Accurate characterization of subducting slab geometry is fundamental to understanding the distribution of earthquakes, the dynamics of arc volcanism, and the assessment of seismic hazards. Well-constrained slab structures also serve as critical inputs for geophysical imaging and geodynamic modeling efforts that aim to resolve key processes in subduction zones. In this study, we present a comprehensive, margin-wide model of the Moho associated with the subducting oceanic plate beneath the Cascadia subduction zone, developed through the integration of publicly available offshore and onshore datasets. We integrate high-resolution seismic reflection data from the offshore CASIE21 expedition with three previously published, lower-resolution onshore slab models (McCrory et al., 2012; Hayes et al., 2018; Bloch et al., 2023) to construct a unified Moho surface. This synthesis produces six alternative Moho geometries, enabling flexibility for studies that require varying structural assumptions. The accompanying open-source workflow offers a transparent and adaptable approach for combining heterogeneous datasets. In areas lacking direct constraints, Moho depths were estimated through interpolation from adjacent regions. The resulting models provide a valuable foundation for analyzing along-strike variations in slab structure and their implications for Cascadia geodynamics.
<p>East Asia has been identified as a key area at risk of precipitation increases resulting from global warming. The East Asian summer monsoon has distinct regional inter-monthly precipitation patterns, and the simulation characteristics of global climate models therefore need to be evaluated closely to obtain reliable projections of future precipitation patterns and associated extreme events. Using metrics of inter-monthly variability in monsoon precipitation over East Asia, this study evaluates the performance of Coupled Model Intercomparison Project Phase 6 (CMIP6) models and analyzes future projections and uncertainty factors. Regional inter-monthly precipitation patterns were simulated reasonably well by the CMIP6 models but with weaker rainfall amplitudes. The CMIP6 models simulated more intense precipitation than their predecessor, the CMIP5 models, and captured observations better. Future projections indicate an overall precipitation increase during both the northward migration of the rain band and the peak monsoon time over East Asia and the three subregions, with stronger changes occurring in the higher emission scenarios. This precipitation increase can be mainly ascribed to a thermodynamic response due to the increased moisture availability associated with global warming. Internal climate variability and model uncertainty largely explained future precipitation uncertainties, which are associated with tropical ocean warming patterns. Dynamic terms explained a large portion of the model uncertainty linked to circulation changes, whereas thermodynamic terms were significantly related to scenario uncertainty.</p>
This study aims to image the crustal structure of the western Tibetan Plateau by analyzing the velocity structure of elastic waves, using manually picked P- and S-wave arrival times from waveform data recorded by temporarily installed seismic stations in western Tibet. Preliminary events located using the VELEST algorithm resulted in the development of a 1-D velocity model through inversion, which was then used in the TomoDD algorithm to relocate earthquakes and generate a high-resolution 3-D velocity structure model. A significant number of events were located between the Karakoram fault (KKF), Main Boundary Thrust, and Main Central Thrust. A low P-wave anomaly of approximately ~8% is noted in the vicinity of the KKF, while a significant low P-wave anomaly is also observed in the crust beneath the western margin. A low P-wave anomaly is concentrated beneath the Lhasa block, whereas a relatively higher P-wave anomaly is evident in the Himalayan terrane. The KKF dips beneath the Tibetan plateau towards the northeast. Evidence of partial melting in the crust beneath the Tibetan plateau and mid-crustal channel flow of slower crustal material from the plateau towards the Himalayan range can also be delineated through the observed velocity structures found in the study.
Geometric Graph Neural Networks (GNNs) and Transformers have become state-of-the-art for learning from 3D protein structures. However, their reliance on message passing prevents them from capturing the hierarchical interactions that govern protein function, such as global domains and long-range allosteric regulation. In this work, we argue that the network architecture itself should mirror this biological hierarchy. We introduce Geometric Graph U-Nets, a new class of models that learn multi-scale representations by recursively coarsening and refining the protein graph. We prove that this hierarchical design can theoretically more expressive than standard Geometric GNNs. Empirically, on the task of protein fold classification, Geometric U-Nets substantially outperform invariant and equivariant baselines, demonstrating their ability to learn the global structural patterns that define protein folds. Our work provides a principled foundation for designing geometric deep learning architectures that can learn the multi-scale structure of biomolecules.
