Pruning Galton-Watson trees and tree-valued Markov processes (Q441242)

Let \(G_p\) be a Galton-Watson tree with offspring distribution \(p\) and \(u\in (0,1)\) be given. The authors introduce the following pruning-at-nodes procedure. If \(\nu\) is a non-leaf node of \(G_p\) that has \(n\) children, one discards this node along with all the subtrees emanating from it with probability \(u^{n-1}\) independently of the other nodes. The resulting tree is still a Galton-Watson tree with offspring distribution \(p^{(u)}\) defined by \(p_n^{(u)}:=u^{n-1}p_n\), \(n\in N\) and \(p_0^{(u)}:=1-\sum_{n\geq 1}p_n^{(u)}\). Using this procedure, the authors construct a tree-valued inhomogeneous Markov process \((G(u): u\in [0,\bar{u}])\) for some \(\bar{u}\geq 1\) such that, for every \(u\in [0,\bar{u}]\), \(G(u)\) is a Galton-Watson tree with offspring probability \(p^{(u)}\), and investigate its properties. Applying the same pruning-at-nodes procedure to a critical or a subcritical Galton-Watson tree conditioned on non-extinction, another tree-valued process \((G^\ast(u):u\in [0,1])\) is constructed. A representation is also given which relates the law of \((G(u))\) stopped at the first time the tree becomes infinite to the law of \((G^\ast(u))\).