Spin Hall effect (SHE) provides a promising solution to the realization of advantageous functionalities for spin-based recording and information processing. In this work, we conduct high-throughput calculations on the spin Hall conductivity (SHC) of antiperovskite compounds with the composition ZXM₃, where Z is a nonmetal, X is a metal, and M is a platinum group metal. From an initial database over 4500 structures, we screen 295 structurally stable compounds and identify 24 compounds with intrinsic SHC exceeding 500 (ℏ/e) (Ω⁻¹ cm⁻¹). We reveal a strong dependence of SHC on spin-orbit coupling-induced energy splitting near the Fermi level. In addition, SHCs can be regulated through proper doping of electrons or holes. The present work establishes high-throughput database of SHC in antiperovskites which is crucial for designing future electric and spintronic devices.

In the rolling production of steel, predicting the performance of new products is challenging due to the low variety of data distributions resulting from standardized manufacturing processes and fixed product categories. This scenario poses a significant hurdle for machine learning models, leading to what is commonly known as the “cold-start problem”. To address this issue, we propose a knowledge graph attention neural network for steel manufacturing (SteelKGAT). By leveraging expert knowledge and a multi-head attention mechanism, SteelKGAT aims to enhance prediction accuracy. Our experimental results demonstrate that the SteelKGAT model outperforms existing methods when generalizing to previously unseen products. Only the SteelKGAT model accurately captures the feature trend, thereby offering correct guidance in product tuning, which is of practical significance for new product development (NPD). Additionally, we employ the Integrated Gradients (IG) method to shed light on the model's predictions, revealing the relative importance of each feature within the knowledge graph. Notably, this work represents the first application of knowledge graph attention neural networks to address the cold-start problem in steel rolling production. By combining domain expertise and interpretable predictions, our knowledge-informed SteelKGAT model provides accurate insights into the mechanical properties of products even in cold-start scenarios.

To meet the demands of advanced electronic devices, inorganic glasses are required to have comprehensive dielectric, thermal, and mechanical properties. However, the complex composition–property relationship and vast compositional diversity hinder optimization. This study developed machine learning models to predict permittivity, dielectric loss, thermal conductivity, coefficient of thermal expansion, and Young’s modulus based on the composition features of inorganic glasses. The optimal models achieve R² values of 0.9614, 0.7411, 0.9454, 0.9684, and 0.8164, respectively. By integrating domain knowledge with model-agnostic interpretation methods, feature contributions and interactions were analyzed. The mixed alkali effect is crucial for property regulation, especially Na-K for dielectric loss and Na-Li for thermal conductivity. Boron anomaly shifts the high-λ region to a balanced composition of alkali metals with rising B%. The multiobjective optimization of properties was realized using a genetic algorithm framework. After 23 iterations, the optimal material in the MgO-Al₂O₃-B₂O₃-SiO₂ system exhibits ε_r = 4.78, tanδ = 0.00063, λ = 2.59 W/(m·K), α = 50.27×10⁻⁷K⁻¹, and E = 82.41 GPa, outperforming all materials in the dataset. The computational effort was reduced to 1/19 of that required using exhaustive search methods. This study provides a model interpretation framework and an effective multiobjective optimization strategy for glass design.

The detrimental Fe element in Al-Si cast alloy can be effectively removed by Fe-containing intermetallics separation. However, the formation temperature of Fe-containing intermetallics can be further improved to increase the removal efficiency of Fe element. The effects of the Cr/Mn atomic ratio on the stability, theoretical melting point, elastic modulus, and thermal properties were calculated with the aim of improving the stability of the α-Al(FeMnCr)Si phase. An increased Cr/Mn atomic ratio effectively increased the stability, theoretical melting point, elastic modulus, isobaric heat capacity, and reduced the volumetric thermal expansion coefficient of α-Al(FeMnCr)Si phase, which can be explained by the strengthened Al-Cr and Si-Cr chemical bonds. The experimental study results revealed that the formation temperature and Young's modulus of the α-Al(FeMnCr)Si phase increase from 673.0°C and 228.5 GPa to 732.0°C and 272.1 GPa with the Cr/Mn atomic ratio increasing from 0.11 to 0.8, which better validates the thermodynamic stability, theoretical melting point and elastic modulus calculation results. These results provide a new strategy for designing Fe-containing intermetallics with the desired properties, which contributes to guiding the development of high-performance recycled Al-Si alloys.

