Sensitivity analysis for Markov reward structures until entrance times. (English) Zbl 0962.60061

van Dijk, N. M.; Korezlioglu, H.

J. Appl. Probab. 37, No. 1, 45-63 (2000).

The sensitivity of a performance measure to model specification can sometimes be determined without first computing the measure itself. In this vein, the authors prove and apply a perturbation result for cumulative rewards up to an entrance time for a continuous-time Markov chain, complementing related discrete-time results by N. M. van Dijk and M. L. Puterman [Adv. Appl. Probab. 20, No. 1, 79-98 (1988; Zbl 0642.60100)]. The result effectively reduces sensitivity analysis here to the bounding of transient reward bias terms. An application to saturation times in circuit switch networks with unreliable links illustrates how this can be achieved in a practical case, albeit with substantial further calculation.

Reviewer: J.Preater (Keele)

MSC:

60J27	Continuous-time Markov processes on discrete state spaces
90B18	Communication networks in operations research
90B25	Reliability, availability, maintenance, inspection in operations research

Keywords:

continuous-time Markov chain; cumulative reward; entrance time; parameter sensitivity; error bound; bias term; uniformization

Citations:

Zbl 0642.60100