Enhancing the simulation accuracy of the COSI anticoincidence system: calibration, benchmarking, and detector effects prediction

COSI is a NASA Small Explorer mission scheduled for launch in 2027. Operating as a Compton telescope in the 0.2–5 MeV range, COSI will provide all-sky observations using germanium detectors (GeDs). The GeDs are surrounded by 22 BGO scintillation panels, readout by SiPMs, forming an Anticoincidence System (ACS). The ACS shields the GeDs from background radiation and can act as a stand-alone detector for gamma-ray transients such as GRBs. The ACS performance is closely linked to scintillation processes in the BGO crystals and to the SiPM readout electronics. The average path length of optical photons, their detection probability, statistical fluctuations, and electronic noise determine the ACS energy resolution, while spatial variations in detection probability can induce a non-uniform energy threshold. These effects are difficult to fully capture with laboratory measurements alone. Although Geant4 includes optical physics, simulating scintillation in every COSI run is computationally prohibitive. We therefore performed dedicated Geant4 optical simulations, benchmarked against calibration data, to encode scintillation effects into a position-dependent correction matrix. This approach enables simulations that closely reproduce real measurements while preserving computational efficiency and allows us to explore fine spatial scales and energy ranges not easily accessible experimentally. We enhanced the Geant4-based BoGEMMS-HPC framework to run efficiently on High-Performance Computing environments, achieving a speed-up of 70 using 112 cores on the Leonardo Tier-0 supercomputer. After validating optical processes against theory and laboratory measurements, we benchmarked simulations against ACS calibration data, reproducing the measured energy resolution (16% at 511 keV). Position-dependent corrections on the measured energy based on the optical processes were derived for each BGO panel and tested on GRB simulations based on Fermi-GBM catalogs. The resulting correction matrix, now integrated into the COSI pipeline, ensures accurate modeling of the ACS response and strengthens its contribution to GRB detection and localization.