Tetrahedron method and vasp5-6 versions

[IBRAHIM]

Dear All,

For semiconductor and insulator with PBE and with ISMEAR=-5, I found that vasp5.4.4 gives E Fermi of 4.02 located at the top of the valence band, while vasp6.3.2 shifted the Fermi level to 6.00 at the bottom of the conduction band. Total energy is the same with both versions!

Why such different behavior (p-n type) with both versions and which one we can rely on?

I see that issue is only with ISMEAR-5, with ISMEAR=0 no such issues for vasp6.3.2 and fermi level is located at the top of the valence band with both versions.

Moreover, for vasp6.3.2, I noticed band occupations of -0.00000, of course it is zero but why do negative occupations appear for semiconductors and insulators with ISMEAR=-5 in vasp6.3.2? or just ignore!

Similar behavior (the Fermi level at the bottom of the conduction band and negative zero occupations) is also found for vasp6.3.2 with HSE and ISMEAR=-5, so the functional may not be the issue. I did not test vasp5.4.4 with HSE and ISMEAR=-5 for those materials.


Attached files are OUTCARs.

Thanks

[fabien_tran1]

Hi,

There have been already several questions related to the position of the Fermi energy, see forum/viewtopic.php?f=4&t=18680.

Note, that in the case of a system with a gap, the position of the Fermi energy is not well defined and can be chosen arbitrarily. Also, with the tetrahedron method with Blöchl corrections, negative values for the occupations can be obtained, in particular for metals. In your case it is a semiconductor, but I do not think that -0.00000 is worrisome.

[IBRAHIM]

Thanks for your reply.

It seems that issue still in vasp6.3.2 with tetrahedron method!

But I am still confused about the band occupations for semiconductors and insulators with vasp6 and ISMEAR=-5,
because in addition to negative zero occupations, I found for some systems very small occupation such as 0.000001, 0.000004, 0.000007 in EIGENVAL using HSE06. but with ISMEAR=0 no such issues and band gap values are OK.

In fact, I am using tetrahedron with HSE06 to show DOS. So as recommended here "forum/viewtopic.php?f=4&t=18525" I can align the Fermi energy at the valence band maximum as a post processing step.

But for the strange occupation of -0.000000 and 0.000001-7, is it a bug for semiconductor and insulator with vasp6 and ISMEAR=-5? or simply can I ignore them? could I trust DOS figures in this case or use DOS and bandgap values from ISMEAR=0?


attached files are EIGENVALs.

Thanks

[fabien_tran1]

These very small occupations should be due to the fact the methods to determine the Fermi energy are not perfect (as far as we know there should be no bug in vasp.6.3.2, but we can not be sure at 100%). Anyway, such tiny artifacts should have no or extremely minor influence on the results (electron density, band structure, etc.). Thus, you can probably ignore them.

However, what is a bit puzzling are the differences between the eigenvalues obtained with ISMEAR=0 and -5 in your HSE06 calculations. They differ by up to 0.1 eV. Did you run these two HSE06 calculations with the same settings (except for ISMEAR)? Could you provide use the input files of the HSE06 calculations, so that we can have a look?

[IBRAHIM]

Thank you very much.

OK I will ignore such occupations and consider them as zero, then align the Fermi energy at the valence band maximum to show DOS with ISMEAR=-5 in those systems.

Sorry by mistake the previous two files belong to two crystal phases of the same material. Please find attached the correct EIGENVAL files for both phases.

For DOS with HSE06 and ISMEAR=-5, as recommended by VASP "ALGO=A and IALGO=5X tend to fail with the tetrahedron method. For DOS calculations use IALGO=53 after pre-converging with ISMEAR>=0".
So, I follow these steps:

Step1: Standard PBE, ISMEAR=0, ALGO=Normal
Step2: HSE06, WAVECAR from step 1, ISMEAR=0, ALGO=All, the other related HSE06 tags
Step3: HSE06, WAVECAR from step 2, ISMEAR=-5, IALGO=53, NEDOS=1000, the other related HSE06 tags

Input files are attached. Please have a look and try to reproduce those issues.

Then I checked the band occupation in step 2 and 3, and found those artifacts only with ISMEAR=-5 and vasp6. Again, I did not test vasp5 with HSE06 and ISMEAR=-5 for my systems. But I did not see those issues with vasp5 before.
So, maybe it is only related to vasp6 and tetrahedron method!

Thanks

[fabien_tran1]

Hi,

What do you mean with "tend to fail with the tetrahedron method"? The SCF convergence can not be reached?

[IBRAHIM]

This warning by VASP:

"ALGO=A and IALGO=5X tend to fail with the tetrahedron method (e.g. |
| Bloechl's method, ISMEAR=-5, is not variational). Please switch to |
| IMSEAR=0-n, except for DOS calculations. For DOS calculations use |
| IALGO=53 after pre-converging with ISMEAR>=0. |
| I HOPE YOU KNOW WHAT YOU ARE DOING!"

[fabien_tran1]

This is what is explained below "How to set ISMEAR" at ISMEAR.

[IBRAHIM]

This is what is explained below "How to set ISMEAR" at ISMEAR.

Yes, I know about issues of relaxation and ISMEAR.

This message for ALGO=A and IALGO=5X and ISMEAR=-5 for DOS.
I have discussed it before "https://www.vasp.at/forum/viewtopic.php?f=4&t=18457"

Please also as I have mentioned in the beginning of this discussion that the issues appear for PBE functional with vasp6 and tetrahedron method.
So, again maybe it is only related to vasp6 and tetrahedron method!

Thanks

[fabien_tran1]

In conclusion, your calculations seem to be ok and there is no real sign of a bug in the code. Some of the occupations differ from normal values, but only by a very tiny amount and with no importance for the results. The differences between vasp.5.4.4 and vasp.6.3.2 are due some changes in the strategy to find the Fermi energy.

[IBRAHIM]

Thanks for the detailed discussion and clarification.