At the moment, I have been successful in running galsim and creating some mock images. I have also made a mock-up version of the pupil (see Pupil00.png). Jim G. should provide me at some point with illumination files which will give correct proportions of struts etc... At the moment I am struggling to understand if I am properly modeling the system, i.e., to set the all the parameters of the simulation (pupil size, pixel scale....) properly, so that the simulated images look somewhat like the LAM images.

I have fully analyzed one of the test images taken at LAM (v3617).

Below we can see one image which shows poststamp images of the defocused spots across the detector. The full black squares are the positions that were either too faint, or for which the simulation on the cluster did not finish (because I put too strict time limits to run my job). The numbers above are x and y positions on the chip. I am showing both the linear and log scaling.

Then I model the post stamps. I have been modeling those with Galsim, with 16 parameters (z4 to z11 Zernike coefficients, size of the central obscuration, size of struts, x and y position of the central obscuration in the focal plane, x and y position on the chi, “size of the telescope”, size of the input fiber image). Results are quite good, capturing the main features of the actual data. There is only one obvious failure (at 3620,2347), and even data in corner of the detector is reproduced. Below I am showing results, with linear and log scaling.

When we look at the residuals in linear scale, some of the images are almost complete erased. Log scale reveals that some structure is still left.

One can also look at the “predicted” images of the pupils and their movement across the detector.

Below I am showing the estimates for the behavior of different Zernike coefficients across the detector (from z5 to z11). Size of the spots is the same for all of the images, the information is only in the color.

Future improvements, roughly in the order of priority that I would suggest:
1. Analysis on better images with larger spots and stronger signal, also use images with different defocus to compare results
2. Check if the images are uniformly illuminated by comparing data taken from both sides of focus
3. Eliminate the pupil parameters from the fit and use the data which will be provided from France about the movement of the central obscuration and struts across the detector
4. I am using “size of the telescope” as a free parameter - obviously this is wrong and I am using it hide some of my lack of knowledge about how the spectrographs work and abusing the code which was done for other purpose. This needs to be better understood (and other parameters in galsim which I am not touching at the moment, e.g., wavelength of the light).
5. Do a bit more realistic modeling of input fiber (currently it is top hat, realistically is that some wings)
6. Analyzing with the images with corrected gains (currently variance in the scientific images is different than what would be inferred from variance image)

This is an old overarching ticket summarizing the development of the PSF pipelne from defocused images. The work has been split to many other tickets.