Dear Professor Graeme Henkelman,
I am trying to do a NEB calculation of He on an W interface (W110/W112). However, even though I decrease the precision of EDIFF and EDIFFG more and more, the calculations do not converge. Also, to avoid overlapping atoms, I used the 'nebavoid.pl' script. I am using IOPT=7, that is suppose to be go for high forces (far from the minimum). I set NSW=100 and I am using 128 processors.
I attached .tar file with all the data used, in addition, all the files generated by 'nebresults.pl' script are append too. It'd be interesting using 'vfin.pl' in this case? What parameters would be interesting to modify?
Sincerely,
Jorge
NEB converge problem
Moderator: moderators
NEB converge problem
- Attachments
-
- 2down_NEB_lowprec.tar
- NEB FILES
- (28.79 MiB) Downloaded 390 times
Re: NEB converge problem
A few suggestions:
1. Do you really need such a large system? If you are interested in a He atom diffusing to a W interface, this is a fairly local event - why do you need such a large system.
2. Check the minimization of your endpoints. In my calculations, the initial state is not stable; there is a more stable state that you can relax to.
3. If you do need such a large system, you will need more images to resolve the path. Even in the calculation that I did, which is not converged, it is difficult to resolve the reaction path. This is coupled to the problem of a large system size and simply how long these calculations take.
Graeme
1. Do you really need such a large system? If you are interested in a He atom diffusing to a W interface, this is a fairly local event - why do you need such a large system.
2. Check the minimization of your endpoints. In my calculations, the initial state is not stable; there is a more stable state that you can relax to.
3. If you do need such a large system, you will need more images to resolve the path. Even in the calculation that I did, which is not converged, it is difficult to resolve the reaction path. This is coupled to the problem of a large system size and simply how long these calculations take.
Graeme