Polyimides (PIs) are widely used in industries for their exceptional mechanical properties and thermal resilience. Despite their benefits, the traditional development process for PIs is time-consuming, often lagging behind the increasing demand for materials with tailored properties. In this study, we introduce a machine learning-based approach to predict and optimize the mechanical properties of PI materials and their composites. We developed six predictive models to assess PI structures under various conditions, aiming to enhance our understanding of PI mechanical behavior and facilitate the discovery of high-performance PI structures. By analyzing the substructures within top-performing PIs, we identified key structural motifs that contribute to improved tensile strength, modulus, and elongation at break. Furthermore, we examined the influence of fillers on PI composites, revealing that rigid fillers such as SiO₂ and graphene oxide (GO) significantly improve mechanical properties, with GO showing versatile enhancement across multiple mechanical properties. We then screened 800,000 virtual PI structures by using our predictive models, identifying several candidates with targeted mechanical properties. These findings provide a basis for the future experimental validation of optimal PI structures and fillers, offering an efficient pathway to accelerate the design of PI materials with targeted mechanical properties. Our study can also be extended to other materials research, serving as a valuable paradigm for the design of polymers and their composites.

Stacking fault energy (SFE) significantly influences plastic deformation, strength, and processing performance, making accurate assessment and prediction of SFE essential for material design and optimization. Traditional SFE calculations mainly rely on experimental measurements and thermodynamic theories, with the former usually being time-consuming and the latter limited in applicability at different compositions. To overcome these limitations, this study proposes a machine learning (ML) strategy introducing physical metallurgy (PM) parameters relevant to SFE, aiming to achieve robust predictions. Specifically, this study evaluates three methods for introducing PM information into ML (as an input, an intermediate parameter, and a transfer source), with transfer learning as the best strategy. Initially, various PM parameters were calculated based on alloy composition and temperature, and subsequently used as outputs to train a convolutional neural network (CNN). This source model was then transferred to the SFE prediction model. The results from the model transfer using different PM information show that incorporating phase-transformation driving force (DF) as a source model for SFE prediction provided the most accurate and reliable results. This approach of introducing PM parameters into ML significantly improves the predictive capability of SFE models, offering a new perspective and solution for the prediction of SFE. Furthermore, this method may also be applicable to the prediction of other material properties during material design and optimization.

The traditional Haber-Bosch process for ammonia synthesis is both energy-intensive and capital-demanding. Electrocatalytic nitrogen reduction reaction (NRR) has emerged as a promising, sustainable alternative, with recent advantages highlighting its potential. Single-atom catalysts (SACs) and single-cluster catalysts (SCCs) are promising catalysts for NRR due to their atomically dispersed active sites, maximized atom utilization, and distinctive coordination and electronic structures, all of which facilitate mechanism insights at the atomic level. Benefiting from efficient atom utilization, for example, the ammonia yield rate on Au_1/C₃N₄ is roughly 22.5 times as high as that of supported Au nanoparticles, fully demonstrating the significant advantages of SACs over nanoparticles. In this review, we focus on the theoretical progress in SACs and SCCs for electrocatalyzing NRR, including nitrogenase-like bio-inspired catalysts and other metal-based catalysts. We further examine key adsorption energy and electronic descriptors that enhance our understanding of catalytic performance. Finally, we discuss the remaining challenges and future directions for advancing SACs and SCCs in electrocatalytic NRR applications.

The partition coefficient (log P) is a critical parameter that measures the balance between hydrophilicity and lipophilicity of molecules, playing a key role in molecular material design and drug development. Developing accurate, efficient, and computationally simple models for log P prediction is essential for advancing drug discovery and materials science. In this study, we introduce the optimized 3D molecular representation of structures based on electron diffraction descriptor (opt3DM) into machine learning (ML) frameworks, achieving highly accurate log P predictions. By fine-tuning key parameters, the scale factor (s_L) and descriptor dimension (N_s), we identified the optimal values of s_L = 0.5 and N_s = 500. Among various ML algorithms tested, automatic relevance determination (ARD) regression, Ridge regression, and Bayesian Ridge regression demonstrated superior predictive performance. These optimized models outperformed the OPEn structure-activity/property relationship app (OPERA) model on the M-dataset and also delivered competitive results in the SAMPL6 and SAMPL9 challenges. Our findings not only establish a robust, fast, and precise approach for log P prediction, but also highlight the potential of opt3DM as a powerful tool for molecular representation. This work lays a foundation for broader applications in molecular material design and drug development.

As global demand for clean energy increases, fuel cells have emerged as a key technology for sustainable power generation. Anion exchange membrane (AEM) fuel cells offer a more economical and environmentally friendly alternative to the popular proton exchange membrane (PEM) fuel cells, which rely on fluorinated polymers and also use expensive platinum group catalysts. However, designing high-performance AEMs is challenging because of the need to balance conflicting material properties. In this study, we employ machine learning to accelerate the design of fluorine-free copolymers for AEMs, focusing on known monomer chemistries. By training models on AEM data from the literature, we predicted key properties, namely, hydroxide ion conductivity, water uptake (WU), and swelling ratio (SR). Screening 11 million novel copolymer candidates using predictive models and heuristic filters, we identified more than 400 promising fluorine-free copolymer candidates with predicted OH^- conductivity greater than 100 mS/cm, WU below 35 wt%, and SR below 50%. This computational approach to AEM design could contribute to developing more efficient and sustainable AEM fuel cells for various energy applications.

