Spectra DM for theoretical spectra?

[bonnarel]

The discussion between Alberto and Carlos stretches the actual difficulty of
the scope of the characterization concept.
   In Alberto's description we admit implicitly that the characterization
axes are Position, time, spectral, polarization and flux.
   This is basically what we have in oBservations datasets.
   Many people (including Alberto, I think) want to restrict this strictly
to these kind of obersvation data products.

   My point is that nothing prevents to have the same general structure with
axes giving other parameters such as Teff, mass, etc ... for a given dataset
(eg a Snapshot), although in practice we only have experimented
Observational data. (This is not at all excluded by the Char reference
document)

    Assume an ObjecTCollection in the sense of SimDB: each object has a
couple of "properties" and these properties show a certain distribution of
values for the whole collection. Characterizing this collection in the sense
of SimDB is giving min/max (very similar to the bounds) and some statistical
moments (mean, mode, etc...) if not the distribution itself.
   I think that a real observation can be seen as a set of "Observing
events". Each "Observing event" is an "Object" in the sense of simDB.
    It has "properties" with the meaning of coordinates on the Observation
axes.
    The statistical distribution of "a priori" efficiencies of these events
is what has been called "Level 4" or "sensitivity" by IVOA characterization.
     I think Gerard had a similar view but was distinguishing "Properties"
and "Values of these properties" in a way incompatible with the actual
design of IVOA Characterization DM. As was stated by Rick "properties" also
include concepts such as resolution which are not axes in the sense of IVOA
Char DM.
     My point is although there are objective difficulties in comparing the
two characterization concepts (IVOA and SimDB) it is a pity that we do not
reached convergence.

     By the way Nothing in this invalidates what has been written by
Alberto, Jesus or myself about Provenance of a theoretical spectrum. 
    It is not specific to simulation data that some stuff in the Provenance
of a dataset has its OWN characterization. Imagine a coadded image: it has
its  characterization. Each individual member is part of its provenance and
has its own charac different from the one of the whole coadded image !!!!

François
  

-----Message d'origine-----
De : Alberto Micol [mailto:alberto.micol at eso.org] 
Envoyé : mercredi 3 juin 2009 14:06
À : Carlos Rodrigo Blanco
Cc : theory at ivoa.net; dm at ivoa.net
Objet : Re: Spectra DM for theoretical spectra?

Carlos Rodrigo Blanco wrote:
>
> All right, I see that I'm getting lost about some concepts and 
> vocabulary in the vosphere... I'll try to catch up.
>
> then I understand that the characterization DM aims to describe all 
> the "characteristics" of an observation (WHAT is in the observation) 
> and the provenance aims to describe the origin of the observation (HOW 
> it was acquired, why, who, whatever)
Sorry, again: CharDM is not about WHAT is observed/simulated, it only 
about some properties on the data axes
(position, coverage, resolution, sampling on the spatial, spectral, 
time, and any other axis the data is subtending).
>
> Then I agree that for a theoretical spectrum a (Teff,logg,metallicity) 
> set is closer to be provenance that to the be characterization.
>
> But I still don't see it IS either.
>
> My question would be:
>
> In SSAP an spectrum is identified by the position in sky of the 
> corresponding object, that is, RA and DEC. Are they included in 
> provenance? 
It depends:
 - where the PI wanted to point the telescope is part of Provenance,
 - which part of the sky is actually subtended by the resulting data is 
in CharDM.
Subtle difference, but an important one.
Ideally the two will coincide, but in real world there is a difference;
the telescope points at a given (ra,dec) then the astrometric pipeline 
corrects
that information using some reference sources/catalogs. If there is no 
astrometric correction
then the two set of (ra,dec) values will coincide.

Provenance describes the "a priori" knowledge (or will) of the PI,
CharDM describes more the "a posteriori" knowledge of what happened
during the observing process (including calibration pipeline).

The archive/vo user who wants to analyse the data will generally care more
about the "a posteriori" info (CharDM), though it is always good to know 
what is that
the PI actually wanted to do "a priori" (Provenance).

> I undertand that what is included in provenance is more "the region of 
> the sky that was observed by the astronomer was a rectangle with these 
> four corners and it was observed this day with this instrument, etc". 
No, that is in CharDM.
> But the actual RA,DEC used by SSAP (POS parameter) are some 
> coordinates associated to that spectrum and not really included in 
> provenance. Am I right? (please, correct me if I'm wrong)
That is up to CharDM, not provenance.
> The point that I want to stress is that, for theoretical spectra (and, 
> of course, from my point of view), the point (Teff,logg,metallicity) 
> is the equivalent to (RA,DEC). They are the coordinates in some 
> "parameter space" of the spectra, not (or not only) some description 
> about how the spectrum was obtained.
>
I see: you claim that the parameter space you want to "observe" is 
divided into
a grid of points in the (Teff, logg, Z) space, and your simulator will 
obtain spectra
for a number of those grid points. 

This very much resembles what a Principal Investigator describes during 
Phase 2
of the proposal preparation at any observatory: all the things the 
observatory must know
to carry out the observation.  In this case the grid of points are in 
the (ra,dec,wavelength,time, etc) space,
and the observatory will obtain spectra/images for a number of those 
grid points
(e.g. a raster map observation to cover with various images a given part 
of the sky).

This to me qualifies as Provenance: it is about where to point the 
(virtual) telescope, and with which camera settings.
It is not about what was actually observed/simulated. It is "a priori" 
knowledge, not "a posteriori".

> What I see equivalent to my idea of provenance is "these theoretical 
> spectra were calculated using kurucz code, calculating a grid with 
> this edges and this step for each axe, considering rotation or not, 
> considering the presence of a disk or not, etc".
>
I agree, that is _also_ part of Provenance.
> I don't really see clearly yet if this makes a difference or not and 
> maybe the provenance data model is the perfect place to consider these 
> things.
>
> But, in Alberto's words, I don't see (teff,logg...) as the equivalent 
> to instrument/telescope, but more the equivalent to data_axes 
> (position,time) for observed spectra.
I already answered to this above, but about data axes:
Data axes are the axes of the resulting dataset.
Given that the data is a spectrum, the axes are spatial, spectral, 
temporal, and the observable (flux),
and not Teff, logg, Z.

Teff, logg, Z are the axes of the parameter space input (provenance) to 
your simulator, not output (chardm).

Alberto

>
> Of course the problem with theoretical spectra is that the parameter 
> space (the one used to identify an spectrum, the one used to search 
> for spectra) is not allways the same because different developers. And 
> that's a problem on its own (the one that started this conversation), 
> either if we call it provenance or something else.
>
> Carlos