Although first-principles calculations offer high precision, they are prohibitively expensive. Conversely, molecular dynamics simulations employing classical interatomic potentials, or force fields, offer quicker but less precise outcomes. To balance between computational speed and accuracy, machine learning (ML) potential functions have been developed and have gained widespread application. The deep potential (DP) method, a type of ML potential, has attracted considerable attention recently. This paper provides a comprehensive review of DP methods in materials science. It begins with an introduction to the theoretical foundation of DP, followed by a detailed exposition of the DP model development and usage. Additionally, the application of DP in various material systems is briefly reviewed. AIS-Square contributes training databases and workflows essential for developing DP models. The paper concludes by assessing the performance of DP models relative to both first-principles calculations and classical potentials in terms of accuracy and efficiency. Finally, a brief outlook on future developments trends is provided.
Ferroelectric materials, which are characterized by tunable spontaneous polarization, show remarkable application potential for nonvolatile information storage; however they present various challenges. The performance of these materials is strongly influenced by their dynamic polarization behavior under multiple external fields. Due to the limited precision of experimental observations, precise atomic-level material simulations are crucial. Although molecular dynamics (MD) offers an ideal method for investigating material dynamics over a wide spatiotemporal range, its application to new materials is often limited by challenges such as low accuracy, complex development, and limited portability of conventional classical force fields. Advances in machine learning have provided new possibilities for developing force fields. Among different machine learning potentials, deep potential (DP) based on deep neural networks stands out. DP offers accuracy comparable to that of density functional theory while providing computational efficiency similar to that of conventional classical force fields. This review primarily focused on the development and application of DP in ferroelectric materials, specifically examining the phase transition mechanisms and polarization reversal processes at the atomic scale. Considerable efforts have been made to develop and evaluate DP for crucial ferroelectric materials such as hafnium dioxide (HfO2) and classic perovskite ferroelectric materials. Furthermore, this review explores the high oxygen-ion migration kinetics in HfO2 using DP and investigates the flexoelectricity induced by polar domain boundaries and the bulk photovoltaic effects in strontium titanate. By highlighting the use of DP molecular dynamics approaches in ferroelectric materials, this review emphasizes the role of machine learning approaches in optimizing and accelerating material simulations to facilitate further breakthroughs and discoveries.
Solidification nucleation is an everlasting research topic in the fields of materials science and condensed matter physics. Molecular dynamics (MD) and enhanced sampling methods provide a powerful means to observe the microscopic mechanisms of solidification processes in situ at the atomic level and to analyze the thermodynamic and kinetic properties of phase transitions. Recent advancements in MD simulations, particularly those incorporating machine learning (ML) techniques, have remarkably advanced our understanding of nucleation across different systems. This paper first reviews the basic theory of solidification nucleation and introduces common methods used in solidification nucleation simulation studies. It then delves into the application of ML techniques in three key areas: force fields, enhanced sampling, and order parameters. The paper further highlights several representative systems to demonstrate the practical applications of these methods. Finally, a summary and outlook on the future of ML-assisted MD simulations for studying material solidification were provided.
With the rapid advancement of artificial intelligence (AI), machine learning is playing an increasingly important role in materials research, development, and design. Traditional machine learning models are often “black box” models that limit researchers' understanding of a model's decision-making and undermines their confidence in the process. Explainable machine learning (XML) can reveal the internal mechanisms of these models and provide insights into their decision-making processes. This study begins with the fundamentals of XML, outlines the development history and notable milestones of XML methods, and discusses the role of XML in AI, emphasizing the Fairness, Accountability, Simplicity, and Transparency (F.A.S.T.) principles that should be followed. Furthermore, this study introduces two major categories of XML methods—those that use model-intrinsic interpretability and those that use external model interpretability—along with their applications in materials science. Specifically, the symbolic regression of XML and visualized XML methods developed by our team offer new tools for materials research and design. Finally, potential directions for XML in the field of materials science are discussed.
