Platonic Representation of Foundation Machine Learning Interatomic Potentials

:microscope: Overview

Foundation machine learning interatomic potentials (MLIPs) are trained on overlapping chemical spaces, yet their latent representations remain model-specific and incommensurable. This work introduces the Platonic representation—a unified framework that aligns embeddings from diverse MLIPs into a common geometric space.

Key Contributions:

:dart: Universal alignment framework: Projects 7+ MLIP models into shared latent space using anchor-based transformation :bar_chart: Model interoperability: Enables direct cross-model comparison via optimal transport and similarity metrics :wrench: Practical applications: Zero-shot model stitching, fine-tuning diagnostics, and symmetry-breaking detection :test_tube: Physical validation: Correlates geometric distortions with prediction failures (phonon dispersions, rotational invariance)

:rocket: Quick Start

Construction flow

Open Notebook :)

1. Load a foundation model
2. Extract embeddings from structures
3. Initialize Platonic transform with K=100 anchors
4. Transform to unified space
5. Compare models
6. Compute similarity

:bar_chart: Reproducing Paper Results

Download Pre-extracted Embeddings

Zenodo (DOI: 10.5281/zenodo.17721681)

Packages needed

:abacus: Already convergence-tested Models

Model	Architecture	Training Data	Status
MACE-MP-0-large	E(3)-equiv. GNN	MPtrj	✅ Supported
MACE-MP-0-medium	E(3)-equiv. GNN	MPtrj	✅ Supported
MACE-MP-0-small	E(3)-equiv. GNN	MPtrj	✅ Supported
MACE-OMat	E(3)-equiv. GNN	OMat24	✅ Supported
SevenNet-OMat	E(3)-equiv. GNN	OMat24	✅ Supported
Orb-v3-conservative	GNN	OMat24	✅ Supported
Orb-v3-direct	GNN	OMat24	✅ Supported
NequIP-OAM-L	E(3)-equiv. GNN	OMat24, Alexandria, MPtrj	✅ Supported
NequIP-MP-L	E(3)-equiv. GNN	OMat24, Alexandria, MPtrj	✅ Supported
MACE-MPA-0	E(3)-equiv. GNN	MPTrj + sAlex	✅ Supported

Made with :heart: by the Walsh Materials Design Group

:world_map: Roadmap v1.0 (Current): Core framework + paper reproduction Contributions toward these goals are welcome!

Last Updated: January 2025