Perfect sampling of ergodic Harris chains

We develop an algorithm for simulating "perfect" random samples from the in variant measure of a Harris recurrent Markov chain. The method uses backwar d coupling of embedded regeneration times and works most effectively for st ochastically monotone chains, where paths may be sandwiched between "upper" and "lower" processes. We give an approach to finding analytic bounds on t he backward coupling times in the stochastically monotone case. An applicat ion to storage models is given.