Fast contracted Clebsch-Gordan tensor products for equivariant graph neural networks
We present an algorithm with an O(L3) operation count for evaluating contracted Clebsch–Gordan tensor products in O(3)-equivariant machine learning potentials at fixed Canonical Polyadic (CP) rank. Mapping the angular integral to a structured Gauss–Legendre and Fourier tensor-product grid decouples the radial channel contractions from the angular transforms. The antisymmetric parity- odd Clebsch–Gordan channels, unreachable by the symmetric pointwise product on a scalar S2 grid, are recovered through the surface-curl pairing ˆ r·[∇S2 A×∇S2 B], the spherical Poisson bracket, which supplies the L = 1 angular momentum on the grid while preserving rotational equivariance. The construction extends to parity-aware equivariant message passing in architectures of the atomic cluster expansion family and is verified by direct numerical quadrature. The full uncontracted Clebsch–Gordan tensor product remains subject to the O(L4) output-size lower bound. Benchmarks show wall-clock scaling approximately as L2 for both operations over the whole range measured, with the runtime on the tested GPU set by data movement on the L2-sized grid tensors rather than by the L3 arithmetic.
Authors
- Ralf Drautz (ORCID: https://orcid.org/0000-0001-7101-8804)
- Yury V. Lysogorskiy (ORCID: https://orcid.org/0000-0003-4617-3188)
- Anton S. Bochkarev (ORCID: https://orcid.org/0000-0001-7229-5758)
Institutions
- Kazan Federal University (RU)
- Ruhr University Bochum (DE)
Publication Details
- Journal
- Machine Learning Science and Technology
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1088/2632-2153/aec40f
- Citations
- 3
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 2.53
Funders
- Deutsche Forschungsgemeinschaft