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

[Pierre de Buyl]

Hi Felix,

On Tue, Jul 31, 2012 at 05:45:23PM +0200, Felix Höfling wrote:
> I apologise for having been silent over such a long time. I've been
> ...

That's fine, it's not like anyone has been pushing things at super speed :-)
altough the specification is now nearing a state that we can push to colleagues.

> 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.

That is my feeling too, I made that explicit in the coming push to the
repository. This is in line with your ulterior message.

> 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.

I agree on that but would like to check something first: before your commit, any
element in "observables" had to obey the H5MD group structure and it was easy to
list all observables. In my opinion, the current way of listing observables is
to recursively go through "observables" and list all groups that contain a
"step", "time" and "value" dataset. It should thus remain that simple in the
future.

> @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).
As of now, it is mostly american. Any preference?

> – 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).
I did that in the coming commit. 

> – 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.

I suggest that you put those in the discussion page (fields are already present
to some extent).

I am preparing a few short commits that go in the direction of this discussion.

Cheers,

Pierre

> 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

> >From 5bd007c352802a1a1e8070e004cacccf8adf818f Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?Felix=20H=C3=B6fling?= <address@hidden>
> Date: Tue, 31 Jul 2012 17:27:04 +0200
> Subject: [PATCH] observables: allow subgroups, add "pressure", matrix of edge
>  vectors
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
> 
> We shall allow subgroups in the "observables" group as we do in "trajectory".
> 
> We shall not forget the pressue, which is of similar importance as the
> potential energy or the temperature (aka kinetic energy).
> 
> For the triclinic box, the set of D D-dimensional vectors may be considered a
> D??D matrix which makes essentially no difference how the data are stored
> finally. But it is clearer that the full set of edge vectors is to be
> considered a single entity, e.g., the determinant of which readily yields the
> box volume.
> ---
>  draft.rst |   16 ++++++++++++----
>  1 file changed, 12 insertions(+), 4 deletions(-)
> 
> diff --git a/draft.rst b/draft.rst
> index 3376c22..9b7cf91 100644
> --- a/draft.rst
> +++ b/draft.rst
> @@ -191,14 +191,14 @@ Cuboid box
>  """"""""""
>  
>  * edges: A vector specifying the length of the box in the D dimensions of
> -    space.
> +  space.
>  * offset: A vector specifying the lower coordinate for all directions.
>  
>  Triclinic box
>  """""""""""""
>  
> -* edges: A set of D D-dimensional vectors specifying the directions and
> -  lengths of the sides of the box.
> +* edges: A set of D??D-dimensional matrices with the rows specifying the
> +  directions and lengths of the edges of the box.
>  * offset: A vector specifying the lower coordinate for all directions.
>  
>  Time dependence
> @@ -230,16 +230,19 @@ Observables group
>  
>  Macroscopic observables are stored as \[variable\] time series for scalar
>  observables and as \[variable\]\[d\] time series for d-dimensional vector
> -observables. The variable dimension allows to accumulate time-steps. The 
> name of
> +observables. The variable dimension allows to accumulate time steps. The 
> name of
>  the group holding these datasets is "observables". This group has the same
>  structure as "trajectory": for each observable there is a group containing 
> three
>  datasets: the actual data in "value" and the step and time datasets.
> +Observables representing only a subset of the particles may be stored in
> +appropriate subgroups similarly to the "trajectory" tree.
>  
>  The following names should be obeyed for the corresponding observables:
>  
>  * total_energy
>  * potential_energy
>  * kinetic_energy
> +* pressure
>  * temperature
>  
>  The content of the observables group is the following::
> @@ -253,6 +256,11 @@ The content of the observables group is the following::
>       |    \-- value
>       |    \-- step
>       |    \-- time
> +     \-- group1
> +     |    \-- obs3
> +     |         \-- value
> +     |         \-- step
> +     |         \-- time
>       \-- ...
>  
>  Program-dependent groups
> -- 
> 1.7.10.2
>