Measure-valued Markov branching processes conditioned on non-extinction (Q2638674): Difference between revisions

Let \(\xi =\{\xi_ t\), \(t\geq 0\}\) be a right Markov process with state space E and transition semigroup \(\{P_ t,\quad t\geq 0\},\) and let \(X=\{X_ t,\quad t\geq 0\}\) be the associated ``Dawson-Watanabe process''. This is the Markov branching process, taking values in the space of finite measures on E, with law determined by \[ E(\exp (-X_ t(f))| X_ 0=\mu)=\exp (-\mu (V_ t(f))), \] where f: \(E\to [0,\infty [\) is bounded, and \(V_ t(f)\) is the unique solution of the integral equation \[ v_ t(x)=P_ tf(x)-\int^{t}_{0}P_ s(x,v^ 2_{t-s})ds. \] The process X may be viewed as a ``high density'' limit of certain branching particle systems, the individual particle motions being determined by the underlying process \(\xi\) ; the zero measure (which is a trap for X) corresponds to extinction. This note is concerned with the limit distribution (as \(t\to \infty)\) of \(t^{-1}X_ t\) conditional on nonextinction up to time \(\beta\) t, for \(\beta\in [1,\infty]\). (When \(\beta =\infty\), the condition \(\{X_ s\neq 0\), \(0\leq s\leq \beta t\}\) must be interpreted in a limiting sense.) It is shown that if \(\xi\) is ``positive recurrent'' in either of two senses, then the distribution of \(\{t^{-1}X_ t| \quad X_ s\neq 0,\quad 0\leq s\leq \beta t\}\) converges to that of \((Y+(1-1/\beta)Z)\nu,\) where \(\nu\) is the invariant distribution of \(\xi\), and Y, Z are independent unit exponential random variables.