A New Class of Feasible Direction Methods.

Abstract

This paper introduces a new class of feasible direction methods for solving convex programming problems with differentiable functions. Unlike most currently used feasible direction methods, the ones presented here are not based on the Kuhn-Tucker theory. One of the methods is designed to solve convex problems without assuming Slater's condition or any other constraint qualification. The other method assumes Slater's condition but it provides at each step a feasible direction of steepest descent which is generally better than the one obtained by the classical Zoutendijk method and its variants.

Document Details

Document Type
Technical Report
Publication Date
Feb 01, 1975
Accession Number
ADA008362

Entities

People

  • A. Ben-tal
  • S. Zlobec

Organizations

  • University of Texas at Austin

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Applied Mathematics
  • Computer Programming
  • Convex Programming
  • Cooperation
  • Evolutionary Algorithms
  • Heuristic Methods
  • Mathematics
  • Operations Research
  • Qualifications

Readers

  • Operations Research