arXiv:2604.23874v3 Announce Type: replace-cross Abstract: The differentiable physics paradigm may be leveraged as an a-posteriori approach for discovering turbulence closure models by embedding a neural network parameterization directly inside the solver and optimizing it given potentially sparse target data. This addresses a key limitation of a-priori learning where direct numerical simulation (DNS) data is used to approximate the subgrid stress with the assumption of a low-pass filter. Closures trained in this a-priori manner frequently lead to unstable deployments due to the mismatch betwee
The increasing computational power and advancements in differentiable programming are enabling more sophisticated AI applications in scientific modeling, allowing for integration of neural networks directly into solvers.
This development allows for more accurate and stable turbulence closure models, critical for fields like climate modeling, aerospace engineering, and energy, which traditionally struggle with computationally expensive or unstable simulations.
The shift from a-priori to a-posteriori learning of turbulence models reduces the reliance on direct numerical simulation data and promises more robust deployment of AI-enhanced physical simulations.
- · Aerospace engineering
- · Climate scientists
- · Fluid dynamics researchers
- · AI/ML researchers in scientific computing
- · Traditional 'a priori' turbulence modeling approaches
- · Brute-force DNS simulation methods
Improved accuracy and stability in complex fluid dynamics simulations, leading to better predictive models.
Accelerated design and optimization cycles for systems dependent on turbulence modeling, such as aircraft, wind turbines, and climate prediction models.
Potential for new materials and energy solutions enabled by highly precise and rapid simulation of fluid interactions, impacting manufacturing and sustainability.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG