relative fluxes

[Petr Skoda]

> In my view, this solution is more useful than setting a "n/a" value to
> the unit field. If we do that, we are loosing a lot of information about
> the continuum shape that is still present for spectra where the relative
> flux is the result to divide the spectra by a single reference value.

Yes but this is exactly the case of known information (simple scaling 
Value = number) so the physical value of measured flux is say P times 
smaller than it really is , but the shape is the same (and I can measure 
power law index as the shape is not changed by the transormation.

But I can imagine that it may be only intentional rescaling and I would 
like to know the use case how this is useful.... perhaps if the detector
is not calibrated in physical flux units but has the known LINEAR 
response.

otherwise all other meaningfull transformations imply the scaling FUNCTION
which is NON-LINEAR and changes the shape of continuum


> continuum as the continuum shape information is already lost.
Exactly - spectrum was divided by unknown function that does not have 
physical prescription.


> The problem I see with this solution is that (if I understand Alberto's
> use case) the reference value looks unknown or lost.

It is not value but function (or set of functions - as still some points 
are missing - gaps between chips and spikes mark missing information in 
given lambda)

>
> In order to cover this case, we were thinking in something like:
>
> scaleSI: 	n/a
> (scale factor unknown)
> dimEquation: 	ML-1T-3
> (F(lambda) flux type)

No I think that Alberto's case is some function of COUNTS
(as it is coming from HARPS the pipeline describes the extracted spectra 
like being expressed in relative flux - but in principle it is the case 
above - the relative means here "with unknown reference function".
The pipeline does not allow (and is not intended) to produce flux 
calibrated data)

> For VOSpec, this kind of spectra has a flag so the user could normalize
> them to a flux value they consider, so it can be compared with absolute
> normalized spectra as it is done between observed spectra and models.
Just by vertical shifting

The case better suiting the problem is something like 
dereddening  (but here the reference scaling function is well described 
by models)

I am afraid that its difficult to judge about the nature of data just by 
looking at the result without knowing the full PROVENANCE (and in fact not 
having observational description) - the discussion started by Alberto is 
the example of type of misunderstandings that may happen just by 
phenomenological description of data without having expertise how it is 
produced (e.g. whole reduction process). And the ambiguity presented for 
one word NORMALIZED is a caveat for whole VO semantics, ucds etc ...


Normalized in sense of Jesus's comment means "scaled by given VALUE"
Normalized in sense of RECTIFIED is the modification by some 
(practically unknown) FUNCTION
Normalization is sense of Miguel's simulation means something like 
estimation of influence of the given variable on whole compound result 
(e.g. reference to unite mass of 1 Sun means what happens with my model 
when I increase or decrease the mass) and I suppose it can work only in 
case of linear response of result on combination of input variables - 
neglecting the synergy (thus nonlinear) effects.
The more mass - or the more stars I put the more intensive is the 
(absolutely calibrated) spectrum.

*************************************************************************
*  Petr Skoda                         Phone : +420-323-649201, ext. 361 * 
*  Stellar Department                         +420-323-620361           *
*  Astronomical Institute AS CR       Fax   : +420-323-620250           *
*  251 65 Ondrejov                    e-mail: skoda at sunstel.asu.cas.cz  *
*  Czech Republic                                                       *
*************************************************************************