Modified nonmonotone Armijo line search for descent method (Q535246)

The authors analyse a matrix-free non-monotone Armijo line search approach for descent methods. For the global convergence of the proposed method purpose, it is assumed that the gradient of the objective function is Lipschitz continuous. For the derivation of the convergence rate, the authors require additionally that the objective function is uniformly convex. Since an estimate of the Lipschitz constant is used in the considered line search the authors discuss some alternatives to determine such an estimate. The section on the numerical results presents a performance analysis of the new method for several well established test cases.