[ase-users] Preconditioned LBFGS - RuntimeError: a1 too small, giving up
Rasmus Karlsson
rasmusk at kth.se
Thu May 18 21:51:46 CEST 2017
Hi Noam and Ulrik,
I've had a very similar experience with PreconLBFGS in GPAW. I've run
into that result for relaxations of adsorbates on magnetic surfaces,
where I had reduced SCF convergence criteria to achieve SCF
convergence.
In a way, I think the PreconLBFGS optimizer just giving up and telling
the user something is wrong is a good thing. The other optimizers would
just continue wasting time and resources not finding a minima. Of
course, fundamentally the problem is not with the individual optimizer
but with my settings, but it is still nice to get a warning that
something is wrong.
Best regards,
Rasmus
Noam Bernstein via ase-users writes:
> On May 18, 2017, at 10:02 AM, Ulrik Leffers via ase-users <
> ase-users at listserv.fysik.dtu.dk> wrote:
>
> Hi,
>
> I've run into some issues when running the preconditioned LBFGS
> optimisation on a YLaS3 structure with 20 atoms in its unit cell. It runs
> about 100 iterations before giving the error, "RuntimeError: a1 too small,
> giving up". However, it only actually seems to do anything within the first
> 75 iterations; I tried to plot the energy of the system and the positions
> of the atoms in the gui, everything is constant from this point. I have
> attached the script as well as the slurm output and the trajectory file
> from the optimisation. Any help as to how this error can be avoided would
> be much appreciated.
>
>
> The most common cause of such behavior is forces and energies that are not self
> consistent, typically because of inadequate convergence w.r.t. some parameter
> or another (SCF tolerance, cutoff can also be an issue for codes that aren’t
> exactly translationally invariant). We’d at least need to know what calculator
> you’re using to be able to say anything more specific.
>
> Noam
>
>
> ____________
> |
> |
> |
> U.S. NAVAL
> |
> |
> _RESEARCH_
> |
> LABORATORY
>
> Noam Bernstein, Ph.D.
> Center for Materials Physics and Technology
> U.S. Naval Research Laboratory
> T +1 202 404 8628 F +1 202 404 7546
> https://www.nrl.navy.mil
--
Rasmus Karlsson, PhD
More information about the ase-users
mailing list