Diffusion Models for Fast Simulation of Cherenkov Detectors

We investigate the application of Diffusion models to achieve fast simulation of Cherenkov detectors, with particular emphasis on optical photon transport in Detection of Internally Reflected Cherenkov (DIRC) light systems. Conventional frameworks like Geant4 incur substantial computational cost when tracing photons through intricate geometries and reflective interfaces, creating a barrier for large-scale studies. Here, we compare Diffusion-based generative approaches against models used in other works, namely Normalizing Flows, to evaluate their fidelity and efficiency in simulation. By training Diffusion methods on photon‐hit distributions and comparing results over the full phase space, we demonstrate some of the limitations in using recursive, non-likelihood based generative methods in low-dimensional, highly conditional physics domains.