27 May 2025 in Physical Review Letters
Polarons are crucial for charge transport in semiconductors, significantly impacting material properties and device performance. The dynamics of small polarons can be investigated using first-principles molecular dynamics. However, the limited timescale of these simulations presents a challenge for adequately sampling infrequent polaron hopping events. Here, we introduce a message-passing neural network combined with first-principles molecular dynamics within the Born-Oppenheimer approximation that learns the polaronic potential energy surface by encoding the polaronic state, allowing for simulations of polaron hopping dynamics at the nanosecond scale. By leveraging the statistical significance of the long timescale, our framework can accurately estimate polaron (anisotropic) mobilities and activation barriers in prototypical polaronic oxides across different scenarios (hole polarons in rocksalt MgO and electron polarons in pristine and F-doped rutile TiO 2 ) within experimentally measured ranges.
6 May 2024 in npj Computational Materials
The multifaceted physics of oxides is shaped by their composition and the presence of defects, which are often accompanied by the formation of polarons. The simultaneous presence of polarons and defects, and their complex interactions, pose challenges for first-principles simulations and experimental techniques. In this study, we leverage machine learning and a first-principles database to analyze the distribution of surface oxygen vacancies (VO) and induced small polarons on rutile TiO2(110), effectively disentangling the interactions between polarons and defects. By combining neural-network supervised learning and simulated annealing, we elucidate the inhomogeneous VOdistribution observed in scanning probe microscopy (SPM). Our approach allows us to understand and predict defective surface patterns at enhanced length scales, identifying the specific role of individual types of defects. Specifically, surface-polaron-stabilizing VO-configurations are identified, which could have consequences for surface reactivity.
4 April 2026 in SSRN Electronic Journal
Deep learning-based surrogate models can substantially accelerate pluvial flood simulations. However, most existing approaches struggle to generalize across heterogeneous terrains or modified topographies, limiting their applicability for real-world flood risk assessment unless retrained for each specific terrain.In this work, we investigate how deep learning flood surrogates can be designed to achieve robust spatial generalization across previously unseen terrain configurations without retraining. We train models on diverse terrain and introduce upstream-aware input maps that propagate terrain properties into downstream regions, allowing the model to capture long-range spatial dependencies and supporting the learning of more generalizable spatial features. In addition, we add architectural design choices to improve the model’s capacity to generalize beyond the training domain. We further propose a two-stage training strategy with an adapter-like rainfall encoder that enables efficient adaptation to previously unseen rainfall intensities while preserving spatial generalizability.The resulting model predicts maximum water depths and is evaluated on unseen catchments in Austria and Denmark, as well as on artificially modified terrains such as dikes. Our results demonstrate that combining a large training dataset with upstream-aware terrain representations and a generalization-oriented architectural design enables accurate flood predictions across unseen terrains and rainfall scenarios, reaching a RMSE below $0.05$ and CSI$_{0.3}$ of up to $0.91$ in test cases, outperforming previous models in both prediction accuracy and Critical Success Index.
25 September 2026 in Journal of Hydrology