NKRED and bandstructure

[IBRAHIM]

Dear All,

For bandstructure calculations using HSE+SOC:

After HSE+SOC calculation with regular K-mesh and NKRED, can we use WAVECAR form that step and run the bandstructure step with that WAVECAR, CONTCAR, and modified KPOINTS (no NKRED)?

Or in that case the WAVECAR will be different due to including and excluding NKRED?

Many thanks in advance,
Ibrahim

[marie-therese.huebsch]

Hi Ibrahim,

Yes, the workflow you descibe is correct.
1. Do a ground-state-DFT calculation to preconverge.
2. Converge HSE+SOC with NKRED
3. Obtain the band structure with KPOINTS including zero-weighted points without NKRED.

Just FYI, for the latest version of VASP 6.3.0, there is an alternative way using KPOINTS_OPT.

Best regards,
Marie-Therese

[IBRAHIM]

Thank you very much.

So just to confirm, in step 3 we can use WAVECAR from step 2?

Also could NKRED affect the accuracy of the DOS calculation? or just the performance will be enhanced without loss of accuracy.

Best regards,
Ibrahim

[marie-therese.huebsch]

Yes, you can use the WAVECAR from step 2 in step 3.

For the use of NKRED, please take into account that there are certainly some caveats to downsampling as discussed on the VASP Wiki. It is a matter of testing and finding a balance between computational time and accuracy you would like to achieve.

Best regards, Marie-Therese

[IBRAHIM]

Thank you very much.

Ibrahim

[IBRAHIM]

Hi,

Please, should ISYM and vasp version be similar in step 1 and 2 for compatible WAVECAR to be used in step 2? Usually I start HSE+SOC from scratch with converged CONTCAR from HSE run.

Best regards,
Ibrahim

Hi Ibrahim,

Yes, the workflow you descibe is correct.
1. Do a ground-state-DFT calculation to preconverge.
2. Converge HSE+SOC with NKRED
3. Obtain the band structure with KPOINTS including zero-weighted points without NKRED.

Just FYI, for the latest version of VASP 6.3.0, there is an alternative way using KPOINTS_OPT.

Best regards,
Marie-Therese

[marie-therese.huebsch]

The purpose of step 1 is to produce a good starting WAVECAR for step 2 in order to reduce the computational time until HSE+SOC is converged. So the short answer is yes, it is best you use vasp_ncl (necessary because of SOC) and the same value for ISYM.

Basically, the INCAR of step 1 and 2 should only differ in the HSE related tags. You can also first converge without SOC and then switch it on. In generell, the convergence is best, when including effects on the largest energy scale first (and on the smallest energy scale last).

By the way, it might be expedient to refine the converged CONTCAR by symmetrizing your structure. During the structure optimization, numerical uncertainties will yield slightly different values for the ionic positions than the high symmetry Wykoff position. Of course, VASP searches for the symmetry operations that can be exploited with a certain tolerance (c.f. SYMPREC), so it is not strictly necessary to refine the structure. Nevertheless, when you symmetrize the structure (using a 3rd party tool for instance spglib or pymatgen) you can avoid these numerical uncertainties to play a role in the subsequent HSE+SOC calculation.

[IBRAHIM]

Thank you very much for the detailed explanation.

Please, in the last paragarph, do you mean that it is reasonable to apply symmetry (for example ISYM 2) to get converged CONTCAR to be used for the SOC steps in which we must use ISYM -1?

Best regards,
Ibrahim

[marie-therese.huebsch]

Actually, I do not know enough about your system to answer this question. The most general calculation would be not to use symmetry during structure relaxation and preceding 3 steps that we discussed above. But if you know certain details about your system from the experiment, then exploiting the symmetry during the relaxation helps to reduce the computational cost.

[IBRAHIM]

Thank you very much. I got what you explained in the last paragraph.

Thanks it is very helpful discussion.

Best regards,

Ibrahim

The purpose of step 1 is to produce a good starting WAVECAR for step 2 in order to reduce the computational time until HSE+SOC is converged. So the short answer is yes, it is best you use vasp_ncl (necessary because of SOC) and the same value for ISYM.

Basically, the INCAR of step 1 and 2 should only differ in the HSE related tags. You can also first converge without SOC and then switch it on. In generell, the convergence is best, when including effects on the largest energy scale first (and on the smallest energy scale last).

By the way, it might be expedient to refine the converged CONTCAR by symmetrizing your structure. During the structure optimization, numerical uncertainties will yield slightly different values for the ionic positions than the high symmetry Wykoff position. Of course, VASP searches for the symmetry operations that can be exploited with a certain tolerance (c.f. SYMPREC), so it is not strictly necessary to refine the structure. Nevertheless, when you symmetrize the structure (using a 3rd party tool for instance spglib or pymatgen) you can avoid these numerical uncertainties to play a role in the subsequent HSE+SOC calculation.