Stability of Epidemic Models over Directed Graphs: A Positive Systems Approach

1 Introduction

• The basic reproduction number separates disease-free equilibrium for R0 ≤ 1 from endemic equilibrium for R0 > 1.The paper uses positive-systems theory to characterize stability on directed topologies.Source: §1 · p. 1
• Earlier stability results focused on strongly connected graphs, while weakly connected digraphs common in practice raise new equilibrium and stability challenges.The paper notes that prior endemic-state results relied on strict-positivity assumptions or advanced graph-theoretic tools.Source: Literature review · p. 2

Statement of Contributions

• For strongly connected digraphs, positive-systems methods establish global stability of the endemic state without assumptions on state evolution, using a quadratic Lyapunov function.This construction avoids advanced combinatorial and graph-theoretic notions used by an earlier approach.Source: Statement of Contributions · p. 2
• For weakly connected digraphs, the paper studies disease-free and endemic equilibria and gives conditions for endemic-state existence, uniqueness, and stability.Weak-graph endemic states can leave some nodes healthy while others are infected.Source: Statement of Contributions · p. 2
• The game-theoretic framework treats the n-intertwined Markov model as best-response dynamics in a concave game and yields a distributed-checkable condition for disease-free stability.The framework can prescribe more general classes of infection dynamics.Source: Statement of Contributions · p. 2

2 Mathematical Preliminaries

• The paper sets conventions for vector ordering, norms, eigenvalues, spectral radius, and matrix notation, including two uses of diag(.).It also introduces the Jacobian and Lyapunov-function concepts used for dynamical systems.Source: §2 · p. 2

Matrix Theory

• For Metzler matrices, Hurwitz stability is equivalent to positive-vector inequalities and to diagonal stability.The section also defines similar, reducible, and irreducible matrices and identifies Metzler matrices by their nonnegative off-diagonal entries.Source: Matrix Theory · p. 3
• Perron–Frobenius results characterize dominant eigenvalues and positive eigenvectors for irreducible Metzler and nonnegative matrices.The dominant eigenvalue is algebraically simple, and its corresponding eigenvector is strictly positive.Source: Matrix Theory · p. 3

Graph Theory

• The graph definitions distinguish strong connectivity, weak connectivity, and strongly connected components, and describe source nodes in directed acyclic graphs.Weak connectivity means the graph becomes connected when its directed edges are treated as undirected.Source: Graph Theory · p. 3

3 The n-Intertwined Markov Model

• The n-intertwined Markov model approximates the influence of neighbors on each node’s infection probability using a mean-field dynamical system.Each node has curing and infection rates, and the state variables are infection probabilities.Source: §3 · pp. 3–4

3.1 Equilibrium States of the n-Intertwined Markov Model

• The basic reproduction number characterizes the model’s equilibria; for connected undirected graphs, R0 ≤ 1 gives a unique disease-free equilibrium, while R0 > 1 also yields a strong endemic equilibrium.A strong endemic state is strictly positive.Source: §3.1 · p. 4

3.2 The n-Intertwined Markov Model as a Concave Game

• The game-theoretic formulation is presented as a basis for broader infection dynamics and a decision-based interpretation of virus spread.The paper also notes potential application to diffusion phenomena such as spam spread in computer networks.Source: §3.2 · p. 4
• The model can be represented as best-response dynamics of a concave game in which each node chooses its infection probability as its decision variable.Each node maximizes an objective that depends on its own choice and other players’ actions.Source: §3.2 · pp. 4–5
• The game’s objective functions are strictly concave in each node’s own decision, and the model dynamics correspond to gradient-flow dynamics for agents maximizing those objectives.The cited proposition gives a diagonal-dominance condition for a unique pure-strategy Nash equilibrium.Source: §3.2 · pp. 4–5

4.1 Stability of the Disease-Free State

• On strongly connected digraphs, the disease-free equilibrium is globally asymptotically stable if and only if R0 ≤ 1.For R0 < 1, global stability and, for R0 > 1, instability do not require strong connectivity; connectivity is used for global stability at R0 = 1.Source: §4.1 · pp. 5–6

4.2 Existence and Stability of an Endemic State

• The endemic state is also locally exponentially stable when R0 > 1 on strongly connected digraphs.The result is established using Lyapunov’s indirect method.Source: §4.2 · p. 6
• When R0 > 1, a unique strong endemic state exists and is globally asymptotically stable for nonzero initial states on strongly connected digraphs.The proof uses a single Lyapunov function and positive-systems theory rather than graph-combinatorial arguments.Source: §4.2 · pp. 6–7

4 Stability of Epidemic Dynamics over Strongly Connected Graphs

• The quadratic Lyapunov proof is presented as useful for convergence-rate analysis and for extending equilibrium stability analysis to weakly connected digraphs.The section introduces the stability study for strongly connected topologies.Source: §4 · p. 5, §4.2 · p. 7

