Weak convergence results for multiple generations of a branching process (Q548150)

Let \(\xi_{n,j}\), \(n,j \geq 1\) be i.i.d.\ \(\mathbb{N}_0\)-valued random variables with distribution \((p_j)_{j \geq 0}\) and mean \(m = \sum_{j \geq 0} j p_j > 0\). The authors consider the Galton-Watson branching process \((Z_n)_{n \geq 0}\) with \(Z_n = \sum_{j=1}^{Z_{n-1}} \xi_{n,j}\). \((Z_n)_{n \geq 0}\) is assumed to be non-degenerate in the sense that \(p_0 + p_1 < 1\). For a given complete metric space \((M,d)\), the existence of a sequence \((H_k)_{k \geq 0}\) of measurable mappings \(H_k: \mathbb{R}^k \to M\) is assumed. Further, a sequence \((r(n))_{n \geq 1}\) with \(1 \leq r(n) \leq n\) is fixed, \(r(n) \to \infty\) is allowed. This sequence is used to define \[ \mathbf{X}_{n,r(n)} := (X_{n,Z_{n-1}},X_{n-1,Z_{n-2}},\ldots,X_{n-r(n)+1,Z_{n-r(n)}},z,z,\ldots), \] where \(z\) is a fixed element in \(M\) and \[ X_{n-j+1,Z_{n-j}} = H_{Z_{n-j}}(\xi_{n-j+1,1},\ldots,\xi_{n-j+1,Z_{n-j}}); \] for convenience, \(z\) is chosen to be \(0\) when \(M\) contains a zero element \(0\), which is the case in all applications presented in the paper. The main theorem of the paper states the following. If the \(H_k\) are such that \(H_k(\xi_{1,1},\ldots,\xi_{1,k}) \to H\) weakly for some random element \(H\) in \(M\) and if \(P(Z_{n-1} > 0)/P(Z_n>0) \to 1\), then the probability measures \(\mu_n := \mathcal{L}(\mathbf{X}_{n,r(n)})|Z_{n-1}>0)\) converge weakly to \(\mathcal{L}(B_1,B_2,\ldots)\), where the \(B_j\) are i.i.d.\ copies of \(H\). Several applications of the main theorem are presented in the paper. Among these applications are a functional CLT, the Darling-Erdős theorem and a new proof of the CLT for a non-parametric maximum-likelihood estimator of the mean of a supercritical branching process.