Unboundedness in Integer and Discrete Programming L.P. Relaxations

Commercial branch and bound codes for solving the general mixed integer linear programming problem commence by solving the linear programming relaxation of the submitted problem, terminating if the relaxation is unbounded. It is assumed that the submitted problem is either unbounded or has no feasible solutions. It is shown that the assumption is correct for all integer programming problems which can be submitted to the currently available codes (though counter examples which cannot be so submitted are given), but that the assumption is generally incorrect for discrete linear programming problems (using for example the special ordered set construct). Sufficient conditions on formulations to ensure its correctness are given. One possible formulation approach, applicable to special ordered set situations, is discussed.