Epidemic Spread in Human Networks

One of the popular dynamics on complex networks is the epidemic spreading. An epidemic model describes how infections spread throughout a network. Among the compartmental models used to describe epidemics, the Susceptible-Infected-Susceptible (SIS) model has been widely used. In the SIS model, each node can be susceptible, become infected with a given infection rate, and become again susceptible with a given curing rate. In this paper, we add a new compartment to the classic SIS model to account for human response to epidemic spread. Each individual can be infected, susceptible, or alert. Susceptible individuals can become alert with an alerting rate if infected individuals exist in their neighborhood. An individual in the alert state is less probable to become infected than an individual in the susceptible state; due to a newly adopted cautious behavior. The problem is formulated as a continuous-time Markov process on a general static graph and then modeled into a set of ordinary differential equations using mean field approximation method and the corresponding Kolmogorov forward equations. The model is then studied using results from algebraic graph theory and center manifold theorem. We analytically show that our model exhibits two distinct thresholds in the dynamics of epidemic spread. Below the first threshold, infection dies out exponentially. Beyond the second threshold, infection persists in the steady state. Between the two thresholds, the infection spreads at the first stage but then dies out asymptotically as the result of increased alertness in the network. Finally, simulations are provided to support our findings. Our results suggest that alertness can be considered as a strategy of controlling the epidemics which propose multiple potential areas of applications, from infectious diseases mitigations to malware impact reduction.

💡 Research Summary

The paper introduces a novel extension to the classic Susceptible‑Infected‑Susceptible (SIS) epidemic model by adding an “Alert” compartment, resulting in a Susceptible‑Alert‑Infected‑Susceptible (SAIS) framework. In this model, a susceptible node that has infected neighbors can transition to an alert state at a rate κ proportional to the number of infected neighbors. While in the alert state, the node’s infection rate is reduced to βₐ (with 0 ≤ βₐ < β₀), reflecting cautious behavior such as wearing masks or reducing contacts. Infected nodes recover to the susceptible state at rate δ, and the model assumes no direct transition from alert back to susceptible, making alert a semi‑absorbing state.

The dynamics are first formulated as a continuous‑time Markov process on a static, possibly directed, graph G = (V,E) with adjacency matrix A. By applying the Kolmogorov forward equations and a mean‑field approximation (replacing the exact neighbor infection indicator with its expectation), the authors reduce the full 3N‑dimensional Markov description to a deterministic system of 2N ordinary differential equations for the infection probability pᵢ(t) and alert probability qᵢ(t) of each node i:  ˙pᵢ = β₀(1 − pᵢ − qᵢ)∑ⱼaᵢⱼpⱼ + βₐqᵢ∑ⱼaᵢⱼpⱼ − δpᵢ,  ˙qᵢ = κ(1 − pᵢ − qᵢ)∑ⱼaᵢⱼpⱼ − βₐqᵢ∑ⱼaᵢⱼpⱼ. The conserved quantity sᵢ + pᵢ + qᵢ = 1 (where sᵢ is the susceptible probability) allows elimination of sᵢ, yielding the compact 2N‑dimensional system.

The analytical core of the paper is the identification of two distinct epidemic thresholds, τ₁ᶜ and τ₂ᶜ, expressed in terms of the infection strength τ = β₀/δ, the spectral radius ρ(A) of the adjacency matrix, and the alert parameters κ and βₐ. The first threshold τ₁ᶜ = 1/ρ(A) coincides with the classic SIS threshold: if τ < τ₁ᶜ, all eigenvalues of the linearized system have negative real parts, guaranteeing exponential decay of infection. The second threshold τ₂ᶜ emerges from a nonlinear stability analysis using the center‑manifold theorem. When τ > τ₂ᶜ, a non‑trivial steady state exists where infection persists. For τ₁ᶜ < τ < τ₂ᶜ, the system initially experiences a surge in infections, but the alert compartment grows rapidly, effectively reducing the infection term (β₀ − βₐ)qᵢ∑ⱼaᵢⱼpⱼ. This feedback drives the infection probabilities down, leading to asymptotic extinction despite the intermediate growth phase.

The authors prove a comparison theorem (Theorem 1) stating that, given identical initial conditions, the infection probabilities in the SAIS model are always bounded above by those of the standard SIS model. The proof relies on the fact that the additional negative term (β₀ − βₐ)qᵢ∑ⱼaᵢⱼpⱼ in the SAIS infection dynamics reduces the instantaneous growth rate relative to SIS.

Numerical simulations on Erdős‑Rényi random graphs, scale‑free networks, and a real‑world social contact network corroborate the theoretical findings. By varying κ and βₐ, the authors demonstrate how the second threshold shifts: higher alerting rates (larger κ) and lower alert infection rates (smaller βₐ) lower τ₂ᶜ, expanding the regime where alertness can suppress an outbreak. The simulations also illustrate the three distinct regimes: (i) rapid die‑out for τ < τ₁ᶜ, (ii) transient outbreak followed by extinction for τ₁ᶜ < τ < τ₂ᶜ, and (iii) endemic persistence for τ > τ₂ᶜ.

The paper discusses limitations, notably the reliance on mean‑field approximations (which provide an upper bound but may miss stochastic fluctuations), the omission of a direct A → S transition, and the assumption of a static contact graph. Future work is suggested to incorporate dynamic networks, heterogeneous alerting behaviors, and data‑driven parameter estimation.

Overall, this work offers a rigorous, graph‑theoretic treatment of human behavioral response in epidemic spreading, revealing that alertness can create a protective feedback loop capable of generating a second, higher epidemic threshold. The findings have practical implications for public‑health interventions (e.g., awareness campaigns, mask mandates) and for cybersecurity (e.g., malware alert systems), where inducing a state of heightened vigilance can substantially mitigate propagation.