Query Efficient Weighted Stochastic Matching

Abstract

In this paper, we study the weighted stochastic matching problem. Let G = (V, E) be a given edge-weighted graph, and let its realization 𝒢 be a random subgraph of G that includes each edge e ∈ E independently with a known probability p_e. The goal in this problem is to pick a sparse subgraph Q of G without prior knowledge of 𝒢, such that the maximum weight matching among the realized edges of Q (i.e., the subgraph Q ∩ 𝒢) in expectation approximates the maximum weight matching of the entire realization 𝒢.
It is established by previous work that attaining any constant approximation ratio for this problem requires selecting a subgraph of max-degree Ω(1/p), where p = min_{e ∈ E} p_e. On the positive side, there exists a (1-ε)-approximation algorithm by Behnezhad and Derakhshan [FOCS'20], albeit at the cost of a max-degree having exponential dependence on 1/p. Within the O(1/p) query regime, however, the best-known algorithm achieves a 0.536 approximation ratio due to Dughmi, Kalayci, and Patel [ICALP'23], improving over the 0.501 approximation algorithm by Behnezhad, Farhadi, Hajiaghayi, and Reyhani [SODA'19]. 
In this work, we present a 0.68-approximation algorithm with the asymptotically optimal O(1/p) queries per vertex. Our result not only substantially improves the approximation ratio for weighted graphs, but also breaks the well-known 2/3 barrier with the optimal number of queries - even for unweighted graphs. Our analysis involves reducing the problem to designing a randomized matching algorithm on a given stochastic graph with some variance-bounding properties. To achieve these properties, we leverage a randomized algorithm by MacRury and Ma [STOC'24] for a variant of online stochastic matching.