
I am testing a new concept for the 2D representation of various values
measured by psphot.  I have been using 2D polynomials to fit the
variations of, eg, the psf parameters as a function of position in the
images.  this has been somewhat successful, but it has also been
fraught with a number of problems.  the worst issue has been the fact
that the polynomial form is not physically motivated.  in some cases,
eg the variation of the psf shape parameters, the true variations for
real distributions are not well represented by a polynomial.  I have
done some ad hoc tricks to get the functional form to look roughly
like a polynomial, but this is really quite a hack.  the other problem
is that it is difficult to decide if the polynomial does an acceptable
job representing the variations.

My new concept for modeling 2D variations in some parameter is to
define a low-resolution 2D image to represent the variation.  The
advantage is that the image may be define with whatever resolution may
be sampled by the input data, and the functional form need not be
constrained by the polynomial basis function.  

** is this a poor substitute for using a chebychev polynomial as a
   better basis function?

psphot parameters which vary by position:

pmPSF.params 
pmPSF.ApTrend
pmPSF.FluxScale (flux given normalization of 1.0)

