I will do this and also reanalyze images from the focus after apodization changes done which are consequence of the work in PIPE2D-463

Ok, what did I learn....

First of all, there are obvious biases in estimation of the parameters from images at different level of defocus. For example, see the images ``CCD_diffusion'', where you can see that amount of defocus is clearly correlated with the inferred CCD diffusion parameter (given that these are all the images taken at the same spot, one after another CCD diffusion should be the same). Another example is ``CCD_diffusion_fiber_r''',  - I plot results from 11 chains from 11 different images. We can see that they do not overlap (even though all of these parameters should be same for all of these images).

One fix that I discovered, in conversations with Peter (Melchior) is that my likelihood  function was slightly wrong (logarithm was in wrong place). I have fixed that but this problem has remained.

My hypothesis is the likelihood surface is multimodal and my minimizer aggressively finds the local minimum and gets stuck there. My current ``solution'' is to find the median of all 11 chains and set that as best value. My hope is that median of ``many'' best values is indeed the best value (where 11 == ``many'').

I am running a large re-analysis of the defocused and focused data with such determined values in order to determine if this is true.

I will open new ticket to summarize the results of this re-analysis.