Accession Number : ADA132599

Title :   Ellipsoid Algorithm Variants in Nonlinear Programming.

Descriptive Note : Doctoral thesis,

Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH

Personal Author(s) : Dziuban,Stephen T

PDF Url : ADA132599

Report Date : Aug 1983

Pagination or Media Count : 274

Abstract : This thesis presents and evaluates variants of the ellipsoid algorithm for nonlinear programming. Two primary types of variants are examined: different deep cut hyperplanes, and different strategies for testing the feasibility constraints. Five of the deep cut variants do not require a linesearch, while three do require a linesearch. Of the five constraint examination methods, three are alternative ways of examining the full list of feasibility constraints. The fourth method is an active set strategy, while the fifth uses a record objective function value constraint. The variants are tested on 13 general and geometric programming problems, both convex and nonconvex. The performance of each variant is measured in combined solution error as a function of solution time. The experimental results show that the lowest solution error achieved is essentially unaffected by the constraint examination strategy. However, all but one of the deep cut variants occasionally converge to non-optimal points if the problem is nonconvex.

Descriptors :   *ALGORITHMS, *NONLINEAR PROGRAMMING, VARIATIONS, PROBLEM SOLVING, ACCURACY, ELLIPSOIDS, PERFORMANCE TESTS, TEST METHODS, COMPUTATIONS, EFFICIENCY, THESES

Subject Categories : Theoretical Mathematics

Distribution Statement : APPROVED FOR PUBLIC RELEASE