GW convergence

Please post here any issue related to convergence parameters in linear response GW and BSE calculations.

Moderators: Davide Sangalli, Daniele Varsano

User avatar
Daniele Varsano
Posts: 2992
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence

Post by Daniele Varsano » Wed Nov 25, 2020 3:14 pm

Dear Alan,
you are right, I got confused with another file.
please provide the report files for the 1200 and 1600 case?
...
in the meanwhile can you repeat few calculations setting the same number of bands in GbndRnge and BndsRnXp?

Best,

Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Wed Nov 25, 2020 7:38 pm

Dear Daniele

I'm uploading the report files for the 1200, 1400 and 1600 case. I should note that for the 1400 and 1600 cases I had to do the calculation over multiple runs, hence several report files for these.

For the calculations you suggest, is it OK to use BG termination? Once you confirm I will get started on them.

Best wishes,

Alan
You do not have the required permissions to view the files attached to this post.
Alan Bowman
University of Cambridge

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Wed Nov 25, 2020 7:38 pm

Second set of files
You do not have the required permissions to view the files attached to this post.
Alan Bowman
University of Cambridge

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Wed Nov 25, 2020 7:39 pm

Final set
You do not have the required permissions to view the files attached to this post.
Alan Bowman
University of Cambridge

User avatar
Daniele Varsano
Posts: 2992
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence

Post by Daniele Varsano » Thu Nov 26, 2020 8:57 am

Dear Alan,
I've noticed nothing strange in your reports, yes you can use the G terminator, it will make the calculation a bit slower, but is done only for two states so it should not be a problem. Set the bands in X and G equal in each run.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Fri Nov 27, 2020 6:33 pm

Dear Daniele

I'm still running calculations with BG, but the results without (but with bands set the same) are:

Bands Evb Ecb Eg
600 0.1829 3.7432 3.5603
800 0.03442 3.566 3.53158
1000 -0.0255 3.4163 3.4418
1200 -0.08738 3.3184 3.40578
1400 -0.0865 3.08 3.1665
1600 -0.4094 3.275 3.6844

Let me know if you have any suggestions. I will carry on fiddiling too.

Also, I'm using FFT=100000 RL which is a little lower than the full value, but not by much.

All the best,

Alan
Alan Bowman
University of Cambridge

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Fri Nov 27, 2020 7:06 pm

Dear Daniele

To follow up further, I plotted some of the charge densities for 1000 bands upwards. I found that especially above 1200 bands there was a significant amount of charge density in the vacuum region. I had been assuming that the RIM settings in yambo would stop this being an issue in GW calculations, but potentially I need to go back to quantum espresso and put a potential correction in here (to prevent any vacuum effects from occuring here)? I expect my system has a dipole which is being felt over the repeating units in the vacuum direction, hence this issue. I will think about dipole corrections too.

If you have any thoughts, or think I might be wrong on this, please do let me know!

All the best,

Alan
Alan Bowman
University of Cambridge

User avatar
Daniele Varsano
Posts: 2992
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence

Post by Daniele Varsano » Sat Nov 28, 2020 9:55 am

Dear Alan,
I found that especially above 1200 bands there was a significant amount of charge density in the vacuum region.

I think that this is expected for high energy states.
This should not be an issue, here you are essentially exploiting closure relation \sum |i><i| = 1 , so the fact the wfs are delocalized is not an issue as far as you have a good orthonormalization.
I had been assuming that the RIM settings in yambo would stop this being an issue in GW calculations,
RIM is not related to that.

How many occupied bands do you have in your system? Is it possible you are still far away from convergence?
So my suggestion is:
1) Try pushing further bands convergence
2) If you have doubt about the quality of your high energy bands, be sure you have set diago_full_acc as .true., and a small conv_thr

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

arb83@cam.ac.uk
Posts: 98
Joined: Thu Jul 02, 2020 3:56 pm

Re: GW convergence

Post by arb83@cam.ac.uk » Mon Nov 30, 2020 11:41 am

Hi Daniele

I have 345 occupied states so I don't think I should be far from convergence at this point.

Do you think adding the line:

"assume_isolated : 2D"

in my quantum espresso input will have knock on effects in my Yambo calculation? Currently I have not done this, but I'm wondering if this is in fact where the problem lies.

Best wishes,

Alan
Alan Bowman
University of Cambridge

User avatar
Daniele Varsano
Posts: 2992
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence

Post by Daniele Varsano » Mon Nov 30, 2020 11:51 am

Dear Alan,
"assume_isolated : 2D", you can have a try, it should not give problem to Yambo.
anyway, I do not see the replica interaction related to bands convergence.
Note that you have a lot of occupied bands, even if it something really system dependent, it is possible you need more bands than you are actually using.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply