[evlatests] Dynamic range of images from Widar.

[George Moellenbrock]

> 1. We usually run the closure-preserving calibration (FRING, BPASS,
>    CALIB) on shortish timescales, reach a floor in the rms, then do
>    non-closing corrections with BLCAL over the whole timerange.
> Q: Are there antenna-based errors left over, that show up on longer
>   integrations (higher SNR)?
> A: No - long CALIB solutions do not make a dent in the corrections
>    needed by the final BLCAL.

This also should not be surprising.  Short-timescale solutions will, _on 
average_, pick up and account for long-timescale effects, even if they 
have insufficient SNR to detect them _per solution_, unless the 
short-timescale solutions are deliberately filtered (e.g., by 
normalization) to avoid this.  In other words, the solution interval alone 
is not such a filter for effects occuring on timescales longer than the 
solution interval.  (Effects on shorter timescales _are_ effectively 
filtered, and we've relied on this explicitly, e.g., "wobbles" and 
"clunking".)

If one wants to detect and state a long timescale effect, one must obtain 
sufficient SNR to be convincing, either by using a long solution interval 
or averaging together many short timescale solutions in the right way. 
But this is not required to achieve properly corrected data from short 
timescale solutions, so long as the short-timescale solutions themselves 
aren't noise-like.  (~Noise-like short-timescale solutions will correct, 
on average, _only_ the long timescale effects that can average coherently, 
interestingly.)

A more interesting question is whether or not the short-timescale 
antenna-based calibration changes in any interesting way by applying a 
constant BLCAL solution.  Presumably the BLCAL solution will change the 
relative contributions of different baselines to the antenna-based 
solution. Of course, by definition, the constant BLCAL solution will 
preclude any post-BLCAL antenna-based _constant_ component.

These considerations are always complicated by the fact that any type of 
solution will enforce some mean consistency of the calibration model and 
the data, and that this changes the mean properties of subsequent 
solutions.  We should also remember that the relative _isolation_ of 
closure-violating effects in antenna-based calibration solutions is about 
half what we might be used to for the full VLA (12 vs 27 antennas). 
NOTHING among the calibration effects is _precisely_ orthogonal to
anything else!

-George


--