Building Improved Directions of Negative Curvature for Constrained Optimization
In this work, we provide an approach to the computation of improved directions of negative curvature for nonlinearly constrained optimization problems. In particular, we focus on the use of low cost procedures to improve directions of negative curvature obtained from a direct factorization of the Hessian matrix of the objective function. The key feature is that the directions are computed within the null subspace of the Jacobian matrix of the constraints. In addition, we show how to include the above procedure within an interior-point algorithm for constrained optimization.Finally, some numerical experiments showing the successful performance of our proposal are presented.
This is a joint work with Javier Cano and Francisco J. Prieto.