GW unoccupied states have strange anomalies

[engelund]

Hi Community,

I am trying to estimate dI/dV curves and apparent band gaps with Yambo. For that purpose, I generate a molecule attached to a cluster representing the substrate and calculate GW0 corrections for many bands (yes it is quite heavy). Although only using GW0, the results should still be much better than bare DFT. I only use the gamma point since it represents a 0D structure.

I see that suddenly, between bands 254 and 255 (see "o-" file), there is a sharp change in the energy correction to the point that these states drop below the Fermi level to become occupied. I am pretty sure this must be an erroneous artifact since these extra occupied bands would push the Fermi Level away from the gap (there are states near the Fermi Level, but very few of them).

I have another system where the problem is more pronounced, and the correction is suddenly -25 eV. However, for various reasons, I cannot publicly show this system.

Do you guys know what might be the problem?
Am I doing something wrong?
Is there a workaround?

[claudio]

Dear Mads

I see you results, but I do not have a clear solution. I give you some hints that you can test

1) Try to use monte-carlo integration for small q-points and G-vectors

RandQpts=3000000 # [RIM] Number of random q-points in the BZ
RandGvec=11 RL # [RIM] Coulomb interaction RS components

2) Can your problem be due to the plasmon pole approximation?
you can try multi-pole approximation that now is implemented in Yambo
https://www.yambo-code.eu/wiki/index.ph ... tion_(MPA)

3) you can check if the problem is due to the exchange only, by doing and HF calculation yambo -g n -x

4) try a different cutoff for the example the "box xyz"

let us know
best
Claudio

[engelund]

Hi Claudio,
Thank you for giving me some leads; I will try them out.
At least the cut-off method seems not to be to blame since I tried it without a cut-off method and the problem persisted.
Will let you know once I have tried the other leads.

Mads

[engelund]

Hi Claudio,

I have tried all the tests you suggested.

1) Monte-Carlo: Still had the issue.
2) MPA: Still had the issue.
3) Exchange-Only: The issue was present.
4) Removing the cut-off entirely did not resolve the issue.

In conclusion, it seems that the problem is due to exchange only.
What do you guys suggest as the next step?

Mads

[claudio]

Dear Mads

may you try to repeat the calculation using DOJO pseudo-potentials
with accuracy = stringent

http://www.pseudo-dojo.org/

and check if the problem is still there

best
Claudio

[engelund]

The calculations are already performed with these pseudo-potentials.

[claudio]

Ok in this case, make a small test with other pseudo-potential for example the PZ-VBC

http://pseudopotentials.quantum-espress ... pp-library

just to be sure that these states are not artifacts created by the pseudos

Claudio

[engelund]

I tried with the pseudos that you described (although I used "O.pz-kjpaw.UPF.txt" as a replacement) - the strange behavior is still there at the same band number.

[Daniele Varsano]

Dear Mads,

the only think I see that it is not consistent is the value of your sphere cutoff radius, this is larger of some of the size of your cell.
You can try to use the box, something as:

Code: Select all

RandQpts=2000000                     # [RIM] Number of random q-points in the BZ
RandGvec= 1                RL    # [RIM] Coulomb interaction RS components
CUTGeo= "box xyz"
 % CUTBox
  17.000000 | 25.000000 | 17.000000 |        # [CUT] [au] Box sides
 % 

Best,
Daniele

[Davide Sangalli]

Dear all,
I had a quick look to the output.

Even the <Vxc> suddenly changes from 254 to 255

Code: Select all

<254|lXC|254> = -2.733166  0.61872E-11
<255|lXC|255> = -19.37973  0.000000

This seems to suggest that the KS states are very different between 254 and 255. Did you try to plot the two orbitals?

Best,
D.