AlloMorph: A Toolkit for Metal Nanoparticle Structure Generation for Machine Learning

AlloMorph is a Python toolkit for generating initial structure models of single, double, and triple metal nanoparticles. It supports multiple geometric shapes and atomic arrangements, enabling the creation of large-scale datasets for atomic simulations and machine learning research. This toolkit addresses key challenges in nanomaterial research: high experimental synthesis costs and data scarcity for ML models.

Nanoparticles have wide applications in catalysis, energy, and biomedicine, but experimental synthesis of specific structures is costly and hard to control. Computational simulation requires high-quality initial structures, while ML in materials science faces data scarcity (lack of diverse structure-property pairs). AlloMorph solves these by systematically generating standardized NP structures for simulations and ML training.

AlloMorph's core capabilities include:

• Supported metals: 1-3 elements (Au, Pt, Pd, Cu, Ni, Ag, etc.)
• Size control: Configurable diameter (default:10-30 Å)
• Geometric shapes: Cube, tetrahedron, rhombic dodecahedron, octahedron, truncated octahedron, cuboctahedron, decahedron, icosahedron, sphere
• Stoichiometry: Flexible ratios (e.g.,20:40:40 for ternary)
• Atomic arrangements: For BNP (L10 ordered, RAL random, RCS random core-shell); for TNP (L10R, CS core-shell, CL10S, CRALS, RRAL, CSRAL, CSL10, CRSR, LL10) These cover ordered to random structures, aiding structure-property relation studies.

Installation:

• Using uv: Clone repo → uv venv → uv pip install -e '.[dev]'
• Using pip: pip install allomorph

Quick Start:

• Generate all NP types: allomorph init-struct --stage all
• Generate only single metal NPs: allomorph init-struct --stage mnp
• Generate BNP with visualization: allomorph init-struct --stage bnp --vis (uses ASE GUI)

AlloMorph is used in:

1. Computational research: Compare size stability, core-shell vs alloy performance, surface coordination effects, phase separation in multi-metal NPs.
2. ML dataset building: Generate diverse structures → compute energy/force via LAMMPS → extract features (NCPac) → build structure-property datasets → train models (potential surfaces, property prediction, structure classification).
3. High-throughput screening: Batch generate structures → auto-submit LAMMPS jobs → collect metrics → identify promising materials.

Current Limitations:

• Only supports FCC lattice (no BCC/HCP)
• Max 3 metals (no quaternary+)
• Limited output formats (modern formats like HDF5/ASE trajectory under development)

Future Plans:

• Support quaternary+ alloys
• Add non-FCC lattice (BCC/HCP)
• Implement modern output formats

AlloMorph integrates with key tools:

• LAMMPS: Directly uses generated structures for molecular dynamics simulations.
• NCPac: Extracts structural features for ML training.
• ASE: Visualizes structures and converts formats. These form a complete workflow for nanomaterial research.