Conjugate gradient methods based on secant conditions that generate descent search directions for unconstrained optimization (Q442712): Difference between revisions

Conjugate gradient methods are very efficient for solving large-scale unconstrained optimization problems. The main advantage of these methods is that no matrices are used which highly decreases storage requirements. When computing a search direction, a lot of various choices for the parameter \(\beta\) (characterizing the conjugate gradient method) and their modifications exist. In order to incorporate the second-order information of the objective function into conjugate gradient methods, many variants based on secant conditions have been proposed. Although such methods have global convergence property, they do not necessarily satisfy the (sufficient) descent condition. In this paper, the authors propose new efficient conjugate gradient methods for solving large-scale unconstrained optimization problems that generate descent search directions and are globally convergent for general objective functions. The methods combine ideas based on the secant condition proposed by \textit{Y.-H. Dai} and \textit{L.-Z. Liao} [Appl. Math. Optim. 43, No. 1, 87--101 (2001; Zbl 0973.65050)] and the formula for \(\beta\) proposed by\textit{W. W. Hager} and \textit{H. Zhang} [SIAM J. Optim. 16, No. 1, 170--192 (2005; Zbl 1093.90085)].