isym = 3 issue

[foyevtsova]

Dear VASP developers,

I'm using HSE06 with LHFCALC = .TRUE. to calculate the non-magnetic band structure of a freestanding FeSe monolayer. With hybrid functionals, the default choice of ISYM is 3. This, however, produces incorrect band splittings inconsistent with the results obtained by setting ISYM to 2. This is illustrated in the attached pdf file.

Could you please explain what is going on and whether it is safe to use ISYM=3.

Thank you,
Kateryna

[svijay]

Dear Kateryna,

Do you have to have the band structure for the same calculation with ISYM=-1 (i.e., all symmetry turned off)? That would be one source of comparison between the two calculations.

Alternatively, it would be great if you could attach all input and OUTCAR files generated from a calculation with ISYM=2 and ISYM=3, especially if they were performed with the NWRITE=3 (https://www.vasp.at/wiki/index.php/NWRITE) INCAR tag to see which operations are performed.

[foyevtsova]

Thank you for your response.

I attach the input and OUTCAR files for the isym=2 and isym=3 runs. Unfortunately, I did not have NWRITE=3 in them.

I'm running the isym=-1 calculation now, but it takes time since this is a hybrid functional calculation and I do not have much computational resources available at the moment.

The attached files are for calculating the bandstructure following the option 1 steps in

https://www.vasp.at/wiki/index.php/Band ... unctionals

But regular (ie automatic) k-mesh calculations with isym=2 and isym=3 also show differences, in DOS and in total energies, in particular.

Looking forward to hearing back from you,
Kateryna

[svijay]

Thanks for the OUTCAR files, it does look like the operations between the two files are the same. To me it looks like the ideal point of comparison between ISYM=3 would be with ISYM=-1 (i.e. no symmetry). As stated in the OUTCAR file, ISYM=2 does not symmetrize the exchange-potential, but does symmetrize the charge density (and hence the Hartree potential) - ISYM=3 on the other hand does not symmetrize the Hartree potential - which would mean that the exchange potential and the Hartree potential are consistent.

Sudarshan

[foyevtsova]

Dear Sudarshan,

thanks for checking the files. While the isym=-1 calculation is running, could you please explain the reasoning behind setting isym=2 and isym=3 as default values for regular PBE and hybrid functionals, respectively?

And still, irrespective of what I get with isym=-1, the 2 and 3 obviously give conflicting results for my system, which, as I understand, should not be happening.

Thanks,
Kateryna

[svijay]

Dear Kateryna,

Looking through the files you posted earlier, it does appear that the energies are the same for both ISYM=2 and ISYM=3. Using,

Code: Select all

grep "energy  without entropy="

for both the OUTCARs gives me a very similar energy. Running a simple test calculation with your system and the following INCAR file

Code: Select all

ISYM = 2 ! or 3
LHFCALC = T
KSPACING = 0.4
GGA = PE
HFSCREEN = 0.2
NWRITE = 3

also gives me the same energy. Could you please let me know how you get a different energy in the figure you posted in the first post?

Sudarshan

[foyevtsova]

Dear Sudarshan,

I appologize for the confusion. Those energies cited in the figure come from two corresponding separate calculations where I use an automatic k-grid and the tetrahedron integration method. I've always regarded the tetrahedron method as more accurate when it comes to comparing the energetics. I attach the input and OUTCAR files for those.

Unfortunately, the isym=-1 calculation diverged instead of converging over the weekend, so I'll have to look into it more carefully and restart.

Thank you,
Kateryna

[svijay]

Dear Kateryna,

I am not entire sure I understand, looking at the INCAR files for the two different calculations ISYM=2 and ISYM=3 shows that the they have different NSW (i.e. there is a relaxation with ISYM=3 but a static calculation with ISYM=2). Looking at the figure you have compared the energies at the end of the relaxation against one where no relaxation was done - both the energies would not be the same in this case.

Sudarshan

[foyevtsova]

Dear Sudarshan,

by doing

grep 'free energy TOTEN' OUTCAR

you can see that the energy in the zeroth step is -32.08324413 eV, whereas with isym=2 static it is -32.14314974 eV.

Thanks,
Kateryna

[svijay]

Yes, but from the files it looks like you are comparing the final energies from a calculation with NSW=30 (isym=3) and NSW=0 (isym=2) - see here (wiki/index.php/NSW). Please compare the energies for the first _converged_ SCF loop.

Sudarshan

[foyevtsova]

NSW=0, isym=3 gives -32.08296422 eV and NSW=0, isym=2 gives -32.14314974 eV, as you can see in the attached files, where the only difference now between the two runs is isym.

Thanks,
Kateryna

[svijay]

Yes indeed, I see, thanks for sharing these files. I suspect that the calculation may not be converged in terms of PRECFOCK (see here: wiki/index.php/PRECFOCK). Have you tried settings PRECFOCK to the default value?

Sudarshan

[foyevtsova]

Dear Sudarshan,

with PRECFOCK = Normal, I still get -32.08322727 eV for isym=3 and -32.14349474 eV for isym=2.

Thanks,
Kateryna

[foyevtsova]

Dear Sudarshan,

earlier I posted that changing PRECFOCK from Fast to Normal did not improve the discrepancy between the total energies from isym=3 and isym=2 runs: -32.08322727 eV for isym=3 and -32.14349474 eV for isym=2.

This looks like a bug to me, could you please investigate further?

Thanks,
Kateryna

[svijay]

Dear Kateryna,

Could you please also send the files for your latest test? Could you also let me know if the ISYM=-1 calculation converged?

Sudarshan