Title: Combining RLT and SDP for nonconvex quadratic programming

Abstract: We consider relaxations for nonconvex quadratic programming based 
on combinations of the Reformulation-Linearization Technique (RLT) and 
Semidefinite Programming (SDP). For pairs of variables that are not near 
their upper or lower bounds, adding the semidefinite constraint provides a 
large reduction in the feasible region for the corresponding product 
variables in the RLT relaxation. We give computational results on nonconvex 
box-constrained and quadratically constrained problems. We also illustrate 
the effect of adding order constraints to RLT and SDP relaxations to 
improve bounds on highly symmetric problems.