No label defined (Q4258741)

The hitting time for a non-homogeneous discrete time Markov chain (NHMC) is studied. Let \(X=\{X_{n}, n\in {\mathbf N}\}\) be a NHMC, \(E=\{1,2,\ldots, s\}\) be the set of states of \(X.\) Let \(p_{n,m}(i,j)=P\{X_{m}=j, X_{n}=i\}\), \(m>n\geq 0\), be the transition probability, \(p_{n}(i,j)=p_{n,n+1}(i,j)\), \(P_{n}=(p_{n,n+1}(i,j))\) be the transition matrix. Let \(T\) be the hitting time in a subset \(B\subset E\), when the chain is initially in a state \(A\) \((A=E\setminus B)\), i.e. \(T(\omega)=\inf\{m>0; X_{m}(\omega)\in B\}.\) The transition matrix \(p_{n}\) of \(X\) is partitioned as follows \[ P_{n}=\begin{pmatrix} P_{n}^{A}&P_{n}^{AB}\\ P_{n}^{BA}&P_{n}^{B}\end{pmatrix}. \] The authors find the distribution of the hitting time \[ P({\mathbf{T}}>n)=\alpha_{A}\prod\limits_{k=0}^{n-1}p_{k}^{A} {\mathbf{1}}_{d},\quad {\mathbf{1}}_{d}=(1,\ldots,1)^{\top}, \] where \(\alpha_{A}\) is the initial distribution of \(X_{n}.\) The authors find the mean hitting time of \(X.\) For example, the next assertion is proved. When at all times \(n\geq 0\), \(B^{c}\) is a transient class, \(B\) is an absorbing class and the sequence \((k_{n})\) verifies \(\lim\limits_{m\to\infty}\sup(k_{m+1}- k_{m})^{1/m}<{1\over\theta}\) (\(k_{n}\) is an increasing sequence from the \(\theta\)-transient class), then the mean hitting time in the subset \(B\) while initially being in \(B^{c}\) is finite and is given by \(E(T)=\alpha_{A}[{\mathbf{1}}_{d}+ \sum_{n\geq 0}\sum_{k=0}^{n}p_{k}^{A} {\mathbf{1}}_{d}].\) A numerical example and an application in reliability evolution are given.