C 4-Equivariant Flow Matching on Anisotropic Power-Diagram Graphs for Microstructure Generation
The model generates realistic polycrystalline microstructures using a continuous normalising flow trained via C4-equivariant flow matching and graph neural networks. It represents grains as anisotropic power diagrams and uses training-free guidance to control microstructure features like layering or welds by optimizing user-defined objectives.