Generative Diffusion Models for Physics Simulation of High-Energy to Multi-Physics Systems: A Brief Survey

Generative diffusion models (GDMs) have emerged as powerful alternatives to traditional physics simulators, offering stable training, high fidelity, and scalability across scientific domains. While numerous applications have been developed in high-energy physics, fluid dynamics, and climate science, the literature remains fragmented. A survey is needed to systematically compare architectures, conditioning strategies, evaluation protocols, and reported results. This paper presents a brief survey of GDMs in scientific simulation. We analyse three representative models—CaloClouds, Jet Diffusion, and the Multi-Physics Diffusion Model (MPDM)—and extend the discussion to emerging domains such as astrophysics, quantum systems, materials science, and geophysics. Comparative tables summarize reported metrics and constraints, while a taxonomy of challenges highlights data scarcity, physical fidelity, generalization, and integration with existing workflows. Finally, we propose open research directions including hybrid physics–ML architectures, uncertaintyaware diffusion, domain adaptation, and real-time deployment.