I am still working on this. I have created the preliminary solutions, but in the meantime we have realized the complexity of our Zernike wavefront implementation across the detector (https://pfspipe.ipmu.jp/jira/browse/PIPE2D-393?filter=-1). The current plan is rerun the solution with ``better'' implementation of the spatial dependence of the Zernike components. This might require opening new sub-tickets.

As already commented, this ticket should be split into smaller tasks if possible.

This has been done. There are solution for spots in focus for 14 out of 16 fibers. The distribution of the points can be seen in Q_40000.png. Short description of figures follows:

1. Example_subtraction_2D shows the final subtraction of one of the Neon spots, in 2d. The panels show the model, the data, the residual and the chi map (residual/sqrt(variance)).

2. Example_subraction_1d shows the 1d result. Orange shows the data, red shows the residual after subtraction. The number denoted the flux at each wavelength (again orange data, red residual). Red numbers also show the difference from zero in sigmas. The number is the right hand side are all some sort of quality measure. They are:

i. Q (my shortcut for ``quality’’) measures the mean absolute sigma of the residual in the grey region (in 4 pixels at each side, outside the inner 5 pixels dominated by the core). This is, at the moment, by working definition of the quality (testing for PIPE2D-353)
ii. Q_40000 is the same, BUT, with some added noise. The noise is added so that it is representative of what we would expect if the max flux in the image is 40000 counts, which is roughly what we expect in the real images (a lot of flux, but below saturation). In this example the result are very similar because the data in this actual example has max flux of around 45000, but the difference can be huge when the flux in the line is 100000, 200000 etc...
iii. chi^2 is just mean chi^2 of the full 20x20 images.
iv. chi^2 is just mean chi^2 of the full 20x20 image, with the added noise as described above.

3. Q_4000 Q.png is showing the quantity Q_40000 as defined above across the focal plane.

4. Comparison HgAr and Neon is showing the comparison between the Neon and HgAr data - the points have been taken from the grey rectangle within Q_4000 Q.png figure. We see that, for the same chi^2 values the Q values are much better for Neon then for HgAr. In other words, for the same quality of 2d subtraction, I get much better result for 1d subtraction for Neon data - I believe that this is probably due to poor continuum subtraction from the HgAr data.

********