Moderate deviations for empirical measures of Markov chains: Lower bounds (Q1356342): Difference between revisions

Let \(\{X_j; j\geq 0\}\) be an ergodic Markov chain with invariant probability measure \(\pi\) and \(\{b_n\}\) be a positive sequence such that \(b_n/n^{1/2} \to\infty\), \(b_n/n \to 0\). Define the random measure \(M_n= {1\over b_n} \sum^{n-1}_{j=0} (\delta_{X_j} -\pi)\). Under the assumption that the Markov chain is ergodic of degree 2 lower bounds for probabilities \(P\{M_n\in B\}\) are obtained in the form \(\lim\inf_n {n\over b^2_n} \log P\{M_n\in B\} \geq \dots\), where the right-hand side depends on certain functional, and the explicit condition for the functional is obtained.