A data-driven view of exterior calculus
Published in AI in Mathematics and Theoretical Physics, Vol. 4, No. 1, 2026
AI in Mathematics and Theoretical Physics, Vol. 4, No. 1
Classical computational tools for differential forms assume symbolic expressions, mesh discretizations, or smoothness close to \(C^1\). In data-driven settings one often has only oracle access: the ability to evaluate a form at arbitrary points and tangent vectors, with no closed-form expression and no mesh.
Building on a flux-based reformulation of the exterior derivative, recently developed by the author with Fadel and Sá Earp, in which the exterior derivative is recovered as a limit of normalized boundary integrals over shrinking blocks, we describe a mesh-free, symbol-free numerical scheme for exterior differentiation, valid under regularity assumptions weaker than \(C^1\).
We demonstrate the method in \(\mathbb{R}^2\), \(\mathbb{R}^3\) and on the sphere \(S^2\), verifying Stokes’ theorem, the identity \(d\circ d=0\), and Gauss–Bonnet purely from pointwise samples, and detecting de Rham cohomology (closed-but-not-exact forms) from data. Companion Jupyter notebooks accompany the paper.
This paper is based on a tutorial delivered at DANGER: Data, Numbers, and Geometry (Banff, April 2026). The combined notes from all three tutorials at that workshop, with Jessica N. Howard and Yidi Qi, are available as arXiv:2610.07220.