Solving differential equations using deep neural networks
Abstract
Recent work on solving partial differential equations (PDEs) with deep neural networks (DNNs) is presented. The paper reviews and extends some of these methods while carefully analyzing a fundamental feature in numerical PDEs and nonlinear analysis: irregular solutions. First, the Sod shock tube solution to the compressible Euler equations is discussed and analyzed. This analysis includes a comparison of a DNN-based approach with conventional finite element and finite volume methods, and demonstrates that the DNN is competitive in terms of degrees of freedom required for a given accuracy. Further, the DNN-based approach is extended to consider performance improvements and simultaneous parameter space exploration. Next, a shock solution to compressible magnetohydrodynamics (MHD) is solved for, and used in a scenario where experimental data is utilized to enhance a PDE system that is a priori insufficient to validate against the observed/experimental data. This is accomplished by enriching the model PDE system with source terms that are then inferred via supervised training with synthetic experimental data. The resulting DNN framework for PDEs enables straightforward system prototyping and natural integration of large data sets (be they synthetic or experimental), all while simultaneously enabling single-pass exploration of an entire parameter space.
- Authors:
-
- Univ. of Texas, Austin, TX (United States)
- Publication Date:
- Research Org.:
- Univ. of Texas, Austin, TX (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC), Fusion Energy Sciences (FES); National Science Foundation (NSF)
- OSTI Identifier:
- 1785492
- Alternate Identifier(s):
- OSTI ID: 1630182
- Grant/Contract Number:
- FG02-04ER54742; SC0018429; AST-1413501
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Neurocomputing
- Additional Journal Information:
- Journal Volume: 399; Journal ID: ISSN 0925-2312
- Publisher:
- Elsevier
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 70 PLASMA PHYSICS AND FUSION TECHNOLOGY; Deep neural networks; Differential equations; Partial differential equations; Nonlinear; Shocks; Data analytics; Optimization
Citation Formats
Michoski, Craig, Milosavljević, Miloš, Oliver, Todd, and Hatch, David R. Solving differential equations using deep neural networks. United States: N. p., 2020.
Web. doi:10.1016/j.neucom.2020.02.015.
Michoski, Craig, Milosavljević, Miloš, Oliver, Todd, & Hatch, David R. Solving differential equations using deep neural networks. United States. https://doi.org/10.1016/j.neucom.2020.02.015
Michoski, Craig, Milosavljević, Miloš, Oliver, Todd, and Hatch, David R. Wed .
"Solving differential equations using deep neural networks". United States. https://doi.org/10.1016/j.neucom.2020.02.015. https://www.osti.gov/servlets/purl/1785492.
@article{osti_1785492,
title = {Solving differential equations using deep neural networks},
author = {Michoski, Craig and Milosavljević, Miloš and Oliver, Todd and Hatch, David R.},
abstractNote = {Recent work on solving partial differential equations (PDEs) with deep neural networks (DNNs) is presented. The paper reviews and extends some of these methods while carefully analyzing a fundamental feature in numerical PDEs and nonlinear analysis: irregular solutions. First, the Sod shock tube solution to the compressible Euler equations is discussed and analyzed. This analysis includes a comparison of a DNN-based approach with conventional finite element and finite volume methods, and demonstrates that the DNN is competitive in terms of degrees of freedom required for a given accuracy. Further, the DNN-based approach is extended to consider performance improvements and simultaneous parameter space exploration. Next, a shock solution to compressible magnetohydrodynamics (MHD) is solved for, and used in a scenario where experimental data is utilized to enhance a PDE system that is a priori insufficient to validate against the observed/experimental data. This is accomplished by enriching the model PDE system with source terms that are then inferred via supervised training with synthetic experimental data. The resulting DNN framework for PDEs enables straightforward system prototyping and natural integration of large data sets (be they synthetic or experimental), all while simultaneously enabling single-pass exploration of an entire parameter space.},
doi = {10.1016/j.neucom.2020.02.015},
journal = {Neurocomputing},
number = ,
volume = 399,
place = {United States},
year = {Wed Feb 19 00:00:00 EST 2020},
month = {Wed Feb 19 00:00:00 EST 2020}
}
Web of Science
Works referenced in this record:
Local bounds preserving stabilization for continuous Galerkin discretization of hyperbolic systems
journal, May 2018
- Mabuza, Sibusiso; Shadid, John N.; Kuzmin, Dmitri
- Journal of Computational Physics, Vol. 361
Data-driven discovery of coordinates and governing equations
journal, October 2019
- Champion, Kathleen; Lusch, Bethany; Kutz, J. Nathan
- Proceedings of the National Academy of Sciences, Vol. 116, Issue 45
Adaptive hierarchic transformations for dynamically p-enriched slope-limiting over discontinuous Galerkin systems of generalized equations
journal, September 2011
- Michoski, C.; Mirabito, C.; Dawson, C.
