using NEB medium accuracy output to perform high accuracy!

[aakbarz]

hi,

I am having trouble with my calcs. on Cu(111) surface.
The system has a vacancy on surface and one on the subsurface in the 1NN distance.
I want to see what the barrier height could be if I remove one atom from surface to the subsurface,
that is inter-plane diffusion of Cu atoms (in reverse how vacancy is moving)
as you see in the attached MEP files, I used the output after an NEB run with medium accuracy to perform
the next calcs. with high accuracy, however the diffusion barrier height is dramatically changed and
I am NOT sure if this can be correct! Can there be an issue if we restart NEB from Medium to High accuracy!
I am puzzled, any suggestions,

Thanks in advance,
Ali

[graeme]

Make sure that your endpoints have been reoptimized using high precision in that run. There could be significant changes in the absolute energies, but it is unlikely that you will find significant changes in the relative energies (the barrier).

We routinely reconverge NEB calculations using different levels of accuracy, energy cutoff, k-point sampling and spin polarization. Just make sure that you endpoint energies are calculated using the same parameters as the intermediate (NEB) images.

[aakbarz]

good point thank you very much,
However I forgot to mention that first I optimized the end points using the high precision accuracy,
and then I started NEB with medium accuracy and after that restarted NEB with high accuracy,
so my end point energies are same in both steps,
I thought this way I would get it converged faster!
Have I broken self-consistency of my calcs.? do you think that this is not a correct way to perform the calcs.?

Thanks,

[graeme]

The endpoints must be converged with the same settings as the NEB to calculate energy barriers.

[andri]

How low are the forces in the original band? Is it possible that you are not using the same POTCAR file in the two runs?

[aakbarz]

thanks for the reply.
But something which I do not understand is that the end points are fixed
for the NEB calculations and thus I thought we can speed up NEB calculations
if we perform the NEB by lower or medium accuracy and after we get some close to the MEP
increase the accuracy to high and thus get good MEP.

Andri, I used same POTCAR in both run., the forces are in the end points are as low as 0.01 eV/A
however when I started NEB I set forces to be as low as 0.03 eV/A,

thanks,