Overview
It supports network development for three-dimensional data. The official MACE repository documents an e3nn backend, establishing atomistic relevance without making e3nn itself a pretrained potential.
Limitations
A downstream model is still required. The README warns that the main branch is unstable and releases can introduce breaking changes.
Key Features
- Equivariant network components
- Tensor products and spherical harmonics
Use Cases
- Study three-dimensional equivariant models
- Develop atomistic network components
How to Use
Prepare PyTorch, follow the installation guide and start with representation and tensor-product examples. Pin a suitable version for downstream work.