Bug? Weird behaviour for FIXED_ATOMS in GEO_OPT (recent CP2K versions)

marci marc... at pci.uzh.ch
Wed Oct 24 11:22:55 CEST 2012


Dear Raphael

You should be  able to recover the right behavior changing the BFGS section 
as follows

    &BFGS
      USE_MODEL_HESSIAN FALSE
      USE_RAT_FUN_OPT TRUE
      TRUST_RADIUS 0.1
      @if ${HESSIAN} == 1
        RESTART_HESSIAN
        RESTART_FILE_NAME ${PROJECT}-BFGS.Hessian
      @endif
    &END

This restores the old defaults. 
I hope the problem will be fixed soon.

kind regards
marcella

On Tuesday, October 23, 2012 11:25:24 PM UTC+2, Dr_Nick wrote:
>
> Hi all,
>
> I am running a GEO_OPT with constraints. My file (attached) contains the 
> following section:
>
>   &CONSTRAINT
>     &FIXED_ATOMS
>      COMPONENTS_TO_FIX XYZ
>      LIST 1 2 11..14 35..38
>     &END FIXED_ATOMS
>   &END CONSTRAINT
>
> I recall that this worked with CP2K 2.2 and 2.3 version: i.e. the xyz 
> coordinates of the fixed atoms do not move during optimization. 
>
> However with 2.4 (today's 12467) I noticed some very strange things:
> - the 1st geometry (i=1) in the -pos-1.xyz file does NOT correspond to the 
> starting structure. All positions are slightly off. This is also true when 
> no constraints are applied and / or when I use other input files! 
> - Fixed atoms move slightly.
> - In one of my systems, some of the fixed atoms move slightly, while other 
> fixed atoms experience large and totally "senseless" deplacements which 
> causes the geometry optimization to fail.
>
> N.B.
> - the constraints section is correctly read in because no force is 
> displayed for the fixed atoms.
> - the starting structure is reasonable because it was preoptimized before 
> with another software.
>
> I ran the same input on different machines (Magny Cours and Xeon) to check 
> for parallelization / optimization / compiler / version issues.
> 2.4 branch 12467, 12431
> 2.3 branch 12343, 12275, 12230, 12229
> AMD: gcc 4.6.3 / 4.6.2 / 4.7.0 + openmpi 1.6.1 + openblas 0.1.1 / 0.2.3 + 
> fftw 3.3
> INTEL: intel 12.1.2 + openmpi 1.6.1 + MKL + intel FFTW / -O2 and -O3 
> optimization
> These versions have been compiled either by me or the admins of our 
> cluster. I guess they know what they are doing.
>
> Here are my findings:
>
> - running on 1 CPU (cp2k.sopt) or using only 1 MPI process (cp2k.popt), 
> everything is OK (*all versions*)
> - running on more than 1 CPU everything is OK for <= 2.3 12230. All higher 
> revisions are affected.
> - OpenMPI has probably no influence because with the same version (1.6.1), 
> old CP2K revisions are ok, new revisions are not. This would require more 
> testing but where can I download old 2.3 revisions? The svn always gives 
> you the most recent version.
> - O2 / O3 optimization tested in the case of the intel compiler: no 
> difference.
>
> So in my understanding, all versions >= 2.3 12230 are affected, when 
> running on more than 1 CPU.
>
> Could you please have a look into this issue or try to reproduce? My input 
> files are attached.
>
> Best regards and thanks for your help,
> Raphael
>
>
>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20121024/989e8b79/attachment.html>


More information about the CP2K-user mailing list