The ABINIT Users Mailing List ( CLOSED )

[Pio Baettig <baettigp@gmx.net>]

Hi,

thank you very much for your e-mail.
I tried your suggestion, but unfortunately, the angles and vectors still 
diverge after a certain number of iterations.
The divergence can become quite large toward the end of the optimization 
(1mdeg) .
As abinit finds the correct symmetry for my input-data (R3c), it should 
not break it at all during the optimization in principle.

This holds true for the atomic positions. (If the symmetry is evaluated 
with 'corrected' vectors and angles (eg 10 10 10bohr, 59 59 59deg) R3c 
is again obtained.

The deviation of the first vector and angle seems always to start after 
about 8 to 10 Broyden steps.

Could I use the following as a temporary workaround:
set ntime to 8 and ndtset to 4 or 5, getwfk to 0
and then use
ngkpt1 6 6 6
ngkpt2 6 6 6
ngkpt3 6 6 6
ngkpt4 6 6 6

to always have the same conditions. The last positons would be read and 
the vectors and angles should hopefully stay equal?

Best regards,

Pio Baettig



Dan Ricinschi wrote:
> Hi!
>
> I cannot confidently point out the cause of this behavior, but in my 
> experience this happens when the optimization goes on for quite a large 
> number of Broyden steps, without being able to find the absolute minimum 
> of the total energy with the symmetry strictly enforced. It could also 
> mean that the imposed symmetry is not that of the ground state, 
> (although in many cases it could be possible to optimize the crystal 
> under other desired symmetry). If you don't suspect a serious physical 
> error, you could restart the optimization with the R3c structure 
> obtained after adjusting those small deviations by hand and see what 
> happens.
>
> On the other hand, by looking at your atomic coordinates I think that it 
> is likely that you have found a local minimum of the total energy 
> surface, as the Bi atoms might be "fixed" to their centro-symmetrical 
> positions. If you have not already done so, I suggest you to restart the 
> optimization from scratch by displacing the Bi atoms slightly
> > from their symmetrical positions, in the starting configuration. That 
> is, use 0.01 0.01 0.01 and 0.51 0.51 0.51 instead of 0 0 0 and 0.5 0.5 
> 0.5. In this way the algorithm might find the optimized structure 
> easier, without breaking the symmetry for lengths and angles.
>
> I hope this helps.
>
> Dan Ricinschi