Eric Darve (Stanford) - Graph Neural Networks for particle accelerators and fluid dynamics
41/2-2162 - Sonoma
SLAC
Authors: Eric Darve; Tiffany Fan; Auralee Edelen; Marta D’Elia; Murray Cutforth; Eric Cropp
Graph Neural Networks (GNNs) offer powerful tools for surrogate modeling of complex physical systems by naturally representing irregular domains and capturing both local and global interactions. We present recent applications of GNNs to fluid dynamics, cardiovascular simulations, and particle accelerators. By learning from high-fidelity simulations, GNN-based reduced-order models provide accurate and computationally efficient approximations that enable rapid design, optimization, and real-time analysis. Examples include hydrodynamic simulations using particle-based methods, reduced-order cardiovascular models, and accelerator beam dynamics. These results demonstrate the versatility and scalability of GNNs for multi-physics and multi-scale problems, highlighting their potential to accelerate discovery and control in computational science and engineering.
Zoom: https://stanford.zoom.us/j/93244635877?pwd=q6mJIZaSBzfX8Jbae3z50zk4c0gI7L.1