Coupling a branching process to an infinite dimensional epidemic process

Branching process approximation to the initial stages of an epidemic process has been used since the 1950’s as a technique for providing stochastic counterparts to deterministic epidemic threshold theorems. One way of describing the approximation is to construct both branching and epidemic processes on the same probability space, in such a way that their paths coincide for as long as possible. In this paper, it is shown, in the context of a Markovian model of parasitic infection, that coincidence can be achieved with asymptotically high probability until o(N^{2/3}) infections have occurred, where N denotes the total number of hosts.

💡 Research Summary

The paper provides a rigorous probabilistic coupling between a high‑dimensional Markovian epidemic model of parasitic infection and a much simpler continuous‑time branching process, thereby quantifying precisely how long the two processes can be made to follow identical sample paths. The authors begin by describing the epidemic model: a population of N hosts, each of which can be in one of several health states (susceptible, infected, recovered, etc.). Infection spreads through random pairwise contacts, and the transition rates are proportional to the product of the numbers of infected and susceptible hosts. Because each host carries a potentially infinite set of internal states (e.g., parasite load, immunity level), the state space is effectively infinite‑dimensional, which the authors refer to as an “infinite‑dimensional epidemic process.”

Next, they construct a continuous‑time branching process that mirrors the epidemic’s transmission mechanism. The branching process starts from a single ancestor (the first infected host). Each infected individual gives birth to new infections at rate β and dies (or recovers) at rate δ, exactly as in a classic birth‑death process. The key idea is to embed both the epidemic and the branching process on a common probability space by using the same underlying Poisson point processes to generate infection and recovery events. In this coupling, whenever the epidemic generates an infection event, the branching process generates a birth, and vice‑versa for recoveries.

The core technical contribution lies in analyzing the discrepancy between the two processes. The authors compare the epidemic’s infection rate λ(N,k)=βk(N−k)/N (where k is the current number of infected hosts) with the branching process’s birth rate βk. Expanding λ(N,k) yields βk+O(k²/N). The O(k²/N) term represents the “collision” error that arises because, in the full epidemic, two infected hosts may attempt to infect the same susceptible host simultaneously—a phenomenon absent in the branching approximation. By defining an error process that tracks the cumulative effect of these collisions, the authors show that it forms a martingale with bounded increments. Applying Doob’s maximal inequality and Azuma–Hoeffding concentration bounds, they obtain an exponential tail bound

 P( sup₀≤t≤τ |Error(t)| > ε ) ≤ exp(−c k³/N²)

for suitable constants c, ε, and stopping time τ defined as the first time the infection count reaches k. Setting k = o(N^{2/3}) makes the exponent tend to −∞, implying that the error probability vanishes as N→∞. Consequently, with asymptotically high probability, the epidemic and branching process trajectories coincide up to the moment when o(N^{2/3}) infections have occurred.

The paper then discusses the implications of this result. Classical epidemiological literature has long used branching‑process approximations to derive stochastic analogues of deterministic threshold theorems (e.g., the basic reproduction number R₀). However, prior work typically offered only heuristic or qualitative justification for the approximation’s validity in the early phase. This study supplies a precise quantitative bound: the approximation remains exact (in the coupling sense) until the infection count reaches the N^{2/3} scale. For realistic large populations (e.g., N ≈ 10⁶), this permits the branching approximation to be trusted for several thousand infections, far beyond the “very first few cases” intuition. Moreover, the result holds despite the underlying epidemic’s infinite‑dimensional nature, demonstrating that the high‑dimensional internal host dynamics do not affect the early‑stage dynamics beyond the O(k²/N) correction.

Finally, the authors outline extensions. The coupling technique can be adapted to other compartmental models (SEIR, SIS with waning immunity) and to network‑based transmission where contacts are not uniformly random. They also suggest investigating non‑Markovian settings (e.g., latency periods with general distributions) where the coupling would require more sophisticated renewal‑process constructions. In summary, the paper delivers a mathematically rigorous bridge between complex epidemic processes and tractable branching approximations, establishing a clear asymptotic regime (up to o(N^{2/3}) infections) where the two are virtually indistinguishable.