A new subspace limited memory BFGS algorithm for large-scale bound constrained optimization (Q870176): Difference between revisions

A new algorithm that combines an active set strategy with the gradient projection method is presented. As in the work by \textit{F. Facchinei, S. Lucidi} and \textit{L. Palagi} [SIAM J. Opt., 12, 1100--1125 (2002; Zbl 1035.90103)] the authors avoid the necessity of finding an exact minimizer of a quadratic subproblem with bound constraints. The algorithm has the following properties: All iterates are feasible and the sequence of the objective function values is decreasing; rapid changes in the active set are allowed; a global convergence theory is established. Moreover, it reserves the advantage of the effective active set identified technique by Facchinei, Lucidi and Palagi [loc. cit.] and uses the superiority of the subspace limited memory Broyden-Fletcher-Goldfarb-Shanno (BFGS) method [see \textit{Q. Ni} and \textit{Y. X. Yuan}, Math. Comp., 66, 1509--1520 (1997; Zbl 0886.65065)] which has been proved much suit for solving large-scale problems. Namely, the active sets are based on a guessing technique to be identified at each iteration, the search direction in the free subspace is determined by a limited memory BFGS algorithm, which provides an efficient means for attacking large-scale optimizatuin problems. The implementations of the method on CUTE test problems are described.