Branching random walk with infinite progeny mean: a tale of two tails (Q6044249)

The main focus of this article is the analysis of the behaviour of a branching random walk in which the offspring distribution has infinite mean. This is a particle system on the real line in which at each step, every particle give birth independently to one another to a random number of children, with probability \(\{p_k\}_{k \geq 0}\). The children move around their parent according to i.i.d. copies of a real-valued random variable \(X\). For any \(n \in \mathbb N\), we denote by \(Z_n\) the number of particles alive at time \(n\) and by \(M_n\) the position of the maximal displacement at time \(n\). Under (a generalization of) the assumption that \(Z_1\) belong to the domain of attraction of an stable random variable with parameter \(\alpha \in (0,1)\), \textit{P. L. Davies} [J. Appl. Probab. 15, 466--480 (1978; Zbl 0387.60093)] proved that \(\alpha^n \log (Z_n+1)\) converges to a non-degenerate limit \(W\) almost surely. Note that contrarily to classical settings for branching random walks, the intensity measure of the offspring distribution possesses does not have any exponential moment. Under this assumption on the reproduction law, the authors study the asymptotic behaviour of \(M_n\) depending on the behaviour of the tail of the displacement of particles, proving notably the following results. \begin{itemize} \item If \(\mathbb P(X > x) \sim x^{-\beta}\) as \(x \to \infty\), then \(\alpha^n \log M_n \to \frac{W}{\beta}\) in probability as \(n\to\infty\). \item If \(-\log \mathbb P(X > x) \sim x^r\) as \(x \to \infty\), then \(\alpha^{n/r} M_n \to c_{\alpha,r}W^{1/r}\) a.s. as \(n\to\infty\). \item If the right-continuous inverse \(L\) of \(-\log \mathbb P(X > x)\) is slowly varying at \(\infty\) and regular enough, then \(M_n / \sum_{k=1}^\infty L(\alpha^{-k}) \to 1\) a.s. as \(n\to\infty\). \end{itemize} Moreover, under the assumption that \(\mathbb P(X > x) \sim x^{-\beta}\), they show that the rescaled empirical measure of the process converges towards a Poisson random measure with intensity \(x^{-\beta - 1} \mathrm d x\).