Question regarding DMFT tutorial on SrVO3

Running Abinit Ground state
1

Dear users,

Can some one help me to understand why the k-resolved spectra was calculated for U=0 while the self energy and real grid frequencies were taken from dataset2 which was DFT+U? Can we make k-resolved spectra with finite U or can we ask to take density from dataset2?

regards,
Rajesh

2

Dear Rajesh,

Generating the k-resolved spectra is a post-processing procedure.

  1. In dataset 1, the DFT simulation is performed to obtain the band eigenvalues and Kohn-Sham states.
  2. In dataset 2, the projectors are constructed and DMFT is performed. In this case, it is one-shot DFT+DMFT since there is only one DFT step. The result is the Green’s function and self-energy of the correlated orbitals.
  3. You then perform the analytical continuation to get the self-energy on the real-frequency axis. It is copied as an input for the dataset 3.
  4. Dataset 3 is launched. It reads the self-energy and embeds it back using the projectors. This allows to solve for the spectral function on the real axis, which is outputted. You don’t need U here because it is not actually solving the impurity problem, it just reads the self-energy file.

This whole procedure is performed in DFT+DMFT in this case, not DFT+U. Otherwise the self-energy would be static. Moreover, if you had U = 0, you would not have a self-energy, since you interacting Hamiltonian would be vanishing.

I hope this helps,
Best,
Olivier

3

Dear Olivier,

Thanks for your clarification. Since it was single step DFT+DMFT and what if it does not converge in one shot like in this SrVO3 case, won’t be a problem? Or needs to increase iterations?

Best regards,
Rajesh

4

Yes, I believe in this case this is simply to show how one would use the k-resolved spectral function feature, but to make an accurate calculation, the self-energy should be converged, that is increasing the number of iterations. By increasing dmft_iter, the DMFT procedure is repeated but the charge is not updated, thus this is one-shot but you can still converge the self-energy in one-shot. To have full charge self-consistency, you increase the nstep.

Best,
Olivier

5

But, can we use full charge self-consistency calculation for k-serolved spectra? Just for having feelings, I increased the iteration number for the dataset 2, and I get better sptectra than one iteration.

Regards,
Rajesh

6

Sure! The goal of the k-resolved spectrum is to use a full charge-self-consistently converged calculation. You get do the last dataset independently. You just have to provide the right files as input files.

Best,
Olivier