高保真模拟和复杂逆问题(如探测器建模和解折叠)是亚原子物理中计算密集的瓶颈,但对于准确的物理解释至关重要。虽然条件流匹配(CFM)提供了一种稳健的加速方法,但我们证明其标准训练损失从根本上具有误导性。具体而言,利用杰斐逊实验室核物理(NP)运动学数据集($γp \ o ρ^0 p \ o π^+π^- p$),我们发现CFM损失过早进入平台期,掩盖了持续的物理改进。为了验证这种脱节是与数据集无关的病理现象,我们引入了ScatterPrism,这是一种高效的生成代理模型,在NP数据和模拟挑战性一维分布拓扑的合成压力测试上进行了评估。结合这些基准测试,我们确定物理信息指标在标准损失收敛后仍持续改进。因此,我们提出了一种多指标诊断协议,以确保在没有数据记忆的情况下实现真正的运动学保真度。受即将到来的电子-离子对撞机(EIC)相关NP挑战的驱动,这一统一机制有潜力扩展到高能物理(HEP)应用,如喷注建模。此外,该框架有望应用于需要严格生成可靠性的更广泛领域,包括医学成像、天体物理学和定量金融。
英文摘要:
High-fidelity simulations and complex inverse problems, such as detector modeling and unfolding, are computationally intensive bottlenecks across subatomic physics, yet essential for accurate physical interpretation. While Conditional Flow Matching (CFM) offers a robust acceleration approach, we demonstrate its standard training loss is fundamentally misleading. Specifically, utilizing a Jefferson Lab Nuclear Physics (NP) kinematic dataset ($γp \to ρ^0 p \to π^+π^- p$), we expose that CFM loss plateaus prematurely, obscuring ongoing physical refinement. To verify this disconnect is a dataset-agnostic pathology, we introduce ScatterPrism, an efficient generative surrogate evaluated against both the NP data and synthetic stress tests modeling challenging 1D distribution topologies. Coupling these benchmarks, we establish that physics-informed metrics continue improving long after standard loss converges. Consequently, we propose a multi-metric diagnostic protocol to ensure true kinematic fidelity without data memorization. Driven by NP challenges relevant to the forthcoming Electron-Ion Collider (EIC), this unified machinery has strong potential to extend to High-Energy Physics (HEP) applications, such as jet modeling. Furthermore, the framework holds promise for broader domains requiring rigorous generative reliability, including medical imaging, astrophysics, and quantitative finance.
Towards Overcoming Data Scarcity in Nuclear Energy: A Study on Critical Heat Flux with Physics-consistent Conditional Diffusion Model
Department of Nuclear Engineering, North Carolina State University Burlington Engineering Laboratories(核工程系,北卡罗来纳州立大学伯灵顿工程实验室);Laboratory for Reactor Physics(反应堆物理实验室);Thermal-Hydraulics, Center for Nuclear Engineering(热工实验室,核工程中心);Sciences, Paul Scherrer Institute(科学,保罗·施特格研究所)
M2PDE: Compositional Generative Multiphysics and Multi-component PDE Simulation
State Key Laboratory of Advanced Nuclear Energy Technology, Nuclear Power Institute of China, China(先进核能技术国家重点实验室,中国核工业研究院);Department of Artificial Intelligence, Westlake University, China(人工智能学院,西湖大学)
Nuclear Diffusion Models for Low-Rank Background Suppression in Videos
Dept. of Electrical Engineering, Eindhoven University of Technology, the Netherlands(埃因霍温理工大学电气工程系,荷兰);Philips Research North America, Cambridge MA, USA(飞利浦北美研究部,马萨诸塞州剑桥)
Quantum statistics from classical simulations via generative Gibbs sampling
Department of Chemistry, University of Chicago(芝加哥大学化学系);James Franck Institute, University of Chicago(芝加哥大学詹姆斯·弗兰克研究所);Courant Institute of Mathematical Sciences, New York University(纽约大学Courant数学科学研究所)
Discriminative versus Generative Approaches to Simulation-based Inference
Leiden Institute of Physics, Universiteit Leiden(莱顿大学莱顿物理研究所);Physics Division, Lawrence Berkeley National Laboratory(劳伦斯伯克利国家实验室物理部);National Energy Research Scientific Computing Center, Berkeley Lab(伯克利实验室国家能源研究科学计算中心);Berkeley Institute for Data Science, University of California, Berkeley(加州大学伯克利分校伯克利数据科学研究所)
Department of Chemistry, University of Chicago(芝加哥大学化学系);James Franck Institute, University of Chicago(芝加哥大学詹姆斯·弗兰克研究所);Courant Institute of Mathematical Sciences, New York University(纽约大学应用数学科学研究所)
A large language model-type architecture for high-dimensional molecular potential energy surfaces
Department of Chemistry, Department of Physics(化学系、物理系);the Indiana University Quantum Science(印第安纳大学量子科学);Engineering Center (IU-QSEC), Indiana University, 800 E. Kirkwood Ave, Bloomington, IN-47405(工程中心(IU-QSEC),印第安纳大学,800 E. Kirkwood Ave,布卢明顿,IN-47405)
Advances in machine-learning-based sampling motivated by lattice quantum chromodynamics
University of Wisconsin-Madison(威斯康星大学麦迪逊分校);University of Bern(伯尔尼大学);Google DeepMind(谷歌DeepMind);Massachusetts Institute of Technology(麻省理工学院);The NSF AI Institute for Artificial Intelligence and Fundamental Interactions(美国国家科学基金会人工智能与基础交互研究所)