Nonmonotone algorithm for minimization on closed sets with applications to minimization on Stiefel manifolds (Q408215)

The authors propose an algorithm for minimizing differentiable functions restricted to an arbitrary closed set by combining a regularization approach and nonmonotone trust-region rules. The method generates a feasible sequence such that all of its accumulation points are stationary points irrespective of the initial approximation chosen. The global convergence of the algorithm is established. A globally convergent nonmonotone gradient-like scheme for minimization on Stiefel manifolds is proposed. Some numerical results on small and medium sized weighted orthogonal Procrustes problems are presented to illustrate the technique applied to minimization problems with orthogonality constraints.