Virus Dynamics on $k$-Level Starlike Graphs

TL;DR

This paper extends the analysis of virus spread dynamics from 2-level to k-level starlike graphs, identifying critical thresholds for infection rates that determine whether the virus dies out or persists.

Contribution

It generalizes previous results to k-level starlike graphs, providing a method to analyze complex hierarchical structures in virus propagation models.

Findings

Derived infection threshold for k-level graphs

Reduced k-level analysis to 3-level case

Identified conditions for virus die-out or persistence

Abstract

Becker, Greaves-Tunnell, Kontorovich, Miller, Ravikumar, and Shen determined the long term evolution of virus propagation behavior on a hub-and-spoke graph of one central node and $n$ neighbors, with edges only from the neighbors to the hub (a $2$-level starlike graph), under a variant of the discrete-time SIS (Suspectible Infected Suspectible) model. The behavior of this model is governed by the interactions between the infection and cure probabilities, along with the number $n$ of $2$-level nodes. They proved that for any $n$, there is a critical threshold relating these rates, below which the virus dies out, and above which the probabilistic dynamical system converges to a non-trivial steady state (the probability of infection for each category of node stabilizes). For $a$, the probability at any time step that an infected node is not cured, and $b$, the probability at any time step that an infected node infects its neighbors, the threshold for the virus to die out is $b \leq (1-a)/\sqrt{n}$. We extend this analysis to $k$-level starlike graphs for $k \geq 3$ (each $(k-1)$-level node has exactly $n_k$ neighbors, and the only edges added are from the $k$-level nodes) for infection rates above and below the critical threshold of $(1-a)/\sqrt{n_1+n_2+\dots+n_{k-1}}$. We do this by first analyzing the dynamics of nodes on each level of a $3$-level starlike graph, then show that the dynamics of the nodes of a $k$-level starlike graph are similar, enabling us to reduce our analysis to just $3$ levels, using the same methodology as the $3$-level case.