Sizhe Ma, Katherine A. Flanigan, Mario Bergés
et al.
Indirect structural health monitoring (iSHM) for broken rail detection using onboard sensors presents a cost-effective paradigm for railway track assessment, yet reliably detecting small, transient anomalies (2-10 cm) remains a significant challenge due to complex vehicle dynamics, signal noise, and the scarcity of labeled data limiting supervised approaches. This study addresses these issues through unsupervised deep learning. We introduce an incremental synthetic data benchmark designed to systematically evaluate model robustness against progressively complex challenges like speed variations, multi-channel inputs, and realistic noise patterns encountered in iSHM. Using this benchmark, we evaluate several established unsupervised models alongside our proposed Attention-Focused Transformer. Our model employs a self-attention mechanism, trained via reconstruction but innovatively deriving anomaly scores primarily from deviations in learned attention weights, aiming for both effectiveness and computational efficiency. Benchmarking results reveal that while transformer-based models generally outperform others, all tested models exhibit significant vulnerability to high-frequency localized noise, identifying this as a critical bottleneck for practical deployment. Notably, our proposed model achieves accuracy comparable to the state-of-the-art solution while demonstrating better inference speed. This highlights the crucial need for enhanced noise robustness in future iSHM models and positions our more efficient attention-based approach as a promising foundation for developing practical onboard anomaly detection systems.
Kai-Yuan Chen, Kai-Hsin Chen, Jyh-Shing Roger Jang
We propose a profitable trading strategy for the cryptocurrency market based on grid trading. Starting with an analysis of the expected value of the traditional grid strategy, we show that under simple assumptions, its expected return is essentially zero. We then introduce a novel Dynamic Grid-based Trading (DGT) strategy that adapts to market conditions by dynamically resetting grid positions. Our backtesting results using minute-level data from Bitcoin and Ethereum between January 2021 and July 2024 demonstrate that the DGT strategy significantly outperforms both the traditional grid and buy-and-hold strategies in terms of internal rate of return and risk control.
We present a theoretical analysis of the DIC-DAC-DOA algorithm, a non-stoquastic quantum algorithm for solving the Maximum Independent Set (MIS) problem. The algorithm runs in polynomial time and achieves exponential speedup over both transverse-field quantum annealing (TFQA) and classical algorithms on a structured family of NP-hard MIS instances, under assumptions supported by analytical and numerical evidence. The core of this speedup lies in the ability of the evolving ground state to develop both positive and negative amplitudes, enabled by the non-stoquastic XX-driver. This sign structure permits quantum interference that produces negative amplitudes in the computational basis, allowing efficient evolution paths beyond the reach of stoquastic algorithms, whose ground states remain strictly non-negative. In our analysis, the efficiency of the algorithm is measured by the presence or absence of an anti-crossing, rather than by spectral gap estimation as in traditional approaches. The key idea is to infer it from the crossing behavior of bare energy levels of relevant subsystems associated with the degenerate local minima (LM) and the global minimum (GM). The cliques of the critical LM, responsible for the anti-crossing in TFQA, can be efficiently identified to form the XX-driver graph. The resulting speedup can be attributed to two mechanisms: in the first stage, energy-guided localization within the same-sign block steers the ground state smoothly into the GM-supporting region, while in the second stage, the opposite-sign blocks are invoked and sign-generating quantum interference drives the evolution along an opposite-sign path. Finally, we derive scalable reduced models that provide a concrete opportunity for verification of the quantum advantage mechanism on currently available universal quantum computers.
