UT theoretical chemistry code forum

Dear Prof. Henkelman and others,

I have been having issues with the convergence of a NEB calculation. I am attaching my calculation folders to this post. Any help to overcome convergence issues are welcome. Here are my calculation details:

I have been performing the NEB calculation in three stages of 75 steps each. My settings for each stage are as follows:

I create new intermediate images and start the NEB calculation with these images.
Stage 1:
IBRION: 3
POTIM: 0.01

I cleaned the previous NEB run with vfin.pl and continue stage 2 NEB with the results of the last stage NEB images.
Stage 2:
IBRION: 1
POTIM: 0.1

In the above two stages, I have not used CI-NEB. In the third stage, I again run vfin.pl on the results of NEB-2 and continue stage 3 NEB. I allow the force to go down to a very low value of (<0.05, My actual EDIFFG is -0.01) with these two steps. Now, I start step 3 CI-NEB with following settings:
Stage :3
IBRION = 3
POTIM = 0.0
LCLIMB = .TRUE.
IOPT = 1
INVCURV = 0.001

I see that forces go back to higher values in stage 3. Also, between the steps in stages 1 and 2, forces and energies decrease very smoothly with no large fluctuations. In stage 3 I see very large fluctuations in the forces.

Also, another point to mention is that after stage I, I saw one surface atom raising up in z-direction. For step-2 I fixed the z-degree of freedom in the POSCAR files.

Thank you in advance for your help. All calculation and results attached.

The biggest problem with these calculations is that you are converging towards a path in which one H atom dissociates from CH4 to the surface leaving a radical CH3 above the surface, which then sequentially adsorbs. That is why you are seeing a barrier over 2 eV -- the surface is not catalyzing the C-H bond breaking. The mechanism of CH4 dissociative adsorption should have a CH bond breaking over a surface atom which is pulled out of the surface.

Don't worry about the optimizer yet - it is tricky to use IBRION=1 or IOPT=1 when you have very soft modes in the system. Instead, go with your Stage1 approach and try to get a sensible path and barrier. You may need to construct an initial path in two segments: initial -> intermediate saddle guess -> final state. When things look reasonable, you can play around with more efficient optimizers.

Other points:

- Make sure that your frozen atoms are in the same position in every image - they are not in your calculation.
- Freeze the bottom two layers
- Unless absolutely necessary, use a minimal (e.g. 2x2x1) kpoint mesh, PREC=Normal, and a cutoff of ~300 eV with soft potentials to efficiently find transition states. It is easy to reconverge a known saddle with more accurate settings.

Dear Prof. Henkelman,

Thank you for your critical analysis of my job and the helpful suggestions to improve them. When you say [quote]
.... one H atom dissociates from CH4 to the surface leaving a radical CH3 above the surface, which then sequentially adsorbs. That is why you are seeing a barrier over 2 eV -- the surface is not catalyzing the C-H bond breaking.
[/quote]
do you mean that my co-adsorbed system CH3 + H, which I am using as a final image in the construction of the intermediate images of this NEB job, is not right? I'll re-submit this co-adsorption job again to see if I can get a better estimate of the end point guess of my job.

Thanks again!

Best regards,
Mmg016

It's not a problem with the final image. The issue is that you are converging to a reaction pathway in which the C-H bond breaks before the molecule adsorbs to the surface. You might have to construct an intermediate image closer to saddle point for dissociative adsorption. You might find the geometries in one of these references helpful:

http://theory.cm.utexas.edu/henkelman/p ... 044706.pdf
http://theory.cm.utexas.edu/henkelman/p ... 01_664.pdf