On \(L^p\)-convergence of the Biggins martingale with complex parameter (Q2320157)

Consider a branching random walk on the real line starting with one initial ancestor at the origin at time \(n=0\) and children located at the points of a random point process \(\mathcal{M}\) on \(\mathbb{R}\). The children of the ancestor form the first generation. All individuals in the process reproduce according to independent copies of the initial point process \(\mathcal{M}\) but shifted by their position. The process is assumed supercritical, i.e., \(\mathbb{E}[\mathcal{M}(\mathbb{R})] > 1\). Denote the Laplace transform of the intensity measure \(\mathbb{E}[\mathcal{M}(\cdot)]\) by \(m\), so \[ m(\lambda) = \mathbb{E}\bigg[\sum_{|u|=1} e^{-\lambda S(u)} \bigg] \] where summation over \(|u|=1\) is summation over the \(1\)st generation individuals and the position of an individual \(u\) is denoted by \(S(u)\). Now if \(\lambda \in \mathbb{C}\) is such that \(m(\lambda)\) is finite and \(\not = 0\), then \[ Z_n(\lambda) = m(\lambda)^{-n} \sum_{|u|=1} e^{-\lambda S(u)} \] where summation over \(|u|=n\) means summation over the \(n\)th generation individuals is a complex-valued martingale called Biggins martingale (with complex parameter). These martingales are important in the study of the asymptotic behavior of the branching random walk. Their convergence almost surely and in \(L^p\) has therefore been studied earlier by \textit{J. D. Biggins} [Ann. Probab. 20, No. 1, 137--151 (1992; Zbl 0748.60080)] and \textit{K. Kolesko} and the reviewer [Electron. Commun. Probab. 22, Paper No. 18, 14 p. (2017; Zbl 1359.60105)]. The paper under review aims at providing equivalent conditions for the convergence in \(L^p\) of \(Z_n(\lambda)\) as \(n \to \infty\) for \(p > 1\). This goal is fully achieved in the case \(p \geq 2\), in which case the authors show that for \(\lambda \not = 0\) with \(0 < |m(\lambda)| < m(\theta) < \infty\) (and \(\theta = \mathrm{Re}\, \lambda\)) the martingale \((Z_n(\lambda))_{n \in \mathbb{N}_0}\) converges in \(L^p\) for \(\lambda \not = 0\) iff \(\mathbb{E}[|Z_1(\lambda)|^p]<\infty\), \[ \frac{m(2\theta)}{|m(\lambda)|^2} \vee \frac{m(p\theta)}{|m(\lambda)|^p} < 1 \] and, additionally in the case \(p>2\), \(\mathbb{E}[Z_1(2\theta)^{p/2}]<\infty\). A similar but simpler condition is provided in the case \(\lambda = 0\). The case \(p \in (1,2)\) is more intricate, yet an equivalent characterization of \(L^p\) convergence in this case is achieved subject to additional assumptions on the branching random walk, namely, that either \(\mathbb{E}[|Z_1(\lambda)|^2]<\infty\) or \(\mathbb{P}(|Z_1(\lambda)|>x) \in \Theta(x^{-\alpha})\) as \(x \to \infty\) for some \(\alpha \in (1,2)\) such that the real martingale \((Z_n(\alpha \theta))_{n \in \mathbb{N}_0}\) is uniformly integrable. (Again, \(\theta\) denotes the real part of \(\lambda\).) If \(\mathrm{Re} \, \lambda, \mathrm{Im} \, \lambda \not= 0\) and \(0 < |m(\lambda)| < m(\theta) < \infty\), then, under the above additional assumptions, the \(L^p\)-convergence of \((Z_n(\lambda))_{n \in \mathbb{N}_0}\) is equivalent to \(\alpha > p\) and \[ \frac{m(\alpha\theta)}{|m(\lambda)|^\alpha} < 1. \] Here, \(\alpha = 2\) in the case where \(\mathbb{E}[|Z_1(\lambda)|^2]<\infty\). There is a similar statement for the case \(\mathrm{Re} \, \lambda = 0\).