Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Understanding latent timescales in neural ordinary differential equation models of advection-dominated dynamical systems

Journal Article · · Physica. D, Nonlinear Phenomena
 [1];  [2];  [2];  [3];  [2];  [4]
  1. Argonne National Laboratory (ANL), Argonne, IL (United States); University of Notre Dame, IN (United States)
  2. Argonne National Laboratory (ANL), Argonne, IL (United States)
  3. University of Notre Dame, IN (United States)
  4. Argonne National Laboratory (ANL), Argonne, IL (United States); Pennsylvania State Univ., University Park, PA (United States)

The neural ordinary differential equation (ODE) framework has shown considerable promise in recent years in developing highly accelerated surrogate models for complex physical systems characterized by partial differential equations (PDEs). For PDE-based systems, state-of-the-art neural ODE strategies leverage a two-step procedure to achieve this acceleration: a nonlinear dimensionality reduction step provided by an autoencoder, and a time integration step provided by a neural-network based model for the resultant latent space dynamics (the neural ODE). This work explores the applicability of such autoencoder-based neural ODE strategies for PDEs in which advection terms play a critical role. More specifically, alongside predictive demonstrations, physical insight into the sources of model acceleration (i.e., how the neural ODE achieves its acceleration) is the scope of the current study. Such investigations are performed by quantifying the effects of both autoencoder and neural ODE components on latent system time-scales using eigenvalue analysis of dynamical system Jacobians. To this end, the sensitivity of various critical training parameters – de-coupled versus end-to-end training, latent space dimensionality, and the role of training trajectory length, for example – to both model accuracy and the discovered latent system timescales is quantified. Furthermore, this work specifically uncovers the key role played by the training trajectory length (the number of rollout steps in the loss function during training) on the latent system timescales: larger trajectory lengths correlate with an increase in limiting neural ODE time-scales, and optimal neural ODEs are found to recover the largest time-scales of the full-order (ground-truth) system. Demonstrations are performed across fundamentally different unsteady fluid dynamics configurations influenced by advection: (1) the Kuramoto–Sivashinsky equations (2) Hydrogen-Air channel detonations (the compressible reacting Navier–Stokes equations with detailed chemistry), and (3) 2D Atmospheric flow.

Research Organization:
Argonne National Laboratory (ANL), Argonne, IL (United States)
Sponsoring Organization:
USDOE Office of Science (SC), Basic Energy Sciences (BES). Scientific User Facilities (SUF); USDOE Laboratory Directed Research and Development (LDRD) Program; USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR)
Grant/Contract Number:
AC02-06CH11357
OSTI ID:
3004189
Journal Information:
Physica. D, Nonlinear Phenomena, Journal Name: Physica. D, Nonlinear Phenomena Vol. 476; ISSN 0167-2789
Publisher:
ElsevierCopyright Statement
Country of Publication:
United States
Language:
English

References (61)

