[Pw_forum] WFC convergence in NEB calculation

[Janos Kiss]

Dear PWSCF community,

I already asked these questions a good while ago, but unfortunately 
none of the resposes solved my problem. This would be the following:

I want to study relative 'simple' proton transfer reaction on a ZnO surface 
(surface slab represented with a  hexagonal supercell and ultrasoft psp). 
When i do an NEB calculation, the wfc for those images, which are close either 
to the reactant, or the product converge really nicely within 7-12 SCF steps.
Such an example, if i make a band with 17 images (15 movable 2 fixed), than 
the wfc for images 2->6 and 9->15 converges flawlessly and really fast. 
Now, for images 7-8-9 i cannot manage to get the wfc to converge, no matter 
what i do. Of course, one could argue, that the position of the nuclei in 
these configuration are way too far from equilibrium. I completely agree with 
this, but if i visualize them, they are completely meaningful TS 
configurations, where an OH group is elongated to around 1.6 A and this 
proton is oriented towards a 'target' C or O atom, which would give me the 
product. What is even more puzzling, is that after the first linear 
interpolation (when the nuclear positions are the worse possible) the wfc 
does converge for all images most of the time. After 4-10 iterations with the 
band the barrier nicely goes down to around 2-3 eV, because the path is 
getting better and better. This is exactly the point, when this strange 
behavior in the wfc convergence kicks in. For images 7-8-9 the Total Energy 
and the gradient on the wfc will just fluctuate up and down within a given 
range, without achieving convergence ever.

What i tried up to now: decrease mixing_beta to values as low as 0.1, tried 
all the mixing modes (my default was mixing mode=’local-TF’), increased 
mixing ndim to 12, increased ecutrho to 5 times ecutwfc. Altough, the ZnO slab 
should be semiconductor, just to try to get a converged wfc i added around 
extra 20% empty bands and a broadening. This did not help, either. Because of 
the available memory i do not really wanna play around to crank up the cutoff, 
but if one say that this gonna help, i can give it a try. Just for testing, i 
plugged the same geometries with the same VDB psp with the same cutoff values 
into CPMD. Although the achieving self-consistency was slow as well, but 
still i could obtain a converged wfc and calculate the forces for these 
configurations without too much additional effort/trick. My point here is not 
that CPMD is better/worse, but the fact that the given configurations were 
not completely nonsense, and the whole setup and set of psp is reasonable.

Is this behavior completely normal in an NEB calculation with PWSCF, or this 
is an issue with my particular system/setup only? 
What else could one try, to get the wfc somehow converged?
Another question: when i restart an NEB calculation, how can i restart the wfc 
for those images, which were already converged for a given NEB iteration?
The output file indicates, that the initial density and the Starting wfc were 
read in 'from file'. I would have expected, that once this is done, the wfc 
will converge within 1-2 iterations, because the already self-consistent wfc 
was written out in the respective *.wfc files. Apparently this is not the 
case, because even for those images, where the wfc was converged, i still 
nead to spend again like 5-9 SCF cycles/image. 

What am i doing wrong?

Yours Sincerely,
 Janos.

 ==================================================================
   Janos Kiss   e-mail: janos.kiss at theochem.ruhr-uni-bochum.de       
 Lehrstuhl fuer Theoretische Chemie  Phone: +49 (0)234/32-26485 
 NC 03/297                                  +49 (0)234 32 26754
 Ruhr-Universitaet Bochum            Fax:   +49 (0)234/32-14045
 D-44780 Bochum            http://www.theochem.ruhr-uni-bochum.de
 ==================================================================