[daip] Corners of dirty beam

[Eric Greisen]

Andy Biggs wrote:
> Hi Eric. I'm making multi-faceted maps with GMRT data (128 channels, 2 
> IFs) and am finding that the values of the dirty beam are quite high in 
> the corners, higher than the peak in fact. Doing a bit of googling I 
> came across a daip entry where you say:
> 
> "The dirty beam has a peak value of 1.4 Jy (in the corner I
> presume) which suggests that some things are not going well in the
> imaging.  The corners can get bad with very large numbers of uv
> samples being gridded into modestly small images - round-off error
> from so many computations, corrected by a large factor in the corners
> does make for trouble.  This then messes up the histogram of things.
> ...
> Try making your image larger, IMSIZE
> of 1024 or 2048, to see if the beam becomes better behaved."
> 
> This is a pretty big dataset and so fits with what you say. I ran SETFC 
> and that recommended 1024 pixel maps with a 0.96 arcsec cellsize, which 
> I rounded up to 1 arcsec. I've tried changing the size of maps and 
> various weighting options, but the corners remain nasty. Can you 
> recommend some way of making my dirty beam flat out to its edge?
> 
> Thanks.
> 
> Andy
> 

Yes but it requires going to a double precision "array processor".  I 
tested this at one time and it in fact works.  But - the amount of 
memory required for a given computation doubles and the performance goes 
down by 10% or more depending on details of the problem and computer.
AIPS works hard to avoid using the corners which are unreliable even 
when they are not nuts and the histogram algorithm stays out of corners 
rather well these days.  I suggest you live with it unless you think it 
is affecting scientific results.  My judgment on the DP array processor 
was that it was satisfying computationally but would not change the 
scientific results.

Eric Greisen