Algorithms for the maximum subarray problem based on matrix multiplication

Abstract

Given an M×N array of reals, we want to find a rectangular contiguous subarray such that the sum of the entries in the subarray is maximized. Since Bentley posed this problem in his Programming Pearls column in 1984 with an O(NM^2) time solution, no progress on the sequential complexity has been reported to date. We give the first subcubic algorithm, by reducing the problem to "funny matrix multiplication", where the scalar product and addition in usual matrix multiplication are replaced by addition and max operations, respectively. We also give a faster ε-approximation algorithm via the same reduction.