tricor

Generate disordered atomic supercells guided by three-body (g3) distributions, spanning the full spectrum from liquid to nanocrystalline.

Silicon, liquid → nanocrystalline. See Static Examples → Silicon.

Overview

tricor builds periodic supercells with controllable disorder, from fully liquid to nanocrystalline, by combining Voronoi grain construction with spring-network relaxation. The resulting structures are characterized by their rooted three-body (g3) distributions, which capture both radial and angular correlations.

Key features:

• Rapidly generate structures from liquid to nanocrystalline with a single call

• Voronoi grain construction with per-species-pair bond topology

• Interactive g3 distribution widgets for Jupyter (Three.js + anywidget)

• Self-contained HTML exporters for trajectories, g3 heatmaps, and multi-cell overviews

• Rotating MP4 / GIF movie export for publication figures

• Works with any crystal structure (Si, SiC, oxides, metals, etc.)

Refining generated structures

tricor generates the initial supercell; a separate relaxation step turns it into a physically realistic structure. Two worked pipelines are documented, in order of accuracy:

• MACE-MP0 refinement (recommended) — relax with a universal machine-learning potential. Near-DFT accuracy, minutes per ~5000-atom cell on a laptop CPU.

• Fast FIRE refinement — the built-in spring-network FIRE quench. Orders of magnitude faster than MACE at reduced accuracy, and the only option that stays practical at 100³ Å and larger. Optionally calibrate the springs against MACE with one call (shell.calibrate_to_mace()) — Morse anharmonicity, per-pair stiffness, and the hard-core wall are measured from the MACE potential on the reference crystal, improving accuracy at zero extra relaxation cost.