# polarization jump in berry phase method using VASP

When I use VASP to compute in-plane polarization of a slab with hexagonal cell (a axis along x axis, b axis is at a(-1/2,$$\sqrt{3}$$/2),polarization is along diagonal direction so I got it using $$p_x$$/cos30°), the result has a jump in polarization at two ends. I computed till 2% and 99% atom displacement. The structure is a slab composed of one quintuple layer and with vacuum region larger than 20A.

Related input parameters for slab and berry phase:
#Dipole correction
LDIPOL = .TRUE.
IDIPOL = 3
DIPOL = 0.5 0.5 0.6

LCALCPOL = .TRUE.

I'm really confused about these and couldn't find a explanation. But if I just use those middle points (2-99%) and extrapolate them to two ends, I can get the correct dipole moment value (same as paper).

• What was the reason for your edit today? Oct 23 at 8:15

This is quite a peculiar scenario and I think more information is needed in order to find the root casue of it.

With the information at hand, I think this might be caused by different k-samplings used for systems under different distortions. i.e., if you are using LCALCPOL=.T. with ISYM=2, VASP will try to reduce the number of k-points used in the calculation by utilizing the symmetry of the system. However, since LCALCPOL doesn't re-construct a string of k-points (unlike LBERRY which does exactly this), the k-points used to calculate the berry phase along each k-string might also be different if the symmetry routine decides that the system is now under a different symmetry than the other system.

I would suggest that you try:

1. switching off the symmetry by setting ISYM=-1, and
2. switching to LBERRY method which gives more infomation in the OUTCAR (e.g., coordinates of k-points along the k-string and the Berry phase for each k-string.)

For more information on how to use LBERRY, see: https://www.vasp.at/wiki/index.php/LBERRY.

I hope this helps.

• Thanks very much for your detailed explanation. I tried LBERRY and switch off symmetry but it doesn't make much difference. I attached the graph at the end of the main text. The jump is still there. All distorted structures have same space group and point group symmetry and jump occurs between distorted structure, so these may rule out effect related to symmetry. k-samplings should be same for all distorted structure. Jul 11 at 18:35
• I added some more details in the main text. the system is a slab. Related input parameters are provided there. I move the top 2 atoms to distort the structure as indicated in the original paper. If you need more info I can provide more. Hope this peculiar scenario can be solved. Jul 11 at 18:41

The problem is a bug in old version VASP that can occur in that unit cell because the non-polar phase polarization is not symmetric around 0. The jump happens where ionic contribution to dipole moment has a jump. After I switched to version 6.4.1, all FE structure are fine. But there's still jump from the 0% non-polar centrosymmetric one to distorted polar one. Although, it's value is exact half polarization, as expected. But to get the correct result, we can extrapolate to 0% position.

• To avoid being flagged as a "very low quality answer", I recommend to elaborate: maybe provide a comparison of the old and new output files. Maybe verify that it's a bug so that the answer doesn't start with "I think...". Jul 22 at 5:54
• Thanks. I've added the link of new results. Jul 22 at 14:08
• Link of new results? Oct 23 at 8:11