We solve a long standing problem with relativistic calculations done with the widely used multiconfiguration Dirac-Fock method. We show, using relativistic many-body perturbation theory (RMBPT), how, even for relatively high-Z, relaxation or correlation causes the nonrelativistic limit of states of different total angular momentum but identical orbital angular momentum to have different energies. We show that only large scale calculations that include all single excitations, even those obeying Brillouin's theorem, have the correct limit. We reproduce very accurately recent high-precision measurements in F-like Ar, and turn then to a precise test of QED. We obtain the correct nonrelativistic limit not only for fine structure but also for level energies and show that RMBPT calculations are not immune to this problem.