The reduced-activation high-entropy alloys (RAHEAs) have promising applications in advanced nuclear systems due to their low activation, excellent mechanical properties and radiation resistance. However, compared to the conventional high-entropy alloys (HEAs), the relatively small datasets of RAHEAs pose challenges for alloy design by using conventional machine learning (ML) methods. In this work, we proposed a framework by incorporating symbolic regression (SR) and domain adaptation to improve the accuracy of property prediction based on the small datasets of RAHEAs. The conventional HEA datasets and RAHEA datasets were classified as source and target domains, respectively. SR was used to generate features from element-based features in the source domains. The domain-invariant features related to hardness were captured and used to construct the ML model, which significantly improved the prediction accuracy for both HEAs and RAHEAs. The normalized root mean square error decreases by 24% for HEAs and 30% for RAHEAs compared to that of the models trained with element-based features. The proposed framework can achieve accurate and robust prediction on small datasets with interpretable domain-invariant features. This research paves the way for efficient material design under small dataset scenarios.

Machine learning (ML) has advantages in studying fundamental properties of materials and comprehending structure-property correlations. In this study, we employed sure independence screening and sparsifying operator (SISSO) method (ML technique) to explore the experimental dielectric constant, temperature coefficient of frequency resonator, and quality factor of inorganic oxide microwave dielectric materials. Among the constructed white-box models, the highest accuracy, with a coefficient of determination (R²) of 0.8, was observed in predicting the dielectric constants of the quaternary materials. Additionally, we proposed a straightforward strategy to merge the ternary and quaternary datasets in a single training, aiming to address the issue of data scarcity in ML research. Although this strategy slightly compromises the model accuracy, it has the advantage of creating a more unified trained model for structural-property relationship understanding. Using the unified and interpretable model trained with the merged dataset, we derived a general rule governing the dielectric constant of materials. Our ML findings regarding the dielectric property provide fundamental insights for designing microwave dielectric materials with diverse dielectric constants.

Intermetallic compounds are crucial in modern industry due to their exceptional properties, where density is identified as a critical parameter determining their potentiality for lightweight applications. In this study, over 7,000 density data points are collected for binary intermetallic compounds from different crystal structures. A new intermetallics graph neural network (IGNN) model is developed to perform regression and classification tasks for density prediction. Compared to traditional machine learning models, the IGNN model demonstrated superior capability in capturing crystal structure and effectively addressing challenges posed by polymorphism. The interpretability of the IGNN model classification process is enhanced through the t-distributed stochastic neighbor embedding (t-SNE) visualization method. Additionally, the IGNN model exhibited excellent performance in predicting the density of multicomponent complex intermetallic compounds, indicating its robustness and generalizability. This study presents a graph neural network (GNN) method suitable for multi-crystal structure data modeling, providing a novel computational framework for density prediction in intermetallic compounds. This advancement represents a significant contribution to this field, paving the way for more targeted material selection and application in lightweight technologies.

Single-atom catalysts (SACs) have emerged as a research frontier in catalytic materials, distinguished by their unique atom-level dispersion, which significantly enhances catalytic activity, selectivity, and stability. SACs demonstrate substantial promise in electrocatalysis applications, such as fuel cells, CO₂ reduction, and hydrogen production, due to their ability to maximize utilization of active sites. However, the development of efficient and stable SACs involves intricate design and screening processes. In this work, artificial intelligence (AI), particularly machine learning (ML) and neural networks (NNs), offers powerful tools for accelerating the discovery and optimization of SACs. This review systematically discusses the application of AI technologies in SACs development through four key stages: (1) Density functional theory (DFT) and ab initio molecular dynamics (AIMD) simulations: DFT and AIMD are used to investigate catalytic mechanisms, with high-throughput applications significantly expanding accessible datasets; (2) Regression models: ML regression models identify key features that influence catalytic performance, streamlining the selection of promising materials; (3) NNs: NNs expedite the screening of known structural models, facilitating rapid assessment of catalytic potential; (4) Generative adversarial networks (GANs): GANs enable the prediction and design of novel high-performance catalysts tailored to specific requirements. This work provides a comprehensive overview of the current status of AI applications in SACs and offers insights and recommendations for future advancements in the field.

