ABSTRACT To overcome the strength‐conductivity trade‐off in Cu–Ni–Sn alloys, this study integrates machine learning, thermodynamic calculations, and experiments to design alloys from a phase‐selectivity perspective. SHAP analysis reveals that the coexistence of Ni 3 Sn and Ni 3 Sn 2 precipitates yields synergistic strengthening superior to single‐phase precipitation. Feature screening identifies the variance of covalent radius and mean Allred–Rochow electronegativity as key physical parameters coupling mechanical and electrical properties. By combining a support vector regression model with thermodynamic phase‐equilibrium calculations, an optimal 350°C aging design window, corresponding to a Sn/(Ni + Sn) mass ratio of approximately 40%–56%, was identified for dual‐phase coexistence. Experimental validation of the predicted Cu–9Ni–9Sn alloy demonstrates a favorable property balance, achieving a tensile strength of 1351 MPa, hardness of 415 HV, and electrical conductivity of 12.43% IACS. Microstructural analysis confirms that the excellent performance originates from the multi‐scale dislocation obstruction by γ‐DO 22 /γ‐L1 2 ordered phases and Ni 3 Sn 2 precipitates, coupled with reduced electron scattering due to solute depletion. This closed‐loop paradigm offers a transferable route for multiphase regulation in precipitation‐strengthened alloys.
ABSTRACT As a high‐performance polymeric material, the thermal stability, mechanical strength, and processability of copolyamides (co‐PAs) are determined by the structural design of diamines, diacids, and their stoichiometric ratios. Labor‐intensive experimental design approaches struggle to efficiently explore the vast structural space of the co‐PAs. Here, we develop the sequence‐sensitive multifidelity (SSMF) hybrid modeling methods to predict the properties of copolyamide chains, and modify the genetic algorithm as a teacher–student dual‐model strategy to achieve the inverse design of copolyamide structures. Based on the long SMILES representations of the copolyamide chain sequences, an SSMF hybrid model is constructed by combining two complementary models, that is, a high‐sequence sensitivity model and a high‐efficiency model, enhancing its accuracy through knowledge distillation. Within the modified genetic algorithm framework, the student model enables a massive population iteration, whereas the teacher model conducts the final optimization of inverse design results. Such a teacher‐student dual‐model strategy effectively identifies candidate structures that simultaneously satisfy the targeted glass transition temperature, melting point, and tensile modulus from over 2 billion virtual structures. The developed SSMF modeling method and the teacher‐student cooperative inverse design approach can promote the design of other copolymer structures, especially their chain sequences, advancing the AI‐assisted design of polymeric materials.
ABSTRACT Biomaterials science literature contains rich experimental data scattered across unstructured text, tables, and figures, making manual extraction unscalable. Existing regex‐based and multi‐agent methods are limited by insufficient semantic understanding and poor multitask consistency. We developed MatInform, a two‐stage multi‐agent information extraction framework integrating preprocessing and paragraph awareness where a Main Agent unifies global variables (e.g., material names and methods) and Sub‐Agents perform six constrained fine‐grained extraction tasks in parallel, following a biomaterial‐domain six‐task unified schema covering composition, processing, characterization, antibacterial activity, biocompatibility, and ion release. Evaluated on 20 annotated Ti‐based papers, MatInform achieves a four‐model average System F1 of 0.826. Ablation studies identify the two‐stage architecture as the primary contributor to performance, whereas preprocessing and paragraph awareness contribute marginal improvements. Applied to 1469 titanium‐based antibacterial material papers, MatInform extracts structured data for 8127 samples. Downstream machine learning and SHAP analysis guide the development of a Ca–AgZn modified Ti material with excellent antibacterial performance and enhanced cell proliferation. With minor code adaptation, the framework yields an F1 of 0.764 on 20 ceramic fiber papers. Overall, the framework enables reusable automated extraction with consistent and interpretable multitask coordination.
