Corrected sequential linear programming for sparse minimax optimization (Q1338528)

An algorithm to solve the large scale nonlinear minimax problem is presented. The proposed method is first-order: Hessian matrices are not calculated. A key feature of the proposed algorithm is the use of a ``corrected'' or ``vertical'' step. A global convergence result is presented; a small collection of computational experiments and comparisons is discussed. Finally, the authors point out that nonlinear inequality constrained optimization problems can be phrased as nonlinear minimax problems. Therefore, this proposed algorithm can be used in this general setting as well.