Yan Wang

Abstract Dual-frequency parametric excitation widely exists in coupled systems. The action of the contact wire on the pantograph in the pantograph–catenary coupled system can be simplified as a typical dual-frequency parametric excitation. The two disproportionate parametric excitation frequencies cause the stiffness of the nonlinear springs to change in quasi-periodic, which make the dynamic characteristics of the pantograph–catenary system more complicated and affect collect currents steadily of the train. In this paper, the dynamics model of pantograph–catenary under dual-frequency parametric excitation is established, and the analytical results of the system stability boundary are obtained by the harmonic balance method. The accuracy of the results with the harmonic balance method in this paper is verified by comparing with the numerical iteration results. Finally, the stability boundary of the pantograph–catenary with different frequency ratios is determined according to the stiffness variation coefficient of the contact wire and the train speed, and the influence of the frequency ratio on the stability boundary of the pantograph–catenary system is analyzed. The influence of the dual-frequency parametric excitation on the stability of the pantograph–catenary system caused by the joint effect of the stiffness fluctuation between the support poles and the vertical droppers during the train running is investigated. The influence of dual-frequency parametric excitation on the stability of the pantograph–catenary system is revealed, which provides guidance for the parameter design and vibration suppression of the pantograph–catenary system.

To address the limitation of using experimental parameters in the macroscopic strength criterion, a micromechanical strength criterion for root-reinforced soil is developed. In this model, a micromechanical model for a three-phase composite (“root—cemented soil matrix—frictional element”) is constructed, and the novel combination of energy equivalence principles with the M-T method is used to determine the meso-scale prestress and strength criterion for root-reinforced soil under freeze–thaw cycles. The representative volume element (RVE) of root-reinforced soil is conceptualized as a composite material consisting of a bonded element (a cemented-soil matrix with root inclusions) and frictional inclusions. By applying micromechanics, along with the Mori–Tanaka method, the LCC method, limit analysis theory, and macro–micro energy equivalence principles (incorporating both strain and dissipated energy), a micromechanical strength criterion is formulated, revealing failure mechanisms at the microscale. The previously used stepwise procedure for deriving the stationary function is improved, and the microscale prestress is determined through the Mori–Tanaka method combined with macro–micro strain-energy equivalence. The proposed micromechanical strength criterion effectively models the primary strength variation in root-reinforced soil under freeze–thaw cycles, extending the existing shear criterion for soil.

An injectable water-responsive hierarchical porous shape-memory material was prepared, featuring ultrahigh blood absorption (>4600%), ultrafast shape recovery (∼6 s), robust antibacterial activity (>99.9%), and coagulation-independent hemostasis.

The paper proposes a method for simulating ghosts in the lens optical systems. The method consists of a vector approach to the formation of a block of optical systems for which the PSF calculation is performed. Also, a method of adaptive selection of points on the entrance pupil with possible interpolation of the ray coordinates on the image plane was implemented. This solution allows for an order of magnitude reduction in the number of traced rays required for correct evaluation of the PSF. A metric for evaluating the effect of ghosts on the quality of the resulting image was proposed. The paper presents the results of simulation and evaluation of the PSF for a number of cases of light interreflections in optical system.

The effectiveness of diagnosis-related groups (DRG) system is pivotal to the implementation of medical insurance payment standards. However, existing methods for constructing DRGs face challenges such as violations of grouping rules and imbalances in multiobjective optimization, which limit their ability to support payment standards that accurately reflect clinical complexities. To address these challenges, this paper proposes a multiconstraint multiobjective optimization model and a tree-structured multiobjective optimization framework, grounded in interpretability theory. The model mathematically defines the dual objectives of enhancing intragroup homogeneity and intergroup heterogeneity, subject to grouping rules. The framework utilizes nonnegative adaptive LASSO regression to accurately quantify clinical complexity, while integrating tree structures with multiobjective optimization algorithms to generate Pareto-optimal DRG sets by solving the model. Empirical results demonstrate that the proposed framework satisfies the grouping constraints and effectively reflects clinical complexity. This framework is expected to provide a paradigm for constructing DRGs, offering decision-makers efficient DRG sets.