ABSTRACT Many real‐world scientific and industrial applications require the optimization of expensive black‐box functions. Bayesian optimization (BO) provides an effective framework but often struggles with local optima and lacks interpretability. This paper introduces reasoning BO, a novel framework leveraging reasoning models to guide BO sampling while incorporating multi‐agent systems and knowledge graphs for online knowledge accumulation. We evaluate our approach across 10 diverse tasks, including synthetic functions and complex real‐world chemical optimizations. Reasoning BO progressively refines sampling strategies through real‐time insights and hypothesis evolution, identifying high‐performing regions effectively. In the direct arylation task, our method significantly outperformed traditional BO, increasing yield from 25.20% to 60.07%. Furthermore, we demonstrate that smaller LLMs, after post‐training, can achieve performance comparable to larger counterparts. This framework establishes an intelligent cost‐effective optimization system for scientific discovery, combining LLM reasoning with structured knowledge management.
ABSTRACT Coarse‐grained (CG) molecular dynamics (MD) can greatly extend the accessible time and length scales for water, provided that the reduced model captures key structural, dynamical, and thermodynamic properties. Here, we introduce a CG machine learning potential (MLP) for water, named CGNEP‐MB‐pol, which integrates a one‐molecule to one‐bead mapping with the neuroevolution potential (NEP) framework and MB‐pol reference data, thereby aiming to alleviate, within the liquid‐water and ice‐Ih states examined here, the state dependence and limited transferability that often constrain conventional CG models. Without altering the NEP descriptor or network architecture, the model is trained on CG coordinates and labels derived from atomistic configurations through a two‐step dataset refinement strategy, combining multi‐state training based on forces, energies, and virials. CGNEP‐MB‐pol remains compatible with the GPUMD + NEP workflow while delivering a substantial speedup over its all‐atom counterpart. It reproduces mapped reference forces, radial distribution functions of liquid water and ice, and the overall temperature dependence of self‐diffusion and viscosity. Two‐phase simulations capture ice growth, melting, and solid–liquid coexistence, yielding a melting temperature of approximately 274 K. This study establishes a practical route for constructing CG‐MLPs from high‐fidelity atomistic reference models.
Hot-deformation flow modeling in steels is dominated by physics-based constitutive equations, whereas machine-learning surrogates are trained in a "one-case, one-model" manner-leveraging fitting capacity rather than reducing the calibration burden. Here we propose a transferable sequence framework which learns shared flow-curve priors across diverse compositions and processing conditions and is then personalized to a new steel using only four extreme-condition curves. The transfer-adapted model achieves high accuracy and substantially improved robustness compared with target-only training, providing a practical alternative to steel-by-steel re-calibration. Strain-resolved global KernelSHAP offers a compact physical interpretation: temperature contributes predominantly negatively (thermal softening) and strain rate positively (rate hardening), and few-shot adaptation systematically shifts attribution from high-dimensional chemistry toward measurable process variables. Practically, this enables fast deployment for new steels, new heats/batches, and new thermomechanical processing windows with only a small set of hot-compression tests, supporting accelerated process-window design and thermomechanical simulation.
ABSTRACT By integrating physics‐informed feature engineering with symbolic regression (specifically, the sure independence screening and sparsifying operator or SISSO), the magnetic properties of Cu‐based alloys have been predicted based on a small experimental dataset ( n = 76). A set of atomic descriptors was constructed, achieving predictive performance for saturation magnetization ( M s ) and remanent magnetization ( M r ). From in‐training set fitting, the coefficients of determination ( R 2 ) of M s and M r are 0.981 and 0.936, respectively. Further leave‐one‐out cross‐validation (LOOCV) on the alloy dataset excluding three pure metal samples demonstrated robust prediction maintaining a cross‐validated R 2 cv of 0.945 for M s . The mathematical expressions generated by SISSO reveal the synergistic effects of processing parameters and atomic properties on magnetic performance. Experimental validation on two newly designed alloy compositions confirmed the generalization ability with relative errors for M s predictions below 12.8%. Moreover, a cross‐system validation on Fe‐based medium manganese steel captured M s of the sample annealed at relatively low temperature (973 K) with low relative error (11.2%), suggesting the model to be transferable to other alloy systems. On the other hand, the model's limited ability to predict coercivity ( H c ) and the properties of these samples that have undergone phase transformations highlight the necessity of incorporating microstructural descriptors in future.
