Re: [h5md-user] box and observables

[Peter Colberg]

On Thu, Sep 19, 2013 at 04:20:01PM +0200, Konrad Hinsen wrote:
> First, some good news: I can convert my protein trajectories to
> H5MD+Mosaic format, and I have a simple molecular viewer to look at
> them. All that seems to work fine for non-periodic systems.

Great! If it's rather small, please feel free to send the compressed
output of "h5dump -A" and the Mosaic file. I am sure some of the
subscribers would be interested to see a typical example of which
data elements are needed for a protein simulation.

Do you already have suggestions for a H5MD protein module?

> While looking at the extension to periodic systems, I noticed what
> looks like a mistake.
> 
> In the particles group, each subgroup has its own box subgroup.
> However, in the observables group, there is only one box subgroup for
> all observables and all subgroups of observables. This implies that
> all observables must use the same time sampling grids. It doesn't
> really make sense to allow different sampling for the coordinates of
> each subgroup but not for the observables, which is why I suppose this
> is a mistake.

This is a difficult topic…

As noted in the specification, all root groups except /h5md are
optional. To analyse observables such as the fluctuations of the
total energy, the box dimension, or more generally the box group,
is needed. Therefore box is replicated in the /observables group.

The observables are of course allowed have different time grids. If
two observables with a different grid require the box information for
interpretation, it is the responsibility of the program to sample the
box at a time grid that is the union of the two observable time grids.

Note that a reader looking up a value at a given time must always
use the time dataset of the respective datagroup (in this case box).
If I am not mistaken, none of the datagroups specified in the
specification require identical time grids among themselves.

So it is not a mistake; but I would agree it seems inconsistent that
box is stored for each subsystem in "particles", and only stored once
in "observables".

Peter