Dear All,
I have a problem of understanding a calculation result. I did a CI-NEB calculation of a phase transition. It turns out that I have barrier more than 1eV using LDA EXC and when I did the same calculation with GGA-PBE, it turns out that I don't have any barrier at all. It a system with 96 atoms of Zirconia.
I always thought in a simple argument, using different exchange and correlation functionals is a matter of accuracy in calculation results not total changing the interpretation of a result in such cases like one. I have attached here the results. So can any one try to understand my problem and give me some comments. I might have missed some setups in my calculations as well.
Here is my INCAR for reference and attached are the plots of the two different results.
I am looking forward to hearing from you all and thank you all in advance.
With best regards,
Abel
LDAvsGGA- CI-NEB
Moderator: moderators
-
abelazeb66
- Posts: 2
- Joined: Tue Sep 01, 2015 1:55 pm
LDAvsGGA- CI-NEB
- Attachments
-
- NEB-POSTED.tar
- result
- (5.67 MiB) Downloaded 730 times
-
- result
- LDAvsGGA-NEB-posted.png (8.39 KiB) Viewed 8193 times
Re: LDAvsGGA- CI-NEB
Hi Abel,
The difference is not in fact due to the functional. Instead, it has to do with the fact that you have no constraints on translation in the cells. In the GGA calculation, you can see that the atoms translate in the cell. The increased length along the path, due to the translation, means that the resolution of the minimum energy path is poor; so poor in fact that you have not resolved the saddle point. Specifically, in the GGA calculation, your first two images are in an initial-state-like geometry and your final five are in a final-state-like geometry.
You have a couple of choices to fix this. Perhaps the most simple is to take your LDA calculation and reconverge it with GGA. Since you have a configuration near the saddle, it should converge nicely without problem giving (I expect) a similar barrier with small differences related to the functional.
A second option is to restart your GGA calculation with at least one atom frozen in the box. This should prevent translation and help converge to a minimum energy path of the shortest length.
The difference is not in fact due to the functional. Instead, it has to do with the fact that you have no constraints on translation in the cells. In the GGA calculation, you can see that the atoms translate in the cell. The increased length along the path, due to the translation, means that the resolution of the minimum energy path is poor; so poor in fact that you have not resolved the saddle point. Specifically, in the GGA calculation, your first two images are in an initial-state-like geometry and your final five are in a final-state-like geometry.
You have a couple of choices to fix this. Perhaps the most simple is to take your LDA calculation and reconverge it with GGA. Since you have a configuration near the saddle, it should converge nicely without problem giving (I expect) a similar barrier with small differences related to the functional.
A second option is to restart your GGA calculation with at least one atom frozen in the box. This should prevent translation and help converge to a minimum energy path of the shortest length.
-
abelazeb66
- Posts: 2
- Joined: Tue Sep 01, 2015 1:55 pm
Re: LDAvsGGA- CI-NEB
Dear Graeme,
Thank you for overlooking my problem and I will implement the suggestions and come back with the result.
With best regards,
Abel
Thank you for overlooking my problem and I will implement the suggestions and come back with the result.
With best regards,
Abel