Re: [h5md-user] topology

[Felix Höfling]

Am 22.04.2014, 13:59 Uhr, schrieb Pierre de Buyl  
<address@hidden>:

> Hi all,
>
> Topology was deliberately left out of H5MD 1.0. The intent was that  
> feedback on
> this point was needed, as well as some experience with relevant  
> simulations.
>
> I have started to work on polymer simulations recently and I have chosen  
> a way
> to store topology information that requires no "conceptual" addition to  
> H5MD.
>
> Depending on the direction of the discussion this could become either a  
> module
> or a part of the specification itself.
>
> Here is how I do:
>
> 1. Create a "/topology" group that contains groups named directly after  
> those in
> "/particles". This is similar to what is done in the thermodynamics  
> module
> http://nongnu.org/h5md/modules/thermodynamics.html#observables
> 2. Within the groups, H5MD elements store bonds as [N_bonds][bond_order]  
> data.
> For pairs, bond_order=2, for instance. This allows to store angles and
> dihedrals. This also allows to store several bond lists (you could have  
> pairs
> for separate potentials, for instance).
>
> As these are H5MD elements, you can store them in a time-dependent  
> fashion. This
> requires that a "fill value" is defined if you need to grow the dataset  
> in
> "N_bonds". I use "-1" but the value itself can be retrieved and does not  
> need to
> be in the specification.
>
> I show below a "h5ls -r" dump of a topology group:
> """
> /chains                  Dataset {4000, 2}
> /crosslinks              Group
> /crosslinks/step         Dataset {100/Inf}
> /crosslinks/time         Dataset {100/Inf}
> /crosslinks/value        Dataset {100/Inf, 417/Inf, 2}
> """
>
> Here, "chains" is a "fixed in time" element of 4000 pairs and  
> "crosslinks" is a
> "time dependent" element with 100 time steps and at most 417 pairs.
>
> The position in "/topology/<group>" could be (in my opinion) changed to
> "/particles/<group>/topology". I wanted to work with minimal  
> modifications,
> which why I use "/topology" now but putting the information in  
> "/particles"
> seems more natural to me.
>
> For completeness, I need to say that I currently use ESPResSo++
> http://www.espresso-pp.de/ . I cannot tell if this influenced my  
> proposition.
>
> Cheers,
>
> Pierre

Hi Pierre,

Indeed, the mailing list (and http://nongnu.org/h5md/discussion.html) is a  
good place to share ideas on extensions of H5MD.

Some time ago, Olaf told me about his thoughts on how to store topological  
information. I think this is pretty much in line with your proposal.  
Olaf's ideas were pretty much inspired by what is stored in the PDB format.

What I'm missing in your post is what information is actually stored? Am I  
right that you define a pair bond by a pair of particle IDs? In /particle  
these IDs are unique only within a subgroup so bonds across these  
subgroups would be difficult to implement. (They may simply not be  
supported, requiring the subgroups to be sufficiently large. E.g., they  
would need to hold the complete macromolecule not only parts of.)

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.

Cheers,

Felix