Generalization Bounds for Physics-Informed Neural Networks for the Incompressible Navier-Stokes Equations A preliminary version of this result was first presented at the meeting DRSciML 2025 (link)
Sebastien Andre-Sloan, Dibyakanti Kumar, Alejandro F Frangi, Anirbit Mukherjee · 2026
Read Generalization Bounds for Physics-Informed Neural Networks for the Incompressible Navier-Stokes Equations A preliminary version of this result was first presented at the meeting DRSciML 2025 (link) by Sebastien Andre-Sloan, Dibyakanti Kumar, Alejandro F Frangi, Anirbit Mukherjee (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-23072.txt.
Description
Andre-Sloan and co-authors prove the first generalization bounds for physics-informed neural networks on fluid flow. The target is the incompressible Navier-Stokes equations. Their networks have depth two and train without labeled data. The bounds come from the Rademacher complexity of the PINN risk. For weight-bounded networks, width does not appear in the bounds. The gap depends on fluid viscosity and loss settings instead. Sample needs show no direct tie to dimension. Tests on the Taylor-Green vortex support the suggested activation choices.
Catalog record
| CREATOR | Sebastien Andre-Sloan, Dibyakanti Kumar, Alejandro F Frangi, Anirbit Mukherjee |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 104499 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.23072