The semidefinite (SDP) relaxation of a quadratically constrained quadratic program (QCQP) is called exact if it has a rank-1 optimal solution corresponding to a QCQP optimal solution. Given an arbitrary QCQP whose SDP relaxation is exact, this paper investigates incorporating additional quadratic inequality constraints while maintaining the exactness of the SDP relaxation of the resulting QCQP. Three important classes of QCQPs with exact SDP relaxations include (a) those characterized by rank-one generated cones, (b) those by convexity, and (c) those by the sign pattern of the data coefficient matrices. These classes have been studied independently until now. By adding quadratic inequality constraints satisfying the proposed conditions to QCQPs in these classes, we extend the exact SDP relaxation to broader classes of QCQPs. Illustrative QCQP instances are provided.
Kojima et al. (2025) studied this question.