There is a growing demand for the use of machine learning (ML) to derive fast-to-evaluate surrogate models of materials properties. In recent years, a broad array of materials property databases have emerged as part of a digital transformation of materials science. However, recent technological advances in ML are not fully exploited because of the insufficient volume and diversity of materials data. An ML framework called “transfer learning” has considerable potential to overcome the problem of limited amounts of materials data. Transfer learning relies on the concept that various property types, such as physical, chemical, electronic, thermodynamic, and mechanical properties, are physically interrelated. For a given target property to be predicted from a limited supply of training data, models of related proxy properties are pretrained using sufficient data; these models capture common features relevant to the target task. Repurposing of such machine-acquired features on the target task yields outstanding prediction performance even with exceedingly small data sets, as if highly experienced human experts can make rational inferences even for considerably less experienced tasks. In this study, to facilitate widespread use of transfer learning, we develop a pretrained model library called XenonPy.MDL. In this first release, the library comprises more than 140 000 pretrained models for various properties of small molecules, polymers, and inorganic crystalline materials. Along with these pretrained models, we describe some outstanding successes of transfer learning in different scenarios such as building models with only dozens of materials data, increasing the ability of extrapolative prediction through a strategic model transfer, and so on. Remarkably, transfer learning has autonomously identified rather nontrivial transferability across different properties transcending the different disciplines of materials science; for example, our analysis has revealed underlying bridges between small molecules and polymers and between organic and inorganic chemistry.
Abstract The discovery of new materials is one of the driving forces to promote the development of modern society and technology innovation, the traditional materials research mainly depended on the trial-and-error method, which is time-consuming and laborious. Recently, machine learning (ML) methods have made great progress in the researches of materials science with the arrival of the big-data era, which gives a deep revolution in human society and advance science greatly. However, there exist few systematic generalization and summaries about the applications of ML methods in materials science. In this review, we first provide a brief account of the progress of researches on materials science with ML employed, the main ideas and basic procedures of this method are emphatically introduced. Then the algorithms of ML which were frequently used in the researches of materials science are classified and compared. Finally, the recent meaningful applications of ML in metal materials, battery materials, photovoltaic materials and metallic glass are reviewed.
The scarcity of predicted magnetic topological materials (MTMs) by magnetic space group (MSG) hinders further exploration towards realistic device applications. Here, we propose a new scheme combining spin space groups (SSGs)--approximate symmetry groups neglecting spin-orbit coupling (SOC)--and MSGs to diagnose topology in collinear magnetic materials based on symmetry-indicator theory, enabling a systematic classification of the electronic topology across 484 experimentally synthesized collinear magnets from the MAGNDATA database. This new scheme exploits a symmetry-hierarchy due to SOC induced symmetry-breaking, so that nontrivial band topology can be revealed by SSG, that is yet invisible by the conventional MSG-based method, as exemplified by real triple points in ferromagnetic CaCu$_3$Fe$_2$Sb$_2$O$_{12}$, Dirac nodal lines at generic $k$-points in antiferromagnetic FePSe$_3$ and Weyl nodal lines in altermagnetic Sr$_4$Fe$_4$O$_{11}$. Notably, FePSe$_3$ is topologically trivial under MSG but hosts Dirac nodal lines within the SSG framework. Upon including SOC, these nodal lines are gapped and generate a sizable anomalous Hall conductivity. Despite a vanishing bulk net magnetism, FePSe$_3$ can host topologically protected surface states with large non-relativistic band spin-splitting. Moreover, topology in MTMs is tunable by rotating the magnetic moment direction once SOC is included, as exemplified in Sr$_4$Fe$_4$O$_{11}$.The interplay of topology with non-relativistic and SOC-induced control of properties via magnetic moment reorientation in the predicted MTMs is worthy of further studies in future.
