Variable metric gradient projection processes in convex feasible sets defined by nonlinear inequalities (Q1101344)

Optimization problems are considered which are defined on a Hilbert space with finitely many inequality constraints. The authors analyze the properties of gradient projection methods, where the gradient is computed with regard to an inner product depending on the iterate. The step-size rule is of the Armijo-Goldstein type. Conditions on nonsingularity of minimal points are given in the context of second order sufficient optimality conditions. These imply certain growth conditions which allow the proof of convergence and a linear rate of convergence.