Absence of epidemic thresholds in a growing adaptive network

The structure of social contact networks strongly influences the dynamics of epidemic diseases. In particular the scale-free structure of real-world social networks allows unlikely diseases with low infection rates to spread and become endemic. However, in particular for potentially fatal diseases, also the impact of the disease on the social structure cannot be neglected, leading to a complex interplay. Here, we consider the growth of a network by preferential attachment from which nodes are simultaneously removed due to an SIR epidemic. We show that increased infectiousness increases the prevalence of the disease and simultaneously causes a transition from scale-free to exponential topology. Although a transition to a degree distribution with finite variance takes place, the network still exhibits no epidemic threshold in the thermodynamic limit. We illustrate these results using agent-based simulations and analytically tractable approximation schemes.

💡 Research Summary

The paper investigates the interplay between network growth and epidemic dynamics by coupling a preferential‑attachment growth process with an SIR disease that removes infected nodes from the network. In the classic Barabási‑Albert model, new nodes attach preferentially to high‑degree vertices, generating a scale‑free degree distribution with a divergent second moment. The authors augment this model with a continuous‑time SIR process: infected nodes transmit the disease to susceptible neighbors with rate β and, upon recovery or death, are permanently removed together with all incident edges. The network therefore evolves adaptively: as the disease spreads, high‑degree “hubs” are eliminated, potentially reshaping the topology.

Two control parameters govern the system: the attachment rate λ (new nodes per unit time) and the infection rate β. By varying β while keeping λ fixed, the authors observe a transition in the degree distribution. For low β the network retains its scale‑free character; for higher β the removal of infected hubs truncates the tail, yielding an exponential (or stretched‑exponential) distribution with finite mean and variance. This structural transition is quantified analytically using mean‑field rate equations for the fraction of infected nodes of degree k, and numerically via dynamic message‑passing (DMP) that tracks the probability of transmission along each edge while accounting for node deletion.

A central finding is that, despite the emergence of a finite‑variance degree distribution, the epidemic threshold β_c disappears in the thermodynamic limit. The authors prove that as the network size N → ∞, the continual influx of new nodes supplies a persistent reservoir of susceptible vertices. Even when β is arbitrarily small, the product of the growth rate and the infection probability ensures a non‑zero steady‑state prevalence. In other words, the basic reproduction number R₀ can be less than one in a static snapshot, yet the dynamic balance between growth and removal sustains endemic infection. This result contrasts with the traditional explanation for threshold‑free spreading on static scale‑free networks (which relies on an infinite variance of the degree distribution) and demonstrates a distinct mechanism rooted in adaptive topology.

Methodologically, the study proceeds in three stages. First, a mean‑field approximation yields differential equations for the degree‑resolved infected fraction i_k(t) and the overall infected density I(t). These equations reveal how the removal term scales with the degree distribution and how the growth term replenishes high‑degree nodes. Second, the DMP framework incorporates correlations introduced by edge‑wise transmission probabilities and node deletions, providing a more accurate description of the stochastic process on finite networks. Third, large‑scale agent‑based simulations (up to one million nodes) validate the analytical predictions. The simulations confirm the degree‑distribution transition, the absence of a finite epidemic threshold, and the agreement between mean‑field and DMP results across a broad parameter range.

The paper contributes several novel insights. It introduces the first unified model that simultaneously captures preferential attachment growth and epidemic‑induced node removal, highlighting how disease dynamics can reshape network topology in real time. It identifies a counter‑intuitive regime where the network becomes “well‑behaved” (exponential degree distribution) yet remains vulnerable to persistent infection because growth continuously feeds the epidemic. The analytical treatment combines classic mean‑field theory with modern DMP techniques, offering a versatile toolkit for studying other adaptive network processes. Finally, the work carries practical implications: public‑health strategies that rely solely on reducing transmission below a static threshold may be insufficient in growing populations. Interventions that limit network expansion (e.g., controlling migration or contact formation) or protect high‑degree individuals could be essential to break the feedback loop that sustains disease.