Combining RLT and SDP for nonconvex QCQP
We consider applications of the reformulation-linearization
technique (RLT) and semidefinite programming (SDP) to obtain bounds on
nonconvex quadratically constrained quadratic programming (QCQP)
problems. From a theoretical standpoint we show that adding the
semidefiniteness condition provides a substantial reduction in the
feasible set corresponding to products of variables in an RLT
relaxation. We also consider computational applications of RLT and SDP
to several different classes of QCQPs, and find that bounds obtained by
combining the two techniques can be substantially stronger than those
obtained using either approach alone.
Chemnitz Workshop
Last modified: Tue Sep 28 8:07:20 CEST 2004