The significant volatility of distributed generation and the uncoordinated charging behavior of Electric Vehicles (EVs) exacerbate the peak-valley disparity in industrial park distribution networks, adversely affecting the stable operation of power systems. To address this issue, this paper proposes a two-stage optimal scheduling strategy for peak shaving and valley filling, taking into account Photovoltaic (PV) systems, EVs, and Battery Energy Storage Systems (BESS). In the first stage, a Long Short-Term Memory (LSTM) network forecasts the short-term PV output. Based on the forecasted PV generation and time-of-use electricity pricing, a heuristic rule-based weight allocation method is introduced to determine the charging schedule of EVs. In the second stage, considering charging/discharging efficiency, capacity constraints, and the full life-cycle cost of BESS, an optimal scheduling model is developed to minimize the variance of grid load. An improved hybrid Genetic Algorithm (GA) is adopted to solve the model, effectively reducing peak-valley load differences. The proposed method is evaluated using real load data from an industrial park, demonstrating that the peak load is reduced from 958.4 to 917.2 kW, while the valley load increases from 61 to 124.5 kW, achieving an 11.7% reduction in peak-valley disparity and significantly improving the load curve.

Icing phenomena occur on aircraft and unmanned aerial vehicles (UAVs) under extreme weather conditions. Ultrasonic detection technology is an effective method for measuring ice formation while maintaining the shape of the structure. However, current ultrasonic sensors, which are large and inflexible, are unsuitable for irregular UAV bodies, limiting their applications in real scenarios. For the detection of icing on curved structure, this study proposes a novel flexible ultra-thin ultrasonic transducer (FUTUT). The transducer exhibits excellent flexibility, making it suitable for use on high-curvature wings. Firstly, the FUTUT was designed based on the material properties of the airframe and the sensitivity requirements for ice detection, following the design guidelines for 1-3 type piezocomposites. The fabrication process for the FUTUT was then investigated, and its flexibility and low-temperature resistance were tested. Finally, icing detection experiments were conducted in an icing wind tunnel (IWT), where the FUTUT of 9.82 MHz demonstrated an ice-thickness-detection sensitivity of 0.29 mm. Experimental results indicate that the FUTUT possesses superior flexibility and exhibits excellent stability in low-temperature environments. These results underscore the FUTUT’s promise for applications in ice detection on curved structures.

In the conventional preparation of medium-manganese high-strength steel, hot rolling or cold rolling processes are primarily employed, interspersed with isothermal tempering to ensure the mechanical properties of the steel. This study aims to enhance austenite stability by controlling the partitioning time to promote carbon diffusion from supersaturated martensite into retained austenite. Subsequent cold rolling induces deformation-induced martensitic transformation in the metastable austenite, converting it into newly formed martensite. Concurrently, this process generates defects such as twin dislocations within the retained austenite, enhancing its mechanical stability and refining the microstructure. Finally, tempering eliminates residual stresses to restore toughness while promoting partial carbon segregation from the newly formed martensite back into the retained austenite, ultimately stabilizing a portion of the austenite. Through an optimized quenching-alloying-cold rolling-tempering process, Fe-10Mn-0.4C steel with outstanding strength and toughness was successfully produced. Research indicates that under optimal process conditions, this steel grade achieves a yield strength of 1450 MPa, a tensile strength of 1720 MPa, and an elongation of 22.5%. Short partitioning times enhance the stability of retained austenite, whereas excessively long partitioning times lead to performance deterioration.

The representative volume element (RVE) of root-reinforced soil is idealized as a composite comprising a bonded element, formed by a cemented-soil matrix with root inclusions, and inclusions of frictional elements. Based on micromechanics, and with the aid of the Mori–Tanaka method, the LCC method, limit analysis theory, and macro–micro energy-equivalence principles (including both strain energy and dissipated energy), a micromechanical strength criterion for root-reinforced soil is established, revealing the failure mechanisms at the microscale. The previously used stepwise procedure for determining the stationary functional is improved, and the microscale prestress is obtained using the Mori–Tanaka method together with macro–micro strain-energy equivalence. The proposed micromechanical strength criterion reproduces well the strength variation of root-reinforced soil under freeze–thaw cycling.