Dislocation glide and/or deformation twinning, which can be uniformly described as the kinetic behaviors of atoms driven by thermodynamics, play important roles in dominating mechanical performance and strength-ductility paradox of metallic alloys. In this work, the physical origins behind the mechanical performance paradox are investigated in light of thermo-kinetic synergy upon materials processing. Combining the classical dislocation theories with common strengthening mechanisms, the quantitative connections among yield strength, plastic strain and dislocation density are bridged by the driving force and energy barrier of dislocation motion. The FCC-Al and Al alloys with Mg, Cu and Si solutes are studied as typical representatives by performing molecular dynamics (MD) simulations to analyze the tensile behaviors. It turns out that the thermo-kinetic synergy is responsible for the strength-ductility exclusive behaviors. The yield strength and flow stress are enhanced with Mg, Cu and Si solutes adding in FCC-Al, due to the increased interactions between dislocations and solute atoms. All the Mg, Cu and Si solutes can benefit the mechanical responses of Al alloys. Increasing Mg content enhances the driving force and yield stress of Al-Mg alloys, but reduces the energy barrier and plastic strain. The solute Si addition can further increase the driving force and yield stress, but decrease the energy barrier and plastic strain of Al-Mg-Si alloys. Our investigation provides an innovative viewpoint for understanding the mechanical performance paradox of metallic alloys, and offers insightful guidance for designing advanced Al alloys with good mechanical performance.

While urea is widely used as a chemical raw material, its precursor ammonia (NH₃) has traditionally been synthesized under high-temperature/pressure conditions, leading to not only huge energy consumption but also serious CO₂ emission. Here, we present a groundbreaking catalyst design approach, which optimizes adsorption configurations and reaction pathways by controlling the adsorption energies of each intermediate in the reaction, thus enhancing catalytic performance. Via density functional theory (DFT) calculations, we designed a triatomic catalyst [i.e., Fe₂Mo@γ-graphdiyne (γ-GDY)] with a limiting potential of -0.22 V and a C-N coupling energy barrier of 0.34 eV. Notably, the Fe₂Mo@γ-GDY catalyst presents a high selectivity and robust antioxidation capabilities under applied potentials. Our comprehensive analysis elucidates the factors affecting the limiting potential and C-N coupling energy barrier. These insights significantly contribute to the advancement of catalyst design strategies for electrocatalytic urea synthesis, offering a more efficient and eco-friendly alternative to traditional methods.

Lattice thermal conductivity (κ_L) is crucial for efficient thermal management in electronics and energy conversion technologies. Traditional methods for predicting κ_L are often computationally expensive, limiting their scalability for large-scale material screening. Empirical models, such as the Slack model, offer faster alternatives but require time-consuming calculations for key parameters such as sound velocity and the Grüneisen parameter. This work presents a high-throughput framework, physical-informed kappa (PINK), which combines the predictive power of crystal graph convolutional neural networks (CGCNNs) with the physical interpretability of the Slack model to predict κ_L directly from crystallographic information files (CIFs). Unlike previous approaches, PINK enables rapid, batch predictions by extracting material properties such as bulk and shear modulus from CIFs using a well-trained CGCNN model. These properties are then used to compute the necessary parameters for κ_L calculation through a simplified physical formula. PINK was applied to a dataset of 377,221 stable materials, enabling the efficient identification of promising candidates with ultralow κ_L values, such as Ag₃Te₄W and Ag₃Te₄Ta. The platform, accessible via a user-friendly interface, offers an unprecedented combination of speed, accuracy, and scalability, significantly accelerating material discovery for thermal management and energy conversion applications.

Machine learning (ML) is revolutionizing alloy design, yet traditional models face challenges with limited data and complex nonlinearities. Our study presents a self-decision design strategy that integrates target property determination, reverse and forward modeling, and feature importance analysis to optimize low-alloyed rare earth (RE)-free magnesium alloys for strength-ductility synergy. The strategy was validated with experimental data, leading to the development of a new Mg-2Al-1Zn-0.6Ca-0.4Mn (wt%) alloy processed at specific conditions, achieving a tensile strength of 344 MPa and an elongation-to-failure (EL) of 21.3% at room temperature. The discrepancies between experimental and predicted results were less than 5%, underscoring the accuracy of this approach. This streamlined design strategy not only promises to accelerate the development of low-cost, high-performance alloys but also minimizes the need for human intervention, thereby enhancing the efficiency and precision of alloy design.

Copper-based electrocatalysts, which hold great promise in selectively reducing CO₂ into multicarbon products, have attracted significant recent interest, both experimentally and theoretically. While many studies have suggested a strong dependence of catalytic selectivity on the concentration of the ^*CO reaction intermediate on the Cu surface, it remains challenging for a direct experimental probe of the CO coverage. This necessitates a reliable computational method that can accurately establish the theoretical coverage-dependent phase diagram of CO adsorbates on the catalyst. Here we propose a scheme composed of density functional theory calculations, machine-learning force fields and graph neural networks as a solution. This method enables a fast screening of 7 million adsorption configurations based on a small set of density functional theory data, with a balance between accuracy and efficiency tuned by the combinatorial use of machine-learning force field and graph neural network models. We have investigated eight different Cu facets and discovered that the high-index facets such as (310), (210) and (322) exhibit a much higher CO coverage than the low-index counterparts such as (111), leading to an increased opportunity for C–C coupling for the former. Our results can provide a new perspective for the understanding of the fundamental role of CO coverage on the Cu surface for electrochemical CO₂ reduction.