Inconsistent dipole moment value for 2d ferroelectric material

[yujia_teng]

Dear admin,
I found that for 2d FE material, the dipole moment along z direction (vacuum direction) obtained by charge integration and berry phase difference w.r.t to the centro-symmetric phase are different. I've attached the tested result here.

In principle these two value should agree with each other, but in my test, p_berry always has larger magnitude than p_charge_integrate, with p_berry ~ 1.5 p_charge_integrate for FE phase. And the difference gets larger with larger structure distortion.

My test is inspired from this post forum/viewtopic.php?p=21643#p21643. They tested on water molecules and obtain reasonable agreement. And the admin there said "polarization perpendicular to the monolayer should be computed with LDIPOL = T". But I'm still don't fully understand why and is still confused.

Best wishes

[fabien_tran1]

Hi,

Could you please also provide the files for the calculations using the berry-phase method?

[yujia_teng]

Sorry I didn't realize that the files are missing for berry phase. I've re uploaded them below. Berry phase method is under polarization directory.

[fabien_tran1]

Hi,

Sorry for the late answer. According to tests, the difference between the two methods seems to come from the vacuum that needs to be very large for the Berry-phase method. Taking the distorted structure "100" as an example, an extrapolation of the dipole moment (with Berry-phase method) at infinite vacuum (using values calculated at c=28.75, 50, 60, 70 and 80 Angstroms) gives 0.096, which is quite close to 0.093 from the charge-integration method.

This large dependency on the vacuum size with the Berry-phase method is at the moment not clear to us. In any case, it is recommended to use the charge-integration method when possible, as in your case.

Besides, I noticed that you use "tt" to comment out lines in your INCAR files. Where do these "tt" come from? The character "#" should be used for all comments and deactivated lines in the INCAR file. Example:

Code: Select all

# Startparameter for this Run:
# NUPDOWN=4
 EDIFF  = 1.0E-06             # stopping-criterion for ELM

[yujia_teng]

Dear admin,
Thanks for your tests, and I also have the following new thoughts on this problem: So the tests are using 1 k-point along vacuum direction, but Berry phase calculation is based on derivative w.r.t to k-points. So 1 k-point may be the source of problem. And I did a test with increased k-points along vacuum direction, changing 1 to 5 and it matches well. The figure is attached. Do you think this is a solution?

The 'tt' is probably a comment in old versions. It works in vasp 5.4.4, but in 6.4.1 it would give warnings. This input file was inherited from my previous group and it might have been in use since way back. So I didn't change this 'tt'.

[fabien_tran1]

You are absolutely right. Using more than 1 k-point in the vacuum direction is a solution, which seems more efficient and practical than increasing the vacuum. This is what Georg Kresse suggested me today after I posted my previous message.

However, keep in mind that the Berry-phase method may possibly lead to strange results, so that using the charge-integration method is preferable (but applicable only in the presence of vacuum).