A Computational Comparison of Gradient Minimization Algorithms.

Abstract

A technique was developed for the comparison of gradient minimization routines in solving the unconstrained optimization problem. The problem of locating the local minimum of a given real-valued, non-negative, differentiable function was used in this study to compare three gradient algorithms, namely Davidon's variance algorithm, the Fletcher-Reeves algorithm, and the Fletcher-Powell algorithm. A cost criterion and an average convergence rate were devised to facilitate the comparisons of these algorithms. Davidon's variance algorithm, a rank-one method, was judged to be the best routine for over fifty-three percent of the total cases tested. The comparisons are graphically presented in an appendix. (Author)

Document Details

Document Type
Technical Report
Publication Date
Mar 01, 1971
Accession Number
AD0730159

Entities

People

  • Craig E. Miller

Organizations

  • Air Force Institute of Technology

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Convergence
  • Heuristic Methods
  • Mathematics

Readers

  • Approximation Theory.
  • Linear Algebra
  • Systems Analysis and Design