zbMATH — the first resource for mathematics

On the maximization of a concave quadratic function with box constraints. (English) Zbl 0801.65058
The maximization of a convex quadratic function, subject to bounds on variables, is considered. The proposed algorithm combines the classical active set and gradient projection methods as many other algorithms do, including the algorithm of J. J. Moré and G. Toraldo [SIAM J. Optim. 1, No. 1, 93-113 (1991; Zbl 0752.90053)]. However, the algorithm uses a new strategy for the decision to leave the current face, making it possible to obtain a finite convergence even for a singular Hessian and in the presence of dual degeneracy. The algorithm, tested using various test problems, seems to perform well in comparison with other existing algorithms.

65K05 Numerical mathematical programming methods
90C20 Quadratic programming
90C25 Convex programming
Zbl 0752.90053
Full Text: DOI