The most-obtuse-angle row pivot rule for achieving dual feasibility: A computational study

We recently proposed several new pivot rules for achieving dual feasibility in linear programming, which are distinct from existing ones: the objective function value will no longer change necessarily monotonically in their solution process. In this paper, we report our further computational testing with one of them, the most-obtuse-angle rule. A two-phase dual simplex algorithm, in which the rule is used as a row selction rule for Phase-1, has been implemented in FORTRAN 77 modules, and was tested on a set of standard linear programming problems from NETLIB. We show that, if full pricing is applied, our code unambiguously will outperform MINOS 5.3, one of the best implementations of the simplex algorithm at present.