Flow reactor studies and kinetic modeling of the H2/O2 reaction journal January 1999
A method for interpolating on manifolds structural dynamics reduced-order models
  • Amsallem, David; Cortial, Julien; Carlberg, Kevin
  • International Journal for Numerical Methods in Engineering, Vol. 80, Issue 9 https://doi.org/10.1002/nme.2681
journal November 2009
Explicit Time-Scale Splitting Algorithm for Stiff Problems: Auto-ignition of Gaseous Mixtures behind a Steady Shock journal May 2001
Atmospheric simulations using a GCM with simplified physical parametrizations. I: model climatology and variability in multi-decadal experiments journal January 2003
Decadal interactions between the western tropical Pacific and the North Atlantic Oscillation journal November 2005
A Comprehensive Deep Learning-Based Approach to Reduced Order Modeling of Nonlinear Time-Dependent Parametrized PDEs journal April 2021
Inertial manifolds for nonlinear evolutionary equations journal June 1988
Deep residual U-net convolution neural networks with autoregressive strategy for fluid flow predictions in large-scale geosystems journal April 2021
Reduced-order fluid/structure modeling of a complete aircraft configuration journal August 2006
Multi-level convolutional autoencoder networks for parametric prediction of spatio-temporal dynamics journal December 2020
A comparison of neural network architectures for data-driven reduced-order modeling journal April 2022
Robust control of uncertain cylinder wake flows based on robust reduced order models journal June 2009
Highly-scalable GPU-accelerated compressible reacting flow solver for modeling high-speed flows journal October 2023
Non-intrusive reduced order modeling of nonlinear problems using neural networks journal June 2018
Model reduction of dynamical systems on nonlinear manifolds using deep convolutional autoencoders journal November 2019
Data-driven POD-Galerkin reduced order model for turbulent flows journal September 2020
Neural-network learning of SPOD latent dynamics journal November 2022
Stabilized neural ordinary differential equations for long-time forecasting of dynamical systems journal February 2023
Multiscale graph neural network autoencoders for interpretable scientific machine learning journal December 2023
Reduced-order modeling: new approaches for computational physics journal February 2004
Towards the use of large eddy simulation in engineering journal August 2008
Time-series learning of latent-space dynamics for reduced-order model closure journal April 2020
Detonation in gases journal January 2009
Emerging trends in numerical simulations of combustion systems journal January 2019
Stochastic estimation of organized turbulent structure: homogeneous shear flow journal May 1988
Optimal LES formulations for isotropic turbulence journal November 1999
Cluster-based reduced-order modelling of a mixing layer journal August 2014
Deep learning in fluid dynamics journal January 2017
Spectral proper orthogonal decomposition and its relationship to dynamic mode decomposition and resolvent analysis journal May 2018
Stiff-PINN: Physics-Informed Neural Network for Stiff Chemical Kinetics journal August 2021
Deep learning for universal linear embeddings of nonlinear dynamics journal November 2018
Learning continuous models for continuous physics journal November 2023
Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators journal March 2021
A dynamic subgrid‐scale eddy viscosity model journal July 1991
Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoencoders journal March 2021
An approximate inertial manifold (AIM) based closure for turbulent flows journal July 2022
Learning data-driven discretizations for partial differential equations journal July 2019
Machine learning–accelerated computational fluid dynamics journal May 2021
Pulsating one-dimensional detonations in hydrogen–air mixtures journal December 2004
Parameterized neural ordinary differential equations: applications to computational physics problems journal September 2021
Data-driven reduction and decomposition with time-axis clustering journal June 2023
Exascale applications: skin in the game
  • Alexander, Francis; Almgren, Ann; Bell, John
  • Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, Vol. 378, Issue 2166 https://doi.org/10.1098/rsta.2019.0056
journal January 2020
Non-Markovian closure models for large eddy simulations using the Mori-Zwanzig formalism journal January 2017
Deep learning for in situ data compression of large turbulent flow simulations journal November 2020
Runge-Kutta neural network for identification of dynamical systems in high accuracy journal March 1998
Centroidal Voronoi Tessellation-Based Reduced-Order Modeling of Complex Systems journal January 2006
On the Choice of Wavespeeds for the HLLC Riemann Solver journal November 1997
Turbulence Modeling in the Age of Data journal January 2019
Dynamic Mode Decomposition and Its Variants journal January 2022
Large-Eddy-Simulation Tools for Multiphase Flows journal January 2012
Two-Dimensional Turbulence journal January 2012
Nonidealities in Rotating Detonation Engines journal January 2023
The Proper Orthogonal Decomposition in the Analysis of Turbulent Flows journal January 1993
DIRECT NUMERICAL SIMULATION: A Tool in Turbulence Research journal January 1998
Large-Eddy Simulation of Turbulent Combustion journal January 2006
Non-intrusive reduced-order modeling for fluid problems: A brief review journal December 2019
AMReX: a framework for block-structured adaptive mesh refinement journal May 2019
Adaptation of Aeroelastic Reduced-Order Models and Application to an F-16 Configuration journal June 2007
Explicit Model Predictive Control for Large-Scale Systems via Model Reduction journal July 2008
Timescale Analysis for Nonlinear Dynamical Systems journal March 2003
New Approaches in Turbulence and Transition Modeling Using Data-driven Techniques conference January 2015

Similar Records

Divide and conquer: Learning chaotic dynamical systems with multistep penalty neural ordinary differential equations
Journal Article · Mon Oct 14 00:00:00 EDT 2024 · Computer Methods in Applied Mechanics and Engineering · OSTI ID:3003857

$\text{GPLaSDI}$: Gaussian Process-based interpretable Latent Space Dynamics Identification through deep autoencoder
Journal Article · Sat Oct 14 00:00:00 EDT 2023 · Computer Methods in Applied Mechanics and Engineering · OSTI ID:2345978