Electronic converangce in a dielectric slab in presence of electric field

[rohan]

Hi,

I have a dielectric (of different supercell sizes) with surfaces. I apply an electric field with
LDIPOL = .TRUE.
IDIPOL = 3
EFIELD = 0.01 ~ 0.0000001 (eV/A, I have tried the range of values specified)

However, I am unable to complete even 1 self-consistent cycle. From the OSZICAR file, everytime the dE value reaches to 1e00 and then shooots up to 1e+6. The energies also go up to very high values of the ( > +1e5).

I dont have any problem relaxing the slab without an electric field.
I have tried the following things, none of which helped :
- decrease the mixing values (amix = 0.2, bmix = 0.0001)
- increases the number of bands (nbands = as high as 4 times the default value for any supercell)
- try different algorithms : IALGO = 38/48, Algo = Fast/Normal
- change the cutoff energies
- use different ISMEAR = -1/1/5
- play with different K-mesh - Monkh, Gamma
In all the cases I use IBRION = 2. But I cant even complete one SCF cycle.

I have read about charge sloshing from previous forum posts. But even decreasing the supercell size does not help.

I would be thankful to get any help. I know similar calculations are not a problem in SIESTA code, but I am unable to do it in VASP!

[rohan]

someone please reply

[mrm]

Hi Rohan,

I have never had to work with applied electric fields so my advice may not work but I will try. Based on the info you gave it leads me to believe that you are working with an "asymmetric" slab since you are using LDIPOL and IDIPOL to account for unphysical dipoles. If this is correct, can you try using a "symmetric" slab which would allow you to avoid the use of LDIPOL and IDIPOL? Is it necessary to use LDIPOL and IDIPOL when applying an external E-field? Perhaps the application of the E-field along with the dipole corrections is causing problems.

Also, what version of VASP are you using? I know that versions older than 4.6.34 (or something like that) had a bug where relaxations could not be performed correctly while dipole corrections were turned on.

Hope this helps.

<span class='smallblacktext'>[ Edited Sun Jul 11 2010, 03:45PM ]</span>

[frg213]

Hi,
I seem to encounter exactly the same problem. LDIPOL and IDIPOL seems necessary and I encounter the problem when performing a static run. My system is a nanoribbon and my VASP version is 4.6.31. Can anyone help? Thanks.

[spakinfo]

I'm encountering the very same problem. As far as I've learned so far, the problem is somehow due to LDIPOL tag. If I replace it with LMONO=.TRUE. convergence is achieved without doing anything with mixing parameters. However, I'm not sure if changing LDIPLE with LMONO would really be a solution when we are going to consider the effect of a sawtooth potential in our calculations. According to the manual, LDIPOLE should be turned on to avoid interactions between the periodically repeated images. So I'm not sure if LMONO can do the same thing! I hope Prof. Kresse or other guys will see this post and give a comment on this issue!

[rohan]

Whew.. Finally, I think I know what may have been going wrong. Although, now working on a different problem, I will go ahead and post my solution, hoping that it may help others. Here it goes:
My current system is an asymmetric slab of a polar oxide, where dipole corrections are crucial (to correct an electric field of 0.5 eV/Ang in the vacuum region, without dipole corrections). Although, I haven't applied any external electric field, I was not able to get any convergence during the SCF cycle.
With only LDIPOL = true and IDIPOL = 3 (vacuum in z-direction), tags, I noticed that the center of the charge distribution, determined automatically by the code, changed during the SCF cycle, due to which, the total energy changed quite drastically.
Once, I fixed the position of the center of charge using the DIPOL tag, I was able to get electronic convergence (very slow, nonetheless, it converged).
You can check, if you have the same problem in your systems, by looking for "pos" in the OUTCAR (grep "pos" OUTCAR). It should remain constant. If not, then use the DIPOL tag in INCAR to set it manually.