Spectral data models

[Ivo Busko]

The following may, or may not be, relevant to the VO. Anyway, 
it came to my mind when reading Frank's and Jonathan's excelent 
write-ups on spectral standards. I apologize for my late 
contribution to this discussion (the Cambridge meeting is taking 
place as I write this).

In section 1.1 (What is a spectrum ?) of Jonathan's document
(http://hea-www.harvard.edu/~jcm/vo/vospec.pdf), the definition
of a fundamental spectrum says it is a special case of a 
one-dimensional histogram. Wouldn't this definition exclude the 
posibility for the VO spectral standard to support analytical 
functions ? I don't know if this is a relevant issue, maybe the 
VO should only support observational data, which always 
intrinsically include some form of sampling. But, if it *is* a 
relevant issue, then we should perhaps step back a little further 
and say that the one-dimensional histogram is already some 
processed form of a "true" spectrum.

This leads to a related issue regarding nomenclature. We tend 
to use the word "spectrum" with a not so accurate meaning. What 
we usually intend to refer to with this word is actually a 
spectrogram. That is, a recording of the spectrum. In this same
context, I recall that, when trying to answer the question "what 
is a spectrum" in the context of Specview's development, we 
attempted to start from the very basics: "spectrum" basically
stands for the dependency (function) of the irradiance (time 
averaged Poynting vector) in a EM wave, with the wave frequency. 
The histogram concept comes into play only after we re-cast the 
wave description into a photon-based one. Then, binning 
the photons generates the spectrogram. This line of thought
bears some parallelism with the historical development in physics,
where the wave characteristics of EM radiation where formalized 
before the quantum characteristics. And makes the continuous 
spectral description a predecessor of the discrete one.

I envision the embodiement of these concepts in Jonathan's 
proposal at the Axis object definition (section 4.3). We would 
have to replace the AXIS_VALUES_XXX and related parameters by
some sort of abstract objects. These would then be subclassed into
the proposed _CENTER, _START, _SIZE, etc. parameters that describe
a sampled data set. They would as well be subclassed into other 
subclasses appropriate for the description of continuous functions.
This second set of classes could be left unspecified for the
moment (I don't have a clear idea on how to do it), but could 
be added any time later on, since the hooks would be in place 
already.

Just some food for thought...

-Ivo