AI-Assisted Crystal Structure Generation
Local Environment Geometry-Oriented Crystal Generator - A symmetry-informed AI generative approach that expands carbon allotropes from 25 known structures to 1,700+ low-energy candidates optimized with machine learning descriptors.
Developed by OG Ridwan, led by Prof. Qiang Zhu | Materials Modelling & Informatics Group (MMI) | UNC Charlotte
Formulae, space group, topology, or energy ranges.
Line up MACE, FF, and VASP energies per structure.
Open a row to view geometry and metadata instantly.
Type a formula, key=value pair, or a boolean expression.
| age | Time since creation | |
| calculator | ASE-calculator name | |
| charge | Net charge in unit cell | |e| |
| density | density | |
| dimension | dimension | |
| dof | dof | |
| energy | Total energy | eV |
| ff_energy | ff_energy | |
| ff_lib | ff_lib | |
| ff_relaxed | ff_relaxed | |
| fmax | Maximum force | eV/Å |
| formula | Chemical formula | |
| id | Uniqe row ID | |
| mace_energy | mace_energy | |
| mace_relaxed | mace_relaxed | |
| magmom | Magnetic moment | μ_B |
| mass | Sum of atomic masses in unit cell | au |
| natoms | Number of atoms | |
| pbc | Periodic boundary conditions | |
| pearson_symbol | pearson_symbol | |
| remark | remark | |
| similarity | similarity | |
| similarity0 | similarity0 | |
| smax | Maximum stress on unit cell | eV/ų |
| space_group_number | space_group_number | |
| topology | topology | |
| topology_detail | topology_detail | |
| unique_id | Random (unique) ID | |
| user | Username | |
| vasp_energy | vasp_energy | |
| vasp_relaxed | vasp_relaxed | |
| volume | Volume of unit cell | ų |
| wps | wps |
LEGO-xtal (Local Environment Geometry-Oriented Crystal Generator) is a symmetry-informed AI generative approach for crystal structure prediction. Our method generates diverse crystal structures using machine learning and optimizes them using structure descriptors rather than traditional energy minimization.
Expanded from 25 known sp2 carbon allotropes to over 1,700 low-energy candidates
Symmetry-informed generative models trained on augmented datasets
Optimized within 0.5 eV/atom of graphite ground state
Adaptable to MOFs, battery materials, and modular building blocks
AI-assisted rapid crystal structure generation towards a target local environment
Ridwan, O.G., Pitie, S., Raj, M.S., Dai, D., Frapper, G., Xue, H., Zhu, Q.
npj Computational Materials (2026)