to invert or not to invert UVW

[o-smirnov]

The help says:

  -iuvw, --invert-uvw   Optional. Invert UVW coordinates. Useful if we want
                        compare our visibilities against MeqTrees.

I wonder at the meaning of this. What I'm trying to do is to predict the contribution of a bright point source into a separate model column. With my latest changes in, I am able to do this. However, if I then make an image from the MODEL_DATA column populated by WSCLEAN by degridding its model image, and compare it to an image made from the visibilities produced by crystalball from the few components associated with the source, the position of the source is mirrored w.r.t. origin (which is exactly what flipping the sign of UVW would do).

So it seems to me that -iuvw is necessary for consistency with WSCLEAN in the first place. Why is it not the default then? @sjperkins?

[sjperkins]

@o-smirnov I wonder if your sign correction to the coordinate parsing flipped a sign (correctly).

I've been thinking of adding invert options to the phase_delay and dft kernels as it's a bit more efficient than flipping each uvw coordinate.

[o-smirnov]

But come to think of it, yeah, the old code would have given you the opposite sign in l and m... which perhaps explains why you needed to flip the sign of u and v to agree with MeqTrees (though not w?? Hmmm...)

At the end of the day it's just the sign of the exponent in the DFT kernel. Rather set that correctly and not worry about flipping coordinates.

[paoloserra]

A while back I found that -iuvw was necessary also with my function for reading of WSclean RA,Dec. That function had no RA,Dec sign error, so I don't think it's that.