Abstract:Deep learning electronic structures from ab initio calculations holds great potential to revolutionize computational materials studies. While existing methods proved success in deep-learning density functional theory (DFT) Hamiltonian matrices, they are limited to DFT programs using localized atomic-like bases and heavily depend on the form of the bases. Here, we propose the DeepH-r method for deep-learning DFT Hamiltonians in real space, facilitating the prediction of DFT Hamiltonian in a basis-independent manner. An equivariant neural network architecture for modeling the real-space DFT potential is developed, targeting a more fundamental quantity in DFT. The real-space potential exhibits simplified principles of equivariance and enhanced nearsightedness, further boosting the performance of deep learning. When applied to evaluate the Hamiltonian matrix, this method significantly improved in accuracy, as exemplified in multiple case studies. Given the abundance of data in the real-space potential, this work may pave a novel pathway for establishing a ``large materials model" with increased accuracy.

What problem does this paper attempt to address?

The main objective of this paper is to propose a new deep learning method, called DeepH-r, for predicting the Hamiltonian of Density Functional Theory (DFT) in real space. Specifically, the research team developed an equivariant neural network-based approach to directly predict the Kohn-Sham potential (\( \hat{V}_{KS}(r) \)), which is a more fundamental and basis-set-independent physical quantity in DFT calculations. This method has the following advantages over existing techniques: 1. **Simplified Equivariance**: The Kohn-Sham potential exhibits simpler equivariant properties under rotational operations, which helps in designing more flexible neural network structures. 2. **Enhanced Locality**: The Kohn-Sham potential in real space shows stronger locality compared to the Hamiltonian matrix elements based on local basis sets, meaning it is less sensitive to distant changes, which is beneficial for improving the model's generalization ability. 3. **Data Richness**: Compared to the Hamiltonian matrix, the Kohn-Sham potential provides more data about DFT calculations. Larger datasets are generally advantageous for training deep learning models. With these improvements, DeepH-r can efficiently and accurately predict DFT Hamiltonians in a basis-set-independent manner and has demonstrated significantly better performance than previous methods in multiple case studies, including studies on SiC alloy systems and carbon allotropes. Additionally, DeepH-r has shown good transferability and generalization ability, being able to make accurate predictions on different types of materials and structures. This work not only opens new avenues for deep learning in electronic structure but also provides significant advancements for high-precision and general-purpose electronic structure calculations.

Deep learning density functional theory Hamiltonian in real space

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

Improving density matrix electronic structure method by deep learning

Deep Neural Network Representation of Density Functional Theory Hamiltonian

Universal materials model of deep-learning density functional theory Hamiltonian

Neural-network Density Functional Theory Based on Variational Energy Minimization

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

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

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

Generalizing deep learning electronic structure calculation to the plane-wave basis

DeepH-2: Enhancing deep-learning electronic structure via an equivariant local-coordinate transformer

Efficient hybrid density functional calculation by deep learning

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

Deep learning-based density functionals empower AI for materials

A deep learning framework to emulate density functional theory

Infusing Self-Consistency into Density Functional Theory Hamiltonian Prediction via Deep Equilibrium Models

Deep-learning density functional perturbation theory

Deep learning and density-functional theory