Advertisement

Asymptotic Complexity of Matrix Multiplication

  • Peter Bürgisser
  • Michael Clausen
  • Mohammad Amin Shokrollahi
Part of the Grundlehren der mathematischen Wissenschaften book series (GL, volume 315)

Abstract

This chapter is devoted to the discussion of fast matrix multiplication algorithms. We define the exponent ω of matrix multiplication, a quantity measuring the asymptotic complexity of this problem. Strassen’s original algorithm gives ω ≤ 2.81 (see Chap. 1). The robustness of the exponent with respect to various cost functions, and its invariance with respect to field extensions is shown. (Further motivation for the investigation of ω is given in Chap. 16.) We introduce the concept of border rank of tensors due to Bini, Capovani, Lotti, and Romani [49], and deduce Schönhage’s asymptotic sum inequality [460], which has become one of the main tools for gaining upper estimates of the exponent. Then we present the laser method [508] and its generalization by Coppersmith and Winograd [134], and prove their estimate ω < 2.39. Finally, an extension of the asymptotic sum inequality to the multiplication of partially filled matrices is considered. As an application, we describe Coppersmith’s construction of astonishingly rapid algorithms for the multiplication of rectangular matrices.

Keywords

Matrix Multiplication Scalar Restriction Scalar Extension Rectangular Matrice Trilinear Form 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag Berlin Heidelberg 1997

Authors and Affiliations

  • Peter Bürgisser
    • 1
  • Michael Clausen
    • 2
  • Mohammad Amin Shokrollahi
    • 2
    • 3
  1. 1.Institut für Mathematik Abt. Angewandte MathematikUniversität Zürich-IrchelZürichSwitzerland
  2. 2.Institut für Informatik VUniversität BonnBonnGermany
  3. 3.International Computer Science InstituteBerkeleyUSA

Personalised recommendations