[daip] multi-scale clean in IMAGR

[Eric Greisen]

Eva Schinnerer wrote:

> I would like to use the multi-scale option for CLEANing in IMAGR for VLA 
> radio continuum data of M51at 20 (ABCD), 6 (BCD) and likely 3.6 (CD) cm 
> data. I was hoping that you could help me.
> 
> I wanted to know which keywords I would need to use and what the 
> important ones are. Also is there a document where I could find more 
> information on how to use this option? And also I was wondering how the 
> IMAGR version of multi-scale clean differs from the one implemented in 
> CASA. I would prefer to work in AIPS, as I haven't used CASA yet.

I thought I had added more "lore" to the IMAGR help file than I 
apparently have.  The first question is the number of Gaussians - 
probably at least the number of configurations, perhaps more.  The 
relative size - find out what your robust 0 beam width is with a quick 
NITER 0 IMAGR.  Then, the first serious user I had tried sqrt(10) 
between the different WGAUS - e.g. wgaus = 0, sqrt(10)W, 10W, ...
where W is the approximate size of the 0 width beam.  This gives factor 
of 10 differences in beam area - much less is hard to differentiate but 
a factor of 8 or 9 would also be good.  The remaining knobs that matter 
are IMAGRP(11) and FGAUS.  The former is critical since a point source 
with a little extended stuff will be cleaned with the fattest model (a 
bad fit) unless you discourage that - and then many, many small 
components will be needed to undo the error.  What I do is to make the 
full set of images with NITER=0.  I look then at the peak flux in 
Jy/beam from each resolution and try to adjust IMAGRP(11) so that all 
resolutions will be about equal after weighting.  FGAUS is important if 
you are going to Clean a spectral cube for example and do not want to 
use the TV for each channel.  I think you can stop each resloution wiith 
the TV without FGAUS is you are doing it only once.  Normally, one wants 
something that stops each resolution at about the same Jy/pixel so 
rather more in Jy/beam at the fatter sizes, e.g 0.0005, 0.003, 0.05, 0.1
(not quite the same Jy/pixel but of that order).  I find the TV 
essential to watch what is happening and to decide when enough is enough 
in each resolution.

Note that it takes > 200000 points to begin to Clean Cygnus A X-band 
while a few 1000s get almost all flux with the multi-scale Clean.  It is 
a good thing.

I think that casa does not implement this option.  It is in the aips++ 
tool set but not yet in casa.  The initial algorithm described by Mark 
Holdaway and Cornwell was an image-plane scheme, but they have probably 
done other schemes in aips++.  I tried to look at their documentation 
once and it was hopeless - it should be better now but I would not know.

Eric Greisen