GW correction is NaN

Concerns issues with computing quasiparticle corrections to the DFT eigenvalues - i.e., the self-energy within the GW approximation (-g n), or considering the Hartree-Fock exchange only (-x)

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

Post Reply
clin
Posts: 9
Joined: Fri Sep 17, 2021 10:27 am

GW correction is NaN

Post by clin » Fri Sep 17, 2021 12:12 pm

Dear developers,

I am calculating GW bands and doing convergence test on NGsBlkXp and BndsRnXp, while GbndRnge is keeped as 50. When I further increased NGsBlkXp from 15 Ry to 20 Ry and keep other settings unchanged, I encountered the problem, NaN GW correction and its self-energy:

# K-point Band Eo [eV] E-Eo [eV] Sc|Eo [eV]
#
19 4 -2.362984 NaN NaN
19 5 1.688877 NaN NaN

What is usually the resaon for this issue? I attach the corresponding report file.

I have also several questions related to GW calculations and it will be great if you can provide any advice.

1) In NSCF calculations of PWSCF, nbnd is set to be large like 1000, there are a few eigenvalues not fully converged. Will this be important and affect the GW calculations a lot? What can I do to solve this?

2) When I truncate Coulomb potential as a typical 2D material in Yambo, I find the calculated E_xc (-8.575967 Ry) by Yambo is not exact the same as the one in PWSCF (-8.57667827 Ry). They can be the same if I don't adopt the truncated c potential and RIM for integrals. Will this small difference affect the subsequent calculations in Yambo? Should I use assume_isolated='2D' to truncate Coulomb interaction in PWSCF as well?

3) What is the good signature of a converged GW correction? Should the variation of GW correction for the top VB and bottom CB be small than 0.01 eV? How is the accuracy of GW correction? I want a accurate GW bands not only the band gap.

4) Is it safe to always keep GDamping and dScStep default for a GW calculation?

5) Is it recommended to use Yambo 4.3 version? Someone told me Yambo 4.3 is more stable than Yambo 5.0.3.

Thanks a lot,
Changpeng
You do not have the required permissions to view the files attached to this post.
Changpeng Lin
Doctoral Assistant, EPFL

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

Re: GW correction is NaN

Post by Daniele Varsano » Fri Sep 17, 2021 12:45 pm

Dear Changpeng,

please add you full name and affiliation, this is the rule of the forum and you can do once for all in your profile setting.

About the NaN we should investigate what is going on as it should not happen.
Anyway, note that the block size has been reduced to 10Ha:

Code: Select all

[WARNING] Response block size in GW reduced to 1369    RL (10010   mHa)
this is due to the use of the terminator, can you try to repeat your calculation without unsigned the BG terminator?

Nest please note that your system is a metal, is it expected? Is it a graphene monolayer? in this case takes into account that the convergence with k points is very demanding.
We have a implemented a strategy to deal with the graphene but it still not present in the public release of the code.

About yours other questions:

1) This should not affect the calculation, even if higher eigenvectors are not properly converged the important fact is they are orthonormal and this should be the case
2) The Exc calculation is totally independent on the coulomb cutoff, please check if the same number of Gvectors are used in the two cases are the same, anyway the two reported value are in a more than reasonable agreement between the two codes.
3) This is really up to you, it depends on the degree of accuracy you want to reach. 0.01 it is a rather tight criterium. Moreover if you are interested in the gap more than the single quasiparticle, in this case the convergence is faster
4) Usually yes.
5) They told you something totally wrong. From 4.3 to 5.0 many bugs have been solved

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/

clin
Posts: 9
Joined: Fri Sep 17, 2021 10:27 am

Re: GW correction is NaN

Post by clin » Fri Sep 17, 2021 2:57 pm

Dear Daniele,

Thanks a lot! I will run a GW calculation without the BG terminator. What do you mean by "unsigned" BG terminator?

Yes, it is monolayer graphene. I know the convergence w.r.t k-grid sampling will be slow, since we had a meeting to discuss this before.

In my case, if I use the BG terminator, the response block size will be reduced, can I use large NGsBlkXp?

As you said that 0.01 would be rather tight criterium, what is the often adopted one for checking convergence? Variations of GW correction smaller than 0.05 or 0.1 eV?

Best,
Changpeng
Changpeng Lin
Doctoral Assistant, EPFL

Post Reply