AtomicAI 0.4.0

Processing and visualization of atomic coordinates; featurizing atomic structures

AtomicAI

AtomicAI is a Python-based package for building machine learning models in the field of computational materials science. It can be used for machine learning force field generation, atomic structure classifications based on temperature, defects, phase, and more.

Table of Contents

1. Features
2. Installation
3. Requirements
4. Usage
5. Examples
6. Documentation
7. Contributing
8. License
9. Authors
10. Contact
11. Cite

Features

AtomicAI offers a wide range of features for computational materials science:

Analysis

• Processing and visualization of atomic coordinates in various file formats (CIF, VASP, XYZ, Conquest, etc.)
• Utilizes 'ase' library for reading and writing structure files
• Radial Distribution Function (RDF) calculation and analysis
• Structure analysis tools

Data Processing

• Classification of atoms in trajectory files
• File format conversion between different atomic structure formats

Machine Learning

• Featurization of atomic structures
• Dimension reduction techniques (PCA, LPP, TsLPP, TsNE, UMAP)
• Machine Learning Force Field (MLFF) generation
• Descriptor generation for atomic environments and forces

Clustering and Visualization

• Various clustering methods
• Data plotting and visualization tools for RDF, molecular dynamics statistics, and clustering results

Installation

Install AtomicAI using pip:

pip install AtomicAI

Requirements

AtomicAI requires Python 3.7 or higher. Other dependencies will be automatically installed during the pip installation process.

Usage

AtomicAI provides several command-line tools for various tasks. Here are some examples:

File Conversion

• cq2vasp: Convert Conquest files to VASP format
• xyz2vasp: Convert XYZ files to VASP format
• vasp2cif: Convert VASP files to CIF format
• vasp2xyz: Convert VASP files to XYZ format
• vasp2lmp_data: Convert VASP files to LAMMPS data format
• vasp2cq: Convert VASP files to Conquest format
• lmp2vasp: Convert LAMMPS trajectory to VASP format
• cif2cq: Convert CIF files to Conquest format
• cq2cif: Convert Conquest files to CIF format
• ase_traj2xyz_traj: Convert ASE trajectory to XYZ trajectory

Structure Manipulation

• supercell: Create supercell structures
• build_interface: Build interfaces between materials
• wrap2unwrap: Convert wrapped coordinates to unwrapped

Analysis

• rdf: Calculate Radial Distribution Function
• structure_analysis: Perform various structural analyses

Visualization

• plot_rdf_data: Plot RDF data
• plot_md_stats: Plot molecular dynamics statistics
• plot_vasp_md: Plot VASP molecular dynamics results
• plot_lammps_md: Plot LAMMPS molecular dynamics results
• plot_clusters: Visualize clustering results

Machine Learning

Atomic Descriptors — generate_descriptors

Computes locally-averaged atomic fingerprints (LAAF) for all element pairs in a trajectory.

generate_descriptors trajectory.xyz [--descriptor TYPE [TYPE ...]] [--n-eta N]

Option	Default	Description
--descriptor / -d	ACSF_G2 ACSF_G2G4 SOAP	One or more descriptor types (see table below)
--n-eta / -n	50	Number of eta decay functions

Available descriptor types:

Type	Description
ACSF_G2	Radial symmetry functions (Behler-Parrinello G2)
ACSF_G3	Cosine basis functions parameterised by kappa
ACSF_G4	Angular symmetry functions including the r_jk cutoff (Behler-Parrinello G4)
ACSF_G5	Angular symmetry functions without r_jk cutoff (G5 variant)
ACSF_G2G4	Combined radial (G2) + angular (G4) descriptor
ACSF_G2G4G5	Combined radial (G2) + angular G4 + angular G5 descriptor
SOAP	Smooth Overlap of Atomic Positions (via DScribe)
MBSF	Many-body symmetry functions (radial gr + angular ga)

Examples:

# G2 radial only, 60 eta functions
generate_descriptors traj.xyz --descriptor ACSF_G2 --n-eta 60

# Multiple descriptor types in one run
generate_descriptors traj.xyz --descriptor ACSF_G2G4 ACSF_G2G4G5 MBSF

# SOAP descriptor
generate_descriptors traj.xyz --descriptor SOAP

Output files are written to ./descriptors/ with naming convention <TYPE>_<rc>_<ra>_<elem1>_<elem2>.dat.

