Asymptotic properties of a singularly perturbed Markov chain with inclusion of transient states. (Q1884830): Difference between revisions

The article is concerned with aggregations in a singularly perturbed nonstationary Markov chain having a finite state space \(M\) and fast and slow components. The generator of the underlying Markov chain is \(Q^{\varepsilon}(t) = \frac {1}{\varepsilon} \widetilde {Q}(t) + \widehat {Q}(t)\), where \(\varepsilon>0\) is a small parameter and generators \(\widetilde {Q}(t)\), \(\widehat {Q}(t)\) describe fast and slow motions, respectively. The generator of fast motions \(\widetilde {Q}(t)=\| \widetilde{q}_{mn}(t)\| \), where \(m,n \in M\), is of the following special type: \(\widetilde{q}_{mn}(t)=0\) if \(m \in M_i\), \(n \in M_j\), \(i \neq j\), \(1 \leq i \leq l\), \(1 \leq j \leq l\). According to this property the state space is partitioned into groups of recurrent states \(M_1,\dots,M_l\) and a group of transient states \(M_*\), so as \(M=M_1 \cup \dots \cup M_l \cup M_*\). Fast motions within recurrent classes \(M_1,\dots,M_l\) are described by generators \(\widetilde {Q}^1(t),\dots,\widetilde{Q}^l(t)\), transitions from transient class \(M_*\) are described by matrices \(\widetilde {Q}^1_*(t),\dots,\widetilde{Q}^l_*(t),\widetilde {Q}_*(t)\). The asymptotic properties are studied through sequences of unscaled and scaled occupation measures. By treating the states within each recurrent class \(M_i\), \(i=1,\dots,l\), as a single state, an aggregated process is defined and shown to be convergent to a limit Markov chain. A sequence of suitably rescaled occupation measures is shown to be convergent to a switching diffusion process weakly. Illustrative examples are given.