[Pw_forum] where is the end of an unconverged NEB calculation

Lorenzo Paulatto paulatto at sissa.it
Fri Nov 6 14:00:03 CET 2009

In data 06 novembre 2009 alle ore 04:35:39, vega lew  
<quantumdft at gmail.com> ha scritto:
> During my NEB calculations, I found there are two possible situation.  
> First, the calculation could be converged after about 100 iterations.  
> Second, the calculation never converged in more than 300 iterations.

Dear Vega,
NEB calculation are a bit tricky in general and require extreme care to be  
setup correctly. NEB also takes easily hunders of iteration to converge,  
of course depending on the number of atoms and of images. Here is some  
free advice I can give, although only practice make prefection.

1. don't use climbing image at the beginning, CI makes convergence slower;  
especially if the special image changes during the convergence process (it  
may happen if CI_scheme='auto') this does ususally mess up everything.  
Converge your calculation than restart from the last configuration with  
the climbing image (note that this will *increase* the barrier!)

2. carefully choose the initial path, remember that QE assumens continuity  
between the first and last image at the initial condition. In other words,  
periodic images are NOT used, you may have to manually translate an atom  
by one or more unit cell base vectors in order to have a meaningful  
initial path.

3. you can try to start the NEB process with most of the atoms fixed in  
position, in order to have the more "problematic" one to converge, before  
freeing all of them.

4. especially for larger systems, you can start the NEB with lower  
accuracy (less k-points, lower cutoff) and then increase it when it has  
converged to refine your calculation.

5. use the Broyden algorithm instead of the default one, it is a bit more  
fragile, but it removes the oscillation problem. If this happens, and you  
cannot afford more images, focus to a smaller problem, decompose it to  

6. you can grossly estimate the number of iterations required as  
number_of_images * number_of_atoms * 3 (the atoms that don't move should  
not be counted). It may take half that many iterations, or twice as many,  
but more or less that's the order of magnitude, unless you start from a  
very accurate or very bad initial guess.

good luck

Lorenzo Paulatto
phone: +39 040 3787 511
skype: paulatz
www:   http://people.sissa.it/~paulatto/

     *** save italian brains ***

More information about the users mailing list