Gillespie algorithms for stochastic multiagent dynamics in populations and network

TL;DR

This paper provides a tutorial on Gillespie algorithms for simulating stochastic multiagent dynamics in populations and networks, emphasizing their advantages over discrete-time models and reviewing recent extensions for increased realism and efficiency.

Contribution

It offers an accessible tutorial on Gillespie algorithms for social dynamics and reviews recent advances extending their applicability and performance.

Findings

Clarifies the benefits of continuous-time models over discrete-time models.

Reviews recent extensions for non-Poissonian dynamics.

Highlights methods to improve simulation speed and realism.

Abstract

Many multiagent dynamics, including various collective dynamics occurring on networks, can be modeled as a stochastic process in which the agents in the system change their state over time in interaction with each other. The Gillespie algorithms are popular algorithms that exactly simulate such stochastic multiagent dynamics when each state change is driven by a discrete event, the dynamics is defined in continuous time, and the stochastic law of event occurrence is governed by independent Poisson processes. In the first main part of this volume, we provide a tutorial on the Gillespie algorithms focusing on simulation of social multiagent dynamics occurring in populations and networks. We do not assume advanced knowledge of mathematics (or computer science or physics). We clarify why one should use the continuous-time models and the Gillespie algorithms in many cases, instead of easier-to-understand discrete-time models. In the remainder of this volume, we review recent extensions of the Gillespie algorithms aiming to add more reality to the model (i.e., non-Poissonian cases) or to speed up the simulations.