Abstract:The combination of deep learning and ab initio materials calculations is emerging as a trending frontier of materials science research, with deep-learning density functional theory (DFT) electronic structure being particularly promising. In this work, we introduce a neural-network method for modeling the DFT density matrix, a fundamental yet previously unexplored quantity in deep-learning electronic structure. Utilizing an advanced neural network framework that leverages the nearsightedness and equivariance properties of the density matrix, the method demonstrates high accuracy and excellent generalizability in multiple example studies, as well as capability to precisely predict charge density and reproduce other electronic structure properties. Given the pivotal role of the density matrix in DFT as well as other computational methods, the current research introduces a novel approach to the deep-learning study of electronic structure properties, opening up new opportunities for deep-learning enhanced computational materials study.

What problem does this paper attempt to address?

This paper aims to improve the density matrix electronic structure method through deep learning. In materials science, the combination of deep learning and ab initio calculations is becoming a frontier for research, especially in deep learning density functional theory (DFT) electronic structure which has shown great potential. The authors propose a neural network approach for modeling the density matrix in DFT, which is a fundamental but previously unexplored quantity. This method utilizes the locality and permutation invariance of the density matrix, demonstrating high accuracy and good generalization ability in multiple example studies, accurately predicting charge densities and reproducing other electronic structure properties. The three fundamental quantities in DFT are the Hamiltonian H, the density matrix ρ, and the charge density n. While there have been works attempting to simulate H or n using neural networks, the neural network representation of ρ has not been addressed yet. The DeepH-DM method described in the paper implements deep learning of the density matrix in a basis set DFT program based on atomic local basis, improving computational efficiency and opening up new avenues for computational material research enhanced by deep learning. The paper also discusses the efficiency advantage of the density matrix compared to the Hamiltonian in computing specific physical quantities, as well as its fundamental role in mixed DFT functionals, density matrix functional theory, and other methods. Through example studies, the high accuracy and generalization performance of the DeepH-DM method on graphene (single layer and bilayer) are demonstrated, further proving its potential applications in structures such as carbon nanotubes. In summary, this paper addresses the problem of effectively predicting and utilizing the density matrix in DFT through deep learning to improve the efficiency and accuracy of electronic structure calculations.

Improving density matrix electronic structure method by deep learning

Deep learning density functional theory Hamiltonian in real space

Deep-learning density functional theory Hamiltonian for efficient ab initio electronic-structure calculation

Improving the efficiency of ab initio electronic-structure calculations by deep learning

Neural-network Density Functional Theory Based on Variational Energy Minimization

A deep equivariant neural network approach for efficient hybrid density functional calculations

Deep-learning density functional perturbation theory

Deep learning-based density functionals empower AI for materials

A Deep Dive into Machine Learning Density Functional Theory for Materials Science and Chemistry

Deep dive into machine learning density functional theory for materials science and chemistry

Efficient hybrid density functional calculation by deep learning

A deep learning framework to emulate density functional theory

Deepks: A Comprehensive Data-Driven Approach Toward Chemically Accurate Density Functional Theory

Universal materials model of deep-learning density functional theory Hamiltonian

Deep Neural Network Representation of Density Functional Theory Hamiltonian

Towards Large-Scale and Spatiotemporally Resolved Diagnosis of Electronic Density of States by Deep Learning

General framework for E(3)-equivariant neural network representation of density functional theory Hamiltonian