Official Resources
- Source Repository: https://github.com/libAtoms/QUIP
- Documentation: https://libatoms.github.io/QUIP/
- License: Open source (GPL-2.0)
Overview
GAP (Gaussian Approximation Potential) with QUIP (Quantum Interatomic Potential) is the original MLIP framework using SOAP descriptors and sparse Gaussian processes. It pioneered the field of ML interatomic potentials and remains widely used for element-specific high-accuracy potentials.
Scientific domain: Gaussian process MLIP with SOAP descriptors
Target user community: Researchers fitting high-accuracy GP potentials for specific elements/systems
Theoretical Methods
- Gaussian Approximation Potential (GAP)
- SOAP (Smooth Overlap of Atomic Positions) descriptors
- Sparse Gaussian process regression
- 2-body, 3-body, and SOAP terms
- Fortran core with Python interface (quippy)
Capabilities (CRITICAL)
- SOAP-GAP fitting
- Multi-body terms (2b, 3b, SOAP)
- LAMMPS integration via quippy
- ASE interface
- Well-tested on many systems
Sources: GitHub repository, Int. J. Quantum Chem. 115, 1051 (2015)
Key Strengths
Pioneering:
- First widely-used MLIP framework
- SOAP descriptor standard
- Well-validated on many systems
- Published potentials library
Accuracy:
- Near-DFT accuracy
- Well-calibrated uncertainty
- Systematic improvement
- Physical constraints
Integration:
- LAMMPS via quippy
- ASE calculator
- Fortran performance
- Python flexibility
Inputs & Outputs
- Input formats: Extended XYZ, VASP, CASTEP
- Output data types: GAP potentials, energies, forces, virials
Interfaces & Ecosystem
- LAMMPS: MD engine
- ASE: Calculator (quippy)
- Fortran: Core
- Python: Interface
Performance Characteristics
- Speed: Moderate (GP evaluation)
- Accuracy: Near-DFT
- System size: 100-10000 atoms
- Automation: Semi-automated
Computational Cost
- Fitting: Hours
- MD: 10-100x faster than DFT
Limitations & Known Constraints
- GPL license: Copyleft
- Fortran build: Complex compilation
- GP scaling: O(N²) training
- Element-specific: Not universal out-of-box
Comparison with Other Codes
- vs MACE: GAP is GP, MACE is NN
- vs FLARE: GAP is batch, FLARE is on-the-fly
- vs SNAP: GAP is GP, SNAP is linear
- Unique strength: Pioneering MLIP framework with SOAP descriptors and sparse GP, well-validated across many systems
Application Areas
High-Accuracy Potentials:
- Carbon (GAP-20)
- Silicon, germanium
- Tungsten, iron
- Water, aqueous systems
Materials MD:
- Phase transitions
- Amorphous materials
- Defect properties
- Thermal transport
Best Practices
- Use SOAP + 2b + 3b terms
- Start from published potentials
- Validate with phonons and elastic constants
- Use quippy for LAMMPS integration
Community and Support
- Open source (GPL-2.0)
- libAtoms community
- Cambridge/Oxford maintained
- Comprehensive documentation
Verification & Sources
Primary sources:
- GitHub: https://github.com/libAtoms/QUIP
Confidence: VERIFIED
Verification status: ✅ VERIFIED
- Source code: ACCESSIBLE (GitHub)
- Specialized strength: Pioneering MLIP framework with SOAP descriptors and sparse GP