Phase-type distributions and the structure of finite Markov chains (Q1801859)

The distribution of any first passage time in a finite-state Markov chain is said to be of phase-type. In some previous papers O'Cinneide and the author have shown how from any distribution of phase-type one may construct a Markov chain (a set of states and a transition matrix) giving rise to it. Also, an algorithm for the solution of this inverse problem was given by the author, and applied to a typical class of phase-type distributions. In this paper the author observes that the algorithm has an unusual property, which may be of interest to implementors and stochastic modellers: the constructed Markov chain always has a rather special structure. The significance of the following theorem for stochastic modelling is emphasized: Theorem (4.1). All phase-type distributions have Markov chain representations of height at most 2.