The separable and non-separable formulations of convex quadratic problems in interior point methods

Csaba Meszaros

Quadratic programming (QP) problems are usually formulated in the general non-separable form. For convex QP problems, however, it is always possible to separate the quadratic function by reformulating the original problem. The two formulations present equvivalent optimization problems but may influence the behavior and efficiency of the optimization techniques applied. In the paper the interior point methods for large--scale convex quadratic programming are concerned. The target of our investigation is to compare how the two formulations of convex QPs influence the computational work of interior point methods and answer the question which formulation is to be used in general.

WP 98-3, Laboratory of Operations Research and Decision Systems, Hungarian Academy of Sciences