Search for: All records

Abstract This study integrates data from all broadband seismic stations in Alaska and northwestern Canada in 1999–2022 to construct a shear‐wave velocity model for south‐central Alaska and northwesternmost Canada, using ambient noise wave propagation simulation and inversion. Our model reveals three key features, including (a) the presence of the subducting Yakutat slab with apparent velocity reductions near the trench and within its flat segment, (b) two slab segments beneath the Wrangell volcanic field, differing in steepness, depth, and seismic velocity, and aligning spatially with the northwestern and southeastern volcano clusters, and (c) the existence of slab windows between the Yakutat and Wrangell slabs and between the northwestern and southeastern portions of the Wrangell slab. Our findings reinforce that the Wrangell volcanoes are predominantly influenced by subduction‐related magmatism. Furthermore, the two slab windows could have induced asthenospheric upwelling, contributing to the volcanism in the Wrangell clustered volcanoes. 

Abstract Current complex prediction models are the result of fitting deep neural networks, graph convolutional networks or transducers to a set of training data. A key challenge with these models is that they are highly parameterized, which makes describing and interpreting the prediction strategies difficult. We use topological data analysis to transform these complex prediction models into a simplified topological view of the prediction landscape. The result is a map of the predictions that enables inspection of the model results with more specificity than dimensionality-reduction methods such as tSNE and UMAP. The methods scale up to large datasets across different domains. We present a case study of a transformer-based model previously designed to predict expression levels of a piece of DNA in thousands of genomic tracks. When the model is used to study mutations in theBRCA1gene, our topological analysis shows that it is sensitive to the location of a mutation and the exon structure ofBRCA1in ways that cannot be found with tools based on dimensionality reduction. Moreover, the topological framework offers multiple ways to inspect results, including an error estimate that is more accurate than model uncertainty. Further studies show how these ideas produce useful results in graph-based learning and image classification. 

A typical subduction of an oceanic plate beneath a continent is expected to be accompanied by arc volcanoes along the convergent margin. However, subduction of the Cocos plate at the Middle American subduction system has resulted in an uneven distribution of magmatism/volcanism along strike. Here we construct a new three-dimensional shear-wave velocity model of the entire Middle American subduction system, using full-wave ambient noise tomography. Our model reveals significant variations of the oceanic plates along strike and down dip, in correspondence with either weakened or broken slabs after subduction. The northern and southern segments of the Cocos plate, including the Mexican flat slab subduction, are well imaged as high-velocity features, where a low-velocity mantle wedge exists and demonstrate a strong correlation with the arc volcanoes. Subduction of the central Cocos plate encounters a thick high-velocity feature beneath North America, which hinders the formation of a typical low-velocity mantle wedge and arc volcanoes. We suggest that the presence of slab tearing at both edges of the Mexican flat slab has been modifying the mantle flows, resulting in the unusual arc volcanism. 

Artificially designed modulators that enable a wealth of freedom in manipulating the terahertz (THz) waves at will are an essential component in THz sources and their widespread applications. Dynamically controlled metasurfaces, being multifunctional, ultrafast, integrable, broadband, high contrasting, and scalable on the operating wavelength, are critical in developing state-of-the-art THz modulators. Recently, external stimuli-triggered THz metasurfaces integrated with functional media have been extensively explored. The vanadium dioxide (VO2)-based hybrid metasurfaces, as a unique path toward active meta-devices, feature an insulator–metal phase transition under the excitation of heat, electricity, and light, etc. During the phase transition, the optical and electrical properties of the VO2 film undergo a massive modification with either a boosted or dropped conductivity by more than four orders of magnitude. Being benefited from the phase transition effect, the electromagnetic response of the VO2-based metasufaces can be actively controlled by applying external excitation. In this review, we present recent advances in dynamically controlled THz metasurfaces exploiting the VO2 phase transition categorized according to the external stimuli. THz time-domain spectroscopy is introduced as an indispensable platform in the studies of functional VO2 films. In each type of external excitation, four design strategies are employed to realize external stimuli-triggered VO2-based THz metasurfaces, including switching the transreflective operation mode, controlling the dielectric environment of metallic microstructures, tailoring the equivalent resonant microstructures, and modifying the electromagnetic properties of the VO2 unit cells. The microstructures’ design and electromagnetic responses of the resulting active metasurfaces have been systematically demonstrated, with a particular focus on the critical role of the VO2 films in the dynamic modulation processes. 

Supervised machine learning approaches have been increasingly used in accelerating electronic structure prediction as surrogates of first-principle computational methods, such as density functional theory (DFT). While numerous quantum chemistry datasets focus on chemical properties and atomic forces, the ability to achieve accurate and efficient prediction of the Hamiltonian matrix is highly desired, as it is the most important and fundamental physical quantity that determines the quantum states of physical systems and chemical properties. In this work, we generate a new Quantum Hamiltonian dataset, named as QH9, to provide precise Hamiltonian matrices for 2,399 molecular dynamics trajectories and 130,831 stable molecular geometries, based on the QM9 dataset. By designing benchmark tasks with various molecules, we show that current machine learning models have the capacity to predict Hamiltonian matrices for arbitrary molecules. Both the QH9 dataset and the baseline models are provided to the community through an open-source benchmark, which can be highly valuable for developing machine learning methods and accelerating molecular and materials design for scientific and technological applications. Our benchmark is publicly available at \url{https://github.com/divelab/AIRS/tree/main/OpenDFT/QHBench}. 

Supervised machine learning approaches have been increasingly used in accelerating electronic structure prediction as surrogates of first-principle computational methods, such as density functional theory (DFT). While numerous quantum chemistry datasets focus on chemical properties and atomic forces, the ability to achieve accurate and efficient prediction of the Hamiltonian matrix is highly desired, as it is the most important and fundamental physical quantity that determines the quantum states of physical systems and chemical properties. In this work, we generate a new Quantum Hamiltonian dataset, named as QH9, to provide precise Hamiltonian matrices for 2,399 molecular dynamics trajectories and 130,831 stable molecular geometries, based on the QM9 dataset. By designing benchmark tasks with various molecules, we show that current machine learning models have the capacity to predict Hamiltonian matrices for arbitrary molecules. Both the QH9 dataset and the baseline models are provided to the community through an open-source benchmark, which can be highly valuable for developing machine learning methods and accelerating molecular and materials design for scientific and technological applications. Our benchmark is publicly available at \url{https://github.com/divelab/AIRS/tree/main/OpenDFT/QHBench}. 

Supervised machine learning approaches have been increasingly used in accelerating electronic structure prediction as surrogates of first-principle computational methods, such as density functional theory (DFT). While numerous quantum chemistry datasets focus on chemical properties and atomic forces, the ability to achieve accurate and efficient prediction of the Hamiltonian matrix is highly desired, as it is the most important and fundamental physical quantity that determines the quantum states of physical systems and chemical properties. In this work, we generate a new Quantum Hamiltonian dataset, named as QH9, to provide precise Hamiltonian matrices for 999 molecular dynamics trajectories and 130,831 stable molecular geometries, based on the QM9 dataset. By designing benchmark tasks with various molecules, we show that current machine learning models have the capacity to predict Hamiltonian matrices for arbitrary molecules. Both the QH9 dataset and the baseline models are provided to the community through an open-source benchmark, which can be highly valuable for developing machine learning methods and accelerating molecular and materials design for scientific and technological applications. Our benchmark is publicly available at https://github.com/divelab/AIRS/tree/main/OpenDFT/QHBench.