Exploring the performance of GPU-based optical photon simulations in Geant4

For future high energy physics (HEP) detectors, dual readout calorimeters (DRC) provide optimal energy resolution by allowing independent measurements of energy deposited via electromagnetic and hadronic interactions. Precise simulations of optical photons from scintillation and Cherenkov processes are essential for the design and calibration of DRCs. Because the number of optical photons generated in DRC simulations is immense and their simulation is impractical on single-threaded CPUs, HEP detector simulations have historically relied on parameterized rather than real-time simulations of optical photons. However, recent advances in Graphical Processing Units (GPUs) and CPU multithreading allow efficient simulation of optical photons in real-time, thus significantly improving the accuracy of DRC simulations. In this project, we explore the performance of the HEP-standard Geant4 simulation software using multithreading and GPUs. The GPU-based simulation relies on the CaTS (Calorimetry and Tracker Simulation) framework, which uses Opticks to translate Geant4 physics to NVIDIA® Optix to run on the GPU and the Geant4 geometry into ray-tracing optimized geometry for GPUs. We observe a speedup of upwards to one hundred times in performance in liquid argon TPC experiments.