Bounding the Size and Probability of Epidemics on Networks

We consider an infectious disease spreading along the edges of a network which may have significant clustering. The individuals in the population have heterogeneous infectiousness and/or susceptibility. We define the out-transmissibility of a node to be the marginal probability that it would infect a randomly chosen neighbor given its infectiousness and the distribution of susceptibility. For a given distribution of out-transmissibility, we find the conditions which give the upper [or lower] bounds on size and probability of an epidemic, under weak assumptions on the transmission properties, but very general assumptions on the network. We find similar bounds for a given distribution of in-transmissibility (the marginal probability of being infected by a neighbor). We also find conditions giving global upper bounds on size and probability. The distributions leading to these bounds are network-independent. In the special case of networks with high girth (locally tree-like), we are able to prove stronger results. In general the probability and size of epidemics are maximal when the population is homogeneous and minimal when the variance of in- or out-transmissibility is maximal.

💡 Research Summary

The paper investigates how heterogeneity in individual infectiousness and susceptibility shapes the probability and final size of epidemics spreading on networks. Each node u has an infectiousness I_u and each neighbor v has a susceptibility S_v, drawn independently from distributions P(I) and P(S). The pairwise transmission probability is a general function T_uv = T(I_u, S_v). From these, the authors define two marginal quantities: the out‑transmissibility T_out(u) = ∫ T(I_u, S) P(S) dS, the probability that u infects a randomly chosen neighbor, and the in‑transmissibility T_in(v) = ∫ T(I, S_v) P(I) dI, the probability that v gets infected by a random neighbor. Both have the same mean h_Ti, which is the only network‑wide average that matters.

To analyze epidemic outcomes, the authors employ the Epidemic Percolation Network (EPN) construction. Starting from the original undirected graph G, each undirected edge {u,v} is replaced by two directed edges (u→v) and (v→u) that are retained independently with probabilities T_uv and T_vu. In a given realization of the EPN, the set of nodes that can be reached from a seed node u forms its out‑component; the set of nodes that can reach u forms its in‑component. The giant strongly connected component (SCC) of the EPN contains a giant in‑component H_in and a giant out‑component H_out. If the seed lies in H_in, the infection spreads to essentially all of H_out, which the authors define as an epidemic. The epidemic probability Y is the expected fraction of seeds belonging to H_in, and the expected epidemic size A is the expected fraction of nodes in H_out.

The central results are distribution‑based bounds that hold for any network, including those with clustering. For a fixed distribution Q_out(T) of out‑transmissibility, the epidemic probability Y and size A are bounded between two extreme cases:

1. Upper bound (maximum epidemic) – All nodes have the same out‑transmissibility equal to the mean, T_out = h_Ti (i.e., a homogeneous population). In this case both Y and A are maximized.
2. Lower bound (minimum epidemic) – The out‑transmissibility distribution is maximally heterogeneous: a fraction h_Ti of nodes have T_out = 1 and the remaining 1 – h_Ti have T_out = 0. This extreme variance yields the smallest possible Y and A.

Exactly analogous bounds hold when the distribution of in‑transmissibility Q_in(T) is prescribed. Thus, knowing only the marginal distribution of one of the transmissibilities suffices to bracket the epidemic outcomes, regardless of the underlying network topology.

When the network is locally tree‑like (high girth, i.e., no short cycles), the authors can strengthen the results. In such networks the percolation process on the EPN reduces to an independent branching process, and the quantities ψ_in(V) = 1 – h_Ti for a single neighbor become exact. Consequently the above homogeneous and maximally heterogeneous cases are not merely bounds but global optima: no other admissible distribution can produce larger (or smaller) Y or A for any network with the same local degree structure.

Mathematically, the proofs proceed in two stages. First, the authors define a survival probability function for the branching process on the EPN and use Jensen’s inequality to relate it to the mean transmissibility h_Ti. This yields the monotonicity of epidemic probability/size with respect to the variance of the transmissibility distribution. Second, they introduce the notion of sequential convergence of local statistics: a sequence of graphs {G_n} with |G_n|→∞ such that the distribution of any finite‑radius neighborhood stabilizes. Under this condition, the probabilities Y_d and A_d that a random node belongs to an in‑ or out‑component of depth d converge, and the limits Y and A coincide with the giant‑component probabilities defined earlier. This framework justifies applying the bounds to any large finite network, even when the exact infinite‑size limit is ambiguous.

From a public‑health perspective, the results imply that heterogeneity is protective: the more variance there is in either infectiousness or susceptibility, the lower the chance of a large outbreak and the smaller its eventual size. Conversely, a perfectly homogeneous population maximizes epidemic risk. Therefore, interventions that increase heterogeneity—such as targeted vaccination of high‑risk individuals, isolation of superspreaders, or measures that reduce contact rates for the most infectious—can be more effective than uniform policies, especially when only marginal transmissibility data are available.

In summary, the paper provides a rigorous, network‑agnostic framework for bounding epidemic probability and size based solely on the distribution of marginal transmissibilities. It unifies earlier results for unclustered configuration‑model networks, extends them to clustered graphs, and identifies the homogeneous and maximally heterogeneous distributions as the universal extremal cases. This contributes both to the theoretical understanding of epidemic percolation on complex networks and to practical risk assessment when detailed network data are lacking.