A Bayesian approach to inferring neutron spectra from projectile fusion

First Light Fusion have confirmed the first example of projectile-driven ICF, reporting a yield of approximately 50 neutrons at the expected 2.45 MeV energy from DD fusion. To support experimental progress from low to high yields, we developed a fully Bayesian method to infer physical properties of the source, including yield and plasma ion temperature.

We modelled the expected neutron arrival times at each detector, from which Poisson likelihood functions were constructed. We accounted for single-neutron coincidence between detector pairs by using a Gaussian copula, with correlation coefficients determined from neutronics simulations. From the joint likelihood function we built a posterior density function for the source parameters, conditioned on the experimental data.

We verified this method using data from the fusion validation campaign (Burdiak et al. 2022), inferring a yield Y = 51⁺³⁹_-21 (95% EQI) and a scattered fraction X = 83⁺¹⁶_-25 %. Current yields and detector resolution are not yet sufficient to measure the ion temperature at significance. However, if a suitably large detector array can be designed, we predict that a factor of ten improvement in timing resolution should enable recovery of temperatures at a few keV for yields of a few thousand neutrons using this method.