The Hubble constant (H0) has always had a significant difference between the measurement results of the early universe and the late universe. This difference is called the Hubble tension, which directly challenges the standard ΛCDM cosmological model. Although the local H0 value can be accurately determined by using the distance ladder method of Cepheid variable stars and Type Ia supernovae, the observation results of cosmic microwave background radiation are often low. Therefore, it is crucial to develop independent detection methods to identify potential system errors or find new physical mechanisms.The cosmic timer (CC) method provides a novel alternative. By analyzing the age differences of massive and passively evolved galaxies, it directly determines the Hubble parameter H(z) without using a model-dependent method. This article reviews the observational and theoretical basis of CC technology, elaborates on the strict criteria for selecting galaxies as reliable cosmic timers, and especially emphasizes the necessity of ancient massive galaxy groups that are not affected by the formation of recent stars. We further discussed the spectral age determination tools, including the physical origin of age-sensitive absorption characteristics and D4000 break, and their application in deriving differential age. With the upcoming data released by DESI, Euclid and JWST, it is expected that the cosmological timer will reach a percentage level of accuracy, thus providing cross-verification of the history of cosmic expansion and the nature of Hubble tension.

Show more

Legal large language models are increasingly entering human AI collaborative judicial scenarios, including legal consultation, similar-case retrieval, judgment assistance, and document generation. While these models provide legal information, their outputs may also influence human judgment through emotional expression, certainty markers, authoritative citation, and responsibility attribution. Focusing on emotional persuasion risks in collaborative justice, this study constructs a legal emotional persuasion corpus, integrates case facts, statutory references, emotion intensity, persuasion strategies, and human judgment changes, designs a multi-feature fusion identification model, and conducts a simulated judicial judgment experiment. The experiment examines how model-generated answers affect responsibility attribution, evidence strength, judgment tendency, and trust level. Preliminary results show that the multi-feature fusion model outperforms BERT, Legal-BERT, and RoBERTa in Accuracy, Macro-F1, and AUC. Composite persuasion answers produce stronger decision shifts in judgment tendency and trust level. These findings indicate that legal LLM risk assessment should consider emotional influence mechanisms in addition to factual accuracy. The proposed framework provides an explainable technical path for judicial AI auditing, risk warning, and human oversight.

Show more

Climate-related information is increasingly integrated with financial, macroeconomic, and market-sentiment data, creating a heterogeneous scientific data management problem in which noisy environmental series, high-frequency market observations, and policy indicators must be aligned, transformed, and analyzed reproducibly. This paper develops and evaluates a reproducible threshold vector autoregression (TVAR) workflow to detect state-dependent climate-risk effects on U.S. equity market volatility. Using monthly data from February 1990 to February 2026, the workflow integrates climate variables, realized volatility constructed from daily S&P 500 returns, VIX-based market-stress regimes, macro-financial controls, stationarity transformations, threshold validation, impulse-response analysis, forecast-error variance decomposition, and robustness checks. The results show that climate risk has no statistically meaningful effect on volatility in low-VIX regimes. At the same time, its first two lags become positive and larger in magnitude in high-VIX regimes. The baseline high-VIX coefficient at the second lag is 0.003152 (p = 0.018), and impulse responses peak around the second month after a climate-risk shock. Although the forecast-error variance share remains modest, it rises to about 1.5% in stressed markets and is nearly zero in calm markets. Beyond the substantive climate-finance result, the paper contributes an auditable scientific data-analysis pipeline for heterogeneous time-series integration, regime-aware uncertainty handling, and reproducible nonlinear modeling, aligning climate-finance analytics with the data-intensive research concerns of scalable scientific data management.

Show more

With the development of the low-altitude economy, unmanned aerial vehicle (UAV) systems have gradually become one of the important tools for environmental monitoring and field observation. Compared with fixed-wing aircraft and quadcopters, flapping-wing robots have significant advantages in maneuverability, environmental adaptability, and low-interference flight. However, most existing flapping-wing platforms face three major challenges: difficulty in scaling up, weak amphibious flight capability, and poor hovering stability. This project proposes a novel hummingbird-like flapping-wing robot that employs reinforcement learning control to achieve stable cross-medium flight and reliable hovering. The design utilizes biomimetic figure-eight wing motion and a simple three-degree-of-freedom mechanical structure to improve aerodynamic efficiency. The robot uses a crank-rocker transmission system, brushless motors, gearboxes, and micro servo motors to achieve wing flapping and attitude adjustment. Its fuselage is made of lightweight carbon fiber tubing, while the wings are made of waterproof and airtight fabric. A specially designed support leg structure enables the robot to float stably on water. To achieve control, we constructed a reinforcement learning framework that takes altitude, pitch angle, roll angle, and vertical velocity as input states and outputs flapping frequency, flapping amplitude, and tail angle. The reward function penalizes unstable altitude and attitude while encouraging efficient lift generation. Test results show that the reinforcement learning model converges well during training. The robot can accurately track the target trajectory in three-dimensional position and attitude, and stably complete takeoff, landing, and stable floating on water. This work demonstrates that biomimetic design combined with reinforcement learning can effectively improve the performance of flapping-wing robots and support cross-medium practical applications, providing a reference for future environmental monitoring and field observation.

Show more