Abstract:Neural network force fields have significantly advanced ab initio atomistic simulations across diverse fields. However, their application in the realm of magnetic materials is still in its early stage due to challenges posed by the subtle magnetic energy landscape and the difficulty of obtaining training data. Here we introduce a data-efficient neural network architecture to represent density functional theory total energy, atomic forces, and magnetic forces as functions of atomic and magnetic structures. Our approach incorporates the principle of equivariance under the three-dimensional Euclidean group into the neural network model. Through systematic experiments on various systems, including monolayer magnets, curved nanotube magnets, and moiré-twisted bilayer magnets of $\text{CrI}_{3}$, we showcase the method's high efficiency and accuracy, as well as exceptional generalization ability. The work creates opportunities for exploring magnetic phenomena in large-scale materials systems.

What problem does this paper attempt to address?

This paper aims to address the challenges of applying Neural Network Force Fields (NNFFs) in magnetic materials. Traditional NNFFs have made significant progress in dealing with non-magnetic materials, but their application in magnetic materials is still in its infancy, mainly due to the intricacy of the magnetic energy landscape and the difficulty in obtaining training data. The paper presents a data-efficient neural network architecture with Euclidean group equivariance for representing the density functional theory total energy, atomic forces, and magnetic forces of atoms and magnetic structures. This approach reduces the need for additional training data by incorporating equivariance principles into the neural network model. The paper demonstrates the efficiency and accuracy of this method through systematic experiments in various systems, including monolayer magnets, curved nanotube magnets, and Moiré twisted bilayer magnets. The experimental results prove the excellent generalization ability of the method and provide possibilities for exploring magnetic phenomena in large-scale material systems. The challenges faced in this study include: (1) the increase in degrees of freedom in magnetic NNFFs, requiring a large amount of additional data for machine learning; (2) the lack of modeling of the total energy derivative with respect to changes in magnetic moment direction (i.e., magnetic forces), which is crucial for atomic kinetic simulations of magnetic materials. To address these issues, the paper proposes a deep learning framework called "MagNet," which combines the degrees of freedom of atoms and magnetic structures and directly maps magnetic forces, thereby achieving efficient and accurate learning of magnetic materials. MagNet's functions include predicting total energy, atomic forces, and magnetic forces, and its design takes into account the equivariance of three-dimensional Euclidean group transformations such as rotation, translation, and reflection. By leveraging these symmetries, the neural network can handle limited data more effectively while preserving the accuracy of magnetic material energy modeling. The paper demonstrates the efficiency and accuracy of MagNet in predicting magnetic dispersion, energy differences of magnetic CrI3 nanotubes, and spin dynamics in simulated Moiré twisted bilayer CrI3.

Equivariant Neural Network Force Fields for Magnetic Materials

Accelerating the electronic-structure calculation of magnetic systems by equivariant neural networks

Deep-learning electronic-structure calculation of magnetic superstructures

General time-reversal equivariant neural network potential for magnetic materials

A deep-learning method for studying magnetic superstructures

Time-reversal Equivariant Neural Network Potential and Hamiltonian for Magnetic Materials

Spin-Dependent Graph Neural Network Potential for Magnetic Materials

Machine Learning in Magnetic Materials

Machine learning magnetism classifiers from atomic coordinates

A Finite Element Based Real-Space Phase Field Model for Domain Evolution of Ferromagnetic Materials

Accelerated Data-Driven Discovery and Screening of Two-Dimensional Magnets Using Graph Neural Networks

Equivariant Neural Networks for Spin Dynamics Simulations of Itinerant Magnets

Predicting magnetic anisotropy energies using site-specific spin-orbit coupling energies and machine learning: Application to iron-cobalt nitrides

How MagNet: Machine Learning Framework for Modeling Power Magnetic Material Characteristics

Deep learning methods for Hamiltonian parameter estimation and magnetic domain image generation in twisted van der Waals magnets

Complex Spin Hamiltonian Represented by an Artificial Neural Network

Interatomic Interaction Models for Magnetic Materials: Recent Advances

Physics-Informed Machine-Learning Prediction of Curie Temperatures and Its Promise for Guiding the Discovery of Functional Magnetic Materials

Deep Learning Illuminates Spin and Lattice Interaction in Magnetic Materials

Magnetic Hamiltonian Parameter Estimation Using Deep Learning Techniques