[alma-config] Comments on Marks proto-memo

[John Conway]

-------------------------------------------------------
This is revised version of an email sent yesterday,
which I think didn't get forwarded to eveyone because it contained
large .ps attachments, here is a new, hopefully more concise
version, without the .ps

---------------------------------------------------------- 


Hi,

 I agree that mosaicing simulations should certainly be done
and they will be a vital part of deciding between array 
styles. However  we are waiting for the AIPS++ software to 
be finalised we can at least do some simulations of sources
which do not fill the primary beam. The reasons are;

1) The size of reconstruction errors for non-mosaicing 
observations are interesting in their own right. These will be a 
signifcant part of ALMA observations and should be an input into the 
decision process between array styles. This simulation effort is 
not wasted.

2) I agree entirely with Ed's comment that one should concentrate 
the  work and simulations on whatever is limiting the 
quality of the reconstructions, whether its uv coveragem
phase calibration or pointing errors. The problem is 
we don't have good esimates of the their relative 
importance, in particular we don't know the level of
reconstruction errors which arise solely due to limited uv 
coverage (!). The assumption in Marks memo was that these are 
likely to less than 1000:1 (for both long track and snapshots)
and so pointing errors are
important, but we really don't know that (...because we havn't 
done  many simulations...) the few simulations I've 
done suggest such pure reconstruction errors due to limited
uv coverage  can be much  larger than this. The only  way to 
find out what they are and their relative importance c.f. phase and
pointing errors (which can be estimated to first order a priori) is 
to do some simulations.

3) We can start the process of developing, arguing and agreeing
the metrics to be used in the final evaluation of array designs.

I argue we go forward now with A and B array simulations (concentrating
on Bb where there is a difference between styles) of 5 relatively 
compact sources, later on add C,D,E, simulations of larger
sources, using mosiacing and AIPS++.


    John.



P.S For those who like reading long emails, below are some 
comments on things I disagree with in Marks protomemo and why.
 

----------------------------------------------------------------

ASSERTION 1)  The Kogan and  Conway styles are similarly tapered 
and imaging differences between them will be small. 

Some new comparisons between the two array styles can be found at
http://www.oso.chalmers.se/~jconway/ALMA/SIMULATIONS/SIM3/

For arrays having the same resolution these plots
show that the maximum baseline length is about a factor or 2 larger
for the spiral, but this is due to the very low density of uv points
at the uv edge.  However as expected for uv coverages giving the same
resolution  the mean and median baseline lengths are the same for
both styles but the spiral is more tapered and close to
gaussian (see plotted model). The Kogan array is
more tapered  than a single ring of course, but there 
is stiil a sharp step at the uv edge, hence the long 
track uv dirty beam is dominated by N-S near-in sidelobes at 
about the 4% level. We can argue (and have argued) which of
two styles is better but we can agree that both the uv
coverages and beams are significantly  differnt; it would therefore
be interesting to see how even non-mosiacing imaging simulations differ
for the two styles.

-----------------------------------------------------------

ASSERTION 2) Reconstruction errors due to incomplete uv coverage
are likley to be less than 1 part in 1000, hence 
pointing and other arrors can become dominat.

At the Tuscon CoDR I showed some VTESS simulations
of the  M51CO image for a 6hr track, for both 
a donut-0.25 and zoom spiral array. To work out 
the fractional errors I have divided the image with the model, 
clipping below a cutoff of only 10% of the peak brightness,
i.e. I looked at the fractional errors only in the
brightest regions of the source. For the donut 
the rms fractional error was about 0.7% and peak about 3%.
For the spiral the corresponding numbers were about a factor
of 2 less, and apears to be limited by the added thermal
noise (I should rerun the simulations with lower additive
noise to get a better idea of the fundemental ratio in 
quality).  Now this is a just a single image and there might
be something special about it, one would
really like to see if we get the same results on a bigger
set of images.  

Note that the above results  was for a long track 
and the fractional reconstrcution errors for 
snapshots (the main style when mosaicing) one can only 
be expected to be larger.

--------------------------------------------------------------