Force Descriptors — generate_force_descriptors

Generates force-projected descriptors for machine learning force field training.

generate_force_descriptors trajectory.xyz [--fp-type TYPE] [--rc R] [--n2b N] [--n3b N]

Option	Default	Description
--fp-type / -f	Split2b3b_ss	Fingerprint type: BP2b or Split2b3b_ss
--rc / -r	10.5	Cutoff radius in Angstrom
--n2b	20	Number of 2-body eta functions
--n3b	10	Number of 3-body eta functions

Available fingerprint types:

Type	Description
BP2b	2-body Behler-Parrinello fingerprint
Split2b3b_ss	Split 2-body + 3-body fingerprint (same species)

Examples:

# Default: Split2b3b_ss with 10.5 Å cutoff
generate_force_descriptors traj.xyz

# BP2b with shorter cutoff and more eta functions
generate_force_descriptors traj.xyz --fp-type BP2b --rc 8.0 --n2b 30

# Split2b3b with custom eta counts
generate_force_descriptors traj.xyz --fp-type Split2b3b_ss --rc 9.0 --n2b 25 --n3b 12

Output is written to ./descriptors/force_descriptors.dat.

Other ML Tools

• laaf: Calculate LAAF descriptors
• dim_reduction: Perform dimension reduction
• dim_reduction_mpi: Perform parallel dimension reduction using MPI
• pca: Perform Principal Component Analysis
• lpp: Perform Locality Preserving Projection
• optimize_tslpp_hyperparameters_without_prediction: Optimize TsLPP hyperparameters
• optimize_tslpp_hyperparameters_with_prediction: Optimize TsLPP hyperparameters with prediction
• predict_tslpp: Predict using optimized TsLPP model

Structure Building

• build_multilayers: Stack multiple VASP structures into commensurate multilayer systems
• build_constrained_multilayers: Build multilayers with lattice constraints
• surfaces: Generate surface slabs for common Miller indices from a bulk structure

LAMMPS Input Generation

• lammps_npt_inputs: Generate LAMMPS NPT input files
• lammps_nvt_inputs: Generate LAMMPS NVT input files

For more details on any tool, run it with the --help flag.

Examples

(This section should be filled with basic examples of how to use the main features of AtomicAI. As the context doesn't provide specific examples, I'll leave this section for you to fill in with relevant use cases.)

Documentation

(Add a link to the full documentation when available. The context doesn't provide this information, so you may want to add it when documentation is ready.)

Contributing

We welcome contributions to AtomicAI! Please see our contributing guidelines for more information. (You may want to add a CONTRIBUTING.md file to your repository with detailed guidelines.)

License

AtomicAI is released under the MIT License. See the LICENSE.md file for details.

Authors

• Selva Chandrasekaran Selvaraj

Contact

• Email: [e-mail]
• Website: https://sites.google.com/view/selvas
• Twitter: https://twitter.com/selva_odc
• LinkedIn: https://www.linkedin.com/in/selvachandrasekaranselvaraj/
• Google Scholar: https://scholar.google.com/citations?user=vNozeNYAAAAJ&hl=en
• Scopus: https://www.scopus.com/authid/detail.uri?authorId=57225319817
• ResearchGate: https://www.researchgate.net/profile/Selva-Chandrasekaran-Selvaraj
• PyPI: https://pypi.org/project/AtomicAI/
• Documentation: https://atomicai.readthedocs.io/en/latest/
• GitHub: https://github.com/SelvaGith/AtomicAI

Cite

If you use AtomicAI in your research, please cite: