[Pw_forum] NEB vs 'direct' approach

[Stefano Baroni]

Dear Pawel:

your findings sound rather weird to me. I would have expected the 
contrary. At convergence, the NEB should produce the "Minimum Energy 
Path" (MEP), i.e. the path whose transition state has minimum energy 
(with respect to the transition states of all the other neighboring 
paths). I think this is discussed at length in the original literature, 
but it is enough to think to the defintinition of the MEP, a path for 
which the force is at any point tangent to the trajectory. Imagine of 
following such a path while climbing a mountain pass, and you will 
convince yourself easily that this path actually passes through the saddle.

I do not know why you find a different result. An error in the QE 
implementation of the NEB would be an obvious explanation. Another 
possibility is that the discretization that you have (by necessity!) 
chosen to sample your "direct" path is so coarse that it actually misses 
the point of largest energy along the path, which is exactly what the 
prefix "CI" (for "climbing image") in the CI-NEB acronym is meant to avoid.

Hope this helps
Yours - Stefano B.

Pawel Jakubas wrote:
> Dear Pwscf Users,
>
> I have started using CI NEB method implemented in espresso. I deal with
> native defects in bulk materials; let's say Cd interstitial in CdTe, which 
> crystalizes in zinc blende structure.
> It is known that it diffuses along [111] direction between tetrahedral 
> sites.
>
> Previously, to calculate barriers for diffusion I used 'direct' approach (I
> coined this name), which means I calculated total energies by fixing the
> position of an interstitial at a number of sites between metastable sites,
> and allowing all other atoms to relax (except one located at the outskirts
> of supercell, i. e., far away form the defect). The barriers heights
> calculated using NEB are usually higher than those obtained by 'direct'
> approach. I try to understand these differences. Because the diffusion 
> occurs along [111] direction in straight line, so the path is to some extent 
> in this case trivial (I am stressing it, because it is ok to choose nstep=1 
> in neb).
>
> Is there a clear, unanimous explanation why it is so? Are you familiarized
> with any paper in literature giving discussion of the dicrepancies between
> NEB results and those obtained using 'direct' approach? What is the physics 
> behind it?
>
> Thank you very much in advance for any comments, hints and help.
>
> Pawel Jakubas
> PhD Student
> Piotr Boguslawski Group
> Institute of Physics
> Polish Academy of Sciences
> AL. Lotnikow 32/46
> PL-02-668 Warsaw, POLAND
> group webpage: http://www.ifpan.edu.pl/SL-3/
> my webpage: http://www.ifpan.edu.pl/SL-3/PJakubas/jakubas.html
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>