Core Concepts
This paper presents several new linearization techniques for quadratic binary optimization problems (QBOPs) using selective aggregation of constraints. The proposed models provide strong linear programming relaxations and have practical value.
Abstract
The paper focuses on the linearization of quadratic binary optimization problems (QBOPs). It starts with a review of existing explicit linearization models, including the Dantzig-Watters (DW), Glover-Woolsey (GW), and Fortet (FT) formulations. The authors then introduce a new basic linearization model called PK.
The key contributions of the paper are:
Weighted aggregation of constraints: The authors present a general technique to generate new explicit linearizations of QBOPs by using weighted aggregation of selected constraints from the basic models. This includes aggregation of both type 1 (product constraints) and type 2 (auxiliary constraints) linearization constraints.
Precise and optimality-restricted models: The authors distinguish between precise models, where the linearized variables precisely represent the product of binary variables, and optimality-restricted models, where the linearized variables only guarantee the correct product at optimality. Both types of models are analyzed.
Theoretical and experimental analysis: The paper provides a detailed theoretical analysis of the proposed linearization techniques, including proofs of validity for the new models. Extensive computational experiments are also reported, offering insights into the relative merits of the different explicit linearization approaches.
Compact explicit linearizations: The authors present new explicit linearization models that match the number of constraints in many compact formulations, while still providing strong LP relaxation bounds.
The paper makes significant contributions to the understanding and development of efficient explicit linearization techniques for QBOPs, which have important applications in various optimization problems.