We establish a new weighted regularity criterion for weak solutions of the 3D Magnetohydrodynamic (MHD) equations, involving only the gradient of the velocity field. It is proved that a weak solution (u,b) is smooth on R3×(0,T) provided supx0∈R3‖|x−x0|β∇u‖Lα(0,T;Lγ(R3))<+∞with 2α+3γ=2−β, 32−β<γ≤94−3β, 3≤α<∞ and −23≤β≤13. Our result improves and extends some known scaling-invariant regularity criteria for the 3D MHD equations, albeit under stronger restrictions on the admissible ranges of α and γ.
All-solid-state lithium-metal batteries (ASSLBs) have emerged as one of the most promising technologies in energy storage research due to their high safety and high energy density. However, current solid-state electrolytes (SSEs) do not meet the requirement of outstanding electrochemical and mechanical properties for practical high-end ASSLBs applications. The previous studies have primarily focused on discovering advanced electrodes and SSE materials, the research on the electrode/SSE interface is rather scarce and remains underexplored for lithium-metal batteries. Herein, a first-principles calculations and machine learning (ML) combined method is proposed to directly predict interfacial properties of lithium metal anode with Li7 L3 Z2 O12 (LLZO)-, Li2 PO2 N (LiPON)-, Li3 PS4 (LPS)- and Li6 PS5 X (LPSX, X = Cl, Br, and I)-based solid-state electrolytes, where diffusion features are introduced as effective novel descriptors with requiring low computation cost. Via first principles calculations, we demonstrate that formation energy and diffusion barrier of lithium are highly correlated to interfacial impedance of lithium/SSEs, and doping strategy can significantly improve stability and lithium diffusion performance of the interfaces. With demanding reduced training data, a high accuracy ( R -square of 0.99) and reliable (mean absolute error of 0.12) gradient-boosted regression tree classifier is developed for screening SSEs. By accelerated discovery of new SSEs from over 876 datasets, we identify 21 additional promising candidates, some of which exhibit superior performance than the reported SSEs. This work provides a high-accuracy first-principles machine learning (FPML) screening approach along with a promising list of solid-state electrolytes, facilitating the future discovery of new-generation solid-state batteries. (c) 2026 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science & Technology.
Ammonia (NH3) is a promising carbon-free energy carrier, yet its low reactivity leads to prolonged ignition delay times (IDTs), constraining its practical deployment in combustion systems. To address this limitation, the present work employs and validates nitromethane (CH3NO2) as a reactivity enhancer for NH3 combustion. This work develops and validates a compact kinetic mechanism (106 species, 1016 reactions) for NH3/CH3NO2 mixtures using a sensitivity-guided multi-objective particle swarm optimization (MOPSO) strategy, simultaneously calibrating against IDTs and species profiles. The optimized mechanism is rigorously evaluated against a comprehensive set of blind validation datasets, including IDTs, species profiles, and laminar burning velocities (LBVs) of pure NH3, pure CH3NO2, and direct NH3/CH3NO2 blended flames. Quantitative benchmarking against recent state-of-the-art mechanisms confirms improved predictive consistency across wide operating conditions. Detailed kinetic analysis reveals that the ignition-promoting effect of CH3NO2 operates primarily through two interconnected channels: enhancement of OH radical production, and acceleration of NH3 consumption via H-abstraction. This synergistic action is driven by a NOx-assisted radical amplification mechanism, centered on three key reactions, namely R680 (H + NO2 <=> NO + OH), R927 (NH2 + NO <=> NNH + OH), and R928 (NH2 + NO <=> H2O + N-2). The proposed kinetic mechanism and the fundamental chemical insights derived herein establish a reliable and validated foundation for simulating NH3/CH3NO2 combustion under realistic conditions.
Incomplete multi-view clustering aims to discover cluster structures from multi-view data in the presence of missing views, which has attracted increasing attention in recent years. However, many existing methods either entangle common and view-specific information in a single latent space, or rely on view completion modules that are weakly coupled with the clustering objective, making it difficult to learn robust and discriminative representations for clustering. To address these challenges, we propose Incomplete Disentangled Multi-View Representation Learning with Graph Propagation (IDMRL-GP). Specifically, it first encodes each view into a latent space and decomposes the resulting latent representation into a common representation and a view-specific representation through shared projection heads. Subsequently, it combines instance-level contrastive constraints, view prediction, and a subspace-based geometric regularizer to develop a structural disentanglement module, which encourages the common branch to capture cross-view semantics while the specific branch preserves complementary view-related information. Furthermore, it uses a graph structure to represent the fused common representations and performs masked graph propagation to complete view-specific representations for missing views. Finally, it then obtains soft cluster assignments in the joint space of common and completed view-specific representations under a cluster-level consistency constraint. Extensive experiments on six public multi-view datasets with different missing rates demonstrate that IDMRL-GP achieves competitive or superior performance compared with representative incomplete multi-view clustering methods. The source code is publicly available at https://github.com/xag2020/IDMRL-GP.
Opportunistic maintenance (OM) is a widely recognized strategy for improving system reliability and cost efficiency. However, its practical application is often constrained by degradation dependencies among subsystems and limited parallel maintenance capacity. Existing research exhibits notable gaps in integrating degradation coupling information into OM decisions, conducting systematic task screening under resource constraints, and developing multi-dimensional benefit evaluation models. To address these gaps, this paper proposes an integrated OM decision-making framework for degradation-dependent systems under limited parallel maintenance capacity. First, a system-level degradation association network is constructed to provide a structured characterization of degradation coupling effects. Second, a resource-aware task screening mechanism is developed to dynamically determine feasible maintenance task sets under parallel capacity and time constraints. Third, a multi-dimensional benefit evaluation model is established that comprehensively incorporates degradation coupling strength, maintenance effectiveness, and OM window utilization. Finally, a multi-objective optimization model is formulated to balance maintenance cost rate and system availability according to decision-maker preferences. Numerical experiments on a civil aircraft air conditioning system show that the proposed method effectively captures degradation interactions, generates robust OM strategies under resource constraints, and achieves a favorable cost–availability trade-off.
