ml-foundation-potentials

What problem does it solve?

It helps you select an appropriate foundation machine-learning interatomic potential (MLIP) for your atomistic research so you balance accuracy, computational cost, and material/composition constraints instead of guessing.

Core Features & Use Cases

• Model selection across major MLIP families: Guides picks among MatGL, FAIRCHEM, and MACE models based on chemistry type (organic vs inorganic), desired fidelity level (e.g., r2SCAN-grade), and intended simulation mode.
• Scenario-driven guidance: Prioritizes cheaper smaller models for expensive/dynamic workflows like MD, NEB, phonons, diffusion, and melting-temperature calculations.
• Registry-first reuse workflow: Instructs you to check the local model registry for existing fine-tuned checkpoints for your chemical system before selecting a foundation model or planning new fine-tuning.

Quick Start

Ask your AI agent to select the best foundation MLIP for a Li-Fe-P-O phase-stability study by first searching the model registry for an existing checkpoint and then choosing an appropriate r2SCAN-grade option if none is available.