Quadratic convex reformulation for nonconvex binary quadratically constrained quadratic programming via surrogate constraint
We investigate in this paper nonconvex binary quadratically constrained quadratic programming (QCQP) which arises in various real-life fields. We propose a novel approach of getting quadratic convex reformulation (QCR) for this class of optimization problem. Our approach employs quadratic surrogate functions and convexifies all the quadratic inequality constraints to construct QCR. The price of this approach is the introduction of an extra quadratic inequality. The “best” QCR among the proposed family, in terms that the bound of the corresponding continuous relaxation is best, can be found via solving a semidefinite programming problem. Furthermore, we prove that the bound obtained by continuous relaxation of our best QCR is as tight as Lagrangian bound of binary QCQP. Computational experiment is also conducted to illustrate the solution efficiency improvement of our best QCR when applied in off-the-shell software.
KeywordsBinary QCQP Semidefinite programming Quadratic convex reformulation Global optimization
The authors would like to thank three anonymous referees for their constructive suggestions and insightful comments, which helped improve the paper substantially.
- 9.Burer, S., Saxena, A.: Old wine in new bottle: the milp road to miqcp. Technical Report, Department of Management Sciences University of Iowa (2009). http://www.optimization-online.org/DB_FILE/2009/07/2338.pdf
- 14.Grant, M., Boyd, S.: CVX: Matlab software for disciplined convex programming, version 2.1, (2014). http://cvxr.com/cvx
- 18.Kolbert, F., Wormald, L.: Robust portfolio optimization using second-order cone programming. In: Satchell, S. (ed.) Optimizing Optimization: The Next Generation of Optimization Applications & Theory, pp. 3–22. Academic Press and Elsevier, Amsterdam (2010)Google Scholar