UT theoretical chemistry code forum

Hi Graeme,

Which one should I believe if the barrier calculated from the energy of a converged climbing image and the barrier from the interpolated MEP (generated by nebresults.pl) show disagreement between them?

If there is any discrepancy, I think you have to figure out what is going wrong. A converged climbing image should have the correct barrier, but if the interpolated MEP is giving a different barrier, it needs to be explained. There are some reasons that the interpolation can give poor results; if you post the data in the neb.dat file, that should be enough to see what's going wrong.

here is the neb.dat:

0 0.000000 0.000000 0.000285 0
1 0.475593 0.022042 -0.050330 1
2 0.951839 0.039069 0.019013 2
3 2.265268 -0.277310 2.775199 3
4 3.573114 -1.033588 0.270024 4
5 4.878093 -1.116706 0.001568 5

Do you have any frozen atoms in the system? If not, check to see if the entire system is translating (or rotating for an isolated system). This will not effect the forces on the images, but it will change the distance between images. In your case, the distance between images 2-3, 3-4, and 4-5 are longer than between 0-1 and 1-2. I would not be surprised if a translation of the system is causing this. If so, the spline will not give a sensible interpolation because the forces along the band are not consistent with the change in energy and the distance between images. You should not trust the interpolated barrier of ~0.35 eV because there are no images there. If you remove the translation, or free an atom and repeat the band, I expect that you will find an interpolated saddle which has an energy of the climbing image (0.04 eV).

I am very sure that my system has no translation since the bottom two layers of the surface slab are fixed. I have also conducted similar calculations with slight different slab which will give higher barrier (~0.2eV) to the same reaction. In that case, the images locate very well on the interpolated MEP, the maximum of MEP coincides with climbing image.

I would have to see the coordinates of the images or an entire .tar.gz of the calculation to see if there is some other problem.

I attached the whole folder. It is pretty big, I am too lazy to clean it up. sorry.

Right, clearly no translation problem. But there is a large distance between images 2 and 3 and a high force (along the band) at image 3 which is leading to a poor spline. I would not trust this barrier without subsequent calculations.

One possibility is to start a minimization at a position corresponding to image 2.1 along the band (interpolate 10% between image 2 and 3) to see if this minimizes to the final state. If it does, then I would believe the low barrier for this largely downhill process. It might also minimize to an intermediate minimum corresponding to hydrogen diffusion. If this is the case and there is an intermediate minimum, you might be missing the dominant saddle entirely because of the poor resolution of the minimum energy path between images 2 and 5. Adding more images, or a separate path between the intermediate minimum and the final state would sort this out

Another possibility is to run a bind with ~7 intermediate images to better resolve the entire path. Any time you have a distance greater than 1 Ang between images, I would be concerned.

thanks a lot. I'll follow your suggestions and give you feedback.

Although the calculations are still running, I am now pretty sure that my previous NEB calculation only spotted a low barrier transition state corresponding to H diffusion and totally miss the hydrogenation barrier. Thanks for your help.