AMP (Atomistic Machine-learning Package)

AMP (Atomistic Machine-learning Package) is an open-source package for fitting neural network potentials to atomistic data. It uses Behler-Parrinello symmetry functions as descriptors and provides a simple interface for training potentia…

6. DYNAMICS 6.3 Machine Learning Potentials VERIFIED
Back to Directory Official Website

Overview

AMP (Atomistic Machine-learning Package) is an open-source package for fitting neural network potentials to atomistic data. It uses Behler-Parrinello symmetry functions as descriptors and provides a simple interface for training potentials from DFT data.

Reference Papers

Reference papers are not yet linked for this code.

Full Documentation

Official Resources

  • Homepage: https://amp.readthedocs.io/
  • Documentation: https://amp.readthedocs.io/
  • Source Repository: https://bitbucket.org/andrewpeterson/amp
  • License: GPL-3.0

Overview

AMP (Atomistic Machine-learning Package) is an open-source package for fitting neural network potentials to atomistic data. It uses Behler-Parrinello symmetry functions as descriptors and provides a simple interface for training potentials from DFT data.

Scientific domain: Neural network potentials, atomistic machine learning
Target user community: Researchers training neural network potentials

Theoretical Methods

  • Behler-Parrinello symmetry functions
  • Neural network regression
  • Gaussian descriptor functions
  • Atomic fingerprints

Capabilities (CRITICAL)

  • Neural network potential training
  • Symmetry function descriptors
  • ASE calculator interface
  • Parallel training
  • Multiple descriptor types

Key Strengths

Simplicity:

  • Easy to use
  • Good documentation
  • ASE integration

Flexibility:

  • Custom descriptors
  • Multiple backends

Inputs & Outputs

  • Input formats: ASE trajectory, VASP OUTCAR
  • Output data types: Trained models, energies, forces

Interfaces & Ecosystem

  • ASE: Calculator interface
  • AmpTorch: PyTorch version

Advanced Features

  • Symmetry functions: G2, G4 types
  • Parallel training: MPI support
  • Custom descriptors: Extensible

Performance Characteristics

  • CPU-based training
  • Good for small datasets
  • ASE integration efficient

Computational Cost

  • Training: Hours (CPU)
  • Inference: Fast
  • Overall: Moderate

Best Practices

  • Use appropriate symmetry function parameters
  • Validate on test set
  • Consider AmpTorch for GPU acceleration
  • Start with default descriptors

Limitations & Known Constraints

  • CPU-based (AmpTorch for GPU)
  • Older architecture (pre-equivariant era)
  • Less accurate than equivariant methods
  • Symmetry function choice critical

Application Areas

  • Surface science
  • Catalysis
  • Materials modeling
  • Educational purposes

Comparison with Other Codes

  • vs AmpTorch: AMP CPU-based original, AmpTorch GPU PyTorch version
  • vs NequIP/MACE: AMP descriptor-based, others equivariant (more accurate)
  • vs N2P2: Both Behler-Parrinello style, different implementations
  • Unique strength: Simplicity, good documentation, educational value, ASE integration

Community and Support

  • Established codebase
  • Good documentation
  • ASE community
  • Published tutorials

Verification & Sources

Primary sources:

  1. Documentation: https://amp.readthedocs.io/
  2. A. Khorshidi & A.A. Peterson, Comput. Phys. Commun. 207, 310 (2016)

Secondary sources:

  1. AMP tutorials
  2. AmpTorch documentation
  3. Published catalysis applications

Confidence: VERIFIED

Verification status: ✅ VERIFIED

  • Source code: OPEN (Bitbucket, GPL-3.0)
  • Academic citations: >500
  • Established codebase

Related Tools in 6.3 Machine Learning Potentials