Bayesian profile reconstruction from multi-monochromatic imager data

The multi-monochromatic imager (MMI) is a diagnostic instrument that provides spectrally resolved images of inertial confinement fusion experiments centered on emission/absorption from the characteristic lines of mid-Z dopants. Previous methods of analysis required some form of binning across either spectral or spatial axes to create composite images or spectra, respectively. This work takes an alternate approach by generating a synthetic MMI image in an autodifferentiable manner, allowing the use of Hamiltonian Monte Carlo for sampling from the complicated posterior distribution of the parameters describing the plasma. The algorithm is tested with artificial data, and the error bars and reconstructions are shown to match well with the ground truth, with 77% of the fit parameters having overlap between the ground truth and the 90% credible interval of the reconstruction. If the fit discrepancies due to dynamic range imposed on the artificial image are accounted for, the agreement increases to 86% of the fit parameters having overlap between the ground truth and the 90% credible interval of the reconstruction, despite the reconstruction assuming spherical symmetry not present in the test image.