Infrared optical materials are critical for numerous applications, yet accurately characterizing their intrinsic optical properties remains challenging. Traditional theoretical approaches—ranging from empirical molecular dynamics to first-principles methods like density functional perturbation theory (DFPT)—face trade-offs between accuracy and computational cost, particularly for complex or low-symmetry material systems. Here, we tackle these challenges by introducing a fast and accurate infrared spectroscopy computational framework using machine learning interatomic potentials. By leveraging machine-learned interatomic forces, this method bypasses costly higher order DFPT calculations, enabling rapid extraction of phonon vibrational parameters. These parameters are then integrated into infrared-active vibration models to compute dielectric functions and infrared optical properties. Validated across diverse materials, our proposed framework demonstrates broad applicability while achieving a drastic reduction in computational cost compared to conventional methods. This framework bridges the gap between experimental characterization and theoretical predictions, offering a scalable tool for high-throughput screening and design of infrared optical materials.

Adsorption on a solid surface is a significant chemical process in the fields of gas sensors, solid catalysts, hydrogen storage materials, and ion batteries. Here, we develop a high-throughput computing package, termed as gas sensors and catalysts automatically screening package (GASCAP), to accelerate the evaluation of adsorption on solid surfaces using integrated computational materials engineering. The aims of GASCAP are to detect unequal adsorption sites, construct coadsorption structures, analyze adsorption energies, calculate work functions, and clarify charge interaction in high-throughput ways. The regulation of CO adsorption on the Pt (111) surface is used as a benchmark to demonstrate the effectiveness of GASCAP. Additionally, the GASCAP is interfaced with the machine learning interatomic potentials (MILP), to accelerate the adsorption energy computations. The calculated results reveal that the MILP can effectively accelerate the adsorption energy screening at 220 times when the calculation accuracy is reliable. To expand the application, a database is built with 5914 adsorbates and substrates. Considering the fast development of high-throughput calculations, the GASCAP will be a promising simulation platform for the future development in solid surface science.

The rapid advancement of large model technology in recent years has ushered in new opportunities for materials science. Leveraging their powerful feature learning capabilities, emergent properties, and flexible fine-tuning mechanisms, large models are gradually being applied to all aspects of materials science research. However, we contend that the deep integration of artificial intelligence and materials science urgently requires a transition from “virtual intelligence assistance” to “embodied intelligence dominance”. Future materials discovery will be autonomously executed in closed-loop operations within physical environments by Embodied AI Chemists. This review first introduces the blossom of large model technology and its multifaceted applications in various materials science-related topics. Subsequently, this review specifically maps the developmental pathways of autonomous experimental platforms, underscoring the comparative advantages of large-model-based systems over traditional machine learning approaches. Furthermore, it discusses the potential of embodied large models in materials science and proposes potential applications by constructing diverse training datasets, enhancing embodied reasoning capabilities, establishing intelligent collaborative environments, and developing multi-agent collaborative frameworks. Through this comprehensive analysis, we aim to pioneer new intelligent pathways for materials science research, motivating an end-to-end intelligent transformation from theoretical exploration to experimental realization.

Low stacking fault energy (SFE) CoCrFeNiMn-based high entropy alloys (HEAs) have garnered widespread attention due to their excellent mechanical properties. These outstanding mechanical properties result from multiple deformation mechanisms during tensile deformation, such as stacking faults, deformation twinning, and martensitic transformation. However, the vast and complex compositional space presents a significant challenge for the design of low SFE HEAs. To address this issue, this study developed an interpretable machine learning (ML) ensemble algorithm framework that integrates three high-accuracy ML models (multilayer perceptron regressor, support vector regressor, extreme gradient boosting regressor, R² > 0.9). In the alloy composition screening stage, the Valence Electron Concentration (VEC) and the proposed ML scoring parameter (Score = A*Mean + B*Std) were employed to constrain the phase composition and screen for low SFE alloy compositions. Ultimately, multiple No-BCC phase CoCrFeNiMn-based HEAs with twinning-induced plasticity/transformation-induced plasticity effects were successfully designed. To overcome the challenge of insufficient model accuracy in data-driven design, correlation-based and importance-based feature selection methods were combined. This approach efficiently processed additional descriptors generated from atomic compositions, improving model accuracy by 13%. Furthermore, the Shapley additive explanation method revealed the influence of individual elements on the SFE, providing valuable guidance for designing low-SFE HEAs.