In the era of big data, the demand for data storage and processing is increasing because of advanced technologies such as artificial intelligence (AI), 5G, and cloud computing. Emerging non-volatile memory materials and devices present remarkable opportunities to enhance computing capacity. Concurrently, the AI-driven scientific research paradigm introduces a new mode for improving device performance. This review focuses on recent advances in phase-change memory materials and devices, emphasizing computational- and data-driven methodologies. Phase-change materials (PCMs) operate based on rapid and reversible phase transitions between amorphous and crystalline states, where differences in electrical and optical properties are used to encode digital information. These materials typically consist of multicomponent alloys, with phase transitions involving melting, quenching, crystallization, glass relaxation, and crystal-crystal structural changes. To achieve a detailed atomistic understanding of PCMs, large-scale density functional theory (DFT) and DFT-based ab initio molecular dynamics (AIMD) simulations are essential. Comparisons between DFT/AIMD simulations and experimental results have clarified many fundamental aspects of PCM. The first part of this review provides an overview of the history and progress in large-scale ab initio simulations of PCMs. With atomic-scale knowledge, rational materials design becomes feasible. The second part explores methods for developing new PCMs with specific properties, such as accelerating crystallization at elevated temperatures while maintaining non-volatile characteristics at room temperature. High-throughput screening's role in discovering new phase change alloys is also discussed. In the third part, we examine multiscale and cross-scale simulations of PCM for various optical and electronic phase change applications. By computing the dielectric functions of PCM during the amorphous-to-crystalline transition, we can track changes in the refractive index and extinction coefficient across visible and infrared spectra over time. These DFT-computed parameters inform coarse-grained device simulations using finite-difference time-domain (FDTD) or finite element method (FEM). Based on these multiscale simulations, we offer optimization guidelines for non-volatile color display and photonic waveguide devices. The machine learning potentials address some performance gaps between the DFT/AIMD and FEM/FDTD calculations. Machine-learning-driven molecular dynamics (MLMD) simulations serve as cross-scale simulations, with recent developments including neural networks, graph convolutional neural networks, and Gaussian approximation potentials. We discuss the role of MLMD in enabling device-scale atomistic simulations, facilitating device design and optimization with atomic-scale information. Finally, we outline future opportunities and challenges in theoretical PCM research. With ongoing AI-driven fundamental research, we anticipate the commercialization of high-performance phase change memory, neuroinspired computing, and reconfigurable nanophotonic devices, which will, in turn, foster the development of more advanced theoretical tools for research.
Entropy is an important concept in science and is ubiquitous from quantum to astronomy. By integrating statistical mechanics, quantum mechanics, and thermodynamics, Professor Zi-Kui Liu proposed the zentropy theory, which stacks entropy over configurations. The zentropy theory takes the configurations in Gibbs' statistical mechanics of a given ensemble as the material gene with the ground state as the basic configuration and additional configurations ergodically derived from its internal degrees of freedom. In the zentropy theory, the total entropy of a system is defined as the weighted average of the entropy of each configuration plus the statistical entropy among all configurations. In this paper, the basic equations and principles of the zentropy theory are introduced, and their typical applications, including magnetic and ferroelectric transformations, thermal expansion mechanisms, and critical phenomenon prediction are outlined. Furthermore, a perspective on the development of this theory, software ecosystems, high-throughput computing, and integration with artificial intelligence is provided in this study.
A comprehensive understanding of the microscopic mechanisms underlying phase transitions and deformations in metallic materials is crucial for developing new materials that meet the nation's essential needs. Molecular dynamic simulation techniques, particularly those powered by machine-learning molecular force fields, are emerging as potent tools for unraveling atomic-scale phenomena. In this study, recent advancements in machine-learning molecular force fields were reviewed to investigate metallic phase transitions and deformations. First, the fundamental principles and evolution of machine-learning molecular force fields were introduced. Then, the phase transformation and deformation of metals were examined, providing insights into the kinetics of phase transitions and microscopic mechanisms. Finally, the challenges faced by current machine-learning molecular force fields in studying metallic phase transformations and deformations were identified, and a glimpse into future research directions was discussed.
Density functional theory (DFT), grounded in the fundamental principles of quantum mechanics, effectively predicts material properties and is now widely used across various research disciplines such as physics, chemistry, materials science, and biology. As research in materials science advances, there is an urgent need to further enhance the accuracy and efficiency of DFT. However, improving accuracy and efficiency is often challenging because these goals can be mutually exclusive. Recently, guided by the concept of AI for science, deep learning-based electronic structure calculation methods have rapidly emerged, offering potential solutions to resolve this accuracy-efficiency dilemma. Nonetheless, developing a stable and reliable DFT software platform remains a substantial challenge in exploring and expanding the use of AI-assisted methods on a broader scale. This paper introduces the open-source DFT package ABACUS (atomic-orbital based ab-initio computation at UStc), focusing on its physical models, deep learning algorithms, and software development aspects. The present discussion emphasizes the progress of the open-source package, highlighting its integration with deep learning algorithms and its evolution from version 2.2 (released in April 2022) to version 3.7 (released in July 2024).
The discovery of hydride superconductors with high critical transition temperature (Tc) under high pressures has received considerable interest in developing superconducting materials that can operate at room temperature and ambient pressure. Although first-principles methods can accurately predict the critical temperature of hydride superconductors, the computational demands are significant because of the expensive calculation of electron-phonon coupling. Hence, constructing an accurate and efficient model for predicting Tc is highly desirable. In this study, a simple and interpretable machine learning (ML) model was developed using the random forest algorithm, which enables the selection of important features based on their importance. Using four physics-based features, namely, the standard deviation of the number of valence electrons, mean covalent radii, range of the Mendeleev number of constituent elements, and hydrogen fraction of the total density of states at the Fermi energy, the optimal ML model achieves high accuracy, with a mean absolute error of 24.3 K and a root-mean-square error of 33.6 K. The ML model developed in this study shows great application potential for high-throughput screening, thereby expediting the discovery of high-Tc superconducting hydrides.
