Re: [h5md-user] New commits, "id" and rc

[Felix Höfling]

Hi Pierre,

I apologise for having been silent over such a long time. I've been pretty  
busy with another project and have done neither much programming nor  
simulations lately, although I did some data analysis of H5MD files. I  
think the proposed H5MD format is in a good shape for a first release and  
has seen some testing in a few applications so far.

Regarding additional datasets in trajectory like "id": we may suggest a  
commonly used name, but not enforce that such datasets are present. This  
would be in line with "species" or "force" and the "observables" group.

I have some minor suggestions for "observables", see the attached commit.  
Mainly, I think we should make it possible to have subgroups for storing  
partial pressures or partial energy contributions. Global quantities shall  
be stored at the top level still. If you agree I will upload the commit.

@Peter: If I remember correctly, in HALMD/master the box group is  
currently created in the trajectory group, which makes the iteration over  
particle groups a bit cumbersome. So I was glad to see that it is not so  
in the H5MD proposal (I already wanted to suggest moving it somewhere  
else.)

Some more comments:

– We shall agree on either British or American orthography (organization  
vs. standardised).

– The group "profiling" may be dropped and be considered as  
program-specific (it is not forbidden by the format to store such groups  
at the root level).

– For the future, I would find the following root groups useful, bearing  
in mind that the format must not be overspecified:

"dynamics" for storing time correlation functions like the mean-square  
displacement, such data can not be interpreted as a time series and the  
datasets will thus have a very different structure

"structure" for storing static correlation functions like the static  
structure factor or the pair distribution function. These data, however,  
may be interpreted as a time series (including averages over finite pieces  
of the simulation run) and we have to discuss about a sensible structure  
and place for such data. In contrast to simple observables, they carry  
additional parameters like a wavenumber grid and do not necessarily belong  
to a single simulation step.

"fields" for storing binned quantities like a one-dimensional density  
profile or a three-dimensional velocity field. Such data sets describe the  
transition from particle-based MD to continuous (albeit discretised)  
fields. From a technical point of view, they have similar requirements  
like the static structure factor.

Best wishes,
Felix


Am 28.07.2012, 22:32 Uhr, schrieb Pierre de Buyl  
<address@hidden>:

> Hi all,
>
> I'd like to know wether there are remarks on my recents commits.
>
> H5MD has stabilized since some time and is in a very usable state. I've  
> cleaned
> a bit the draft.
>
> I'd like to release a "0.1" version of H5MD. As a first step, I'd like  
> to tag
> the current (modulo "id", see below) version as 0.1-rc1.
>
> Before that, I think that the "id" information should be included.  
> Defining it
> already seems a way to future proof a building block that matters in the  
> case of
> spatial sorting of particles, including parallel coding.
>
> "id" would be included in a similar way to "position": in the  
> "trajectory"
> group, and of shape [var,N] where var is the time stepping and N the  
> number of
> particles.
>
> Comments and input welcome :-)
>
> Best,
>
> Pierre
>
> PS: despite the low activity on the list, I believe that at least Felix,  
> Peter
> and me are still using the file format for our software, which gives  
> sense to
> agree on a release.


--
Dr Felix Höfling
Max Planck Institute for Intelligent Systems
(formerly Max Planck Institute for Metals Research)
Heisenbergstr. 3
70569 Stuttgart
Germany

Phone:  +49 711 689 1938
Fax:    +49 711 689 1922

txtCXkY3Rv0MU.txt
Description: Text Data