Spectra DM for theoretical spectra?

[Miguel Cerviño]

Dear Alberto,

> Dear Miguel,
>
> May I ask you what is missing in the SpectrumDM?
> What is SimDB offering extra, specific to spectra?

yes, of course :)
You can find an extensive description related with the needs for the  
"class of theoretical spectra
corresponding to single stellar populations" in the astro-ph paper  
0711.1353

http://arxiv.org/pdf/0711.1353v1

In summary, most of the issues are in the characterization/provenance.  
I have seen
in this interop a lot of effort related with the characterization of  
how the observations has been
taken, which is the extinction curve at the moment the observation is  
taken, or even the details
of the photometric system. Most of these details (but with theoretical  
names), are needed.

In the observational point of view, all these things are needed to be  
able to "repeat" the experiment if
needed under similar conditions (i.e. the usual scientific method).  
Isn't it? In theoretical models it is needed
something similar: an exahustive description about how the "data" has  
been produced


As example: Kuruzc ATLAS 12 programs allow produce stellar spectra  
with user specific
isotopic abundances. Hence, to understand (and choose) a given ATLAS  
12 atmosphere model it should
be needed to access  the particular isotopic abundances used in the  
particular model. This information is
even more relevant than the resolution of the final spectra (in most  
cases, it is by far more cheaper repeat a
simulation than a observation). A lot of the focus in SimDB (or any  
SimDB-like description) is in this "description/characterization".


There are other examples adressed in the quoted paper. For example a  
lot of spectral theoretical models that
should be in the VO are "generic representations" of object classes: a  
kurucz A0 V spectra is a generic spectra that do not necessary apply  
to any real star, but as an average description with a dispersion...
In this case the dispersion is not an "error", but the description of  
an underling probability density distribution.
But the "StatError" can be used, although with a "non usual" meaning.
However, such a kind situations with an underling distributions are,  
by far, better defined if high order moments of such a distribution  
are quoted (like skewnes and kurtosis). But there no place in the  
Spectra data model for that: is is assumed an error with its underling  
gaussian statistics, and gaussian statistics is an issue that can
work to describe the error of observational data (although also  
Poisson statistics, I not sure how much
a modelization of errors are implicit in the current data model or if  
it can be specified somewhere....)

I can develop a bit more these examples if needed. I hope this would  
illustrate part of the problem in any case

cheers :)

m