Minimizing quadratic functions with semidefinite Hessian subject to bound constraints

Abstract

The MPRGP (modified proportioning with reduced gradient projections) algorithm for minimization of the strictly convex quadratic function subject to bound constraints is adapted to the solution of problems with a semidefinite Hessian A. The adapted algorithm accepts the decrease directions that belong to the null space of A and generates the iterates that are proved to minimize the cost function. The paper examines specific features of the solution of the problems with convex, but not necessarily strictly convex Hessian. The performance of the algorithm is demonstrated by the solution of a semi-coercive contact problem of elasticity and a 3D particle dynamics problem. The results are compared with those obtained by the spectral projected gradient method and the projected-Jacobi method.