[daip] [Fwd: cross check DFTPL-IMAGR]

[Eric Greisen]

I am tired so my brain is working slowly.

DFTPL does a rather simple summation with a result legitimately in Jy.

IMAGR does a summation in a very complex way - each sample is convolved 
with a convolving function (XTYPE controls this) so the FFT is a 
summation over convolved samples.  Then the sum is divided by the 
equivalent peak value in the beam to make "Jy/beam" units - but beam 
area is a poorly defined concept since technically it is 0.000000 in a 
dirty image lacking a zero spacing.  Converting a brightness value in a 
dirty IMAGR output back to Jy is not likely to be very accurate.

You could get the convolution down to less of a problem by specifying 
XTYPE=1 and YTYPE=1 but I still would not be sure about the units issue.

DFTPL does use the data weights in the same way as IMAGR I hope but even 
that might be different.  DFTPL sums the RR and LL weights which would 
be okay if IPOL was a weighted sum of RR and LL.  But it is not - so the 
correct weights are not so simple and IMAGR uses the correct ones:

VISOUT(3,I) = 4.0 * WT1 * WT2 / (WT1 + WT2)

If the weights are equal which they would be if you had used compressed 
data, then the result is the same.  Otherwise it will differ.

Tell the referee that the DFTPL result is likely to be better - but 
worry about it some if the IMAGR result does not suggest similar science 
to the DFTPL one.

Eric