Erratic behaviour of adsorbate during NEB optimization

[julien_steffen]

Dear VASP community,

I'm currently trying to calculate the activation barrier of different elementary reaction steps on several metal surfaces, i.e. adsorption,
dissociation etc., using VASP's NEB module.

Interpolation of the relaxed start/end points yields a sensible reaction path, however starting the NEB run quickly leads to the N2
molecule doing anything but what would be expected.
The first few SCF cycles converge normally, though after a few cycles, the forces between the images become extremely large and the nitrogen atoms are
anywhere but where they're supposed to be after basically exploding.
After some more steps, the optimizer will start having trouble converging until eventually the calculation aborts.

I've attached the inputs/outputs/trajectories of the N2 adsorption on a Co surface from the gas phase.
You can view the trajectory of the interpolated POSCARS in trajectoryPOS.xyz, as well as the VASP-optimized trajectory in trajectoryCONT.xyz.

In an attempt to troubleshoot, I've increased ENCUT above what's recommended for N atoms and also increased the number of k-points to 5 5 1.
The order of atoms is identical in the start/end POSCAR.
Selective dynamics is also switched on for the same atoms in both files (upper two layers and N2 are allowed to relax).
As visible from my INCAR, I normally utilize VTST's climbing image NEB , but switching it off didn't improve the behaviour of N2 during the NEB run.


The problem occurs for all other metals and reaction steps (i.e. dissociation from the adsorbed state), so none of my calculations on the metal surfaces are currently working.
This could indicate that something with my general settings might be faulty, but I've went over my inputs several times now and they seem fine to me.
I've preciously run the same type of NEB calculations on the surface of metal clusters using almost identical settings without any issues, so it's beyond me
what could lead to such an erratic behaviour of the N2.


Best regards

[manuel_engel1]

Hi Julien,

Sorry for the late reply. As you say, the nitrogen atoms move in a really unpredictable way. After analyzing the trajectory, it seems like the forces are already extremely large in the first step. I believe that this might be caused by MAGMOM. Try setting the magnetic moment of the nitrogen atoms to 0. This would also explain why the calculation fails for all other types of surfaces.

Let me know if it works.

[julien_steffen]

Hello manuel,
thank you for the suggestion.
I can see how specifying MAGMOM in this way (22*2), instead of seperately for each element (20*2 2*0) might be problematic.

I've tried setting MAGMOM = 20*2 2*0 in the INCAR, but the NEB calculation seems to be bahaving the same as before.


Besides Co, I'm also doing the same calculations with Fe, Mn, Mo and Ru surfaces. For these, I'm using:
Co, Fe: ISPIN = 2 with MAGMOM
Mn, Mo, Ru: ISPIN = 1 without MAGMOM

There's something I've noticed when looking at the individual trajectories:
For the spin-polarized calculations, the nitrogen atoms behave much more unpredictable than in the non-spin-polarized systems.
So the trajectory of N2 on the Co and Fe surfaces is much more erratic, i.e. the N atoms jump all over the cell.
Meanwhile for the non-spinpolarized cases, the N atoms somewhat stay where they're supposed to be, but they seem to oscillate around their center of mass.
Depending on the element, this behaviour is either more or less pronounced. Mn and Mo oscillate less strongly, Ru very strongly (to the point of N atoms fusing together).
Looking back at the trajectory on the Co surface, I might also be seeing some sort of oscillating behaviour.

I'll attach all of the relevant files for the trajectories I've observed:
Co: spin-polarized, N atoms jump all over the cell
Mn: non-spinpolarzied, N atoms lightly oscillate around COM
Ru: non-spinpolarized, N atoms strongly oscillate around COM

Best regards