The effects of initial conditions on the accuracy of mean-field approximations of Markov processes on large random graphs

Abstract

We study the evolution of a general class of stochastic processes (containing, e.g. SIS and SIR models) on large random networks, focusing on a particular general class of random graph models. To approximate the expected dynamics of these processes, we employ a mean-field technique known as the N-Intertwined Mean-Field Approximation (NIMFA). Our primary goal is to quantify the impact of the randomness in the network topology on the accuracy of such approximations. While classical work by Kurtz established strong approximation results for density-dependent Markov chains using mean-field on the complete-graph of order $N$, yielding error bounds of order $\frac{1}{\sqrt{N}}$, we demonstrate that in the context of NIMFA on random graphs, the error admits a refined characterization depending on initial conditions. Specifically, for generic initial conditions, the worst case error is of order $\frac{1}{\sqrt{d}}$, where $d$ is the expected average degree of the graph. This result highlights how network sparsity contributes an additional source of variability not captured by classical mean-field approaches. Furthermore, when the initial conditions are taken to be fairly homogeneous, we show that the error term is of order $\frac{1}{\sqrt{N}}+\frac{1}{d}$, where $N$ is the number of nodes, underscoring the intricate role of the choice of initial conditions in the error bounds. Our analysis connects with recent interest in quantifying the role of graph heterogeneity in epidemic thresholds and dynamics.