The development of magnesium alloy gigacastings puts forward higher demand on alloy fluidity because of more complex structure and longer filling distance. The fluidity is closely connected with atomic diffusion and melt viscosity. This work investigates the melt structure, self-diffusion coefficient, and viscosity of the Mg‒6wt.%Al alloy at different temperatures through molecular dynamics simulations. The melt structure is characterized by pair distribution functions, coordination numbers, H‒A index, and Voronoi polyhedron analysis. The self-diffusion coefficient is determined by the mean squared displacement method, whereas the viscosity is evaluated through the Green–Kubo equation and the Muller-Plathe algorithm. By establishing relationships among structure, diffusion, and viscosity, the microscopic mechanisms behind the temperature-induced changes in transport properties are revealed, in which the change of the degree of order at different temperatures is highlighted. The findings provide an atomic-scale theoretical basis for understanding the rheological behavior of magnesium alloy melt.

The discovery of topological materials is severely hampered by fragmented research workflows that cause information loss, inconsistent reasoning, and frequent computational failures. To overcome these barriers, we present TopoMAS, an interactive multi-agent framework that unifies the entire discovery pipeline through human–AI collaborative intelligence. TopoMAS seamlessly integrates natural language processing, knowledge retrieval from literature and databases, crystal structure generation, and automated first-principles validation. At its core, is a multi-level reasoning and coordination mechanism coupled with a self-refining knowledge graph. This architecture enhances query understanding and ensures computational robustness by adaptively allocating tasks, monitoring execution, and recovering from failures. In collaboration with human experts, TopoMAS has accelerated the identification of candidate topological phases and successfully guided the discovery of new materials. Benchmark evaluations show that TopoMAS's coordinated intelligence enables smaller, more efficient models to rival or even surpass the performance of substantially larger counterparts at a fraction of the computational cost. Ultimately, TopoMAS offers not only a powerful accelerator for materials research but also a transferable blueprint for building next-generation, AI-augmented discovery platforms across scientific disciplines.

Additive friction stir deposition (AFSD) is an effective method for fabricating high-performance deposits, with process parameters directly influencing the mechanical properties of the resulting samples. In this study, three machine learning models, that is, multilayer perception (MLP), radial basis function (RBF), and back propagation (BP), are developed to predict the ultimate tensile strength (UTS) and elongation (EL) of AFSD Al2219 samples. The input variables include set parameters (rotation speed, traverse speed, layer thickness, and the presence or absence of a preheating system as well as data obtained from an in situ process monitoring kit (temperature, feedstock force, deposition interface force, and deposition interface torque)). Results show that the BP-trained neural network provides the best fit to the experimental data, achieving the highest coefficient of determination (R² = 0.821 for UTS and 0.817 for EL), the lowest mean absolute error (MAE = 8.692 for UTS and 1.003 for EL), and the lowest root mean square error (RMSE = 13.773 for UTS and 1.266 for EL). These findings demonstrate the effectiveness and advantages of BP-trained neural networks in predicting mechanical properties based on various input parameters. Recommendations are provided on how the prediction model can be applied in the field of additive manufacturing.

The fundamental mechanisms of solute segregation and their impacts on material properties remain elusive, primarily due to the complexity and computational challenges in modeling. To address this, we present a specialized GPU implementation of highly efficient hybrid Monte Carlo and molecular dynamics (MCMD) algorithms in the open-source GPUMD package. Using this efficient MCMD approach, combined with a general-purpose machine-learning-based neuroevolution potential for 16 elemental metals and their alloys, we simulate the segregation of 15 solutes in polycrystalline Al. Our results reveal distinct segregation patterns for these solutes (Ag, Al, Au, Cr, Cu, Mg, Mo, Ni, Pb, Pd, Pt, Ta, Ti, V, W, Zr) in polycrystalline Al. We further investigate the impact of solutes on the strength of polycrystalline Al, analyzing the mechanisms of solute strengthening and embrittlement at the atomistic level. Our findings indicate the critical roles of grain boundaries cohesion and the nucleation and movement of Shockley dislocations in determining the material's strength. We anticipate that our efficient GPU-accelerated MCMD implementation in GPUMD, along with the insights into solute segregation behavior in polycrystalline Al, will be valuable for the design of Al alloys and other multi-component materials, including medium-entropy materials, high-entropy materials, and complex concentrated alloys.

