[Pw_forum] relaxation of ice Ic with qe404

[Paolo Giannozzi]

merlin meheut wrote:

> I have had a symetry problem during the relaxation of ice Ic, 
> which happened on two platforms where I have compiled the  
> 4.0.4 version, whereas the 3.0 version seemed to work fine.

the problem you mention appeared in v.3.1 (and is still there),
apparently after a change of the damped-dynamics algorithm.
In lines 824-826 of PW/vcsubs.f90, note the following:

      ! if(.true. ) = original version modified by Cesar Da Silva
      ! if(.false.) = modified algorithm by SdG
      if (.true.) then

The last line was "if(.false.) then" in v.3.0 and it was change
to the current one in v.3.1, because CDS said that his version
was more tested. If you revert to the previous version, the
problem you mention disappears. I cannot say whether this means
that there is something fishy in the current algorithm.

In any case: cell_dynamics='bfgs' is faster, at least in this
case. By the way, diagonalization='david_overlap' is obsolete
and will issue an error message in the new version. Use
diagonalization='david'

Paolo
-- 
Paolo Giannozzi, Democritos and University of Udine, Italy