Machine learning Hubbard parameters with equivariant neural networks

2024-06-04

npj Computational Materials

AI Summary

Accurately and efficiently determining Hubbard $U$ (and $V$) parameters for strongly correlated materials remains a longstanding challenge. To address this, we propose the first SE(3)-equivariant graph neural network framework for predicting Hubbard parameters, explicitly incorporating 3D rotational and translational symmetries as physical priors. The method takes crystalline atomic structures as input and is trained on high-fidelity GW-calculated $U$ values, thereby unifying physical constraints with cross-material generalizability. On a diverse set of transition metal oxides, our model achieves a mean absolute error in $U$ prediction below 0.15 eV—improving upon empirical DFT+$U$ fitting by over 50%. It enables interpretable, first-principles–level parameter generation without system-specific tuning. This provides a robust, efficient, and transferable parameterization strategy for DFT+$U$+$V$ calculations, significantly advancing predictive capability for strongly correlated systems.

Technology Category

• Machine Learning: Probabilistic Circuits and Graphical Models
• Reasoning under Uncertainty: Relational Probabilistic Models
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Problem

• DFT+U+V
• Transition Metals
• Hubbard Model Parameters

Innovation

• Artificial Intelligence
• Neural Networks
• DFT+U+V Optimization