Comparison of the Computational Performance of Three Quadratic Programming Algorithms

Abstract

The main objective of this study is to compare the computational perforinance of three quadratic programming algorithms. A quadratic programming problem is one in which the objective function to be minimized is quadratic and the constraint functions are linear. The three algorithms are Wolfe's reduced gradient method (implemented in the MINOS package), Lemke's complementary pivot method, and Fletcher's active set method. Fletcher's method was shown to be superior to the other two methods. In this paper, a random-problems generator is used. In addition, a translator program has been written which tranforms a given input data into MPS and SPECS files which are needed for the MINOS package. In a recent study, it was shown that Lemke's algorithm terminated with an infeasible solution in a convex quadratic programming problem. 'Ibis claim was investigated to know the reason for such an abnormal behavior. 'Ibis investigation is a secondary objective of the study.