Competitive algorithms for online matching and vertex cover problems

Abstract

The past decade has witnessed an explosion of research on the online bipartite matching problem. Surprisingly, its dual problem, online bipartite vertex cover, has never been explicitly studied before. One of the motivation for studying this problem is that it significantly generalizes the classical ski rental problem. An instance of such problems specifies a bipartite graph G = (L, R, E) whose left vertices L are offline and right vertices arrive online one at a time. An algorithm must maintain a valid vertex cover from which no vertex can ever be removed. The objective is to minimize the size of the cover. In this thesis, we introduce a charging-based algorithmic framework for this problem as well as its generalizations. One immediate outcome is a simple analysis of an optimal 1/1-1/e- competitive algorithm for online bipartite vertex cover. By extending the charging-based analysis in various nontrivial ways, we also obtain optimal l_1 e-competitive algorithms for the edge-weighted and submodular versions of online bipartite vertex cover, which all match the best performance of ski rental. As an application, we show that by analyzing our algorithm in the primal-dual framework, our result on submodular vertex cover implies an optimal (1/1-1/e)-competitive algorithm for its dual, online bipartite submodular matching. This problem is a generalization of online bipartite matching and may have applications in display ad allocation. We consider also the more general scenario where all the vertices are online and the graph is not necessarily bipartite, which is known as the online fractional vertex cover and matching problems. Our contribution in this direction is a primal-dual 1.901-competitive (or 1/1.901 ~~ 0.526) algorithm for these problems. Previously, it was only known that they admit a simple well-known 2-competitive (or 1/2) greedy algorithm. Our result is the first successful attempt to beat the greedy algorithm for these two problems. Moreover, our algorithm for the online matching problem significantly generalizes the traditional online bipartite graph matching problem, where vertices from only one side of the bipartite graph arrive online. In particular, our algorithm improves upon the result of the fractional version of the online edge-selection problem in Blum et. al. (JACM '06). Finally, on the hardness side, we show that no randomized online algorithm can achieve a competitive ratio better than 1.753 and 0.625 for the online fractional vertex cover problem and the online fractional matching problem respectively, even for bipartite graphs.