Science, 17 Sep 2026, VOL 393, ISSUE 6817

《科学》2026年9月17日，第393卷，6817期

物理学Physics

Disorder-promoted stability

无序促进的稳定性

▲ 作者：Arthur N. Montanari, Pietro Zanin, Adilson E. Motter

▲链接：

https://www.science.org/doi/10.1126/science.aeg3946

▲摘要：此前关于网络动力学的研究认为，节点间的异质性会削弱系统稳定性，这与自然界及工程系统普遍存在固有异质性的现象相悖。

研究组表明，该结论源于为简化数学求解而采取的模型简约化处理。当节点动力学为高维形式，进而产生非厄米雅可比矩阵时，原有结论不再成立。在包括神经网络、电力网络与材料网络的这类系统中，即便参数呈随机无序分布，节点异质性也能够提升系统稳定性。

非厄米性也是网络异质性稳定效应的基础，正如生态网络所示，这种稳定效应甚至可通过非互易相互作用在一维节点动力学中产生。该研究的理论框架揭示，无序并非系统的缺陷，而是稳定复杂系统的一种普适性资源。

▲ Abstract：Previous studies of network dynamics have suggested that heterogeneity among nodes inhibits stability, which is at odds with the ubiquity of inherently heterogeneous natural and engineered systems. Here, we show that this conclusion arises from model reductions introduced for mathematical tractability and breaks down when nodal dynamics are higher-dimensional, yielding non-Hermitian Jacobians. In such systems, including neural, power-grid, and material networks, nodal heterogeneity can instead enhance stability, even when parameters are randomly disordered. Non-Hermiticity also underlies the stabilizing effects of network heterogeneity, which can arise even in one-dimensional nodal dynamics through nonreciprocal interactions, as shown for ecological networks. Our framework reveals disorder not as a liability but as a general resource for stabilizing complex systems.

Extreme loss suppression in an ultracold molecular gas with widely tunable dipolar interactions

在偶极相互作用大范围可调的超冷分子气体中实现极端损耗抑制

▲ 作者：Weijun Yuan, Siwei Zhang, Niccolò Bigagli, Haneul Kwak, Claire Warner, Tijs Karman, et al.

▲链接：

https://www.science.org/doi/10.1126/science.adz0521

▲摘要：超冷偶极分子有望用于制备奇异的物质量子态，但要获得具有强偶极—偶极相互作用、长寿命的分子体样品仍颇具挑战。

研究组实验制备了碰撞稳定的超冷基态分子气体，其寿命为数秒钟。采用双微波缀饰技术，实现了对二体损耗和三体损耗的极大抑制，抑制倍数分别超过10000倍和1000倍。

研究组发现，在广泛的偶极—偶极相互作用范围内，损耗仍受到抑制，据此可将偶极长度在21000至12000玻尔半径（a0）之间连续调谐。结合近期偶极分子玻色—爱因斯坦凝聚的实验成果，该研究结果为探索强偶极量子液体开辟了道路。

▲ Abstract：Ultracold dipolar molecules hold great promise for the creation of exotic quantum states of matter, but the realization of long-lived molecular bulk samples with strong dipole-dipole interactions has remained challenging. In this work, we realized a collisionally stable gas of ultracold ground state molecules with a lifetime of several seconds. Utilizing double-microwave dressing, we achieved an extreme suppression of inelastic two- and three-body losses by factors of more than 10,000 and 1000, respectively. We found that losses remained suppressed across a wide range of dipole-dipole interactions, allowing the continuous tuning of the dipolar length from 21,000 to 12,000 a0, where a0 is the Bohr radius. Combined with the recent realization of Bose-Einstein condensation of dipolar molecules, our findings open the door to the exploration of strongly dipolar quantum liquids. 

Quantum jumps of sound

声音的量子跃迁

▲ 作者：Takuma Makihara, Erik Szakiel, Matthew P. Maksymowych, Oliver A. Hitchcock, Kaveh Pezeshki, Rachel G. Gruenke-Freudenstein, et al.

▲链接：

https://www.science.org/doi/10.1126/science.aeh7535

▲ 摘要：量子力学预言，振动物体的能量以离散包形式存在，但对其位置的任何测量都无法观测到这种离散性。分辨单个能级需要采用定性不同的测量方案：即耦合谐振子的能量，而非其位移。研究组采用与纳米机械谐振器色散耦合的超导量子比特，对声子数开展重复量子非破坏性测量。借助对齐转移打印工艺集成量子比特与谐振器，该研究实现了T₁=2.1 ms的机械寿命，以及每声子2χ/2π=328 kHz的色散频移。

研究组以85%的保真度预测了单声子态，并观测到谐振子在第一激发态与基态之间的量子跃迁。这类不连续跃迁，正是量子力学在宏观振动物体上的显著体现。

▲ Abstract：Quantum mechanics predicts that a vibrating object’s energy comes in discrete packets, yet no measurement of its position reveals this discreteness. Resolving individual energy levels requires a qualitatively different measurement, one coupling to the resonator’s energy rather than its displacement. We use a superconducting qubit dispersively coupled to a nanomechanical resonator to perform repeated quantum nondemolition measurements of the phonon number. An aligned transfer-print technique integrating the qubit and resonator yields a mechanical lifetime of T₁=2.1 milliseconds and a dispersive shift of 2χ/2π=328 kilohertz per phonon. We heralded single-phonon states with 85% fidelity and observed quantum jumps between the resonator’s first excited state and ground state. These discontinuous transitions are a striking manifestation of quantum mechanics in a massive, vibrating object.

