[daip] JMFIT

[Eric Greisen]

Ian Pattison writes:
 > Using JMFIT I'm a little uncertain as to what effect the size of the
 > fitting regions has (i.e. setting blc and trc). For example when I use a
 > small box, of similar size to the source at it's 3 sigma level, I get a
 > deconvolved size of 0 in the minor axis. The source does look (fairly)  
 > un-resolved (it's not exactly circular, like the beam). If I set the
 > fitting region so that it encompasses all the 3 sigma emmision and above,
 > then presumably I'm missing out all the emission weaker than this (though
 > it'll be hidden within the noise). When I use a slightly larger area I see
 > that the deconvolved axes increase in size, as I increase the box size.
 > How can I decide what the 'best' or 'correct' size fitting region to use
 > is?
 > 

I do not know if there is a best size.  You have discovered something
which my Monte Carlo testing showed up but which I have not been able
to characterize in any detail.  Basically, as the window gets larger,
the uncertainties in the fit parameters increase (as shown by Monte
Carlo testing).  When doing the fit in a larger window the
minimization routine must deal with beam-size lumps separated from the
obvious source.  This is likely to decrease the fit peak brightness
while increasing the fit size.  Larger sizes allow for the
deconvolution to succeed more often.  The 0's => that the
deconvolution needed to take a square root of a negative number.

Of course, if your source is not Gaussian, a larger box is likely to
show that more than a tight box around the main point.

Nonetheless, results from smaller windows are more likely to be
reliable.

Eric Greisen