Large deviations for vector-valued functionals of a Markov chain: Lower bounds (Q1110900): Difference between revisions

Let \(X_ 0,X_ 1,..\). be a Markov chain with general state space S and transition probability \(\pi\). Let E be a topological vector space and f: \(S\to E\). Under certain assumptions on E and f and the irreducibility of \(\pi\), lower bounds of the type \[ \lim_{n\to \infty}\inf n^{-1}\log P_ x\{n^{-1}\sum^{n-1}_{k=0}f(X_ k)\in G\}=-\inf_{u\in G} \lambda (u) \] are obtained. Here G is an open subset of E, \(x\in S\), \(\lambda\) is the convex conjugate \(of\quad \phi (\xi)=-\log R(K_{\xi}),\quad \xi \in E^*,\) and \(R(K_{\xi})\) is the convergence parameter of the kernel \[ K_{\xi}(x,A)=\int_{A}e^{<\xi,f(y)>}\pi (x,dy). \] Essential tools in the proof are properties of irreducible kernels, especially a continuity property of the convergence parameter and the approximation by quasinilpotent kernels. As an application, lower bounds for large deviations of occupation times of Markov chains are derived under somewhat weaker assumptions than those of \textit{M. D. Donsker} and \textit{S. R. S. Varadhan} [Commun. Pure Appl. Math. 29, 389-461 (1976; Zbl 0348.60032)]. An ergodic theorem for Banach space-valued functionals of a Markov chain and a continuity result for the Fenchel transform are also proved.