Response

Frank Valdes wrote on Oct 22, 1998

Q:	I'm reducing some spectro(photometry) that I just took at the coude-feed 
and have run into a bit of a snag.  Because I took spectra that stretched all 
the way from 3000-6500A, the intensity of the quartz lamp varies very much over 
the flat field (Daryl found a filter that helped some, but there still isn't 
that much flux at the short wavelengths).  Now when I try to divide out the lamp 
(in RESPONSE) I can't seem to not get bad "ringing" in the spectrum, and then 
primarily in the blue end (where there are few flat field counts, but where the 
most important part of the object spectra are).  I have tried increasing the 
binning size (up to 300 - out of 3000 pixel) and that helped some, but not 
fully.  I also tried different functions, but it seems that in all cases the 
fits go "bonkers" at the ends of the spectra, or leave very large 
residuals/rations (if I use a low order fit).  Any wise and experienced advice 
that you can offer?  Is there a way to just fit a set of connected lines?
	

A: The problem you are seeing is not so much that the fit wiggles more
when the signal is low but that the response ratio, data/fit, amplifies
the wiggles.  There is not much you can do with a single
response operation.  Either the fit is of high order with problems associated
with that or of low order where you can't fit some of the shape.  However,
you touch on a possiblity of doing the response in pieces.  It happens
that RESPONSE actually supports doing this by using image sections.

	cl> response data{*,1:512] data[*,1:512] response
	cl> response data[*,513:2048] data[*,513:2048] response
	
In the first step you say to do the response only on the first 512 lines.
Unlike most IRAF tasks this task will still create a full size output
based on the true size of data and not the section.  The region beyond
the section is filled with ones.  The second step adds other pieces.
You might get a better result if you overlap the two regions.

Of course, the importance of the wiggles depend on whether you will flux
calibrate the data with standard stars.  In that case it really doesn't
matter much what you normalize the flat field since whatever shape is used
will then be removed during the sensitivity function determination.

---

Last post on Oct 22, 1998