I coded a simple version of the angular misalignment averaging method we discussed yesterday - without optimization on my slower computer, it adds ~5 seconds to the run time. So it's not nearly as fast as I hoped it would be and too slow to be incorporated yet. I'm sure there are ways that it can be optimized, though.

I have been working on the issue of angular misalignment from before, modeling angular misalignment as a function of 2D displacement (incorporating vignetting + nontelecentricity effects) and comparing profiles with expected lab drop-off. Given that the simplest model of the focal plane on the instrument could be well-replicated by a simple modification to the illumination model, I tested for qualitative agreement between the modified model and Dan's Zemax model (example 1D plots from his work below). My work is able to reproduce the shapes of these profiles although more work can be put into the model if it is found to be useful. 

The overall 2D model process is indicated in the "2D Production of Angular Misalignment" image.

More interesting, I have found that my simple model is reproducible from a 1D integration. An example of the comparison of the two models is in the "1D Analytical Profile Comparison" image.  The 1D code runs in about 1/3 of the time it takes for the 2D analysis. It seems to me that the 1D method is a promising way to incorporate more complicated illumination profiles without taking too much time for pupil generation, but this will depend on my implementation in the 2D-DRP.

Finally, I investigated the variation between a profile generated with the normal error function and one convolved with a Gaussian to determine if not convolving has any noticeable effect on the profile. After accounting for the sigma^2/radius shift Jim Gunn found in his analysis in his PFS-SPS-PRU000210-01_slit2frd file on the SuMIRe workspace, errors are on the order of 0.1%, but without correcting for this in the profile, errors of order 1% can occur (with a clear ring residual due to the misalignment). In practice, if the effective radius of the 2D-DRP is fit to the size of the observed profile, this would effect an angular scale measurement on the order of (FRD/radius)^2 << 1, however, so the current error function as implemented in the 2D-DRP is excellent in implementation.

As discussed with Neven on a call on Thursday, I'm currently trying to replicate that analysis for a Lorentzian contribution and incorporating it into my 1D code; the Lorentzian component is important for Neven to match the large-angle intensities, but the Lorentzian's long tails make the convolution effect between 1D and 2D harder to compare directly. Afterward I will be working to implement this analysis into the 2D-DRP directly to compare the time savings and compare angularly misaligned data to the 2D-DRP images.

I apologize for the extra emails that were sent out to everyone following this issue due to my method of uploading files.

It has been decided to focus on the combination of FRD and angular misalignment (sig_tot effectively) first and consider angular misalignment if the sig_tot implementation is not precise enough for our needs. Work into extracting angular misalignment effectively continues in PIPE2D-623.