• Corpus ID: 220280657

Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers

  title={Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers},
  author={Kiwon Um and Yun Fei and Philipp Holl and Robert Brand and Nils Th{\"u}rey},
Finding accurate solutions to partial differential equations (PDEs) is a crucial task in all scientific and engineering disciplines. It has recently been shown that machine learning methods can improve the solution accuracy by correcting for effects not captured by the discretized PDE. We target the problem of reducing numerical errors of iterative PDE solvers and compare different learning approaches for finding complex correction functions. We find that previously used learning approaches are… 

NeuralSim: Augmenting Differentiable Simulators with Neural Networks

Through extensive experiments, the ability of the hybrid simulator to learn complex dynamics involving frictional contacts from real data, as well as match known models of viscous friction, is demonstrated, and an approach for automatically discovering useful augmentations is presented.

Learning Incompressible Fluid Dynamics from Scratch - Towards Fast, Differentiable Fluid Models that Generalize

A novel physics-constrained training approach that generalizes to new fluid domains, requires no fluid simulation data, and allows convolutional neural networks to map a fluid state from time-point t to a subsequent state at time t + dt in a single forward pass is proposed, which simplifies the pipeline to train and evaluate neural fluid models.

A Latent space solver for PDE generalization

The solver uses an iterative inferencing strategy combined with solution initialization to improve generalization of PDE solutions and can generalize well to several PDE conditions.

Global Transport for Fluid Reconstruction with Learned Self-Supervision

A novel method to reconstruct volumetric flows from sparse views via a global transport formulation that reconstructs its motion based on a single initial state and introduces a learned self-supervision that constrains observations from unseen angles.

Leveraging Stochastic Predictions of Bayesian Neural Networks for Fluid Simulations

It is shown that BNN-based corrector networks can stabilize coarse-grained simulations and successfully create multimodal trajectories and two use-cases of a combination of BNNs with fluid simulations: uncertainty estimation and multi-modal synthesis.

A Differentiable Solver Approach to Operator Inference

This work reformulates the problem of operator inference as a constrained optimization problem allowing to relax on the required regularization, and produces a robust model order reduction approach for real-world industrial applications, validated along a dynamics complex 3D cooling process of a multi-tubular reactor using a commercial software package.


This work reformulates the learning problem by introducing an upper bound on the prediction error of a physical-statistical model, and generalizes several existing hybrid schemes proposed in the literature.

Half-Inverse Gradients for Physical Deep Learning

This work analyzes the characteristics of both physical and neural network optimizations to derive a new method that converges more quickly and yields better solutions, which is based on a halfinversion of the Jacobian and combines principles of both classical network and physics optimizers to solve the combined optimization task.

Hybrid Neural Network PDE Solvers for Reacting Flows

It is demonstrated that results obtained from non-reactive PDEs can be efficiently corrected at every time step by a hybrid neural network – PDE solver model so that the effect of the chemical reactions on the flow field is correctly accounted for.


  • 2022



Learning Neural PDE Solvers with Convergence Guarantees

This work proposes an approach to learn a fast iterative solver tailored to a specific domain by learning to modify the updates of an existing solver using a deep neural network, and achieves 2-3 times speedup compared to state-of-the-art solvers.

Learning data-driven discretizations for partial differential equations

Data-driven discretization is proposed, a method for learning optimized approximations to PDEs based on actual solutions to the known underlying equations that uses neural networks to estimate spatial derivatives.

Learning to Control PDEs with Differentiable Physics

It is shown that by using a differentiable PDE solver in conjunction with a novel predictor-corrector scheme, this work can train neural networks to understand and control complex nonlinear physical systems over long time frames.

Accelerating Eulerian Fluid Simulation With Convolutional Networks

This work proposes a data-driven approach that leverages the approximation power of deep-learning with the precision of standard solvers to obtain fast and highly realistic simulations of the Navier-Stokes equations.

PDE-Net: Learning PDEs from Data

Numerical experiments show that the PDE-Net has the potential to uncover the hidden PDE of the observed dynamics, and predict the dynamical behavior for a relatively long time, even in a noisy environment.

End-to-End Differentiable Physics for Learning and Control

This paper demonstrates how to perform backpropagation analytically through a physical simulator defined via a linear complementarity problem, and highlights the system's ability to learn physical parameters from data, efficiently match and simulate observed visual behavior, and readily enable control via gradient-based planning methods.

DGM: A deep learning algorithm for solving partial differential equations

Learning to Simulate Complex Physics with Graph Networks

A machine learning framework and model implementation that can learn to simulate a wide variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another, and holds promise for solving a wide range of complex forward and inverse problems.

A Machine Learning Strategy to Assist Turbulence Model Development

This work investigates the feasibility of a new data-driven approach to turbulence model development by attempting to reproduce, through a machine learning methodology, the results obtained with the well-established Spalart-Allmaras model.

Hidden Fluid Mechanics: A Navier-Stokes Informed Deep Learning Framework for Assimilating Flow Visualization Data

HFM is a physics informed deep learning framework capable of encoding an important class of physical laws governing fluid motions, namely the Navier-Stokes equations, and can be used in physical and biomedical problems to extract valuable quantitative information for which direct measurements may not be possible.