Physics-Guided Diffusion Models for Calorimeter Simulation
A new arXiv preprint introduces Lantern, a method for physics-guided diffusion models in calorimeter simulation. Monte Carlo simulation is a bottleneck for the High-Luminosity LHC, and diffusion models offer fast surrogates but can minimize statistical objectives while violating physics. Existing physics-informed methods require closed-form laws or per-sample constraints unavailable for stochastic showers. The authors propose the Correlation Frobenius Distance (CFD) to measure correlation fidelity across layers and voxels, and encode soft per-sample structure. The paper is available at arXiv:2607.25060.
Key facts
- arXiv:2607.25060v1
- Announce Type: cross
- Monte Carlo simulation is a principal bottleneck for High-Luminosity LHC
- Diffusion models can minimize denoising objective while placing physics wrong
- Existing physics-informed methods assume closed-form law, PDE residual, or hard per-sample constraint
- No per-sample PDE governs stochastic cascade
- Energy conservation fixes only one scalar per shower
- Correlation Frobenius Distance (CFD) is introduced for correlation fidelity
Entities
Institutions
- arXiv