Stationary analysis of certain Markov-modulated reflected random walks in the quarter plane (Q2115768)

Consider a random walk in the quarter plane \(\mathbb{Z}^2_+\) the dynamics of which depends on the state of a finite random environment \(S_0=\{0,1,\dots,N\}\). The joint process admits a Markov process modelling with increments \(\{-1,0,1,2,\dots\}\) for the random walk in both directions. The increments are homogeneous in space as long as the random walk stays in the interior of the quarter plane. If the environment is \(\geq 1\) the random walk can not jump downwards, so a drift to the origin needs the environment to stay in \(0\). Formally, in case of increments of size \(\leq 1\) this process is a two-dimensional version of a quasi-birth-death proces with phase space \(S_0\). In general the stationary distribution of this process is not explicitly accessible. Therefore application of the power series method is applied to obtain (asymptotically) access to the equilibrium behaviour of the system. A second approach is via generating functions which leads to a Riemann boundary value problem for matrix-valued functions. In a second step this is reduced to a similar problem for scalar function. The proofs are involved and lead to (rather technical) stability condition. Natural applications of the theoretical findings are to determine the stationary behaviour of two-queues systems with a supplementary component. An example is given: A retrial system with a finite capacity priority line and two coupled orbit queues.