[alma-config] reply to mark

[John Conway]

Hi,

After sending my email yesterday I was casually Googling
and found the following review in the Annual Reviews of
Astronomy and Astrophysics 2005 - entitled 'Digital Image
Reconstruction: Deblurring and Denoising' at

http://www.pixon.com/publications/Puetter_05_ARAA.pdf.

probably everyone knows about this already  except me.

On page 3 they say '..After decades of development the literature
on this subject is still unusually editorial and even contentious in
tone...' just so

On Marks point about when stopping cleaning extended emission
prematurely and leaving the extended structure in the residuals,
I agree there is in practice a real problem, the 'clean bean' and 'dirty
beam' are obvioulsy not the same. This can mean that that before and after
CLEANing deeply extented emission the brightness of extended emission can
change quite a bit which might effect the astrophysical numbers at the
end. I don't think the average user is aware of this, he/she stops when
there is 'nice looking' image. This is why I said in this case that users
'should' really  in this  case be cleaning much deeper. In the end one
wants  the final  image model  to  fit *all* the uv data which is
determined to be signal,  not just to  give up and not fit the short
baselines at all. If that is not  achieved you end up with some mongrel
map which  not really good for anyone. Maybe we should soon retire the
30 year old classical clean and only offer multiscale type cleans. This
would give some degree of denoisng for all scales and would revert to
classical clean when the image really was bunch of dots.

Another issue that arises from the modern view that one estimates
both image and noise - is why do we in radio astronomy always add the
noise back at the end?? (another discussion that never dies)  This is as
far as I know only  done in radio-astronomy. The
reason I think is that our field is sensitivity dominated and when
we had images with a narrow ranges of spatial scales additting the
noise back was by far the the  simplest way of showing the significance
of a feature, one  could see with ones eyes wheter a  compact feature was
several times the noise background. It however has lots of disadvantage,
in particualar  when the image has many scales, one does not see the
smooth weak emisison abouve the noise in the full resolution map even
though it is highly significant in teh uv data. In this case radio
astronomers  then resort to publishing maps at seveal resolutions. An
alternative might be to just publish a single image  containinng
all that is   signifcant above noise  from components in all scales, and
then estimates of the noise at  different  resolutions.

      John