A Globally Convergent Condensation Method for Geometric Programming.
Abstract
To solve a posynomial geometric program, rather than solve its dual directly, one solves the duals of a sequence of approximating geometric programs. Each approximating program, obtained by condensing certain terms of the given problem, has fewer degrees of difficulty than the given problem. Therefore, this method is to solve a sequence of problems, each posed in the same low dimensional space, rather than solve one problem in a higher dimensional space. The method requires a special sub-matrix of the matrix of exponents, and procedures to find such a sub-matrix are presented. A line search on an exact penalty function is employed. Under mild conditions, from any arbitrary starting solution estimate the method generates a sequence of estimates converging to a solution. The method also generates a sequence of lower bounds, thus providing an effective stopping criterion. A numerical example is presented.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1979
- Accession Number
- ADA081067
Entities
People
- Eric Rosenberg
Organizations
- Stanford University