A new SOCP relaxation of nonconvex quadratic programming problems with a few negative eigenvalues
Introduction
In this paper, we consider a nonconvex generalized quadratic programming problem with convex quadratic constraints in the following form:
is an symmetric indefinite matrix with negative eigenvalues, is a positive semidefinite matrix for , and . The feasible region of (1.1) is supposed to be bounded with nonempty relative interior points. Problem (1.1) covers many important engineering problems and combinatorial optimization problems [1], [2], [3]. In this paper, we focus on the (1.1) with a few negative eigenvalues (). In fact, several well-known problems, such as the concave quadratic programming problems [4], the low rank DC programs [5], [6], the optimal spectrum sharing problems in MIMO cognitive radio networks with multiple-input multiple-output(MIMO) links [7] and the two-trust-region subproblems [8], could be recast into the model of (1.1). (1.1) is NP-hard even with one negative eigenvalue [9]. Cambini et al. [10] stated some theoretical properties and characterized the existence of minimizers of DC programs with one negative eigenvalue and no quadratic constraints. They [5], [6] also proposed global algorithms for DC programs with only a few negative eigenvalues. As a special nonconvex quadratically constrained quadratic programming problem, many researches have paid more attention to the design of convex relaxations of (1.1). In the past few decades, semidefinite programming (SDP) relaxation has been regarded as an attractive technique for solving (1.1) [11], [12]. However, though the SDP relaxation could provide a good lower bound, the computational complexity is high, especially for large scale problems due to the fact that the SDP relaxation enlarges the dimension of the problem by transforming the original -dimensional variable vector to an variable matrix. Therefore, Luo et al. [13] used the classical SOCP relaxation to achieve a lower bound for (1.1) when designing a branch and bound algorithm and they showed that the SOCP relaxation is effective if the number of negative eigenvalues is few.
Note that the convex relaxation effect is critical when designing a global algorithm, thus, this paper aims to present a new SOCP relaxation of (1.1) via a simultaneous diagonalization tool in order to enhance the bound quality of the classical SOCP relaxation. Ben-Tal et al. [14] first introduced the simultaneous diagonalization tool to solve the quadratic programs with one or two parallel quadratic constraints when the objective and constraint matrices are simultaneously diagonalized. They proved that the simultaneous diagonalization based SOCP relaxation is tight since the auxiliary variables could be bounded from above by one convex constraint. Then this tool has been effectively extended to solve the convex quadratic programs with linear complementary constraints and the generalized trust-region problems [15], [16]. All these researches illustrated that the approach of applying the simultaneous diagonalization tool into the SOCP relaxation could improve the relaxation effect more often than not. In this paper, we first add a redundant constraint into (1.1). Then the objective matrix are decomposed to the subtraction between two semidefinite matrices according to the signs of its eigenvalues. Afterwards, we find a nonsingular matrix such that and the redundant constraint matrix are simultaneously diagonalizable. Finally, we design a new SOCP relaxation of (1.1). It is worth mentioning that the auxiliary variables are bounded by one more convex quadratic constraint in the proposed SOCP relaxation than that in the classical SOCP relaxation, thus, the proposed SOCP relaxation offers a better lower bound. Moreover, it can be proved that the proposed SOCP relaxation is strictly tighter than the classical SOCP relaxation when is positive definite.
The rest of the paper is organized as follows. In Section 2, we design a simultaneous diagonalization based SOCP relaxation of (1.1) and make comparisons with the classical SOCP relaxation. In Section 3, we carry out numerical experiments to show the effectiveness of the proposed SOCP relaxation both in the relaxation effect and computing time. Conclusions are given in Section 4.
Some notations are adopted throughout this paper. represents a real symmetric matrix that is positive semidefinite. Given a vector , indicates an diagonal matrix with its diagonal elements equal to .
Section snippets
A simultaneous diagonalization based SOCP relaxation
We first give a brief introduction of the classical SOCP relaxation [17] for (1.1) in this section. Let denote the number of negative eigenvalues of . Without loss of generality, we assume the first eigenvalues of are negative. We decompose into , where and , in which is an eigenvalue and is the corresponding eigenvector of for . By introducing auxiliary variables , and relaxing into for , the
Numerical experiments
In this section, we compare the proposed SOCP relaxation with the classical SOCP relaxation. Numerical examples are carried out in MATLAB R2013b on a PC (2.50 GHz, 8 GB RAM) and the SOCP relaxations are solved by Cplex solver 12.6.3. The lower bound and its CPU time (in secs) are displayed for each relaxation in Table 1, Table 2, Table 3, Table 4. For convenience, we refer to the (1.1) with RLT constraints as the classical SOCP relaxation and the (2.3) with RLT constraints as the proposed SOCP
Conclusion
This paper derives a new SOCP relaxation for convex quadratically constrained nonconvex quadratic programs with a few negative eigenvalues. The new relaxation is proved to provide a better lower bound than the classical SOCP relaxation and three types of examples are tested to verify the conclusion. Moreover, when the number of variable dimension is larger than 250, the advantage is amplified both in the lower bound and computing time. Therefore, the DC decomposition and simultaneous
Acknowledgments
Zhou’s research has been supported by the National Natural Science Foundation of China (Grant #11701512), Xu’s research has been supported by the National Natural Science Foundation of China (Grants #11704336), and Wang’s research has been supported by the Youth Innovation Team of Capital University of Economics and Business .
References (20)
- et al.
Global optimization algorithm for mixed integer quadratically constrained quadratic program
J. Comput. Appl. Math.
(2017) - et al.
A branch and reduce approach for solving a class of low rank dc programs
J. Comput. Appl. Math.
(2009) - et al.
Solving a class of low rank dc programs via a branch and bound approach: A computational experience
Oper. Res. Lett.
(2010) - et al.
Binary partitioning, perceptual grouping, and restoration with semidefnite programming
IEEE Trans. Pattern Anal. Mach. Intell.
(2003) - et al.
A new rank-two semidefnite programming relaxation method for multiuser detection problem
Wirel. Pers. Commun.
(2012) A new algorithm for concave quadratic programming
J. Global Optim.
(2019)- et al.
Optimal spectrum sharing in MIMO cognitive radio networks via semidefinite programming
J. Sel. Areas Commun.
(2011) - et al.
A computational study of global optimization solvers on two trust region subproblems
J. Global Optim.
(2018) - et al.
Quadratic programming with one negative eigenvalue is NP-hard
J. Global Optim.
(1991) - et al.
A finite algorithm for a particular D.C quadratic programming problem
Ann. Oper. Res.
(2002)