Low-energy reconstruction and AI methods

Liquid argon detectors record far more information than standard event reconstructions can take advantage of, but this information can be used to lower the energy threshold of detectors, improve particle identification, and generally help us learn more from each interaction. Accomplishing this requires accurate detailed modeling of the interactions and detector response, in addition to sophisticated machine learning reconstruction methods. We have built and continue to develop the machine learning and differentiable programming methods that produce these advances. Recent results include the first event-by-event identification of Cherenkov light from sub-MeV particles in liquid argon Phys. Rev. Lett. 135, 171804 (2025) and a liquid argon calibration measurement using a fully differentiable simulation Phys. Rev. D 112, 072010 (2025). The payoff is MeV-scale supernova, solar, and dark matter physics in detectors built for GeV neutrinos.