Explanatory note for the model spectra (from P. Hammersley)
___________________________________________________________

The models are interpolated in Te, log g and metallicity. Where measured 
values for log g and [Fe/H] exist these are used. All of the files have a 
fits like header which give the parameters used
for generating the models.  The data is then given in 4 columns :

wavelength   in microns
flux         in W/m2/micron  (it really is this time)
flux         in Jansky
error        percentage error in flux.

As the Kurucz models are poorly sampled longwards of 10 microns, data is 
interpolated in f(lambda)*lambda^4 space to give points at 0.25 microns between
10 and 20 microns; at 0.5 microns between 20 and 40 microns and every micron
between 40 and 160 microns. In principle the curves could be extrapolated to
260 microns in the same way.

The models beyond about 7 microns are basically continuum models. So there is
no SiO. The Hydrogen lines disappear at about 5 microns.

The errors are considered to be made up of two parts: 

a) A global multiplier for each spectrum coming from the error
in the absolute calibration then normalizing the spectrum to a filter
magnitude. This is called ERRFILT which I've put as 3% for all stars.
b) The errors in temperature, log g and [Fe/H] which will alter the
shape of the spectrum. In order to estimate the effects models have been 
run  changing the temperature by 3%, log g by 0.5 and metallicity by 0.3 and
then the percentage error caused by each change has been calculated. 
In fact, these changes are in general larger than the real error in 
these values.

The final errors were found by summing the individual errors in quadrature.
Currently, the individual errors values are rather ad hoc and should be
refined further. The final errors calculated seem reasonable or even optimistic
(despite the fact the individual errors are probably pessimistic).  However,
there is nothing in the errors for errors in the model tables.