Small gap semiconductor issue

[C.Long]

Dear Yambo developers,
I calculated a system with a rather small band gap which is 0.2993eV at PBE level. I ploted the band structure of GW correction which looks like a mess and it is metallic instead of semiconducting.[img]

Gfn=1.jpg

[/img] After I changed the value of GfnQP_N in the ypp input file, the band structure became normal but the band gap was affacted by the value of GfnQP_N. [img]

Gfn=30.jpg

[/img]When I calculated the BSE absorption spectrum with previous QP correction, yambo warn me that the system is metallic but this warning didn’t show up if I use a simple scissor operator.
I want to know how to fix the incorrectness of GW band structure and the metallic warning of BSE calculation. Whether it is inevitable in a small gap system or there is something I have not noticed in the input parameters.
Thanks

Best regards,
Chen Long

[Daniele Varsano]

Dear Chen Long,
please sign your post with your affiliation, this is a rule of the forum and you can do once for all by filling your signature in the user profile.

After I changed the value of GfnQP_N in the ypp input file, the band structure became normal but the band gap was affected by the value of GfnQP_N.

This is not a problem as the interpolation tool is meant only for plotting band structure, their values are not used in the following calculations as BSE.

When I calculated the BSE absorption spectrum with previous QP correction, yambo warns me that the system is metallic but this warning didn’t show up if I use a simple scissor operator.

So this means either you really have a metallic QP band structure (you should look at your o.qp file if this is the case), but I doubt about that as the interpolation indicate it is not, either you are using more bands than the ones you calculated the QP correction: in this case yambo interpolates the missing bands (k points) and there is the possibility is not doing it correctly for some reason. Anyway, this is just speculation as I cannot tell you more without looking at input and reports. If you ant you can post them and I will have a look.

Best,
Daniele

[C.Long]

Dear Daniele,
Thank you for your kind reply!
This bothered me for a while. I have not complete the BSE calculation but i put the report in attachment in case of need. The system I used has spontaneous polarity, could it be the reason for the abnormality?
Please check input and report in the attachment.
Thanks again!
Best,
Chen Long

[Daniele Varsano]

Dear Chen,
I had a look to your output and indeed it seems your system turns metallic when considering QP corrections.
It seems to me that something strange is happening.
If you look the QP energies you realize that QP corrections for selected points are very small. Here in attachment a plot showing Eqp vs Eks. I think you should inspect deeply what is happening to that points having a correction much smaller than the orest (e.g. k=1 b=139, k=10 b=150). Unfortunately the report file for the GW calculation you uploaded is not complete so it is not possible to see the different contributions to the QP energy (exchange self-energy, correlation or Z-factor).
I suggest you to concentrate on one of that points, check carefully the convergence (e.g. EXXRLvcs seems a bit low. but probably you already checked it), inspect wfs shape with respect adjacent k points having more regular correction etc.
I would also look at what happen by removing the GTermKind= "BG" option.

Best,
Daniele

[C.Long]

]Dear Daniele,
Thank you for your kind reply!
I have done the GW calculation without GTermKind = 'BG' option, and the situation has improved. The GW band structure without changing GfnQp_N finally became a semiconductor, but there is still some strange points. The valence bands still such a mess but better, however the conduction bands became normal.
Please check input and report in the attachment.
Thanks again!
Best,
Chen Long

[Daniele Varsano]

Dear Chen,
thanks for reporting, it seems that in your case the Gterm has some problem.
The problem you have QP corrections which are not in line with the others it is also plausible, may be you can inspect the nature of that bands: it is possible they have a different character so the correction is different from the others, also it is possible that they do convergence with different parameters.
I had a quick look to your input file and what I suggest you is to carefully check the RandQpts value. This is meant for a Monte Carlo stochastic integration, the more the random q points the more the error is reduced. Usually, it is set to a few million and it is rather costless. A simple check is the measure of the BZ volume:

Code: Select all

RIM  RL volume             [a.u.]:   0.0088
Real RL volume             [a.u.]:   0.0091

Note that these integrals are then used to build the cutoff coulomb potential, so inaccuracy propagates.

Best,
Daniele