Differentiable Turbulence

Differentiable turbulence focuses on integrating machine learning (ML) models directly into computational fluid dynamics (CFD) solvers to improve the accuracy and efficiency of turbulent flow simulations, particularly for large eddy simulations (LES). Current research emphasizes developing and benchmarking various ML architectures, including convolutional neural networks, diffusion models, and graph neural networks, often within a differentiable framework to enable end-to-end training and optimization of the coupled physics-ML system. This approach aims to address the persistent challenge of accurately modeling unresolved scales in turbulent flows, potentially leading to significant speedups in CFD simulations and improved predictions across various applications.