The multiplicative penalty function method for linear
programming is extended in several ways.  An elegant duality
theorem between primal and dual trajectories is shown.