Simulation algorithms for Markovian and non-Markovian epidemics

TL;DR

This paper compares two algorithms for simulating non-Markovian epidemics, proving their statistical equivalence and analyzing their performance on complex network models to guide researchers in selecting efficient simulation methods.

Contribution

It establishes the statistical equivalence of the Next Reaction Method and the non-Markovian Gillespie algorithm and evaluates their performance on complex epidemic models.

Findings

Both algorithms are statistically equivalent.

Next Reaction Method performs well on time-varying networks.

Next Reaction Method is a viable alternative to Gillespie.

Abstract

Researchers have employed stochastic simulations to determine the validity of their theoretical findings and to study analytically intractable spreading dynamics. In both cases, the correctness and efficiency of the simulation algorithm are of paramount importance. We prove in this article that the Next Reaction Method and the non-Markovian Gillespie algorithm, two algorithms for simulating non-Markovian epidemics, are statistically equivalent. We also study the performance and applicability under various circumstances through complexity analyses and numerical experiments. In our numerical simulations, we apply the Next Reaction Method and the Gillespie algorithm to epidemic simulations on time-varying networks and epidemic simulations with cooperative infections. Both tasks have only been done using the Gillespie algorithm, while we show that the Next Reaction Method is a good alternative. We believe this article may also serve as a guide for choosing simulation algorithms that are both correct and efficient for researchers from epidemiology and beyond.