Sufficient descent nonlinear conjugate gradient methods with conjugacy condition (Q849150)

The authors consider unconstrained optimization problems with a continuously differentiable objective function \(f: \mathbb{R}^n\to\mathbb{R}\). A class of modified conjugate gradient methods is proposed for solving the problems. The methods in this class have a common property that the direction \(d_k\) generated at iteration \(k\) and corresponding gradient \(g_k\) of function \(f\) satisfy the equality \(g_k^Tdk= -\| g_k\|^2\). Global convergence for modified methods \(YT\) and \(YT\)+, which belong to the proposed class of methods, is proved under suitable conditions. Extensive numerical experiments show the efficiency of the proposed methods. The numerical experiments are carried out using test problems from CUTE library.