4.3 A Simplified Stability Condition through a Game-Theoretic Perspective

• A sufficient disease-free stability condition derived through the game-theoretic connection is linear and distributively checkable, but more restrictive than μ(A^T B − D) < 0.The paper relates the conditions through the Gershgorin Circle Theorem.Source: §4.3 · p. 7

• For a DAG with positive curing rates at every node, the disease-free equilibrium is the unique equilibrium and is globally asymptotically stable.Source: §5 · p. 7
• When G1 ≺ G2 and R1^0 > 1, any R2^0 is allowed, and the disease-free state and a unique strong endemic equilibrium across both components are the only equilibria.Source: §5 · p. 8
• When R1^0 ≤ 1 and R2^0 > 1, the only equilibria are the disease-free state and an endemic state confined to G2.Source: §5 · p. 8
• When R1^0 ≤ 1 and R2^0 ≤ 1, the disease-free state is the only possible equilibrium.Source: §5 · p. 8
• For two ordered SCCs, G2 is input-to-state stable, and the resulting equilibrium over G is globally asymptotically stable for all values of R1^0 and R2^0.Source: §5 · pp. 8–9

Two-SCC stability

• Since G1 is GAS and G2 is ISS, [13, Lemma 4.7] implies that the equilibrium of the cascaded system is GAS under those subsystem conditions.Source: §5 · p. 10
• When R1^0 ≤ 1 and R2^0 ≤ 1, the disease-free equilibrium is GAS; when R1^0 > 1 and R2^0 ≤ 1, the strong endemic equilibrium is GAS.The strong-endemic result assumes qi(0) ≠ 0 for all i ∈ [2].Source: §5 · p. 10
• When R2^0 > 1 and R1^0 ≤ 1, the weak endemic equilibrium is GAS if q2(0) ≠ 0; when both exceed one, the strong endemic equilibrium is GAS.For the both-supercritical case, the stated initial-condition assumption is qi(0) ≠ 0 for i ∈ [2].Source: §5 · p. 10

Corollary 1

• If every SCC has R_i^0 ≤ 1, the disease-free state is GAS; if the first supercritical SCC k follows only subcritical SCCs, the stated endemic state is GAS.The corollary assumes all initial infection components are nonzero.Source: §5 · p. 10

Weakly connected network

• In the 105-node, 66-SCC network, 13 nodes in G1 have high infection probabilities, four unreachable nodes become healthy, and other nodes have positive probabilities that decrease with distance from G1.G1 remains asymptotically stable despite receiving input from other SCCs.Source: §6 · p. 10

Ring graph

• In the 20-node ring, the state converges to the strictly positive endemic state while the Lyapunov function decreases monotonically to zero.The study presents this simulation as illustrating global asymptotic stability over connected undirected graphs.Source: §6 · pp. 10–11, Figure 4 · p. 11

Random graph

• In the 100-node random graph with 1704 edges, the state converges to the endemic state faster than in the ring-graph simulation.The random graph was connected and undirected.Source: §6 · p. 11

Stability results

• For strongly connected digraphs, the disease-free state is GAS when the basic reproduction number is at most 1, while the endemic state is GAS above 1 and locally exponentially stable.The endemic state is globally asymptotically stable when the network is not initialized at the disease-free state.Source: §7 · p. 11

Weakly connected graphs

• Viewing a weakly connected graph as a cascade of SCCs yields existence and uniqueness results for weak and strong endemic states, alongside stability analysis using input-to-state stability.The conclusion also states that the disease-free state's stability is studied for weakly connected graphs.Source: §7 · p. 11

Game characterization

• The n-intertwined Markov model is a special case of a proposed concave-game dynamical model of interactions among infected nodes.The resulting condition for disease-free equilibrium stability can be checked collectively by agents.Source: §7 · p. 11

Future work

• Future work is directed toward SIS stability on time-varying networks and optimal dynamic curing mechanisms.Source: §7 · p. 11

Matrix lemma

• For an irreducible Metzler matrix X with μ(X)=0, a positive diagonal R can be chosen so that X^T R + RX is negative semidefinite.The proof constructs R from positive left and right null vectors of X.Source: A Appendix · p. 12

Mapping properties

• The mapping T is monotonic for nonnegative X and 0 ⪯ y ≪ 1, and any strictly positive fixed point of T is unique.The appendix proves monotonicity directly and establishes uniqueness by comparing two proposed positive fixed points.Source: A Appendix · pp. 12–13

Fixed-point result

• When X is nonnegative and irreducible with ρ(X)>1 and 0 ⪯ y ≪ 1, T has a unique strictly positive fixed point.The proof uses an invariant closed subinterval and continuity to establish existence, then the uniqueness result for positive fixed points.Source: A Appendix · pp. 12–13