On the ergodic theory of critical branching Markov chains (Q1318328)

The author considers a system of particles moving randomly and branching on a countable set \(X\). Each particle lives an exponentially distributed time, and at the end of its lifetime it is replaced (at the same site) by a random number of particles according to a critical offspring distribution with finite second moment. During its lifetime, a particle moves at rate 1 according to a Markov transition kernel \(p(x,y)\) on \(X\). The (measure-valued) process under study is \(\eta_ t(y)\), which represents the number of particles occupying site \(y\) at time \(t\). Under some natural conditions, such a model obeys an extinction/stability dichotomy, i.e., as \(t\to\infty\), either \(\eta_ t\) becomes extinct or it converges in law towards a non-degenerate invariant measure. This dichotomy is related to recurrence/transience properties of the kernel \(p(x,y)\). The paper deals only with the stable (i.e. \(p(x,y)\) transient) case. The main objective is to answer the question: for what initial states \(\eta_ 0\) does the process \(\eta_ t\) converge in law as \(t\to\infty\)? For a function \(\pi: X\to{\mathbb{R}}^ +\), under certain conditions on the kernel \(p\), if \(\eta_ 0p_ t\to\pi\) in probability as \(t\to\infty\) (i.e. the law of the initial state is carried towards \(\pi\) by the transition probability of the process \(\eta_ t\)), the author proves that the process \(\eta_ t\) converges in law as \(t\to\infty\) towards a measure \(\mu_ \pi\) (on a certain class of configurations), which is determined by \(\pi\) by means of a Laplace transform formula. In the case of deterministic initial states, the result is that \(\eta_ 0p_ t\to\pi\) if and only if \(\eta_ t\) converges in law towards \(\mu_ \pi\). The usual situation is that the measure \(\mu_ \pi\) is invariant for \(\eta_ t\), but this need not occur, and the author gives an example, due to Tom Liggett, of this phenomenon. Previous work on the subject dealt with random initial states which are either translation invariant and/or satisfy a domination condition involving a reference measure. Such conditions are not assumed in this paper.