In Figure 1 I am showing what are the definitions of the regions that I will be using in the rest of the text below. 

Figure 2. shows the expected flux as a function of radius. In the full lines I am showing results from data. Spot 55 is my internal name for the spot (5th out of 10 lines I analyze, in the 5th fiber) which is in the middle of the detector. In dashed lines I am showing result from the Zemax simulations for the same spot (more details in the report mentioned in the description of the ticket). We see (as claimed in the report) there is more flux in the wings of the data than predicted by optical model.

Figure 3. shows result after adding grating, with effective number of lines=40 000, following Equation 16 from http://adsabs.harvard.edu/abs/2008MNRAS.386...47E. We see that this effect is able to described observed flux in the data in the vertical direction (and only in vertical direction).

Figure 4. I then investigated if I there is wavelength dependence of the excess flux(as diffraction and scattered light should have different wavelength dependence). Figure 4 shows results from considering spots from a single fiber but different wavelengths. Note that is averaged of all directions. No clear trend is seen.

Figure 5. shows how much is the vertical component stronger than what I would expect it to be from Zemax modeling in the center of the detector, as a function of the position on the chip. Note the log scale. Obviously there are certain spots which are behaving very differently than others and it seems that same data as same wavelength is exhibiting the same behaviour across fibers. 

Figure 6. I (and Robert) were worried that different spots show very different vertical component is because I perhaps did a wrong subtraction of the continuum from the lamps. Figure 6. shows comparison of the two spots in the same fiber ('35' is denoted by red in Figure 5 and '55' is blue in 'Figure 5'). One can see that there is no error in the subtractions (i.e., the continuum seems to be well subtracted in both of the cases far away from the spot); yet, we can see obvious difference in the strength of the vertical component, as claimed in Figure 5. I tested one spot in Neon data (there are no more obvious lines available) and it is very symmetric, i.e., looks more like spot '55'/

Figure 5: can you remind me what the 0-1 scale for z (color) corresponds to again? I know it's a vertical component strength indicator, but would like to know a little more.

Also for my info: for figure 5 which is the wavelength direction?

It is by how much I have to multiply vertical component of the zemax to get the data (in dex)! So red value of 1 = need to multiply Zemax value by 10; blue value of 0.5 means I have to multiply Zemax model by 10**0.5 = 3.3 to get the data. That is the point of Figure 6, which shows this explicitly so that you do not have to just trust my code - you can actually see the difference in the vertical component of two spots.

Wavelength direction is the vertical direction. Blue part of detector is up, red is down.

All clear and what you had explained yesterday to Craig (I just missed the start of that discussion) thanks.

I have performed some additional tests and added few more Figures.

7. The same figure as Figure 5., but now with more spots across the detector. The results are still very similar.

8. Same as Figure 7., but now done with Neon lamp. The advantage of using this lamp is that this lamp does not contamination from continuum, but the flux in the lines is much smaller. This is especially true for the red part of the detector (longer wavelengths, upper half of the Figure), so I am not completely convinced by the results for the 3 longest wavelengths.

9. Figure showing spots in the same Fiber. The differences in the shapes of the PSF is clearly visible. This is extension of Figure 6.

10. Same for Neon. I have stacked images from all the 10 fibers the same wavelength and from 4 dithering positions together create these images. You can see that for the 3 largest wavelengths the noise is dominating and the signal is very weak, as mentioned above. Otherwise all others spots seems to show similar structure, as seen in Figure 8.

11. Condensed 1d results. Orange lines shows results for a HgAr (single fiber in the center of the detector), while blue for Neon (images from all fibers and dithering positions stacked together). Title above each plot shows to which component is the particular figure referring (horizontal, vertical, diagonal). Total number of counts in the images shows you that that the flux in the neon stamps, especially in the red part is much lower - note each neon stack is composed out of 40x more images than the HgAr image (10 fibers and 4 dither positions), so the corresponding noise is also larger.

Kjetil Dohlen (at LAM) was not impressed with my explanation that this effect is due to grating. His main argument is that all wavelengths go through the grating. I have to clarify this with JEG.

I have started looking at this problem, with the new data taken in 2019.

I am still extremely confused. I have added two plots:

12. summary_new - I tried to create a similar plot as plot ``Summary (figure 11)'', showing the most important vertical/horizontal ratio. I used the images taken with Neon lamp in February. We took some special exposure for this experiment. One special set was 30 exposures for Neon, which we stack together to increase signal to noise. Other set consists of 3 overexposed images. The problem is that there the weak Neon continuum starts to be significant on these levels, but I cleaned the images using the same algorithm as in PIPE2D-401.
As we can see the results from the new neon data is a all over the place, but I think that the main result (especially from HgAr) is the same as before. We see that at some wavelength the vertical component (around 830 nm) is much more pronounced - even though we do not see the same effect in Neon data!!!

13. Direct comparison of the far wings from HgAr and Neon data at very similar wavelengths - observe the large difference! I have verified that this is not due to continuum subtraction (I have to add images showing this as well).

Resolved in PIPE2D-439