AlN and its alloys are the state-of-the-art materials for deep-ultraviolet (DUV) light emitting diodes (LEDs). The limited number of materials known acts as a bottleneck for the design of efficient DUV light emitters. Previous computational screening of DUV light emitters has yielded a few promising candidates without rare-earth elements in light of the fabrication cost. However, it remains unclear if there exist any rare-earth compounds that may efficiently emit DUV light. Here, based on a high-throughput computational screening, we identify two promising DUV light emitters (Na₃TmBr₆ and Na₃LuBr₆) with potentially comparable performance with AlN. Our analysis reveals that hole effective masses in these rare-earth materials critically depend on the hybridization between rare-earth 4f orbitals and anion p orbitals, providing a valuable framework for targeted optimization. This work thus constitutes a basis for developing novel rare-earth-based DUV LEDs with expanded material diversity.

The graded density impactor (GDI) dynamic loading technique serves as a crucial method for achieving controllable stress/strain-rate loading, where the loading velocity and adaptability of GDI structural design critically govern the loading results. Numerical simulations of layer-wise modulation and shock wave transmission revealed a decoupling mechanism for stress and strain-rate parameters. Specifically, the loading velocity determines the overall magnitude, whereas variations in interlayer thickness modulate the specific strain-rate loading path. Building on this, a branched convolutional neural network (CNN)-bidirectional long short-term memory model (BLSTM) is developed to simultaneously predict stress/strain-rate curves achieving R² = 0.95 and loading velocity achieving R² = 0.99 while enabling GDI thickness design. This methodology resolves multi-physics coupling challenges in curve prediction and offers solutions for time-dependent issues in extreme conditions.

In the development of advanced nuclear fuels, we investigated how transition metals (TM = Nb, Ta, Zr) affect the thermodynamic and electronic properties of thorium monocarbide (ThC) using first-principles calculations. We modeled six ternary carbide compositions (Th_1−xTM_x)C with x = 0.1 and 0.2 to predict key properties including heat capacity, entropy, Gibbs free energy, and bulk modulus across 0 K–1800 K. Results show that (Th_1−xTa_x)C has the lowest equilibrium energy and smallest volume, whereas (Th_1−xZr_x)C maintains the highest thermal conductivity and mechanical rigidity. Notably, (Th_0.8Ta_0.2)C exhibits significant phonon scattering and structural softening. Bonding charge density analysis reveals strong covalent Nb-C bonds and intensive Zr-C interactions, providing critical atomic-level insights to accelerate development of next-generation Th-based nuclear fuels.

This study has developed a physically interpretable machine learning framework for predicting coercivity of Sm-Co-based alloys by integrating principles of permanent magnetic materials. Key features governing coercivity were systematically reconstructed using a developed two-step symbolic regression algorithm combining frequency statistics, and individual contributions of these reconstructed features were elucidated by sensitivity analysis. A high-throughput predictive model was set up for coercivity evaluation with exceptional accuracy enabling data-driven composition design of Sm-Co-based permanent magnetic alloys with high coercivity. Taking SmCo₇-based alloys as an example, ternary doping with Ti, In, and Al was identified as optimal for coercivity enhancement. Guided by these predictions, novel multielement doped nanocrystalline Sm-Co-based alloys were prepared exhibiting record high coercivity. This work established a paradigm shift from empirical optimization to mechanism-guided data-driven design of advanced permanent magnetic materials, demonstrating the potential of interpretable machine learning in materials innovation.

Key phase diagram studies of Li₂O–V₂O₅, K₂O–V₂O₅, and Rb₂O–V₂O₅ systems were conducted using X-ray diffraction and differential thermal analysis within Pt crucibles. The XRD results confirmed the existence of stoichiometric phase Li₄V₃₄O₈₇ in the Li₂O–V₂O₅ system. In the K₂O–V₂O₅ system, the melting temperatures of K₂V₈O₂₁ and KVO₃ were experimentally determined to be 532.4°C and 516.5°C, respectively. The eutectic reaction between liquid, Rb₃V₅O₁₄, and RbVO₃ in the Rb₂O–V₂O₅ system was identified at 496°C with a composition of 42 mol% Rb₂O. The modified quasichemical model (MQM), which accounts for the short-range ordering of the second-nearest neighbors of cations in molten oxide solutions, was employed to describe the liquid phase, and compound energy formalism (CEF) was applied to model the Li_1+XV₃O₈ solid solution at elevated temperatures. Thermodynamic modeling of the R₂O–V₂O₅ (R = Li, Na, K, Rb, and Cs) systems was developed using the CALculation of PHAse Diagrams (CALPHAD) methodology. The experimental data across the entire composition range of the R₂O–V₂O₅ systems were successfully reproduced, and thermodynamic properties for all solid and liquid phases within all binary systems were obtained. The developed thermodynamic database was further applied to simulate vanadium extraction processes, with the optimal operation windows.

