Hi Jaap,
   A nice document. Here are a few minor comments or suggestions.

1) First paragraph.  The missing spacings addressed by the ACA, for the
benefit of the regular ALMA array, are something like 6 to 12 m, rather
than 0 - 12m.  There's no advantage - in fact a major disadvantage - in
trying to use the smaller antennas to get the uv spacings extending
from, say, 2m up to a few m.  The 12 m dishes, in total power mode, will
be much better at that.  Roughly, the responsivity of a 12 m dish in
total power has fallen to ~50% at a 6-meter baseline, while the
responsivity of an interferometer with 18m separation of 12m antennas
has fallen to ~50% at uv spacings of 12 m. So, the ACA addresses just
6-12m approximately.    
   Roughly, 6-m antennas in total power will only be useful for uv
spacings up to 3m, so don't really bring anything at the shortest
spacings <6m, over the existing 12m antennas.

2) Notwithstanding point (1) above, if the ACA is used as a standalone
array above 1 THz, then of course some of the antennas will be needed
for the short baselines 0 - ~3m in that standalone mode, assuming the
12m antennas aren't usable at all at the highest frequencies.  This has
been addressed elsewhere by Stephane and others.

3) Sections 2.1 amd 3.1, Ad 4..  For adding in the shorter spacings, the
ABSOLUTE pointing is what counts, not the pointing as a fraction of the
antenna beamwidth. If we relax the pointing specs on the ACA antennas in
terms of arc secs, we'd get better images and better data over the
relevant u-v spacings by using the 12m and ignoring the 6m antennas. So,
the ACA pointing in arc sec has to be at least as good, preferably
better, than the ALMA antennas.  As a fraction of a beamwidth, that
means at least twice as good as the ALMA antennas.

4) Section 4.3.  I agree, but if we ever give up the idea of THz
frequencies with the smaller dishes, there would be no point at all in
having nutators on the ACA antennas. (See my point (1)).

5) Section 5.1.  Optics.  Would we ever want to do terrestrial
holography with the ACA antennas?  If so, then would we want to use the
same holography receiver?  If the answer is "yes" to both of these
questions, then the prime focus optics (i.e. f/D ratio) should ideally
be similar between the ACA and ALMA antennas (no big deal - we could
always change a feed) but more importantly the prime focus mount should
be strong enough to hold the 12m holography receiver.
  I suspect that by the time we have the ACA, we'll be able to do
astronomical interferometric holography using the ALMA array as the
reference antenna, so obtaining sufficient s/n on astronomical sources. 
So, I doubt whether we'll ever want to use existing coherent holography
receivers on the ACA.  However, someone should do the sums to ensure
that sufficient s/n will indeed be available with astrometric holography
and ACA antennas in order to get sufficiently precise surface
measurements.

The talk of the ACA and possible operation above 1 THz reminds me of the
following detail:

   The baseline spec. for the ALMA interferometer local oscillator
coherence (i.e. fast phase jitter, too fast to be corrected by
calibration) is set such that 90% coherence is obtained between an
interferometer pair at 950 GHz.  This spec. is analagous to the antenna
surface accuracy spec;  the frequency dependence is the same as for the
Ruze law applied to antenna surface errors.  This 90% coherence at 950
GHz spec. is extremely difficult to reach, whether by multipliers or by
a direct photonic system.  Of course atmospheric phase noise adds to
this loss of coherence.

   For reference, if the same local oscillator system were to be used at
1.5 THz, the coherence would become 77%, while at 2.5 THz it becomes
48%.  So, with the current specs, a system using even perfect antennas
would be degraded to less than half the theoretical sensitivity when
used as an interferometer with this local oscillator system at that
frequency.  If serious interferometer operation above 1 THz is
contemplated one day, either this loss has to be accepted, or at least
parts of the LO system will have to be re-engineered - if indeed that's
technically feasible.

   Another detail, that's probably too obvious to mention: the
sensitivity calibration in single dish mode, such as made with some
hot/hotter load system or whatever, has to be tied somehow to the
sensitivity in interferometer mode, which will always be degraded by
finite LO coherence compared to single dish sensitivity.  The baseline
LO coherence spec implies a 1% degradation at 293 GHz, which needs to be
taken into account if an overall absolute calibration accuracy of 1% is
to be obtained.