The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

On 27 Aug 2007, at 14:24, jzwanzig@dal.ca wrote:

> Dear colleagues,
>
> I find that when minimizing the forces in a crystal structure, say  
> by doing a run with "ionmov 2" and "optcell 2" the resulting output  
> structure (acell, rprim, xred) together breaks the original input  
> symmetry. Never by much, but certain symmetry operations will have  
> residuals of around 1E-7, which is worse than the built-in 1E-8  
> threshold. So the symmetry of the "optimized" structure is lower  
> than the original input. Sometimes this is easy to detect, for  
> example when all the atoms are on high symmetry positions and the  
> acell variable is "optimized" to (for example) "1.11111111  
> 1.11111111 1.11111112". However, some unit cells I work with have  
> complicated shapes but still some symmetry. Why does the  
> optimization run break symmetry in this way? Is there a setting I  
> should be using that I don't know about?

Dear Joe,

As far as I understand, the problem comes from the following :
the symmetrization is imposed on the stress (routine strsym.F90),
and not imposed on the unit cell shape,
and there is a building up of small numerical inaccuracies.
Back in 1999, Razvan and I discussed the imposition of symmetry
on the unit cell shape versus on the stress, and it was much easier
to impose on the stress (the strsym.F90 is quite short !).
Now, we see the limitations of this approach ...

So, one needs to write a routine that is able to symmetrize the unit  
cell ...

Best regards,
Xavier