To overcome the limitations of small-sample data in establishing microstructure–property linkages, this study introduces a deep transfer learning framework with a dynamic weighting mechanism. By transferring the perceptual capabilities of the ResNet-18 model and adaptively fusing them with material domain knowledge, the model effectively captures complex features such as phase distribution. Using silicon nitride ceramics as the primary research object, the framework achieves an average cross-validation prediction accuracy (R²) of 0.73, representing a 104.3% relative improvement compared to the traditional CNN framework, and the optimal model reaches an accuracy of R² = 0.89. Furthermore, this framework also demonstrates exceptional predictive accuracy on silicon carbide ceramics (R² = 0.84) and sintered nano silver (R² = 0.93), indicating its strong generalization capabilities. By employing multilevel gradient-weighted class activation mapping (Grad-CAM) and sliding occlusion analysis, the decision-making process of the model is elucidated, thereby validating the logical soundness of its predictions. Additionally, symbolic regression is utilized to identify the influence of different microstructural features on thermal conductivity and to establish their quantitative relationships. This research holds broad application prospects in the rapid development and design of thermal management materials, analysis of material microstructure images, and the establishment of structure–performance relationships between microstructural features and macroscopic properties.

Advances in artificial intelligence (AI) and large language models (LLMs) are transforming materials research by enabling automated data extraction, knowledge integration, and property prediction. This study presents a dual-stage, LLM-assisted framework for magnesium alloy design that combines semantic extraction, thermodynamic reasoning, and machine learning (ML). Using Qwen-2.5, alloy chemistry, processing details, and thermal and mechanical property data are automatically extracted from full-text literature and converted into structured records. The extracted information is expanded with thermodynamic phase descriptors predicted by DeepSeek-R1 and numerical processing features generated from CLIP-based embeddings. The feature set is optimized using sequential backward selection (SBS), and predictive models are developed using support vector machines (SVM), random forest (RF), and eXtreme Gradient Boosting (XGB). The proposed workflow effectively integrates chemistry, thermodynamics, and processing history, achieving robust predictions for thermal conductivity, yield strength, and ultimate tensile strength. The best performing models yielded R² values of ∼ 0.80 (RMSE ∼ 9.98 W·m⁻¹ K⁻¹), ∼ 0.69 (RMSE ∼ 37.2 MPa), and ∼ 0.73 (RMSE ∼ 31.5 MPa) for TC, YS, and UTS, respectively. Validation against CALPHAD calculations shows that DeepSeek-R1 reproduces equilibrium phase fractions within 1 wt.% deviation. Overall, this work shows that semantic intelligence can link literature-derived knowledge with predictive modeling, providing a pathway for processing-informed alloy design.

Potassium sodium niobate (KNN)-based ceramics have attracted significant interest due to their strong piezoelectric response, distinct photochromic, and photoluminescent behaviors, demonstrating great potential for applications in medical devices and optical securities. Recent breakthroughs in artificial intelligence have facilitated the use of machine learning (ML) in KNN-based ceramics. However, conventional global ML modeling approaches tend to overlook the decisive role of local atomic environments, especially those introduced by dopants, which critically govern the ceramic properties. Herein, a robust database containing 300 entries for key properties of KNN-based ceramics is constructed through high-throughput density functional theory calculations, whose reliability is benchmarked against experimental data. Furthermore, we implement an ML approach that specifically emphasizes the features describing the local coordination of dopants to map the relationships between doping behaviors and the structural stability and electronic structure of KNN-based ceramics. Moreover, the analysis of feature importance yields physically meaningful design rules that directly link atomic scale to functional performance. This work accelerates the development of KNN-based ceramics with electro-optical multifunctional coupling by establishing the critical influence of local structure on macroscopic properties.