FitSNAP

**FitSNAP** is software for generating machine-learning interatomic potentials for LAMMPS. It implements SNAP, qSNAP, and other linear/nonlinear potentials with tight LAMMPS integration for production MD simulations.

10. NICHE & ML 10.2 MLIPs ACE Linear VERIFIED
Back to Directory Official Website

Overview

**FitSNAP** is software for generating machine-learning interatomic potentials for LAMMPS. It implements SNAP, qSNAP, and other linear/nonlinear potentials with tight LAMMPS integration for production MD simulations.

Reference Papers

Reference papers are not yet linked for this code.

Full Documentation

Official Resources

  • Source Repository: https://github.com/FitSNAP/FitSNAP
  • Documentation: https://fitsnap.github.io/
  • License: Open source (MIT)

Overview

FitSNAP is software for generating machine-learning interatomic potentials for LAMMPS. It implements SNAP, qSNAP, and other linear/nonlinear potentials with tight LAMMPS integration for production MD simulations.

Scientific domain: SNAP/qSNAP potential fitting for LAMMPS
Target user community: Researchers fitting SNAP potentials for LAMMPS MD

Theoretical Methods

  • SNAP (Spectral Neighbor Analysis Potential)
  • qSNAP (quadratic SNAP)
  • Linear and nonlinear fitting
  • Bispectrum descriptors
  • LAMMPS mliap integration

Capabilities (CRITICAL)

  • SNAP potential fitting
  • qSNAP quadratic extension
  • LAMMPS production integration
  • Multi-element support
  • Parallel fitting
  • Uncertainty quantification

Sources: GitHub repository

Key Strengths

LAMMPS Integration:

  • Direct mliap pair_style
  • Production MD ready
  • Parallel execution
  • No format conversion

SNAP Framework:

  • Well-tested SNAP implementation
  • qSNAP for improved accuracy
  • Linear regression (fast fitting)
  • Physics-informed constraints

Production Quality:

  • Published potentials available
  • Sandia National Labs maintained
  • Extensive testing
  • Documentation

Inputs & Outputs

  • Input formats: Training data (LAMMPS dump, VASP, etc.)
  • Output data types: LAMMPS potential files, SNAP coefficients

Interfaces & Ecosystem

  • LAMMPS: MD engine
  • Python: Core language
  • NumPy: Computation

Performance Characteristics

  • Speed: Very fast (linear model)
  • Accuracy: SNAP-level (~100 meV/atom)
  • System size: Any (LAMMPS)
  • Automation: Full

Computational Cost

  • Fitting: Minutes
  • MD: Very fast (linear evaluation)

Limitations & Known Constraints

  • SNAP accuracy: Lower than NN potentials
  • Descriptor fixed: SNAP only
  • LAMMPS only: No other MD engines
  • Training data: Needs diverse configurations

Comparison with Other Codes

  • vs ACE1pack: FitSNAP is SNAP, ACE1pack is ACE basis
  • vs DeePMD-kit: FitSNAP is linear, DeePMD is NN
  • vs PACE: FitSNAP is SNAP, PACE is ACE in LAMMPS
  • Unique strength: SNAP/qSNAP fitting with direct LAMMPS mliap integration

Application Areas

SNAP Potentials:

  • Tungsten, tantalum, uranium
  • BCC/FCC metals
  • High-temperature MD
  • Radiation damage

LAMMPS MD:

  • Production MD with MLIP
  • Large-scale simulations
  • Multi-million atom runs

Best Practices

  • Use diverse training data
  • Validate with elastic constants
  • Test phonon spectra
  • Compare with DFT MD

Community and Support

  • Open source (MIT)
  • Sandia National Labs maintained
  • Comprehensive documentation
  • Published potentials library

Verification & Sources

Primary sources:

  1. GitHub: https://github.com/FitSNAP/FitSNAP

Confidence: VERIFIED

Verification status: ✅ VERIFIED

  • Source code: ACCESSIBLE (GitHub)
  • Specialized strength: SNAP/qSNAP fitting with direct LAMMPS mliap integration

Related Tools in 10.2 MLIPs ACE Linear