ABSTRACT We investigate high‐order, energy‐dissipative schemes for simulating the directed self‐assembly (DSA) of diblock copolymer melts, which plays a crucial role in materials science, nanotechnology, and soft matter. The DSA process is primarily modulated through strategies such as external physical fields, substrate interfacial engineering, and geometric confinement, all of which are fundamentally described by Ohta–Kawasaki energy functionals. To preserve the intrinsic energy dissipation property of the corresponding gradient flow equations, we develop implicit‐explicit Runge‐Kutta (IMEXRK) schemes of up to third order and approximately fourth order, ensuring energy stability for any time step size. Under the uniform boundedness assumption of solutions, a novel criterion to justify the energy stability is established by rewriting the IMEXRK schemes in a unified matrix‐vector framework. To address the time delay effect in stabilization schemes, an optimized, time‐step‐dependent selection of stabilization parameters is proposed, which shows significant accuracy improvement compared to a constant stabilization parameter. Numerical experiments validate the superior accuracy and stability, while simulations of directed self‐assembly of diblock copolymer melts under different situations demonstrate the universality and broad applicability of the proposed schemes.

BACKGROUND AND OBJECTIVE Solid pancreatic lesions (SPLs) represent one of the most lethal forms of gastrointestinal malignancies, and Rapid on-site evaluation (ROSE) serves as an important component of intraoperative diagnosis. However, efficient and accurate ROSE slide interpretation remains challenging due to the gigapixel scale of whole-slide images, sparse distribution of diagnostically relevant regions, and the need for real-time feedback. METHODS To address challenges, we propose a novel two-stage framework for fast and precise ROSE WSI classification, following the clinical diagnostic workflow of cytopathologists. In the first stage, we design a lightweight Transformer-based object detection network named as RoF DETR, which detects key cell clusters at 5x magnification. To further enhance detection performance, we incorporate domain-specific medical foundation model features and design a multi-scale feature fusion module for effective feature extraction. In the second stage, we design a prototype-guided multiple instance learning network (PG-MIL) based on pseudo-bag augmentation for 20x magnification patch extraction, improving feature discrimination and robustness under class imbalance. RESULTS For comprehensive evaluation, we establish a dedicated ROSE WSI dataset and a cell cluster detection dataset. Our method achieves an AP@0.5 of 0.482 in cell cluster detection and an AUC of 92.36% in WSI-level classification. Compared to conventional WSI-level classification pipelines, the proposed framework reduces computational overhead by approximately 100× and halves the inference time. CONCLUSION The proposed framework provides a scalable and efficient solution for rapid cytological assessment of ROSE slides, showing potential to support real-time intraoperative decision-making in clinical workflows.

• Integration and future directions of technologies required for mass production of cultured meat. • Expanding the possibility of chemically pluripotent stem cells for cultured meat seed cell banks. • The establishment of the current low-cost serum-free cell culture technology system and the precise regulation technology and rational amplification of the magnitude in large-scale cell culture. • Cost breakdown of key stages in large-scale cultured meat production and current commercial development status. In summary, this review provides a comprehensive guide and valuable insights for researchers and companies in the field of the future of cultured meat mass production. Cultured meat offers an efficient, safe, and sustainable solution for the supply of animal protein; however, due to the lack of systematic research and the integration of mass production technology, no products are currently available on the market on a large scale. This review summarizes key factors for mass cultured meat production and discusses opportunities, challenges, and future directions of key technologies: Constructing a cultured meat seed cell repository to support basic research and application of cultured meat; establishment of a low-cost serum-free cell culture technology system and updates on edible biomaterial scaffold in order to reduce the cost of cultured meat; development of a large-scale bioreactor for cultured meat, realizing precise regulation of key culture parameters and rational amplification of large-scale cell culture, and food processing techniques were the foundation for realizing product iterative upgrades and market acceptance. In summary, this review offers insights into cultured meat production. Schematic diagram of the general process of cultured meat production and traditional animal husbandry. A. Seed cells for cultured meat. In this step, cells were isolated from different sources, including chicken, cow, porcine, ovine or fish. B. Large-scale cell expansion. In this step, cells were expanded through different methods. C. Tissue maturation. In this step, cells undergo maturation and differentiation on scaffolds. D. Processing into food products. The last step the matured tissue was converted into final product.

