[daip] IMAGR and msclean

[Eric Greisen]

Elias Brinks wrote:
> Hi Eric,
> 
> I have a bit a detailed question regarding multi-scale clean, as it has 
> been
> implemented in IMAGR. I'm a bit confused, after reading the description
> in the Appendix of the NGC6503 paper and the explain file. Basically,
> if you clean HI data (so just 1 facet, but 5 or so different scales), how
> does IMAGR go about finding for each iteration the next clean component?
> I am led to think that IMAGR finds the scale with the highest flux and 
> subtracts
> this from all scaled maps (in the map plane with a beam patch a la 
> Clark-clean).
> IMAGR then searches for the next highest peak (this can be on any 
> scale), etc.,
> until it decides that it is time to do a proper subtraction in the 
> uv-plane...it then
> recalculates all scaled maps and starts another minor cycle, etc. And what
> is the criterion (in broad terms) to do a major cycle?
> 
> I'm asking as my collaborators seem to think that IMAGR does a 
> multi-resolution
> type clean, where it takes the largest scale and cleans this down to the 
> flux
> limit, then it attacks the next scale down, etc. So this would be the 
> old Wakker & Schwarz
> scheme. I think they reach this conclusion because the explain file 
> actually uses
> the term multi resolution clean, but in your NGC6503 paper you clearly 
> talk about
> multi scale clean (and in the explain file you refer to the Holdaway & 
> Cornwell
> presentation on multi scale clean which prompted you to incorporate this 
> into
> IMAGR).
> 
> Any light you might be able to shine on this issue would be very much 
> appreciated!

The nomenclature seems to be confusinng people.

I initially called the multiple scale/resolution Clean a 
multi-resolution Clean to avoid conflict with Cornwell and because I was 
unaware of the silly scheme which apparently goes by that name. 
(Cornwell still formally accused me of plagiarism to NRAO's chief 
scientist!)

And none of you are actually reading the words that are written - or 
reading the messages that come from IMAGR.  Frankly I am amazed that you 
have such confusion.

IMAGR selects a resolution to Clean for the next major cycle based on 
the strongest one after some scaling to down-weight the larger beams.
It then finds in and takes from that image a number of components in the 
usual Clark-Clean fashion.  These new components are subtracted from the 
residual UV data and the process cycles.  So more than 1 CC is taken at 
a given cycle without looking at the other resolutions but all 
resolutions are decreased together (in a jerky sort of way).  At any 
time, we are always dealing with the residual image/uv for the set of 
CCs found over all resolutions (and spatial facets).  I suspect that 
CASA does what you suggest - subtracting each single CC from each 
resolution image with appropriate "cross-beams" in an image plane way 
although it may then cycle to the UV plane occasionally in a Clark-Clean 
way - but I do not know.  That would explain some of its extreme slowness.

Frankly I am really upswet at how badly your group has interpreted what 
is written - I think it is written rather clearly and yet none of you 
have understood it at all.

Eric Greisen