NequIP (Neural Equivariant Interatomic Potentials)

Official Resources

• Homepage: https://github.com/mir-group/nequip
• Documentation: https://github.com/mir-group/nequip (README/Wiki)
• Source Repository: https://github.com/mir-group/nequip
• License: MIT License

Overview

NequIP is a code for building E(3)-equivariant neural network interatomic potentials. It uses the e3nn library to ensure that the learned potentials respect rotation and translation symmetries and parity by construction. This data efficiency allows NequIP to achieve high accuracy with very small training sets compared to invariant models.

Scientific domain: Machine learning potentials, equivariant neural networks
Target user community: MD users, ML researchers

Capabilities (CRITICAL)

• Equivariance: E(3)-equivariant features (vectors, tensors) used throughout the network.
• Data Efficiency: High accuracy with few (100-1000) training structures.
• LAMMPS: Interface for running MD with NequIP potentials via pair_nequip.
• ASE: ASE calculator interface.

Sources: NequIP GitHub, Nat. Commun. 13, 2453 (2022)

Inputs & Outputs

• Input formats: Extended XYZ (for training), YAML config
• Output data types: PyTorch model (.pth), deployed model for LAMMPS

Interfaces & Ecosystem

• PyTorch: Core framework.
• e3nn: Library for equivariant operations.
• LAMMPS: Plugin available.
• ASE: Integration.

Workflow and Usage

1. Prepare training data (extxyz with energy/forces).
2. Create config YAML.
3. Train: nequip-train config.yaml
4. Deploy: nequip-deploy build --train-dir results/ model.pth
5. Run LAMMPS MD.

Performance Characteristics

• Slower inference than simple invariant models (due to tensor products).
• Extremely high accuracy and stability.
• Excellent for complex materials where angular dependence is critical.

Application Areas

• Phase transitions
• Reaction dynamics
• Complex oxides
• Liquid structures

Community and Support

• Developed by Kozinsky Group (Harvard)
• Active development

Verification & Sources

Primary sources:

1. GitHub: https://github.com/mir-group/nequip
2. Publication: S. Batzner et al., Nat. Commun. 13, 2453 (2022)

Confidence: VERIFIED

Verification status: ✅ VERIFIED

• Website: ACTIVE
• Documentation: AVAILABLE
• Source: OPEN (GitHub)
• Development: ACTIVE
• Applications: Equivariant ML potentials