Science, 22 October 2026, Volume 658, Issue 8137

《自然》2026年10月22日，第658卷，8137期

材料科学Materials Science

Wave coarsening drives time crystallization in mechatronic metamaterials

波粗化驱动机电超材料中的时间晶化

▲ 作者：Jonas Veenstra, Jack Binysh, Vito Seinen, Roger Solé Borràs, Rutger Naber, Damien Robledo-Poisson, et al.

▲链接：

https://www.nature.com/articles/s41586-026-11163-6

▲摘要：当金属被磁化、乳液相分离或是星系成团时，会形成畴壁与斑图，并随时间发生不可逆粗化。这类粗化过程普遍由扩散弛豫驱动，向平衡态演化。

研究组在活性弹性介质中发现了对应的振动类现象，即波粗化：振动模式自发产生，其波长、周期与振幅逐步增大，最终形成一种被称为时间晶体的全局同步态。研究组在一维与二维机电超材料中观测到波粗化现象，并捕捉到其动力学标度。

通过破坏动量守恒，研究组进一步中止该粗化过程，还发现了一种远离平衡态、与手性拓扑边缘模对应的非线性类比物。该研究结果引出了有关非势相互作用体系瞬态物理的新问题，同时为声学、光机、活性物质以及软体机器人领域的非线性波提供了一种组织机制。

▲ Abstract：When metals are magnetized, emulsions phase separate, or galaxies cluster, domain walls and patterns form and irremediably coarsen over time. This coarsening is universally driven by diffusive relaxation towards equilibrium. Here we discover a vibrational counterpart—wave coarsening—in active elastic media: vibrations emerge and spontaneously grow in wavelength, period and amplitude, before a globally synchronized state called a time crystal forms. We observe wave coarsening in one- and two-dimensional mechatronic metamaterials and capture its dynamical scaling. We further arrest the process by breaking momentum conservation and reveal a far-from-equilibrium nonlinear analogue to chiral topological edge modes. Our results open new questions about the transient physics of systems with non-potential interactions and suggest an organizing principle for nonlinear waves in acoustics, optomechanics, living matter and soft robotics.

Competing mechanisms toughen bcc binary and medium-entropy alloys at 20 K

竞争机制实现体心立方二元中熵合金在20 K下增韧

▲ 作者：David H. Cook, Xiaoyu Chen, Michael O’Leary, Satish Rao, Wenqing Wang, Calvin H. Belcher, et al.

▲链接：

https://www.nature.com/articles/s41586-026-11065-7

▲ 摘要：体心立方（bcc）金属与合金的塑性变形通常由螺型位错主导。其温度敏感运动归因于材料强度随温度降低而急剧升高，从而压缩裂纹尖端塑性区，进而引发韧脆转变。

研究组测试了电子结构相近的IV-V族bcc难熔合金在20K下的力学性能，发现这类材料的拉伸屈服强度与断裂韧性随温度变化呈现平台状特征，不存在明显的韧脆转变。

该行为源于孪生与ω相变的全域激活：这两种机制会先在界面处形核产生位错，之后才发生位错的整体运动。这种初始机制对温度不敏感，并控制屈服过程，因此裂纹尖端塑性区不会随温度降低而收缩。

▲ Abstract：Plastic deformation in body-centred cubic (bcc) metals and alloys is conventionally mediated by screw dislocations. Their temperature-sensitive motion is attributed to a steep rise in strength with decreasing temperature that constricts plastic zones at crack tips, leading to a ductile-to-brittle transition. Here we evaluated the mechanical properties of group IV–V bcc refractory alloys with similar electronic structures at 20?K and found that these materials instead exhibit a plateau in tensile yield strength and fracture toughness as a function of temperature, with no apparent ductile-to-brittle transition. This behaviour arises from the global activation of twinning and an ω-phase transformation, which nucleate dislocations at interfaces before global dislocation motion. The initial mechanism is temperature insensitive and controls yielding, thereby preventing a crack-tip plastic zone from constricting with decreasing temperature.

信息科学Information Science

Retrofitting language models to operate over bytes

改造语言模型实现字节级运算

▲ 作者：Benjamin Minixhofer, Tyler Murray, Tomasz Limisiewicz, Anna Korhonen, Luke Zettlemoyer, Noah A. Smith, et al.

