about output handling in cp2k

Axel akoh... at gmail.com
Mon Jul 9 18:44:11 CEST 2007


hi all!

i have a couple more questions about how cp2k handles output:

it may be my ignorance towards the inner workings of cp2k, but to me
the flags ADD_LAST, EACH, COMMON_ITERATION_LEVELS and
the filename naming conventions are rather confusing or maybe the
the description in the documentation needs to be 'translated' into
layman's terms,
perhaps with a few examples. in particular, i'd like to understand
what
is meant by 'iteration level', as it obviously does not refer to the
SCF iterations
alone. i also noticed that the example for ADD_LAST: ADD_LAST NO
SYMBOLIC
does generate an parser error.

also, is there a way to have the (quickstep) RESTART file (or whatever
name i assign it)
been written only at the end of every n-th MD step (not SCF step)? as
far as i
understood, a postitive number writes the restart at every n-th scf
iteration and 0
when convergence is reached or SCF_MAX. how about having negative
values
being matched against the time step number?

finally, i noticed that .dcd files are missing the header part
when i am using &EXT_RESTART to continue a calculation but
rename the files for each run. the only workaround i found so far
is to set RESTART_COUNTERS to .false., but since my jobs can
have different length (due to dynamic backfill queueing they can
be allowed to run for anything between 2 and 12 hours) it would be
extremely nice to have my cake and eat it, i.e. .dcd files with
headers
and restarted counters. is there an alternative way to get the header
written?

since cp2k already seems to append some sort of serial number to
outputs, the most desirable (to me) way of writing output would
actually
be to have a parameter that specifies how many steps would go into
a single file (e.g. 2000, derived from the time step number) and then
start a new file when that is reached and step up the serial number.
could that be implemented with reasonable effort?

thanks,
   axel.




More information about the CP2K-user mailing list