Page 1 of 1

NEB CONVERGENCE Not achieved (NSW=1000) and Internal Error with VASP 6.4

Posted: Sat Jul 15, 2023 4:15 am
by mjakhar
Dear vtst developers,

I am working on an 2D system on which a Li2S absorbed on surface. I am calculating the Li2S decomposition Barrier but some problems appeared.

First Issue:-
My calculation is not able to converged until 1000 iteration (NSW=1000).
I used ISIF=2 with optimization of intial and final steps.

Second Issue:-
| internal error in: vhdf5.F at line: 116 |
| |
| HDF5 call in vhdf5.F:116 produced error: 2 |
| |
| If you are not a developer, you should not encounter this problem. |
| Please submit a bug report.

Can you please check my INCAR files and suggest me the right path.

Re: NEB CONVERGENCE Not achieved (NSW=1000) and Internal Error with VASP 6.4

Posted: Thu Aug 31, 2023 6:48 pm
by naman_katyal
Hi

I looked at your calculation, and it's optimizing perfectly, though slowly, but that's the choice of the optimizer in your INCAR. If you look at the forces in each directory, they are all close to convergence. Usually, when I run NEB, I recommend optimizing the band without a CLIMBING IMAGE tag because its easier to converge regular NEB and then push one of the images to the peak of the band. Also, because you set POTIM = 0.1, you aren't using any VTST optimizer, which is fine if that's your choice. I would recommend turning off LCLIMB, using one of the VTST optimizers by setting POTIM = 0 and choosing IOPT fire or quick min or steepest decent.

Beyond that, forces started to fluctuate after a while instead of going down smoothly; another probable reason is your system.
I looked closely at your system, and it's possible that the band may never converge to ediffg criteria because you only have one layer of CN, making the potential energy surface extremely rough. I know it is a 2D system. Try the optimizer and climb solution first; if not, think about your system and if you can add another layer below to smoothen the convergence of band to ediffg because right now, the CN layer can bend in any way it wants which may/may not represent reality. Your optimized band looks good though.

The second error is not your fault and should not happen because you reached the limit of ionic optimization steps—maybe a supercomputer related issue.