Rapidly Convergent Algorithms for Nonsmooth Optimization.
Abstract
The research supported under this grant has led to new developments for solving nonlinear optimization problems involving functions that are not everywhere differentiable and/or are implicitly defined, such as those that arise from dual formulations of optimization models. A new result has been proved showing rapid convergence of an algorithm for the single variable case where generalized derivations are available. For the single variable case where only function values are used a safeguarded bracketing technique has been introduced which guarantees convergence for lower semicontinous functions and preserves rapid convergence of polyhedral and/or polynomial fitting algorithms. A safeguarded polyhedral/quadratic fitting algorithm has been developed which has better than linear convergence for certain piecewise twice continously differentiable functions. Ideas from the single variable case are being extended to the multi-variable case. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 14, 1986
- Accession Number
- ADA182531
Entities
People
- Robert Mifflin
Organizations
- Washington State University