The first Laurent series coefficients for singularly perturbed stochastic matrices (Q1434426): Difference between revisions

Dynamics in singularly perturbed Markov chains provide some specific time scales. One time scale may correspond to the more frequent transitions occurring among states, which communicate also in the unperturbed case. Another time scale, considered as a higher scale, may correspond to the less frequent transitions occurring among states which do not communicate in the unperturbed case. The general purpose of the present paper is to investigate the processes within the highest time scale, and to show that in this time scale there is a process among subsets (as opposed to states) which can be analyzed after constructing a suitable transition (stochastic) matrix. The particularly important concrete result obtained by the authors is that, in the process of solving the fundamental equations of the system, the most singular term of the Laurent expansions (for the mean passage time matrix) can be computed by solving successive Markov chains defined on shrinking state spaces. Briefly, Section 2 of the paper reviews the regular perturbation case, where the unperturbed process is ergodic, too. Section 3 explores the case when the unperturbed process does not possess transient states, having a singular perturbation that is referred to as the nearly completely decomposable (or nearly uncoupled) perturbation. There are achieved new expressions for the singular part in the Laurent expansion for the deviation and mean passage time matrices that are based on the aggregated process between subsets. Section 4 considers the general case, when absorbing states communicate via transient states. In particular, the matrix governing transitions at the highest time scale is computed, and from this matrix there are derived the most singular matrices in the Laurent series expansion for the mean passage time matrix and for the deviation matrix.