[ase-users] [EXTERNAL] Dimer convergence

[Hermes, Eric]

On Sunday, January 19, 2020 11:52:57 AM PST Alaithan, Zainab via ase-users 
wrote:
> Dear ASE-users,
> 
> My dimer calculations hasn't converged after 130 steps and the forces are
> fluctuating. I read in the article that 20 steps were the average number of
> steps required for convergence. but how about the behaviour of the forces
> and the curvature should I pay attention to them or just wait and see after
> 200 steps?  I started from a converged NEB calculation, and I was expecting
> that it will converge fairly quickly.
> 
> Best Regards,

Zainab,

While I don't have too much experience with ASE's implementation of the Dimer 
method, my experience indicates that Dimer can  be quite unreliable at times. 
I have developed a package for saddle point optimization called Sella. Sella 
is much more efficient and robust than Dimer. You can find Sella on github:

https://github.com/zadorlab/sella. 

The README has a brief example, and the github wiki page goes into more detail 
about Sella's various options. Here is a paper describing the method 
implemented in Sella:

https://pubs.acs.org/doi/abs/10.1021/acs.jctc.9b00869.

Regards,
-- 
Eric Hermes
Postdoctoral Researcher
Sandia National Laboratories