Lossless Online Rounding for Online Bipartite Matching (Despite its Impossibility)

TL;DR

This paper explores the possibility of lossless online rounding in bipartite matching, introducing constraints to overcome the general impossibility and achieving improved algorithms and thresholds.

Contribution

It characterizes non-convex constraints enabling lossless online rounding, leading to better competitive ratios and new insights into randomness requirements in online bipartite matching.

Findings

Lossless online rounding is generally impossible, but can be achieved under specific constraints.

Introduces non-convex constraints that enable lossless rounding and improved competitive ratios.

Provides thresholds for randomness needed to outperform deterministic algorithms.

Abstract

For numerous online bipartite matching problems, such as edge-weighted matching and matching under two-sided vertex arrivals, the state-of-the-art fractional algorithms outperform their randomized integral counterparts. This gap is surprising, given that the bipartite fractional matching polytope is integral, and so lossless rounding is possible. This gap was explained by Devanur et al.~(SODA'13), who showed that \emph{online} lossless rounding is impossible. Despite the above, we initiate the study of lossless online rounding for online bipartite matching problems. Our key observation is that while lossless online rounding is impossible \emph{in general}, randomized algorithms induce fractional algorithms of the same competitive ratio which by definition are losslessly roundable online. This motivates the addition of constraints that decrease the ``online integrality gap'', thus allowing for lossless online rounding. We characterize a set of non-convex constraints which allow for such lossless online rounding, and better competitive ratios than yielded by deterministic algorithms. As applications of our lossless online rounding approach, we obtain two results of independent interest: (i) a doubly-exponential improvement, and a sharp threshold for the amount of randomness (or advice) needed to outperform deterministic online (vertex-weighted) bipartite matching algorithms, and (ii) an optimal semi-OCS, matching a recent result of Gao et al.~(FOCS'21) answering a question of Fahrbach et al.~(FOCS'20).