The accuracy of Bartlett's small-fluctuation approximation for stochastic-difference-equation population models (Q1079529)

In a variety of stochastic-difference equation population models, a method to approximate the stationary moments based on the assumption of ''small'' fluctuations has been successfully used without analytic validation. It consists ''of approximating the dynamics by a truncated Taylor series about a stable equilibrium of the corresponding deterministic model, and manipulating the approximate dynamics to obtain sets of equations that can be solved for the moments.'' It was first used by \textit{M. S. Bartlett} [Proc. third Berkeley Sympos. math. Statist. and Probability 4, 81-109 (1956; Zbl 0070.150), and Biometrika 44, 27-42 (1956; Zbl 0080.363)]. This paper shows that Bartlett's approximation is valid for models of the form \[ (*)\quad X_ r(t+1)=F(X_ r(t),r\alpha_ r(t)),\quad t=0,1,2,..., \] where \(X_ r(t)\in {\mathbb{R}}^ n\) is population size, \(\alpha_ r(t)\in {\mathbb{R}}^ m\) is the state of the random environment at time t (assumed to have bounded moments uniformly on r), \(r>0\) is a real parameter (small r meaning small noise), \(X_ r(0)\) is given, the distribution of the process \(X_ r(t)\), \(t\geq 0\), is assumed stationary and other suitable general assumptions are made. It is shown, using Lyapunov functions, that the error in Bartlett's approximation for the stationary moments of (*) is of order \(r^ 3\) as \(r\to 0\). The problem of existence of stationary solutions of (*) is adressed and, if \(\alpha_ r(t)\) is Markovian and satisfies some assumptions, such an existence (but not necessarily uniqueness) can be shown.