Abstract:Atomistic simulations hold significant value in clarifying crucial phenomena such as phase transitions and energy transport in materials science. Their success stems from the presence of potential energy functions capable of accurately depicting the relationship between system energy and lattice changes. In magnetic materials, two atomic scale degrees of freedom come into play: the lattice and the spin. However, accurately tracing the simultaneous evolution of both lattice and spin in magnetic materials at an atomic scale is a substantial challenge. This is largely due to the complexity involved in depicting the interaction energy precisely, and its influence on lattice and spin-driving forces, such as atomic force and magnetic torque, which continues to be a daunting task in computational science. Addressing this deficit, we present DeepSPIN, a versatile approach that generates high-precision predictive models of energy, atomic forces, and magnetic torque in magnetic systems. This is achieved by integrating first-principles calculations of magnetic excited states with deep learning techniques via active learning. We thoroughly explore the methodology, accuracy, and scalability of our proposed model in this paper. Our technique adeptly connects first-principles computations and atomic-scale simulations of magnetic materials. This synergy presents opportunities to utilize these calculations in devising and tackling theoretical and practical obstacles concerning magnetic materials.

What problem does this paper attempt to address?

The main objective of this paper is to address the precise modeling of spin-lattice interactions in magnetic materials. Specifically, the researchers have developed a method called DeepSPIN, which is a deep learning-based technique aimed at accurately predicting the energy, atomic forces, and magnetic moment forces (magnetic torques) in magnetic systems. By combining first-principles calculations and deep learning techniques, especially utilizing active learning strategies to optimize the dataset, DeepSPIN can efficiently describe the interactions between non-collinear magnetic configurations and complex lattice structures. Key points mentioned in the paper include: - **Problem Background**: Traditional methods such as first-principles calculations are limited by scale, while phase-field methods may lack spatial and energy resolution. Existing atomic-scale simulation techniques often focus on either lattice or spin changes, making it difficult to handle both simultaneously. - **Solution**: The DeepSPIN method is proposed, which uses the concept of "pseudo-atoms" to map lattice and spin configurations into interdependent features, and designs specific neural network structures and loss functions to capture these interactions. - **Technical Details**: Constructing a cost-minimized dataset through active learning strategies; using deep neural networks to predict the local atomic energy corresponding to arbitrary lattice and spin configurations; designing loss functions to balance the accuracy of energy, atomic force, and magnetic moment force predictions. - **Validation and Application**: The effectiveness and accuracy of DeepSPIN are demonstrated through cases of two antiferromagnetic insulating materials, NiO and BiFeO3. Additionally, the method's performance in predicting lattice dynamics (e.g., phonon spectra) and spin dynamics (e.g., magnon spectra) is validated, as well as its applicability in complex crystal structures, such as high-angle NiO grain boundary systems and the polarization reversal process in BiFeO3. In summary, DeepSPIN is a general model capable of accurately describing spin-lattice coupling phenomena in magnetic materials, with significant implications for both theoretical and practical applications.

Deep Learning Illuminates Spin and Lattice Interaction in Magnetic Materials

Learning Spin Liquids on a Honeycomb Lattice with Artificial Neural Networks.

A deep-learning method for studying magnetic superstructures

Deep-learning electronic-structure calculation of magnetic superstructures

Synergistic Optimization of Lattice and Spin in Atomic Scale Simulation of Magnetic Materials

Magnetic Hamiltonian Parameter Estimation Using Deep Learning Techniques

Spin-Dependent Graph Neural Network Potential for Magnetic Materials

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

Large-Scale Simulation of the Spin-Lattice Dynamics in Ferromagnetic Iron

SPILADY: A Parallel CPU and GPU Code for Spin–lattice Magnetic Molecular Dynamics Simulations

Spin-lattice-electron Dynamics Simulations of Magnetic Materials

Application of deep learning methods to the study of magnetic phenomena

Exploring lattice thermal conductivity models via interpretable deep learning to accelerate the discovery of novel materials

Machine Learning Magnetic Parameters from Spin Configurations

Non-equilibrium dynamics of spin-lattice coupling

Complex Spin Hamiltonian Represented by an Artificial Neural Network

Advancements in High‐Throughput Screening and Machine Learning Design for 2D Ferromagnetism: A Comprehensive Review

Quantifying the Magnetic Interactions Governing Chiral Spin Textures Using Deep Neural Networks

Emergent Dynamics of Artificial Spin-Ice Lattice Based on an Ultrathin Ferromagnet.

The prediction of single-molecule magnet properties via deep learning

A Self-Adaptive First-Principles Approach for Magnetic Excited States