A unique coherent bremsstrahlung (CB) regime is proposed for the generation of an isolated half-cycle attosecond pulse (AP), even for the case of a multicycle driving laser pulse, which is realized by the laser pulse interacting with the double-foil target. When the rising edge of the laser pulse interacts with the double-foil target, the AP train is generated in the reflected direction due to the fact that the relativistic electron sheet (RES) is generated periodically twice per cycle of the laser pulse. While the peak cycle of the laser pulse interacts with the target, the electrons of the second foil target were nearly completely removed, which leads to the formation of an electrostatic potential well, and then the double-foil target becomes an electrostatic storage target (EST) that can capture and store electrons coming from the second foil target. Accordingly, only one RES can be generated by the peak cycle of the laser pulse interacting with those electrons stored in the EST, and is accelerated to ultra-relativistic velocity in the transmission direction, which emits the isolated AP by the CB regime. The isolation regime of AP can be determined by the scaling law of the depletion of electrons, which is described by the ratio of the areal charge density of the double-foil target and the amplitude of the peak-cycle laser pulse. The ratio is called the generalized similarity parameter. The generalized similarity parameter is about 2/π, which is predicted by the theoretical model and confirmed by the simulation results. The robustness of the isolation regime of AP is confirmed by considering different parameters of the laser pulse and plasmas, and by considering the presence of preplasma and two-dimensional effects.

Despite extensive research on the effect of nitrogen flow rate on titanium nitride (TiN) coating properties, its influence on esthetic and engineering performance through microstructure control remains insufficiently explored. To simultaneously meet the requirements for surface strengthening and decorative esthetics in high-end stainless-steel crafts, TiN coatings were deposited on 304L stainless-steel substrates using multi-arc ion plating. The regulatory mechanisms and synergistic evolution laws of nitrogen flow rates (100, 200, and 300 sccm) on the microstructure, decorative properties (color and gloss), and engineering performance (adhesion strength, hardness, wear resistance, and corrosion resistance) of the coatings were investigated. At a nitrogen flow rate of 200 sccm, the coating exhibited a uniform and dense columnar crystal structure, as well as a saturated golden hue with high surface gloss. Additionally, the film–substrate adhesion, microhardness, tribological properties, and corrosion resistance reached optimal levels. In contrast, excessively low (100 sccm) or high (300 sccm) nitrogen flow rates resulted in coarse particles and blurred grain boundaries owing to uncontrolled droplet ejection or target poisoning, respectively, thereby deteriorating both appearance and engineering performance. These findings can inform the optimization of TiN coatings, enabling the design of surfaces that simultaneously meet esthetic and high-performance engineering requirements.

The large-scale utilization of renewable energy challenges the stability and safety of the grid; thus, the flexibility of coal-fired power plants should be increased to balance unstable renewable energies. To achieve this, a heat storage system (HSS) is integrated into a power plant. This is the first study utilizing furnace flue gas to drive a molten-salt-heat-exchanger (MSHE). Compared to steam-vapor-driven MSHE, flue gas-driven technology avoids the pinch temperature limitation (PTL) and simplifies the system configuration. In this study, we demonstrate the concept, design, fabrication, and experiments of the MSHE. The novelties include: ① finned tubes to balance the thermal resistances between the flue gas side and the molten salt side; ② a weak angle design to ensure gravity-driven recession of the molten salt; and ③ a modular design to ensure even temperature distribution at the outlet of the tube bundles. A heat transfer correlation is developed for molten salt, covering a wide range of Reynolds numbers. An experimental setup is constructed to collect data and verify the effectiveness of the MSHE. The measured overall heat transfer coefficients matched the predictions well, with deviations of less than 10 %. The measured heat power reached 320 kW, exceeding the 300 kW design target. We demonstrate the heat transfer between the flue gas and molten salt to compensate for the heat release from the HSS to the environment, reducing electricity consumption in the standby stage of the system. The modular design of the MSHE ensures minimal temperature deviations of < 4 K among different tubes, avoiding local overheating-induced decomposition of the molten salt. Based on the 300 kW MSHE results, a 10 MW MSHE is designed, fabricated, and integrated into a 350 megawatt electric (MWe) coal-fired plant to achieve a higher load variation rate of 6 % Pe·min −1 for a coal-fired power plant.

