[Pw_forum] What to do when NEB calculation is difficult to converge?

Tone Kokalj tone.kokalj at ijs.si
Wed Aug 30 09:24:04 CEST 2006

On Sun, 2006-08-27 at 09:39 +0200, Xunlei Ding wrote:
> Dear all,
> I am now doing a neb calculation on a surface adsorption system with 
> pwscf3.0

> It has more than 400 iterations and has not converged. 

This is a very large number of iterations, and the error on the 3rd,
4th, and 5th image is still large. It is difficult to guess why it does
not converge from your data. The time step ds=1.0 seems reasonable (as
far as my experience go). 

One thing that you may try is to increase the number of images!

How to do that? You need to do a new calculation with larger number of
images, but you can use the currently calculated path as the initial
path for this new calculation.

You may either try with the path_int.sh script that is located in
pwtools/ directory (edit it to suit your needs). This will generate a
new path file will requested number of images, and in the pw.x input
file you specify then new number of images and restart_mode='restart'.
The other possibility is to insert the current path (for example, the
axsf file the pw.x has produced) into the input file of pw.x (specify
new number of images and restart_mode='from_scratch'). You need to do
some hand-editing then:

   (atomic position of image-1)
   (atomic position of image-1)
   (atomic positions of last image)

Regards, Tone

More information about the users mailing list