Convergence of Markov chains in information divergence (Q1014048): Difference between revisions

Some convergence theorems in probability theory can be reformulated as ``the entropy converges to its maximum''. A. Rényi used information divergence to prove convergence of Markov chains to equilibrium on a finite state space. Later I. Csiszar and D. Kendall extended Rényi's method to provide convergence on countable state spaces. Here, the authors establish convergence in information divergence for a large class of Markov chains. The basic result is that information divergence is continuous under the formation of the intersection of a decreasing sequence of \(\sigma\)-algebras. The same technique can be used to obtain a classical result of Pinsker about continuity under an increasing sequence of \(\sigma\)-algebras.