Search Results (295)

Large language models (LLMs) built upon the Transformer architecture have achieved remarkable performance in natural language understanding, text generation and logical reasoning, while their internal working mechanisms remain poorly interpreted. This paper establishes a systematic mathematical analysis framework tailored for decoder-only Transformer LLMs, based on linear algebra, tensor analysis, probability theory, information theory, optimization dynamics and geometric deep learning. We conduct rigorous mathematical modeling and theoretical deduction on core modules including word embedding, position encoding, self-attention, feed-forward networks, training optimization and generalization reasoning, and explore the mathematical nature of semantic representation, contextual correlation, knowledge storage and logical inference within models. In this paper, we strictly distinguish between classic established Transformer theories and our original mathematical derivations and conclusions. Distinct from existing fragmented theoretical studies, this work presents six targeted novel contributions beyond conventional Transformer theories: (1) we construct the first full-process unified mathematical framework covering all core modules and the entire lifecycle of Transformer-based LLMs; (2) we provide strict mathematical proof to verify that single-head self-attention is essentially a kernel weighted average operation in reproducing kernel Hilbert space and derive the low-rank and sparse properties of attention weights; (3) we establish a high-dimensional non-convex optimization dynamics model for pre-training and mathematically prove that model training converges to flat local minima; (4) we derive a tighter upper bound of generalization error and quantify the quantitative relationship among model parameters, sequence length, training data scale and generalization performance; (5) we characterize the latent space as a low-curvature smooth Riemannian manifold and model logical reasoning as geometric transformation on this manifold; (6) we design multi-group controlled experiments on mainstream datasets to quantitatively validate all above theoretical conclusions. This paper further summarizes the inherent mathematical limitations of current Transformer LLMs and proposes feasible theoretical optimization paths, referring to state-of-the-art research published from 2021 to 2026. The outcomes of this research can provide solid mathematical theoretical support for improving model interpretability, optimizing network structures and boosting practical performance, and facilitate the transition of LLM research from empirical engineering practice to theory-driven development. Full article

This study presents a systematic terahertz time-domain spectroscopy (THz-TDS) investigation of hardened cement paste, framed as a complex-optical measurement in which the real and imaginary parts of the response probe distinct microstructural attributes. Transmission-mode measurements were made on pastes with water-to-cement (w/c) ratios of 0.3, 0.4, and 0.5 at curing ages of 7, 14, 28, and 56 days. The effective refractive index, obtained from the time-domain pulse delay (7, 28, and 56 days, paired with mercury intrusion porosimetry), correlates strongly and linearly with porosity over nine porosity-paired conditions spanning 15.1–30.4% (pooled R² = 0.94, p < 0.001). In a quasi-static effective-medium framework—where the pores a re far smaller than the THz wavelength—this reflects the dependence of the effective permittivity on the solid volume fraction: the Bruggeman model outperforms the Maxwell–Garnett model, and all data fall within the Wiener bounds, lying close to the upper bound, indicating a continuously connected solid matrix with isolated pores. Cross-validated porosity estimation is reliable to within about ±2 percentage points (refractive-index uncertainty ±0.02–0.04). The absorption follows a power law (β ≈ 1.0–1.3) characteristic of disorder-activated vibrational absorption, in which the loss of long-range order in the amorphous C–S–H relaxes the crystalline selection rules and couples the THz field to the full vibrational density of states. The refractive index (structure-sensitive, governed by volume fraction) and the absorption (material-sensitive, governed by solid disorder; estimated loss tangent of order 0.1) thus form two complementary channels. Combining the THz-derived porosity with the Powers hydration model gives a degree of hydration consistent with literature ranges—an indirect comparison rather than direct validation. These results establish THz-TDS as a non-contact, non-ionizing technique for rapid porosity estimation and hydration assessment of cementitious materials. Full article

