Perfect simulation and monotone stochastic bounds

International audienceWe combine monotone bounds of Markov chains and the coupling from the past to obtain an exact sampling of a strong stochastic bound of the steady-state distribution for a Markov chain. Stochastic bounds are sufficient to bound any positive increasing rewards on the steady-state such as the loss rates and the average size or delay. We show the equivalence between st-monotonicity and event monotonicity when the state space is endowed with a total ordering and we provide several algorithms to transform a system into a set of monotone events. As we deal with monotone systems, the coupling technique requires less computational efforts for each iteration. Numerical examples show that we can obtain very important speedups. Copyright 2007 IC