A small dataset of ~300 datapoints of zinc (Zn) alloys were collected and 125 entries containing alloying elements, extrusion parameters (temperature (ET), speed (ES) and ratio (ER)), and mechanical properties (yield strength (YS), ultimate tensile strength (UTS), and final elongation (EL)) were selected. Machine learning models were applied to predict mechanical properties, in which random forest (RF) model exhibited the best performance and further validated by a new experimental sample of Zn-0.05Mg-0.5Mn, with the mean absolute percentage error (MAPE) less than 10%. [Correction added on 13 May 2025, after first online publication: In the preceding sentence, the value ‘12%’ has been changed to ‘10%’.]. Moreover, an empirical formula was induced by the clustering model (CL). Control over strain softening/hardening behavior was achieved through only process parameter adjustment. Finally, by combining multi-objective genetic algorithm and RF models, the optimization alloy composition and extrusion parameters was carried out, targeting high-strength, strength/plasticity synergy, and high plasticity for biodegradable purpose. A notable optimized scheme for strength/plasticity synergy in Zn-0.20Mg-0.60Mn (wt.%) achieves the YS of 303 MPa, UTS of 354 MPa, and EL of 25.1% with the MAPE less than 10%, and exhibits the strain-hardening response, associated with ER of 16, ET of 170°C, and ES of 3.21 mm/s. [Correction added on 13 May 2025, after first online publication: In the preceding sentence, the value ‘345 MPa’ has been changed to ‘354 MPa’. and the value ‘3.33 mm/s’ has been changed to ‘3.21 mm/s’].

Using first-principles-based machine-learning potential, molecular dynamics (MD) simulations are performed to investigate the micro-mechanism in phase transition of NbO₂. Treating the DFT results of the low- and intermediate-temperature phases of NbO₂ as training data in the deep-learning model, we successfully constructed an interatomic potential capable of accurately reproducing the phase transitions from low-temperature (pressure) to high-temperature (pressure) regimes. Notably, our simulations predict a high-pressure monoclinic phase (>14 GPa) without treating its information in the training set, consistent with previous experimental findings, demonstrating the reliability of the constructed interatomic potential. We identified the Nb-dimers as the key structural motif governing the phase transitions. At low temperatures, the displacements of the Nb-dimers drive the transition between the I4₁/a (α-NbO₂) and I4₁ (β-NbO₂) phases, while at high temperatures, Nb ions are prone to being equally distributed and the disappearance of Nb-dimers leads to the stabilization of a high-symmetry P4₂/mnm phase. These findings elucidate the structural and dynamical mechanisms underlying the structural properties of NbO₂ and highlight the utility of combining DFT and deep potential MD methods for studying complex phase transitions in transition metal oxides.

Antiferroelectric (AFE) materials have received great attention because of their potential applications in the energy sector. Nevertheless, the properties of AFE materials have not been explored for a long time, especially the atomic-scale understanding of AFE domain walls. Here, using first-principles-based machine learning potentials, we identify the atomic structures, energies, and dynamic properties of the domain walls for AFE lead zirconate. It is found that the domain wall can reduce the critical antiferroelectric-ferroelectric transition field. During the electric field-driven polarization switching process, the domain wall is immobile. Importantly, we observe that a distinct domain structure spontaneously forms in bulk lead zirconate upon annealing at 300 K. The domain structure exhibits an alternating array of clockwise–anticlockwise vortexes along radial with continuous polarization rotation. This anomalous AFE vortex is derived from the energy degeneracy in four possible orientations of the polarization order, which can enhance the dielectric response in the terahertz. The current results give an implication for the emergence of AFE vortex in AFE materials as well as ferroelectric materials.

