From: 	astrocat1@gmail.com
	Subject: 	Re: Test run results
	Date: 	March 12, 2010 12:20:38 PM PST
	To: 	fmasci@ipac.caltech.edu
	Cc: 	tim@ipac.caltech.edu, hlm@ipac.caltech.edu, jwf@ipac.caltech.edu, roc@ipac.caltech.edu

Frank wrote:
Could you remind me on what made this work. My recollection from previous discussions: 
the linear+cross terms from gndstr were explicitly defined in the SIPs (along with the CD matrix) so SFPRex could get a better solutions for a subsequent gndstr run to give better non-linear coeffs?

John wrote:
We just have to have a very vanilla CD matrix to generate the RvB files, then let gndstr fit everything, then use everything gndstr computed in the FITS header, i.e., all SIP coefficients, including A_0_0, A_0_1, A_1_0, etc. for B and AP and BP.
I think this means that W3 had some skew, and now we realize that skew can be absorbed into the linear SIP coefficients, since they include the cross terms.

Tim wrote:
The problem was, the polynomial fit had power in the cross terms, which could not be translated to the b2b+scale info that went into the CD matrix. So we punted and went back to the fops scales and b2b and simply took all the SIP coefficients, including the linear terms. Thus what the pipeline did and what John's poly fitter did were exactly the same.
John and Howard are now working on a refinement to try and improve the b2b and scales to better vlaues that we already have and then do a fit based ont he improved values. That will allow the highly visible WCS CDELT and CVAL terms to be more meaningful.

Howard wrote:
This test consisted of letting gnDSTR absorb all corrections needed (linear & non-linear) to tie each band to 2MASS.  That takes care of b2b changes as well.  The problem with that is that it masks the real scales and b2b.  Am starting work on a technique which still uses the both the linear & non-linear terms coming out of gnDSTR, but pre-linearizes the b2r differences in each band.  That should provide the same results without having so much of the b2b and scales hidden in the distortion coefficients.