Optimizing Low-Energy Nuclear Reaction Measurements using Machine Learning

Low-energy (p,n) reactions on unstable nuclei play an important role in astrophysical nucleosynthesis, particularly in processes such as explosive silicon burning and the vp-process in core-collapse supernovae. These reactions directly influence the production of proton-rich isotopes, but their rates remain uncertain due to the difficulty of measuring them experimentally, especially for short-lived nuclei.

To access these reactions, measurements may be performed in inverse kinematics using radioactive ion beams, such as those available at the Facility for Rare Isotope Beams (FRIB). However, (p,n) reactions in inverse kinematics present a unique challenge; the reaction products and the unreacted beam have nearly identical masses, rendering their separation particularly challenging.

In this talk, I will present how machine learning aided a new experimental approach for the measurement of such reactions with a recoil separator. Specifically, I will describe a framework that combines multi-objective evolutionary algorithms with ion-optical simulations to optimize the recoil separator configuration for (p,n) measurements. This method has been successfully validated using a stable-beam ⁵⁸Fe(p,n)⁵⁸Co experiment, demonstrating its viability for future measurements with radioactive beams.

I will elaborate on the machine learning methodology, the experimental validation, and its significance for future (p,n) studies at FRIB. This work highlights how data-driven techniques are expanding experimental capabilities in nuclear physics and helping to address long-standing challenges in reaction measurements relevant to astrophysics.