Large Steps Discrete Newton Methods for Minimizing Quasiconvex Functions : Notas de Matemática, 53

This paper develops an idea of H. D. Scolnik for minimizing quasiconvex functions concerning the use of "large steps" based upon geometrical properties of the objective function derived from its level sets. Basically speaking, the increment will be computed by means of auxiliary points suc...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Echebest, Nélida Ester, Guardarucci, María Teresa, Scolnik, Hugo Daniel, Vacchino, María Cristina
Formato: Publicacion seriada
Lenguaje:Inglés
Publicado: 1993
Materias:
Acceso en línea:http://sedici.unlp.edu.ar/handle/10915/170315
Aporte de:
Descripción
Sumario:This paper develops an idea of H. D. Scolnik for minimizing quasiconvex functions concerning the use of "large steps" based upon geometrical properties of the objective function derived from its level sets. Basically speaking, the increment will be computed by means of auxiliary points such that f(xk + hp p) = f(xk) where f(x) is a quasiconvex function and p is an element of a set of directions chosen according to an algorithm which guarantees global and local quadratic convergence. The search direction in each iteration is obtained as the solution of a finite differences system of equations involving the auxiliary directions p and the increments hp , and is Newton's when the function is quadratic. Computational tests are given showing the efficiency of the new algorithm.