Epidemic threshold for the SIS model on networks

We derive an analytical expression for the critical infection rate r_c of the susceptible-infectious-susceptible (SIS) disease spreading model on random networks. To obtain r_c, we first calculate the probability of reinfection, pi, defined as the probability of a node to reinfect the node that had earlier infected it. We then derive r_c from pi using percolation theory. We show that pi is governed by two effects: (i) The requirement from an infecting node to recover prior to its reinfection, which depends on the disease spreading parameters; and (ii) The competition between nodes that simultaneously try to reinfect the same ancestor, which depends on the network topology.

💡 Research Summary

The paper presents a novel analytical framework for determining the epidemic threshold (critical infection rate r_c) of the susceptible‑infectious‑susceptible (SIS) model on random networks. Traditional mean‑field approaches estimate r_c simply as μ/⟨k⟩, where μ is the recovery rate and ⟨k⟩ the average degree, but they ignore two crucial aspects: (i) the temporal requirement that an infected node must recover before it can reinfect its infector, and (ii) the structural competition among multiple descendants that simultaneously attempt to reinfect the same ancestor. To capture these effects, the authors introduce the “reinfection probability” π, defined as the probability that a node successfully reinfects the node that originally infected it.

π is decomposed into two multiplicative factors. The first factor, r/(r + μ), reflects the disease dynamics: an infected node can only transmit back after it has recovered, and the likelihood of this sequence depends on the infection rate r and recovery rate μ. The second factor, ⟨1/k⟩_excess, encodes the network topology. It is the average of the reciprocal of the excess degree (the degree of a node reached by following a random edge) and quantifies how many competing descendants are trying to reinfect the same ancestor. In highly heterogeneous networks, especially those with hub nodes, ⟨1/k⟩_excess becomes small, dramatically reducing π. Consequently, the authors approximate

 π ≈ (r/(r + μ)) · ⟨1/k⟩_excess.

Having obtained π, the SIS spreading process is mapped onto a dynamic percolation problem. In percolation theory, an epidemic persists indefinitely if the branching factor of the infection tree exceeds unity. The branching factor for SIS on a network can be expressed as ⟨k⟩·π·(r/μ). Setting this equal to one yields the critical condition, which can be rearranged to give a closed‑form expression for the epidemic threshold:

 r_c = μ / (⟨k⟩·π − 1).

This formula reduces to the classic mean‑field result when π ≈ 1 (i.e., when the network is homogeneous and the disease dynamics are fast relative to recovery). However, for heterogeneous networks where π < 1, the threshold is higher, indicating that stronger transmissibility is required for sustained spread.

The theoretical predictions are validated through extensive Monte‑Carlo simulations on two canonical random graph ensembles: Erdős‑Rényi (ER) graphs and scale‑free networks with power‑law degree distributions. In ER graphs, the degree distribution is narrow, ⟨1/k⟩_excess is close to 1/⟨k⟩, and the measured r_c aligns closely with the mean‑field estimate. In contrast, scale‑free networks with degree exponent γ ≤ 3 exhibit diverging second moments ⟨k²⟩, which would make the mean‑field threshold vanish (r_c → 0). The authors’ π‑based formula, however, predicts a finite r_c because the competition among many low‑degree descendants of a hub suppresses reinfection. Simulation results confirm that a non‑zero threshold persists, thereby reconciling the apparent discrepancy between earlier theoretical claims and empirical observations.

The paper also discusses practical implications. Since π decreases with increasing heterogeneity, targeted interventions that immunize or isolate high‑degree nodes (hubs) effectively raise π for the remaining network, thereby lowering the required r_c for epidemic control. Moreover, the explicit dependence of π on disease parameters (r, μ) highlights that shortening the infectious period (increasing μ) raises the threshold, providing a quantitative justification for treatment strategies that accelerate recovery.

Limitations are acknowledged: the analysis assumes static, uncorrelated random networks and does not incorporate clustering, degree‑degree correlations, or temporal changes in contact patterns. Extending the framework to include these features, as well as multi‑stage infection processes, constitutes a promising direction for future work. Nonetheless, by introducing the reinfection probability as a bridge between disease dynamics and network topology, and by leveraging percolation theory to obtain an exact expression for r_c, the study offers a significant conceptual advance in the understanding of SIS epidemics on complex networks.