[ase-users] atomic symbols in netcdf file get lost

[John Kitchin]

ncdump shows this is exactly what the data is in the ncfile

 DynamicAtomSpecies =
  "",
  "",
  "",
  "\001T",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu",
  "Cu" ;
}

so, the python code is reading the file correctly, it is just the first few
symbols have gotten mangled. Everything else in the ncfile looks fine. This
makes me think it is related to the fortran code somehow. The dacapo code
looks like it has exited just fine, and the only reason the jobs fail is
because the DynamicAtomSpecies causes a KeyError in a dictionary in python.
very strange.

j


-----------------------------------
John Kitchin
Assistant Professor
NETL-IAES Resident Institute Fellow
Doherty Hall A207F
Department of Chemical Engineering
Carnegie Mellon University
Pittsburgh, PA 15213
412-268-7803
http://kitchingroup.cheme.cmu.edu




On Thu, Jan 22, 2009 at 3:43 AM, Jakob Schiotz <schiotz at fysik.dtu.dk> wrote:

> On Thursday 22 January 2009, John Kitchin wrote:
> > was there ever any resolution to what caused the problem at:
> >
> > https://listserv.fysik.dtu.dk/pipermail/campos/2007-June/002261.html
> >
> > the issue is related to the final error:
> >     Z = numbers[symbol]
> > KeyError: '\x00\x00'
> > which is caused because in the netcdf file the atomic symbols are
> replaced
> > by '\x00\x00'
> > I have seen this problem intermittently over the years, and am seeing it
> > again intermittently lately, especially with parallel NEB calculations.
> >
> > I am suspicious that somehow this is a problem with the netcdf library
> > because the text in the key error above looks alot like
>
> I cannot really help you, I have not seen the problem myself.  But a few
> thoughts:  The \x00\x00 characters are two 0-bytes, so it is most likely
> caused by uninitialized data.  Does ncdump show something useful for the
> atomic symbols?  Could it be that the writing script crashed before the
> NetCDF files was properly closed?
>
> /Jakob
>
>
> >
> > NC_CHAR      = '\x00\x00\x00\x02'  (from the pupynere documentation at).
> I
> > am not using pupynere here, but the character strings look similar. it
> > almost looks like the netcdf library is reading from the wrong place in
> the
> > file. I can't explain why the rest of the ncfile looks fine though.
> > I also suspect the timing of when dacapo fortran does something to the
> > netcdf file and when python does something, but as I mentioned, the
> problem
> > is intermittent.
> >
> > any ideas? or solutions from the past?
> >
> > thanks,
> >
> > j
> >
> > -----------------------------------
> > John Kitchin
> > Assistant Professor
> > NETL-IAES Resident Institute Fellow
> > Doherty Hall A207F
> > Department of Chemical Engineering
> > Carnegie Mellon University
> > Pittsburgh, PA 15213
> > 412-268-7803
> > http://kitchingroup.cheme.cmu.edu
>
>
>
> --
> Jakob Schiotz, Ph.D., Associate professor (lektor)
> CINF, Department of Physics
> Technical University of Denmark
> DK-2800 Kongens Lyngby, Denmark
> http://www.cinf.dtu.dk/~schiotz/ <http://www.cinf.dtu.dk/%7Eschiotz/>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/ase-users/attachments/20090122/c4779b22/attachment.html>