Moments based approximation for the stationary distribution of a random walk in Z+ with an application to the M/GI/1/n queueing system

From MaRDI portal
Publication:4503229





The paper deals with Lindley processes \(X=(X_n)_{n\geq 0}\) on the set \(\mathbb{N} _0\) of nonnegative integers, that are defined by \(X_{n+1}=\max (0, X_n+A_n)\), where \((A_n)\) is a sequence of i.i.d. integer valued random variables with a negative first moment and \(E((A_n^+)^{1+s})<\infty\) for some \(s>0\). The sequence \(X\) forms an ergodic Markov chain on \(\mathbb{N}_0\), provided it is irreducible. The steady state distribution \(\pi\) of \(X\) is to be approximated by the steady state distribution \(\pi _n\) of some Markov chain \(Y_n=(Y_{nk})_{k\geq 0}\) with finite state space \(\{1,\ldots ,n\}\) and transition probabilities satisfying \[ P(Y_{n k+1}=j\mid Y_{n k}=i)\geq P(X_{k+1}=j\mid X_{k}=i) \quad \text{for }0\leq i,j \leq n. \] Using a coupling technique it is proved that the distance \(||\pi -\pi _n||:=\sum_{k=0}^\infty |\pi (k)-\pi _n(k)|\) is of order of magnitude \(o(n^{-s})\). In case \(s\geq 1\) the exponent \(s\) is best possible. This result is applied to determine the convergence rate of the time stationary distribution of an \(\text{M/GI}/1/n\)-queue to the time stationary distribution of the corresponding \(\text{M/GI}/1/\infty\)-queue as \(n\) tends to \(\infty\).











This page was built for publication: Moments based approximation for the stationary distribution of a random walk in Z+ with an application to the M/GI/1/n queueing system

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4503229)