Faster high-accuracy multicommodity flow in dense graphs

We give a fast algorithm for solving min-cost $k$-commodity flow. The basic idea is to construct an auxiliary linear program that has low rank and whose minimum value is at most $1/k$ times the minimum value of the original problem (thus, solving this auxiliary linear program will make at least $1/k$ fraction of progress in the original problem). Low-rank linear programs can be solved quickly using black-box techniques. Thus, our algorithm runs in time $\tilde{O}(\operatorname{poly}(k)(n^{2.5}+m\sqrt{n}))$ on a directed graph with $n$ vertices and $m$ edges. We do not rely on any fast matrix multiplication.

Source: arXiv:2610.06665v1 - http://arxiv.org/abs/2610.06665v1 PDF: https://arxiv.org/pdf/2610.06665v1 Original Link: http://arxiv.org/abs/2610.06665v1