Quite often i am running into a problem with one or two images fluctuating between two clearly resolved states, making the whole NEB calculation not converged to a satisfying extent. I read through the old posts regarding this but still have some questions, calling for professional help.
Generally speaking, there is no problem for my pre-convergence (force down to 0.1 ev/A for each image); but such problem occurs quite often in further calculation if no parameters are changed from my pre-converge INCAR file;
Based on the old post, such fluctuation happens quite often with quasi-Newton or LBFGS due to the force error therefore could be fixed by tightening the ediff. However, in all my trials, this strategy does not work. I even see the similar fluctuation with quick-min or FIRE. The only thing works (occationally) is to lower small step size (eg, in FIRE, MAXmove=0.02)
what else strategy one might come up with?
One idea coming up to me now is to pick up the problematic image + the two neighboring image do a one-image plain (not climbing) NEB. That is use the two neighboring ones as endpoints to force the problematic one to a good force standard; one this is done, re-do the whole NEB calculation with small step. would this make sense? I am trying it and would post the result later once it finishes (hopefully) or hits the wall time.
structure fluctuation
Moderator: moderators