Propagation on networks: an exact alternative perspective

By generating the specifics of a network structure only when needed (on-the-fly), we derive a simple stochastic process that exactly models the time evolution of susceptible-infectious dynamics on finite-size networks. The small number of dynamical variables of this birth-death Markov process greatly simplifies analytical calculations. We show how a dual analytical description, treating large scale epidemics with a Gaussian approximations and small outbreaks with a branching process, provides an accurate approximation of the distribution even for rather small networks. The approach also offers important computational advantages and generalizes to a vast class of systems.

💡 Research Summary

The paper introduces a novel “on‑the‑fly” framework for modeling propagation processes on configuration‑model networks that eliminates the need to pre‑construct the entire graph. Traditional simulations first generate a static network by randomly pairing stubs according to a prescribed degree sequence and then run a contagion dynamics (e.g., SI, SIR) on that fixed structure. This two‑step approach is memory‑intensive and computationally costly, especially for large‑scale systems where the number of nodes N can be millions.

The authors observe that, during an infection event, the exact identity of the neighbor of an infectious node is irrelevant until the moment the infection actually occurs. Exploiting this, they define a stochastic process whose state is a compact vector x = (x_{‑1}, x_0, x_1, …, x_{k_max}). Here x_{‑1} is the total number of unmatched stubs, x_k counts susceptible vertices of degree k, and λ(x) = x_{‑1} – Σ_k k x_k denotes the number of unmatched stubs belonging to infectious vertices. An infection attempt corresponds to selecting one of the λ(x) infectious stubs and randomly pairing it with another unmatched stub. If the partner stub belongs to another infectious vertex, the state changes by r_{‑1} = (‑2,0,…,0); if it belongs to a susceptible vertex of degree k, the state changes by r_k = (‑2,‑δ_{k0},…,‑δ_{kk},…,0).

These two families of transitions (j = –1 and j = k ≥ 0) define a continuous‑time Markov birth‑death process with transition rates

q_{‑1}(x) = β λ(x) (λ(x)‑1)/(x_{‑1}‑1),
q_k(x) = β λ(x) k x_k/(x_{‑1}‑1).

The master equation

∂_t P(x,t) = Σ_j