Generating cutting planes through inequality merging for integer programming problems

Abstract

Integer Programming (IP) problems are a common type of optimization problem used to solve numerous real world problems. IPs can require exponential computational effort to solve using the branch and bound technique. A popular method to improve solution times is to generate valid inequalities that serve as cutting planes.

This dissertation introduces a new category of cutting planes for general IPs called inequality merging. The inequality merging technique combines two or more low dimensional inequalities, yielding valid inequalities of potentially higher dimension. The dissertation describes several theoretical results of merged inequalities.

This research applies merging inequalities to a frequently used class of IPs called multiple knapsack (MK) problems. Theoretical results related to merging cover inequalities are presented. These results include: conditions for validity, conditions for facet defining inequalities, merging simultaneously over multiple cover inequalities, sequentially merging several cover inequalities on multiple variables, and algorithms that facilitate the development of merged inequalities. Examples demonstrate each of the theoretical discoveries.

A computational study experiments with inequality merging techniques using benchmark MK instances. This computational study provides recommendations for implementing merged inequalities, which results in an average decrease of about 9% in computational time for both small and large MK instances. The research validates the effectiveness of using merged inequalities for MK problems and motivates substantial theoretical and computational extensions as future research.