Search for: All records

Graphs have emerged as one of the most important and powerful data structures to perform content analysis in many fields. In this line of work, node classification is a classic task, which is generally performed using graph neural networks (GNNs). Unfortunately, regular GNNs cannot be well generalized into the real-world application scenario when the labeled nodes are few. To address this challenge, we propose a novel few-shot node classification model that leverages pseudo-labeling with graph active learning. We first provide a theoretical analysis to argue that extra unlabeled data benefit few-shot classification. Inspired by this, our model proceeds by performing multi-level data augmentation with consistency and contrastive regularizations for better semi-supervised pseudo-labeling, and further devising graph active learning to facilitate pseudo-label selection and improve model effectiveness. Extensive experiments on four public citation networks have demonstrated that our model can effectively improve node classification accuracy with considerably few labeled data, which significantly outperforms all state-of-the-art baselines by large margins. 

Using a starlike Be 6 Au 7 − cluster as a building block and following the bottom-up strategy, an intriguing two-dimensional (2D) binary s-block metal Be 2 Au monolayer with a P 6/ mmm space group was theoretically designed. Both the Be 6 Au 7 − cluster and the 2D monolayer are global minima featuring rule-breaking planar hexacoordinate motifs (anti-van't Hoff/Le Bel arrangement), and their high stabilities are attributed to good electron delocalization and electronic-stabilization-induced steric force. Strikingly, the Be 2 Au monolayer is a rare Dirac material with two perfect Dirac node-loops in the band structure and is a phonon-mediated superconductor with a critical temperature of 4.0 K. The critical temperature can be enhanced up to 11.0 K by applying compressive strain at only 1.6%. This study not only identifies a new binary s-block metal 2D material, namely Be 2 Au, which features planar hexacoordination, and a candidate superconducting material for further explorations, but also provides a new strategy to construct 2D materials with novel chemical bonding. 

Solid molecular hydrogen has been predicted to be metallic and high-temperature superconducting at ultrahigh hydrostatic pressures that push current experimental limits. Meanwhile, little is known about the influence of nonhydrostatic conditions on its electronic properties at extreme pressures where anisotropic stresses are inevitably present and may also be intentionally introduced. Here we show by first-principles calculations that solid molecular hydrogen compressed to multimegabar pressures can sustain large anisotropic compressive or shear stresses that, in turn, cause major crystal symmetry reduction and charge redistribution that accelerate bandgap closure and promote superconductivity relative to pure hydrostatic compression. Our findings highlight a hitherto largely unexplored mechanism for creating superconducting dense hydrogen, with implications for exploring similar phenomena in hydrogen-rich compounds and other molecular crystals. 

Lithium-ion batteries (LIBs) are widely used energy storage devices, and sodium-ion batteries (SIBs) are promising alternatives to LIBs because sodium is of high abundance and low toxicity. However, a dominant obstacle for the advancement of LIBs and SIBs is the lack of high capacity anode materials, especially for SIBs. Here, we propose that three characteristics, namely appropriate pore size, suitable pore distribution, and an entirely planar topology, can help achieve ultrahigh capacity 2D anode materials. Under such guidelines, we constructed a B 7 P 2 monolayer, and investigated its potential as a LIB/SIB anode material by means of density functional theory (DFT) computations. Encouragingly, the B 7 P 2 monolayer possesses all the essential properties of a high-capacity LIB/SIB anode: its high stability ensures the experimental feasibility of synthesis, its metallicity does not change upon Li/Na adsorption and desorption, the Li/Na can well diffuse on the surface, and the open-circuit voltage is in a good range. Most importantly, the B 7 P 2 monolayer has a high storage capacity of 3117 mA h g −1 for both LIBs and SIBs, and this capacity value ranks among the highest for 2D SIB anode materials. This study offers us some good clues to design/discover other anode materials with ultrahigh capacities, and serves us another vivid example that (implicit and hidden) trends/rules in the literature can guide us in the design of functional materials more efficiently. 

The anomalous nondipolar and nonaxisymmetric magnetic fields of Uranus and Neptune have long challenged conventional views of planetary dynamos. A thin-shell dynamo conjecture captures the observed phenomena but leaves unexplained the fundamental material basis and underlying mechanism. Here we report extensive quantum-mechanical calculations of polymorphism in the hydrogen–oxygen system at the pressures and temperatures of the deep interiors of these ice giant planets (to >600 GPa and 7,000 K). The results reveal the surprising stability of solid and fluid trihydrogen oxide (H 3 O) at these extreme conditions. Fluid H 3 O is metallic and calculated to be stable near the cores of Uranus and Neptune. As a convecting fluid, the material could give rise to the magnetic field consistent with the thin-shell dynamo model proposed for these planets. H 3 O could also be a major component in both solid and superionic forms in other (e.g., nonconvecting) layers. The results thus provide a materials basis for understanding the enigmatic magnetic-field anomalies and other aspects of the interiors of Uranus and Neptune. These findings have direct implications for the internal structure, composition, and dynamos of related exoplanets. 

The 3.6‐Myr sedimentary record of Lake El'gygytgyn is crucial for understanding the response of the sensitive ecosystems in the Arctic to Quaternary climate variations at orbital timescales. In this study, we synthesize previously published pollen records and biome reconstructions and perform pollen diversity analysis of the deep‐drilling core ICDP 5011‐1 from Lake El'gygytgyn for periods during the Early Pleistocene (MIS 82 – MIS 79), Early–Middle Pleistocene (MIS 31 – MIS 18) and late Middle Pleistocene (MIS 7e – MIS 6f). The results indicate that the predominance of herb tundra in the regional vegetation was most characteristic during glacials/stadials. Interglacials, in contrast, can be distinguished by the expansion of shrub communities mainly composed of birch, alder and willow. The expansion of forest biomes in the region was influenced by peaks in obliquity values, which led to increases in daylight length, which was essential for plant growth in high latitudes. An apparent long‐term decreasing trend in the tree and shrub population, accompanied by a reduction in floristic richness, was induced by stepwise cooling and drying since the Mid‐Pleistocene Transition (MPT), which is linked to the modulation of extended global ice volume during the MPT via strong snow‐ and ice‐albedo feedback effects.