A computational procedure for the asymptotic analysis of homogeneous semi-Markov processes (Q1060493)

A semi-Markov process with a finite space of states is considered. A one- dimensional stationary distribution of such a process depends on a set of transient and recurrent classes of an imbedded Markov chain, on stationary probabilities of the Markov chain and on conditional expectations of first exit times given initial states. An algorithm is proposed for evaluation of this distribution. The main step of the algorithm is obtaining equivalence classes with respect to a set of circles in a transition graph of the Markov chain. The algorithm is considered in its application to social security problems.