国防科技大学

High-order time-stepping schemes are crucial for simulating incompressible fluid flows due to their ability to capture complex turbulent behavior and unsteady motion. In this work, we propose a third-order accurate numerical scheme for the two-dimensional incompressible Navier-Stokes equation. Spatial and temporal discretization is achieved using Fourier pseudo-spectral approximation and the BDF3 stencil, combined with the Adams-Bashforth extrapolation for the nonlinear convection term, resulting in a semi-implicit, fully discrete formulation. This approach requires solving only a single Poisson-like equation per time step while maintaining the desired temporal accuracy. Classical numerical experiments demonstrate the advantage of our scheme in terms of permissible time step sizes. Moreover, we establish uniform-in-time bounds for the vorticity in both L2 and higher-order Hs norms (s >= 1), provided the time step is sufficiently small. These bounds, in turn, facilitate the derivation of optimal convergence rates.

Large Language Models (LLMs) have exhibited remarkable proficiency in textual understanding and generation, leading to the widespread application across various domains. However, their inherent tendency to hallucinate critically hinders the implementation especially in high-stakes scenarios. Existing classification-based detection methods often rely on static internal states from a specific layer, thereby overlooking the dynamic semantic evolution and the hierarchical divergence of hallucination patterns throughout LLM layers. To address these limitations, we propose TrackLGD (Tracking Layer-wise Graph Dynamics), a novel hallucination detection framework grounded in multiplex graph neural networks. Specifically, we model the LLMs’ hierarchical inference process as a multiplex graph, enabling the tracking of intra-layer semantics and inter-layer information dynamics. To identify the critical knowledge layers and enhance interpretability, we introduce an adaptive layer-wise attention awareness mechanism to discern the most influential tokens. Furthermore, a calibration module is integrated to align model confidence with actual accuracy, thereby ensuring the reliability of the detection process. Extensive experiments across multiple mainstream LLMs and diverse benchmarks consistently demonstrate the effectiveness of our framework. Notably, TrackLGD achieves 3.2% improvement in AUROC on TruthfulQA dataset compared to competitive baselines, which underscores the potential of graph-based signal probing in fostering more transparent and reliable LLM ecosystems.

Active decoys equipped with digital radio frequency memory can reproduce radar waveforms and reduce the reliability of recognition based on a single station or a single physical domain. This study develops a distributed radar spatial–polarimetric joint-domain recognition framework based on a confidence-weighted heterogeneous graph neural network. Single-station polarimetric descriptors and observation-quality attributes are assigned to graph nodes. Inter-station geometry, aligned-signal correlation, high-resolution range-profile correlation, relation type, and association confidence are assigned to multi-relational graph edges. Active polarization switching provides effective-rank, multi-pulse consistency, co-polarized, cross-polarized, and phase descriptors. Relation-specific message passing and station-level bidirectional cross-attention form the graph representation used for classification. The evaluation combines a public-data-augmented simulation dataset, a pure simulation dataset, and an enterprise-collaborative semi-physical radar dataset. The revised protocol also tests association errors, variable receiver layouts, advanced dual-channel and polarization-controllable relay conditions, feature redundancy, hyperparameter stability, and execution on an NVIDIA A100 GPU. Under the default PADS setting (SNR = 15 dB, medium diversity), the model attained 92.1% accuracy. Under cross-batch EC-Radar evaluation, it achieved a Macro-F1 of 0.847. At a combined association-error rate of 0.25, Macro-F1 decreased by 6.9 percentage points relative to the clean graph. On the NVIDIA A100 platform, the processing latency excluding acquisition was 23.8 ms per target graph, and the algorithmic latency including the polarization acquisition window was 43.8 ms.

Poverty mapping is increasingly important for monitoring Sustainable Development Goal 1 (SDG 1) of the United Nations 2030 Agenda, which aims to end poverty in all its forms everywhere. Yet timely and fine-resolution poverty estimation remains difficult because conventional census- and survey-based approaches are costly, infrequent, and often sparse precisely where deprivation is most severe. As poverty emerges from complex socioeconomic systems shaped by human mobility, social interactions, infrastructure, and economic activities, emerging computational methods and nontraditional data sources have created new opportunities for poverty estimation and mapping. At the intersection of statistical physics, complex systems science, and data science, these approaches enable poverty estimation at finer spatial and temporal resolutions. This review summarizes the main concepts of poverty and the principal frameworks used to measure it, and examines recent advances on poverty estimation and mapping using satellite imagery, mobile phone data, social media data, and multisource data fusion. The review also discusses persistent challenges related to representativeness, transferability across regions, interpretability, and uncertainty quantification. Finally, the review clarifies both the analytical promise and the practical limits of contemporary poverty mapping.

Ammonia decomposition is a promising route for COx-free hydrogen production, but the development of low-cost catalysts with high activity and long-term stability at medium temperatures remains challenging. Ni-based catalysts are attractive alternatives to Ru-based systems; however, they often suffer from active-site sintering, insufficient regulation of metal-support interfaces, and limited stability. Herein, a series of one-dimensional Ni-Co/Ce0.6Zr0.4O2 nanofiber catalysts, denoted as Ni-Co/CZ64, were prepared by an in situ electrospinning strategy to regulate the bimetallic composition and oxide interface simultaneously. Structural and surface analyses show that the fibrous CZ64 framework effectively maintains the one-dimensional morphology after H2 reduction, while Ni and Co species remain spatially adjacent and closely coupled with the Ce-Zr oxide matrix. Among the investigated catalysts, 10Ni10Co/CZ64 exhibits the best overall performance, achieving 98.62% NH3 conversion and an H2 formation rate of 33.02 mmol H2 gcat−1 min−1 at 500 °C under a GHSV of 30,000 mL gcat−1h−1. It also shows the lowest apparent activation energy of 36.23 kJ mol−1 and maintains approximately 97% NH3 conversion during a 100 h stability test. XPS, quantitative H2-TPR deconvolution , NH3-TPD, CO2-TPD, and in situ DRIFTS results indicate that the superior activity is associated with spatially adjacent Ni-Co sites, a defect-rich CZ64 interface, a reducible-species distribution dominated by low-temperature components while retaining a non-negligible interfacial/support-related fraction, balanced acid–base properties, and limited accumulation of NHx/H-related intermediates. These results identify the equimolar Ni-Co composition as the optimal formulation within the investigated fibrous catalyst series and highlight the importance of coordinating bimetallic active sites with a defect-rich Ce-Zr oxide interface for stable ammonia decomposition.