USNCCM18 Plenary: Marta D'Elia, Atomic Machines & Stanford ICME

On the Use of Graph Networks in Scientific Applications

Marta D'Elia, Atomic Machines & Stanford ICME

In the context of scientific and industrial applications, we often have to deal with unstructured spatial and temporal data obtained from numerical simulations and the real world. The data is usually in the form of a mesh or a point cloud. In this context, graph neural networks (GNNs) have proved to be effective tools to reproduce the data behavior; however, depending on the physical nature of the datasets, variations of vanilla GNNs have to be considered to ensure accurate results. Furthermore, when only a point cloud is available, one is faced with the question of how (and whether) to build a corresponding graph.

In this presentation we go over general challenges in the use of GNNs in computational mechanics and fluid dynamics. Special attention will be given to particle-accelerator simulations; a computationally demanding class of problems for which rapid design and real-time control are challenging. We propose a machine learning-based surrogate model that leverages both graph and point networks to predict particle-accelerator behavior across different machine settings. Our model is trained on high-fidelity simulations of electron beam acceleration, capturing complex, nonlinear interactions among particles distributed across several initial state dimensions and machine parameters. Our results show the model’s to accurately track electron beams at downstream observation points, outperforming baseline graph convolutional networks. This framework accommodates key symmetries inherent in particle distributions, enhancing stability and interpretability. We also go over the extension of these architectures to autoregressive tracking across multiple timesteps. This research offers a powerful approach to reducing computational demands in particle-accelerator simulations, contributing to advancements in real-time optimization and control. This work had been performed at Stanford in collaboration with SLAC.