Parametrised branching processes: a functional version of Kesten \& Stigum theorem

Abstract

Let $(Z_n,n\geq 0)$ be a supercritical Galton-Watson process whose offspring distribution $\mu$ has mean $\lambda>1$ and is such that $\int x(\log(x))_+ d\mu(x)<+\infty$. According to the famous Kesten \& Stigum theorem, $(Z_n/\lambda^n)$ converges almost surely, as $n\to+\infty$. The limiting random variable has mean~1, and its distribution is characterised as the solution of a fixed point equation. \par In this paper, we consider a family of Galton-Watson processes $(Z_n(\lambda), n\geq 0)$ defined for~$\lambda$ ranging in an interval $I\subset (1, \infty)$, and where we interpret $\lambda$ as the time (when $n$ is the generation). The number of children of an individual at time~$\lambda$ is given by $X(\lambda)$, where $(X(\lambda))_{\lambda\in I}$ is a c\`adl\`ag integer-valued process which is assumed to be almost surely non-decreasing and such that $\mathbb E(X(\lambda))=\lambda >1$ for all $\lambda\in I$. This allows us to define $Z_n(\lambda)$ the number of elements in the $n$th generation at time $\lambda$. Set $W_n(\lambda)= Z_n(\lambda)/\lambda^n$ for all $n\geq 0$ and $\lambda\in I$. We prove that, under some moment conditions on the process~$X$, the sequence of processes $(W_n(\lambda), \lambda\in I)_{n\geq 0}$ converges in probability as~$n$ tends to infinity in the space of c\`adl\`ag processes equipped with the Skorokhod topology to a process, which we characterise as the solution of a fixed point equation.