Two-sided taboo limits for Markov processes and associated perfect simulation

In this paper, we study the two-sided taboo limit processes that arise when a Markov chain or process is conditioned on staying in some set A for a long period of time. The taboo limit is time-homogeneous after time 0 and time-inhomogeneous before time 0. The time-reversed limit has this same qualitative structure. The precise transition structure at the taboo limit is identified in the context of discrete- and continuous-time Markov chains, as well as diffusions. In addition, we present a perfect simulation algorithm for generating exact samples from the quasi-stationary distribution of a finite-state Markov chain.