NGsBlkXp convergence

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

Moderators: Davide Sangalli, Daniele Varsano

Post Reply
Online
8813204602
Posts: 24
Joined: Tue Jun 15, 2021 5:53 pm

NGsBlkXp convergence

Post by 8813204602 » Sun Oct 24, 2021 4:45 pm

Dear forum,

I am doing convergence tests to calculate the GW band structure. The convergences related to RandQpts,RandGvec,EXXRLvcs and BndsRnXp were well done, but in the NGsBlkXp convergence , as seen in the attached figures, with increasing the value of NGsBlkXp for BndsRnXp = 90, the VBM starts to decrease again after reaching the convergence, and as a result, the gap also starts to increase. What is the cause of this behavior? Do I still have to do the calculation for NGsBlkXp greater values to achieve convergence?
In calculating the correct GW band structure, the gap is the criterion for convergence, or should convergence be checked relative to VBM and CBM?

Thanks,
Mitra Helmi
You do not have the required permissions to view the files attached to this post.
Mitra Helmi,Ph.D student
Ferdowsi university of Mashhad
Mashhad,Iran

User avatar
Daniele Varsano
Posts: 3131
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: NGsBlkXp convergence

Post by Daniele Varsano » Mon Oct 25, 2021 8:45 am

Dear Mitra,

there is no criterion to argue that the convergence should be monotone, so yes, you should keep increasing the NGsBlkXp value.
The gap energy is a good criterion if you are interested in optical properties, while you should look at the single bands if you need to converge them separately e.g. band alignments etc.

Please note also that NGsBlkXp and BndsRnXp are not dependent variable, ie if NGsBlkXp converge for a BndsRnXp value you may need larger value for NGsBlkXp when increasing the number of bands.

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/

Online
8813204602
Posts: 24
Joined: Tue Jun 15, 2021 5:53 pm

Re: NGsBlkXp convergence

Post by 8813204602 » Tue Oct 26, 2021 12:59 pm

Dear Daniele

In these convergence tests, I set the NGsBlkXp in the Ry at the inputs, but when I checked the value in RL at the output, the equivalent values are different from what is given in the Yambo site tutorial. For example, in the tutorial up to NGsBlkXp= 4Ry, the size of the dielectric matrix has not yet reached 40RL, but in the output of my calculations, the size of the dielectric matrix in NGsBlkXp = 4Ry has been reported to be 139RL. What is the reason for this difference?
Is this conversion dependent on the type of material?


Best regards,
Mitra Helmi
Last edited by 8813204602 on Tue Oct 26, 2021 1:10 pm, edited 1 time in total.
Mitra Helmi,Ph.D student
Ferdowsi university of Mashhad
Mashhad,Iran

User avatar
Daniele Varsano
Posts: 3131
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: NGsBlkXp convergence

Post by Daniele Varsano » Tue Oct 26, 2021 1:07 pm

Dear Miytra,

when you provide the size of the matrix in Ry, ie using an energy cutoff, this means all the G's such that |G|^2/2 < Ecut
The value of the G depends on the geometry and size of your unit cell being the G_i=2pi*n/L_i ;(i=x,y,z)

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/

Online
8813204602
Posts: 24
Joined: Tue Jun 15, 2021 5:53 pm

Re: NGsBlkXp convergence

Post by 8813204602 » Fri Oct 29, 2021 4:11 pm

Dear Daniele

According to the attached figure, is it correct to consider the values of NGsBlkXp = 10 and BndsRnXp = 70 as convergent values, or should higher values be considered?

Best,
Mitra
You do not have the required permissions to view the files attached to this post.
Mitra Helmi,Ph.D student
Ferdowsi university of Mashhad
Mashhad,Iran

User avatar
Daniele Varsano
Posts: 3131
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: NGsBlkXp convergence

Post by Daniele Varsano » Sat Oct 30, 2021 10:28 am

Dear Mitra,

of course it depends on the accuracy you need. It seems to me pretty converged. You can try to extrapolate the last points with a reasonable function e.g. a+1/Ngblk and see if the fit is reasonable and estimate your error wrt the asymptotic value.

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