- Journal of Computational Physics, Vol. 230, Issue 22
Error estimates for the moving least-square approximation and the element-free Galerkin method in n-dimensional spaces
journal, January 2016
- Li, Xiaolin
- Applied Numerical Mathematics, Vol. 99
Local adaptive mesh refinement for shock hydrodynamics
journal, May 1989
- Berger, M. J.; Colella, P.
- Journal of Computational Physics, Vol. 82, Issue 1
Survey of Multifidelity Methods in Uncertainty Propagation, Inference, and Optimization
journal, January 2018
- Peherstorfer, Benjamin; Willcox, Karen; Gunzburger, Max
- SIAM Review, Vol. 60, Issue 3
Hierarchical reconstruction for spectral volume method on unstructured grids
journal, September 2009
- Xu, Zhiliang; Liu, Yingjie; Shu, Chi-Wang
- Journal of Computational Physics, Vol. 228, Issue 16
Discrete equations for physical and numerical compressible multiphase mixtures
journal, April 2003
- Abgrall, Rémi; Saurel, Richard
- Journal of Computational Physics, Vol. 186, Issue 2
Physics-Driven Regularization of Deep Neural Networks for Enhanced Engineering Design and Analysis
journal, September 2019
- Nabian, Mohammad Amin; Meidani, Hadi
- Journal of Computing and Information Science in Engineering, Vol. 20, Issue 1
Hydrodynamic instabilities in supernova remnants - Self-similar driven waves
journal, June 1992
- Chevalier, Roger A.; Blondin, John M.; Emmering, Robert T.
- The Astrophysical Journal, Vol. 392
Performance Comparison of HPX Versus Traditional Parallelization Strategies for the Discontinuous Galerkin Method
journal, May 2019
- Bremer, Maximilian; Kazhyken, Kazbek; Kaiser, Hartmut
- Journal of Scientific Computing, Vol. 80, Issue 2
Management of discontinuous reconstruction in kinetic schemes
journal, June 2004
- Ohwada, Taku; Kobayashi, Seijiro
- Journal of Computational Physics, Vol. 197, Issue 1
Artificial neural networks for solving ordinary and partial differential equations
journal, January 1998
- Lagaris, I. E.; Likas, A.; Fotiadis, D. I.
- IEEE Transactions on Neural Networks, Vol. 9, Issue 5
Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
journal, February 2019
- Raissi, M.; Perdikaris, P.; Karniadakis, G. E.
- Journal of Computational Physics, Vol. 378
Optimal regularity for the Poisson equation
journal, January 2009
- Wang, Lihe; Yao, Fengping; Zhou, Shulin
- Proceedings of the American Mathematical Society, Vol. 137, Issue 06
A Centre Manifold Description of Contaminant Dispersion in Channels with Varying Flow Properties
journal, December 1990
- Mercer, G. N.; Roberts, A. J.
- SIAM Journal on Applied Mathematics, Vol. 50, Issue 6
Foundations of the blended isogeometric discontinuous Galerkin (BIDG) method
journal, June 2016
- Michoski, C.; Chan, J.; Engvall, L.
- Computer Methods in Applied Mechanics and Engineering, Vol. 305
Validating predictions of unobserved quantities
journal, January 2015
- Oliver, Todd A.; Terejanu, Gabriel; Simmons, Christopher S.
- Computer Methods in Applied Mechanics and Engineering, Vol. 283
Parametrized maximum principle preserving flux limiters for high order schemes solving hyperbolic conservation laws: one-dimensional scalar problem
journal, December 2013
- Xu, Zhengfu
- Mathematics of Computation, Vol. 83, Issue 289
Data-driven discovery of partial differential equations
journal, April 2017
- Rudy, Samuel H.; Brunton, Steven L.; Proctor, Joshua L.
