Screening of Coulomb interactions in transition metals

We discuss different methods of calculation of the screened Coulomb interaction U in transition metals and compare the so-called constraint local-density approximation (LDA) with the GW approach. We clarify that they offer complementary methods of treating the screening and therefore should serve for different purposes. The analysis is illustrated by calculations for the ferromagnetic Ni. In the ab initio GW method, the renormalization of bare on-site Coulomb interactions between 3d electrons (being of the order of 20--30 eV) occurs mainly through the screening by the same 3d electrons, treated in the random-phase approximation (RPA). The basic difference of the constraint-LDA method from the GW method is that it deals with the neutral processes, where the Coulomb interactions are additionally screened by the ``excited'' electron, since it continues to stay in the system. This is the main channel of screening by the itinerant (4sp) electrons, which is especially strong in the case of transition metals and missing in the GW approach, although the details of this screening may be affected by additional approximations, which typically supplement these two methods. The major drawback of the conventional constraint-LDA method is that it does not allow us to treat the energy dependence of U, while the full GW calculations require heavy computations. We propose a promising approximation based on the combination of these two methods. First, we take into account the screening of Coulomb interactions in the 3d-electron-like bands located near the Fermi level by the states from the orthogonal subspace, using the constraint-LDA methods. The obtained interactions are further renormalized within the bands near the Fermi level in RPA. This allows the energy-dependent screening by electrons located near the Fermi level, including the same 3d electrons.