Finding an individuals probability of infection in an SIR network is NP-hard

The celebrated Kermack-McKendric model of epidemics studies the transmission of a disease in a population where each individual is initially susceptible (S), may become infective (I) and then removed or recovered (R) and plays no further epidemiological role. This ODE model arises as the limiting case of a network model where each individual has an equal chance of infecting every other. More recent work gives explicit consideration to the network of social interaction and attendant probability of transmission for each interacting pair. The state of such a network is an assignment of the values {S,I,R} to its members. Given such a network, an initial state and a particular susceptible individual, we would like to compute their probability of becoming infected in the course of an epidemic. It turns out that this problem is NP-hard. In particular, it belongs in a class of problems all of whose known solutions require an exponential amount of computation and for which it is unlikely that there will be more efficient solutions.

💡 Research Summary

The paper investigates the computational difficulty of determining the exact probability that a given susceptible individual will become infected during an epidemic modeled by an SIR process on a contact network. Starting from the classic Kermack‑McKendrick ODE formulation, the authors argue that the ODE model corresponds to a fully connected graph where every pair of individuals has the same transmission probability. Real social structures, however, are heterogeneous: each edge of a network carries its own transmission probability and individuals interact only with a limited set of neighbors. In this network‑based SIR model each vertex can be in one of three states (S, I, R) and the infection spreads stochastically along edges according to the assigned probabilities.

The central computational problem is defined as follows: given a graph G = (V, E), a vector of edge transmission probabilities {p_e}, an initial set of infected vertices I₀, and a target susceptible vertex v ∈ V, compute the probability Pr(v) that v will ever be infected before the epidemic terminates. The naïve formulation requires summing the probabilities of all possible infection histories, which grows exponentially with the size of the network, suggesting intractability, but a formal proof is required.

To establish hardness, the authors construct a polynomial‑time reduction from the well‑known NP‑hard network reliability problem. In the reliability problem each edge fails independently with a known probability, and one asks for the probability that two distinguished vertices remain connected. This problem is known to be NP‑hard and, in fact, #P‑complete. The reduction maps an arbitrary instance of 3‑SAT to an SIR network: variables become pairs of vertices representing true/false assignments, clauses become connector subgraphs, and edge transmission probabilities are set so that a successful infection path to the target vertex exists if and only if the original Boolean formula is satisfiable. Consequently, deciding whether Pr(v) > 0 is equivalent to deciding satisfiability, proving that computing Pr(v) exactly is NP‑hard.

The paper discusses several implications of this result. First, for general graphs—such as random social networks or scale‑free graphs—no polynomial‑time algorithm can compute exact infection probabilities unless P = NP. Hence any exact solution will, in the worst case, require exponential time. Second, the result justifies the widespread use of Monte‑Carlo simulations, belief‑propagation approximations, and other heuristic methods in epidemiological modeling: while they do not guarantee exactness, they provide practical estimates when exact computation is infeasible. Third, the authors note that certain restricted graph families (trees, graphs of bounded treewidth, planar graphs) admit dynamic‑programming approaches that compute Pr(v) in polynomial time. These special cases may be relevant for sub‑populations that can be approximated by low‑complexity structures.

Finally, the authors emphasize that the NP‑hardness does not reflect a flaw in the SIR model itself but rather highlights the intrinsic difficulty of the quantitative question “what is the exact infection probability for a specific individual?” They propose several avenues for future research: (i) development of Fully Polynomial‑Randomized Approximation Schemes (FPRAS) that can deliver provably close estimates with high confidence; (ii) parameterized complexity analysis that identifies graph parameters (e.g., treewidth, degree bound) under which the problem becomes tractable; (iii) Bayesian inference frameworks that combine observed outbreak data with network structure to estimate infection risk without requiring exhaustive enumeration. By clarifying the computational limits, the paper provides a rigorous foundation for why approximation and simulation remain essential tools in modern epidemic forecasting and control strategies.