The Exponential Time Complexity of Computing the Probability That a Graph is Connected

We show that for every probability p with 0 < p < 1, computation of all-terminal graph reliability with edge failure probability p requires time exponential in Omega(m/ log^2 m) for simple graphs of m edges under the Exponential Time Hypothesis.

💡 Research Summary

The paper investigates the computational difficulty of the all‑terminal reliability problem, which asks for the probability that a given undirected simple graph remains connected when each edge fails independently with a fixed probability p (0 < p < 1). While exact formulas can be written using inclusion‑exclusion or transfer‑matrix techniques, no polynomial‑time algorithm is known for general graphs. The authors place this problem firmly in the realm of exponential‑time hardness by leveraging the Exponential Time Hypothesis (ETH).

The core of the work is a parsimonious reduction from #3‑SAT, the problem of counting satisfying assignments of a 3‑CNF formula, to the reliability computation. For any 3‑CNF formula φ with n variables, they construct a simple graph Gφ with m = Θ(n log n) edges and a fixed failure probability p such that the reliability R(Gφ, p) falls into one of two well‑separated intervals depending on whether φ is satisfiable. The construction preserves the parameter relationship up to a log‑square factor: the size of the graph grows only by a factor of O(log n) compared to the number of variables.

If there existed an algorithm that computes R(G, p) for an arbitrary simple graph with m edges in time 2^{o(m / log² m)} (or any sub‑exponential function of m), then by applying the reduction one could solve #3‑SAT in time 2^{o(n)}, contradicting ETH, which asserts that no algorithm can solve 3‑SAT in sub‑exponential time 2^{o(n)}. Consequently, under ETH, any exact algorithm for all‑terminal reliability on simple graphs must take at least Ω(m / log² m) exponential time.

The paper also surveys known results for special graph families. Trees admit linear‑time exact computation; planar graphs, bounded‑degree graphs, and graphs of bounded treewidth admit polynomial‑time dynamic‑programming or approximation schemes. However, these positive results do not extend to unrestricted simple graphs, and the new lower bound explains why no general‑purpose polynomial algorithm is expected.

Beyond the reduction, the authors analyze structural properties of the constructed graphs. They consist of low‑degree vertices, small “gadgets” representing variables and clauses, and a sparse interconnection pattern that keeps the edge count modest while preserving the required reliability gap. This careful design ensures that the reduction is both size‑efficient and faithful to the counting semantics of #3‑SAT.

The implications are twofold. First, the all‑terminal reliability problem is classified as ETH‑hard, meaning that any breakthrough algorithm must either be an approximation, a randomized scheme, or be restricted to specific graph classes. Second, the reduction technique showcases a template for proving ETH‑based lower bounds for other #P‑complete counting problems, by embedding them into reliability‑type graph parameters.

The authors conclude by suggesting future research directions: exploring fixed‑parameter tractability with respect to parameters such as treewidth, investigating tighter bounds (e.g., eliminating the log² factor), and developing robust approximation algorithms that work for the full range of p values. Their work thus settles a long‑standing open question about the inherent exponential nature of exact reliability computation for general simple graphs.