CrySPY

Overview

CrySPY (pronounced "crispy") is a crystal structure prediction tool written in Python. It supports multiple search algorithms including random search (RS), Bayesian optimization (BO), Look Ahead based on Quadratic Approximation (LAQA), and evolutionary algorithms (EA).

Theoretical Basis

• Random structure searching with symmetry constraints
• Bayesian optimization for efficient sampling
• LAQA for accelerated local optimization
• Evolutionary/genetic algorithms
• Interface with ML potentials for fast screening

Key Capabilities

• Multiple CSP algorithms in one package
• Interface with VASP, QE, soiap, LAMMPS
• Support for ML potentials (M3GNet, MACE)
• Automatic structure generation with PyXtal
• Parallel execution support

Sources: CrySPY documentation, Sci. Technol. Adv. Mater. Methods 1, 87 (2021)

Key Strengths

Algorithm Variety:

• Random search, Bayesian optimization
• LAQA, evolutionary algorithms
• Flexible algorithm selection

Interfaces:

• VASP, Quantum ESPRESSO
• LAMMPS, soiap
• ML potentials (M3GNet, MACE)

Usability:

• Python-based, easy installation
• Good documentation
• Active development

Inputs & Outputs

• Input formats: cryspy.in (configuration), initial structures (optional)
• Output data types: Optimized structures, energy rankings, logs

Interfaces & Ecosystem

• DFT codes: VASP, Quantum ESPRESSO, OpenMX
• ML potentials: M3GNet, MACE, CHGNet
• Structure generation: PyXtal integration

Workflow and Usage

1. Prepare cryspy.in configuration file
2. Set up calculator (VASP/QE/ML potential)
3. Run: cryspy
4. Monitor progress and collect results
5. Analyze lowest energy structures

Performance Characteristics

• Efficient with ML potential pre-screening
• Scales with number of atoms and algorithm choice
• Bayesian optimization reduces required evaluations

Computational Cost

• Depends on calculator (DFT vs ML)
• BO significantly reduces iterations
• Parallelizable across structures

Best Practices

• Use ML potentials for initial screening
• Validate with DFT for final structures
• Choose algorithm based on system complexity
• Use symmetry constraints when appropriate

Limitations & Known Constraints

• Performance depends on calculator choice
• Large systems require ML pre-screening
• Algorithm selection requires experience

Application Areas

• Inorganic crystal structure prediction
• High-pressure phase discovery
• Materials screening with ML potentials
• Alloy structure prediction

Comparison with Other Codes

• vs USPEX: CrySPY more algorithms, USPEX more mature
• vs CALYPSO: CrySPY Python-based, easier customization
• vs XtalOpt: CrySPY more ML integration
• Unique strength: Multiple algorithms, ML potential integration, Python flexibility

Community and Support

• Open-source (MIT License)
• GitHub repository with active development
• Documentation and tutorials available
• Growing user community

Verification & Sources

Primary sources:

1. Homepage: https://tomoki-yamashita.github.io/CrySPY_doc/
2. GitHub: https://github.com/Tomoki-YAMASHITA/CrySPY
3. Publication: Sci. Technol. Adv. Mater. Methods 1, 87 (2021)

Secondary sources:

1. CrySPY tutorials
2. PyXtal documentation
3. Published applications

Confidence: VERIFIED

Verification status: ✅ VERIFIED

• Website: ACTIVE
• Documentation: COMPREHENSIVE
• Source: OPEN (GitHub, MIT)
• Development: ACTIVE
• Applications: Crystal structure prediction, ML-accelerated CSP