We present a convex conic relaxation for a problem of maximizing an indefinite quadratic form over a set of convex constraints on the squared variables. We show that for all these problems we get at least 12/37-relative accuracy of the approximation. In the second part of the paper we derive the conic relaxation by another approach based on the second order optimality conditions. We show that for lp-balls, p>=2, intersected by a linear subspace, it is possible to guarantee (1- 2/p)-relative accuracy of the solution. As a consequence, we prove (1 - 1/eln-n)-relative accuracy of the conic relaxation for an indefinite quadratic maximization problem over an n-dimensional unit box with homogeneous linear equality constraints. We discuss the implications of the results for the discussion around the question P = NP.
[1]
Yurii Nesterov,et al.
Interior-point polynomial algorithms in convex programming
,
1994,
Siam studies in applied mathematics.
[2]
David P. Williamson,et al.
Improved approximation algorithms for maximum cut and satisfiability problems using semidefinite programming
,
1995,
JACM.
[3]
Mihir Bellare,et al.
The complexity of approximating a nonlinear program
,
1995,
Math. Program..
[4]
Y. Ye.
Approximating Quadratic Programming With Bound Constraints
,
1997
.
[5]
Y. Nesterov.
Quality of semidefinite relaxation for nonconvex quadratic optimization
,
1997
.
[6]
Y. Nesterov.
Semidefinite relaxation and nonconvex quadratic optimization
,
1998
.