Abstract. These notes provide an elementary and self-contained introduction to branching random walks.Section 1 gives a brief overview of Galton-Watson trees, whereas Section 2 presents the classical law of large numbers for branching random walks. These two short sections are not exactly indispensable, but they introduce the idea of using size-biased trees, thus giving motivations and an avant-goût to the main part, Section 3, where branching random walks are studied from a deeper point of view, and are connected to the model of directed polymers on a tree.Tree-related random processes form a rich and exciting research subject. These notes cover only special topics. For a general account, we refer to the St-Flour lecture notes of Peres [47] References 38
Galton-Watson treesWe start by studying a few basic properties of supercritical Galton-Watson trees. The main aim of this section is to introduce the notion of size-biased trees. In particular, we see in Subsection 1.3 how this allows us to prove the well-known Kesten-Stigum theorem. This notion of size-biased trees will be developed in forthcoming sections to study more complicated models.
Galton-Watson trees and extinction probabilitiesWe are interested in processes involving (rooted) trees. The simplest rooted tree is the regular rooted tree, where each vertex has a fixed number (say m, with m > 1) of offspring. For example, here is a rooted binary tree:Let Z n denote the number of vertices (also called particles or individuals) in the n-th generation, thenIn probability theory, we often encounter trees where the number of offspring of a vertex is random. The easiest case is when these random numbers are i.i.d., which leads to a Galton-Watson treeA Galton-Watson tree starts with one initial ancestor (sometimes, it is possible to have several or even a random number of initial ancestors, in which case it will be explicitly stated). It produces a certain number of offspring according to a given probability distribution. The new particles form the first generation. Each of the new particles produces offspring according to the same probability distribution, independently of each 1 Sometimes also referred to as a Bienaymé-Galton-Watson tree. other and of everything else in the generation. And the system regenerates. We write p i for the probability that a given particle has i children, i ≥ 0; thus To avoid trivial discussions, we assume throughout that p 0 + p 1 < 1.As before, we write Z n for the number of particles in the n-th generation. It is clear that if Z n = 0 for a certain n, then Z j = 0 for all j ≥ n. One of the first questions we ask is about the extinction probability q := P{Z n = 0 eventually}.(1.1) 4
ESAIM: PROCEEDINGSIt turns out that the expected number of offspring plays an important role. Let(1.2) Theorem 1.1. Let q be the extinction probability defined in (1.1).(i) The extinction probability q is the smallest root of the equation f (s) = s for s ∈ [0, 1], whereis the generating function of the reproduction law.(ii) In particular, q = 1 if m ≤...