Project and Generate: Divergence-Free Neural Operators for Incompressible Flows — Quantapedia

Learning-based models for fluid dynamics often operate in unconstrained function spaces, leading to physically inadmissible, unstable simulations. While penalty-based methods offer soft regularization