Abstract:Deep-learning density functional theory (DFT) shows great promise to significantly accelerate material discovery and potentially revolutionize materials research. However, current research in this field primarily relies on data-driven supervised learning, making the developments of neural networks and DFT isolated from each other. In this work, we present a theoretical framework of neural-network DFT, which unifies the optimization of neural networks with the variational computation of DFT, enabling physics-informed unsupervised learning. Moreover, we develop a differential DFT code incorporated with deep-learning DFT Hamiltonian, and introduce algorithms of automatic differentiation and backpropagation into DFT, demonstrating the capability of neural-network DFT. The physics-informed neural-network architecture not only surpasses conventional approaches in accuracy and efficiency, but also offers a new paradigm for developing deep-learning DFT methods.

What problem does this paper attempt to address?

The main objective of this paper is to propose a new theoretical framework, namely Neural-network Density Functional Theory (NN-DFT) based on variational energy minimization, aiming to combine deep learning with Density Functional Theory (DFT) to achieve more efficient and accurate material simulations. Specifically, this study attempts to address the following core issues: 1. **Separation of existing methods**: Current research typically treats DFT calculations and neural network training as two independent processes. First, material datasets are obtained through DFT calculations, and then these data are used to train neural network models. This approach leads to a separation between DFT and neural networks. 2. **Collaborative optimization**: To solve the above problem, the paper proposes a theoretical framework that unifies the energy functional optimization in DFT with the loss function optimization in neural networks, allowing both to promote each other's development. 3. **Integration of physical information**: By directly incorporating physical knowledge into deep learning, researchers hope that the trained neural network models can better simulate real physical phenomena, rather than just learning in a data-driven manner. To achieve this goal, the authors have undertaken the following work: - **Theoretical framework construction**: Established the theoretical framework of NN-DFT, which allows DFT calculations to be performed simultaneously during neural network training. - **Development of computational code**: Developed a computational code named AI2DFT, which combines equivariant neural networks to learn the mapping from material structures to DFT Hamiltonians, and makes the DFT algorithm differentiable, thereby introducing modern techniques such as automatic differentiation and backpropagation. - **Validation and testing**: Conducted comprehensive testing of AI2DFT, including verifying its performance on different types of materials and comparing it with traditional supervised learning methods. Through the above work, the paper demonstrates that this unsupervised learning method guided by physical information not only surpasses traditional supervised learning methods in accuracy but also shows significant advantages in predicting derived physical quantities. This indicates that the new approach of combining deep learning with DFT is of great significance for advancing materials science research.

Neural-network Density Functional Theory Based on Variational Energy Minimization

Deep Neural Network Representation of Density Functional Theory Hamiltonian

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

Deep learning density functional theory Hamiltonian in real space

Improving density matrix electronic structure method by deep learning

Deep-Learning Density Functional Perturbation Theory

Deep learning-based density functionals empower AI for materials

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

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

A deep learning framework to emulate density functional theory

Neural network distillation of orbital dependent density functional theory

Pushing the frontiers of density functionals by solving the fractional electron problem

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

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

D4FT: A Deep Learning Approach to Kohn-Sham Density Functional Theory