Abstract:Calculating perturbation response properties of materials from first principles provides a vital link between theory and experiment, but is bottlenecked by the high computational cost. Here, a general framework is proposed to perform density functional perturbation theory (DFPT) calculations by neural networks, greatly improving the computational efficiency. Automatic differentiation is applied on neural networks, facilitating accurate computation of derivatives. High efficiency and good accuracy of the approach are demonstrated by studying electron-phonon coupling and related physical quantities. This work brings deep-learning density functional theory and DFPT into a unified framework, creating opportunities for developing ab initio artificial intelligence. https://doi.org/10.1103/PhysRevLett.132.096401 © 2024 American Physical Society

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the efficiency of Density Functional Perturbation Theory (DFPT) calculations while maintaining high precision. DFPT is a method used to predict the response properties of materials to various perturbations, such as phonons and Electron - Phonon Coupling (EPC). These properties play a crucial role in many physical phenomena, such as BCS superconductivity, ferroelectricity, electron and heat transport, infrared and Raman spectroscopy, etc. However, traditional DFPT calculations are very time - consuming, especially when dealing with larger - sized systems, which limits the development of high - throughput materials research. To solve this problem, the paper proposes a general framework based on deep learning to perform DFPT calculations. This framework uses neural networks to learn the changes in Kohn - Sham potential energy and efficiently and accurately calculates the derivatives of physical quantities through automatic differentiation techniques. This method not only significantly improves the computational efficiency but also maintains good accuracy, thus opening up new avenues for the development of ab initio artificial intelligence. Specifically, the main contributions of the paper include: 1. **Improving computational efficiency**: By using neural networks and automatic differentiation techniques, the computational cost in traditional DFPT calculations is greatly reduced. 2. **Maintaining high precision**: By training the neural network model, it is able to accurately predict physical quantities such as electron - phonon coupling. 3. **Unifying framework**: Unifying deep - learning Density Functional Theory (DFT) and DFPT into one framework, providing new ideas for future algorithm development. These improvements enable DFPT calculations to be applied to larger - scale systems and promote the development of high - throughput materials research.

Deep-Learning Density Functional Perturbation Theory

Deep-learning density functional perturbation theory

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

Neural-network Density Functional Theory Based on Variational Energy Minimization

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

Improving density matrix electronic structure method by deep learning

Deep learning density functional theory Hamiltonian in real space

A deep learning framework to emulate density functional theory

Density functionals from deep learning

Accelerating finite-temperature Kohn-Sham density functional theory with deep neural networks

Deep learning-based density functionals empower AI for materials

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

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

Accurate Deep Learning-Aided Density-Free Strategy for Many-Body Dispersion-Corrected Density Functional Theory

Efficient hybrid density functional calculation by deep learning

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

A Time-Dependent Random State Approach for Large-Scale Density Functional Calculations

Deep Neural Network Representation of Density Functional Theory Hamiltonian