What's the reasonable number of bands from DFT calculation?

[Longjun Xiang]

Hi, dear developers:

I am following your tutorials to learn Yambo for studying excitation state physics. I have roughly understood how to choose the reasonable parameters to perform GW level calculation in terms of your nice guides. But in your supported examples, how to choose the number of bands from DFT calculation is not explicitly explained and the number, of course, is important because it is related to NGsBlkXp and GbndRange. For bulk hBN, the number of bands is 100 and for bulk GaSb this number has been set to 200. How can I choose the number for my own system? The best strategy is trying my best to choose a large number, then converge the gaps obtained by LUMO minus HOMO at Gamma point and finally reduce this number to a reasonable number when considering the GW corrections in all Brillouin zone?

From physical respects, the number of bands just governs the convergence of correlation self-energy part. The other soft-package BerkeleyGW https://berkeleygw.org/ is more demanding for this number. So what's the main difference for this number between BerkeleyGW and Yambo?

Best regards.

[Daniele Varsano]

Dear Dr. Longjun Xiang,

is important because it is related to NGsBlkXp and GbndRange.

Please note NGsBlkXp governs the dimension of the dielectric matrix and it is not directly related to the number of bands included in the response, even if it is indirectly in the sense they are not independent. Probably you were meaning the

Code: Select all

BndsRnXp

variable.

For bulk hBN, the number of bands is 100 and for bulk GaSb this number has been set to 200

Note that the tutorials are meant to run in simple platforms in a reasonable time, they are thought for tutorials for teaching purposes and not to have accurate converged results.
Anyway, as you can argue the number of bands needed to reach convergences it is system dependent and needs to be carefully checked. As you say, if possible it is a good strategy to calculate a large number in the nscf calculation in order to perform convergence tests.

what's the main difference for this number between BerkeleyGW and Yambo?

As far as I know, the different Plasmon model adopted, this is something that has a big influence on the number of bands needed to have converged results, anyway I'm pretty sure that also in BerkeleyGW it is now possible to use the same PP model of Yambo (Godby-Needs).

Best,
Daniele

[claudio]

Dr. Longjun Xiang

you can have a look at this webpage:

http://www.attaccalite.com/reasonable-p ... culations/

best regards
Claudio Attaccalite

[Longjun Xiang]

Dear Dr.Daniele:

Many thanks for you to correcting my mistake for the variables for NGsBlkX and BndsRnXp. In fact, as for these two variables, I have one more question. Can I consider BndsRnXp [GbandRange...... ] represents the first class variables which are closely related to the number of bands in PW calculation and NGsBlkX [EXXRLvcs......] represents the second class variables which are closely related to the Kinect energy cutoff in PW calculation (such as the encutwf in quantum espresso) ?

[Daniele Varsano]

Dear Dr. Longjun Xiang,
in the post of Claudio you can find a strategy to perform convergences.
Of course, in yambo calculations, BndsRnXp and GbndRange cannot exceed the number of bands stored by the QE nscf calculations and similarly NGsBlkX and EXXRLvcs cannot exceed the kinetic energy density cutoff of QE. Usually, EXXRLvcs requires a large value, but the calculation is rather fast while NGsBlkX it is in general much smaller than the kinetic energy cutoff used in QE.

Best,
Daniele

[Longjun Xiang]

Dear Dr.claudio:

Thanks for your reply. Your blog is fantastic and I will carefully look at your advice posted on your homepage.