Contact processes and moment closure on adaptive networks

Contact processes describe the transmission of distinct properties of nodes via the links of a network. They provide a simple framework for many phenomena, such as epidemic spreading and opinion formation. Combining contact processes with rules for topological evolution yields an adaptive network in which the states of the nodes can interact dynamically with the topological degrees of freedom. By moment-closure approximation it is possible to derive low-dimensional systems of ordinary differential equations that describe the dynamics of the adaptive network on a coarse-grained level. In this chapter we discuss the approximation technique itself as well as its applications to adaptive networks. Thus, it can serve both as a tutorial as well as a review of recent results.

💡 Research Summary

The chapter provides a comprehensive treatment of contact processes on adaptive networks, focusing on how the interplay between node states and network topology can be captured by low‑dimensional ordinary differential equations (ODEs) through moment‑closure approximations. It begins by defining a contact process as a stochastic rule set in which each node occupies one of a small number of discrete states (e.g., susceptible/infected, opinion A/opinion B) and may change state by “contact” with neighboring nodes. In static networks, only the node states evolve, while the edge set remains fixed. Real‑world systems, however, often exhibit simultaneous evolution of both states and connections: individuals may break ties with disagreeing neighbors and form new links with like‑minded ones, a phenomenon captured by adaptive network models.

To formalize this, the authors write a master equation for the joint probability distribution over node states and edge types (e.g., SS, SI, II for a two‑state model). The state space grows combinatorially with network size, rendering direct analytical treatment infeasible. The key methodological contribution is the systematic application of moment‑closure techniques. The authors select the first‑order moments (node‑state densities) and second‑order moments (pair densities) as the primary variables. Higher‑order moments (triplets, quadruplets, etc.) are approximated as functions of these lower‑order moments, thereby closing the hierarchy. Two principal closure schemes are examined: the classic pair approximation, which assumes that the probability of a triplet factorizes into the product of two overlapping pairs, and a normalized‑cluster approximation that enforces consistency with the network’s average degree.

With the closed ODE system in hand, the authors explore a range of dynamical regimes by varying two key parameters: the infection (or opinion‑conversion) rate β and the rewiring probability p. When p is low, the system behaves like a standard SIS model on a static graph, converging to an endemic equilibrium if β exceeds a critical threshold βc. As p increases past a second threshold, the network self‑organizes into a bipartite structure that isolates infected nodes from susceptibles, dramatically reducing prevalence and leading to a disease‑free equilibrium. Between these two thresholds, the model exhibits bistability: both endemic and disease‑free fixed points coexist, and the ultimate outcome depends on initial conditions. Moreover, the nature of the rewiring rule (assortative versus disassortative) determines whether the system settles into a steady state, displays sustained oscillations, or even chaotic fluctuations. These qualitative behaviors are captured quantitatively by the ODEs, which predict the locations of bifurcations and the size of basins of attraction.

To validate the analytical predictions, extensive Gillespie‑algorithm simulations of the full agent‑based model are performed. The simulation results align closely with the ODE solutions across most of the parameter space, confirming that the second‑order closure captures the essential macroscopic dynamics. Small discrepancies appear only in regimes of extreme rewiring rates or highly heterogeneous degree distributions, where third‑order correlations become non‑negligible. Nonetheless, the overall agreement demonstrates that moment‑closure provides a reliable coarse‑graining tool for adaptive networks.

The chapter concludes by outlining extensions and open challenges. The same closure framework can be applied to multi‑state epidemic models (SIR, SEIR), multilayer or multiplex networks, and systems with explicit time delays. Higher‑order closures, data‑driven inference of closure functions, and hybrid approaches that combine analytical ODEs with machine‑learning surrogates are suggested as promising avenues for future research. By bridging microscopic interaction rules with macroscopic dynamical equations, the moment‑closure methodology offers a powerful, scalable approach for analyzing and controlling complex adaptive networks in epidemiology, sociology, and beyond.