Piecewise Linear-Quadratic Programming and Its Applications
Abstract
Piecewise linear-quadratic programming problems are a fundamental class of optimization problems in the mathematical modeling of multistage decision-making and large-scale dynamic systems, with or without the presence of uncertainty. Patterns of mathematical structure in such problems have been identified that cover wide areas of application and are conductive to the development of solution methodology. Work has gone forward on utilizing this structure in new numerical procedures, which include finite-envelope methods and a double conjugate gradient method:, as well as a simplex-like algorithm for solving small scale subproblems. Preliminary tests have been made of these procedures on problems of modest size. To pave the way for experiments with larger examples, programs modules for handling discrete-time dynamics have been coded in part. For decision problems involving scenarios, a progressive hedging algorithm have been devised. This provides a systematic approach to optimization in cases where uncertainty cannot be modeled in terms of standard random variables.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1988
- Accession Number
- ADA214489
Entities
People
- R. T. Rockafellar
Organizations
- University of Washington