A preconditioner for solving linear programming problems with dense columns

The Interior-Point Methods are a class for solving linear programming problems that rely upon the solution of linear systems. At each iteration, it becomes important to determine how to solve these linear systems when the constraint matrix of the linear programming problem includes dense columns. In this paper, we propose a preconditioner to handle linear programming problems with dense columns, and we prove theoretically that the final linear system to solve is uniformly bounded when the Interior-Point Method is converging to an optimal solution. This result is illustrated through computational experiments, which show that our proposed method is robust and competitive in terms of running time and/or number of iterations compared with existing methods.