climbing image tag

[kai]

I am now facing a problem that I am suspecting to be related with the "climbing" part of the NEB calculation.

My current cacluation seems to show two barriers between my initial and final states. Anoyingly, these two barriers seem to be quite close in energy (~0.1 eV), making it difficulty for NEB to find the right image to "climb". From the calculation history, i believed that NEB is struggling to climb either image #2 or #4, preventing the whole calculation to go down to 0.05 ev/A in force from a pre-converged whole set of NEB (0.1 eV/A)

I have two ideas in my mind that welcomes any comment:

#1
Is is a good idea in this case to swictch off the climbing part of the NEB to let the system converged to 0.05 first, and then from there re-run the calculation with the climbing on?

#2
or shall I break down this overal process into two separate NEB calculation, that is one for barrier #1 and one for barrier #2; if this works, do i then need to combine those two sets and re-do an overal NEB?

Thanks

[graeme]

Breaking the path into two is probably the better strategy since you know there will be no competition between the two barriers, and each path will be shorter so the saddle can be found with fewer images.

At one point we talked about putting in a flag to climb/descend to all local extrema along the path. If this problem is showing up a lot, we can put that in.

[kai]

thx, graeme

could you tell me a bit more how it determines to push which image to saddle point in the current versioned neb? is it as simple as pushing the highest energy image at certain iteration?

it would be lovely to have such a tag to find all the extremes; but based on my reading of the old posts, isnt that gonna cause some problem with the convergence?

[graeme]

The climbing image simply moves the highest energy image to a saddle. Optimizing all extrema should not reduce the convergence rate if there are sufficient images. It would, however, make the path somewhat longer which could increase the number of required images. We also felt that you can efficiently do minimization calculations separately from an NEB calculation. But having it as an option could be useful.

[kai]

indeed, now i managed to converge everything down to 0.02 eV/A by breaking them into two parts for two individual climbing NEB;

Thanks