Final Report on Contract N00014-87-K-0163 (1 January 1987-31 May 1989)
Abstract
Implementation began for DECOMPAR, a linear programming decomposition code in Fortran to run on the VAX/UNIX based CRYSTAL multicomputer. Documentation continued for DECOMPAR. A large part of our computational work is based on DECOMP, a FORTRAN implementation of the Dantzig-Wolfe decomposition algorithm for block-angular linear programs. This code has evolved into a robust and relatively portable experiment tool for large-scale mathematical programming. Analysis and empirical study of computational strategies for parallel decomposition were made. The multistage, multiproduct material requirements planning problem with capacity constraints. The stochastic dynamic traffic assignment problem. Previously, the PI has shown that a class of non-linear, non-convex dynamic network flow problems can be solved as a sequence of linear programs. The LP's have the staircase structure typical of time-phased problems. A generalization to the case where input traffic at the nodes are stochastic was proposed and analyzed. Computing true shadow prices in linear programming. The optimal values of the dual variables in a LP are true shadow prices (marginal values) only under nondegeneracy. nonprocedural implementation of mathematical programming algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 31, 1989
- Accession Number
- ADA216039
Entities
People
- James K. Ho
Organizations
- University of Tennessee system