计算科学Computational Science

Fast, continuous, and coherent atom reloading in a neutral-atom qubit array

中性原子量子比特阵列中的快速、连续且相干的原子重新装载

▲ 作者：Yiyi Li, Yicheng Bao, Michael Peper, Chenyuan Li and Jeff D. Thompson 

▲链接：

https://www.science.org/doi/10.1126/science.adz9952

▲摘要：中性原子量子处理器是可扩展量子计算颇具前景的平台。深度量子线路实现的一大障碍是原子损耗问题。现有方案替换损耗原子所需的时间，至少比栅极操作与测量操作高出一个数量级。

研究组基于亚稳态171Yb量子比特，实现了快速、连续的原子重新装载。计算区附近设置持续加载的原子储备库，借助光镊可按需提取原子，最高速率可达每秒500次。

当包含单原子制备、非破坏性成像与态初始化时，新型随机填充量子比特阵列的初始化速率可达每秒30次。处于亚稳态的现有量子比特不会受到原子补给过程的干扰。该工作为实现深度无上限、快速且完全容错的量子线路奠定了完整基础。

▲ Abstract：Neutral-atom quantum processors are a promising platform for scalable quantum computing. An obstacle to implementing deep quantum circuits is managing atom loss. Current approaches require at least an order of magnitude longer time than gate and measurement operations to replace lost atoms. In this work, we demonstrate fast, continuous atom reloading that leverages the metastable 171Yb qubit. A continuously loaded reservoir near the computation zone enables on-demand atom extraction with tweezers up to 500 times per second. New, stochastically filled qubit arrays can be initialized 30 times per second when including single-atom preparation, nondestructive imaging, and initialization. Existing qubits in the metastable state are undisturbed by the reloading process. This work establishes a complete foundation for the implementation of fast, fully fault-tolerant quantum circuits with unlimited depth.

地球科学Earth Science

Atmospheric methane lifetime during the Last Glacial Maximum was reduced owing to dust-mediated chlorine chemistry

末次冰盛期的大气甲烷寿命因粉尘介导的氯化学而缩短

▲ 作者：Daphne Meidan, Carlos A. Cuevas, Julián Villamayor, Rafael P. Fernandez, Nicolás J. Cosentino, Samuel Albani, et al.

▲链接：

https://www.science.org/doi/10.1126/science.aec4071

▲ 摘要：大气甲烷（CH₄）在地球气候系统中起着核心作用，但在末次冰盛期（LGM）等高粉尘的冰期环境下，甲烷浓度下降的驱动因素仍不明确。此前解释均假定大气甲烷寿命与现代相当，认为甲烷变化由排放源驱动。近期研究表明，矿物粉尘与海盐气溶胶之间的相互作用会生成可消耗甲烷的氯自由基。

研究组表明，末次冰盛期大气甲烷寿命缩短至7.8年，较现代缩短20%。氯贡献了全球约15%的甲烷减少，为当前的4倍。该结果再现了冰芯甲烷同位素证据，证明无需大幅改变甲烷排放源，仅依靠强于以往认知的大气汇就可解释甲烷的波动变化，凸显了氯化学在冰期甲烷收支中这一长期被忽视的作用。

▲ Abstract：Atmospheric methane (CH₄) plays a central role in Earth’s climate, yet the drivers of its decline during the high-dust conditions of glacial periods, such as the Last Glacial Maximum (LGM), remain uncertain. Previous explanations imply source-driven changes, assuming an atmospheric lifetime comparable to that of present day. Recent work shows that interactions between mineral dust and sea salt aerosols produce CH₄-removing chlorine radicals. In this work, we show that during the LGM, CH₄ lifetime shortened to 7.8 years, 20% lower than that of present day. Chlorine contributed ~15% of global CH₄ loss, fourfold that of present day. Our results reproduce ice core CH4 isotopic evidence, demonstrating that stronger-than-assumed atmospheric sinks can explain CH₄ variability without invoking substantial source changes, highlighting the overlooked role of chlorine chemistry in the glacial CH₄ budget.

影像学Medical Imaging

An expert-level generalist AI for abdominal CT diagnosis

面向腹部CT诊断的专家级通用AI

▲ 作者：Qi Zhang, Jianpeng Zhang, Weiwei Cao, Zilin Lu, Wanxing Chang, Haonan Ding, et al.

▲链接：

https://www.science.org/doi/10.1126/science.aec6129

▲ 摘要：放射学领域的人工智能（AI）旨在各类临床任务中实现专家级诊断，但现有监督学习方法适用范围仍存在局限。

研究组开发了RADAR这一通用视觉语言模型，基于超40万例腹部增强计算机断层扫描（CT）影像、1500万份解剖学图像文本对完成训练，无需人工标注即可直接从临床报告学习。经多中心、多样化临床场景的内部与外部评估，RADAR针对18类解剖结构、146种影像结果均取得优异诊断性能，泛化能力稳定。

阅片对照研究显示，借助RADAR辅助，26名放射科医师的诊断灵敏度提升约10%。RADAR为腹部CT提供了一种可规模化、多功能且可解释的解决方案，证明通用AI在常规及复杂放射学检查中能够与人类专家相媲美。

▲ Abstract：Artificial intelligence (AI) in radiology aspires to deliver expert-level diagnosis across diverse clinical tasks, yet existing supervised strategies remain limited in scope. We developed RADAR, a generalist vision-language model trained on more than 400,000 contrast-enhanced abdominal computed tomography (CT) examinations and 15 million anatomy-wise image-text pairs, learning directly from clinical reports without manual annotation. Throughout internal and external evaluations across multiple centers and varied clinical scenarios, RADAR achieved high diagnostic performance and robust generalization for 18 anatomical structures and 146 imaging findings. In a reader study, RADAR assistance increased the diagnostic sensitivity of 26 radiologists by ~10%. RADAR offers a scalable, versatile, and interpretable solution for abdominal CT, demonstrating that generalist AI can match human experts in general and complicated radiology tasks.