[Wien] 'SECLR4' - POTRF error (again)

[Laurence Marks]

Unfortunately, the fact that it ran better with NMATMAX changed was
pure luck. You should google potrf (it's a lapack routine) and you
will see from looking at seclr4 in SRC_lapw1 that the error is due to
the matrix not being completely positive definite. I have a suspicion
that this is probably due to rounding errors rather than being real,
although a bad case.in1 (duplicate values) might lead to this.

On 10/10/06, Roberto Iglesias <roberto.iglesias at psi.ch> wrote:
> Hello again!
>
> Looking at the thoughts expressed in your latest email, I thought about the possibility that some of the
> parameters in the compilation, as NMATMAX could have been set to their default values...and voila! So I told
> the computing support staff to try to change it and set it to something higher, like NMATMAX=10000. With that,
> the calculation ran smoothly to convergence. As I noticed I had high values for the forces, I decided to try a
> structural minimization. The first mini loop was OK, but in the second one there was a crash in dnlapw1, with
> the following dnlapw1.error file:
>
> > **  Error in Parallel LAPW1
> > **  LAPW1 STOPPED at Tue Oct 10 13:17:15 CEST 2006
> > **  check ERROR FILES!
> >  Cholesky INFO =          7392
> >  'SECLR4' - POTRF (Scalapack/LAPACK) failed.
> >  Cholesky INFO =          7391
> >  'SECLR4' - POTRF (Scalapack/LAPACK) failed.
>
> so the same error again. This time I could not find any warnings in .scf, neither in the energies, nor
> something as "RKMAX reduced due to NMATMAX".
> Should we set NMATMAX to a higher value? According to the UG, with NMATMAX = 10000 we will honor the
> hardware limitation (2GB per CPU) even for complex calculation. Should we also change NUME from its present
> value of 1000?
>
> Thanks in advance
>
> Roberto
>
> Laurence Marks wrote:
> > Thoughts, which may not work:
> > 1) Look at the sections in the UG/FAQ on setting the energies in case.in1.
> > 2) Try the -in1new switch
> > 3) Check your Fermi energy (grep -e :FER) and put your valence states,
> > not the lo's at about -0.3 Rdy below the Fermi energy.
> > 4) Do "ldd lapw1" with the appropriate path to lapw1 and see what
> > libraries are being used.
> > 5) Try ifort 9.1 if it is available rather than 9.0 -- the later has some bugs.
> > 6) Try changing compilation options.
> >
> > On 10/6/06, Roberto Iglesias <roberto.iglesias at psi.ch> wrote:
> >> Dear Laurence
> >>
> >> Thanks a lot for your suggestions. Unfortunately, these calculations are run on a Cray XT3 system, with more
> >> than 1000 computing CPUs, Opteron 2.6 GHz. The system runs the UNICOS/lc 1.4.26 operating system, implementing
> >> the ACML libraries. How would you change the case.in1 file to avoid these rounding errors? I don't think I
> >> have any illegal content in there...after running x nn, x sgroup and x symmetry, the inequivalent atoms are
> >> reduced to just 8. I don't know what do you mean precisely with "duplicated lo's". I only have one lo, for a
> >> Fe 3d orbital. But of course, this repeats for every atom.
> >>
> >> Regards
> >>
> >> Roberto
> >>
> >>
> >>
> >>
> >> Laurence Marks wrote:
> >>> I had something similar recently, and moved from mkl version 8.0.1 to
> >>> version 8.1 and it seems to have gone away. I think it is probably due
> >>> to rounding errors and sometimes just changing the case.in1 will
> >>> eliminate it. (If you have an illegal case.in1, for instance with
> >>> duplicate lo's I think it can also occur.)
> >>>
> >>> On 10/6/06, Roberto Iglesias <roberto.iglesias at psi.ch> wrote:
> >>>> The uplapw1.err file looks like:
> >>>>
> >>>> **  Error in Parallel LAPW1
> >>>> **  LAPW1 STOPPED at Fri Oct 6 09:23:58 CEST 2006
> >>>> **  check ERROR FILES!
> >>>>   Cholesky INFO =          4985
> >>>>   'SECLR4' - POTRF (Scalapack/LAPACK) failed.
> >>>>   Cholesky INFO =          4997
> >>>>   'SECLR4' - POTRF (Scalapack/LAPACK) failed.
> >>>>   Cholesky INFO =          4999
> >>>>   'SECLR4' - POTRF (Scalapack/LAPACK) failed.
> >>>>
> >>>> --
> >>>> ------------------------------------------
> >>>> Roberto Iglesias
> >>>> High Temperature Materials Project
> >>>> Laboratory for Materials Behaviour
> >>>> Nuclear Energy and Safety Department
> >>>> OHLD/013
> >>>> PAUL SCHERRER INSTITUT
> >>>> CH-5232 Villigen PSI
> >>>> phone: +41 (0)56 310 54 81
> >>>> fax:   +41 (0)56 310 35 65
> >>>> e-mail: roberto.iglesias at psi.ch
> >>>> Internet: www.psi.ch
> >>>> -----------------------------------------
> >>>>
> >>>> _______________________________________________
> >>>> Wien mailing list
> >>>> Wien at zeus.theochem.tuwien.ac.at
> >>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> >>>>
> >>>
> >> _______________________________________________
> >> Wien mailing list
> >> Wien at zeus.theochem.tuwien.ac.at
> >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> >>
> >
> >
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>


-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
http://www.numis.northwestern.edu