Enabling Rare Event Discovery with Surrogate Modeling and Simulation AI Agents
Computationally demanding Monte Carlo simulations of optical photons are a major bottleneck for optimizing the design of the Large Enriched Germanium Experiment for Neutrinoless ββ-Decay (LEGEND) experiment for sensitivity to neutrinoless double-beta decay.
Enabling Rare Event Discovery with Surrogate Modeling and Simulation AI Agents, led by the University of North Carolina, aims to accelerate this process by developing a Rare Event Surrogate Model (RESuM) with simulations accelerated by GPUs using the ORNL-developed Celeritas project, and orchestrated by an agentic AI system. This project will accelerate the generation of liquid argon optical response maps used by LEGEND-1000 a hundredfold and enable optimizations of the design of LEGEND’s liquid argon detector system. The techniques developed by this project will have transformative potential for design optimization and improving precision of high energy and nuclear physics experiments.
