A Quasi-Newton Method for Unconstrained Minimization Problems.
Abstract
A method is described for the minimization of a function F(x) of n variables. Convergence to a stationary point is shown without assumptions on second order derivatives. If the sequence generated by this method has a cluster point in a neighbourhood of which F(x) is twice continuously differentiable and has a positive definite Hessian matrix, then the convergence is superlinear. It is shown that under appropriate assumptions n consecutive search directions are conjugate. No computation of second order derivatives is required.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1975
- Accession Number
- ADA011013
Entities
People
- K. Ritter
Organizations
- University of Wisconsin–Madison