A feasible trust-region algorithm for inequality constrained optimization (Q2489190)

For solving smooth nonlinearly ineqality constrained optimization problems an globally convergent algorithm is developed, in which a sequence of feasible iterates is generated by a trust-region sequential quadratic programming subproblem at each iteration. Many practical trust-region methods do not insist on satisfaction of the constraints of subproblem, but achieve some reduction in the infeasibility, while staying within the trust-region and reducing the objective function in the subproblem. In this paper a feasible trust-region algorithm is proposed, in which all iterates are feasible. Firstly, it is shown, if the algorithm does not terminate finitely, it has a limit point which meets Karush-Kuhn-Tucker (KKT) conditions. Secondly, under a stronger assumption on the approximate Hessian, it is stated that all limit points satisfy KKT conditions. Some results of computational experiments showing the efficiency and robustness of the proposed method are given.