A Non-edge Following Method for Solving Linear Programs

선형계산문제의 비정변형해법의 연구

In this paper, we propose a non-edge following method for linear programs. Unlike alledged poor performance of algorithms of this type, this method performs well at least with 25 randomly generated problems. This method is comparable to Rosen's gradient projection method as applied to the dual formulation. The latter is of general purpose, and no implementation rules are available for linear program applications. This paper suggests ways of finding improving dual feasible directions, and of allowing to move across the extreme faces of a higher dimension polyhedron. Rather simple computational rules are provided for projection operations needed at each iteration.