[daip] imagr/clean problem

[Eric Greisen]

Caroline E Simpson writes:

 > We've been having trouble with the clean in imagr in dec01. We have a 
 > 127 channel line data set, and on some (approx 8?) channels, clean 
 > issues a warning (abandon clean), but then continues on the same 
 > channel until it hits the iteration limit -- and the result is 
 > essentially a large checkerboard pattern for that channel. 
 > 
 > We are running this on a linux system with redhat 7.2 (I believe). 
 > 
 > When we run the clean using a unix system running dec99, we get the 
 > abandon clean messages, but the clean does seem to end, and the channel 
 > looks normal. 
 > 
 > Is there some data problem with these few channels that is causing this 
 > behavior? Or is it truly a bug in dec01? 
 > 
 > I'm including the header and input files. I can also send an image of  
 > the checkerboard channel if you like.


The abandon Clean message comes from inside the inner loop where it is
Cleaning in memory in the image plane.  It is triggered by the next
component going to higher value in a significant way.  I have looked
through the CHANGE.DOC entries and cannot find a reason why 31DEC00
and 31DEC01 should be different in their response to this, but those
entries are often incomplete.  Often, it is possible to resume the
Clean after the current components have been subtracted, so going on
for a while usually seems to be a good idea.

Have you actually watched the Clean progress?  I think the DOTV option
is the most pwerful option in IMAGR.  Usually this problem arises when
one is Cleaning too deeply into pure noise, but it is important to
watch and see what is happening.

Eric Greisen