▲链接：

https://www.nature.com/articles/s41586-026-11111-4

▲ 摘要：人工智能（AI）的最新进展很大程度上由大语言模型（以词元为离散处理单元的深度神经网络）来推动。为了表示文本，大多数大语言模型使用单词或单词片段作为词元，这被称为子词分词。

但子词分词会掩盖细粒度信息，这会带来一系列问题，尤其是对于计算机代码或生物序列等依赖单字符或字节承载语义的科学数据很不友好。直接处理文本字节编码的模型避免了这些限制，但性能长期不及基于子词的模型。

研究组提出字节化方法搭建字节级大语言模型，其性能接近基于子词的模型。借助两阶段转换策略，仅需少量额外训练即可把现有基于子词的模型改造成字节级模型。

由此生成的模型优于早期字节级方案，尤其擅长字符级推理任务；通过高效处理字节级信息达到实用推理速度，并通过复用源大语言模型周围的现有生态获得适应性。

该研究工作消除了端到端字节级语言建模的长期性能短板，说明原始文本编码模型在规模化上具备竞争力，在需要细粒度文本理解的场景中更具优势。

▲ Abstract：Recent advances in artificial intelligence (AI) have largely been driven by large language models, deep neural networks that operate over discrete units called tokens. To represent text, most large language models use words or word fragments as the tokens, known as subword tokenization. Subword tokenization obscures fine-grained information, which is problematic, especially for scientific data—such as computer code or biological sequences—where meaning depends on the individual characters or bytes. Models that instead operate directly on the byte encoding of text avoid these limitations, but until now they have lagged behind subword-based models in performance. Here we introduce a general method for creating byte-level large language models through byteification that approach the capabilities of subword-based systems. We use a two-stage conversion procedure to retrofit existing subword-based models into byte-level models with minimal extra training. The resulting models outperform earlier byte-level approaches and excel on character-level reasoning tasks, achieving practical inference speeds by efficiently processing byte-level information and adaptability by reusing the existing ecosystem around the source large language model. Our results remove a long-standing performance barrier to end-to-end byte-level language modelling, demonstrating that models operating on raw text encodings can scale competitively while offering advantages in domains requiring fine-grained textual understanding.

化学Chemistry

Hydrogen bonding in water under extreme confinement

极端受限条件下水中的氢键合

▲ 作者：Xintong Xu, Matthias Kuehne, Harrison A. Walker, De-Liang Bao, Xin Jin, Yu-Ming Tu, et al.

▲链接：

https://www.nature.com/articles/s41586-026-10858-0

▲摘要：极端受限或界面附近流体的分子结构、分子间键合均区别于其体相流体。光学振动谱表明，界面水、受限水的分子内O-H伸缩频率会发生偏移，是分子间氢键变化的灵敏光谱信号。但在分子间力和表面力的尺度上实验表征受限水仍颇具困难。

研究组利用纳米级分辨率的原位振动电子能量损失谱（vEELS），直接在分子尺度上观测单根碳纳米管内受限水的氢键。大管径碳纳米管内的水表现出体相水的键合O-H振动；管径缩小后频率蓝移，趋近水蒸气和疏水界面水的自由O-H伸缩，对应高度分散的非氢键状态。

量子振子理论分析将观察到的光谱特征与局部氢键构型联系起来，与实验观察结果一致。低温实验进一步呈现了受限水复杂的结构相变。该工作揭示了受限条件下氢键的量子与动力学属性，为解析受限流体分子结构和键合提供新支撑。

▲ Abstract：Fluids under extreme confinement or near interfaces exhibit molecular structures and intermolecular bonding distinct from their bulk analogues. Optical vibrational studies have shown that interfacial and confined water can exhibit altered intramolecular O–H stretching frequencies, a sensitive spectral signature of changes in intermolecular hydrogen bonding. Investigating confined water experimentally at the length scale of intermolecular and surface forces has, however, remained a challenge. Here we report direct molecular-level observations of hydrogen bonding in water confined inside individual carbon nanotubes (CNTs), enabled by in situ vibrational electron energy loss spectroscopy (vEELS) with nanoscale resolution. Water in larger CNTs exhibits the bonded O–H vibrations of bulk water, but at smaller diameters, and the frequency blueshifts to near the free O–H stretch found in water vapour and water located near hydrophobic surfaces, indicating a highly dispersed, non-H-bonded environment. Theoretical analysis based on quantum vibrational oscillators links the observed spectral features to local hydrogen-bonding configurations, consistent with the experimental observation. Furthermore, cryogenic experiments provide insights into complex structural phase transitions of confined water. This research reveals the quantum and dynamic nature of hydrogen bonds under confinement and the potential impact of unveiling molecular-level structure and bonding in confined fluids.

地球科学Earth Science

Interhemispheric monsoon dynamics over the past 450,000 years

过去45万年间的半球间季风动力学

▲ 作者：E. J. Olson, D. P. Gillikin, D. Parmenter, S. Burns, A. Verheyden, O. Miyawaki, et al.