Large Language Models (LLMs) have demonstrated remarkable abilities in scientific reasoning, yet their reasoning capabilities in materials science remain underexplored. To fill this gap, we introduce MatSciBench, a comprehensive college-level benchmark comprising 1,340 problems that span the essential subdisciplines of materials science. MatSciBench features a structured and fine-grained taxonomy that categorizes materials science questions into 6 primary fields and 31 sub-fields, and includes a three-tier difficulty classification based on the reasoning length required to solve each question. MatSciBench provides detailed reference solutions enabling precise error analysis and incorporates multimodal reasoning through visual contexts in numerous questions. Evaluations of leading models reveal that even the highest-performing model, Gemini-2.5-Pro, achieves under 80% accuracy on college-level materials science questions, highlighting the complexity of MatSciBench. Our systematic analysis of different reasoning strategie--basic chain-of-thought, tool augmentation, and self-correction--demonstrates that no single method consistently excels across all scenarios. We further analyze performance by difficulty level, examine trade-offs between efficiency and accuracy, highlight the challenges inherent in multimodal reasoning tasks, analyze failure modes across LLMs and reasoning methods, and evaluate the influence of retrieval-augmented generation. MatSciBench thus establishes a comprehensive and solid benchmark for assessing and driving improvements in the scientific reasoning capabilities of LLMs within the materials science domain.
Scientific research faces high costs and inefficiencies with traditional methods, but the rise of deep learning and large language models (LLMs) offers innovative solutions. This survey reviews transformer-based LLM applications across scientific fields such as biology, medicine, chemistry, and meteorology, underscoring their role in advancing research. However, the continuous expansion of model size has led to significant memory demands, hindering further development and application of LLMs for science. This survey systematically reviews and categorizes memory-efficient pre-training techniques for large-scale transformers, including algorithm-level, system-level, and hardware-software co-optimization. Using AlphaFold 2 as an example, we demonstrate how tailored memory optimization methods can reduce storage needs while preserving prediction accuracy. By bridging model efficiency and scientific application needs, we hope to provide insights for scalable and cost-effective LLM training in AI for science.
Reshma Devi, Keith T. Butler, Gopalakrishnan Sai Gautam
Abstract A pathway to overcome limited data availability in materials science is to use the framework of transfer learning, where a pre-trained (PT) machine learning model (on a larger dataset) can be fine-tuned (FT) on a target (smaller) dataset. We systematically explore the effectiveness of various PT/FT strategies to learn and predict material properties and create generalizable models by PT on multiple properties (MPT) simultaneously. Specifically, we leverage graph neural networks (GNNs) to PT/FT on seven diverse curated materials datasets, with sizes ranging from 941 to 132,752. Besides identifying optimal PT/FT strategies and hyperparameters, we find our pair-wise PT-FT models to consistently outperform models trained from scratch on target datasets. Importantly, our MPT models outperform pair-wise models on several datasets and, more significantly, on a 2D material band gap dataset that is completely out-of-domain. Finally, we expect our PT/FT and MPT frameworks to accelerate materials design and discovery for various applications.
Materials of engineering and construction. Mechanics of materials, Computer software
Trond Brandvik, Louis Gosselin, Zhaohui Wang
et al.
Refractory flue walls in anode baking furnaces are exposed to harsh conditions during operation, affecting the structural properties of the material. The flue walls in industrial furnaces degrade over time to the point where they no longer perform as intended and must be replaced. Earlier studies of spent refractory lining from anode baking furnaces have shown considerable densification of the flue wall bricks, where the densification varies significantly from the anode side to the flue side of the brick. The observed densification is proposed to be caused by high-temperature creep, and the aim of this work was to determine whether the uneven densification across the brick could be modeled using a finite element method (FEM) implementing high-temperature steady-state creep. Finite element modeling was used to model steady-state creep for a material similar to that used in the baking furnace. Thermal and physical parameters and boundary conditions were chosen to simulate the conditions in an anode baking furnace. Refractory samples of pristine and spent lining from the baking furnace were also analyzed with X-ray computed tomography (CT), with a reduction in the porosity confirming the densification during operation. The FEM modeling demonstrated that high-temperature creep could explain the observed densification in the spent flue walls. The present findings may be useful in relation to increasing the lifetime of industrial flue walls.
Abstract The semiconductors industry has put its eyes on two-dimensional (2D) materials produced by chemical vapour deposition (CVD) because they can be grown at the wafer level with small thickness fluctuations, which is necessary to build electronic devices and circuits. However, CVD-grown 2D materials can contain significant amounts of lattice distortions, which degrades the performance at the device level and increases device-to-device variability. Here we statistically analyse the quality of commercially available CVD-grown hexagonal boron nitride (h-BN) from the most popular suppliers. h-BN is of strategic importance because it is one of the few insulating 2D materials, and can be used as anti-scattering substrate and gate dielectric. We find that the leakage current and electrical homogeneity of all commercially available CVD h-BN samples are significantly worse than those of mechanically exfoliated h-BN of similar thickness. Moreover, in most cases the properties of the CVD h-BN samples analysed don’t match the technical specifications given by the suppliers, and the sample-to-sample variability is unsuitable for the reproducible fabrication of capacitors, transistors or memristors in different batches. In the short term, suppliers should try to provide accurate sample specifications matching the properties of the commercialized materials, and researchers should keep such inaccuracies in mind; and in the middle term suppliers should try to reduce the density of defects to enable the fabrication of high-performance devices with high reliability and reproducibility.
