[Passband] a useful self-contained model?

[Martin Hill]

Hi Anita!

Thanks for that - though I haven't explained some things nearly clearly enough :-(

What I would like to try and do is consider passbands *outside* the context of 
SEDs/spectra - if this is possible.  This 'divide and conquer' approach should 
make it simpler to model each part without trying to involve everything 
everywhere, which seems to be making each model attempted so far become and more 
generalised and less and less of a model of our data.

Similarly I'd like to see if we can get an 'agreed' distillation of this 
component from the various existing documents - including the SSA model, 
Jonathon's docs, etc.

Anita Richards wrote:

>>The intention is 1) to provide a passband model that can be 'plugged in' to the
>>SED model (from SSA) and ObsData when it is ready, and 2) to practice modelling
>>a relatively self-contained concept that is familiar to everyone, yet likely to
>>be viewed and stored and used in different ways by different disciplines.
> 
> 
> Number one need is to define terms. To me, a spectral energy distribution
> (SED) means "a series of flux v. wavelength measurements where each single
> point is measured across a bandpass which is not necessarily linear" and
> has a fairly small (lambda/delta lambda) e.g.  <10^3, and the points in
> the SED are "often separated by large/irregular gaps"; and the SED is
> usually used to measure the average or continuum emission from objects.
> In contrast a spectrum has a higher resolution, is made up of a large
> number of points in narrow channels which are usually contiguous or
> regularly spaced (or is a series thereof), and is usually used to measure
> specific spectral lines coresponding to molecular or atomic transitions.

So this gives us non-linear (narrow) bandpasses that might be described by a set 
of points on a graph, and channel bandpasses, each of which might be considered 
linear but very narrow?

> 
> This is not a rigid distinction but I propose that we use SED to cover the
> situation I ahve put in """" above, and spectrum to cover the further
> subdivision into channels where this can be described as follows ''

So I understand the terms are coming to mean!

> 
> 'A spectral region described by a bandpass function can be further divided
> into channels, where the channel spacing and width are regular in the
> appropriate units and the sensitivity of the individual channel is
> described by its location in the spectral region bandpass' that is, you
> only need one bandpass function.  The 'appropriate units' bit is because a
> 0.5-MHz bandpass divided into 512 equal ~1 KHz channels will only give
> approximately linearly-spaced channels in wavelength etc.  I think that
> the individual channels may sometimes be best described as Gaussian
> functions, but all by the same Gaussian (bar the centre ref)?
> 
>  1 2 3 4 5 6 7 8     eight channels
> | | | | | | | | |
> -----------------
> |----BANDPASS---|
> 
> (here I use flux to include energy etc. and wavelength/frequency to
> include freq., wavel., energy, wavenumber)

Would this work if each channel was a separate Passband?  We might have a 
Passband that was a collection of other Passbands (ie what you've labelled as 
BANDPASS above) to make it easier to use collections of them.  I'll try and get 
some diagrams together on the page to make this clearer.

I should think we can have a GaussianPassband that has a central wavelength and 
sigma and calculates passrates from that.  And presumably an 'effective' min/max 
wavelength...

> 
> 
>>  - Who needs errors on passrate()?
> 
> 
> REFERENCE FREQ UNCERTAINTY
> In some circumstances it is preferable to use a reference wavelength in a
> given part of the band (etc.) and a bandpass function.  e.g.
> reference wavelength 600 nm  in channel 256 out of 512
> or even
> band start nominal 590 nm;  band end nominal 610 nm; reference wavelength
> 600 nm  in 0.5 of band
> 
> becaues the spectral alignment is done using some reference like an
> iodine(?) crystal for optical or a hydrogen maser+GPS for radio.  This is
> usually sufficiently accurate that it is ignorred but can become an issue
> e.g. for looking for extra-solar planets by Doppler wobble.
> 
> BAND SHAPE UNCERTAINTY
> Depending on how complex a bandpass we can handle, there will certainly be
> an uncertainty if we are e.g. using linear bandpasses only.

Hmmm; what I meant was that assuming getPassrate(wavelength) returns a 
probability value between 0 and 1 (inc), will you want that probability value to 
have an associated error? Will you want to give an uncertain wavelength?

Let us say we are using shaped bandpasses - would you need an error on the band 
shape then?

>>  - Is 'passrate(Fravergy)', returning a value 0-1 probability, sufficient for
>>Radio, X-Ray, etc?
> 
> 
> Do you mean
>  0 1 1 1 1 1 1 0    transmission probability
> | | | | | | | | |
> -----------------

No I actually meant:

   1|      xxx
    |     x   x
0.5|    x     x
    |  x       x
   0 -----------------> Wavelength

Which hopefully will clear up a lot of other comments you made!  Will people 
want any other characteristic than probability of transmission?

Thanks Anita

MC
-- 
Martin Hill
www.mchill.net
+44 7901 55 24 66