Source-linked AI summary
On the Dynamics of Deterministic Epidemic Propagation over Networks
Wenjun Mei, Shadi Mohagheghi, Sandro Zampieri, Francesco Bullo
TL;DR
The paper addresses how deterministic epidemic dynamics can be analyzed when propagation occurs over structured contact networks rather than well-mixed populations. It reviews and analyzes network SI, SIS, and SIR models, introducing algorithms and results for equilibria, thresholds, stability, convergence, and transient behavior. Its central conclusion is that these network results generalize established scalar-model properties while adding network-specific characterization and computation.
Problem
Scalar models overlook contact-network structure and heterogeneity, while network propagation raises questions about model analysis and transient and asymptotic dynamics.
Method
The paper provides a comprehensive nonlinear treatment of deterministic SI, SIS, and SIR models over strongly connected contact networks, combining proofs, spectral analysis, and iterative algorithms.
Results
The paper establishes network SI, SIS, and SIR results covering thresholds, equilibria, stability, positivity, convergence, transient behavior, endemic-state computation, and asymptotic-state computation.
Takeaways & Limitations
Network epidemic models provide generalizations of scalar SI, SIS, and SIR properties while capturing contact-network effects and supporting computation of endemic or asymptotic states.
Takeaways & Limitations
The framework assumes propagation over strongly connected contact networks and interprets node states through probabilities or population fractions.
Abstract
from arXiv · showhide
In this work we review a class of deterministic nonlinear models for the propagation of infectious diseases over contact networks with strongly-connected topologies. We consider network models for susceptible-infected (SI), susceptible-infected-susceptible (SIS), and susceptible-infected-recovered (SIR) settings. In each setting, we provide a comprehensive nonlinear analysis of equilibria, stability properties, convergence, monotonicity, positivity, and threshold conditions. For the network SI setting, specific contributions include establishing its equilibria, stability, and positivity properties. For the network SIS setting, we review a well-known deterministic model, provide novel results on the computation and characterization of the endemic state (when the system is above the epidemic threshold), and present alternative proofs for some of its properties. Finally, for the network SIR setting, we propose novel results for transient behavior, threshold conditions, stability properties, and asymptotic convergence. These results are analogous to those well-known for the scalar case. In addition, we provide a novel iterative algorithm to compute the asymptotic state of the network SIR system.
1 Introduction
The paper reviews deterministic SI, SIS, and SIR epidemic models over strongly connected contact networks, addressing network structure, heterogeneity, thresholds, and dynamic behavior. It contributes analyses and algorithms that extend scalar epidemic-model results to network settings.
- Motivation: Scalar propagation models overlook contact-network structure and individual heterogeneity, despite both shaping propagation dynamics.
- Scope and approach: The paper focuses on deterministic mean-field network models for SI, SIS, and SIR epidemics and characterizes their dynamical properties.
- Overall contribution: The network results are presented as generalizations of corresponding scalar-model properties.
- Network SI: The network SI model has no epidemic threshold, and all trajectories converge to the full-contagion state.
- Network SIS: For network SIS systems above threshold, the paper gives a provably correct iterative algorithm for the endemic state and derives expansions near threshold and at high infection rates.
- Network SIR: For network SIR systems, the paper develops transient and threshold results, proves asymptotic infection vanishing, and supplies an iterative algorithm for the asymptotic state.
2 Model Set-Up and Notations
The models describe epidemic propagation on strongly connected weighted directed contact networks, with node-level state probabilities and standard vector and spectral notation. Nodes may represent individuals or homogeneous populations.
- Contact Network: Epidemics propagate over a strongly connected weighted digraph whose adjacency entries encode directed contact strengths.
- Node States and Probabilities: SI and SIS nodes are susceptible or infected, while SIR nodes additionally have a recovered state.
- Node States and Probabilities: State variables represent probabilities for individuals or population fractions when nodes represent homogeneous populations.
- Frequently Used Notations: The notation defines nonnegative vectors, componentwise inequalities, all-ones and zero vectors, diagonal matrices, and Perron-Frobenius spectral quantities.
3 Susceptible-Infected Model
The SI model describes infection spreading through a strongly connected network: infection probabilities remain valid and increase, while every nonzero initial infection converges to full contagion. Its early and late dynamics are governed by different network centralities.
- 3.1 Scalar SI model: The scalar SI model produces monotonically increasing infection fractions that converge to the unique equilibrium 1.Its trajectories have a logistic-like evolution.
