Unifying Calibration and Reconstruction for Optical Particle Detectors using Differentiable Simulations

Monolithic detectors with optical readouts are widely used for particle detection, with Cherenkov and scintillator neutrino detectors as prominent examples of this technology. Experimental physics analyses depend on high-fidelity simulators for these detectors that are used to translate sensor-level information into physics variables of interest. This task depends intimately both on calibration, used to align the simulator behavior with the real detector, and tracking, used to infer particle properties from the detector-level information. The talk presents a differentiable simulator of an optical particle detector capable of optimising calibration parameters as well as track parameters via gradient-based optimization. I will discuss how to model light emission with surrogate light-emission models and utilize differentiable ray-tracing including discrete random operations while preserving meaningful gradient information. The talk describes the simulation, calibration and tracking procedures, as well as its applicability to experiments relying in photon ray-tracing for reconstruction. The performance trade-offs of our methodology compared to existing approaches will also be discussed, as well as possible continuations to this work.

Zoom: https://dipc-org.zoom.us/j/98949496221

Youtube: https://youtube.com/live/BXiZE-xvYHk