Speaker
Description
Optical photon tracking in Geant4 is a major computational bottleneck for simulating Liquid Argon (LAr) detectors, where scintillation light is central to event reconstruction and triggering. The cost of traditional CPU-based optical simulation severely limits the production of high-statistics samples needed for training deep learning models and developing AI/ML-based reconstruction techniques. We present PhoXim, a GPU-accelerated optical photon simulation framework developed at BNL/NPPS, and its ongoing adaptation to LAr neutrino detectors including the Deep Underground Neutrino Experiment (DUNE). PhoXim is based on the Opticks framework developed for the JUNO experiment and it leverages NVIDIA OptiX ray tracing to offload optical photon propagation to the GPU while running within a standard Geant4 simulation, delivering orders-of-magnitude speedup over conventional tracking. Originally developed for Electron-Ion Collider (EIC) detector simulations, the framework is being extended to support the optical physics processes relevant to LAr, including wavelength-shifting and packaged as a framework-independent module for production use. We report on the status of physics validation in LAr geometries, performance benchmarks. By removing the optical simulation bottleneck, PhoXim aims to be a shared tool that enables the neutrino community to fully exploit AI/ML methods, from detector design optimization to data-driven reconstruction that demand large-scale, high-fidelity simulated datasets.