Re: [h5md-user] Varying particle number

[Felix Höfling]

Am 08.10.2014, 04:46 Uhr, schrieb Peter Colberg
<address@hidden>:

> On Tue, Oct 07, 2014 at 09:35:57AM +0200, address@hidden wrote:
> > Before going further with "particle_number", did you test your  
> > situation with a
> > compressed "id" element?
>
> No, the particles are indistiguishable, so there is not id. The
> compression should reduce the size dramatically; for a 64-bit “step”
> dataset and a chunk size of 32 kB the compression ratio can be >100.
>
> > Alternatively, a better compression could be achieved if we decided to  
> > allow
> > "species" to contain also a fill value. This way, you'd compress a  
> > dataset of
> > fill_value, 0, 1 (for two species, for instance) and that should be more
> > efficient.
>
> That is a good idea. We should add this as a minimal solution.
>
> In any case, I am not concerned with the efficiency of storage, but
> with the convenience of reading the particle data. It would be nice if
> one would not have to loop over species/id to find non-empty regions.
> Instead one reads the particle number, checks that both species and id
> (if present) do not define a fill value (= no holes), and then selects
> a single hyperslab for reading a particle dataset.
>
> Peter

I find the particle_number is a useful extension. It features both
ease-of-use and efficiency, which both are goals of H5MD.

With respect to having a fill value in "species" too I'm concerned about
possible inconsistencies: if both "id" and "species" contain a fill value,
it is not guaranteed that the same particles are marked as placeholders.

Even for indistinguishable particles, one may assign some consecutive IDs
(or simply omit the ID). Though the trajectories may look a bit random (if
somebody tries to visualise the data). So there is no real need to extend
"species".

Regards,

Felix