Computational results obtained when solving a subset of NETLIB problems by using a dense projected gradient implementation of the non-simplex active-set sagitta method presented in [12] are summarized. Two different addition rules for its initial phase are considered and, for each problem solved, two corresponding graphs are reported to illustrate the variations of the objective value along the active-set path. The comparison of our code for the sagitta method versus MATLAB code linprog shows that this sagitta method outperforms the simplex method in number of iterations and reliability and can be competitive in overall speed.
Citation
MA-05-05, Department of Applied Mathematics, University of Málaga, 29071 Málaga, Spain September/2005