A Kernel Method for Exact Tail Asymptotics-Random Walks in the Quarter Plane

Abstract

In this paper, we propose a kernel method for exact tail asymptotics of a random walk to neighborhoods in the quarter plane, which is a general model for two-dimensional queueing systems with many important applications in various areas including service management. This is a two-dimensional method, which does not require a determination of the unknown generating function(s). Instead, in terms of the asymptotic analysis and a Tauberian-like theorem, we show that the information about the location of the dominant singularity or singularities and the detailed asymptotic property of the unknown function at a dominant singularity is sufficient for the exact tail asymptotic behaviour for the marginal distributions and also for joint probabilities along a coordinate direction. We provide all details, not only for a "typical" case, the case with a single dominant singularity for an unknown generating function, but also for all non-typical cases which have not been studied before. A total of four types of exact tail asymptotics are found for the typical case, which have been reported in the literature. We also show that on the circle of convergence, an unknown generating function could have two dominant singularities instead of one, which can lead to a new periodic phenomena. Examples are illustrated by using this kernel method. This paper can be considered as a systematic summary and extension of existing ideas, which also contains new and interesting research results.