Stationary distributions and mean first passage times of perturbed Markov chains (Q2575710)

This paper is concerned with the effects of perturbations of the transition probabilities on the stationary distributions of irreducible Markov chains, in the magnitude of stationary probabilities. The author accomplishes this by analyzing the mean first passage times of the original irreducible Markov chain. The main tools used to involve generalized inverses, through which a general relationship between the stationary probabilities for the initial and perturbed Markov chains and also a relationship between the corresponding mean first passage times. Using these relationships, the author is deriving component-wise and relative error bounds between the stationary probabilities of the two Markov chains. Furthermore, special cases, involving perturbations in a single row (for the two, three and multiple element perturbations) are considered. In the last section the focus is on mean first passage time under perturbations, and a general procedure for the joint computation of the transition probability vector and the mean first passage time matrix is analyzed using a succession of two element perturbations.