Pushing the limits of unconstrained machine-learned interatomic potentials

2026-01-22

AI Summary

This work investigates whether machine learning interatomic potentials (MLIPs) trained without explicit enforcement of physical constraints can outperform traditional models that strictly adhere to physical laws under large-scale training regimes. We propose a novel paradigm that forgoes hard-coded physical symmetries and conservation laws during training, instead leveraging massive datasets and highly parameterized architectures to enhance representational capacity, while restoring essential physical consistency through lightweight post-processing at inference time. Experimental results demonstrate that this unconstrained MLIP achieves significantly higher prediction accuracy than constrained counterparts in static simulation tasks such as geometry optimization and lattice dynamics, all while maintaining superior computational efficiency. These findings validate the effectiveness and practicality of a “large data plus inference correction” approach for atomic-scale modeling.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition
• Natural Language Processing: (Large) Language Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Machine-learned interatomic potentials (MLIPs) are increasingly used to replace computationally demanding electronic-structure calculations to model matter at the atomic scale. The most commonly used model architectures are constrained to fulfill a number of physical laws exactly, from geometric symmetries to energy conservation. Evidence is mounting that relaxing some of these constraints can be beneficial to the efficiency and (somewhat surprisingly) accuracy of MLIPs, even though care should be taken to avoid qualitative failures associated with the breaking of physical symmetries. Given the recent trend of \emph{scaling up} models to larger numbers of parameters and training samples, a very important question is how unconstrained MLIPs behave in this limit. Here we investigate this issue, showing that -- when trained on large datasets -- unconstrained models can be superior in accuracy and speed when compared to physically constrained models. We assess these models both in terms of benchmark accuracy and in terms of usability in practical scenarios, focusing on static simulation workflows such as geometry optimization and lattice dynamics. We conclude that accurate unconstrained models can be applied with confidence, especially since simple inference-time modifications can be used to recover observables that are consistent with the relevant physical symmetries.

Problem

• machine-learned interatomic potentials
• unconstrained models
• physical symmetries
• scaling up
• accuracy

Innovation

• unconstrained machine-learned interatomic potentials
• scaling up
• physical symmetries
• inference-time modifications
• accuracy-efficiency trade-off