[Wien] lapw1 crash with SECLIT - Error in Cholesky
Laurence Marks
L-marks at northwestern.edu
Fri Oct 10 16:03:57 CEST 2014
I forgot that your case has no inversion symmetry -- you need to use "x
RMTCheck -c". Please send me that output so I can make educated guesses.
If you are using -it then increasing nband and emax can help. The iterative
methods use an expansion in terms of the previous eigensolutions, both
occupied and some unoccupied. If this expansion is not "adequate" I am
pretty certain one starts to get ghostbands and many other problems. (This
is more intuition and experience than any proper math.)
I do know that for MSR1a one can often improve things a little by using
more solutions, the speed cost is very minor so long and you do not use
extreme increases. I also prefer -noHinv, but that is my personal view not
a general suggestion.
On Fri, Oct 10, 2014 at 4:03 AM, Pavel Ondracka <pavel.ondracka at email.cz>
wrote:
> On Thu, 2014-10-09 at 08:02 -0500, Laurence Marks wrote:
> > I am not sure what exactly you are trying to do. It looks like you
> > have some approximation to a Si doped amorphous TiO2 structure. The
> > BVS looks reasonable, so this may have come from some other code.
>
> Yeah, the structure was produced by simulated annealing done by other
> code and I want to calculate band gap and optical constants with Wien2k
> using mBJ. At the moment I'm just trying to converge the case with LDA
> before I initialize mBJ.
>
> >
> > One thing odd is the RMT for Si of 1.44 which may very well lead to
> > problems. This is actually close to what setrmt is giving. I think
> > there may be a bug here in setrmt, Peter can say more. The small Si
> > RMT is why you are losing core electrons.
> >
> >
> > One thing you can do is (after at least one pass) do "x RMTCheck".
> > This will show the magnitude of the discontinuity at the RMT. I have
> > seen that the discontinuities of different types of atoms should be
> > roughly the same, and I suspect that in your case they are not.
> >
> Looking at the output of x RMTCheck (attached) the Si atom doesn't seem
> to stand out too much, but I don't have any idea what "roughly the same"
> means in this case. Can you have a look please? At the moment I have a
> converged LDA case with the Si RMT = 1.44 and ecut -10.2 (no warnings)
> and I'm wondering if I can continue with mBJ or rather restart with
> different RMTs and higher ecut (e.g. -7.2).
> >
> > Without running your case myself, I would want to use
> >
> >
> > cp case.struct Hold.struct
> > setrmt case -a Ti:1.7,Si:1.6,O:1.4 ; cp *set* *.struct
> >
> >
> > case is the name and the initial cp is because sometimes setrmt can be
> > slightly buggy and get confused about decimal points.
> >
> >
> > This initialized for me without warning with -ecut -7.2 and only the
> > Si 2p as semicore not the 2s as well.
>
> After playing with this a little bit more it seems that the old scheme
> in setrmt is actually giving better results in my case:
>
> new: O:1.57 Ti:1.74 Si:1.44
> old: O:1.56 Ti:1.76 Si:1.56
>
> The O and Ti RMTs are quite similar so I have no idea why the Si RMT is
> so much lower with the new scheme.
> BTW with the old scheme I can get no warning also with only 2p states.
> >
> > N.B., you may want to increase nband at the bottom of case.in1c to
> > something like 480 or 512, increase emin to 2.5, only use one k-point
> > and for LDA with these RMTs I suspect that a RKMAX of 6 is fine.
>
> Did you meant increase emax to 2.5?
>
> Anyway this is really helpful, thank you very much.
> >
> >
> > On Thu, Oct 9, 2014 at 7:01 AM, Pavel Ondracka
> > <pavel.ondracka at email.cz> wrote:
> > On Thu, 2014-10-09 at 06:23 -0500, Laurence Marks wrote:
> > > Why are you using P1? You have made everything much slower
> > and less
> > > efficient.
> > >
> > > Beyond this it is hard to guess.
> > >
> > Well, P1 is what I get during the initialization with sgrup.
> >
> > In the meantime I managed to get it running by removing -it
> > switch (two
> > successful iterations so far), so will see if it actually
> > stays working.
> >
> > Best regards
> > Pavel Ondračka
> >
> > > __________________________
> > > Professor Laurence Marks
> > > Department of Materials Science and Engineering
> > > Northwestern University
> > > www.numis.northwestern.edu
> > > MURI4D.numis.northwestern.edu
> > > Co-Editor, Acta Cryst A
> > > "Research is to see what everybody else has seen, and to
> > think what
> > > nobody else has thought"
> > > Albert Szent-Gyorgi
> > >
> > > Dear Wien2k mailing list,
> > >
> > > I have a problem with crash in parallel lapw1. It crash with
> > SECLIT -
> > > Error in Cholesky output in stderr. Looking at tail of
> > corresponding
> > > case.output1_2 I see:
> > >
> > > Time for los (hamilt, cpu/wall) : 0.8
> > 5.6
> > > Time for alm (hns) : 4.2
> > > Time for vector (hns) : 14.8
> > > Time for vector2 (hns) : 14.0
> > > Time for VxV (hns) : 211.3
> > > Wall Time for VxV (hns) : 2.8
> > > reading Afacts -1 0.000000000000000E+000
> > > :seclit: estimate of singular value, factor: 0.6527E+00
> > 0.1000E-14
> > > :seclit: min(sproj(ne+1:2ne)) 0.3110E-02
> > > WARNING: INFO (Cholesky) = 679
> > >
> > > I found some suggestions for Cholesky errors here:
> > > http://www.mail-archive.com/wien%
> > > 40zeus.theochem.tuwien.ac.at/msg02400.html
> > > however I'm quite sure my struct is OK, RKmax is default 7
> > (which
> > > should
> > > be reasonable for compound with Si Ti and O) and neither can
> > I spot
> > > any
> > > problems in my in1 file.
> > >
> > > This happens in second scf cycle. I'm using a LDA potential
> > and
> > > everything was initialized to the default in init_lapw
> > (except for
> > > energy seperation between core/valence which was set to
> > -10.2 because
> > > some Si core electrons were leaking out of MT sphere, and
> > reduced
> > > number
> > > of k-points).
> > >
> > > I'm attaching my struct and in1 files. Any ideas?
> > >
> > > Best regards
> > > Pavel Ondračka
> > >
> > > _______________________________________________
> > > Wien mailing list
> > > Wien at zeus.theochem.tuwien.ac.at
> > > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > > SEARCH the MAILING-LIST at:
> >
> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
> >
> >
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > SEARCH the MAILING-LIST at:
> >
> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
> >
> >
> >
> >
> > --
> > Professor Laurence Marks
> > Department of Materials Science and Engineering
> > Northwestern University
> > www.numis.northwestern.edu
> > Corrosion in 4D: MURI4D.numis.northwestern.edu
> > Co-Editor, Acta Cryst A
> > "Research is to see what everybody else has seen, and to think what
> > nobody else has thought"
> > Albert Szent-Gyorgi
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > SEARCH the MAILING-LIST at:
> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>
>
--
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu
Corrosion in 4D: MURI4D.numis.northwestern.edu
Co-Editor, Acta Cryst A
"Research is to see what everybody else has seen, and to think what nobody
else has thought"
Albert Szent-Gyorgi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20141010/7333f6c4/attachment.html>
More information about the Wien
mailing list