We want to recover the transition kernel P of a Markov chain X when only a sub-sequence of X is available. The time gaps between the observations are iid with unknown distribution. We propose a method to build an estimator of P under the assumption that it has some zero entries. Its asymptotic...