mat-random-structure-search

What problem does it solve?

It solves the problem of discovering promising low-energy crystal structures for a given chemical composition without imposing a strong structural prior.

Core Features & Use Cases

• Random structure generation (AIRSS-style): Produces chemically correct candidate crystals by sampling space groups, lattice parameters, and atomic positions under minimum-distance filters.
• MLIP-based relaxation: Relaxes the generated candidates using a machine-learning interatomic potential to obtain low-energy minima efficiently.
• Energy ranking and validation: Ranks relaxed structures by energy, removes duplicates via structure matching, and validates top candidates using thermodynamic stability (E_hull) against known references.

Use cases include searching for ground-state or metastable polymorphs of inorganic solids (e.g., cathodes, salts, ionic conductors) and exploring the energy landscape when no reliable structural starting guess exists.

Quick Start

Generate 100 random NaCl structures and relax them with an MLIP to rank the lowest-energy candidates.