Bound reduction using pairs of linear inequalities

We describe a procedure to reduce variable bounds in Mixed Integer Nonlinear Programming (MINLP) as well as Mixed Integer Linear Programming (MILP) problems. The procedure works by combining pairs of inequalities of a linear programming (LP) relaxation of the problem. This bound reduction technique extends the implied bounds procedure used in MINLP and MILP and can be seen as a special case of optimality based bound reduction, a method to infer variable bounds from the LP relaxation. For an LP relaxation with m constraints and n variables, the number of pairs of constraints is O(m^2), while a naive implementation of our proposed bound reduction scheme is O(n^3) for each pair. Therefore, its overall complexity is O(m^2 n^3), which can be prohibitive for relatively large problems. To this purpose, we have developed a more efficient procedure that has complexity O(m^2 n^2), and embedded it in two Open-Source solvers: one for MINLP and one for MILP. We provide computational results which substantiate the utility of this bound reduction technique for several instances.



View Bound reduction using pairs of linear inequalities