ABSTRACT Key phase diagram studies of Li 2 O–V 2 O 5 , K 2 O–V 2 O 5 , and Rb 2 O–V 2 O 5 systems were conducted using X‐ray diffraction and differential thermal analysis within Pt crucibles. The XRD results confirmed the existence of stoichiometric phase Li 4 V 34 O 87 in the Li 2 O–V 2 O 5 system. In the K 2 O–V 2 O 5 system, the melting temperatures of K 2 V 8 O 21 and KVO 3 were experimentally determined to be 532.4°C and 516.5°C, respectively. The eutectic reaction between liquid, Rb 3 V 5 O 14 , and RbVO 3 in the Rb 2 O–V 2 O 5 system was identified at 496°C with a composition of 42 mol% Rb 2 O. 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+X V 3 O 8 solid solution at elevated temperatures. Thermodynamic modeling of the R 2 O–V 2 O 5 (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 R 2 O–V 2 O 5 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.
ABSTRACT 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.
ABSTRACT Ru‐based solid‐solution alloys have emerged as attractive substitutes for Pt‐based electrocatalysts in the alkaline hydrogen evolution reaction (HER). However, the vast compositional space of these alloys hinders rapid catalyst discovery. Existing machine‐learning design strategies are largely based on idealized simulation data, limiting their experimental relevance and screening efficiency. Herein, we develop an interpretable machine‐learning framework informed by experiment to facilitate the rapid screening of Ru‐based HER electrocatalysts. Built on an experimental dataset, the framework integrates electronic descriptors, interpretability analysis, a Bayesian‐optimized surrogate model and uncertainty‐aware ranking to efficiently identify promising compositions. Interpretability analysis identifies V, Co, and Ru as the key elements governing catalytic performance, whereas the surrogate model maps HER activity across the ternary alloy space to define a confined high‐performance composition window. Guided by this framework, the optimized V 17 Co 31 Ru 52 catalyst exhibits an overpotential of 61.7 mV at a current density of 100 mA·cm −2 in 1 M KOH, while maintaining stable operation for 260 h under the same current density. This work provides a rapid, interpretable and experimentally relevant strategy for discovering high‐performance alkaline HER electrocatalysts.
ABSTRACT Low‐melting liquid metals, especially Ga–In alloys, are essential for flexible electronics, soft robotics, and adaptive thermal interfaces. Predictive atomistic modeling of their melting behavior is challenging because experiments generally provide limited microscopic insight, whereas first‐principles simulations are computationally expensive. In this work, Neuroevolution Potential (NEP) potentials for the Ga–In system are developed using density functional theory (DFT) datasets of 200–700 atom supercells generated with the local density approximation (LDA), Perdew‐Burke‐Ernzerhof (PBE), and PBE + D3 functionals. Eighteen independent NEP models with different energy/force/virial weights were trained and benchmarked against density, radial distribution function (RDF), self‐diffusion coefficient (SDC), and melting temperature ( T m ). Radar‐metric analysis shows that PBE + D3‐based models provide the most balanced overall performance for density, liquid structure, and diffusion, whereas LDA‐based models give the best T m predictions for Ga and EGaIn. We further design a local Lindemann parameter for solid–liquid identification in disordered alloys, extending the two‐phase method to compositionally complex liquid‐metal systems. Together, these results provide a benchmarked NEP model set and a scalable framework for phase‐transition and transport simulations of Ga–In liquid metals.
