Multiple image gravitational lens systems, and especially quads are invaluable in determining the amount and distribution of mass in galaxies. This is usually done by mass modeling using parametric or free-form methods. An alternative way of extracting information about lens mass distribution is to use lensing degeneracies and invariants. Where applicable, they allow one to make conclusions about whole classes of lenses without model fitting. Here, we use approximate, but observationally useful invariants formed by the three relative polar angles of quad images around the lens center to show that many smooth elliptical+shear lenses can reproduce the same set of quad image angles within observational error. This result allows us to show in a model-free way what the general class of smooth elliptical+shear lenses looks like in the three dimensional (3D) space of image relative angles, and that this distribution does not match that of the observed quads. We conclude that, even though smooth elliptical+shear lenses can reproduce individual quads, they cannot reproduce the quad population. What is likely needed is substructure, with clump masses larger than those responsible for flux ratio anomalies in quads, or luminous or dark nearby perturber galaxies.

In a quadruply imaged lens system the angular distribution of images around the lens center is completely described by three relative angles. We show empirically that in the 3D space of these angles, spanning 180 x 180 x 90 degrees, quads from simple two-fold symmetric lenses of arbitrary radial density profile and arbitrary radially dependent ellipticity or external shear define a nearly invariant 2D surface. We give a fitting formula for the surface using SIS+elliptical lensing potential. Various circularly symmetric mass distributions with shear up to 0.4 deviate from it by typically, rms~0.1 deg, while elliptical mass distributions with ellipticity of up 0.4 deviate from it by rms~1.5 deg. The existence of a near invariant surface gives a new insight into the lensing theory and provides a framework for studying quads. It also allows one to gain information about the lens mass distribution from the image positions alone, without any recourse to mass modeling. As an illustration, we show that about 3/4 of observed galaxy-lens quads do not belong to this surface within observational error, and so require additional external shear or substructure to be modeled adequately.