the paper under review, new drift conditions to control the \((f,r)\)-modulated hitting time moments (2) are found and several applications (e.g.\ to the random walk Hastings Metropolis algorithm or to general nonlinear state-space models) are discussed. The drift conditions are used to obtain explicit, computable bounds for polynomial ergodicity by evaluating moments of an appropriately defined coupling time.