- Science Advances, Vol. 3, Issue 4
Solving high-dimensional partial differential equations using deep learning
journal, August 2018
- Han, Jiequn; Jentzen, Arnulf; E., Weinan
- Proceedings of the National Academy of Sciences, Vol. 115, Issue 34
Data-driven discovery of PDEs in complex datasets
journal, May 2019
- Berg, Jens; Nyström, Kaj
- Journal of Computational Physics, Vol. 384
AD Model Builder: using automatic differentiation for statistical inference of highly parameterized complex nonlinear models
journal, April 2012
- Fournier, David A.; Skaug, Hans J.; Ancheta, Johnoel
- Optimization Methods and Software, Vol. 27, Issue 2
A Comparison of Artificial Viscosity, Limiters, and Filters, for High Order Discontinuous Galerkin Solutions in Nonlinear Settings
journal, April 2015
- Michoski, C.; Dawson, C.; Kubatko, E. J.
- Journal of Scientific Computing, Vol. 66, Issue 1
Theory of shock wave propagation during laser ablation
journal, June 2004
- Zhang, Zhaoyan; Gogos, George
- Physical Review B, Vol. 69, Issue 23
DGM: A deep learning algorithm for solving partial differential equations
journal, December 2018
- Sirignano, Justin; Spiliopoulos, Konstantinos
- Journal of Computational Physics, Vol. 375
A divergence-free semi-implicit finite volume scheme for ideal, viscous, and resistive magnetohydrodynamics: A divergence-free semi-implicit finite volume scheme for ideal, viscous and resistive magnetohydrodynamics
journal, September 2018
- Dumbser, M.; Balsara, D. S.; Tavelli, M.
- International Journal for Numerical Methods in Fluids, Vol. 89, Issue 1-2
Wave Propagation Algorithms for Multidimensional Hyperbolic Systems
journal, March 1997
- LeVeque, Randall J.
- Journal of Computational Physics, Vol. 131, Issue 2
Survey of meshless and generalized finite element methods: A unified approach
journal, May 2003
- Babuška, Ivo; Banerjee, Uday; Osborn, John E.
- Acta Numerica, Vol. 12
Novel approach to nonlinear/non-Gaussian Bayesian state estimation
journal, January 1993
- Gordon, N. J.; Salmond, D. J.; Smith, A. F. M.
- IEE Proceedings F Radar and Signal Processing, Vol. 140, Issue 2
Element-free Galerkin methods
journal, January 1994
- Belytschko, T.; Lu, Y. Y.; Gu, L.
- International Journal for Numerical Methods in Engineering, Vol. 37, Issue 2
Optimization by Simulated Annealing
journal, May 1983
- Kirkpatrick, S.; Gelatt, C. D.; Vecchi, M. P.
- Science, Vol. 220, Issue 4598
Robust Estimation of a Location Parameter
journal, March 1964
- Huber, Peter J.
- The Annals of Mathematical Statistics, Vol. 35, Issue 1
A survey of several finite difference methods for systems of nonlinear hyperbolic conservation laws
journal, April 1978
- Sod, Gary A.
- Journal of Computational Physics, Vol. 27, Issue 1
Spectral element-FCT method for scalar hyperbolic conservation laws
journal, March 1992
- Giannakouros, John; Karniadakis, George Em
- International Journal for Numerical Methods in Fluids, Vol. 14, Issue 6
Least-squares finite element methods for compressible Euler equations
journal, April 1990
- Jiang, Bo-Nan; Carey, G. F.
- International Journal for Numerical Methods in Fluids, Vol. 10, Issue 5
Learning partial differential equations via data discovery and sparse optimization
journal, January 2017
- Schaeffer, Hayden
- Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, Vol. 473, Issue 2197
Quadrature-free non-oscillatory finite volume schemes on unstructured meshes for nonlinear hyperbolic systems
journal, September 2007
- Dumbser, Michael; Käser, Martin; Titarev, Vladimir A.
- Journal of Computational Physics, Vol. 226, Issue 1
Works referencing / citing this record:
An overview on deep learning-based approximation methods for partial differential equations
text, January 2020
- Beck, Christian; Hutzenthaler, Martin; Jentzen, Arnulf
- arXiv
Using analog computers in today's largest computational challenges
journal, December 2021
- Köppel, Sven; Ulmann, Bernd; Heimann, Lars
- Advances in Radio Science, Vol. 19