OnsagerNet: Learning Stable and Interpretable Dynamics using a Generalized Onsager Principle

Abstract

We propose a systematic method for learning stable and physically interpretable dynamical models using sampled trajectory data from physical processes based on a generalized Onsager principle. The learned dynamics are autonomous ordinary differential equations parameterized by neural networks that retain clear physical structure information, such as free energy, diffusion, conservative motion and external forces. For high dimensional problems with a low dimensional slow manifold, an autoencoder with metric preserving regularization is introduced to find the low dimensional generalized coordinates on which we learn the generalized Onsager dynamics. Our method exhibits clear advantages over existing methods on benchmark problems for learning ordinary differential equations. We further apply this method to study Rayleigh-Benard convection and learn Lorenz-like low dimensional autonomous reduced order models that capture both qualitative and quantitative properties of the underlying dynamics. This forms a general approach to building reduced order models for forced dissipative systems.

arXiv comment29 pages, 19 figures

References in corpus (26)

• Discovering governing equations from data: Sparse identification of nonlinear dynamical systems
• On the Convergence of Adam and Beyond
• Data-driven discovery of coordinates and governing equations
• Active Learning of Uniformly Accurate Inter-atomic Potentials for Materials Simulation
• Extended dynamic mode decomposition with dictionary learning: a data-driven adaptive spectral decomposition of the Koopman operator
• A variational approach to the moving contact line hydrodynamics
• Data-assisted reduced-order modeling of extreme events in complex dynamical systems
• Multistep Neural Networks for Data-driven Discovery of Nonlinear Dynamical Systems
• Symplectic ODE-Net: Learning Hamiltonian Dynamics with Control
• Deep learning to discover and predict dynamics on an inertial manifold
• Uniformly Accurate Machine Learning Based Hydrodynamic Models for Kinetic Equations
• Structure-preserving neural networks
• Observing the Sun with Atacama Large Millimeter/submillimeter Array (ALMA): High Resolution Interferometric Imaging
• Data-driven Discovery of Closure Models
• Tracing Electron Beams in the Sun's Corona with Radio Dynamic Imaging Spectroscopy
• From microscopic theory to macroscopic theory: a systematic study on static modeling for liquid crystals
• Microwave Quasi-periodic Pulsations in Multi-timescales Associated with a Solar Flare/CME Event
• Combining Machine Learning with Knowledge-Based Modeling for Scalable Forecasting and Subgrid-Scale Closure of Large, Complex, Spatiotemporal Systems
• Microwave Zebra Pattern Structures in the X2.2 Solar Flare on Feb 15, 2011
• Better Approximations of High Dimensional Smooth Functions by Deep Neural Networks with Rectified Power Units
• Dissipative SymODEN: Encoding Hamiltonian Dynamics with Dissipation and Control into Deep Learning
• Radio Spectroscopic Imaging of a Solar Flare Termination Shock: Split-Band Feature as Evidence for Shock Compression
• Application of Onsager Machlup integral in solving dynamic equations in non-equilibrium systems
• Coronal magnetic field and the plasma beta determined from radio and multiple satellite observations
• GPU-Based High-Performance Imaging for Mingantu Spectral RadioHeliograph
• Generalized Lorentz reciprocal theorem in complex fluids and in non-isothermal systems

Cited by in corpus (12)

• Automated discovery of generalized standard material models with EUCLID
• GFINNs: GENERIC Formalism Informed Neural Networks for Deterministic and Stochastic Dynamical Systems
• ChebNet: Efficient and Stable Constructions of Deep Neural Networks with Rectified Power Units via Chebyshev Approximations
• Learning Thermodynamically Stable and Galilean Invariant Partial Differential Equations for Non-equilibrium Flows
• Structure-preserving Sparse Identification of Nonlinear Dynamics for Data-driven Modeling
• DeePN${}^2$: A deep learning-based non-Newtonian hydrodynamic model
• A Data Driven Method for Computing Quasipotentials
• Inverse modified differential equations for discovery of dynamics
• Efficient Spectral Methods for Quasi-Equilibrium Closure Approximations of Symmetric Problems on Unit Circle and Sphere
• Discovery of Governing Equations with Recursive Deep Neural Networks
• Learning Dynamics Models with Stable Invariant Sets
• Multiscale and Nonlocal Learning for PDEs using Densely Connected RNNs