New quasi-Newton methods for unconstrained optimization problems (Q2493696)

Some new quasi-Newton methods for unconstrained minimization based on a new quasi-Newton equation are proposed. Quasi-Newton methods use only first derivatives to build an approximate Hessian over a number of iterations. This approximation is updated in each iteration by a matrix of low rank. The authors give two choices of matrices which are included in the quasi-Newton equation. Three corresponding Broyden-Fletcher-Goldfarb-Shanno type algorithms are proved to possess the global convergence property. The superlinear convergence of the one algorithm is proved. The convergence results and the computational experiment show that the approach given in this paper is very successful.