Bayesian evidence adaptive pursuit to identify neutron sources with scatter-based spectrometers

Abstract

Reliable neutron source identification is essential for nuclear nonproliferation, safeguards, and homeland security, but remains challenging because neutron spectral inversion is often ill-conditioned, especially for mixed-source fields with overlapping spectral signatures. Here, we present a scalable Bayesian framework for neutron source identification from recoil spectroscopy measurements using evidence-based model selection. The method introduces a Bayesian Evidence Adaptive Pursuit (BEAP) algorithm that efficiently searches the combinatorial space of candidate source ensembles by iteratively ranking, retaining, and pruning source mixtures according to their Bayesian evidence. We validate the framework experimentally with controlled Cf-252 and deuterium--deuterium neutron-generator measurements, complemented by high-fidelity Monte Carlo simulations spanning representative fission, $(α,\text{n})$, and fusion sources with varying emission rates and mixture complexities. BEAP correctly identifies single- and multi-source mixtures with decisive statistical support ($>\!4σ$), requiring between $\mathcal{O}(10^1)$ and $\mathcal{O}(10^6)$ detected recoil events depending on source-mixture complexity, spectral similarity, and emission-rate imbalance. These findings establish BEAP as a practical, scalable, and robust tool for quantitative source identification in mixed neutron fields, significantly extending the operational capabilities of scatter-based neutron spectrometers in nuclear security and emergency response applications.