EFIELD ignored when changing atomic ordering in POSCAR (VASP 6.5)

[soungmin_bae1]

Dear VASP team,

We have encountered a bug in VASP version 6.5.1 (and earlier versions, e.g., VASP 6.4.1) when applying an external electric field via the EFIELD tag. Although we only modify the atom ordering in our POSCAR (leaving the underlying geometry completely unchanged), reordering the atoms causes a significant decrease of the electric field applied to the slab, which appears to be a bug. I have illustrated the details in the attached PNG file (bug‑illustration.png).

We have also attached the input and output files used for these calculations (input‑output‑files.zip) and the macroscopic average potential plot (plot‑macroscopic‑pot‑averaged‑b‑axis.pdf).

We are reporting this issue because it can lead to significant errors when calculating slabs with substitutions or vacancies under an electric field: the field is similarly reduced (or effectively nullified) when using the rearranged POSCARs.

Thank you for your time and assistance.

Best regards,
SB

[ahampel]

Dear Soungmin,

thank you for reaching out to us on the official VASP Forum and reporting this problem. I can confirm the problematic behavior with your input files. I am concentrating now on the original slab POSCAR at the one in which you just split the Si atoms into two groups (POSCAR-02) and added a corresponding entry in the POTCAR. When removing dipole corrections and efield (as you write) calculations converge to the same energy at least, but I can already see that the convergence and mixer works differently which is unexpected.

With e-field applied the original slab POSCAR converges fine (also with ALGO=Normal and adjusted mixing parameters). Even when starting from the resulting WAVECAR the POSCAR-02 calculation diverges. While the ALGO=All calculations seemingly converge they do so to different solutions. I will keep analyzing the problem but it seems that there is already a problem without applied e-field. I do not see a reason why the calculation should behave numerically differently when splitting the Si atoms into two subsets. I will keep digging and keep you posted.

Best,
Alex

[ahampel]

Dear Soungmin,

we figured out the problem. I created a known issue: https://www.vasp.at/wiki/index.php/Known_issues :

EFIELD tag and symmetries: When applying an EFIELD the inversion symmetry along the EFIELD direction specified via IDIPOL is not removed if present and no dipole moment can form. Results of such calculations are wrong and manually removing symmetries via ISYM=0 is strongly advised.

So the problem is actually that the calculation that you successfully converged with the POSCAR-perfect is giving wrong results. This can be quickly verified by checking the reported dipolemoment in the OUTCAR file which is 0 . This is clearly wrong, since applying an electric field, even in a symmetric slab, should create a dipolemoment. What happens when you split your atomic species in two groups is that this breaks symmetry of the slab and the wrongly assumed inversion symmetry along the field direction is removed. I realized this when performing a calculation of the "perfect" POSCAR with ISYM=0. This gives the same results as POSCAR-02! Then I realized that ALGO=ALL incorrectly signaled convergence here, which led you to believe the calculation is okay.

I attach corrected input files that should be performed in the following order:

• run folder 1. which uses now ALGO=Normal with adjusted mixing for best convergence. Copy the resulting WAVECAR file to folder 2-...
  run folder 2. which now activates dipole corrections. Here the dipole moment is 0, since the slab is symmetric along axis b. Copy WAVECAR to folder 3-
  run folder 3. which activates additionally the e field. This has now a dipole moment that converges slowly to 1.15... copy WAVECAR to folder 4
  run folder 4. which is the same as 3 but holds the changes to the POSCAR (note that here is no ISYM=0 tag in the INCAR), this calculation should now converge quickly to the same result as folder 3

Here is the plotted locpot from these calculations:

locpot_fix.png

You can see a stark difference between POS0 (perfect) ISYM=2 and ISYM=0 results. ISYM=2 is the default and what you found. In contrast the potential is much weaker (but still tilted!) with ISYM=0. Now also POS0 and POS2 results lay on top of each other.

Let me know if this makes sense to you. We internally implementing a fix that automatically will remove the symmetry along the efield direction in the upcoming VASP version. Thank you again for bringing this to our attention!

Best,
Alex