[Pw_forum] GIPAW problems
kucukben at sissa.it
Wed Sep 5 18:21:34 CEST 2012
Just to update the forum:
The problem mentioned in this thread has been solved by the use of
occupations='fixed' instead of smearing methods.
This might be a good time to remind interested users that the QE-GIPAW
implementation is only for insulators for the time being. Although not
always, but with some compilers, the unexpected insulator+smearing
input configuration could give wrong results as we have seen here.
Currently, Davide Ceresoli is updating the code to include metallic
systems as well, with this development the mentioned 'bug' would no
more appear starting from the next release.
Thanks for your feedback!
postdoc, epfl, switzerland
Quoting Zbigniew Lodziana <Zbigniew.Lodziana at ifj.edu.pl>:
> Dear Forum Users & Developers,
> I am new in QE Forum - welcome to All & congratulations for excellent
> information exchange here and software.
> Being new in GIPAW calculations I might have missed the answer browsing the
> Forum but did not found solution yet.
> The problem is that I am somehow puzzled by GIPAW calculations:
> - Using QE 5.0 I did compile it with Intel Fortran, MKL, openmp (more details
> below). No tuning, no tricks etc. just followed installation guide
> (./configure & make all).
> - While pw.x performs nicely, gipaw.x gives results that are difficult for
> interpretation, and I cannot really locate where the problem is.
> - Both pw.x and gipaw.x pass self-test; examples were no problem too.
> However, trying to calculate NMR spectra for slightly complex system
> calculations with GIPAW give results that seems to be machine
> dependent to the
> large extend.
> Input files and part of output are at the end; shortly "Total NMR chemical
> shifts in ppm:" is either 16.58 or 134.05 for the same system/element (inputs
> are simply copied) when calculated on two similar machines.
> Both computers are Intel based, compiler and libraries differ however:
> a) intel 11.1 (ifort); mkl 10.3
> b) intel 11.1 (ifort); mkl 10.2.5.035
> Playing with pseudopotentials, k-points make things even worse.., denser fft
> grid, density cutoff does not seems to improve results either). I use several
> other numerical codes on both machines - never had similar problems.
> I would highly appreciate any suggestions/help toward resolving this problem.
> best regards & thank you in advance,
> Zbigniew Lodziana
> Polish Academy of Sciences,
> ul.. Radzikowskiego 152,
> PL-31-342 Krakow
More information about the users