Modified subspace limited memory BFGS algorithm for large-scale bound constrained optimization (Q955062)

This method is based on modification to the one proposed by \textit{Q. Ni} and \textit{Y. X. Yuan} [Math. Comput. 66, No.~220, 1509--1520 (1997; Zbl 0886.65065)]. A new parameter is used to estimate the set of active or inactive variables in every step. This parameter automatically becomes zero as the iterations progress. The global convergence is established under suitable conditions. The advantages of the Broyden-Fletcher-Goldfarb-Shanna (BFGS) method are preserved in the proposed method. It is pointed out that numerical experiments on a set of large problems are promising.