Matroids provide one of the most elegant structures for algorithm design. This is best identified by the Edmonds-Rado theorem relating the success of the simple greedy algorithm to the anatomy of the optimal basis of a matroid [Edm71; Rad57]. As a response, much energy has been devoted to understanding a matroid's computational properties. Yet, less is understood where parallel algorithms are concerned. In response, we initiate the study of parallel matroid optimization in the adaptive complexity model [BS18]. First, we reexamine Bor\r{u}vka's classical minimum weight spanning tree algorithm [Bor26b; Bor26a] in the abstract language of matroid theory, and identify a new certificate of optimality for the basis of any matroid as a result. In particular, a basis is optimal if and only if it contains the points of minimum weight in every circuit of the dual matroid. Hence, we can witnesses whether any specific point belongs to the optimal basis via a test for local optimality in a circuit of the dual matroid, thereby revealing a general design paradigm towards parallel matroid optimization. To instantiate this paradigm, we use the special structure of a binary matroid to identify an optimization scheme with low adaptivity. Here, our key technical step is reducing optimization to the simpler task of basis search in the binary matroid, using only logarithmic overhead of adaptive rounds of queries to independence oracles. Consequentially, we compose our reduction with the parallel basis search method of [KUW88] to obtain an algorithm for finding the optimal basis of a binary matroid terminating in sublinearly many adaptive rounds of queries to an independence oracle. To the authors' knowledge, this is the first algorithm for matroid optimization to outperform the greedy algorithm in terms of adaptive complexity in the independence query model without assuming the matroid is encoded by a graph.
翻译:暂无翻译