The trend in the power electronics industry toward higher power density and efficiency has brought high-frequency transformers (HFTs) to the forefront of critical applications, including isolated DC–DC converters, electric vehicle chargers, and solid-state transformers. This paper focuses on the leakage inductance of HFTs and presents a systematic comparative framework that evaluates five surrogate modeling and hybrid optimization approaches for the rapid and accurate estimation of leakage inductance. A comprehensive parametric dataset was constructed, comprising 1210 finite element analysis simulations conducted via finite element analysis in the ANSYS Maxwell 2024 R1 environment, varying the number of winding turns, primary winding thickness, and secondary winding thickness of the HFT. All five methods were trained and evaluated on the same dataset under identical conditions. The comparative evaluation demonstrates that the proposed hybrid Gray Wolf optimizer–artificial neural network (GWO-ANN) framework achieved the highest prediction accuracy (R² = 0.9832, MSE = 0.01780, MAE = 0.0935 µH) and the fastest convergence among all tested approaches. The generalization capability of the proposed model was confirmed through blind validation tests across six geometric configurations spanning the full range of the design space, yielding a maximum prediction error of 8.15% and an average error of 2.14%. The functional validity of the proposed parameters was further tested in a third validation layer using MATLAB/Simulink R2024b transformer circuit studies, demonstrating a theoretical efficiency of 96.06%. This three-layer validation approach proves both the parametric and functional reliability of the proposed framework for HFT designs. Full article

TIDEs (therapeutic peptides, oligonucleotides, and related molecules) represent a rapidly expanding market that has gained significant momentum due to the recent success of Glucagon-like peptide-1 (GLP-1) receptor agonists for the treatment of obesity, diabetes and as cardiovascular and kidney diseases. Chemical synthesis remains the dominant manufacturing route for candidates containing approximately 10–40 amino acids and includes non-proteinogenic amino acids. Consequently, various combinations of solid-phase peptide synthesis (SPPS), liquid-phase peptide synthesis (LPPS), hybrid approaches, or tag-assisted peptide synthesis (TAPS) can be applied to achieve full-sequence assembly. However, identifying the most eco-efficient pathway through experimental trials alone is impractical because of the vast number of possible process combinations and the growing variety of green solvent alternatives. Therefore, process simulation studies—widely established in chemical engineering—must be adapted to the specific physicochemical characteristics of these large, multi-component molecules. This paper provides an overview of the current state of research and illustrates potential process improvements enabled by digital twin technologies as exemplified for the first manufacturing steps of tirzepatide. Full article

Mining activities generate substantial volumes of solid waste materials during exploration and processing. These residuals pose environmental and geotechnical concerns due to their large spatial footprints and associated risks but may also contain potentially valuable resources. These characteristics highlight the necessity and opportunity of effective management and monitoring strategies. In recent years, a diverse range of technologies and methods have been applied to characterize mine waste compositions and analyze their spatial–temporal variability. These include remote sensing systems, ground-based sensors, and advanced data-driven methods. Despite the rapid advancement, the existing literature provides limited insight into the critical evaluation of how these techniques are applied in practice. This review systematically examines peer-reviewed journal articles published between 2021 and 2024 to highlight the state of the art in characterization, modeling, and monitoring techniques for mine waste. The review identifies recent trends, key gaps, advantages, and limitations of these techniques. The summary suggests that mining companies and research communities are increasingly adopting innovative technologies, transitioning from conventional methods to more sustainable practices. However, it also reveals ongoing challenges and persistent limitations. Further efforts, such as real-time monitoring capabilities, are required to achieve full implementation and integration across the industry and academia. Full article

The design of active pneumatic upper limb exoskeletons is complicated by the challenge of reliably determining a kinematically safe workspace. Existing analytical kinematic methods are not sufficient to predict geometric collisions between elements of closed kinematic chains, which poses risks of mechanical damage and threats to user safety during exoskeleton operation. This paper proposes a hybrid algorithm for verifying the workspace of a pneumatic exoskeleton, combining analytical modelling in MATLAB R2020b based on the Product of Exponentials (PoE) method with high-performance static simulation in the Unity environment. At the initial stage, a discrete set comprising 758 million positions of the upper exoskeleton manipulator was generated. Subsequently, a multithreaded two-stage filtering process was implemented: analytical verification of rod stroke limits and angular constraints, followed by the detection of physical intersections of solid-state meshes using the PhysX engine. The results indicate that while the analytical model filters out 99.6% of invalid configurations. Yet, among the remaining positions—formally correct from a mathematical standpoint—up to 50% lead to critical geometric collisions or breaks in the kinematic chain. The computational efficiency of the proposed architecture enabled full static workspace verification in under 20 min. A reachable zone topology was established, revealing pronounced asymmetry and the presence of a “manoeuvrability core” in the user’s anterior hemisphere. The developed algorithm generates a verified set of kinematically safe exoskeleton states, providing a foundation for the kinematic safety layer of a hierarchical control system. These findings demonstrate the necessity of complementing analytical kinematics with physical collision detection when designing hybrid kinematic mechanisms, and the approach can be applied to verify collision-free movement trajectories in various robotic systems. The approach can be applied to verify collision-free movement trajectories in simulation, with physical validation deferred to future work. Full article

