Page 1 of 1

DMFT difference in Nb. of corr. elec. G_imp and G_loc

Posted: Mon Nov 09, 2015 9:53 am
by PhilippBa
Dear all,

I am using the DMFT implementation of ABINIT to solve the AIM with CT-HYB. I am interested in the total energy of my system and I am happy that ABINIT can calculate it for DMFT.

However, I obtain a difference in the occupation between the impurity Greens function and the local Greens function that I have no explanation for. I want to solve the impurity model just once so that i only run the DMFT loop once. Therefore the mixing of the self-energy was set to 1. For further details see the attached input file. I set nstep to 200, but this was a mistake. I canceld the calculation after the first step, so don't wonder why the log-file is not complete (I don't want charge self consistency for now).

After the CT-QMC I obtain a local occupation of 7.81 (G(tau) in line 3932 of the log file).

After the mixing of the self-energy and post-treatment the occupation of G(_loc??) is 7.40. The occupation of 7.40 could also be obtained after the search of the new Fermi level.

If that was G_loc in this case (line 4076 and 4157 of the log file) shouldn't it be the same value as G_imp?

At least if I set U = 0 it is the case (unfortunately I lost the results, but if required I can reproduce them).

Best regards,


Re: DMFT difference in Nb. of corr. elec. G_imp and G_loc

Posted: Tue Nov 10, 2015 9:10 am
by amadon
Dear Philipp,

The number of electrons obtained after the Monte Carlo (line 3932) is the trace of the impurity Green's function G_imp.
The number of electrons written on line 4076 and 4157 is the trace of the local Green's function G_loc.

These two Green's functions are in general different.

However, at convergence of the DMFT loop, G_loc=G_imp.
But in your calculation, you did not converge the DMFT loop because you did only one iteration of the DMFT loop.
Try dmft_iter=10 and the two numbers should be closer....

Of course, if U=0, G_imp=G_loc=G_locLDA is constant.

Best regards

Re: DMFT difference in Nb. of corr. elec. G_imp and G_loc

Posted: Tue Nov 10, 2015 10:50 am
by PhilippBa
Dear Mr. Amadon,

I totally agree with you. For a lattice problem like SrVO3 the convergence is, of course, required to have G_imp = G_loc. I tested this with SrVO3 and it works fine. But for a non-lattice problem, like in my case (a single Co atom on a cupper surface), I would assume that the convergence is not required, because this is an example for a single impurity coupled to an electronic bath (i.e. an Anderson Impurity Model).

However, more important for me is the calculation of the total energy. Do you think that even with one iteration the energy is meaningful? How would you handle a single adatom on a metallic surface with ABINIT, if you are interested in the total energy?

Best regards,