AGROCAMPUS OUEST (Institut Supérieur des Sciences Agronomiques, Agroalimentaires, Horticoles et du Paysage - 65, rue de St Brieuc - CS 84215 - 35042 Rennes cedex - France)
Abstract : Let be a continuous time Markov chain with finite or countable state space S and let T be its first passage time in a subset D of S. It is well known that if mu is a quasi-stationary distribution relative to T, then this time is exponentially distributed under . However, quasi-stationarity is not a necessary condition. In this paper, we determine more general conditions on an initial distribution mu for T to be exponentially distributed under . We show in addition how quasi-stationary distributions can be expressed in terms of any initial law which makes the distribution of T exponential. We also study two examples in branching processes where exponentiality does imply quasi-stationarity.