Learned discretizations for passive scalar advection in a 2-D turbulent flow

Abstract

The computational cost of fluid simulations increases rapidly with grid resolution. This has given a hard limit on the ability of simulations to accurately resolve small scale features of complex flows. Here we use a machine learning approach to learn a numerical discretization that retains high accuracy even when the solution is under-resolved with classical methods. We apply this approach to passive scalar advection in a two-dimensional turbulent flow. The method maintains the same accuracy as traditional high-order flux-limited advection solvers, while using 4x lower grid resolution in each dimension. The machine learning component is tightly integrated with traditional finite-volume schemes and can be trained via an end-to-end differentiable programming framework. The solver can achieve near-peak hardware utilization on CPUs and accelerators via convolutional filters. Code is available at https://github.com/google-research/data-driven-pdes.

14 pages, 13 figures

References in corpus (7)

• Hidden Physics Models: Machine Learning of Nonlinear Partial Differential Equations
• Deep Learning for Physical Processes: Incorporating Prior Scientific Knowledge
• Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers
• TensorFlow Eager: A Multi-Stage, Python-Embedded DSL for Machine Learning
• Physical invariance in neural networks for subgrid-scale scalar flux modeling
• Using Machine Learning to Augment Coarse-Grid Computational Fluid Dynamics Simulations
• Computers and turbulence

Cited by in corpus (18)

• Machine learning accelerated computational fluid dynamics
• Machine Learning in Aerodynamic Shape Optimization
• JAX-FLUIDS: A fully-differentiable high-order computational fluid dynamics solver for compressible two-phase flows
• Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations
• Unsupervised Discovery of Inertial-Fusion Plasma Physics using Differentiable Kinetic Simulations and a Maximum Entropy Loss Function
• Explicit physics-informed neural networks for non-linear upscaling closure: the case of transport in tissues
• Error-correcting neural networks for semi-Lagrangian advection in the level-set method
• Variational Data Assimilation with a Learned Inverse Observation Operator
• A learned conservative semi-Lagrangian finite volume scheme for transport simulations
• Deep neural network for solving differential equations motivated by Legendre-Galerkin approximation
• Reducing Spatial Discretization Error on Coarse CFD Simulations Using an OpenFOAM-Embedded Deep Learning Framework
• Physics-aware Spatiotemporal Modules with Auxiliary Tasks for Meta-Learning
• STENCIL-NET: Data-driven solution-adaptive discretization of partial differential equations
• Learnable-Differentiable Finite Volume Solver for Accelerated Simulation of Flows
• A Model-Constrained Tangent Slope Learning Approach for Dynamical Systems
• Solving continuum and rarefied flows using differentiable programming
• A Latent space solver for PDE generalization
• Error-Correcting Neural Networks for Two-Dimensional Curvature Computation in the Level-Set Method