Sublinear-Time Decremental Algorithms for Single-Source Reachability and Shortest Paths on Directed Graphs

TL;DR

This paper introduces a randomized decremental algorithm for single-source reachability and approximate shortest paths in directed graphs, significantly improving the total update time over the long-standing classic algorithms, especially for sparse and dense graphs.

Contribution

It presents the first sublinear-time randomized decremental algorithms for SSR and SSSP, improving upon the classic $O(mn)$ total update time and extending to SCCs with efficient performance.

Findings

Expected total update time of $O(m n^{9/10 + o(1)})$ for SSR and approximate SSSP.

Algorithms are most efficient for both sparse ($m=O(n)$) and dense ($m=O(n^2)$) graphs.

Achieves constant query time with high probability against an oblivious adversary.

Abstract

We consider dynamic algorithms for maintaining Single-Source Reachability (SSR) and approximate Single-Source Shortest Paths (SSSP) on $n$-node $m$-edge directed graphs under edge deletions (decremental algorithms). The previous fastest algorithm for SSR and SSSP goes back three decades to Even and Shiloach [JACM 1981]; it has $ O(1) $ query time and $ O (mn) $ total update time (i.e., linear amortized update time if all edges are deleted). This algorithm serves as a building block for several other dynamic algorithms. The question whether its total update time can be improved is a major, long standing, open problem. In this paper, we answer this question affirmatively. We obtain a randomized algorithm with an expected total update time of $ O(\min (m^{7/6} n^{2/3 + o(1)}, m^{3/4} n^{5/4 + o(1)}) ) = O (m n^{9/10 + o(1)}) $ for SSR and $(1+\epsilon)$-approximate SSSP if the edge weights are integers from $ 1 $ to $ W \leq 2^{\log^c{n}} $ and $ \epsilon \geq 1 / \log^c{n} $ for some constant $ c $. We also extend our algorithm to achieve roughly the same running time for Strongly Connected Components (SCC), improving the algorithm of Roditty and Zwick [FOCS 2002]. Our algorithm is most efficient for sparse and dense graphs. When $ m = \Theta(n) $ its running time is $ O (n^{1 + 5/6 + o(1)}) $ and when $ m = \Theta(n^2) $ its running time is $ O (n^{2 + 3/4 + o(1)}) $. For SSR we also obtain an algorithm that is faster for dense graphs and has a total update time of $ O ( m^{2/3} n^{4/3 + o(1)} + m^{3/7} n^{12/7 + o(1)}) $ which is $ O (n^{2 + 2/3}) $ when $ m = \Theta(n^2) $. All our algorithms have constant query time in the worst case and are correct with high probability against an oblivious adversary.