The taphole is the only visible window for observing the blast furnace hearth state, and hot metal flow carries key hearth information. To address the problems of current hot metal flow monitoring, such as reliance on manual work, difficulty in quantification, poor real-time performance, as well as insufficient perception stability and low data utilization in existing research, this study proposes a full-chain intelligent solution for blast furnace taphole hot metal flow monitoring: by building an image acquisition system adapted to extreme working conditions, selecting ResNet50 as the state perception model, and combining Canny edge detection with the local morphological extremum analysis algorithm to extract core contour parameters; supplemented by the anti-vibration self-adjustment algorithm and the multi-taphole automatic switching strategy, the robustness and operation efficiency of the system are significantly improved. On this basis, a coke sticking early-warning model is constructed, splashing in different periods is quantitatively classified, and the spatiotemporal difference in hot metal flow is revealed. Finally, a full-chain technical system of “data acquisition–intelligent perception–working condition diagnosis–decision support” is formed, which promotes the digital and intelligent upgrading of hot metal flow monitoring and provides solid support for the safe operation. Full article

Reconfigurable radio frequency (RF) attenuators are critical passive components for 5G-Advanced and emerging 6G wireless systems. Conventional tunable attenuators rely on solid-state switches combined with fixed resistor networks, which suffer from unavoidable static power consumption and severe parasitic degradation at high frequencies. Here, we systematically demonstrate HfO₂-based non-volatile memristors as RF switches with tunable ON-state resistance (R_ON), enabling a switching-attenuation-integrated multistate attenuator. The fabricated Au/HfO₂/Ag devices exhibit stable bipolar resistive switching with an ON/OFF ratio exceeding 10⁹, reliable retention of 10⁵ s, and programmable R_ON continuously tuned from 5.8 Ω to 197.5 Ω. On-wafer RF characterizations from 10 MHz to 43.5 GHz reveal low insertion loss (−0.53 dB), high isolation (−26.8 dB), and clear scaling laws governing the effects of device geometry and R_ON on RF performance. Leveraging these unique characteristics, we propose a symmetric π-type programmable all-memristor attenuator architecture with a cascaded 2-unit configuration. The design achieves 12 discrete attenuation levels from 2 dB to 24 dB, a return loss better than 10 dB across the full band, and zero static power consumption without additional passive components or bias networks. This work establishes the fundamental material-device-RF performance relationship in HfO₂-based RF switches and provides a compact, low-power, and highly integrable solution for next-generation reconfigurable RF front-ends. Full article

The development of reliable and scalable control software is a key requirement for the Cherenkov Telescope Array Observatory, where distributed subsystems must operate coherently and support increasingly automated observing strategies. This paper presents the architecture and design of the Telescope Control System of the Small-Sized Telescopes of the observatory, addressing the need for modularity, deterministic behavior, and long-term maintainability. The proposed solution adopts a set of software managers implementing well-defined interfaces and state machines, enabling predictable control flows and consistent interaction with heterogeneous hardware. Modern software engineering practices were applied, including containerized services, automated deployment workflows, and a comprehensive simulation environment. These elements were evaluated through prototypes and pathfinder activities that allowed us to explore design alternatives, validate the behavior of individual components, and assess the scalability of the overall architecture. Results from these exploratory tests indicate that the interface-driven and modular design supports robust operation, facilitates integration, and reduces the effort required for system evolution. While full implementation is currently in progress, the findings confirm that the proposed architecture provides a solid foundation for the test readiness review phase (the phase preceding formal integration testing) and can be effectively extended to future facilities requiring flexible, maintainable, and resilient control software. Full article

