Page 1 of 1

Electronic convergence failures

Posted: Wed Oct 25, 2023 7:22 am
by KWH
I have a couple of slab systems that will not converge electronically after 1 to 3 cycles of ionic optimization. I tried the recommended procedure of setting ALGO= NORMAL and ICHARG=12 and it did converge (see files ending in .12). I then set AMIX=0.1 and BMIX=0.01, ICHARG=2 and started with the WAVECAR from the 12 run. This stopped converging on the third cycle (see other files). I also tried BMIX=3.0 and AMIN=0.01 with no success.

Can you suggest a process that will lead to convergence? Of course, It may also be my mistake and I will be happy to be informed of what I am doing wrong.

Can you please notify me by email?

Convergence Failure problem additional information

Posted: Wed Oct 25, 2023 7:30 am
by KWH
My apologies. I forgot to add all the files to the rar. The one below is complete

Re: Electronic convergence failures

Posted: Thu Nov 02, 2023 9:27 am
by marie-therese.huebsch
Hi,

Thanks for your query about slab systems with a molecule on the surface!

The procedure to fix the charge density and see if the system converges is exactly the recommended first step.

From the .out file you uploaded, I can see that the first electronic minimization reached the maximum number of steps set by NELM instead of reaching convergence. Hence, the energies and forces used for the first ionic step are wrong. In fact, when you look at the column "dE", it is of the order of -0.47384E+00 after 180 electronic steps. In other words, the system is far away from convergence. There is no sense in performing an ionic step based on those values.

It might be that the incorrect setting for the dipole corrections is causing the bad convergence. I have centered the surface such that there is an equal amount of vacuum padding on either side.

Then, I have set

Code: Select all

  LDIPOL  = T 
  IDIPOL  = 3 
  DIPOL   = 0.5 0.5 0.5 
The last line was missing in your INCAR files. After 30 steps in the electronic minimization using the blocked Davidson algorithm with default mixing, the system showed reasonable convergence behavior:

Code: Select all

          N                 E                             dE                d eps        ncg        rms            rms(c)
...
DAV:  21    -0.974478735274E+02   -0.11938E-02   -0.68608E-04  4950   0.219E-02    0.231E-02
DAV:  22    -0.974495134916E+02   -0.16400E-02   -0.80145E-04  5038   0.268E-02    0.152E-02
DAV:  23    -0.974507110576E+02   -0.11976E-02   -0.53274E-04  5060   0.190E-02    0.819E-03
DAV:  24    -0.974512287959E+02   -0.51774E-03   -0.34607E-04  5192   0.162E-02    0.795E-03
DAV:  25    -0.974515953118E+02   -0.36652E-03   -0.37438E-04  4928   0.164E-02    0.450E-03
DAV:  26    -0.974517117399E+02   -0.11643E-03   -0.47358E-04  5016   0.176E-02    0.338E-03
DAV:  27    -0.974518163716E+02   -0.10463E-03   -0.32663E-04  5170   0.164E-02    0.387E-03
DAV:  28    -0.974519093428E+02   -0.92971E-04   -0.96430E-04  5170   0.182E-02    0.310E-03
DAV:  29    -0.974520778141E+02   -0.16847E-03   -0.20920E-03  5060   0.222E-02    0.236E-03
DAV:  30    -0.974519279494E+02    0.14986E-03   -0.27628E-02  4928   0.190E-02    0.197E-03
This is also not yet converged, but it is going in the right direction. By the way, at step 30 my test run displayed a dipole moment of the order of -0.6 eÅ along c. That is much less than the value you obtained by not setting DIPOL (-1.510774 eÅ along c in step 179).

I recommend the following:
  • Center the slab and set the proper tags for the dipole correction: LDIPOL, IDIPOL, DIPOL.
  • First, focus on the first electronic minimization. I have seen some evidence of charge sloshing in the test run using the blocked Davidson algorithm with default mixing that I performed. Thus, you may also want to adjust the mixing as described on the Wiki, see AMIX. dE should be monotonously decreasing. Small deviations from that are ok, but at least there should be some progress in the rms(c) column, i.e., the value should decrease.
  • We also recommend simplifying your calculation by reducing the number of tags specified in the INCAR to a minimum when facing convergence issues. For instance, in line 49, you set LVTOT = .FALSE. and in line 54, you set LVTOT = .TRUE. Be aware that VASP will use the first occurrence.
Could you please let me know if this solves the issue? The calculation is quite heavy, so I did not converge it all the way.

Cheers,
Marie-Therese
Screenshot 2023-11-02 at 10.17.37.png

Re: Electronic convergence failures

Posted: Fri Nov 03, 2023 5:47 pm
by KWH
Worked fine, thank you. Is it generally a good idea to center slabs in the supercell?

Re: Electronic convergence failures

Posted: Mon Nov 06, 2023 7:10 am
by marie-therese.huebsch
Hi,

Thank you for your confirmation!

Good question! Two aspects come to my mind:
  • It may break a symmetry if the slab is not centered. Hence, the calculation is more expensive. However, there are two buts: (1) For polar slabs (with a molecule attached to the surface), the symmetry is broken anyway. (2) And if you relax the structure, you should not enforce symmetry.
  • For polar slabs, it is easy to set DIPOL = 0.5 0.5 0.5. However, you can also determine the center of your slab and set DIPOL to that value.
In summary, there is no strong reason to center the slab that I can think of. So, my recommendation to center the slab was rather out of habit when using the DIPOL tag. If someone reading this thread has a different experience, I would be curious to hear it.

Marie-Therese