A globally and R-linearly convergent hybrid HS and PRP method and its inexact version with applications (Q326018)

This paper proposes a sufficient descent hybrid method to improve the computational efficiency and convergence properties of conjugate gradient methods in solving unconstrained optimization problems of general nonlinear functions. The proposed algorithms are based on the three-term Hestenes-Stiefel method and the three-term Polak-Ribiére-Polyak method. The global and R-linear convergence of the proposed method with a descent backtracking line search for nonconvex optimization is established. An inexact conjugate gradient method by using approximate gradient and function values is proposed, and its global convergence by the use of some approximate function-value descent line search is shown. The inexact algorithm is applied to the solution of nonsmooth convex optimization problems by using the Moreau-Yosida regularization.