[daip] Gaussian fitting/correlated errors

[Patrick Leahy]

Hi Folks,	
	I was glad to read in the 30 June AIPS Letter that someone is
working on correlated errors in Gaussian fitting. I hope you will extend
this to the 1-D fitting in SLFIT and XGAUS, which definitely give
unreasonably low errors if there are correlations between data points
(always, in the case of slices). My application is attempts to measure
the width of a jet, i.e. I'm fitting to rows of a map.

SLFIT is a particularly big offender because the initially oversampled map
is oversampled again along the slice; if the original slice had a fairly
small number of pixels the oversampling becomes fabulous with a minimum of
512 pts per slice.  Pending more detailed action, it might be an idea to
update the EXPLAIN files for Gaussian fitting routines to say how the
"errors" are actually calculated.  It would be nice to extend (reduce?)
Jim Condon's analysis to the 1-D case! 

I also noticed that when calculating errors in SLFIT, the value of
TOL passed to GETERR by GAUFIN doesn't seem to be initialised. Looking at
what TOL is actually used for when it eventually arrives at COVAR it might
be an idea to set it to zero but it would be nice to do so explicitly!
(XGAUS sets TOL to 1D-5, the fitting tolerance in LMSDR).

cheers,
		Paddy Leahy