Key points are not available for this paper at this time.
We propose a systematic procedure for constructing effective lattice fermion models for narrow-band compounds on the basis of first-principles electronic-structure calculations. The method is illustrated for the series of transition-metal (TM) oxides: SrVO₃, YTiO₃, V₂O₃, and Y₂Mo₂O₇, whose low-energy properties are linked exclusively to the electronic structure of an isolated t₂₆ band. The method consists of three parts, starting from the electronic structure in the local-density approximation (LDA). (i) Construction of the kinetic-energy Hamiltonian using formal downfolding method. It allows us to describe the band structure close to the Fermi level in terms of a limited number of (unknown yet) Wannier functions (WFs), and eliminate the rest of the basis states. (ii) Solution of an inverse problem and construction of WFs for the given kinetic-energy Hamiltonian. Here, we closely follow the construction of the basis functions in the liner-muffin-tin-orbital (LMTO) method, and enforce the orthogonality of WFs to other bands. In this approach, one can easily control the contributions of the kinetic-energy term to the WFs. (iii) Calculation of screened Coulomb interactions in the basis of auxiliary WFs. The latter are defined as the WFs for which the kinetic-energy term is set to be zero. Meanwhile, the hybridization between TM d and other atomic states is preserved by the orthogonality condition to other bands. The use of auxiliary WFs is necessary in order to avoid the double counting of the kinetic-energy term, which is included explicitly in the model Hamiltonian. In order to calculate the screened Coulomb interactions we employed a hybrid approach. First, we evaluate the screening caused by the change of occupation numbers and the relaxation of the LMTO basis functions, using the conventional constraint-LDA approach, where all matrix elements of hybridization connecting the TM d orbitals and other orbitals are set to be zero. Then, we switch on the hybridization and evaluate the screening of on-site Coulomb interactions associated with the change of this hybridization in the random-phase approximation. The second channel of screening appears to be very important, and results in relatively small values of effective Coulomb interactions in the t₂₆ band (of the order of 2--30. 3em{0ex}eV, depending on the material). We discuss details of this screening and consider its band-filling dependence, frequency dependence, influence of the lattice distortion, proximity of other bands, as well as the effect of dimensionality of the model Hamiltonian.
I. V. Solovyev (Tue,) studied this question.