AI and the World's Hard Problems · Physics · arXiv · 2026-10-22
Agentic AI Detector Co-design and Optimization in Vertically Integrated Differentiable Full Simulations
Researchers report the first implementation of AI agents in the design and optimization of high-energy-physics detectors, using a bi-level optimization framework that vertically integrates detector geometry, front-end digitization, and high-level reconstruction-algorithm parameters within differentiable full simulations. Using a dual-readout segmented crystal electromagnetic calorimeter with a baseline resolution of 3%/√E as an example, they demonstrate an AI agent finding an optimal design point among three competing performance criteria. Frontier LLM reasoning models, even without experiment-specific context, executed complex workflows and proactively suggested further avenues, showing that agents can yield higher performance on key physics goals while reducing labor and compute.