We consider the CONGEST model on a network with $n$ nodes, $m$ edges, diameter $D$, and integer costs and capacities bounded by $\text{poly} n$. In this paper, we show how to find an exact solution to the minimum cost flow problem in $n^{1/2+o(1)}(\sqrt{n}+D)$ rounds, improving the state of the art algorithm with running time $m^{3/7+o(1)}(\sqrt nD^{1/4}+D)$ [Forster et al. FOCS 2021], which only holds for the special case of unit capacity graphs. For certain graphs, we achieve even better results. In particular, for planar graphs, expander graphs, $n^{o(1)}$-genus graphs, $n^{o(1)}$-treewidth graphs, and excluded-minor graphs our algorithm takes $n^{1/2+o(1)}D$ rounds. We obtain this result by combining recent results on Laplacian solvers in the CONGEST model [Forster et al. FOCS 2021, Anagnostides et al. DISC 2022] with a CONGEST implementation of the LP solver of Lee and Sidford [FOCS 2014], and finally show that we can round the approximate solution to an exact solution. Our algorithm solves certain linear programs, that generalize minimum cost flow, up to additive error $\epsilon$ in $n^{1/2+o(1)}(\sqrt{n}+D)\log^3 (1/\epsilon)$ rounds.
翻译:本文考虑在有$n$个节点,$m$条边,直径为$D$且容量和代价的整数都受到$\text{poly} n$的限制的网络上的CONGEST模型。我们展示了如何在$n^{1/2+o(1)}(\sqrt{n}+D)$轮内求解最小费用流问题,这一结果比文章Forster等人FOCS 2021中的算法运行时间$m^{3/7+o(1)}(\sqrt nD^{1/4}+D)$要优秀,后者仅适用于单位容量图这种特殊情况。对于某些图,我们取得了更优秀的结果,尤其是对于平面图、展开图、$n^{o(1)}$-genus图、$n^{o(1)}$-treewidth图和排除小图的图形,我们的算法仅需要$n^{1/2+o(1)}D$轮。我们通过将CONGEST模型中的Laplacian求解器的最新结果[Forster等人FOCS 2021, Anagnostides等人DISC 2022]与Lee和Sidford [FOCS 2014]提出的线性规划求解器的CONGEST实现相结合,并最终证明我们可以对近似解四舍五入为准确解来获得这些结果。我们的算法可以在$n^{1/2+o(1)}(\sqrt{n}+D)\log^3(1/\epsilon)$轮内,将一定的线性规划问题(这些问题推广了最小费用流问题)的解求得到$\epsilon-$精度。