Comparação empírica de algoritmos de busca linear inexata em otimização não linear irrestrita

Abstract

Inexact linear search emerges as a crucial strategy for nonlinear optimization, being fundamental in determining step size in descent direction methods. This work implements and compares the Armijo, Wolfe, and Goldstein algorithms in different optimization scenarios. Results obtained through benchmark function tests demonstrated the superiority of the Wolfe algorithm in terms of general consistency, while Goldstein stood out in specific multimodal functions. The experiments showed that the effectiveness of each method is intrinsically linked to the characteristics of the objective function, providing important guidelines for selecting inexact linear search algorithms in practical optimization problems