ABSTRACT Accurate physicochemical property prediction is critical for the rational design of energetic materials (EMs), yet limited high‐quality experimental data and property‐wise data imbalance restrict the application of conventional single‐task and multi‐task machine learning models. Here, we propose QGeoSEP, a multi‐task learning framework for the multi‐property prediction of EMs under data imbalance and scarce annotation. Integrating three‐dimensional geometric structures, electronic structure, features and endpoint semantic information, QGeoSEP captures attribute interdependencies via the transformer's multi‐head attention mechanism. It outperforms single‐task baselines and the multi‐task model MTL‐GNN on the QM9 dataset. On a curated dataset of 211,890 EMs‐like C–H–O–N organic small molecules, it achieves R 2 of 0.949, 0.878, 0.891, and 0.647 for density, melting point, heat of combustion, and decomposition temperature, respectively, with a 14.37% average improvement over MTL‐GNN. On an independent 661‐EMs test set, it yields superior prediction accuracy for density and decomposition temperature, with MAE of 0.090 g·cm −3 and 21.641 K. QGeoSEP trained on the C–H–O–N dataset addresses EMs data scarcity, enables collaborative multi‐property prediction under data sparsity and imbalance, and exhibits excellent transfer generalization for real EMs. The model is deployed on a web platform for barrier‐free use without programming or local installation.
ABSTRACT Energetic molecules are the key components in energetic materials, and the huge screening space of molecular structures make the design of energetic molecules complicated. In this work, we adopt a research method that combines High‐Throughput Computation (HTC), Chemical Informatics, classic Machine Learning (ML) and Artificial Intelligence (AI) generative models to achieve high‐throughput design of molecular structures for energetic materials. Based on the HTC results and molecular structural features extracted from chemical informatics, ML models for the structure–activity relationship of energetic molecules was obtained. Using the ML models, detonation parameters were predicted and target energetic molecules were screened in a short time. Meanwhile, using AI generative models, it was found that novel low‐sensitivity high‐energy molecules could be generated automatically; and their performance as well as synthetic accessibility were verified theoretically. Furthermore, the applications of ML models and AI in synthetic route design of energetic molecules are prospected. This research demonstrates the important role of AI/ML models in accelerating the development of energetic materials.
ABSTRACT Machine learning and AI assistants are reshaping materials research; however, their day‐to‐day application in experimental and computational workflows is still inconsistent, limited by insufficient software documentation and software engineering practices, fragmented software engineering ecosystem and brittle integration between computational tools. Especially in the field of materials informatics, the lack of user‐friendly tool interfaces has restricted the popularization of data‐driven methods in daily scientific research. In this study, we present MatterMind, a user‐friendly agentic interface that closes the loop between first‐principles computation, generative structure design, and large language model (LLM) analysis. The platform unifies (i) plane‐wave first‐principles workflows through Vienna Ab initio Simulation Package (VASP), (ii) crystal generation via diffusion models, and (iii) LLM assistance for error diagnosis, result interpretation, and report drafting. With simple “button‐click” operations or natural‐language prompts, users can: (1) build, launch, and monitor VASP jobs with automated parsing and recovery; (2) sample candidate crystals using modern generative models (MatterGen); and (3) obtain LLM‐guided summaries, comparisons, and next‐step suggestions for screening decisions. We illustrate end‐to‐end case studies that couple crystal generation to DFT relaxation and LLM‐assisted assessment, reducing scripting overhead and improving transparency and reuse from a software engineering perspective. It provides an intelligent scientific tool with comprehensive software documentation, enhancing the efficiency and reliability of scientific exploration in chemical and materials science research.
ABSTRACT The general predictive approach established in our previous work Qiao et al., Materials Genome Engineering Advances. 2025;3(3):e70021. was employed to study the diffusion behavior of interstitial B and N atoms in FCC_CoNiV multi‐principal element alloy (MPEA) based on sublattice preference, with comparative C data from prior work, to enrich the diffusion genome database of lightweight interstitial elements. Furthermore, we employed the Kabsch algorithm to describe the lattice distortion of the octahedra containing interstitial atoms quantitatively. The results show that the number of V atoms in the local octahedral environment exerts a different regulatory effect on the diffusion behavior of interstitial atoms B and N; that is, B and C exhibit a higher diffusion barrier when migrating into V‐rich sites, whereas N exhibits such higher barrier when leaving these sites. Electron localization function (ELF) analysis shows the difference is due to the diverse bonding strengths between V atoms and interstitial atoms B, N, and C. Nonperiodic diffusion barrier waves and diffusion parameters were quantitatively predicted in detail. The fundamental understanding of interstitial diffusion mechanisms and quantitative characterization of the diffusion parameters of B, N, and C in FCC_CoNiV MPEA provide a benchmark and critical insights for tailoring alloy properties through interstitial engineering.
