Turbulence Scaling from Deep Learning Diffusion Generative Models DOI Creative Commons
Tim Whittaker, Romuald A. Janik, Yaron Oz

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Complex spatial and temporal structures are inherent characteristics of turbulent fluid flows comprehending them poses a major challenge. This comprehesion necessitates an understanding the space flow configurations. We employ diffusion-based generative model to learn distribution vorticity profiles generate snapshots solutions incompressible Navier-Stokes equations. consider inverse cascade in two dimensions diverse that differ from those training dataset. analyze statistical scaling properties new profiles, calculate their structure functions, energy power spectrum, velocity probability function moments local dissipation. All learnt exponents consistent with expected Kolmogorov scaling. agreement established turbulence provides strong evidence model's capability capture essential features real-world turbulence.

Language: Английский

On the spatial prediction of the turbulent flow behind an array of cylinders via echo state networks DOI Creative Commons
Mohammad Sharifi Ghazijahani, Christian Cierpka

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 144, P. 110079 - 110079

Published: Jan. 23, 2025

Language: Английский

Citations

2
Accelerating the Discovery of Steady‐States of Planetary Interior Dynamics With Machine Learning DOI Creative Commons
Siddhant Agarwal, Nicola Tosi, Christian Hüttig

et al.

Journal of Geophysical Research Machine Learning and Computation, Journal Year: 2025, Volume and Issue: 2(1)

Published: March 1, 2025

Abstract Simulating mantle convection often requires reaching a computationally expensive steady‐state, crucial for deriving scaling laws thermal and dynamical flow properties benchmarking numerical solutions. The strong temperature dependence of the rheology rocks causes viscosity variations several orders magnitude, leading to slow‐evolving “stagnant lid” where heat conduction dominates, overlying rapidly evolving strongly convecting region. Time‐stepping methods, while effective fluids with constant viscosity, are hindered by Courant criterion, which restricts time step based on system's maximum velocity grid size. Consequently, achieving steady‐state large number steps due disparate scales governing stagnant regions. We present concept accelerating simulations using machine learning. generate data set 128 two‐dimensional mixed basal internal heating, pressure‐ temperature‐dependent viscosity. train feedforward neural network 97 predict profiles. These can then be used initialize time‐stepping methods different simulation parameters. For an example application, required reach is reduced factor 2.8, compared typically initializations. benefit this method lies in requiring very few on, providing solution that numerically accurate as we method, posing minimal computational overhead at inference time. demonstrate effectiveness our approach discuss its potential advancing research.

Language: Английский

Citations

0
Turbulent mesoscale convection in the Boussinesq limit and beyond DOI Creative Commons
Shadab Alam, Dmitry Krasnov, Ambrish Pandey

et al.

International Journal of Heat and Fluid Flow, Journal Year: 2025, Volume and Issue: 115, P. 109856 - 109856

Published: May 21, 2025

Language: Английский

Citations

0
Turbulence scaling from deep learning diffusion generative models DOI Creative Commons
Tim Whittaker, Romuald A. Janik, Yaron Oz

et al.

Journal of Computational Physics, Journal Year: 2024, Volume and Issue: 514, P. 113239 - 113239

Published: July 2, 2024

Language: Английский

Citations

3
The atmospheric boundary layer: a review of current challenges and a new generation of machine learning techniques DOI Creative Commons
Linda Canché-Cab, Liliana San-Pedro, A. Bassam

et al.

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(12)

Published: Oct. 17, 2024

Language: Английский

Citations

0
On the Spatial Prediction of the Turbulent Flow Behind an Array of Cylinders Via Echo State Networks DOI
Mohammad Sharifi Ghazijahani, Christian Cierpka

Published: Jan. 1, 2024

Language: Английский

Citations

0
Turbulence Scaling from Deep Learning Diffusion Generative Models DOI Creative Commons
Tim Whittaker, Romuald A. Janik, Yaron Oz

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Complex spatial and temporal structures are inherent characteristics of turbulent fluid flows comprehending them poses a major challenge. This comprehesion necessitates an understanding the space flow configurations. We employ diffusion-based generative model to learn distribution vorticity profiles generate snapshots solutions incompressible Navier-Stokes equations. consider inverse cascade in two dimensions diverse that differ from those training dataset. analyze statistical scaling properties new profiles, calculate their structure functions, energy power spectrum, velocity probability function moments local dissipation. All learnt exponents consistent with expected Kolmogorov scaling. agreement established turbulence provides strong evidence model's capability capture essential features real-world turbulence.

Language: Английский

Citations

0