Re: [h5md-user] Tuples and topology, the return

[Felix Höfling]

Am 07.07.2014, 11:50 Uhr, schrieb Pierre de Buyl
<address@hidden>:

> Hi all,
>
> Following http://thread.gmane.org/gmane.science.simulation.h5md.user/590  
> and the
> ensuing discussion, I modified my proposal.
>
> 1. a module "tuple", defining n-tuples
> 2. a module "topology", similar to the previous proposal but made more  
> precise.
>
> They can be found at http://pdebuyl.be/tmp/h5md/modules/index.html
> (note, it is just a temporary working version)
>
> The idea is that "tuple" would be included somehow in the main  
> specification so
> that it can serve as the basis to further functionality. A specific  
> topology
> module remains needed to provide the link to particle groups, etc. It  
> would
> remain a module.
>
> If tuple goes into the specification, topology would refer to it do avoid
> repetition.
>
> I've put TODO items in each module:
> Tuples
> 1. Provide a way link unambiguously to a particle group.
> 2. Require that tuple data is found at least for every step of the  
> particle
>    position data.
> I don't know if "1" can be solved generically (it could be solved only  
> when a
> specific situation is defined). "2" is indeed very related :-)
>
> Topology
> - Build on top of pdebuyl_tuples.
>
> All in all, I've tried to separate (fully on the basis of the  
> aforementioned
> thread) the container and content :-)
> With respect to the discussion, I've dropped the dimension attribute (it  
> is
> redundant w r t the dataset shape) and also the type attribute.
> Additional attributes are always allowed but they should not be required  
> (they
> can be added in the related module).
>
> The names are currently "pdebuyl_tuples" and "pdebuyl_topology" to avoid  
> any
> possible confusion if they get included later on with the versions that  
> I put
> online.
>
> P

Hi Pierre,

Good to see progress here :-) I'm not (yet) using molecules with a
topology, so I can't comment on the semantics.

Your draft seems to repeat the definition of the H5MD element (twice), so
it may be made more concise by defining a list of N k-tuples simply as
H5MD element of rank 2 with dimensions [N, k]. (Especially the distinction
between time-dependent and -independent should be clear.) The speciality
that the values refer to particle indices goes beyond a general data
structure and should be part of the topology module.

topology module:

* one may want to store different bond types for the same group (implying
that each tuple list collects bonds of the same type)

* can one define a set of names for the bond lists which imply the kind of
bond (similarly to the fields in the thermodynamics module)? The PDB
format may give an inspiration for possible bond types.

*  the wording "The step and time datasets must match exactly" should be
replaced by "... must be hard links to". Compare to the specification of
particles/image: "For time-dependent data, the step and time datasets of
image must equal those of position, which must be accomplished by
hard-linking the respective datasets."

Regards,

Felix