BRICKS: Compositional Neural Markov Kernels for Zero-Shot Radiation-Matter Simulation

2026-05-07

AI Summary

This work addresses the high computational cost and poor zero-shot generalization of conventional radiation–matter interaction simulations by proposing a composable neural surrogate kernel grounded in particle locality and Markovian dynamics. The approach integrates a hybrid discrete–continuous Transformer architecture with Riemannian flow matching and product manifold modeling, introducing Riemannian flow matching to radiation simulation for the first time. This yields a differentiable neural Markov kernel capable of likelihood estimation and zero-shot generation of large-scale, unseen material configurations. Experiments demonstrate that a single-GPU implementation significantly outperforms traditional CPU-based simulators in speed, while multi-step autoregressive rollouts remain stable and reliable. The study also releases a new dataset comprising 20 million events.

Abstract

We introduce a new strategy for compositional neural surrogates for radiation-matter interactions, a key task spanning domains from particle physics through nuclear and space engineering to medical physics. Exploiting the locality and the Markov nature of particle interactions, we create a \emph{next-particle prediction} kernel using hybrid discrete-continuous transformer models based on Riemannian Flow Matching on product manifolds. The model generates variable-sized typed sets of particles and radiation side effects that are the result of the interaction of an incident particle with a material volume. The resulting kernel can be composed to simulate unseen large-scale material distributions in a zero-shot manner. Unlike mechanistic simulators, our model is designed to be differentiable, provides tractable likelihoods for future downstream applications. A significant computational speed-up on GPU compared to CPU-bound mechanistic simulation is observed for single-kernel execution. We evaluate the model at the kernel level and demonstrate predictive stability over multi-round autoregressive rollouts. We additionally release a novel 20M-event radiation-matter interaction dataset for further research.

Problem

• radiation-matter interaction
• zero-shot simulation
• compositional modeling
• neural surrogate
• particle physics

Innovation

• compositional neural surrogates
• Riemannian Flow Matching
• zero-shot simulation
• hybrid discrete-continuous transformer
• differentiable radiation-matter modeling