A note on the local asymptotic mixed normality of a controlled branching process with a random control function (Q2070638)

This paper treats a controlled branching process with a random function (CBP), which is a generalized notion of a fundamental Galton-Watson branching process (GWBP). CBP has been extensively discussed, for example, by \textit{M. González} et al. [J. Appl. Probab. 40, No. 4, 995--1006 (2003; Zbl 1054.60087)]. Let \(Z_n\) denote the number of individual in \(n\)-th generation in the model, \(\phi_n(k)\) a control random variable. \(p_k = p(k; \psi)\) denotes the common probability distribution of non-negative integer-valued i.i.d. random variable \(X_{ni}\) defined on a probability space \(( \Omega, \mathcal{F}, P)\). When \(\Lambda_n(\psi_n, \psi_0)\) denote the log of likelihood ratio for testing \(H_0 : \psi = \psi_0\) against \(H_n : \psi = \psi_n := \psi_0 + \frac{h}{\sqrt{ \delta_n} }\), with a fixed real number \(h\) and the unconditioned Fisher information function \(\delta_n = I_{ N_n}\), then under some conditions on \(Z_n\) and \(\phi_n(k)\), and under the condition on smoothness of \(p(j; \psi)\), it can be proved that \(\Lambda_n( \psi_n, \psi_0)\) is permitted to possess an asymptotic expansion, and also that \(\Lambda_n(\psi_n, \psi_0)\) is asymptotically mixed normal as \(n \to \infty\) in the sense that \[ \Lambda_n(\psi_n, \psi_0) \stackrel{\mathcal{L}}{\longrightarrow} h \sqrt{ W^*} Z - \frac{1}{2} h^2 W^* ,\] where \(Z \sim N(0,1)\) and \(W^*\) is an almost sure finite and positive random variable, and \(W^*\) is obtained by the limit \[\lim_{n \to \infty} \frac{N_n}{ \mathbb{E} [ N_n] } = W^* \] almost surely, where \(N_n\) is a random sample of size. Thus they attain the main result, the local asymptotic mixed normality property of a controlled branching process with a random control function, under the assumption that the process is supercritical and on a non-extinction path. For other related works, see, e.g., [\textit{M. González} et al., Comput. Stat. Data Anal. 93, 209--227 (2016; Zbl 1468.62067)] for maximum likelihood (ML) estimation and expectation maximization algorithm for controlled branching processes, where the ML estimation of some parameters of a CBP has been discussed, and a fascinating result that estimators are asymptotically normal with random scaling is suggested. It would be possible to say that this provides a leading philosophy to the authors of the present paper.