Variable Metric Secant Updates from Matrix Factorizations.
Abstract
The major thrust of the research accomplished under this grant is toward algorithms for large problems, although there have been other results along the way, like connections between accuracy attainable by Newton-like methods and noise in residual computations. The main thrusts were toward finding a variable-metric technique for large sparse nonlinear optimization problems by the novel approach of defining the sparse derivative matrix approximations in terms of their sparse factorization which would be updated at each iteration, and toward designing algorithms that exploit the systolic array architecture in numerical linear algebra. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 21, 1983
- Accession Number
- ADA128771
Entities
People
- Franklin Luk
- John E. Dennis
Organizations
- Rice University