Anisotropy calculation for single polarization may be skewed due to sampling from beamstop region

[KSRandazzo]

The built-in anisotropy calculation for a single polarization can return skewed values because it samples from a region that may overlap with the beamstop.

Specifically, when calling:

rsoxs.AR(calc2d=False, two_AR=False, chi_width=45, calc2d_norm_energy=None)

the function appears to sample from wedges centered at 0° and -90°. In the data taken in April 2025, the beamstop is located around -50°, so it's very easy for the wedges to include part or all of the beamstop, which can distort the resulting anisotropy calculation.

While I could be mistaken about the exact wedge centers (e.g., if the image is flipped or rotated internally) some basic testing confirmed that the angles near chi=-50° in the unwrapped data do seem to correspond to the beamstop.

Using two polarizations is generally more robust, but single-polarization scans are still sometimes used at the beamline, so it may be more helpful to make this calculation on the basis of the wedges centered at, for example, 90° and 180°.

[pbeaucage]

Thanks for another good issue!

I think there are really two sub-issues here:

1. The anisotropy calculation should recruit not only 0 / 90, but also their mirrors, i.e, 180 and -90/270.

2. The shape of a beamstop in "cake-plot" space is not even in chi, so some values of chi will have no data at low-q and more data at high-q, which skews the overall calculation:

2a) More generally, the beam stop occludes data in only one "wedge" or "half" of the image, which makes the integral of intensity in a certain wedge fundamentally biased unless you mask all equivalent areas, i.e, mask all pixels in (mask, mask + 90, mask+180, mask+270).

(1) is pretty straightforward to fix, and may have already been done in the stalled "new AR" PR #147 which @delongchamp has promised us a review on.

(2) is tricky. Stamping the mask around every 90 degrees is one approach, but I also think we need to mask all values of chi with significant missing/NaN data for a given slice of q. For instance, in the above dataset, if we consider q=0.02 - 0.08 A^-1, we should be excluding the green values of chi and only computing using the yellow-boxed values of chi. (forgive the crude mspaint)