[ase-users] converging nudged elastic band's in ASE

[Jens Jørgen Mortensen]

On Mon, 2009-09-21 at 15:06 +0100, Felix Hanke wrote:
> Hi all,
> I was wondering if there is a generally recommended strategy to converge
> Nudged Elastic band calculations within ASE.
> 
> I have spent a fair amount of time on the scenarios provided in the
> tutorials, but only got the simple Au diffusion on Al(100) to actually
> give me a reasonable transition state. The latter I would like to define
> as a transition state obtained with climbing images and with a maximal
> force (on the transition state image) of, say, less than 0.01 for starters.
> 
> I tried a number of things on the dissociation tutorial (N2 on copper),
> but did not manage to get anywhere near convergence. Things I have tried
> include:
> - all relaxation algorithms
> - various spring constants
> - more images (for better tangent resolution)
> (- auto-refinement of the NEB path based on the energy profile)
> Not only is there no convergence in the system, but in the end it
> actually seems to diverge.
> What else can one do?

This dissociation tutorial (N2 on Cu(111) with EMT) looks like it needs
some more work ...  The EMT potential gives a very unrealistic
description of this reaction - a 10 eV barrier.  This exercises has a
more realistice NEB example:

 https://wiki.fysik.dtu.dk/gpaw/exercises/neb/neb.html

I can get a decent N2-dissociation path if I use 20 images and
climb=True with MDMin and dt=0.1.

If you need the transition state with fmax<0.01, I would try to relax
the climbing image with a constraint that will only allow it to move
perpendicular to the path.

> The attachment contains my latest iteration of the program. Basically
> there is a simple running procedure at the bottom, with a few
> definitions for NEB and BFGS that clarify the auto-refining procedure
> and some additional output. Maybe this helps.
> 
> Cheers
> Felix
> 
> 
> 
> 
> 
> _______________________________________________
> ase-users mailing list
> ase-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users