Optical Data Reconstruction at the DUNE Near Detector 2x2 Demonstrator Using a Differentiable Surrogate

DUNE is a next-generation long Baseline Neutrino oscillation experiment in the US. The 2x2 demonstrator is located at Fermilab, and is a prototype liquid argon time projection chamber (LArTPC) for the DUNE near detector. Optical signal propagation in this detector is modeled with a look-up table containing the probability of observing a photon at each optical detector given the discretized source location in the active volume. The table is constructed by simulating O(10^4) photons in the detector volume, which is voxelized into 5cm cubes. Recently, a sinusoidal representation network (SIREN) has been successfully implemented as an alternative model for the physics of photon transport in LArTPCs. This new method addressed the challenge of scalability for the large future DUNE LArTPC and enabled model optimization from real data. Due to SIREN's differentiable nature, it can also be useful for data reconstruction tasks. The two key tasks we have applied it towards are reconstructing the depth of particle tracks within the detector volume from their detections on an anode plane, and associating particle tracks with their corresponding optical signals. In this talk, I will demonstrate the performance of SIREN on these data reconstruction techniques for the 2x2 demonstrator.