Re: [h5md-user] topology

[Felix Höfling]

Am 24.04.2014, 14:35 Uhr, schrieb Pierre de Buyl  
<address@hidden>:

> Hi Felix,
>
> On Wed, Apr 23, 2014 at 10:00:17AM +0200, Felix Höfling wrote:
> > Am 22.04.2014, 13:59 Uhr, schrieb Pierre de Buyl
> > <address@hidden>:

> > We better keep the topology separate from the /particles tree. The
> > information complements the information in /particles, but is quite
> > different in structure. In /particles, there is one entry per
> > particle while the topology is about subsets of pairs, triples, etc.
>
> It is true that the information in /topology is structured quite  
> differently
> than in /particles. I don't see a hard reason to choose any of the  
> solutions
> yet, though. An advantage of putting it in /particles could be that the
> particles' positions and topologies could be copied more simply from a  
> file to
> another.

Eventually, everything is somehow related to /particles. But following  
this argument we end up in putting everything in this tree, which is  
probably not intended. Let's stick to the (informal) definition that the  
data in /particles scale as O(N), one entry per particle.

Second argument: as I will outline below, the topology is not a single  
H5MD element, but a group of such elements. And I would like to avoid too  
nested structures for clarity. It should be as simple as
/<top level>/<group>/<H5MD element>

> I realize that I forgot also that topology information could be pairs of
> particles combined with an equilibrium/constraint distance. I have no  
> idea on
> how to add that yet.

What is missing is to specify interaction parameters for each bond (even  
if the functional form of the interaction is the same). The problem is  
similar to the 'species' field in /particles and we could implement it  
similarly as an additional H5MD element parallel to the list of bonds,  
indicating the bond parameters. (It would require an additional  
intermediate group though.) The bond parameters themselves have to be  
specified elsewhere, as per bond species. Alternatively, there is an H5MD  
element 'bond_length' which gives the parameter for each bond in the list.  
The analogy would be the "mass" field in /particles (which could also be  
specified on a per-species basis).

Actually, the spec 1.0 does not contain the possibility of providing  
per-species properties. An generic extension would be most welcome. Such  
data, of course, would not be stored inside /particle (no O(N) scaling),  
but in a different top-level group. As of 1.0, the user may store such  
information in /parameters.

Cheers,

Felix