Wright--Fisher kernels: from linear to non-linear dynamics, ergodicity and McKean--Vlasov scaling limits

We consider a population of hosts infected by a pathogen that exists in two strains. We use a two-parameter family of Markov kernels on $[0,1]$ to describe the discrete-time evolution of the pathogen type-composition within and across individuals. First, we assume that there is no interaction between pathogen populations across host individuals. For a particular class of parameters, we establish moment duality between the type-frequency process and a process reminiscent of the \emph{Ancestral Selection Graph}. We also show convergence, under appropriate scaling of parameters and time, to a Wright-Fisher diffusion with drift. Next, we assume that pathogen-type compositions are correlated across hosts by their empirical measure. We show a propagation of chaos result comparing the pathogen type-composition of a given host with the evolution of a non-linear chain. Furthermore, we show that under appropriate scaling, the non-linear chain converges to a McKean-Vlasov diffusion. To illustrate our results, we consider a population affected by mutation rates that depend on the instantaneous distribution across multiple hosts. For this example, we study the uniform-in-time propagation of chaos and the ergodicity of the limiting McKean-Vlasov SDE.
Comment: 43 pages, 2 figures