Strong memoryless times and rare events in Markov renewal point processes. (Q1889787): Difference between revisions

The paper is concerned with rare events in a stationary finite-state Markov renewal point process (MRPP). Each point of MRPP has an associated mark or state. The distance between two successive points and the state of the second point are jointly conditionally independent of the past given the state of the first point. If the points with a given set of states are rare, then it should be approximately compound Poisson distributed. The main result of the paper is to give an upper bound for the total variation distance between the distribution of this quantity and a particular Poisson distribution. It is expressed in terms of the first two moments of the interrenewal time conditional distributions and two constants obtained from each Radon-Nikodym derivative of an interrenewal time conditional distribution with respect to an exponential distribution. To solve the problem one studies a pair of random variables \((\zeta ,V)\), where \(\zeta \) is the distance between two successive points and \(V\) is the state of the second point. There is constructed a probability space containing \((\zeta , V)\) and a third random variable \(\widehat \zeta \) such that for all \(t\), conditional on \(\{\widehat \zeta \leq t, \zeta >t\}\), the pair \((\zeta -t,V)\) has the distribution \(\nu _{\gamma }\times \mu \), where \(\nu _{\gamma }\) is an exponential (or geometrical) distribution with mean \(\gamma ^{-1}\) and \(\mu \) is a fixed distribution, so the event \(\{\widehat \zeta \leq t, \zeta >t\}\) indicates a loss of memory at or before \(t\). Using this property the stationary MRPP is embedded into another MRPP with an additional state 0. The points with state 0 represent losses of memory in the original MRPP, the bound is then derived by an application of the author's theorem [Probab. Theory Relat. Fields 117, No. 2, 145--161 (2000; Zbl 0961.60084)] to the accumulated reward of a renewal process embedded into the new MRPP, for which the points 0 serve as renewals.