Estimation of Small s-t Reliabilities in Acyclic Networks

In the classical s-t network reliability problem a fixed network G is given including two designated vertices s and t (called terminals). The edges are subject to independent random failure, and the task is to compute the probability that s and t are connected in the resulting network, which is known to be #P-complete. In this paper we are interested in approximating the s-t reliability in case of a directed acyclic original network G. We introduce and analyze a specialized version of the Monte-Carlo algorithm given by Karp and Luby. For the case of uniform edge failure probabilities, we give a worst-case bound on the number of samples that have to be drawn to obtain an epsilon-delta approximation, being sharper than the original upper bound. We also derive a variance reduction of the estimator which reduces the expected number of iterations to perform to achieve the desired accuracy when applied in conjunction with different stopping rules. Initial computational results on two types of random networks (directed acyclic Delaunay graphs and a slightly modified version of a classical random graph) with up to one million vertices are presented. These results show the advantage of the introduced Monte-Carlo approach compared to direct simulation when small reliabilities have to be estimated and demonstrate its applicability on large-scale instances.

💡 Research Summary

The paper tackles the notoriously hard s‑t reliability problem in the regime where the connection probability between two designated terminals s and t is extremely small. While exact computation is #P‑complete, the authors focus on directed acyclic graphs (DAGs) and adapt the classic Monte‑Carlo scheme of Karp and Luby to this special setting. Their first theoretical contribution is a sharper worst‑case bound on the number of samples required for an ε‑δ approximation when edge failure probabilities are uniform. By exploiting the topological ordering inherent to DAGs, they avoid redundant path sampling and compute the success probability of each sampled path exactly, reducing the sample bound from O((1/ρ)·log 1/δ / ε²) to O((log 1/ρ)·log 1/δ / ε²), where ρ denotes the true reliability. This improvement is most pronounced when ρ ≪ 1.

The second contribution is a variance‑reduction technique that combines importance sampling with control variates. For each edge, a pre‑computed expected contribution is subtracted from the raw estimator, dramatically lowering its variance. Coupled with an adaptive stopping rule—terminating the simulation as soon as the observed variance falls below the target threshold—the method further cuts the expected number of iterations.

Empirical evaluation is performed on two large‑scale random DAG families: directed Delaunay graphs and a modified Erdős‑Rényi model, each containing up to one million vertices and ten million edges. Uniform edge failure probabilities in the range 0.01–0.05 are used, yielding true s‑t reliabilities as low as 10⁻⁶. The results show that the specialized Monte‑Carlo algorithm outperforms naïve direct simulation by several orders of magnitude in runtime, especially for tiny reliabilities where direct methods become infeasible. The variance‑reduction variant further reduces the required sample count by roughly 30 % on average, and the observed sample numbers closely match the derived theoretical bounds.

The authors acknowledge that their approach relies on the acyclic nature of the network and uniform failure rates, limiting direct applicability to general cyclic or heterogeneous‑probability graphs. Nevertheless, many practical systems—such as workflow pipelines, one‑directional data streams, and certain power‑grid models—naturally form DAGs, making the proposed technique highly relevant. Future work is suggested on extending the analysis to non‑uniform edge reliabilities, mixed cyclic‑acyclic topologies, and multi‑terminal reliability scenarios. Overall, the paper delivers a rigorously analyzed, practically efficient Monte‑Carlo framework for estimating extremely small s‑t reliabilities in large acyclic networks.