Deep Learning Approach for Turbulence Closure Models Using Data Assimilation
Researchers developed a deep learning method for discovering turbulence closure models that combines the efficiency of a-priori training with the stability of a-posteriori approaches using continuous data assimilation. The method trains neural networks on sparse DNS data without modifying the LES solver and explicitly conditions models on numerical schemes to improve generalization. This addresses a critical challenge in computational fluid dynamics where existing closure models often fail due to mismatches between training assumptions and actual solver behavior.
The study presents a novel framework for turbulence closure modeling that leverages the differentiable physics paradigm and continuous data assimilation. Traditional a-priori approaches train closures on direct numerical simulation data but often produce unstable deployments due to filter mismatches with numerical discretizations. Conversely, a-posteriori methods maintain stability but require expensive backpropagation through large eddy simulation solvers and significant solver modifications. The proposed approach enables stable, efficient training using sparsely observed DNS data without solver modification, while explicitly conditioning the learned correction on numerical schemes to track discretization errors. The framework was validated on two- and three-dimensional canonical test cases, demonstrating systematic adaptation to different discretizations.
What's missing
The paper does not discuss computational cost comparisons between the proposed method and existing a-posteriori approaches, nor does it provide quantitative metrics for generalization performance across different numerical schemes beyond the canonical test cases presented.
What different sources said
- arXiv cs.LGCenter
Deep Learning of Solver-Aware Turbulence Closures from Nudged LES Dynamics
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.