On the genealogical structure of critical branching processes in a varying environment (Q2135132)

This paper treats the most recent common ancestor given survival for critical branching processes (CBP) in a varying environment, and the author tries to confirm the similarity of behaviors between CBP in a varying environment and critical Galton-Watson (GW) processes via the underlying genealogical structures. More precisely, let \(Y_1, Y_2, \dots\) denote a sequence of random variables with values in \(\mathbb{N}_0\) and \(f_1, f_2, \dots\) their distributions. Let \(Y_{in}\), \(i, n \in \mathbb{N}\), be independent random variables such that \(Y_{in}\) and \(Y_n\) coincide in distribution for all \(i, n \geq 1\). Define the random variables \(Z_n\), \(n \geq 0\), with values in \(\mathbb{N}_0\) recursively as \[ Z_0 := 1, \qquad Z_n := \sum_{i=1}^{Z_{n-1}} Y_{in}, \qquad n \geq 1. \] Then the process \((Z_n)_{n \geq 0}\) is called a branching process (BPVE) in the varying environment \(v = (f_1, f_2, \dots)\) with initial value \(Z_0 =1\). The condition \(0 < \mathbb{E} [Y_n] < \infty\) is always assumed for all \(n \geq 1\). Note that the BPVE \((Z_n)_{n \geq 0}\) is critical if both conditions \[ \mathbb{P} \left(Z_n > 0\right) \to 0 \text{ and }\mathbb{E} [Z_n \mid Z_n > 0] \to \infty \] as \(n \to \infty\) are satisfied. The first main result asserts: \textbf{Theorem 1.} Assume that \[ \mathbb{E}[Y_n^2; Y_n > c_{\varepsilon} (1 + \mathbb{E} [Y_n])] \leqslant \varepsilon \cdot \mathbb{E} [Y_n^2; Y_n \geq 2] < \infty \tag{\(*\)} \] for all \(n \geq 1\). For a critical BPVE \((Z_n)_n\), we have \[ \max_{0 \leqslant k \leqslant n}]\left\vert \mathbb{P} (G_n \leqslant k \mid Z_n > 0) - \frac{\rho_k}{\rho_n} \right\vert \longrightarrow 0 \text{ as }n \to \infty, \] where \(\rho_n\) is defined by \[ \rho_n := \frac{\mathbb{E} [Z_n (Z_n -1)]}{\mathbb{E} [Z_n]^2}, \qquad n \geq 0. \] \textit{A. M. Zubkov} [Theory Probab. Appl. 20, 602--612 (1975; Zbl 0348.60118); translation from Teor. Veroyatn. Primen. 20, 614--623 (1975)] proved that for critical GW processes \(W_n := G_n / n\) has asymptotically a uniform distribution on the interval \([0,1]\), conditional on the event that \(Z_n > 0\), where \(G_n\) is the generation of the most recent common ancestor of the \(Z_n\) particles in generation \(n\). The above first main result (= Theorem 1) generalized the above-mentioned remarkable result by Zubkov for GW processes \((W_n)\) to BPVEs \((Z_n)\). From the generation of the most recent common ancestor, the author steps forward to the more elaborate reduced branching trees given the event \(\{Z_n > 0 \}\). As introduced by \textit{K. Fleischmann} and \textit{R. Siegmund-Schultze} [Math. Nachr. 79, 233--241 (1971; Zbl 0299.60065)] in the case of GW trees, there are the subtrees containing all individuals which possess descendants in generation \(n\). Accordingly, the reduced branching process \((Z_{k,n})_{0 \leqslant k \leqslant n}\) is made up of the numbers \(Z_{k,n}\) of individuals in generation \(k \leqslant n\) having descendants in generation \(n\). Just as in the GW case, this process is a non-homogeneous Markov chain, actually a BPVE. It turns out that asymptotically the process \((Z_{[nt], n})_{0 \leqslant t \leqslant 1}\) coincides with a properly time-changed Yule process \(Y\). Here is the second main result of this paper. \textbf{Theorem 2.} Assume the condition (\(*\)). Then for the reduced process of a critical BPVE we have \[ \left((Z_{k_n(t), n})_{0 \leqslant t \leqslant 1} \mid Z_n > 0\right) \quad \stackrel{d}{\Longrightarrow} \quad \left(Y \left(\log \frac{1}{1-t} \right)\right)_{0 \leqslant t < 1}, \ (\text{as }n \to \infty) \] where \((Y(u))_{u \geq 0}\) denotes a standard Yule process, and \(k_n: [0,1] \to \{0,1,2, \dots, n-1\}\) is given by \[ k_n(t) := \max \{k \geq 0: \rho_k \leqslant t \rho_n\}, \qquad 0 \leqslant t < 1. \] Here the notation \(\stackrel{d}{\Longrightarrow}\) indicates that for any \(0 < \eta < 1\) we have the convergence in the Skorohod sense of the sequence of processes restricted to time \(t \in [0, \eta]\). This Theorem 2 contains the result of Fleischmann and Siegmund-Schultze [loc. cit.] already referred to in the above) as a special case. The author of this paper states in his previous paper [J. Appl. Probab. 57, No. 1, 196--220 (2020; Zbl 1434.60242)], that under the assumption (\(*\)), if the probability of extinction is \(1\), that is, if \(\mathbb{P}(Z_n > 0) \to 0\) as \(n \to \infty\), then the following statements are equivalent: \begin{itemize} \item[(i)] there is a sequence \(b_n\), \(n \geq 0\), of positive numbers such that \(Z_n/b_n\) conditioned on the event \(\{Z_n > 0 \}\) converges in distribution to a standard exponential distribution as \(n \to \infty\); \item[(ii)] \(\mathbb{E} [Z_n \mid Z_n > 0] \to \infty\), (as \(n \to \infty\)); \item[(iii)] \(\mu_n \rho_n \to \infty\), (as \(n \to \infty\)), where \(\mu_n := \mathbb{E}[Z_n]\) for all \(n \geq 0\); and also that for the case of \(b_n := \mathbb{E} [Z_n \mid Z_n > 0]\), \begin{align*} \mathbb{E} [Z_n \mid Z_n > 0] \quad &\sim \quad \frac{1}{2} \mu_n \rho_n \\ \text{or equivalently } \mathbb{P} (Z_n > 0) \quad &\sim \quad \frac{2}{\rho_n} \text{ as } n \to \infty. \end{align*} \end{itemize} This result characterizes in the context of BPVEs the range where \textit{A. M. Yaglom}'s asymptotics for critical GW processes on the conditional distribution of \(Z_n\) holds [Dokl. Akad. Nauk SSSR, n. Ser. 56, 795--798 (1947; Zbl 0041.45602)]. Therefore, assuming the condition (\(*\)) it is quite natural to transfer by means of the above previous result the notion of criticality from GW processes to BPVEs. The author makes use of this trick very well to derive the main results of this paper.