Time-homogeneous birth-death processes with probability intervals and absorbing state (Q2431658): Difference between revisions

Consider a discrete-time birth-death process on a finite state space consisting of a single absorbing state and an irreducible, aperiodic set of transient states with one-step transition probabilities. In the classical case, where the transition probability matrix is precisely known, the process becomes eventually extinct, but may, however, exhibit a quasi-stationary behaviour, which can be described by the limiting conditional distribution, as time tends to infinity, given that extinction has not yet occurred. This distribution is closely related to the quasistationary distribution [\textit{J. N. Darroch} and \textit{E. Seneta}, J. Appl. Probab. 2, 88--100 (1965; Zbl 0134.34704)], which is an initial distribution on the non-absorbing states such that the conditional distribution of the process, given that absorption has not yet occurred, is independent of time. In applications, the transition probability matrix may not be known precisely, thus allowing for imprecision in the model is of interest. There are various ways of modelling imprecision. The approach taken here, which assumes constant probabilities known to exist within constant intervals, is in the spirit of \textit{K. Weichselberger} [Int. J. Approx. Reasoning 24, No. 2--3, 149--170 (2000; Zbl 0995.68123)]. The theory of imprecise probability in this sense is combined with the classical theory of quasi-stationarity to analyse the behaviour of imprecise absorbing birth-death processes. The authors show that set of all possible limiting conditional distributions is the set of all possible quasi-stationary distributions. They give an approximation of the possible infinite set of conditional distributions at time \(n\) together with an example.