Fabricating non-noble metal-based carbon air electrodes with highly efficient bifunctionality is big challenge owing to the sluggish kinetics of oxygen reduction/evolution reaction (ORR/OER). The efficient cathode catalyst is urgently needed to further improve the performance of rechargeable zinc-air batteries. Herein, an activation-doping assisted interface modification strategy is demonstrated based on freestanding integrated carbon composite (CoNiLDH@NPC) composed of wood-based N and P doped active carbon (NPC) and CoNi layer double hydroxides (CoNiLDH). In the light of its large specific surface area and unique defective structure, CoNiLDH@NPC with strong interface-coupling effect in 2D-3D micro-nanostructure exhibits outstanding bifunctionality. Such carbon composites show half-wave potential of 0.85 V for ORR, overpotential of 320 mV with current density of 10 mA cm−2 for OER, and ultra-low gap of 0.70 V. Furthermore, highly-ordered open channels of wood provide enormous space to form abundant triple-phase boundary for accelerating the catalytic process. Consequently, zinc-air batteries using CoNiLDH@NPC show high power density (aqueous: 263 mW cm−2, quasi-solid-state: 65.8 mW cm−2) and long-term stability (aqueous: 500 h, quasi-solid-state: 120 h). This integrated protocol opens a new avenue for the rational design of efficient freestanding air electrode from biomass resources.
Gregor Pieplow, Cem Güney Torun, Charlotta Gurr
et al.
The detection of individual charges plays a crucial role in fundamental material science and the advancement of classical and quantum high-performance technologies that operate with low noise. However, resolving charges at the lattice scale in a time-resolved manner has not been achieved so far. Here, we present the development of an electrometer with 60 ns acquisition steps, leveraging on the spectroscopy of an optically-active spin defect embedded in a solid-state material with a non-linear Stark response. By applying our approach to diamond, a widely used platform for quantum technology applications, we can distinguish the distinct charge traps at the lattice scale, quantify their impact on transport dynamics and noise generation, analyze relevant material properties, and develop strategies for material optimization.
Using Monte Carlo electronic transport simulations, coupled self-consistently with the Poisson equation for electrostatics, we explore the thermoelectric power factor of nanoengineered materials. These materials consist of alternating highly doped and intrinsic regions on the scale of several nanometers. This structure enables the creation of potential wells and barriers, implementing a mechanism for filtering carrier energy. Our study demonstrates that by carefully designing the nanostructure, we can significantly enhance its thermoelectric power factor compared to the original pristine material. Importantly, these enhancements stem not only from the energy filtering effect that boosts the Seebeck coefficient but also from the utilization of high-energy carriers within the wells and intrinsic barrier regions to maintain relatively high electronic conductivity. These findings can offer guidance for the design and optimization of new-generation thermoelectric materials through improvements in the power factor.
We show that single-pulse burst fabrication will produce a flatter and smoother profile of axicons milled on sapphire compared to pulse overlapped fabrication which results in a damaged and much rougher surface. The fabrication of large-area (sub-1 cm cross-section) micro-optical components in a short period of time (∼10 min) and with less processing steps is highly desirable and would be cost-effective. Our results were achieved with femtosecond laser fabrication technology which has revolutionized the field of advanced manufacturing. This study compares three configurations of axicons such as the conventional axicon, a photon sieve axicon (PSA) and a sparse PSA directly milled onto a sapphire substrate. Debris of redeposited amorphous sapphire were removed using isopropyl alcohol and potassium hydroxide. A spatially incoherent illumination was used to test the components for imaging applications. Non-linear reconstruction was used for cleaning noisy images generated by the axicons.
