An Active-Learning Prioritization Engine for Nuclear Data Relevant to X-Ray Bursts
The compressibility of nuclear matter — the fundamental property that helps determine the structure of neutron stars — remains an important unanswered question in nuclear astrophysics. Type-I X-ray bursts, recurring thermonuclear explosions on neutron-star surfaces, provide an observational window into this question. However, accurately modeling these bursts requires extensive nuclear data, much of which remains uncertain. Because accelerator-based experiments at national user facilities have limited beam time, researchers need a rigorous way to determine which measurements are most likely to produce the greatest scientific benefit.
An Argonne-led team is developing an artificial intelligence (AI) forecasting workflow to address this challenge. The system will combine a multi-tier knowledge graph of nuclear reaction data, uncertainties, simulations and experimental constraints with a high-speed surrogate model capable of reproducing key predictions from detailed X-ray-burst calculations. An active-learning prioritization engine will then evaluate prospective nuclear measurements and rank them according to their expected value of information — that is, how effectively each measurement could reduce uncertainty in X-ray-burst observations and inferred neutron-star properties.
The team will test the approach by reconstructing the nuclear data available decades ago and asking which measurements an AI-guided strategy would have prioritized. The results will be compared with the experiments that were actually conducted, providing a quantitative assessment of whether the system can reach modern scientific precision with fewer measurements and more efficient use of accelerator time. The Phase I effort is designed as a transferable foundation for prioritizing experiments across other nuclear reactions, astrophysical environments and, ultimately, the broader national experimental physics portfolio.