An Active Set Trust-Region Method for Bound-Constrained Optimization

Abstract This paper discusses an active set trust-region algorithm for bound-constrained optimization problems. A sufficient descent condition is used as a computational measure to identify whether the function value reduced or not. To get our complexity result, critical which computationally better than other known measures. Under positive definiteness of approximated Hessian matrices restricted subspace non-active variables, it will be shown that unlimited zigzagging cannot occur. It competitive in comparison with state-of-the-art solvers solving ill-conditioned least-squares problem.