[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [h5md-user] Subgroups

From: Felix Höfling
Subject: Re: [h5md-user] Subgroups
Date: Mon, 23 Sep 2013 16:10:53 +0200
User-agent: Opera Mail/12.15 (Linux)

Am 23.09.2013, 15:53 Uhr, schrieb Konrad Hinsen <address@hidden>:

Felix Höfling writes:

> I'm not sure whether overlapping subgroups are also useful for the data in > "particles". (Maybe one example for a spatial selection: a user wants to > analyse the interfacial region between two liquids, so only a shallow slab
 > along the interface is stored instead of all particles.) Do you see a
 > problem here w.r.t. your applications?

Overlapping subgroups are quite relevant for me, in particular the
special case of one subgroup being a subset of another.

In protein simulations, the total system is often decomposed as a tree

                  /      \
                 /        \
            solvent      protein
                        /       \
                       /         \
                   chain A     chain B
              active site

Observables may be provided at any level. Particle positions are often
stored more frequently at the lower levels than at the upper levels.

It is quite common not to store anything at all for the complete
system, so reconstruction is not my goal either. However, I would like
to be able to say which of the atoms in "chain A" are the ones that
are also in "active site".

Thank you for providing an example with nested subgroups, which I didn't have in mind. The H5MD spec is not very explicit about it, but may be interpreted in this direction.

I think the particle "id" would serve a good purpose here. Finding common particles boils down to taking the intersection of two ID datasets. The ID is defined to be unique across the whole system (in contrast to the index in the subgroup's particle array). Maybe the spec should be more precise here.

I have solved this problem for my own applications in the
"H5MD-Mosaic" module that I am working on, which is why I didn't
insist on this question any longer. Mosaic has an "atom selection"
data item which handles the job quite nicely. I simply require that
for every subgroup in "particles", there is an entry with the same
name in the "mosaic" group at the root level, which is either a Mosaic
universe (i.e. the whole system) or a Mosaic atom selection.

> Automatic reconstruction of a complete simulation from a H5MD file would
 > be nice, but there is much more to it than just particle coordinates.
> Then, one would have to enter into specifying particle interactions in a
 > generic way which appears impossible to me.

Well, I am working on that, so I hope you are wrong ;-)

It's clearly a non-trivial problem and I don't expect to finish this
any time soon. Since interactions are arbitrary mathematical
functions, it comes down to defining a domain-specific language for
interactions. The real difficulty is to design such a language in a
way that is both sufficiently flexible to handle all cases and
sufficiently simple to implement for the common cases to make the
approach efficient.


Even more, interactions may be given in a tabulated form and refer to two, three, many particles ... The approach with the conventions seems to point an elegant way out of this mess. A certain set of interactions is then defined by a certain convention, and specifying an interaction could be reduced to a single string (instead of an arbitrary function) and a few parameters.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]