Researchers present a fast algorithm for working with manifold harmonics, the natural basis functions that come from the Laplace operator on a surface. In plain terms, these basis functions are the analogues of sines and cosines on a circle or spherical harmonics on a sphere. The new method lets one quickly form linear combinations of those basis functions on arbitrary surfaces, which is a common step in many numerical tasks.

Manifold harmonics are the eigenfunctions of the Laplace–Beltrami operator — the Laplace operator adapted to a curved surface. On familiar shapes the eigenfunctions have closed forms: complex exponentials on the circle and spherical harmonics on the sphere. For general surfaces there is no simple formula. The paper aims to make transforms using these general eigenfunctions as fast and memory‑efficient as the special‑case transforms.

The authors build a fast algorithm using a technique called butterfly factorization. This is a hierarchical compression method. It works by splitting the full transform matrix into carefully chosen blocks and approximating those blocks by low‑rank pieces arranged in nested layers. The compressed representation lets the algorithm compute linear combinations of manifold harmonics much faster and using less memory than the full dense matrix would require. The paper reports numerical examples that show speedups and memory reduction across several geometries, discretizations, and applications.

The authors also give a detailed mathematical analysis for the special case when the underlying manifold is the flat periodic square. That provides theoretical backing in one important setting. Outside of that case the paper reports numerical evidence, but the rigorous analysis is not claimed more generally. In addition, the compression is an approximation: the method depends on the transform matrix having subblocks that admit low‑rank approximations, so performance and accuracy can vary with the surface geometry and the discretization used.