Guillain-Barré syndrome (GBS) is a rare neurological disorder characterized by muscle weakness and paralysis. Although the exact etiology remains unclear, the current standard treatments include intravenous immunoglobulin (IVIG) and plasma exchange (PLEX) therapy. While the majority of GBS patients respond well to immunotherapy, some severe cases can be fatal. Efgartigimod, an Fc receptor antagonist, has been utilized in the treatment of various autoimmune diseases. However, its clinical efficacy in acute GBS has been rarely documented. In this study, we administered intravenous efgartigimod to four patients with different subtypes of acute GBS, two of whom received efgartigimod monotherapy without concomitant glucocorticoids, IVIG, or PLEX. The treatment outcomes were favorable, suggesting that intravenous efgartigimod may represent a promising therapeutic option for acute GBS. Further research is warranted to validate these preliminary findings.

Enhancing user emotional experience is crucial for improving product appeal and user satisfaction. However, current systems often struggle to integrate explicit and implicit interactions, particularly in capturing subtle emotional signals and dynamically adjusting feedback based on user behavior. To address these challenges, XFormerLite, a lightweight transformer-based model, is proposed to optimize emotional experiences via explicit–implicit signal integration, temporal modeling, and adaptive feedback. By combining explicit signals (e.g., voice and gestures) with implicit signals (e.g., facial expressions and posture), XFormerLite improves emotional engagement and personalizes real-time feedback. Experimental results indicate that XFormerLite achieves an emotion recognition accuracy of 86.2% on the Emo-DB dataset and 84.5% on IEMOCAP, along with significant improvements in response speed and emotion onset delay (EOD). This model presents a promising solution for optimizing emotional experiences in human–product interactions.

Traditional rolling bearing intelligent fault diagnosis methods using a single signal input often face challenges such as low diagnostic accuracy and poor robustness to noise in practical applications. This paper proposes a multi-branch convolutional neural network (CNN) model based on multi-scale and multi-modal input data. The model realizes multimodal multiscale inputs by means of variational modal decomposition (VMD) and STFT transforms while combining the original signals. Compared with a single signal input, richer and complementary fault features can be extracted, fully reflecting the multidimensional information of the fault. By designing multiple parallel CNN branches, the model is able to extract key features from the signal at different scales and modalities and perform feature fusion through the attention mechanism, which further improves the diagnostic accuracy and robustness to noise.

An ultraintense, isolated circularly polarized (CP) attosecond x-ray pulse is often required for many application of pump-probe techniques. A quantum electrodynamics (QED) effect dominated coherent synchrotron emission (CSE) regime is proposed for the generation of an ultraintense isolated, nearly CP attosecond x-ray pulse. Due to the QED effects, the density of the target plasma increases due to Breit-Wheeler pair generation and the radiation reaction effects of gamma photons through the nonlinear Compton effect. As a result, the transparent target becomes an opaque target. Accordingly, the relativistic electron sheet (RES) can be accelerated toward a reflected direction only once. After that, the RES was pushed tempestuously forward continuously under the Lorentz force of the driving laser pulse. Therefore, an isolated, nearly CP attosecond x-ray pulse can be obtained in the reflected direction without filtering. The intensity of the nearly CP attosecond x-ray pulse can reach 2.48×10^{24}W/cm^{2} with ellipticity of 0.83, and the corresponding power can reach up to ∼100PW, which opens the door for nonlinear attosecond studies.

To address the issue of dual prescribed performance (convergence speed and accuracy) control for underactuated surface vessels (USVs) under multi-time scales, this paper proposes a new simplified tracking control approach founded on singular perturbation. To resolve the underactuation constraint, a dynamic feedback coordinate transformation is employed. This transformation maps the velocity composite variable and the position variable into the fast and slow subsystems respectively, thus establishing a singular perturbation model suitable for USVs. Then, the singular perturbation system is decoupled in time scale to obtain non-coupled fast and slow subsystems. For the fast subsystem, the corresponding control law is designed using a predefined-time convergence method. To govern tracking error dynamics in the slow subsystem, a performance function enforces both transient and steady-state performance. Further fortifying robustness, a nonlinear disturbance observer provides continuous estimation and real-time compensation for unknown environmental disturbances. The effectiveness of the proposed control strategy was verified through Lyapunov stability analysis.

