Search for: All records

Climate changes during the mid- to late-Holocene, after the last vestiges of glacial ice sheets dwindled, provide important context for climate change today. In the tropical Andes, most of the continuous paleoclimate records covering the late Holocene are derived from the oxygen isotope composition of ice cores, speleothems, and lake carbonates. These archives are powerful recorders of large-scale changes in circulation and monsoon intensity, but they do not necessarily capture local moisture balance, and so reconstructions of local precipitation and aridity remain scarce. Here we present contrasting histories of local effective moisture vs. regional circulation from several new biomarker records preserved in lakes and peat in the Colombian and Peruvian Andes. We focus on the hydrogen isotope composition of long-chain plant waxes, which reflects precipitation δ2H similarly to δ18O from ice cores and speleothems; and the δ13C of waxes and the δ2H of mid-chain waxes, which reflect local water stress and effective moisture. In both the Northern and Southern Hemisphere tropical Andes, fairly gradual δ2H shifts during the late Holocene indicate a progressive intensification of circulation in the South American lowlands. On the other hand, plant wax δ13C and mid-chain δ2H records indicate abrupt transitions into and out of intervals of water stress and aridity – similar to findings from pollen and sediment lithology from elsewhere in the tropical Andes. We draw on climate models and proxy data syntheses to help reconcile these curiously different accounts of effective moisture in the tropical Andes since the mid-Holocene and discuss implications for modern climate research. 

We introduce SignNet and BasisNet---new neural architectures that are invariant to two key symmetries displayed by eigenvectors: (i) sign flips, since if v is an eigenvector then so is -v; and (ii) more general basis symmetries, which occur in higher dimensional eigenspaces with infinitely many choices of basis eigenvectors. We prove that under certain conditions our networks are universal, i.e., they can approximate any continuous function of eigenvectors with the desired invariances. When used with Laplacian eigenvectors, our networks are provably more expressive than existing spectral methods on graphs; for instance, they subsume all spectral graph convolutions, certain spectral graph invariants, and previously proposed graph positional encodings as special cases. Experiments show that our networks significantly outperform existing baselines on molecular graph regression, learning expressive graph representations, and learning neural fields on triangle meshes. Our code is available at https://github.com/cptq/SignNet-BasisNet. 

Abstract High-pressure electrical resistivity measurements reveal that the mechanical deformation of ultra-hard WB 2 during compression induces superconductivity above 50 GPa with a maximum superconducting critical temperature, T c of 17 K at 91 GPa. Upon further compression up to 187 GPa, the T c gradually decreases. Theoretical calculations show that electron-phonon mediated superconductivity originates from the formation of metastable stacking faults and twin boundaries that exhibit a local structure resembling MgB 2 (hP3, space group 191, prototype AlB 2 ). Synchrotron x-ray diffraction measurements up to 145 GPa show that the ambient pressure hP12 structure (space group 194, prototype WB 2 ) continues to persist to this pressure, consistent with the formation of the planar defects above 50 GPa. The abrupt appearance of superconductivity under pressure does not coincide with a structural transition but instead with the formation and percolation of mechanically-induced stacking faults and twin boundaries. The results identify an alternate route for designing superconducting materials.