Extinction of metastable stochastic populations

TL;DR

This paper develops a WKB-based theoretical framework to analyze the extinction times of metastable stochastic populations, considering different scenarios depending on the deterministic fixed points and incorporating various approximation methods.

Contribution

It introduces a comprehensive WKB approach for calculating extinction probabilities and times in complex stochastic population models, including multi-step processes and bifurcations.

Findings

Derived quasi-stationary distributions for different extinction scenarios

Calculated exponential extinction times and entropic barriers

Validated methods near bifurcation points and for single-step processes

Abstract

We investigate extinction of a long-lived self-regulating stochastic population, caused by intrinsic (demographic) noise. Extinction typically occurs via one of two scenarios depending on whether the absorbing state n=0 is a repelling (scenario A) or attracting (scenario B) point of the deterministic rate equation. In scenario A the metastable stochastic population resides in the vicinity of an attracting fixed point next to the repelling point n=0. In scenario B there is an intermediate repelling point n=n_1 between the attracting point n=0 and another attracting point n=n_2 in the vicinity of which the metastable population resides. The crux of the theory is WKB method which assumes that the typical population size in the metastable state is large. Starting from the master equation, we calculate the quasi-stationary probability distribution of the population sizes and the (exponentially long) mean time to extinction for each of the two scenarios. When necessary, the WKB approximation is complemented (i) by a recursive solution of the quasi-stationary master equation at small n and (ii) by the van Kampen system-size expansion, valid near the fixed points of the deterministic rate equation. The theory yields both entropic barriers to extinction and pre-exponential factors, and holds for a general set of multi-step processes when detailed balance is broken. The results simplify considerably for single-step processes and near the characteristic bifurcations of scenarios A and B.