SOC and LDIPOL = T

[aedstrom]

I get unexpected results when calculating the electric polarization for non-collinear spin conigurations, including SOC, with LDIPOL = TRUE.
Are these options well tested in combination with each other and compatible?

[marie-therese.huebsch]

Hi,

thank you for sharing your concerns. I will investigate the extent to which this combination of tags has been tested.
In the meantime, could you please specify unexpected results and describe your workflow? Please also upload the appropriate files according to the forum guidelines, so that we can reproduce the problem and debug (if there is a bug).

Cheers,
Marie-Therese

[aedstrom]

Hello,

Thank you Marie-Therese for the response.
I attach example input files for my calculation.
The calculation is for a monolayer of CrI3, with vacuum separating the periodic images of monolayers.

I first performed the calculation with LDIPOL = F, and then started the calculation with LDIPOL = T from the previously converged calculation.

The unexpected results are as follows:

When using LDIPOL = T, I get one polarization from the Berry Phase calculation (LCALCPOL = .TRUE.) and another from the dipole correction routine. These do not exactly coincide, and the values do not seem to converge to each other when increasing nr of k-points or amount of vacuum. Note, however, that the polarization I calculate is tiny and maybe there is other numerics limiting this accuracy?

Moreover, the polarization value (from the Berry phase calculation) obtained with LDIPOL = F does not coincide with the values mentioned above. This may be expected, and indeed the convergence with respect to amount of vacuum is worse when LDIPOL = F. However, I would expect the large vacuum limit to converge towards the values obtained with LDIPOL = T. This is not the case.

thanks and best regards!

[marie-therese.huebsch]

So, I see you not only consider noncollinear magnetism with SOC and LDIPOL=T. You also use LDAU and constrained magnetic moments. That should be fine, but what worries me is that your structure has the space group P-31m (no. 162). That is a trigonal crystal system and not cubic. On the VASP Wiki, it says:

VASP will stop if the supercell is not cubic and LDIPOL=.TRUE.

I need to investigate if this restriction is still present and, if so why you did not get an error message.

Could you please share the stdout and OUTCAR file?

Best regards,
Marie-Therese

[marie-therese.huebsch]

I checked the code and the system only needs to be cubic if the system is charged. So it should be fine for your calculation.

Could you please upload the output your calculation generated with LDIPOL = T and LDIPOL = F?

[marie-therese.huebsch]

Hi Aedstom,

So, after some internal discussion, we expect that the in-plane polarization should be computed with LCALCPOL = T, and the polarization perpendicular to the monolayer should be computed with LDIPOL = T. We have thus far not tested how the Berry Phase implementation (LCALCPOL = T) behaves in case of extended vacuum, e.g., for slabs or isolated molecules though it should work.

We would like to investigate this problem, but before we invest the time, we would like to kindly ask that you upload the stdout and OUTCAR in accordance with the forum guidelines. Thank you for understanding.

Best regards,
Marie-Therese

[aedstrom]

Hello Marie-Therese,

Thanks a lot for your help!
I attach stdout and OUTCAR for the calculation.

Best regards,
Alexander

[aedstrom]

Hello Marie-Therese,

The previous output I posted was with LDIPOL=.TRUE. Here I attach output also for the same calculation but with LDIPOL=.FALSE.

Best regards,
Alexander

[marie-therese.huebsch]

Dear Alexander,

Just an update: I am still investigating this issue.

Previously, I mentioned that we had not tested how the Berry Phase implementation (LCALCPOL = T) behaves in case of extended vacuum, e.g., for slabs or isolated molecules, though it should work. Therefore, I have set up a test where two isolated water molecules are in a box. Initially, they are facing each other such that the system is centro-symmetric, the relative angle between them is 180dgr and the dipole moments cancel each other. Then, one of the water molecules is rotated slowly by π which yields twice the dipole moment of a single isolated water molecule. We can confirm that the total dipole moments computed by LDIPOL and LCALCPOL are within reasonable agreement.

plot.png

I hope this update is somewhat helpful to you already, although it is not immediately concerning your system. I will now continue to look into your specific output. Thank you for your patience!

Best regards,
Marie-Therese

[marie-therese.huebsch]

Dear Alexander,

Note, however, that the polarization I calculate is tiny and maybe there is other numerics limiting this accuracy?

Regarding the monolayer CrI3, as you mentioned the polarisation is tiny. And actually I would say it is vanishing, so the remaining comments are only relevant for future calculations where you expect a non-zero dipole moment.

The calculation is for a monolayer of CrI3, with vacuum separating the periodic images of monolayers.

First, you did not set IDIPOL = 3. This is necessary because the integral expression evaluated for LDIPOL=T is not applicable in plane of the monolayer. There only the Berry-phase expression (LCALCPOL=T) is valid.

Moreover, the polarization value (from the Berry phase calculation) obtained with LDIPOL = F does not coincide with the values mentioned above.

Second, there is a difference in the SCF solution between LDIPOL=T or F because it switches on corrections to the potential and forces. In fact, due to the periodic boundary conditions the total energy, the potential and the forces converge slowly with respect to the size of the supercell.This effect can be counterbalanced by setting adding dipole corrections. The biggest advantage of this mode is that leading errors in the forces are corrected, and that the work-function can be evaluated for asymmetric slabs. The disadvantage is that the convergence to the electronic groundstate might slow down considerably (i.e., more electronic iterations might be required to obtain the required precision).

Finally, do not forget that the Berry-phase expression can only compute a change in polarisation. So, it is necessary to first find a reference system with zero polarisation, as I did for the water molecule, and then adiabatically change the system into the system of interest.

Do these explanations make your results less unexpected? Let me know if there are any more questions. I will give my best to follow up more quickly.

Best regards,
Marie-Therese