Probing One Neutron from Pb Using the LHC

In ultra-peripheral PbPb collisions at the LHC, electromagnetic excitation via the Giant Dipole Resonance often produces single forward neutrons detected in the ATLAS ZDC, whose transverse momentum is sensitive to the excitation dynamics and nucleon momentum distributions inside the nucleus. For Run 3, ATLAS installed a transversely segmented Reaction Plane Detector with a mixed-depth, pan-flute fiber geometry, where traditional subtraction-based reconstruction becomes unstable in low-light single-neutron events. We present a two-stage neural-network reconstruction trained on Monte Carlo simulations that infers effective photon yields from raw RPD signals and then reconstructs the neutron transverse momentum, achieving significantly improved resolution and extending the sensitivity of forward-neutron measurements in ATLAS.