[daip] jmfit versus imfit

[Eric Greisen]

On 08/26/2016 04:49 AM, Sadie Jones wrote:
> Hi Eric,
> I emailed you many years ago about AIPS during my PhD, I also met you at
> a VLA summer school, I hope you are doing well!
>
> I was wondering if you could answer some questions I have about jmfit in
> AIPS and imfit in casa. I've read the documentation but I am still not
> entirely clear.
>
> In my paper, I am trying to do a variability study of an AGN core, but
> the VLA is changing array during the observations (A,B, BnC, C,C,C).
>
> I am trying to argue that a team who use imfit in casa but use a large
> box size of 8'' by 6'' to measure the flux of a nearly resolved core
> radio blob in a VLA A map will get an  'inaccurate' value for the
> integral flux density for the core, because the core blob that we are
> trying to measure the flux of is only 0.5'' x 0.5'' arc sec but it does
> have extended structure around it, out to 3'' from the centre, which
> will be enclosed in the 8'' by 6'' box. So when they use imfit it will
> try fit a gaussian to the whole of the structure within the box and
> drive up the measured flux. And this will happen again when they use the
> same box size for their B and C array images too, but these maps will
> have more extended structure in them, which will drive up the value of
> the integrated flux. i.e. the flux increase they measure will not be
> intrinsic to the core it will be a result of the VLA lowering resolution
> & they fact they use integral flux (over peak) & the the large box size
> they use...
>
> In our analysis of the same VLA archive data we use a different box size
> around a radio core for each of the different configurations, and try to
> enclose just the core flux and exclude as much of the extended emission
> as possible,  the box sizes we used are below:
>
> 'The box size for each array configuration was chosen to
> just encompass the synthesised beam for that configuration so that we
> can use JMFIT to measure accurately the peak and total flux density of
> the core
> Gaussian components. Thus we used boxes of size $ 1''\times1''$,
> $2.55''\times1.69''$ and $8.4''\times8.4''$ for images from the A, B and C
> configurations respectively, all centred on the position of peak flux.'
>
> So my question is, do you agree with what I have said, do you think our
> argument is sound or is there something that jmfit and imfit do that I
> am missing/not understanding?

I am inclined to agree with you although I think that it must be hard to 
measure the core flux in the C array because it will blend with the 
extended flux.  I also found with a small-diameter source that the flux 
was less accurate with larger boxes even when there was only noise in 
the larger box.  The Gaussian fitting tries to take that noise into 
account and this makes the fits get strange.  In fact, I encourage users 
of SAD (JMFIT to automatically find sources) to include only higher 
brightnesses to solve for the strong sources first and then go to lower 
brightnesses.  When one does not do this, fits on 100:1 objects can fail 
because of the noise at their lowest levels.

I do not know anything about CASA's imfit but I assume it is a simple 
Gaussian fitter in the map plane.  To get a better core flux you might 
try looking at the visibility data at the longest spacings rather than 
the images, esp at C array.

Eric Greisen