Periodicity of asymptotically attainable structures in nonhomogeneous Markov systems (Q1200573)

Consider a nonhomogeneous Markov system with state vector \(N(t)=(N_ 1(t),N_ 2(t),\dots,\) \(N_ k(t))\). Let \(\{P(t)\), \(t=0,1,2,\dots\}\) be the sequence of transition-probability matrices, \(\{p_{k+1}(t)\), \(t=0,1,2,\dots\}\) the vector-valued sequence of loss probabilities, and \(\{p_ 0(t)\), \(t=0,1,2,\dots\}\) the vector-valued sequence of input probabilities. Let \(Q(t)=P(t)+p_{k+1}'(t)p_ 0(t)\), and \(Q(s,m)=Q(s)Q(s+1)\dots Q(m)\). It is assumed that \(Q(t)\) converges in norm to a stochastic matrix \(Q\) of period \(d\). The authors prove that \(Q(kd+m,nd+r)\) converges at geometric rate uniformly in \(k\). Let \(T(t)=N_ 1(t)+N_ 2(t)+\dots+N_ k(t)\), and \(q(t)=N(t)/T(t)\), which is called the relative structure. It is proved that under certain conditions \(\{q(t)\), \(t=0,1,2,\dots\}\) splits into \(d\) convergent subsequences, and that each converging subsequence has a geometric rate of convergence.