refReport2.txt

Many thanks to the reviewer for the quick turnaround on this
manuscript.  We hope that we have adequately addressed the
remaining comments below.

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

    1)
    This seems not to be the right reference; section III does
    not have a subsection A, and section III in Lui et al 2014b
    does not discuss the issue (neither
    http://journals.aps.org/prd/abstract/10.1103/PhysRevD.90.023019
    nor http://arxiv.org/abs/1404.4372 )

Apologies for this mistake.  It was Section IV, subsection
A.  We've changed this in the manuscript.

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

    2) I agree with the new footnote's contents, and I think its
    text is somewhat helpful. However, I still don't agree with
    the main text. In the way the main text describes this, the
    authors seem to incorrectly generalize that rescaling Vxx
    and Vyy with a scalar gives the least possible leakage when
    uv-samples don't overlap. However, using a kernel instead of
    a scalar will provide information about more modes.
    Therefore, in this sentence: "As such, for a chosen uv
    coordinate, the level of leakage in the Stokes I uv plane
    can be no better than [a scaling factor]" -- One can choose
    a uv coordinate that would be inside a beam correction
    kernel, but is not at the centre of the kernel or sampled
    otherwise, and according to this sentence it would still be
    corrected. However, a single scaling factor at the centre
    will of course not accurately correct this uv-sample (but
    rather leave it zero). So this statement is not fully
    correct, but only holds for the uv-samples at the centre of
    the kernel.

We have changed the sentence in question to:
"As such, for a chosen uv coordinate, the level of leakage in
the Stokes I uv plane effectively reduces to what can be
achieved using a single number to rescale VXX and VYY before
summing."
We point out to the reviewer that the scaling factor can be
chosen to match the XX and YY kernel amplitudes at a 
uv point that is not necessarily the center.  This
equivalently means that the responses of the XX and YY beams 
can be chosen to match in any one chosen direction, but not
arbitrarily across the beam.

    This also relates to this response:

    "Page 11, col 1, line 8: "In the case of... products" this is
    dependent on the above points, and thus needs to be changed
    too if above points can't be clarified.

    This point has been addressed in the clarifications above. "

    I disagree with this sentence: "[..] the result will be
    identical to the best that can be achieved by independently
    imaging [..]", because for scalar correction, information is
    lost for the modes that are contained in the kernel. As an
    example, if one observes one strong, dominating source away
    from zenith (e.g. one of the 'A' sources), even the 'image'
    of a single visibility will accurately correct the
    (modelled) instrumental polarization when convolving with
    the proper kernel, while a single scaling factor will not.
    That using fringe filters with polarization correction are
    an improvement over no correction is well shown, and that
    actually doesn't require these statements.

We have changed the sentence in question to:
"In the case of sparse array sampling, fringe-rate filtering 
reproduces what can be achieved by independently imaging"
As per our response above, a scaling factor between XX and YY
can be chosen to match the beam responses in a chosen direction,
in this case, the direction of a single source.  

    Finally, personally because of its length I would not have
    made a footnote of footnote 8, but that's up to the authors
    / editor.

We have gone ahead and made this footnote part of the main
text, per request.