Autoregressive approximation in branching processes with a threshold (Q1332314): Difference between revisions

A classical Galton-Watson branching process is generalized in such a way that the offspring distribution is no longer constant, but now depends on the population density \(x\) with respect to a threshold \(K > 0\). Denote by \(Z^K_n\) the population size of generation \(n\) when the threshold is \(K\), and by \(X^K_n = Z^K_n/K\) the corresponding density. The behavior of this density process is investigated as \(K \to \infty\) under some regularity conditions on \(m(x)\) and \(\sigma^2 (x)\), the mean and variance of the offspring distribution for population density \(x\), and if the third moments are bounded. \(X^K_\bullet\) admits a representation which is in a certain sense a stochastic analogon to the logistic or other deterministic models with limited growth having the form \(x_{n + 1} = f(x_n)\). As \(K \to \infty\), \(X^K_\bullet\) converges in distribution to such a deterministic process with \(f(x) = xm(x)\). Denote by \(Y^K_n = (X^K_n - x_n) \sqrt K\) the rescaled deviation from the deterministic limit where \(X^K_0 \to x\) and \(x_n = f_n(x)\) \((f_n\) is the \(n\)-th iterate of \(f)\). \(Y^K_\bullet\) now converges weakly to a Gaussian process where the mean and covariance function can be explicitly represented by \(m(x)\) and \(\sigma^2 (x)\). In the last two theorems some more details of the behavior of \(Y^K_\bullet\) are given if \(f\) has a stable \(d\)-cycle.