Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction

Xiang Fu, Brandon Wood, Luis Barroso-Luque, Daniel S. Levine, Meng Gao, Misko Dzamba, Larry Zitnick

International Conference on Machine Learning 2025 · Oral

This article delves into a pivotal talk presented at ICML 2025 by Brandon Wood and Meng Gao from Meta's Fair Chemistry team, highlighting groundbreaking work led by Xiang Fu. The presentation, titled "Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction," addresses a fundamental challenge in computational materials science: accelerating the discovery and design of novel materials using artificial intelligence. Specifically, it focuses on developing highly accurate and stable AI surrogates for computationally intensive quantum chemistry methods like **Density Functional Theory (DFT)**.

AI review

This is competent, honest applied ML work that makes a clear and useful engineering argument: smooth interatomic potentials, enforced through gradient-derived forces and smooth cutoff functions, are necessary for physically meaningful downstream property prediction. The central claim is well-motivated and the phonon prediction results are genuinely compelling. However, as presented, this sits closer to principled engineering than theoretical ML — the 'theorem' bounding energy drift by smoothness is invoked but not scrutinized, the architectural contribution (ESAN) is described without enough…