- 3.2 Network SI model: The network SI model has exactly two equilibria: 0^n, representing no epidemic, and 1^n, representing full contagion.The equilibrium characterization follows from the strongly connected network structure.
- 3.2 Network SI model: The zero equilibrium has linearization x_dot = βAx and is exponentially unstable, whereas the full-contagion equilibrium is approached by every trajectory with nonzero initial infection.The convergence result identifies 1^n as the asymptotic state for all nonzero initial conditions.
- 3.2 Network SI model: During initial growth, infection increases exponentially at rate βλmax and follows the dominant eigenvector; near full contagion, susceptible fractions decay at rates proportional to node degrees.Thus initial infection rate is associated with eigenvector centrality, while final infection speed is associated with degree centrality.
4 Susceptible-Infected-Susceptible model
The SIS analysis characterizes scalar and network epidemic behavior through equilibria, stability, positivity, convergence, and threshold conditions. For strongly connected networks, the epidemic threshold separates extinction from a unique endemic state with global convergence properties.
- Scalar SIS model: The scalar SIS model has infection rate β, recovery rate γ, and infected and susceptible fractions satisfying x(t) + s(t) = 1.The model assumes infected individuals recover to the susceptible state at rate γ.
- Scalar SIS model: If β ≤ γ, every scalar SIS trajectory converges to the unique equilibrium x = 0, so the epidemic disappears.If β > γ, every trajectory with x(0) > 0 converges to an exponentially stable positive equilibrium representing persistent contagion.
- Behavior below the threshold: Below the threshold, the zero infection equilibrium is unique, trajectories remain in [0, 1]n, and all trajectories converge to zero.The linearized system is exponentially stable when βλmax − γ < 0.
- Network SIS model: For a strongly connected network, the epidemic threshold is βλmax/γ = 1, where λmax is the dominant eigenvalue of the adjacency matrix.The network basic reproduction number is R0 = βλmax/γ.
- Behavior above the threshold: The paper adds a provably correct iterative algorithm for computing the endemic state, novel Taylor expansions, and alternative proofs based on the map F+.The monotone iteration can increase or decrease depending on whether its dominant-eigenvector initialization lies below or above the stated boundary.
5 Network Susceptible-Infected-Recovered Model
The network SIR model preserves positivity and population normalization, while its infected population ultimately vanishes. Threshold behavior is governed by the spectral radius of diag(s(t))A, and the asymptotic susceptible state is characterized as a unique fixed point computable by iteration.
- Asymptotic state and algorithm: The asymptotic susceptible state is the unique fixed point of H, and iterating y(k + 1) = H(y(k)) converges to it.The fixed point lies in {s ∈ R^n | 0_n ≤ s ≤ 1_n − r(0)}.
- Threshold behavior: If βλmax(τ) < γ, the dominant-eigenvector-weighted infected average decreases monotonically and exponentially after time τ.This provides the below-threshold decay condition for the network model.
- Threshold behavior: If βλmax(0) > γ, the dominant-eigenvector-weighted infected average grows exponentially for small time, indicating an epidemic outbreak.The initial growth measure uses vmax(0)⊤x(t).
- Threshold behavior: The effective reproduction number is R(t) = βλmax(t)/γ; it eventually falls below one, after which infection decreases exponentially to zero.The model guarantees a finite time after which the below-threshold condition holds.
- Asymptotic state and algorithm: Along every trajectory, conserved functions Vi constrain the dynamics to their level curves, supporting characterization of the limiting state.At the limit, x(∞) = 0_n and r(∞) = 1_n − s(∞).
- Asymptotic state and algorithm: The iterative fixed-point scheme converges from any initial y(0) satisfying 0_n ≤ y(0) ≤ 1_n − r(0), under strong connectivity.Monotonicity and boundedness establish convergence, while comparison yields uniqueness.
- Numerical illustration: Numerical trajectories on an undirected graph show a unimodal time evolution for the average infected fraction, as in the scalar model.The experiment starts above the effective reproduction threshold with R(0) = 3.57.
6 Conclusion
The paper consolidates deterministic network SI, SIS, and SIR analyses, covering epidemic asymptotics and transient behavior. For network SIR, it adds threshold and convergence results alongside an iterative method for computing the asymptotic state.
- The paper consistently analyzes deterministic nonlinear SI, SIS, and SIR propagation models over contact networks.The analysis addresses asymptotic behaviors including vanishing infection, steady-state epidemic, and full contagion.
- Network SIR results include new transient behavior, threshold conditions, system properties, and asymptotic convergence.
- An iterative algorithm computes the asymptotic state of the network SIR system.