My Community

Hello,

My question involves a follow-up to this thread: forum/viewtopic.php?f=4&t=17981&p=19779 ... ral#p19779

martin.schlipf wrote:There was a bugfix for the tetrahedral integration in 6.1, perhaps that had unintended side effects. I'll investigate to see how this occurs.

Our team is still investigating, but we've found significantly more "incorrect" Fermi levels reported by VASP as of v6.1.

Could you offer any more information on what motivated the bug fix in v6.1, what this change entailed, and if any further changes to this algorithm have been made as of v6.2 or v6.2.1? I see the basic description of the algorithm in the previous thread.

This would be very helpful for us to track down the issues we're seeing.

Many thanks!

Matthew

Dear Matthew,

Sorry to hear this is causing issues. I will look into what this change entailed and if any further changes to this algorithm have been made.
Meanwhile, it would be helpful if you could define "incorrect" Fermi levels. For example, is your main concern backward compatibility? Or is there a certain definition of the Fermi level you compare to?

Cheers,
Marie-Therese

Hi Marie-Therese,

Thank you very much for your reply!

This is partly a problem of our own making. Essentially, to determine the band gap of a material, we typically assume that for a semiconductor that the Fermi level reported by VASP will be somewhere in the gap, and that states below this level are occupied and states above this level are unoccupied. This allows us to reduce storage requirements when we're storing 100,000s of band structures in our database since we don't have to store the occupancies explicitly.

Historically, this approach seems to have worked well. However, after running calculations with VASP 6 we started to notice that the band gap would change radically for existing materials in a small number of cases. It appears to be that the Fermi level lies within a band even for semiconductors where it clearly shouldn't. Examining these calculations with reference to the actual occupancies reported in the EIGENVAL file allows us to determine the correct band gap.

When comparing band gaps determined with reference to the Fermi level, and band gaps determined by looking at occupancies explicitly, these values largely agree for VASP v5.x, but for VASP 6.1.0 we see a large discrepancy. This is really only something we see in aggregate. It only affects a small % of calculations, perhaps hundreds out of our whole dataset.

We're still investigating this and a colleague is looking to find a small, reproducible example, so we don't have the full story here yet. I'll keep you updated if we learn more.

Best regards,

Matthew

Hi Matthew,

Thank you for the additional details and your patience.

As you know and as was discussed in the previous post, the bisection algorithm is causing the Fermi level to be either at the top or bottom of the band gap. Although it's not really a bug, it causes undesired behavior and we now plan to set it to the middle of the band gap in a future version of VASP. This will be compatible with your current workflow, so you may consider waiting for that update to fix the issue. Sorry, this is not an immediate solution.

Best regards,
Marie-Therese

Many thanks Marie-Therese! No worries that this is not an immediate solution, it's very helpful to know this will be changed in an upcoming version.

All the best,

Matt

Hi Matt,

Perhaps you have noticed that a new release (VASP 6.3) is out. Unfortunately, the changes we discussed are not part of this release. It is still work in progress as it involves careful checking of our tests etc.

Best regards,
Marie-Therese

Hi Marie-Therese,

I actually just logged on to ask this very question, so thank you for updating the thread! I appreciate the need for testing etc. before making changes so will look forward to the changes in a future update.

Congratulations on the VASP 6.3 release!

All the best,

Matt

Hi Marie-Therese,

I was wondering if there was any update on this issue? Asking on behalf of a colleague.

Many thanks,

Matthew

Hi Matt,

Thank you for asking again and for your patience in this issue. In the next release, we plan to have a new INCAR setting: EFERMI = MIDGAP. Then, all smearing methods will set the Fermi energy in the center of the band gap. It is reasonable to expect 6.4 around New Year, but we don't announce exact dates. I am sorry if this causes a delay in your projects.

Best regards,
Marie-Therese

VASP 6.4 is now released. The Fermi energy calculation now reverts back to the traditional algorithm if you don't set anything. With a new INCAR tag EFERMI = MIDGAP, you can switch on a more deterministic algorithm that will put the Fermi energy in the middle of the band gap if possible.