▲链接：

https://www.nature.com/articles/s41586-026-11166-3

▲摘要：全球季风系统为地球上包含人类在内的多数生物群提供淡水。米兰科维奇理论指出地质时期太阳辐射发生改变，进而驱动全球季风变化。冰盖、海表温度等气候内部强迫因子也会影响该系统。但南半球跨越冰期-间冰期旋回的季风记录匮乏，制约了学界对全球季风驱动机制的认识。

研究组表明，45万年以来南北半球季风均受米兰科维奇强迫和内部边界条件的调制。依托亚马逊盆地洞穴堆积物的氧同位素数据，研究组重建了南美夏季风（SASM）序列：南美夏季风与北半球主要季风（即亚洲季风）的变化明显异相。

研究组确定了一个同步性阈值，共有十次事件（含冰消期）超出该阈值。对比这些时段的古气候强迫因素可知，温室气体浓度、跨赤道热力梯度以及大西洋、太平洋海表温度可能是这些事件的主控因素。

该记录表明，全球季风系统能够发生重组，在行星边界条件快速变化阶段尤为突出。

▲ Abstract：The global monsoon system provides freshwater to much of Earth’s biota, including its human population. The Milankovitch theory predicts changes in Earth’s solar insolation over geologic time, and these variations are expected to result in changes to the global monsoon. Internal climate forcings, such as ice cover and sea surface temperatures, may also affect the system. However, few records of monsoon variability in the Southern Hemisphere over glacial–interglacial cycles exist to test our understanding of the factors that drive variations in the global monsoon. Here we show that monsoons in both the Southern and Northern Hemispheres vary with Milankovitch forcing, and with internal boundary conditions over the past 450,000 years. We present a record of South American Summer Monsoon (SASM) variability, based on speleothem oxygen-isotope data from the Amazon Basin. South American Summer Monsoon variability and that of a major Northern Hemisphere monsoon—that is, the Asian monsoon—have been strikingly out of phase. However, we observe occasional synchronous behaviour in regions impacted by both monsoons. We identify a synchronicity threshold that is exceeded during ten events, including during glacial terminations. Comparison with palaeoclimatic forcing factors during these intervals indicates that greenhouse gas concentrations, cross-equatorial thermal gradients and sea surface temperatures in both the Atlantic and Pacific probably influence these events. This record demonstrates that the global monsoon can reorganize itself, especially during intervals of rapidly changing planetary boundary conditions.

Advancing subseasonal forecasting with machine learning

机器学习助力次季节天气预报

▲ 作者：Hannah Guan, Soukayna Mouatadid, Paulo Orenstein, Judah Cohen, Haiyu Dong, Zekun Ni, et al.

▲链接：

https://www.nature.com/articles/s41586-026-11126-x

▲ 摘要：决策者依靠天气预报开展作物种植、野火管控、水与能源调配以及极端天气防范工作。得益于基于物理的动力学模型与数据驱动人工智能（AI）模型的稳步发展，当前这类预报在两周时效内已达到前所未有的精度。但受误差累积、模型系统性偏差以及大气混沌特性的影响，模型的预报能力在次季节尺度（未来2-6周）会急剧下降。

为解决预报性能下降，研究组提出概率偏差校正，即一种通过学习校正历史概率预报，从而大幅降低系统性误差的机器学习框架。将其应用于欧洲中期天气预报中心（ECMWF）的主流动力学与AI模型后，概率偏差校正使AI预报系统的适度次季节预报技能提升一倍。

对于操作去偏动力学模型，其气压、气温、降水预报目标的提升占比分别达到91%、92%、98%。

研究组为操作部署设计了概率偏差校正，在ECMWF 2025年实时预报竞赛中，基于该方法的全球预报在所有气象要素与预报时效上均排名第一，性能优于6家业务预报中心的动力学模型、一套国际动力学多模型集合、ECMWF AI预报系统以及全球34支队伍的预报系统。

这类概率预报技能提升有助于更精准地预测极端事件，有望助力脆弱地区优化农业规划、能源调度与防灾准备。

▲ Abstract：Decision-makers rely on weather forecasts to plant crops, manage wildfires, allocate water and energy and prepare for weather extremes. Today, such forecasts enjoy unprecedented accuracy as far as 2?weeks thanks to steady advances in physics-based dynamical models and data-driven artificial intelligence models. However, model skill drops precipitously at subseasonal timescales (2–6 weeks ahead) due to compounding errors, systemic model biases and the chaotic nature of the atmosphere. To counter this degradation, we introduce probabilistic bias correction, a machine learning framework that substantially reduces systematic error by learning to correct historical probabilistic forecasts. When applied to the leading dynamical and artificial intelligence models from the European Centre for Medium-Range Weather Forecasts (ECMWF), probabilistic bias correction doubles the modest subseasonal skill of the AI Forecasting System and improves the skill of the operationally debiased dynamical model for 91% of pressure, 92% of temperature and 98% of precipitation targets. We designed probabilistic bias correction for operational deployment, and, in the 2025 real-time forecasting competition of ECMWF, its global forecasts were the best for all weather variables and lead times, outperforming the dynamical models from six operational forecasting centres, an international dynamical multimodel ensemble, the AI Forecasting System of ECMWF and the forecasting systems of 34 teams worldwide. These probabilistic skill gains translate into more accurate prediction of extreme events and have the potential to improve agricultural planning, energy management and disaster preparedness in vulnerable communities.