The article talks about the development of equivariant neural networks for the E(3) group which is important for modeling 3D data. The primary challenge is the computational complexity that increases with the use of higher-order tensors. The authors propose a systematic approach to accelerate these computations by connecting Clebsch-Gordan coefficients to Gaunt coefficients, which are integral products of three spherical harmonics. This connection allows the tensor product of irreducible representations to be equivalent to the multiplication between spherical functions. The work introduces a new method, the Gaunt Tensor Product, to construct efficient equivariant operations across different model architectures. The approach is tested on the Open Catalyst Project and 3BPA datasets, demonstrating improved efficiency and performance.
Publication date: 19 Jan 2024
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2401.10216
