DFT+DMFT example for Ce with VASP

Here we perform a DFT+DMFT calculation for Ce in its high temperature gamma phase, as discussed, e.g., in references [1] and as a tutorial in dft_tools for charge self consistent DFT+DMFT calculations with Wien2K and the old Hubbard-I version 1.4. For simplicity, we do no charge self-consistency. We rely on DFT calculations done with VASP and the VASP interface to triqs implemented in the dft_tools app. For all DFT related details have a look in the detailed documentation of dft_tools and especially the VASP interface.

We perform a VASP calculation with input files INCAR, KPOINTS, POSCAR, and an appropriate POTCAR. Non standard flags in the INCAR are:

LORBIT = 14
EMIN = 0
EMAX = 14
LOCPROJ = 1 : f : Pr 1
ISYM = -1

These select a special way of calculating projectors which optimize the overlap of bands in an energy window (EMIN and EMAX) and the Ce f states. We also switch off all symmetries (ISYM=-1).

After executing VASP we orthonormalize the raw projectors in the VASP output using plovasp and a corresponding input file plo.cfg which selects the correlated subspace and an energy window in which the orthonormalization is done:

plovasp plo.cfg

We convert the plovasp output to a h5 archive called ce.h5 using the script converter.py which serves as an input to triqs/dft_tools:

python converter.py

We can finally perform the dmft loop using a quite plain setup as implemented in the script ce.py:

python ce.py

For each iteration we calculate the self energy on Matsubara frequencies and the real axis by the calc_gw flag:

S.solve(h_int = H, calc_gw=True )

All results are saved in the archive ce.h5. To compare with results from literature, let us calculate the local lattice green function. We use the script ce_local_lattice.py, where we load the self energy on the real axis from the last dmft iteration and plug it into the equation for the lattice green function. The result is again saved in the archive. Let’s look at the result, which can, e.g., be generated by the script plot_dos.py.

We can nicely see, that the non-interacting DOS shows a large density of states close to the Fermi energy and that these states mostly consist of Ce f states. Including local atomic interactions leads to a splitting of these states into multipletts, which end up far away from the Fermi energy.