Abstract Background Metabolism reprogramming plays a vital role in glioblastoma (GBM) progression and recurrence by producing enough energy for highly proliferating tumor cells. In addition, metabolic reprogramming is crucial for tumor growth and immune‐escape mechanisms. Epidermal growth factor receptor (EGFR) amplification and EGFR‐vIII mutation are often detected in GBM cells, contributing to the malignant behavior. This study aimed to investigate the functional role of the EGFR pathway on fatty acid metabolism remodeling and energy generation. Methods Clinical GBM specimens were selected for single‐cell RNA sequencing and untargeted metabolomics analysis. A metabolism‐associated RTK‐fatty acid‐gene signature was constructed and verified. MK‐2206 and MK‐803 were utilized to block the RTK pathway and mevalonate pathway induced abnormal metabolism. Energy metabolism in GBM with activated EGFR pathway was monitored. The antitumor effect of Osimertinib and Atorvastatin assisted by temozolomide (TMZ) was analyzed by an intracranial tumor model in vivo. Results GBM with high EGFR expression had characteristics of lipid remodeling and maintaining high cholesterol levels, supported by the single‐cell RNA sequencing and metabolomics of clinical GBM samples. Inhibition of the EGFR/AKT and mevalonate pathways could remodel energy metabolism by repressing the tricarboxylic acid cycle and modulating ATP production. Mechanistically, the EGFR/AKT pathway upregulated the expressions of acyl‐CoA synthetase short‐chain family member 3 (ACSS3), acyl‐CoA synthetase long‐chain family member 3 (ACSL3), and long‐chain fatty acid elongation‐related gene ELOVL fatty acid elongase 2 (ELOVL2) in an NF‐κB‐dependent manner. Moreover, inhibition of the mevalonate pathway reduced the EGFR level on the cell membranes, thereby affecting the signal transduction of the EGFR/AKT pathway. Therefore, targeting the EGFR/AKT and mevalonate pathways enhanced the antitumor effect of TMZ in GBM cells and animal models. Conclusions Our findings not only uncovered the mechanism of metabolic reprogramming in EGFR‐activated GBM but also provided a combinatorial therapeutic strategy for clinical GBM management.
Neoplasms. Tumors. Oncology. Including cancer and carcinogens
Looking back on the short and yet very rewarding history that has shaped the game-changing lithium-ion batteries over the past three decades, among the main technical milestones and on-going challenges are the energy density, cycling-ability and yet poor safety, which are largely related to the use of organic liquid electrolytes. While the holey grail for energy storage is all-solid-state batteries, there is the inevitable transition from liquid electrolytes to solid ones, in the long term. Nevertheless, currently known solid-state electrolytes face problems of their own and also the largely incompatible interfaces with the solid electrodes. Then an interesting question arises: At least in the near-future term, what would be the electrolytes that can replace the organic liquid electrolytes? In this short Comment, we examine the transition from liquid electrolytes to solid states, where a potential newcomer is the frameworked, which is expected to address the sluggish kinetics of ion transport and poor interfaces with solid electrodes.
Seung‐Hwan Kwon, Prof. Dr. Sang‐il Kim, Prof. Dr. Weon Ho Shin
et al.
Abstract Zintl compounds are promising thermoelectric materials for power generation as their electronic and thermal transport properties can be simultaneously engineered with anion/cation alloying. Recently, a peak thermoelectric figure‐of‐merit, zT, of 1.4 was achieved in a (Yb0.9Mg0.1)Cd1.2Mg0.4Zn0.4Sb2 Zintl phase at 700 K. Although the effects of alloying Zn in lattice thermal conductivity had been studied thoroughly, how the Zn alloying affects its electronic transport properties has not yet been fully investigated. This study evaluates how the Zn alloying at Cd sites alters the band parameters of (Yb0.9Mg0.1)Cd1.6−xMg0.4ZnxSb2 (x=0‐0.6) using the Single Parabolic Band model at 700 K. The Zn alloying increased the density‐of‐states effective mass (md*) from 0.87 to 0.97 m0. Among Zn‐alloyed samples, the md* of the x=0.4 sample was the lowest (0.93 m0). The Zn alloying decreased the non‐degenerate mobility (μ0) from 71 to 57 cm2 s−1 V−1. Regardless of Zn alloying content, the μ0 of the Zn‐alloyed samples were similar (∼57 cm2 s−1 V−1). Consequently, the x=0.4 with the highest zT exhibited the lowest weighted mobility (μW). The lowest μW represents the lowest theoretical electronic transport properties among other x. The highest zT at x=0.4 despite the lowest μW was explained with a significant lattice thermal conductivity reduction achieved with Zn alloying with x=0.4, which outweighed the deteriorated electronic transport properties also due to the alloying.