Dear all,
I have been using the NEB method for a while now, and I am wondering what are the pro's and con's to run NEB calculations between images of previous runs (provided these images are converged). My question arise from the need of a more fine sampling of the reaction path only around a certain region. I am aware of the existence of other methods that could solve this problem in a different manner, yet I would like to find out how valid it is to do so using the NEB/cNEB method. As an example I made an attempt to sample the reaction path of a radical attack on a slab (~200 atoms), that led to the energy profile (in attachment).
A first 5-images NEB run was performed between initial and final states as usual, then two 5-images additional ones between images 1-2 and 2-3, then three additional one between images 1-2, 2-3 and 3-4 of the "sub" NEB 2-3 previously mentioned. Eventually I found what I was looking for (i.e. the "bump" you see on the energy profile) and ran a cNEB to accurately determine the position and energy of this local maximum. Every NEB calculation was performed with EDIFF=5E-09 and EDIFFG=-1E-03, but endpoints of "sub" NEB were not relaxed and taken as they were computed on the previous NEB calculations (but of course endpoints of the initial NEB calculation were converged using the same criteria).
An analysis of the coordinates of the hydrogen atoms involved in this reaction shows that the reaction is well defined, with a smooth displacement of the atoms along the reaction path as one would expect for this particular reaction.
Are there concerns that should be raised with respect to the methodology I used ?
Best regards,
Emerick
PS: I am sorry if this is a duplicate topic, if it is could you refer me to the original question ?
NEB calculation between previously converged images.
Moderator: moderators
NEB calculation between previously converged images.
- Attachments
-
- Energy_profile.png (65.81 KiB) Viewed 26455 times
Re: NEB calculation between previously converged images.
Hi Emerick,
I think the method is fine, but if your goal is to find the minimum energy pathway after you have a saddle (which it seems that you do) then there could be an easier way. What could be more efficient is to run 4 minimization calculations: 1. from your initial state to the minimum at 0.4 (reaction coordinate value); 2. from the final state to RC ~0.57, and then 3,4 from either side of your saddle at RC 0.41. If you set a small value for maxmove in these minimizations, you can trace out the minimum energy paths to an arbitrary precision. Just make sure to use steepest descents (or quickmin) for the minimizations so that you stay as close as possible to the minimum energy pathway (RK4 would be better, but then you would need to use something outside of vasp to do the optimization). An example of this approach can be found here: https://theory.cm.utexas.edu/henkelman/ ... 134106.pdf (fig 10).
I think the method is fine, but if your goal is to find the minimum energy pathway after you have a saddle (which it seems that you do) then there could be an easier way. What could be more efficient is to run 4 minimization calculations: 1. from your initial state to the minimum at 0.4 (reaction coordinate value); 2. from the final state to RC ~0.57, and then 3,4 from either side of your saddle at RC 0.41. If you set a small value for maxmove in these minimizations, you can trace out the minimum energy paths to an arbitrary precision. Just make sure to use steepest descents (or quickmin) for the minimizations so that you stay as close as possible to the minimum energy pathway (RK4 would be better, but then you would need to use something outside of vasp to do the optimization). An example of this approach can be found here: https://theory.cm.utexas.edu/henkelman/ ... 134106.pdf (fig 10).
Re: NEB calculation between previously converged images.
Thank you for your quick answer and for the suggestion!