MLIP Arena: Advancing Fairness and Transparency in Machine Learning Interatomic Potentials through an Open and Accessible Benchmark Platform

Yuan Chiang, Tobias Kreiman, Elizabeth Weaver, Ishan Amin, Matthew Kuner, Christine Zhang, Aaron Kaplan, Daryl Chrzan, Samuel M Blau, Aditi S. Krishnapriyan, Mark Asta

Submission Track: Paper Track (Tiny Paper)

Submission Category: Automated Synthesis

Keywords: machine-learning interatomic potentials, benchmark, simulations

Abstract: Machine learning interatomic potentials (MLIPs) have revolutionized molecular and materials modeling, but existing benchmarks suffer from data leakage, limited transferability, and an overreliance on error-based metrics tied to specific density functional theory (DFT) references. We introduce MLIP Arena, a benchmark platform that evaluates MLIPs based on physics awareness, chemical reactivity, stability under extreme conditions, and predictive capabilities for thermodynamic properties and physical phenomena. Our evaluation challenges previous assumptions about model architectures and performance. MLIP Arena provides a reproducible framework to guide MLIP development toward improved predictive accuracy and runtime efficiency while maintaining physical consistency. The Python package and online leaderboard are available at https://huggingface.co/spaces/atomind/mlip-arena

Submission Number: 57