Accurate representation of large earthquake sources is required for understanding rupture dynamics and improving seismic hazard assessments. While capable of capturing complex spatio-temporal slip scenarios, traditional finite-fault models often suffer from over-parameterization, require strong regularization, and pose significant computational challenges, especially in rapid-response scenarios. Conversely, multiple point source (MPS) models reduce the rupture as a sequence of point sources but are inadequate to simulate short-period wavefield and static displacement. We introduce a hybrid source representation that leverages moment tensor interpolation to bridge the gap between these models. By treating moment tensors as "key" centroids of a tensor field, we construct geometrically coherent slip models that retain the spatial complexity of finite-fault models while maintaining MPS's computational efficiency and simplicity. Our method extends existing 2D tensor-field reconstruction techniques to moment tensors, allowing source-type-preserving interpolation and enabling sparse model approximation and source upscaling for numerical simulations. We demonstrate how our approach can benefit both the inverse and forward problems on the January 2024 Noto earthquake, computing a sparse approximation of the USGS NEIC source model with fewer than ten key tensors and computing full wavefield and static deformation from upscaled source distributions in a realistic 3D regional tomographic model using spectral-elements method.
This paper proposes a multitask learning framework for probabilistic model updating by jointly using strain and acceleration measurements. This framework can enhance the structural damage assessment and response prediction of existing steel frame structures with quantified uncertainty. Multitask learning may be used to address multiple similar inference tasks simultaneously to achieve a more robust prediction performance by transferring useful knowledge from one task to another, even in situations of data scarcity. In the proposed model-updating procedure, a spatial frame is decomposed into multiple planar frames that are viewed as multiple tasks and jointly analyzed based on the hierarchical Bayesian model, leading to robust estimation results. The procedure uses a displacement-stress relationship in the modal space because it directly reflects the elemental stiffness and requires no prior knowledge concerning the mass, unlike most existing model-updating techniques. Validation of the proposed framework by using a full-scale vibration test on a one-story, one-bay by one-bay moment resisting steel frame, wherein structural damage to the column bases is simulated by loosening the anchor bolts, is presented. The experimental results suggest that the displacement-stress relationship has sufficient sensitivity toward localized damage, and the Bayesian multitask learning approach may result in the efficient use of measurements such that the uncertainty involved in model parameter estimation is reduced. The proposed framework facilitates more robust and informative model updating.
Nan Zhang, Christian Vergara-Marcillo, Georgios Diamantopoulos
et al.
Dynamic data-driven Digital Twins (DDTs) can enable informed decision-making and provide an optimisation platform for the underlying system. By leveraging principles of Dynamic Data-Driven Applications Systems (DDDAS), DDTs can formulate computational modalities for feedback loops, model updates and decision-making, including autonomous ones. However, understanding autonomous decision-making often requires technical and domain-specific knowledge. This paper explores using large language models (LLMs) to provide an explainability platform for DDTs, generating natural language explanations of the system's decision-making by leveraging domain-specific knowledge bases. A case study from smart agriculture is presented.
This article introduces a novel numerical scheme within the finite element method (FEM) to study soil heterogeneity, specifically focusing on the root–soil matrix in fracture treatments. Material properties, such as Young’s modulus of elasticity, cohesion, and the friction angle, are considered as randomly distributed variables. To address the inherent uncertainty associated with these distributions, a Monte Carlo simulation is employed. By incorporating the uncertainties related to material properties, particularly the root component that contributes to soil heterogeneity, this article provides a reliable estimation of the factor of safety, failure surface, and slope deformation, all of which demonstrate a progressive behavior. The probability distribution curve for the factor of safety (FOS) reveals that an increase in the root area ratio (RAR) results in a narrower range and greater certainty in the population mean, indicating reduced material variation. Moreover, as the slope angle increases, the sample mean falls within a wider range of the probability density curve, indicating an enhanced level of material heterogeneity. This heterogeneity amplifies the level of uncertainty when predicting the factor of safety, highlighting the crucial importance of accurate information regarding heterogeneity to enhancing prediction accuracy.