The knowledge regarding the structure-activity relationships of nickel-based single crystal superalloys is mainly stored in the form of unstructured text in the vast published scientific literature, and its effective utilization can accelerate the design of high-performance materials. Named entity recognition (NER) methods can be used to extract vital information from unstructured text, thus contributing to automatically achieving tedious text mining tasks. However, existing NER methods typically rely on a large amount of corpus data, especially of the deep-learning-based type, and can hardly tackle cross-domain tasks. To the best of our knowledge, no prior research has been conducted for the knowledge discovery of nickel-based single crystal superalloys based on deep-learning-based NER; thus, it is difficult to adapt existing methods to this field. Here, a semantic-features-fused NER (SF-NER) method based on deep learning was proposed, aiming to accurately extract knowledge from abstract text concerning nickel-based single crystal superalloys. Specifically, as data quality determines the performance of NER models, a high-quality annotated corpus dataset for the above-mentioned superalloys (containing 19405 entity data of eight entity types) was constructed. This was created via remote supervision using domain-specific materials dictionary under the domain knowledge's guidance. To accurately capture the terms related to specific materials from the high-quality corpus dataset, a encoding fusion strategy for word representation was proposed for encoding the essential semantic features of materials from various perspectives. Then, based on these semantic features, a deep learning model, i.e., bidirectional long short-term memory-cenditional random field (Bi-LSTM-CRF), was built to capture key semantic information in sentence sequences, thus accurately predicting entity types. The results of the experiment demonstrated that the proposed SF-NER method could accurately distinguish the entity categories of nickel-based single crystal superalloys (i.e., F1 = 0.84) and effectively identify the key factors influencing their service performance. Lastly, descriptors with high importance were recommended, as they can be employed for machine learning modeling to explore the structure-activity relationships of high-performance materials.
Exploring ultrafast structural transitions in materials triggered by femtosecond laser pulses—from their condensed states to high-temperature, high-pressure conditions, and potentially to ideal plasmas—is a crucial scientific endeavor with profound implications for fields such as inertial confinement fusion, metal additive manufacturing, and laser processing. These extreme conditions, which are challenging to replicate and directly observe in experiments due to temporal and spatial resolution limitations, require theoretical models and simulations to decode the underlying microscopic mechanisms. Molecular dynamics (MD) simulations, especially when paired with advanced potential energy surfaces, are effective tools for addressing these challenges. However, maintaining a balance between computational efficiency and physical accuracy, particularly when simulating excited states induced by laser interactions, remains a formidable task. In this context, neural network potential energy surfaces (NNPES) have demonstrated exceptional capability for capturing the complex interactions and properties of materials under extreme conditions, providing vital links between quantum mechanics and macroscale phenomena. Using Cu as a prototypical example, the ability of NNPES to accurately depict lattice vibrations, thermophysical properties, and complex dynamics during laser-matter interactions has been demonstrated. By seamlessly integrating NNPES with a two-temperature MD model, this study directly simulates the atomic-scale dynamics of Cu thin films subjected to intense pulsed laser irradiation. This innovative approach, which combines quantum-level accuracy with large-scale thermodynamics and detailed microstructural evolution, provides unprecedented insights into the fundamental mechanisms of laser-induced melting. Our findings reveal two distinct melting behaviors in Cu, dependent on laser fluence. At fluences near the melting threshold, a heterogeneous melting process initiated at the film surface because of the lower free energy barrier was observed. The solid-liquid interface then moves inward at velocities of tens of meters per second, requiring hundreds of picoseconds for melting to complete. Conversely, at fluences well above the threshold, Cu films experience rapid and homogeneous melting, markedly different from conventional heating-induced melting. Here, the lattice temperature almost instantaneously exceeds the thermal stability limit, leading to uniform liquid nucleation and rapid growth throughout the film, culminating in complete melting within just tens of picoseconds. This study not only illuminates the atomic-scale dynamics of laser-induced melting but also underscores the transition from heterogeneous to homogeneous melting mechanisms as a function of laser fluence. This study serves as an invaluable research tool for enhancing our understanding of laser-matter interactions and their potential applications in optimizing laser-based manufacturing processes and predicting material behavior under extreme conditions. Moreover, the reliability and versatility of NNPES set the stage for extending the research to more complex systems, including alloys and amorphous materials. This expansion fosters robust connections between microscopic theories and macroscale applications, deepening our understanding of material responses to intense laser irradiation. Future studies employing this framework could explore complex physical phenomena such as explosive boiling and material disintegration during laser ablation, offering unique atomic-scale insights that could pave the way for groundbreaking discoveries and technological advancements.
