An unconstrained optimization method using nonmonotone second order Goldstein's line search (Q2464310): Difference between revisions

The authors consider the problem of unconstrained minimization of a real-valued twice continuously differentiable function of \(n\) variables. They propose a new unconstrained optimization method using the non-monotone second-order Goldstein linear search, which is different from both the nonmonotone Goldstein step size rule and the curvilinear line search rule. Under mild conditions, the sequence generated by the proposed algorithm is proved to converge to a stationary point satisfying the second-order optimality conditions. Numerical tests presented in the concluding part of the paper show a good numerical behavior of the method in comparison with some other methods from the literature. The method is efficient and robust especially for some ill-conditioned problems.