The utilization of coal-based solid waste materials (CSW) in road engineering is an important pathway for reducing stockpiling pressure, mitigating environmental risks, and promoting resource recycling. However, their large-scale diffusion is still constrained by residual engineering risk, misaligned cost and risk allocation between upstream and downstream actors, and imperfect regulatory and incentive mechanisms. To address these issues, this study develops a tripartite evolutionary game model involving the regulator, the waste producer, and the waste utilizer. The model incorporates pretreatment investment, residual engineering risk, government rewards and penalties, and green collaborative benefits to examine the evolutionary dynamics of the three parties and the stability of the system under different conditions. The results show that deep pretreatment by waste producers is a key prerequisite for the diffusion of CSW materials, as it reduces material instability and downstream engineering risk and increases the utilizer’s willingness to adopt such materials. The effects of rewards and penalties are differentiated across actors: effective penalties play a stronger role in constraining low-cost disposal by waste producers, whereas rewards are more effective in encouraging adoption by waste utilizers. The interaction analysis further shows that residual engineering risk significantly constrains the positive effect of green collaborative benefits, indicating that benefit enhancement cannot substitute for risk governance. In addition, the total amount of green collaborative benefits and their release and distribution structure jointly affect behavioral convergence and system stability. The system is more likely to evolve toward a stable state characterized by deep pretreatment, active adoption, and routine regulation when benefit sharing is consistent with the costs and risks borne by each party. Based on these findings, this study suggests that differentiated policy design is needed, including stronger source pretreatment and quality control, a coordinated reward–penalty mechanism for different actors, more targeted incentives and acceptance requirements for waste utilizers, and an improved governance framework featuring quality standards, full-process traceability, and risk warning mechanisms. These measures are essential for promoting the stable and large-scale utilization of CSW materials in road engineering. By translating model results into staged regulatory, quality-control, and supply-chain actions, the findings also support broader sustainable development goals, including responsible consumption and production, resilient infrastructure, climate action, and ecosystem protection. Full article

High-performance full-color displays and white lighting require stable and efficient red, green, and blue emitters; however, they are often limited by wide bandgaps, imbalanced carrier injection/transport, complex device structures, and high material costs. To address these challenges, we designed and synthesized a multifunctional deep-blue molecule (PPI-F-PO) integrating a phenanthroimidazole moiety, a 9,9-diphenylfluorene unit, and a phosphine oxide group. The twisted structure of fluorene, featuring a sp³-hybridized carbon, effectively suppresses conjugation extension and aggregation-caused quenching, whereas the electron-withdrawing phosphine oxide group enhances electron transport. Consequently, it exhibits good thermal stability, high solid-state photoluminescence quantum yield (58.8%), and high triplet energy (E_T = 2.54 eV). Non-doped blue OLEDs based on this emitter achieve a maximum external quantum efficiency (EQE) of 2.52% with deep-blue CIE coordinates of (0.16, 0.06). Moreover, using this material as a host, green and orange-red phosphorescent OLEDs exhibit maximum EQEs of 15.4% and 9.7%, respectively, along with low efficiency roll-off. This work demonstrates that a bipolar deep-blue emitter with high triplet energy can act both as a high-efficiency standalone emitter and as a universal host for lower-energy phosphors, thereby simplifying device architecture and reducing material costs for full-color OLEDs. Full article

Geometric deviations are inevitably generated during tunnel lining construction. These deviations result from construction inaccuracies. They pose potential risks to long-term structural safety and engineering quality. Traditional numerical simulations are based on idealized design cross-sections. This approach is limited in reflecting actual mechanical behavior. In this study, a refined modeling and safety assessment method is developed. Construction-induced geometric deviations are incorporated into the analysis. Optimized geometric fitting and mesh reconstruction algorithms are employed. Large-scale irregular point cloud data are efficiently processed. A full-scale solid finite element model is constructed. Actual construction deviations are represented in this model. The results are systematically compared with those from the conventional design model. It is revealed that construction-induced geometric deviations alter internal force transmission paths. Asymmetric deformation is induced. Localized stress concentrations are observed. The ideal stress state is predicted by the design model. In contrast, stiffness degradation is observed in the as-built model. This degradation is significant in vulnerable regions such as the haunch on the heavily loaded side. A considerable reduction in the local safety factor is also observed. The overestimation of safety redundancy is quantified when geometric variations are neglected. The results indicate that incorporating field-measured point cloud data into structural simulations can improve the geometric realism of tunnel-lining assessment and assist in identifying potential high-risk zones. Full article

