NEB Convergence

[aarondesk]

I know this has been discussed in previous threads, but it's still not entirely clear to me. When is convergence reached for a NEB calculation?

I have a calculation that converged (EDIFFG = -.1E-1). The following are in one of the OUTCAR files. How do I interpret this?

NEB: forces: par spring, perp REAL, dneb 0.004116 0.024379 0.000000
NEB: distance to prev, next image, angle between 1.360366 1.361189 150.233483
NEB: projections on to tangent (spring, REAL) 0.004116 -0.053914

FORCES: max atom, RMS 0.007235 0.001949
FORCE total and by dimension 0.024724 0.005644


Thank you,
Aaron

[dano]

When you set (EDIFFG = -.1E-1) in the INCAR this means that the NEB will converge when the maximum Fneb (force with neb projections and spring forces) in any direction falls below that threshold of 0.01 eV/Ang

The line in the OUTCAR that can be used to monitor this convergence is:

FORCES: max atom, RMS 0.007235 0.001949

Here first number is the maximum force on any atom in the image. For convergence to be reached that number in every image has to fall below the prescribed force criteria.

You can monitor convergence of an neb by running the nebef.pl script (from vtstscripts) in the same directory as the neb.

Dan

[aarondesk]

[quote="dano"]The line in the OUTCAR that can be used to monitor this convergence is:

FORCES: max atom, RMS 0.007235 0.001949
[/quote]

Thank you for the response.

It's still not clear to me where exactly this number is coming from (force max atom). I'm trying to see how all the numbers in OUTCAR add up and give this number.

For example, here is the total forces on one atom.
4.68002 1.97888 13.78231 -0.033048 -0.042499 0.006040

Here's the line for the tangent force on that same atom.
0.66678 0.69167 -0.03756

How do these numbers relate to the criteria for convergence (Fneb)?

Thank you.

[graeme]

We don't print the full (atom-by-atom) NEB force, although we could do this. You can calculate it from the information in the OUTCAR. To do this, you need the force (F), the tangent (T), and the distance to the next and previous images (Rprev,Rnext) and the NEB spring constant (k). Then you can evaluate:

Fneb = F - (F*T) T + k (Rprev-Rnext)^2 T

We could also give you a print statement for NEB.F to print this force.

Also, note that Fneb is different for the climbing-image.

[aarondesk]

Thank you for that explanation. It makes sense now.

Aaron