New Methods for Nonlinear Optimization.
Abstract
This research project has investigated topics in unconstrained and constrained optimization, solving systems of nonlinear equations, nonlinear least squares, and parallel optimization. We have continued our development of tensor methods for nonlinear equations and extended these methods to unconstrained optimization and nonlinear least squares. In all cases, the tensor methods appear to provide significant practical improvements over the best currently known methods, on both singular and nonsingular problems. We have developed new trust region methods for equality constrained optimization problems that have strong convergence properties, and have begun to implement these methods. We have also developed new analysis techniques that provide local convergence results for constrained optimization problems. We have developed and analyzed an efficient method for orthogonal distance regression, intended for problems where there are errors in independent as well as dependent variables, and have developed a robust code that implements this method. Finally, we have developed, implemented, and analyzed parallel methods for global optimization and for unconstrained optimization. (KR)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 11, 1988
- Accession Number
- ADA195982
Entities
People
- Richard H. Byrd
- Robert B. Schnabel
Organizations
- University of Colorado Boulder