Speaker
Description
Machine learning is rapidly transforming computational science by replacing expensive first-principles calculations with accurate, scalable surrogate models. A major challenge, however, is incorporating the fundamental physical symmetries that govern scientific simulations. We present a gauge-equivariant graph neural network (GNN) that enables scalable machine learning for lattice gauge theories by embedding exact local non-Abelian gauge symmetry directly into the message-passing architecture. Rather than relying on handcrafted gauge-invariant descriptors, the network propagates matrix-valued gauge-covariant features, allowing nonlocal Wilson-line correlations and many-body interactions to emerge naturally through local symmetry-preserving operations. We benchmark the framework on three representative scientific workloads: prediction of observables in 3+1D SU(3) lattice gauge theory, structure-property learning in SU(2) and SU(3) gauge-matter systems, and force-field learning for semiclassical quantum-link dynamics. Across these problems, the model accurately predicts gauge-invariant observables, captures fermion-mediated long-range correlations, and replaces repeated exact diagonalization with a learned gauge-equivariant surrogate for dynamical force calculations. By combining exact physical symmetry with scalable graph neural networks, this work establishes a general scientific machine learning framework for accelerating lattice gauge simulations and illustrates how structure-preserving AI can bridge first-principles theory and large-scale computational discovery across high-energy physics, quantum many-body systems, and quantum simulation.
| Do you plan to submit a 4-page extended abstract on OpenReview (only for Presentations/Posters)? | Yes |
|---|