This paper proposes a dynamic regression (DR) framework that enhances existing deep spatiotemporal models by incorporating structured learning for the error process in traffic forecasting. The framework relaxes the assumption of time independence by modeling the error series of the base model (i.e., a well-established traffic forecasting model) using a matrix-variate autoregressive (AR) model. The AR model is integrated into training by redesigning the loss function. The newly designed loss function is based on the likelihood of a non-isotropic error term, enabling the model to generate probabilistic forecasts while preserving the original outputs of the base model. Importantly, the additional parameters introduced by the DR framework can be jointly optimized alongside the base model. Evaluation on state-of-the-art (SOTA) traffic forecasting models using speed and flow datasets demonstrates improved performance, with interpretable AR coefficients and spatiotemporal covariance matrices enhancing the understanding of the model.
M. G. Kleinhans, L. Roelofs, S. A. H. Weisscher
et al.
<p>Rivers and estuaries are flanked by floodplains built by mud and vegetation. Floodplains affect channel dynamics and the overall system's pattern through apparent cohesion in the channel banks and through filling of accommodation space and hydraulic resistance. For rivers, effects of mud, vegetation and the combination are thought to stabilise the banks and narrow the channel. However, the thinness of estuarine floodplain, comprised of salt marsh and mudflats, compared to channel depth raises questions about the possible effects of floodplain as constraints on estuary dimensions. To test these effects, we created three estuaries in a tidal flume: one with recruitment events of two live vegetation species, one with mud and a control with neither. Both vegetation and mud reduced channel migration and bank erosion and stabilised channels and bars. Effects of vegetation include local flow velocity reduction and concentration of flow into the channels, while flow velocities remained higher over mudflats. On the other hand, the lower reach of the muddy estuary showed more reduced channel migration than the vegetated estuary. The main system-wide effect of mudflats and salt marsh is to reduce the tidal prism over time from upstream to downstream. The landward reach of the estuary narrows and fills progressively, particularly for the muddy estuary, which effectively shortens the tidally influenced reach and also reduces the tidal energy in the seaward reach and mouth area. As such, estuaries with sufficient sediment supply are limited in size by tidal prism reduction through floodplain formation.</p>
The dynamics of materials failure is one of the most critical phenomena in a range of scientific and engineering fields, from healthcare to structural materials to transportation. In this paper we propose a specially designed deep neural network, DyFraNet, which can predict dynamic fracture behaviors by identifying a complete history of fracture propagation - from cracking onset, as a crack grows through the material, modeled as a series of frames evolving over time and dependent on each other. Furthermore, this model can not only forecast future fracture processes but also backcast to elucidate the past fracture history. In this scenario, once provided with the outcome of a fracture event, the model will elucidate past events that led to this state and will predict the future evolution of the failure process. By comparing the predicted results with atomistic-level simulations and theory, we show that DyFraNet can capture dynamic fracture mechanics by accurately predicting how cracks develop over time, including measures such as the crack speed, as well as when cracks become unstable. We use GradCAM to interpret how DyFraNet perceives the relationship between geometric conditions and fracture dynamics and we find DyFraNet pays special attention to the areas around crack tips, which have a critical influence in the early stage of fracture propagation. In later stages, the model pays increased attention to the existing or newly formed damage distribution in the material. The proposed approach offers significant potential to accelerate the exploration of the dynamics in material design against fracture failures and can be beneficially adapted for all kinds of dynamical engineering problems.
Biological functions of RNAs are determined by their three-dimensional (3D) structures. Thus, given the limited number of experimentally determined RNA structures, the prediction of RNA structures will facilitate elucidating RNA functions and RNA-targeted drug discovery, but remains a challenging task. In this work, we propose a Graph Neural Network (GNN)-based scoring function trained only with the atomic types and coordinates on limited solved RNA 3D structures for distinguishing accurate structural models. The proposed Physics-aware Multiplex Graph Neural Network (PaxNet) separately models the local and non-local interactions inspired by molecular mechanics. Furthermore, PaxNet contains an attention-based fusion module that learns the individual contribution of each interaction type for the final prediction. We rigorously evaluate the performance of PaxNet on two benchmarks and compare it with several state-of-the-art baselines. The results show that PaxNet significantly outperforms all the baselines overall, and demonstrate the potential of PaxNet for improving the 3D structure modeling of RNA and other macromolecules. Our code is available at https://github.com/zetayue/Physics-aware-Multiplex-GNN.