Thanks Aidan for your advice...
Looking more carefully at my calculations I found some very weird result. In one of my neb calculation I got one atom which did desorb from the surface after around 200 neb steps... This could happen because of a too large force at some point (it is strange that it did not happen at the first steps though)... but the worse is that the force on that atom is very large (around 6eV/A) and is taking the atom away from the surface even for the image for which the atom is at its maximal altitude...
The cutting distance for the interatomic interaction is very short so this atom away from the surface should just feel the string force and the string force should bring it back to the surface... I really do not understand what's going on. I suspect that it could somehow be related with my convergence issue...
Concerning your remarks Aidan, I agree with you that depending on the system size we might have paths of different nature which might make the convergence more or less difficult. But I think that whatever the initial path, the NEB should ultimately converge...
Actually the fact that I keep having a large value of "maximum force per replica" is not my main concern.
The real problem, I think is that I have a large value for "maximum force per atom" (which can be larger than 1 eV/A). From the documentation I understand (although it is not explicitly said) that it corresponds to the maximum
force component of any atom in any replica and that it does include the inter-replica forces. So when a NEB is converged, it should definitely be small.
Could please someone confirm that the maximum force per atom include the inter-replica forces.
Thanks again
Emile