A regularized limited memory subspace minimization conjugate gradient method for unconstrained optimization (Q6141538)

The authors present a regularized limited memory subspace minimization conjugate gradient method containing two types of iteration with a modified regularized quasi-Newton method in a small dimensional subspace to correct the orthogonality, an improved initial step size selection strategy and some simple acceleration criteria. The global convergence of the proposed algorithm is established by utilizing generalized nonmonotone Wolfe line search under certain assumptions. The efficiency of the proposed method in comparison to other algorithms is illustrated by some numerical experiments.