The Franck-Hertz experiment, a landmark experiment in the history of modern physics, plays a significant role in undergraduate laboratory courses for elucidating atomic structure models. This paper presents an improved design for the Franck-Hertz experiment, encompassing six aspects: integration of historical context, automated data acquisition, design of experiments targeting high-excitation states, fostering of team collaboration, training in scientific software usage, and progressive formulation of thought-provoking questions. With these enhancements, students are relieved from the burden of tedious manual data recording, thereby freeing up time to substantially expand and deepen the experimental content. In addition to maintaining a balance between theoretical instruction and hands-on training, the revised approach further cultivates scientific inquiry skills and reinforces basic competencies in scientific literacy. This reform aligns with the general-education philosophy underlying experimental physics teaching at Tsinghua University.

While AI shows great potential in empowering physics experiments, the highly innovative and rapidly iterating nature of these experiments poses challenges for traditional Large Language Models (LLMs). Reasoning on specific methods typically requires full-context injection of local knowledge bases, resulting in excessive context consumption and high inference costs. This study proposes an intelligent agent framework for physics experiments integrating dynamic context-aware management and collaborative reasoning. Adaptable to existing LLMs, the framework dynamically extracts and injects high-value information into the context, significantly reducing inference costs and improving efficiency without compromising reasoning quality compared to full-context methods. By supporting rapid secondary development and diverse scenario adaptation, the framework effectively overcomes computational and cost bottlenecks, offering an efficient and flexible paradigm for the automation and intelligent transformation of physics experiments.

This study systematically investigates the effects of water level, cup height, and bottom diameter on the fundamental frequency of a glass water chime (liquid bell) system. By establishing a standardized experimental platform and applying FFT spectrum analysis, we obtained primary frequency data for three types of glass cups under various water levels. The results show that the resonance frequency increases monotonically and then tends to level off as the water level decreases. The bottom diameter of the cup is the most sensitive parameter affecting the frequency, while the cup height primarily determines the response range. Theoretical modeling agrees well with the experimental results, revealing the sound response mechanism under multivariable coupling in such liquid bells. This work provides an experimental and theoretical foundation for the acoustic optimization and engineering applications of water cup chimes and related liquid containers.

To address the insufficient naturalness of timbre in audio synthesis, this study proposes a multi-stage framework integrating feature analysis and intelligent optimization. First, a dataset is constructed by synthesizing three types of sound waves, from which acoustic features are extracted. Dimensionality reduction and clustering are employed to quantify the differences between synthetic and natural sounds. Second, Bayesian optimization is applied to adaptively assign feature weights, identifying key discriminative indicators. Finally, reinforcement learning dynamically adjusts synthesis parameters (e.g., frequency, decay factor) using clustering center distance as a reward to drive synthetic timbre closer to natural sound distributions. Experimental results demonstrate that this method significantly enhances the naturalness of synthesized timbre, providing an efficient data-driven solution for audio optimization.

In competitive sports and humanoid robotics, individuals often employ rapid arm-retraction strategies to enhance angular velocity and adjust body posture. However, under the constraint of angular momentum conservation, there is a lack of systematic modeling and evaluation frameworks to design such strategies for achieving both accelerated rotation and attitude stability. This paper, grounded in rigid body dynamics, investigates how different retraction strategies affect angular velocity enhancement and posture control. We propose a unified modeling framework that integrates physical mechanism interpretation with strategic performance optimization. By constructing two-dimensional and three-dimensional dynamic models and analyzing stability through principal axis deviation, we demonstrate the influence of strategy parameters on rotational performance via simulations and high-frame-rate video validation. The introduction of the Angular Head Control Index (AHCI) offers a unified criterion for evaluating and optimizing retraction strategies.