How to converge CI-NEB calculations faster?

[nrathnayake]

Dear Prof. Henkelman,

First, Thank you for developing the VTST tools.

I am trying to run CI-NEB calculations for expanded graphite system. I am using 3 TS for this calculation. For my calculations I considered diffusion of Na-ions in EG from hollow to hollow position along the C-C bond. However, this calculation did not converge and I experienced that, calculation takes lots of time to converge which I did not experienced in previous calculations. Following is the INCAR file I used to run these calculation.

ISYM = 0
ISTART = 0
ICHARG = 2
ENCUT = 600 eV
ALGO = FAST
EDIFF = 1.0E-5
EDIFFG = -1.0E-2
NSW = 1000
ISIF=2
ISMEAR = 0
SIGMA = 0.05
PREC = ACCURATE
LREAL = AUTO
LWAVE = .FALSE.
LCHARG = .TRUE.
IMAGES = 3
SPRING = -5
ICHAIN = 0
LCLIMB = .TRUE.
LNEBCELL = .FALSE.
IBRION = 3
POTIM = 0.5
TIMESTEP = 0.1
LSCALAPACK = .FALSE.

Could you please suggest, how to run CI-NEB calculations to converge systems faster using parallelization?

Thank you.

Kind regards,

Nimali

[graeme]

If you can post a .tar.gz file of your calculation, without CHG* WAV* files, I should be able to debug and suggest ways to speed things up.

[nrathnayake]

Dear Prof. Henkelman,

Thank you.

Please find the files of my calculation.

Kind regards,

Nimali

[graeme]

There are a few problems with your NEB calculation. First, you have frozen C atoms which are in different geometries along your band. Second, you have put a lot of resources into the electronic structure precision, but have only 3 images to find a saddle geometry. Finally, your POTIM is likely too large so that combined with the poor resolution along your path, convergence was unstable in your calculation.

In the attached I have used more reasonable settings to find a saddle. I have relaxed all of the C atoms and used 5 images to resolve the path. Interestingly, your initial and final states are 2 sites away and you can see in the relaxed path that there is an intermediate minimum. So the diffusion does not follow the path that you expected. Instead, the Na atom hops from hollow to neighboring hollow. You can see that the climbing image (04) found what I think is a monkey saddle, based upon the 3-fold symmetry. The start of the path goes from the initial state to the intermediate minima, but it is not forced to find a saddle because there is only one climbing image.

From here, I suggest that you break up the pathway into two, use more reasonable settings to find saddles and then you can increase kpoints, cutoff, and accuracy to check for convergence.

[nrathnayake]

Dear Prof. Henkelman,

Thank you very much for your explanation on my NEB calculations.

Also, for helping me to correct the INCAR file. Now, I understand the problem with my NEB calculation. I will try running the calculation as suggested and check.

Once again I am so grateful for you for your help to solve this problem.

Kind regards,

Nimali