Symmetry in the Wild: The Role of Equivariance in Neural Fluid Surrogates

In three linesGroup-equivariant architectures improve neural CFD surrogates when training data lacks strong regularities, but degrade performance on strongly aligned datasets. AB-GATr, an E(3)-equivariant geometric algebra transformer, outperforms data augmentation on automotive aerodynamics and hemodynamics benchmarks.

Summary generated by Claude — human-verified