Fast simulation for slow paths in Markov models

Inspired by applications in the context of stochastic model checking, we are interested in using simulation for estimating the probability of reaching a specific state in a Markov chain after a large amount of time tau has passed. Since this is a rare event, we apply importance sampling. We derive approximate expressions for the sojourn times on a given path in a Markov chain conditional on the sum exceeding tau, and use those expressions to construct a change of measure. Numerical examples show that this change of measure performs very well, leading to high precision estimates in short simulation times.