NEB "converged" while |max(F)|>EDIFFG

[jeffrey_zom]

Hello,

I am doing NEB VASP calculations, and I don't understand why my calculations stop.
The OUTCAR says convergence has been reached. But comparing the forces printed in the OUTCAR compared to the EDIFFG set in the INCAR, it should not be considered converged yet, if I understand the settings correctly.

See for instance this example OUTCAR from 1 of the IMAGES (IMAGE 1 of the 7 I used):

Code: Select all

 NEB: forces: par spring, perp REAL, dneb     0.006448    0.010138    0.000000
 NEB: distance to prev, next image, angle between     0.067991    0.069281  158.366578
 NEB: projections on to tangent (spring, REAL)     0.006448    0.669386

 stress matrix after NEB project (eV)
      7.16373     -0.00000      0.07076
     -0.00000      8.12942      0.00000
      0.07076      0.00000      6.58806
  FORCES: max atom, RMS     0.005766    0.001343
  FORCE total and by dimension    0.012015    0.005625
  Stress total and by dimension   12.681446    8.129421
 OPT: skip step - force has converged

FORCE total and by dimension says 0.012015, while the EDIFFG says -0.01.
0.012015 > 0.01, yet the final line of that block says "OPT: skip step - force has converged"
if only perp REAL is used, 0.010138 > 0.01, so that would also not explain it.

• Why does VASP think the calculation is converged?

• Which value printed in the collective OUTCAR is considered for convergence?

I have attached:
- the example OUTCAR (of Image 1)
- associated INCAR
- figure showing "convergence" of the Images (perp REAL and energy as function of IMAGE and step index)
I used IBRION=3, POTIM=0, and IOPT = 7 (FIRE)
Note that I have seen this weird behaviour consistently for different IOPT settings, also for default VASP (IOPT=0, POTIM>0).

Kind regards,

Jeffrey

[michael_wolloch]

Hi Jeffrey,

Please confirm a couple of things for me, so I can help you better:

• Are you using transition state tools for VASP here (VTST)?

• Are the forces on the non-clamped images (not 1 and 7) also larger than EDIFFG? I am not sure from your graph, but it seems so. The clamped images might have larger forces, since they cannot move freely.

• You said that you have seen this on vanilla VASP NEB as well. Can you provide an example (does not have to be a production calculation, a minimal reproducible example would be ideal)?

Generally, my hunch is that the force you see is either the raw DFT-force or the elastic-band-force, but not the addition, which is what is actually checked, but since your OUTCAR is not standard VASP output, I don't know and wanted to ask what extension you are using.

For vanilla VASP, the usual

Code: Select all

TOTAL-FORCE (eV/Angst)

block is the raw DFT force and not checked in NEB calculations. However, you should see a block after

Code: Select all

CHAIN + TOTAL  (eV/Angst)

in your OUTCAR, that reports the summed up forces, which are then compared to EDIFFG.
They should be smaller than EDIFFG once the NEB clalculation is converged.

Please let me know if this already helps you with your problem, or provide some more information. I am not sure how far I am able to get with VTST itself; however, since we do not really provide support for that package, which is not maintained by us.

Cheers, Michael