Photocatalytic degradation of volatile organic compounds (VOCs) offers a sustainable strategy for air purification; however, developing efficient photocatalysts is limited by trial-and-error methods due to complex composition-structure-performance relationships. To address this, this work identifies charge carrier transport, a key microscopic process governing photocatalytic efficiency, and proposes a high-throughput characterization-based screening strategy using this descriptor. A compositional gradient thin-film library in the TiO2/ZnFe2O4/Cu system is fabricated via inclined gradient sputtering, enabling continuous composition variation on one sample. High-throughput Hall effect measurements map carrier mobility across the library, rapidly identifying the composition (TiO2)58(ZnFe2O4)31(Cu)11 with superior charge transport properties. The same composition powder catalyst is synthesized and achieve 82.9% toluene degradation within 4 h under 30% relative humidity, confirming the screening effectiveness. A strong correlation between the photocatalytic degradation performances of thin films and the corresponding powder catalysts (Spearman r = 0.94) confirms that catalytic activity trends are maintained across material forms, thereby validating the transferability of the thin-film screening results to practical powder catalysts. By bridging high-throughput thin-film libraries with practical powder catalysts, this work establishes a transferable screening framework that links charge transport properties with photocatalytic performance, providing a general pathway for accelerated discovery of photocatalytic materials beyond conventional trial-and-error approaches.
Fabricating gamma '-strengthened nickel-based superalloys via laser powder bed fusion (LPBF) faces significant obstacles due to their severe susceptibility to cracking, which strictly limits their industrial application. To address this, a physics-informed machine learning (ML) framework based on a generative design concept and high-throughput thermodynamic calculations is proposed. A thermodynamic database containing 210,000 samples was constructed, and a conditional variational autoencoder (CVAE) was trained and subsequently used to generate potential target compositions. Guided by critical physical metallurgy features, including gamma ' phase content, carbide characteristics, and cracking indices, a novel superalloy entitled SHA800, tailored for service temperatures of 800 degrees C-900 degrees C, was designed and experimentally validated. Experimental results demonstrated a wide crack-free processing window. Additionally, the alloy attained a 43% gamma ' volume fraction after heat treatment, achieving an exceptionally high hardness of 587 HV0.2. This integrated approach, leveraging large-scale thermodynamic data and ML models, significantly accelerates the alloy design process for laser additive manufacturing and has important implications for the development of superalloys.
This study enhanced the optoelectronic properties of BIT (Bi4Ti3O12) through Zr doping at the B site, constructing an Au/Bi4Ti2.85Zr0.15O12/Nb: SrTiO3 (Au/BTZ/Nb:STO) device. Zr doping induces a red shift in the optical absorption band of BIT, narrowing the bandgap to similar to 3.22 eV. Under 365 nm illumination, the device achieves optically controlled SET processes and switching ratio modulation, demonstrating short-term optical synaptic plasticity that successfully simulates learning and forgetting processes. Through optical coupling signal modulation, the device exhibits long-term potentiation (LTP) and long-term depression (LTD) characteristics. An differential convolutional neural network (DCNN), which was constructed based on this device, achieved classification accuracies of 94.2% for the handwritten digit dataset (MNIST) and 83% for the Fashion-MNIST dataset after 50 iterations of training, incorporating the device's conductance response parameters. The study demonstrates that the Au/BTZ/Nb:STO device combines outstanding optoelectronic performance with high-precision neuromorphic computing capabilities, offering a novel pathway for next-generation neural network computing.
