Dave:  Thanks for your thoughtful comments on all of the key issues.  I agree with what
you wrote.  I believe that the science that we will do with this instrument will demand observinglarge fields much of the time.  This means that for best point source sensitivity in these fieldswe will be in the regime where we need to optimize nD.  Indeed, the optimization does not 
produce a strongly peaked function.  However, going in this direction does push for smaller
antennas.   On the whole, future improvements will be more dramatic in the electronics than in
the development of mechanical structures.  Also, although we are hearing that some designers of 
antennas are confident that large antennas are no problem, I would feel more comfortable about it
if there were any recent examples of large millimeter antennas that came close to meeting the 
specs that we are hoping to meet.  This fact makes me think that there is less risk building
more electronics than building larger antennas.  

Jonas' developments are good examples of the future in front-end electronics.  As you know,
wideband HEMT amplifiers with low temperatures are becoming quite inexpensive.  Sandy Weinreb
showed a remarkable mimic chip amplifier that he had recently made.  The chip was about 20
mills accross and had two or three HEMTS inside.  With almost no tuning structures, he got a
low pass amplifier that was flat out to 25 GHz and then trailed off to 50 Ghz. The future here
is going to at least 300 GHz, he believes.  It all comes about because of the small size and
corresponding tiny amounts of random capacitance and inductance.  And all of this is coming 
along very quickly.  Also, this kind of technology can be prototyped at modest cost and then
built for very little cost.  It does not require thousand of devices to get into the game.  And, as you well know, the FPGA's are making a revolution in correlator construction.

My proposal would be to make a compromise on the antenna size, so that no one is going to feel
uneasy about the choice.  The logical compromise between 7m (to optimize for nD) and 13m (to
optimize for nD**2) is 10m, the compromise that we have already made.  That is, I don't think
that the choice is between 10m or 12m, but rather 8m and 12m.