Bayesian inference of deceleration-phase Rayleigh-Taylor growth rates in laser-driven cylindrical implosions

An iterative forward modeling approach has been developed and applied to the analysis of radiography data from cylindrical Rayleigh-Taylor instability studies performed at the Omega laser facility. Synthetic radiographs are generated and iterated using the Bayes’ Inference Engine [1] to produce maximum likelihood estimates for the time-dependent amplitude of deceleration-phase Rayleigh-Taylor modes seeded by a sinusoidal perturbation in the aluminum marker layer. This iterative forward modeling approach self-consistently fits the magnification and parallax in the image, both of which are sensitive at this scale (~13x magnification) to misalignments of the pinhole smaller than 200 μm. Systematic errors in the inference of growth factor as large as 10% have been identified and eliminated by this technique. Furthermore, the self-consistent modeling of image parallax reveals that these implosions are not elliptical as they may appear to the casual observer, but indeed highly symmetric. The regularizing priors adopted here, and further constraints that might be applied in future work, reduce experimental uncertainties and lend greater statistical significance to comparisons with hydrodynamic modeling.