Page 1 of 1

spline

Posted: Mon Sep 11, 2023 12:21 am
by ibrahim
Dear prof. Henkelman,

I used CI-NEB to find the diffusion path with 5 images, in one calculation after convergence to <0.01 eV/Ang:

from neb.dat, the transition state is at image3:

0 0.000000 0.000000 0.000000 0
1 0.684453 0.240176 -0.443828 1
2 1.367635 0.405261 -0.069436 2
3 2.048908 0.445257 0.009086 3
4 2.758948 0.404111 0.059959 4
5 3.467756 0.263926 0.433952 5
6 4.175574 0.009124 0.000000 6

However, the spline shows one point with very little higher energy than the TS image
spline:
2.7 1.8445261 0.44050800777003 -0.0504218558159505
2.75 1.87858975 0.442095516162281 -0.0426438355499558
2.8 1.9126534 0.44340349785408 -0.0340098330026289
2.85 1.94671705 0.444402794964589 -0.0245198481739699
2.9 1.9807807 0.44506424961297 -0.0141738810639788
2.95 2.01484435 0.445358703918386 -0.00297193167265546
3 2.048908 0.445257 0.009086
3.05 2.08441 0.444768682577125 0.018306159166385
3.1 2.119912 0.4439655062966 0.0268235278942031
3.15 2.155414 0.442872421624575 0.0346381061834545

I am using CI-NEB and the calculation is converged, so it is supposed that the TS image will have the highest energy (image 3) as in neb.dat.

In another calculation, the spline shows more points (three or four) with very little higher energy (about 0.0004 eV) than the TS image, however, the calculation is also converged.

How can I get the TS image in these cases, is it that from neb.dat?, and what do those points in the spline mean?

Thanks

Re: spline

Posted: Mon Sep 11, 2023 1:02 pm
by graeme
I have to guess a little without seeing all the files, but I'm almost sure that I know what this issue is related to. The spline interpolation uses the force and energy for every pair of adjacent images along the path, as well as the distance between them. If there is a displacement along any soft modes in the system, such as translation, rotation, or low frequency modes, then the distance between images can be inconsistent with the forces and the change in energy. A clear example of this is for a system with no frozen atoms - I could just translate an image and increase the distance to the neighbors without changing any energy or force. But then the spline still needs to match the forces and the energy difference but over a longer distance, and this can lead to spurious extrema. Bottom line, I would not worry about this. The spline can help to identify local minima but it does not contain any data apart from the energy and forces at the images, as well as the distance between them. If you have a converged saddle, which you seem to, that is what matters.