We study the so-called Celis-Dennis-Tapia (CDT) problem to minimize a non-convex quadratic function over the intersection of two ellipsoids. Contrasting with the well-studied trust region problem where the feasible set is just one ellipsoid, the CDT problem seems to be not yet fully understood. Our main objective in this paper is to narrow the difficulty gap defined by curvature of the Lagrangian. We propose apparently novel sufficient and necessary conditions for global optimality and hint at algorithmic possibilities to exploit these.