Galton-Watson trees with vanishing martingale limit (Q2511525): Difference between revisions

The authors prove that a Galton-Watson tree with positive martingale limit \(W\) conditioned on \(W < \varepsilon\) coincides a with regular \(\mu\)-ary tree in the first \(K\) generations, where \(\mu\) is the minimal number of offspring per individual and \(K\) is a random variable that is strongly concentrated around a function growing like a constant times \(\log (1/\varepsilon)\) as \(\varepsilon \to 0\). The precise statement of the main result is the following. Let \((Z_n)_n\) denote the sequence of generation sizes of a (single-ancestor) Galton-Watson tree with non-degenerate offspring variable \(N\) satisfying \(\mu = \min\{n \in \mathbb{N}_0: \operatorname{P}(N=n) > 0\} > 0\) and mean offspring number \(a = \operatorname{E} N > 1\). Further, assume that \(\operatorname{E} N \log N < \infty\). The assumptions imply that the martingale limit \(W = \lim_{n \to \infty} Z_n/a^n\) is positive with probability one. Define \(K := \min\{k \in \mathbb{N}: Z_k > \mu^k\}\). In the case \(\mu=1\), the authors prove that there is some \(\zeta > 0\) such that \[ \limsup_{\varepsilon \downarrow 0} \operatorname{P}\Big(\Big|K-\frac{\log(1/\varepsilon)}{\log(a)} \Big| \geq x \, \Big| \, W < \varepsilon \Big) ~\leq~ e^{-\zeta x} \quad \text{for all } x \geq 1. \] The main result of the paper is on the case \(\mu \geq 2\), sometimes called Böttcher case. For this case, it is shown that \[ \lim_{\varepsilon \downarrow 0} \operatorname{P}(K \in \{\lceil \gamma(\varepsilon) \rceil, \lceil \gamma(\varepsilon) \rceil + 1\} \mid W < \varepsilon ) ~=~ 1 \] for \[ \gamma(\varepsilon) = \frac{\log(1/\varepsilon)}{\log(a/\mu)} - \frac{\log\log(1/\varepsilon)}{\log\mu} + H(\varepsilon), \] where \(H\) is a deterministic continuous and multiplicatively periodic function with period \(a/\mu\). Finally, let \(\mathcal{T}\) denote the space of rooted trees in which every vertex has a finite degree, and let \(d\) be the metric on \(\mathcal{T}\) defined by \(d(T_1,T_2) = e^{-n}\), where \(n\) is the maximal number \(k\) such that \(T_1\) and \(T_2\) coincide up to the \(k\)-th generation. Then, the main results imply that (under the assumptions stated above) a Galton-Watson tree conditioned on the event \(\{W < \varepsilon\}\) converges in distribution on \((\mathcal{T},d)\) to the regular \(\mu\)-ary tree as \(\varepsilon \to 0\).