Solving the Incompressible Navier-Stokes Equations on Oriented Curved Surfaces Discretized by Point Clouds
arXiv:2609.00216
2026
Geometry
2 ideas extracted · analyzed Sep 2, 2026
What the math gives to ML
The transferable asset is not the particular Navier–Stokes solver, but its meshfree intrinsic calculus for vector fields stored on irregular samples of a curved manifold. Tangent projection provides a simple way to prevent neural outputs from acquiring nonphysical normal components, while projected covariant derivatives give geometry-aware local interactions without constructing a mesh. The artificial-compressibility analysis also suggests a pressure-like feedback state that suppresses local divergence through a tunable stiffness parameter and has a measurable oscillation-frequency signature. The strongest initial tests are therefore a tangent-constrained point-cloud network and a divergence-feedback decoder for predicting surface vector fields.
Ideas from this paper
✓✓ Beats tuned baseline
2026
Modify a point-cloud message-passing or neural-operator layer so that scalar gradients, vector features, and vector-to-vector interactions are computed only in the estimated tangent plane of the surface. Projecting both feature values and derivative directions prevents the network from using arbitrary ambient-space normal directions and should improve transfer across differently embedded but intrinsically similar surfaces.
Useful7/10
Difficulty4/10
Novelty5/10
Unverified
2026
Add a pressure-like recurrent state to a neural surface-flow decoder and update it from the predicted local divergence, creating a learned or fixed feedback loop that drives vector outputs toward local incompressibility. Unlike a static divergence penalty, the state can accumulate constraint violations and produce corrective tangent gradients at each refinement step.
Useful6/10
Difficulty5/10
Novelty6/10