[CP2K:6529] Re: reading dcd files

Steve Schmerler elco... at gmail.com
Thu May 21 14:56:50 UTC 2015

On May 20 08:13 -0700, Matthias Krack wrote:
> Hi Steve
> the DCD file header is only written once at the beginning of the simulation 
> run. Thus no way to guess how many frames will be dumped during the run 
> eventually. 

Yes that makes sense, I don't know how the lammps people do that. I
didn't look at their code -- I assume they just write `nstep` from the
input file into the header and assume that the run always finishes as

> It is most important that common tools like VMD are able to digest the
> DCD files dumped by CP2K even if one could certainly devise a better
> binary format.

Of course, compatibility counts.

> Note, that dumpdcd.f90 has also some nice extra feature like
>    - definition of first and last frame dumped
>    - stride (i.e. skipping frames regularily)
>    - applying PBC (at box centre or at 0,0,0), e.g. useful for movies
>    - tracing of atoms which leave the original cell

This is a nice tool indeed. I'm planning to read the coordinates into
numpy arrays for post-processing and storage and wanted to avoid
unfolding the data to a text-based format if possible. At the moment we
are still working with xyz output, which is OK for medium-sizes systems
and time scales but may become problematic in the future. So, thanks for
the tool. I think I know how to proceed here, based on how dumpdcd does


Steve Schmerler
Institut für Theoretische Physik
TU Freiberg, Germany

More information about the CP2K-user mailing list