Dear Yang Ding, care is needed when estimating the crystal field splitting from the LDA+U calculation using the DOS. There are two reasons for it. First, the LDA+U lower the energy of more occupied states and increase the energy of less occupied states. Even if the bands are above Fermi energy, they contain nonzero fraction of electrons (cf :QTL in scf file), which is different for eg and t2g states, hence LDA+U distorts the splitting. Second, the selfinteraction of the d-electrons is present, again distorting the crystal field splitting.
Regards Pavel Novak On Tue, 15 Sep 2009, Yang Ding wrote: > Dear WIEN2k users, > > I am really new to WIEN2k, and wondering if you could give your advice and > experience on following question concerning the crystal filed splitting > calculated from WIEN2k. > > In order to understand if the pre-edge splitting appearing in the Fe K-edge > spectra (1s-4p transition) measured by emission-XANES on Fe2O3 [Groot et al. > J. Phys.: Condens. Matter 21 (2009) 104207 > <http://www.iop.org/EJ/abstract/0953-8984/21/10/104207/>], is linked to > crystal-filed splitting in 3d empty band. We did a very preliminary ground > state calculation using WIEN2k based on GGA+U (and LSDA+U) with U = 4 eV > structure to check the crystal field splitting in empty d band above Fermi > level. > > As a result, we found that above 2-6 eV above Fermi level, the energy of t2g > is higher than that of eg. This result is similar to what reported by Rollsman > et al (PHYSICAL REVIEW B 69, 165107 (2004) > <http://prola.aps.org/abstract/PRB/v69/i16/e165107>) on Fe2O3. In his > calculation (GGA/LSDA+U , U= 4eV), the energy of t2g is also higher than that > of eg. So my question is why the t2g and eg are reversed in DFT, but the > Multiplet calculation gives contradictory results (i.e from Groot et al.). > > I noticed that Glatzel et al (PHYSICAL REVIEW B 77, 115133 (2008) > <http://prola.aps.org/abstract/PRB/v69/i16/e165107>) reported that they > obtained the "right" crystal field splitting using (LDA+U, U=6 eV) from > WIEN2k. So we wonder if we might missed something in the calculations? > > Thanks in advance for your help, > --