[Wien] 'SECLR4' - SYEVX

[Peter Blaha]

a) Maybe NUME in param.inc_r/c of SRC_lapw1 needs to be incresed!
b) The emin/emax energy window is specified in Ry (not eV).
c) You are running a "medium size" calculations for a metallic system.
Thus you should have many k-points (more than your 8 ?) and 
matrixsize/memory should be small enough to hold the H,S matrices in 
memory on a single cpu. Therefore, you should definitely use the k-point 
parallel version, not the fine-grain mpi-based parallel one !!
For this, your .machines file should NOT look like:

8:hostname:8

but contain 8 (identical) lines like

1:hostname
... (8 times)

This will span 8 lapw1 jobs, each one runs one k-point.

Eventually, on your Altix you set "shared memory" during siteconfig.


Robertson Burgess schrieb:
> Hi all,
> I'm running Wien2k/07 on an Altix Cluster on a Linux OS using a PBS
> queueing system. My structure is Aluminium slabs ranging from 5 to 15
> atoms thick. I succesfully run the SCF cycle through to completion on
> all the structures. I then modify the case.in1 to go up to higher
> energies than the default 2eV. Here's a sample of my case.in1 file
> 
> WFFIL        (WFPRI, SUPWF)
>   6.00       10    4 (R-MT*K-MAX; MAX L IN WF, V-NMT
>  -.39071   4   0      global e-param with N other choices, napw
>  0   -0.443     0.000 CONT 1
>  1   -0.300     0.000 CONT 1
>  1   -4.866     0.002 CONT 1
>  2   -0.280     0.000 CONT 1
>  -.39071   4   0      global e-param with N other choices, napw
>  0   -0.502     0.000 CONT 1
>  1   -0.355     0.000 CONT 1
>  1   -4.864     0.002 CONT 1
>  2   -0.274     0.000 CONT 1
>  -.39071   4   0      global e-param with N other choices, napw
>  0   -0.507     0.000 CONT 1
>  1   -0.355     0.000 CONT 1
>  1   -4.866     0.002 CONT 1
>  2   -0.276     0.000 CONT 1
>  -.39071   4   0      global e-param with N other choices, napw
>  0   -0.503     0.000 CONT 1
>  1   -0.354     0.000 CONT 1
>  1   -4.864     0.002 CONT 1
>  2   -0.274     0.000 CONT 1
>  -.39071   4   0      global e-param with N other choices, napw
>  0   -0.505     0.000 CONT 1
>  1   -0.353     0.000 CONT 1
>  1   -4.865     0.002 CONT 1
>  2   -0.280     0.000 CONT 1
> K-VECTORS FROM UNIT:4   -9.0       5.0      emin/emax window
> 
> I then run the following jobscript file.
> 
> #!/bin/bash
> #PBS -l walltime=12:00:00
> #PBS -l vmem=8000MB
> #PBS -l jobfs=2000MB
> #PBS -l software=wien2k
> #PBS -l ncpus=8
> #PBS -wd
> 
> module load wien2k/07
> export SCRATCH=$PBS_JOBFS
> export MPI_GROUP_MAX=128
> 
> KPOINTS=8
> 
> echo $KPOINTS:`hostname`:$PBS_NCPUS > .machines
> x lapw1 -p
> x lapw2 -fermi
> x optic -p
> 
> 
> LAPW1 then crashes with the following message
> **  Error in Parallel LAPW1
> **  LAPW1 STOPPED at Mon Mar 12 09:25:42 EST 2007
> **  check ERROR FILES!
>  SEP INFO =            5
>  'SECLR4' - SYEVX (Scalapack/LAPACK) failed.
> 
> This only happens on parallel processes, and the error can be avoided by
> reducing the energy parameters on the last line of case.in1. For
> instance, I get the error with a 6 atom slab going from -9 to 7 eV, but
> not -9 to 5 eV. Can anyone help me out with this?
> I've looked in the mailing list, and I find a lot of references to a
> 'SECLR4' - POTRF error but not this one.
> 
> Thanks
> Bob Burgess
> University of Newcastle
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: 
http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------