Deep learning models for global coordinate transformations that linearise PDEs

DOI10.1017/S0956792520000327zbMATH Open1479.35021arXiv1911.02710OpenAlexW3088290697WikidataQ114116642 ScholiaQ114116642MaRDI QIDQ5014841FDOQ5014841

J. N. Kutz, S. L. Brunton, Bethany Lusch, Craig Gin

Publication date: 8 December 2021

Published in: European Journal of Applied Mathematics (Search for Journal in Brave)

Abstract: We develop a deep autoencoder architecture that can be used to find a coordinate transformation which turns a nonlinear PDE into a linear PDE. Our architecture is motivated by the linearizing transformations provided by the Cole-Hopf transform for Burgers equation and the inverse scattering transform for completely integrable PDEs. By leveraging a residual network architecture, a near-identity transformation can be exploited to encode intrinsic coordinates in which the dynamics are linear. The resulting dynamics are given by a Koopman operator matrix $mathbfK$. The decoder allows us to transform back to the original coordinates as well. Multiple time step prediction can be performed by repeated multiplication by the matrix $mathbfK$ in the intrinsic coordinates. We demonstrate our method on a number of examples, including the heat equation and Burgers equation, as well as the substantially more challenging Kuramoto-Sivashinsky equation, showing that our method provides a robust architecture for discovering interpretable, linearizing transforms for nonlinear PDEs.

Full work available at URL: https://arxiv.org/abs/1911.02710

Recommendations

• Deep hidden physics models: deep learning of nonlinear partial differential equations
• Data-driven discovery of coordinates and governing equations
• Data-driven discovery of PDEs in complex datasets
• PDE-Net 2.0: learning PDEs from data with a numeric-symbolic hybrid deep network
• Linearly recurrent autoencoder networks for learning dynamics

zbMATH Keywords

Cole-Hopf transformdeep neural networksKoopman theoryresidual networksautoencoder architecturelinearising transforms

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Transform methods (e.g., integral transforms) applied to PDEs (35A22) Theoretical approximation in context of PDEs (35A35)

Cites Work

• Pattern recognition and machine learning.
• Title not available (Why is that?)
• Dynamic mode decomposition of numerical and experimental data
• Spectral analysis of nonlinear flows
• Analysis of Fluid Flows via Spectral Properties of the Koopman Operator
• Fourth-Order Time-Stepping for Stiff PDEs
• On a quasi-linear parabolic equation occurring in aerodynamics
• On some dissipative fully discrete nonlinear Galerkin schemes for the Kuramoto-Sivashinsky equation
• Title not available (Why is that?)
• The partial differential equation u_t + uu_x = μ_xx
• Introduction to Applied Nonlinear Dynamical Systems and Chaos
• Approximation by superpositions of a sigmoidal function
• Comparison of systems with complex behavior
• Spectral properties of dynamical systems, model reduction and decompositions
• A kernel-based method for data-driven Koopman spectral analysis
• Geometry of the ergodic quotient reveals coherent structures in flows
• Title not available (Why is that?)
• A data-driven approximation of the koopman operator: extending dynamic mode decomposition
• A Variational Approach to Modeling Slow Processes in Stochastic Dynamical Systems
• Title not available (Why is that?)
• Hamiltonian Systems and Transformation in Hilbert Space
• Title not available (Why is that?)
• The Method of Near-Identity Transformations and Its Applications
• Data-driven model reduction and transfer operator approximation
• Data-Driven Science and Engineering
• Extended dynamic mode decomposition with dictionary learning: A data-driven adaptive spectral decomposition of the Koopman operator
• Applied Koopman theory for partial differential equations and data-driven modeling of spatio-temporal systems
• Parsimonious representation of nonlinear dynamical systems through manifold learning: a chemotaxis case study
• Physics-Informed Probabilistic Learning of Linear Embeddings of Nonlinear Dynamics with Guaranteed Stability
• Data-driven discovery of coordinates and governing equations
• Linearly Recurrent Autoencoder Networks for Learning Dynamics

Cited In (8)

• Parsimony as the ultimate regularizer for physics-informed machine learning
• Connections between deep learning and partial differential equations
• Koopman neural operator as a mesh-free solver of non-linear partial differential equations
• \(\Phi\)-DVAE: physics-informed dynamical variational autoencoders for unstructured data assimilation
• Machine learning methods for reduced order modeling
• Modern Koopman Theory for Dynamical Systems
• Semi-supervised invertible neural operators for Bayesian inverse problems
• Bridging the gap: machine learning to resolve improperly modeled dynamics

Uses Software

• VAMPnets
• PRMLT

This page was built for publication: Deep learning models for global coordinate transformations that linearise PDEs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5014841)