Abstract One of the challenging scientific computing problems is topology optimization (TO), where the two tasks of searching through the combinatorially complex configurations and solving the constraints of partial differential equations need to be done simultaneously. In this paper, a novel variational quantum algorithm for constrained TO is proposed, which allows for the single-loop parallel search for the optimal configuration that also satisfies the physical constraints. The optimal configurations and the solutions to physical constraints are encoded with two separate quantum registers. A constraint encoding scheme is also proposed to incorporate volume and connectivity constraints in optimization. The gate complexity of the proposed quantum algorithm is analyzed. The algorithm is demonstrated with compliance minimization problems including truss structures and Messerschmitt–Bölkow–Blohm beams.

A Mg-Gd-Y alloy prepared by surface mechanical attrition treatment (SMAT) was annealed at 450 °C combined with peak aging. The deformation and fracture mechanisms were investigated using in situ tensile tests. Through quantitative calculations of the geometrically necessary dislocation (GND) densities, it was found that the fine-grained (FG) layer in the gradient structure carried greater plastic strain than the coarse-grained (CG) layer during tension. The calculation results of the geometric compatibility parameter (m') and microstructure characterization during in situ tests showed that crack initiation and propagation were prone to occur between adjacent coarse grains. However, the hetero-deformation-induced (HDI) strengthening and strain hardening induced by the strain gradient between the FG and CG layers effectively improved the strength-ductility synergy of the gradient-structured (GS) alloy. In addition, the synergistic effect of intrinsic and extrinsic toughening mechanisms in the GS alloy played a significant role in delaying premature failure.

BACKGROUND: Adolescence represents a sensitive developmental window for brain maturation, particularly for executive functions, which are crucial for goal-directed behavior and long-term mental health. While physical activity is associated with brain health, the potentially differential impacts of specific ball games on executive cognitive processes remain insufficiently examined. OBJECTIVE: This study aims to explore the sport-specific associations of eight commonly investigated ball games, selected from a broad literature base, on adolescent executive function, through a comparative analysis across distinct cognitive domains. METHODS: A systematic review and network meta-analysis were conducted following searches in PubMed, Embase, Cochrane Library, and Web of Science for studies published between January 2000 and March 2025. Eligible cross-sectional studies assessed executive function using one or more of five validated cognitive tasks: Go/No-Go Task, Stop-Signal Task, Stroop Task, Trail Making Test, and N-back Task. The analysis compared adolescents participating in baseball, handball, tennis, badminton, basketball, volleyball, soccer, or table tennis with non-athlete controls. Reaction time was selected as the primary outcome. Effectiveness rankings were established using surface under the cumulative ranking curve (SUCRA), and network consistency and potential publication bias were evaluated using standard diagnostic plots. RESULTS: Thirty-two studies were included. Across all five tasks, adolescents involved in ball sports demonstrated significantly better executive function performance than non-athletes. Relatively higher SUCRA rankings were observed for: table tennis and baseball in the Go/No-Go Task (80.0%, 79.6%), soccer in the Stop-Signal Task (89.3%), volleyball in the Stroop Task (99.1%), table tennis in the Trail Making Test (95.4%), and basketball in the N-back Task (76.2%). However, not all sports were represented across every task, and these rankings should be interpreted with caution. No substantial publication bias was identified. CONCLUSION: Participation in ball games was associated with sport-specific differences in executive function performance among adolescents. These findings highlight potential links between structured physical activity and neurocognitive outcomes, supporting its consideration in youth health promotion strategies aimed at fostering brain development, mental resilience, and lifelong cognitive health. SYSTEMATIC REVIEW REGISTRATION: PROSPERO CRD420251036950.

To ensure the navigational safety of LNG ships entering and leaving ports, a modeling method for the dynamic safety domain of LNG ships under the interference of environmental factors is proposed. This method considers the impacts of wind and current on the navigation of LNG ships, utilizing car-following theory, wind-induced drift laws, and current-induced drift laws to construct models for safety distances in different orientations. In the scenario of head-on encounters during two-way navigation, the repulsive force principle in the artificial potential field method is applied to establish a dynamic safety domain model for LNG ships. For a specific LNG ship, a series of simulations are conducted under varying environmental conditions, and the results are compared with the theoretical calculations derived from the "Design Code of General Layout for Sea Ports" and the experiments using the NT-PRO 5000 simulator. The comparison validates the reasonableness and applicability of the model. The results demonstrate that the model accurately reflect the differential changes in the ship's safety domain under the influence of wind and current, providing a reference for optimizing the utilization of spatiotemporal resources in two-way navigation scenarios for LNG ships.