I hope that I am close to creating the first version of the solution. I am adding Figure ``Overview.png'' which shows that is the minimal value of predicted LSF across the detector (as a function of fiber and y-value on the detector). As you can see there is not a single position where the LSF is negative. At the moment I am studying the strange cluster of the high value at y=650. When I solve this, I think this can be the first version of new solutions. 

I am almost certain that the problem is that I have been extra stupid when doing extrapolation...

I have added  ``Overview_interpolation.png'' which shows that is the minimal value of predicted LSF across the detector (as a function of fiber and y-value on the detector). This new figure is created with the new interpolation schema that I am proposing.

Beforehand the algorithm was searching for two nearest points to either interpolate or extrapolate the solution. 

The new schema searching for two spots which straddle the position that you are asking for, i.e., it always does the interpolation. If you are asking for a position that is outside the last available spot (at the edges of the wavelength coverage) the algorithm at the moment returns the last available spot, without any extrapolation. Therefore the solutions at the edges will be deficient in this type of algorithm. Covere_space.png shows the part of the space where we interpolate (dark blue) and where we do not interpolate (yellow).

Solutions are placed in `/tigress/ncaplar/PIPE2D-521`. The description is the same as in PIPE2D-450. 

positions_of_simulation_00 array contains positions of each PSF model. The columns are: 1. fiberId, 2. approximate x coordinate, 3. approximate y coordinate, 4. wavelength.

array_of_simulation_00 array contains arrays with 189x189 PSF models (9x times oversampled).

At the moment there are *_Apr15_v1 and *_Apr_v2 versions.

*_Apr15_v1 version is created with the lines as they appear on the detector, i.e., the with broad HgAr lines identified in PIPE2D-439.

*_Apr15_v2 version is created by artificially narrowing these lines, i.e, how I believe they should appear if they were not broadened due to atomic issues. This narrowing is done quite crudely at the moment, I just set the argument controlling the Lorentzian width component so that is roughly similar to the other nearby lines. This needs to be improved. 

These PSFs are valid for F/2.8 data taken in July 2019 (e.g.,  runs 21400, 21604 and 21808.)

The updated method is available at https://github.com/Subaru-PFS/dev_2ddrp/blob/master/2d_PSF_code/PIPE2D-521/Provide_PSF_2D.py

I have added *_Apr15_v3 solutions. The v3 version includes the artificial narrowing of the Hg lines, identified in PIPE2D-564. 

price identified that the new implementation did not handle the case when the requested spot is at the exact position of the input spot correctly. This has been rectified and pushed to https://github.com/Subaru-PFS/dev_2ddrp/blob/master/2d_PSF_code/PIPE2D-521/Provide_PSF_2D.py.