English

Matrix Multiplication and Binary Space Partitioning Trees : An Exploration

Data Structures and Algorithms 2020-12-11 v1

Abstract

Herein we explore a dual tree algorithm for matrix multiplication of ARM×DA\in \mathbb{R}^{M\times D} and BRD×NB\in\mathbb{R}^{D\times N}, very narrowly effective if the normalized rows of AA and columns of BB, treated as vectors in RD\mathbb{R}^{D}, fall into clusters of order proportionate to Ω(Dτ)\Omega(D^{\tau}) with radii less than arcsin(ϵ/2)\arcsin(\epsilon/\sqrt{2}) on the surface of the unit DD-ball. The algorithm leverages a pruning rule necessary to guarantee ϵ\epsilon precision proportionate to vector magnitude products in the resultant matrix. \textit{ Unfortunately, if the rows and columns are uniformly distributed on the surface of the unit DD-ball, then the expected points per required cluster approaches zero exponentially fast in DD; thus, the approach requires a great deal of work to pass muster.}

Keywords

Cite

@article{arxiv.2012.05365,
  title  = {Matrix Multiplication and Binary Space Partitioning Trees : An Exploration},
  author = {CNP Slagle and Lance Fortnow},
  journal= {arXiv preprint arXiv:2012.05365},
  year   = {2020}
}
R2 v1 2026-06-23T20:51:32.482Z