On the ergodic decomposition for a class of Markov chains (Q2485833)

This paper focuses on the question of the ergodic decomposition of Markov chains and the existence of invariant probability measures for Markov chains. For discrete-time Markov chains, there are known sufficient conditions, based on a Foster-Lyapunov type criterion, for a given Markov chain on a general state space to have an invariant probability measure (S. Meyn and R. Tweedie; 1993). Continuing their previous work (2000), the authors propose an answer to the question of necessary conditions holding for the above mentioned problem. There are explored further properties of the so-called T\('\)-condition that represents an abstract generalization of T-chains and irreducible chains (since T-chains and irreducible chains satisfy T\('\)-condition but the reverse may not be true). Under no separability assumption on the \(\sigma\)-field (of the Markov chain defined on a measurable space), the paper proves that the T\('\)-condition is sufficient for the requirement that there are no uncountable disjoint absorbing sets and, under some hypothesis, T\('\)-condition is also necessary. For the case when \(\sigma\)-field is countably generated and separated, the T\('\)-condition is equivalent to the existence of a T-continuous component (everywhere non-trivial) for the considered Markov chain. Furthermore, under the assumption that the space is a compact separable metric one, it is shown that the Foster-Lyapunov criterion represents a necessary and sufficient condition for the existence of an invariant probability measure for the Markov chain, and that every probability measure for the Markov chain is non-singular, in this case.