Structure-Preserving Learning Improves Geometry Generalization in Neural PDEs

Researchers introduce Geo-NeW, a neural network method that solves Partial Differential Equations while preserving physical laws and generalizing to unseen geometries. The approach combines learned differential operators with finite element spaces that explicitly encode geometry information, achieving state-of-the-art performance on PDE benchmarks with significant improvements on out-of-distribution test cases.

This research addresses a critical limitation in current physics-informed neural networks: their struggle to generalize beyond training geometries. Geo-NeW tackles this by integrating geometry directly into the learning architecture through both transformer-based mesh encoding and learned finite element spaces, creating an inductive bias that forces the model to respect underlying physical structure.

The work builds on growing interest in foundation models for scientific computing. Traditional neural networks trained on PDEs often memorize specific geometries rather than learning transferable physics principles. By anchoring the method in Finite Element Exterior Calculus, the authors guarantee conservation laws hold exactly—a critical requirement for engineering applications where energy, mass, or momentum conservation cannot be approximate.

For the computational science and engineering sectors, this represents meaningful progress toward deployable AI systems. Real-time PDE solving with guaranteed physical correctness could accelerate simulation workflows in aerospace, climate modeling, and materials science. The demonstrated generalization to unseen domains suggests practical utility beyond controlled academic benchmarks.

The technical contribution—providing a parameterization ensuring solution existence and uniqueness—strengthens theoretical foundations that practitioners need for production systems. However, the method's reliance on mesh representation may limit scalability to extremely high-dimensional problems compared to pure neural approaches. Near-term impact focuses on specialized physics simulation applications rather than broad market disruption.

• →Geo-NeW jointly learns differential operators and finite element spaces encoded with geometry information, enabling better generalization to unseen domains.
• →The method exactly preserves physical conservation laws through Finite Element Exterior Calculus, ensuring physically valid solutions.
• →Transformer-based mesh encoding explicitly connects domain geometry and boundary conditions to PDE solutions.
• →State-of-the-art performance on steady-state PDE benchmarks with significant out-of-distribution generalization improvements over baselines.
• →Novel constitutive model parameterization guarantees existence and uniqueness of solutions for practical deployment.