[alma-config] BOUNCE alma-config at majordomo.cv.nrao.edu: Non-member submission from ["David Woody" <dwoody at caltech.edu>]

[Mark Holdaway]

Dave, as usual I bypass most of your arguments and just
address the little bits I know about.

> It seems to me that the question at this point becomes:  
> What taper do we want in the raw UV-coverage?
> Uniform (no taper), linear, 1/UV, gaussian, cosine....

Concerning uniform coverage:  if you fit a Gaussian to the Lambda function
PSF, its a very good fit in the inner part, but the Lambda function falls
off very fast and goes quickly to the first negative sidelobe (due to the
"excess" long baselines), while the Gaussian goes gently for a while to
zero.  To correct an image for the Lambda function PSF, you deconvolve
(extrapolate in the UV plane, ala Conway); and THEN when you restore, you
convolve with the gaussian, which is equivalent to tapering in the Fourier
plane: so, you are throwing away much of the extra long baseline
information you worked so hard to get in the first place.

Another question:  how often will superresolution be used/trusted?
I think that is the main argument for uniform (ie, the heaviest we
can weight long baselines) coverage.  If superresolution remains
a dirty word as it largely is today, I don't see compelling arguments
for uniform coverage.

> Clearly we want to maximize Ed Fomalont's smoothness
> metric.  All of these tapers will yield "complete"
> coverage out to some maximum configuration diameter with
> uniform giving the largest diameter configuration with
> complete coverage.  As noted above, earth rotation
> can be effective in increasing the long BL coverage. 
> Combining several configurations also gives tapering.

Unanswered question: how will we use multi-configuration
ALMA data?  Tapered coverage can be "simulated" from arrays
with uniform coverage by combining them in a "wedding cake"
multi-tiered manner.

> It may be more important to have configurations that
> cover the two approaches of uniform and tapered UV 
> coverage than it is to have many different scale sizes.

Simon Radford likes to make this point: we can't agree, so
build pads that allow both and let the users decide.  Some merit,
but its also a cop out.

> It is possible that images produced using uniform UV data
> will have resolution function (FT of disk = J1(r)/r )
>  artifacts that degrade the image, 
> but this is not the same as saying the science information
> is degraded.  A close call science conclusion should be
> tested against the raw UV measurements and not against
> a heavily processed image.  Of course we all live and 
> die by the visual image on the cover of Nature, so it
> is important that we present a nice picture that backs
> up our conclusions.  This will probably require some
> tapering of the UV data to remove resolution function
> artifacts.  

Tim Cornwell was the first to make this point, in 1991: there is no great
need for higher resolution in any but the most extended configuration.  
If you need higher resolution, go to the *^^#%#$-ing next larger array.  
It is more important to be relaxed about resolution envy and go for good
imaging quality.  Of course, when you are at the biggest *^^#%#$-ing
array, there ARE compelling arguments to have the highest resolution
possible --> hence Conway's transition from spiral to circle.  Conway's
gently increasing array design makes this argument more compelling: if you
need more resolution, you don't have to swallow it in factors of 4.
(Hey John, I think I'm becoming a convert.)

	-Mark