Limited ionic conductivity and unstable interfaces, primarily caused by poor solid–solid contact, pose significant challenges to the stable cycling of solid-state batteries. In this study, an interfacial in situ polymerization strategy is proposed to construct a poly(1,3-dioxolane) (PDOL) gel electrolyte layer between a poly(vinylidene fluoride) (PVDF)-based solid polymer electrolyte and the electrodes. This approach aims to address interfacial compatibility issues in solid-state lithium metal batteries. By precisely tuning the composition of the gel precursor and employing characterization techniques such as FTIR and NMR, the efficient ring-opening polymerization of 1,3-dioxolane (DOL) was confirmed, achieving a high conversion rate of 90%. The precursor was drop-cast onto the PVDF-based electrolyte/electrode interfaces before cell assembly. Electrochemical evaluations revealed that the in situ formed solidified interlayer significantly enhanced interfacial compatibility and ion transport, yielding a high Li⁺ transference number (0.341), an exceptional critical current density (1.4 mA cm⁻²), and remarkable cycling stability exceeding 1600 h in Li||Li symmetric cells. Furthermore, full cells incorporating LiFePO₄ cathodes demonstrated excellent rate capability and long-term cyclability, retaining 98.7% of their capacity after 1000 cycles. These results collectively underscore the effectiveness of this in situ solidification strategy in optimizing the interface structure and improving the overall performance of PVDF-based solid-state batteries. Full article

Metal-functionalized boron nitride nanostructures represent promising platforms for lightweight solid-state hydrogen storage. In this work, we perform a comprehensive density functional theory (DFT) investigation of pristine and metal-decorated B₄N₄ quantum dots (M = Li, Ti) to evaluate their structural stability, adsorption energetics, and near-ambient storage performance. Pristine B₄N₄ is highly stable but interacts weakly with H₂ (E_ads ≈ −0.12 eV), leading to negligible uptake under operating conditions. Li decoration moderately enhances adsorption through charge-induced polarization (E_ads ≈ −0.15 eV) but offers limited stabilization beyond the first few molecules. In contrast, Ti decoration fundamentally reshapes the interaction landscape, strengthening electrostatic, polarization, and dispersion contributions and enabling significantly stronger yet reversible H₂ binding (E_ads ≈ −0.36 eV). Sequential adsorption calculations predict maximum theoretical capacities of 14, 18, and 20 H₂ molecules for pristine, Li-, and Ti-decorated systems, respectively. Grand canonical thermodynamics show that Ti–B₄N₄ retains nearly its full loading at 30 bar and 298 K, while pristine and Li-decorated clusters store only negligible amounts. Under desorption conditions (3 bar, 373 K), Ti–B₄N₄ releases most of its stored hydrogen, yielding an exceptional reversible capacity of 15.1 wt%. Energy decomposition analysis attributes this performance to cooperative electrostatic, polarization, and dispersion enhancements. Ti–B₄N₄ emerges as a highly promising theoretical candidate, warranting future experimental validation. Full article

The formation of lithium-ion batteries (LIBs) directly affects the properties of the solid electrolyte interphase (SEI) layer, which in turn affects cell performance, lifetime, and safety. Therefore, measurement of SEI properties during formation is a topic of great interest for LIB manufacturing. EIS has previously been applied to half-cell and three-electrode configurations for this purpose; however, these results have been questioned due to the potential non-linearity of the EIS measurement. Additionally, the limited application of the method to half cells and three-electrode cells limits the application of this method to production lines, where only two-electrode full cells are manufactured. In this work, we compare dynamic and steady-state EIS measurements during the formation cycling of NMC532/graphite coin cells. DRT analysis is used to distinguish the time constants of the two electrodes for equivalent circuit modeling. The main findings of this work are that dynamic EIS (DEIS) measurements only significantly affect the frequency response below ~30 Hz. Additionally, time constants and effective capacitance are unaffected by DEIS. We conclude that DEIS remains a promising technique for studying SEI formation in a two-electrode configuration and may be applicable on production lines for rapid diagnostics or even tracking SEI growth in real time. Full article