Page 1 of 1

Convergence criterion of climb NEB calculation

Posted: Wed Mar 31, 2021 8:51 am
by can_liu
Dear Support,

I am doing a climb NEB calculation with 5 images. The convergence criterion is set as: EDIFF=1E-05. EDIFFG is defaulted, which should be 1E-04. Now the change in the total energy between two ionic steps is often smaller than EDIFFG, however, the calculation doesn't stop. I'm wondering whether the energy change is the only criterion for convergence, or, there is any other problem. It will be really appreciated if I can get some suggestions.

The input files, OUTCAR and OSZICAR of each images are attached. Due to the size limitation, they are packed into two zip files (
report1.zip
report2.zip
). Thank you.

Re: Convergence criterion of climb NEB calculation

Posted: Mon Apr 12, 2021 10:48 am
by ferenc_karsai
In a NEB calculation not only the forces but also the stress tensor components need to converge to be below EDIFFG. But they are not so easy to read for the user since they are scaled to be able to be compared with EDIFFG.
So don't wonder if the calculation continues after the energies and forces converged, most likely your stress is not converged.

We also know that EDIFFG for stresses is not well transparent for the user and we put it on our to do list to have a separate converge criterion for the stress tensors.

Re: Convergence criterion of climb NEB calculation

Posted: Tue Apr 13, 2021 8:25 am
by can_liu
Thank you very much for your kind reply.
So in this case, should I wait until stress is converged, or may I stop the calculation and use the result when only energies and forces reach my criterion? Does the slow convergence (or unable to converge) in stress tensors mean the model structures are not good?

Re: Convergence criterion of climb NEB calculation

Posted: Tue Apr 13, 2021 9:36 am
by ferenc_karsai
I never conducted NEB calculations myself, but experienced colleagues who did reported the same behaviour as you.
I would wait until the stress components are also converged, otherwise you may get some uncontrollable instabilities in your calculation.

There is nothing wrong with your calculation only that EDIFFG is a bad tag for monitoring of the stress components.

Re: Convergence criterion of climb NEB calculation

Posted: Wed Apr 14, 2021 7:01 am
by can_liu
Thank you very much for your suggestion. I will wait for genuine convergence.