Problem with convergence

Vasp transition state theory tools

Moderator: moderators

Post Reply
shs28
Posts: 1
Joined: Thu May 18, 2017 2:25 pm

Problem with convergence

Post by shs28 »

Dear Prof Graeme Henkelman
I’m new to NEB calculations. I am studying the migration of an interstitial atom from a less stable site to a most stable site in an Al-layer. I’m using VASP 5.4.1. Some of my calculations converged without a problem. But others are taking unexpectedly longer (more than 5 24h-runs on 960 cores ). I *think* that in my calculations the path is relatively simple. I always start with IOPT=3 (QM) and switch to IOPT=1(LBFGS) when the forces are around 1 eV/A. I would be grateful if you could have a look at my calculations and suggest some solution for the slow convergence. I have also included (vaspgr for each continuation run) to check the convergence for different runs if it is of any use.
Please let me know if you need any further information.
Thanks in advance,
Shafqat
Attachments
C1.tgz
(64.37 MiB) Downloaded 892 times
Al2.tgz
(55.47 MiB) Downloaded 950 times
Al1.tgz
(62.82 MiB) Downloaded 970 times
Zr1.tgz
(52.42 MiB) Downloaded 1000 times
graeme
Site Admin
Posts: 2256
Joined: Tue Apr 26, 2005 4:25 am
Contact:

Re: Problem with convergence

Post by graeme »

The geometry of these paths are all over the place. It's not clear to me how they were generated to know if there was a problem right from the start.

Anyway, make sure you look at a movie of your initial path to make sure that it makes sense. Check that there are no errors in the labeling (or ordering) of atoms between your initial and final states. Make sure that there is no net translation of the cell. You might want to freeze one atom away from where the diffusion is taking place. This atom should be frozen in the same position in your initial and final structures and then in every neb image. Finally, run with IBRION=3 and make sure that your initial convergence steps have a reasonable force (i.e., no atoms running into each other) and that the force drops steadily.

I expect that if you solve any problems with how the paths were constructed then convergence will go better.
Post Reply