Number of Matsubara frequencies; some help needed

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
PhilippBa
Posts: 15
Joined: Tue May 26, 2015 7:22 pm

Number of Matsubara frequencies; some help needed

Post by PhilippBa » Wed Jun 10, 2015 10:54 am

Dear all,

since I am dealing with a large system (91 atoms in supercell) some warning occurs during my DMFT run:

== Compute Band Energy terms for DMFT
Warning !:Differences between band energy from LDA occupations
and LDA green function is: -0.882458
which is larger than 0.00001
Action: increase number of frequencies, or reduce the number of high energies_dmft bands



This error is of course due to a small value of Matsubara frequencies in the linear mesh. In this case is used 10^6 frequencies and the difference decreases to -0.088246 if i increase the number of frequencies to 10^7. But to reach a difference of just 0.00001 it would need a very high value of Matsubara frequencies in the linear mesh.

The question is, can i ignore this warning? Do i do some serious error if i do so? I guess the computational costs would increase rapidly if i increase the Matsubara frequencies e.g to 10^11


I should mention that i do not converge the DMFT over the charge and i only want to solve the Andersion impurity model once (thats why only one DMFT step).


Inputs for LDA and DMFT are given below.


Best regards,

Philipp
Attachments
co.in
(3.72 KiB) Downloaded 263 times
DMFT.in
(4.3 KiB) Downloaded 269 times
Philipp Bahlke
University of Hamburg
Institute for inorganic and applied chemistry
Germany

amadon
Posts: 46
Joined: Mon Aug 24, 2009 10:58 am

Re: Number of Matsubara frequencies; some help needed

Post by amadon » Thu Jun 11, 2015 8:45 am

Dear Philipp,

Indeed, in order to be useful, the difference of energy could be renormalized by the number of atoms.
In your case, as you have a very high number of states, the difference is large.
More generally, this warning is here to warn the users about the absolute precision of the Matsubara grid used in view of doing
total energy calculations.
Nevertheless, what is indeed useful most of the time is energy differences and not absolute values.
So, the best solution to deal with this parameters, is to converge the quantity you want to compute with respect
to the number of frequencies (dmft_nwli, and dmft_nwlo). For lower temperatures, you need more frequencies.

I hope it helps
Best regards
Bernard
Bernard Amadon
CEA
France

PhilippBa
Posts: 15
Joined: Tue May 26, 2015 7:22 pm

Re: Number of Matsubara frequencies; some help needed

Post by PhilippBa » Thu Jun 11, 2015 10:01 am

Thanks to you! That helped a lot.


Best regards,

Philipp
Philipp Bahlke
University of Hamburg
Institute for inorganic and applied chemistry
Germany

Locked