Improved Bounds for Rectangular Monotone Min-Plus Product and Applications

TL;DR

This paper generalizes a recent fast algorithm for Monotone Min-Plus Product to rectangular matrices and applies it to improve the efficiency of several related problems in graph algorithms and matrix computations.

Contribution

It extends the state-of-the-art algorithm for Monotone Min-Plus Product to rectangular matrices and enhances the running times of multiple applications.

Findings

Improved bounds for Rectangular Monotone Min-Plus Product.

Enhanced algorithms for Single Source Replacement Path and Batch Range Mode.

Faster solutions for k-Dyck Edit Distance and All Pairs Shortest Path approximation.

Abstract

In a recent breakthrough paper, Chi et al. (STOC'22) introduce an $\tilde{O}(n^{\frac{3 + \omega}{2}})$ time algorithm to compute Monotone Min-Plus Product between two square matrices of dimensions $n \times n$ and entries bounded by $O(n)$. This greatly improves upon the previous $\tilde O(n^{\frac{12 + \omega}{5}})$ time algorithm and as a consequence improves bounds for its applications. Several other applications involve Monotone Min-Plus Product between rectangular matrices, and even if Chi et al.'s algorithm seems applicable for the rectangular case, the generalization is not straightforward. In this paper we present a generalization of the algorithm of Chi et al. to solve Monotone Min-Plus Product for rectangular matrices with polynomial bounded values. We next use this faster algorithm to improve running times for the following applications of Rectangular Monotone Min-Plus Product: $M$-bounded Single Source Replacement Path, Batch Range Mode, $k$-Dyck Edit Distance and 2-approximation of All Pairs Shortest Path. We also improve the running time for Unweighted Tree Edit Distance using the algorithm by Chi et al.