Large language models (LLMs) have demonstrated effectiveness in interpreting complex data. However, they encounter challenges in specialized applications, such as predicting material properties, due to limited integration with domain-specific knowledge. To overcome these challenges, we introduce MatAgent, an artificial intelligence (AI) agent that combines computational chemistry tools, such as first-principles (FP) calculations, with the capabilities of LLMs to predict key properties of materials. By leveraging prompt engineering and advanced reasoning techniques, MatAgent integrates a series of tools and acquires domain-specific knowledge in the field of material property prediction, enabling it to accurately predict the properties of materials without the need for predefined input structures. The experimental results indicate that MatAgent achieves a significant improvement in prediction accuracy and efficiency. As a novel approach that integrates LLMs with FP calculation tools, MatAgent highlights the potential of combining advanced computational techniques to enhance material property predictions, representing a significant advancement in computational materials science.

The development of cost-effective noble-metal-free cocatalysts with exceptional hydrogen evolution reaction (HER) activity is critical for advancing scalable and sustainable photocatalytic hydrogen production. Although platinum (Pt) remains a benchmark HER catalyst, its scarcity and high cost stimulates the search for viable alternatives. In this work, a machine learning (ML)-accelerated strategy is presented to screen highly active ternary CrNiCu alloys. Combining with density functional theory calculations, XGBoost regression models were trained to predict hydrogen adsorption energies and water dissociation energy barriers on CrNiCu alloy surfaces. Consequently, the theoretical exchange current densities were predicted for all possible compositions of CrNiCu alloys, enabling the identification of alloy catalysts with composition of 10~30 at.% Cr, 30–50 at.% Ni, and 20–60 at.% Cu that exhibits superior HER activity than Pt. Stability assessment of optimal ternary CrNiCu alloys further confirms their excellent resistance to element segregation and hydroxyl poisoning under operational conditions. This work not only identifies promising ternary CrNiCu alloys of non-noble HER catalysts but also establishes an efficient ML-accelerated computational framework for the discovery of durable high-activity alloys for renewable energy applications.

Determining thermodynamic properties in disordered systems remains a formidable challenge because of the difficulty in incorporating nuclear quantum effects into large-scale and nonperiodic atomic simulations. In this study, we employ a machine learning deep potential model in conjunction with the quantum thermal bath method, enabling machine learning molecular dynamics to simulate thermodynamic quantities of liquid materials with satisfactory accuracy without significantly increasing computational costs. Using this approach, we accurately calculate the variations in various thermodynamic quantities of liquid metal gallium at temperatures ranging from zero to room temperature. The calculated thermodynamic properties accurately capture the solid-liquid phase transition behavior of gallium, whereas classical molecular dynamics methods fail to reproduce realistic results. Through this approach, we offer a potential method for accurately calculating the thermodynamic properties of liquids and other disordered systems.

PyGAMD (Python GPU-accelerated molecular dynamics software) is a molecular simulation platform developed from scratch. It is designed for soft matter, especially for polymer by integrating coarse-grained/multi-scale models, methods, and force fields. It essentially includes an interpreter of molecular dynamics (MD) which supports secondary programming so that users can write their own functions by themselves, such as analytical potential forms for nonbonded, bond, angle, and dihedral interactions in an easy way, greatly extending the flexibility of MD simulations. The interpreter is written by pure Python language, making it easy to be modified and further developed. Some built-in libraries written by other languages that have been compiled for Python are added into PyGAMD to extend it's features, including configuration initialization, property analysis, etc. Machine learning force fields that are trained by DeePMD-kit are supported by PyGAMD for conveniently implementing multi-scale modeling and simulations. By designing an advanced framework of software, graphics processing unit-acceleration achieved by the Numba library of Python and compute unified device architecture reaches a high computing efficiency.

Refractory high-entropy alloys (RHEAs) represent a promising class of structural materials with significant potential for various applications. However, their limited plasticity at room temperature restricts their deformability, posing challenges for processing and industrial implementation. Traditional experimental methods for characterizing this property are time-consuming and resource-intensive, necessitating the development of efficient predictive models. In this study, we propose a machine learning approach to predict the fracture strain of RHEAs. A dataset comprising 128 RHEAs fracture strain samples is compiled from the literature and classified into two categories: “high plasticity” and “low plasticity.” Through feature selection techniques, a critical subset of features is identified, enabling a support vector classification model to achieve 96% prediction accuracy. Additionally, an interpretable machine learning algorithm is employed to derive explicit functional expressions describing the relationship between key features and fracture strain, achieving 88% accuracy. Although slightly less accurate, it provides valuable insights into the underlying mechanisms, making it a useful tool for materials design and optimization.