NRAO Home  >  Green Bank  |  Wiki Topic:    GB > Pennarray > OrionMapStability

Changes | Index | Contents | Search | Go

Map Reconstruction Tests

bsm 13jun08

Parameter Dependence

I ran runs 1, 2, and 6 Orion maps through with a variety of reconstruction parameters, varying chiefly:

• the number of clean components (50,000 --> 300,000)
• the minimum flux to clean to (5 mJy or 10 mJy)
• baseline and atmospheric timescales (20,10) sec or (30,20) sec
• set of iteration timescales-- [5,2,1] or [6,3,1,1] (times solInt=1 sec)
• starting from no clean model vs starting with a prior (in the cases tested so far, the prior was always our best total orion map, Orion1.6.fits from the protected orion results wiki page)

First, it was determined that for the orion maps on our good days, a minimum flux of about 10 mJy is needed to get convergence in a reasonable number of iterations (10 mJy gives convergence in 73,000 clean components with a gain of 3%; 5 mJy didn't converge by 300,000 cc's). Most comparisons here use the 10 mJy minflux.

All contours are spaced from the map peak, down by sqrt(2) each, all the way down to about 8 mJy.

Following is run 1, comparing:

• 1e5 cc's; (20,10) and [5,2,1] timescales; 10 mJy minflux (white)
• 1e5 cc's; (30,20) and [6,3,1,1] timescales; 10 mJy minflux (red)

Folllowing is run 2, comparing:

• 1e5 cc's; (20,10) and [5,2,1]; 10 mJy minflux (white)
• 1e5 cc's; (30,20) and [6,3,1,1]; 10 mJy minflux (red)

Following is run 6, comparing:

• 5e4 ccs with (20,10) and [5,2,1], 5 mJy minflux (white)
• 1e5 ccs with (20,10) and [5,2,1], 5 mJy minflux (Green)
• 1e5 ccs with (30,20) and [6,3,1,1], 5 mJy minflux (purple)
• 1e5 ccs with (30,20) and [6,3,1,1], 5 mJy minflux, starting from a prior (magenta)

The run 6 tests could stand to be run again with a 10 mJy cutoff.

Clean vs Dirty

The clean and dirty maps, even for reconstructions that turn out quite differently, trace each other very well. What this says to me is that the important effect is the degeneracy between the astronomical model and the instrument model, and not the ability of the models to represent the data -- in retrospect perhaps obvious, though it was still possible that we would have had effects that were not being captured well in the model.

Here is run 6, 100,000 ccs (30,20) [6,3,1,1] 5 mJy minflux, started from the prior (clean is white)

Here is the image resulting from the same data and reconstruction parameters, but started from the prior-- again clean (white) vs dirty

Comparing Days/Pointing

My original goal was to compare the maps from our best runs. By plotting the maps from runs 1 and 2 on top of each other it is clear that pointing must be accounted for in this comparison! There's a 1 arcminute offsets between them which is bigger than I would have expected. Rerunning now with pointing put in.

Following is run 1 (white) vs run 2 (magenta) with no pointing corrections applied

-- BrianMason - 13 Jun 2008

Attachment:	Action:	Size:	Date:	Who:	Comment:
orion-run1-cc3e5vsCc1e5tweak.png	action	28882	13 Jun 2008 - 12:35	BrianMason
orion-run2-cc1e5vsCc1e5tweak.png	action	26328	13 Jun 2008 - 12:35	BrianMason
orion-run6-clean-nomVsCc1e5vsTweakVsTweakWprior.png	action	58107	13 Jun 2008 - 12:37	BrianMason
orion-run6-cc1e5tweakWprior-cleanVsDirty.png	action	23841	13 Jun 2008 - 12:38	BrianMason
orion-run6-cc1e5-cleanVsDirty.png	action	40083	13 Jun 2008 - 12:39	BrianMason
orion-cc1e5trunc-run1vsRun2.png	action	27973	13 Jun 2008 - 12:40	BrianMason