The ABINIT Users Mailing List ( CLOSED )

[Matthieu Verstraete <mjv500@york.ac.uk>]

No, there are no hidden options. I have observed this as well. Abinit 
symmetrizes the forces and stresses before applying them, so I don't know 
how it manages to break them in the end. Maybe some accumulated round-off. 
Of course it keeps the original symops from the initial structure, even if 
they are broken somehow. It would be nice to see when and where it 
appears. It might even be necessary to re-symmetrize the positions 
periodically (or at every step - it's cheap).

In my case it happens progressively (angles deviating from 90 deg etc). 
Same for you? Does it happen if you don't relax the cell? I doubt it - 
probably the stresses are to blame. If you could do a thorough search or 
look at the last decimals of the stresses that are applied (maybe beyond 
what is printed in the normal output) that would be great. In the Broyden 
algorithm they are also piped through the Hessian, so the numerical error 
could appear there. Dunno, maybe the Hessian itself has to be symmetrized.

Bye!

Matthieu

--
================================================================
Dr. Matthieu Verstraete                 mailto:mjv500@york.ac.uk
Dept. of Physics, University of York,     tel: +44 1904 43 22 08
Heslington, YO10 5DD York, United Kingdom fax: +44 1904 43 22 14