[gpaw-users] LCAO DZP basis set for Cr

[Ask Hjorth Larsen]

Hi

2012/7/18 Ask Hjorth Larsen <asklarsen at gmail.com>:
> Hi
>
> 2012/7/18 Ulrik Grønbjerg Andersen <ugan at fysik.dtu.dk>:
>> Hi Ask
>>
>> The two lower layers are fixed and the two upper layers are not constrained. This run was done with BFGS, but I started out by using BFGSLineSearch, which behaved in the same way. You are correct that this run was terminated by exceeding the walltime limit, not by convergence.
>>
>> I have tried to run it using BFGSLineSearch and in LCAO mode without constraints, and in this case there were no problems.
>
> Okay, this is the strangest optimization I have seen since 2009.  I
> hereby CC this to the ASE list with the logfile reattached.
>
> Atom #10 sinks symetrically down between the fixed atoms #4, 5 and 7.
> The force which pushes atoms 4, 5 and 7 outwards is around 20 eV/Å on
> each!  However these atoms are fixed and thus not part of the
> optimization.  Meanwhile, the three atoms act back on #10 with equally
> large but opposite force contributions that entirely cancel each other
> in the xy directions because of the symmetry.  Only in the z direction
> is there a "comparatively modest" remaining upward force on atom 10 of
> 3 eV/Å.  The optimizer then inexplicably proceeds to move #10 down,
> straight against this force.
>
> I don't see any problems with the actual force calculations.
>
>> FD  does not show unphysical behaviour either.
>
> This might be slightly befuddling, but the issue seems rare and it
> could be that it does not happen due to "luck".
>

In fact, maybe if you slightly randomize the initial positions of the
non-fixed atoms, this pathological symmetry-related phenomenon won't
be possible.  Differences in the egg-box effect could be the reason
why this is not observed in the grid-based case.  Is the strangely
behaving atom by any chance located on top of a grid point?

Regards
Ask