The k-shortest path problem is a generalization of the fundamental shortest path problem, where the goal is to compute k simple paths from a given source to a target node, in non-decreasing order of their weight. With numerous applications modeling various optimization problems and as a feature in some learning systems, there is a need for efficient algorithms for this problem. Unfortunately, despite many decades of research, the best directed graph algorithm still has a worstcase asymptotic complexity of Õ(k n(n + m)). In contrast to the worstcase complexity, many algorithms have been shown to perform well on small diameter directed graphs in practice. In this paper, we prove that the average-case complexity of the popular Yen’s algorithm on directed random graphs with edge probability p = Ω(log n)/n in the unweighted and uniformly distributed weight setting is O(kmlog n), thus explaining the gap between the worst-case complexity and observed empirical performance. While we also provide a weaker bound of O(kmlog4 n) for sparser graphs with p ≥ 4/n, we show empirical evidence that the stronger bound should also hold in the sparser setting. We then prove that Feng’s directed k-shortest path algorithm computes the second shortest path in expected O(m) time on random graphs with edge probability p = Ω(log n)/n. Empirical evidence suggests that the average-case result for the Feng's algorithm holds even for k > 2 and sparser graphs.
CITATION STYLE
Schickedanz, A., Ajwani, D., Meyer, U., & Gawrychowski, P. (2019). Average-case behavior of k-Shortest path algorithms. In Studies in Computational Intelligence (Vol. 812, pp. 28–40). Springer Verlag. https://doi.org/10.1007/978-3-030-05411-3_3
Mendeley helps you to discover research relevant for your work.