Energy&Force Regression on DFT Trajectories is Not Enough for Universal Machine Learning Interatomic Potentials

2025-02-05

AI Summary

This work addresses three critical bottlenecks hindering the practical deployment of general-purpose machine-learned interatomic potentials (MLIPs) in materials discovery: (1) DFT-dependent training data limiting accuracy and transferability; (2) poor reliability in large-scale molecular dynamics (MD) simulations; and (3) weak mechanistic interpretability. To overcome these, we propose a systematic solution: first, replacing DFT with high-accuracy CCSD(T) to generate reference training data; second, establishing the first metrology-inspired MLIP evaluation framework—integrating large-scale benchmarking across diverse materials systems with visualization-enhanced interpretability analysis; and third, developing a lightweight, scalable neural network potential architecture. Our approach enables reproducible, physics-grounded MLIP development, advancing beyond single-component systems toward device-scale, multicomponent material simulations. It significantly improves prediction accuracy, cross-system generalizability, and physical fidelity—particularly for complex, multi-element materials.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Multiagent Systems: Multiagent Learning

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Security and privacy of machine learning and AI applications

Abstract

Universal Machine Learning Interactomic Potentials (MLIPs) enable accelerated simulations for materials discovery. However, current research efforts fail to impactfully utilize MLIPs due to: 1. Overreliance on Density Functional Theory (DFT) for MLIP training data creation; 2. MLIPs' inability to reliably and accurately perform large-scale molecular dynamics (MD) simulations for diverse materials; 3. Limited understanding of MLIPs' underlying capabilities. To address these shortcomings, we aargue that MLIP research efforts should prioritize: 1. Employing more accurate simulation methods for large-scale MLIP training data creation (e.g. Coupled Cluster Theory) that cover a wide range of materials design spaces; 2. Creating MLIP metrology tools that leverage large-scale benchmarking, visualization, and interpretability analyses to provide a deeper understanding of MLIPs' inner workings; 3. Developing computationally efficient MLIPs to execute MD simulations that accurately model a broad set of materials properties. Together, these interdisciplinary research directions can help further the real-world application of MLIPs to accurately model complex materials at device scale.

Problem

• Overreliance on DFT for MLIP training data creation.
• MLIPs' inaccuracy in large-scale molecular dynamics simulations.
• Limited understanding of MLIPs' underlying capabilities.

Innovation

• Employ Coupled Cluster Theory
• Create MLIP metrology tools
• Develop efficient MLIPs MD simulations