The Galton-Watson process with killing (Q794353)

Recently \textit{S. Karlin} and \textit{S. Tavaré} [ibid. 59, 57-75 (1982; Zbl 0503.92015)] introduced the notion of a killed Galton-Watson process. In their model an ordinary Galton-Watson process \(\{Z_ n\}\) is killed (this amounts to the introduction of a second absorbing state) at time n with probability \(\alpha^ j\) if \(Z_ n=j\). They show that this population must either become extinct or be killed and obtain the conditional limit for the population size given that neither has yet occurred. The present paper shows how these results can be obtained quite simply from known results on the total progeny of the branching process. A substantial extension of the model is also considered in which killing occurs at time n with probability a(j) when \(Z_ n=j\). It is shown that if a(j) approaches one sufficiently quickly the process can avoid both extinction and killing and then, if the offspring mean is finite, the usual norming leads to convergence. Conditions for the convergence in distribution of the population size conditional on neither extinction nor killing having occurred are also given. Use is made of the R- classification of the processes transition matrix, discussed by \textit{E. Seneta} [Non-negative matrices (1973; Zbl 0278.15011)] to obtain the latter results.