[CP2K-user] Cell optimization using ASE

Maxime Van den Bossche maxime.cp.... at gmail.com
Mon May 17 08:13:56 UTC 2021


Dear Lucas,

(I'm not a CP2K developer, but I have worked quite a bit with ASE.)

Your code snippet looks fine, except that this will simply relax the ionic 
positions.
If you also want to optimize the cell vectors, you will need to look at 
e.g. the StrainFilter
or UnitCellFilter in ase.constraints. An "fmax" value of 0.01 eV/Angstrom 
should
generally give well-converged minima. Fmax=0.05 eV/Angstrom is also 
sufficiently
accurate in many cases (but please check it yourself to be sure).

But the fact that the forces do not readily decrease in this calculation 
suggests
that something may be wrong with your CP2K input. But without a minimal 
working
example (including all input and output files you get when running the 
example),
it is impossible to diagnose that further.

Best,
Maxime
Op woensdag 12 mei 2021 om 19:13:26 UTC+2 schreef ban... at gmail.com:

> Dear cp2k group, 
>
> I am trying to perform some cell optimization using cp2k interfaced with 
> ASE. The command lines I am using are the following:
>
> sys=bulk('Cu', 'fcc', a=3.7, cubic=True)
> sys.calc = calc
> opt=BFGS(sys, trajectory='CuBulk.traj', logfile='CuBulk.log', 
> restart='Cubulk.rst')
> opt.run(fmax=0.01)
>
> This is based on this ASE tutorial: 
> https://wiki.fysik.dtu.dk/ase/gettingstarted/surface.html.
>
> However, as I have seen, the maximum force on each atom oscillates about 
> the same range 0.6-0.8 eV/A (for a Cu unit cell), and it also depends on 
> the metal I am working with. To be honest, I chose this value for fmax 
> because I thought it was the convergence criteria for the maximum force 
> between two consecutive optimization steps. My question is: How do I choose 
> the maximum value for the force on each atom, or any other convergence 
> criteria? Is this the right way to perform cell optimization using ASE?
>
> Yours faithfully,
>
> Lucas Bandeira
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210517/aa306ea6/attachment.htm>


More information about the CP2K-user mailing list