GW convergence rim_W method

You can find here problems arising when using old releases of Yambo (< 5.0). Issues as parallelization strategy, performance issues and other technical aspects.

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano, Conor Hogan

Locked
rreho
Posts: 13
Joined: Mon Apr 26, 2021 3:39 pm

GW convergence rim_W method

Post by rreho » Sat Sep 17, 2022 4:52 pm

Dear developers,
I am trying to converge my GW gap for WS2 with the new rim_W method with respect to k-mesh as explained in https://arxiv.org/abs/2205.11946. Unfortunately, it does not seems to converge as nicely as the case for MoS2 presented in the paper. The main difference with respect to the procedure presented in the reference is the inclusion of spin-orbit coupling and the convergence parameter used to perform the calculation (low number of bands etc..). I attach here the r_setup, input, and output files of my calculations. I am worried that I am not following the procedure correctly. Could you please help me?
Thank you in advance for your help,

Kind regards,
Riccardo Reho
You do not have the required permissions to view the files attached to this post.

Alberto Guandalini
Posts: 8
Joined: Thu Jun 24, 2021 9:10 am

Re: GW convergence rim_W method

Post by Alberto Guandalini » Mon Sep 19, 2022 9:51 am

Dear Riccardo Reho,
by looking at your results, it seems that also the HF self-energy (the valence one) shows some convergence issues.

Please find in attachment a modified version of your output file where I collected self-energy values.

As the calculation of the HF self-energy is not related with the rim_W part of the code, my guess is that there is something
unexpected that happens already at the HF level.

As the HF self-energy converges rapidly for the conduction state but slowly for the valence state, the problem may be related to the
self-interaction removing term (for nk QP state, the Sx term where n=m, where m is the band index in the summation
to compute Sx).

By looking at your input file, there are some input variable that seems to be very low:
EXXRLvcs = 1 Ry #Number of G vectors included in the Sx summation
FFTGvecs = 10 Ry #Number of G-vectors included in the FFT of wavefunctions

I will try to increase these parameters and see if the problem persists.
My advice is to focus on the HF gap at first, then move to GW calculations.

Let us know if the problem persists.

Kind regards,
Alberto Guandalini
You do not have the required permissions to view the files attached to this post.

rreho
Posts: 13
Joined: Mon Apr 26, 2021 3:39 pm

Re: GW convergence rim_W method

Post by rreho » Mon Sep 19, 2022 3:27 pm

Dear Alberto,
Thank you for your reply. I tried to calculate the HF corrections for my system varying the number of k-points and keeping EXXRLvcs and FFTGvecs to a converged value = 70 Ry. Unfortunately, it does not seem to converge. Do you have any suggestions?
I share the input, output, and report files.
Thank you in advance.

Best regards,
Riccardo Reho
Utrecht University
You do not have the required permissions to view the files attached to this post.

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence rim_W method

Post by Daniele Varsano » Tue Sep 20, 2022 7:34 am

Dear Riccardo,

from what I can see, in your input is missing the rim_cut keyword, ie the MC integration of the Coulomb potential which is needed to accelerate the convergence wrt k points.

Out of topic: even if it is not used for the HF calculation, I discourage the use of XTermKind: it is a memory intensive procedure without much gain in time-to-solution.

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/

rreho
Posts: 13
Joined: Mon Apr 26, 2021 3:39 pm

Re: GW convergence rim_W method

Post by rreho » Tue Sep 20, 2022 10:06 am

Dear Daniele,
I have included the rim_cut keyword and deleted XTermiKind but I do not see any difference at the GW level. Of course, I can always try to go higher with the k points but I have the impression that something important may cause this issue. Please note that I have not used the assume_isolated=
2D' flag in QE.

Thank you for your help.
Best,
Riccardo
You do not have the required permissions to view the files attached to this post.

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: GW convergence rim_W method

Post by Daniele Varsano » Tue Sep 20, 2022 4:35 pm

Dear Riccardo,

the new calculations do not seem to have a bad behaviour:

Code: Select all

nk hf_gap
7 7.71474
19 7.56194
37 7.57195
61 7.59565
91 7.61578
127 7.62763
169 7.63214
What I find unusual anyway is the very large value of "RandGvec" you are using. Besides that, as next you want to sue the RIM I would use the CUTGeo with slab and not the box. You can also perform an MC integration using the slab. Also note that when using the box the MC is not performed and the coulomb integral are used to build the box-shaped potential.

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/

rreho
Posts: 13
Joined: Mon Apr 26, 2021 3:39 pm

Re: GW convergence rim_W method

Post by rreho » Wed Sep 21, 2022 7:29 am

Dears,
Thank you for your help and for your nice explanation.

Best regards,
Riccardo Reho

Locked