Learning the nonlinear flux function of a hidden scalar conservation law from data

Qing Li; Steinar Evje; Qing Li; Steinar Evje

doi:10.3934/nhm.2023003

Networks and Heterogeneous Media

2023, Volume 18, Issue 1: 48-79. doi: 10.3934/nhm.2023003

Previous Article Next Article

Research article

Learning the nonlinear flux function of a hidden scalar conservation law from data

Qing Li ,
Steinar Evje ^,

University of Stavanger, Department of Energy and Petroleum, Group of Computational Engineering, Stavanger, Norway

Received: 08 July 2022 Revised: 24 August 2022 Accepted: 31 August 2022 Published: 20 October 2022

Nonlinear conservation laws are widely used in fluid mechanics, biology, physics, and chemical engineering. However, deriving such nonlinear conservation laws is a significant and challenging problem. A possible attractive approach is to extract conservation laws more directly from observation data by use of machine learning methods. We propose a framework that combines a symbolic multi-layer neural network and a discrete scheme to learn the nonlinear, unknown flux function $ f(u) $ of the scalar conservation law

$ \begin{equation} u_t + f(u)_x = 0 \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (*) \end{equation} $

with $ u $ as the main variable. This identification is based on using observation data $ u(x_j, t_i) $ on a spatial grid $ x_j, \, \, j = 1, \ldots, N_x $ at specified times $ t_i, \, \, i = 1, \ldots, N_{obs} $. A main challenge with Eq (*) is that the solution typically creates shocks, i.e., one or several jumps of the form $ (u_L, u_R) $ with $ u_L \neq u_R $ moving in space and possibly changing over time such that information about $ f(u) $ in the interval associated with this jump is sparse or not at all present in the observation data. Secondly, the lack of regularity in the solution of (*) and the nonlinear form of $ f(u) $ hamper use of previous proposed physics informed neural network (PINN) methods where the underlying form of the sought differential equation is accounted for in the loss function. We circumvent this obstacle by approximating the unknown conservation law (*) by an entropy satisfying discrete scheme where $ f(u) $ is represented through a symbolic multi-layer neural network. Numerical experiments show that the proposed method has the ability to uncover the hidden conservation law for a wide variety of different nonlinear flux functions, ranging from pure concave/convex to highly non-convex shapes. This is achieved by relying on a relatively sparse amount of observation data obtained in combination with a selection of different initial data.
- nonlinear conservation law,
- flux function,
- machine learning method,
- discrete scheme,
- entropy condition
Citation: Qing Li, Steinar Evje. Learning the nonlinear flux function of a hidden scalar conservation law from data[J]. Networks and Heterogeneous Media, 2023, 18(1): 48-79. doi: 10.3934/nhm.2023003

Related Papers:

Abstract

Nonlinear conservation laws are widely used in fluid mechanics, biology, physics, and chemical engineering. However, deriving such nonlinear conservation laws is a significant and challenging problem. A possible attractive approach is to extract conservation laws more directly from observation data by use of machine learning methods. We propose a framework that combines a symbolic multi-layer neural network and a discrete scheme to learn the nonlinear, unknown flux function $ f(u) $ of the scalar conservation law

$ \begin{equation} u_t + f(u)_x = 0 \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (*) \end{equation} $

with $ u $ as the main variable. This identification is based on using observation data $ u(x_j, t_i) $ on a spatial grid $ x_j, \, \, j = 1, \ldots, N_x $ at specified times $ t_i, \, \, i = 1, \ldots, N_{obs} $. A main challenge with Eq (*) is that the solution typically creates shocks, i.e., one or several jumps of the form $ (u_L, u_R) $ with $ u_L \neq u_R $ moving in space and possibly changing over time such that information about $ f(u) $ in the interval associated with this jump is sparse or not at all present in the observation data. Secondly, the lack of regularity in the solution of (*) and the nonlinear form of $ f(u) $ hamper use of previous proposed physics informed neural network (PINN) methods where the underlying form of the sought differential equation is accounted for in the loss function. We circumvent this obstacle by approximating the unknown conservation law (*) by an entropy satisfying discrete scheme where $ f(u) $ is represented through a symbolic multi-layer neural network. Numerical experiments show that the proposed method has the ability to uncover the hidden conservation law for a wide variety of different nonlinear flux functions, ranging from pure concave/convex to highly non-convex shapes. This is achieved by relying on a relatively sparse amount of observation data obtained in combination with a selection of different initial data.

References

[1]	J. Bongard, H. Lipson, Automated reverse engineering of nonlinear dynamical systems. Proc. Natl. Acad. Sci., 104 (2007), 9943–9948. https://doi.org/10.1073/pnas.0609476104
[2]	M. Schmidt, H. Lipson, Distilling free-form natural laws from experimental data, Science, 324 (2009), 81–85. https://doi.org/10.1126/science.1165893
[3]	H. Owhadi, Bayesian numerical homogenization, Multiscale. Model. Sim., 13 (2015), 812–828. https://doi.org/10.1137/140974596 doi: 10.1137/140974596
[4]	M. Raissi, P. Perdikaris,, G.E. Karniadakis, Inferring solutions of differential equations using noisy multi-fidelity data, J. Comput. Phys., 335 (2017), 736–746. https://doi.org/10.1016/j.jcp.2017.07.050
[5]	M. Raissi, P. Perdikaris, G.E. Karniadakis, Machine learning of linear differential equations using Gaussian processes, J. Comput. Phys., 348 (2017), 683–693. https://doi.org/10.1016/j.jcp.2017.07.050
[6]	C.E. Rasmussen, C.K. Williams, Gaussian processes for machine learning, Cambridge: MIT press, 2006.
[7]	S.L. Brunton, J.L. Proctor, J.N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proc. Natl. Acad. Sci., 113 (2016), 3932–3937. https://doi.org/10.1073/pnas.1517384113 doi: 10.1073/pnas.1517384113
[8]	H Schaeffer, Learning partial differential equations via data discovery and sparse optimization, Proc. Math. Phys. Eng. Sci, 473 (2017), 20160446. https://doi.org/10.1098/rspa.2016.0446 doi: 10.1098/rspa.2016.0446
[9]	S.H. Rudy, S.L. Brunton, J.L. Proctor, J.N. Kutz, Data-driven discovery of partial differential equations, Sci. Adv., 3 (2017), e1602614. https://doi.org/10.1126/sciadv.1602614 doi: 10.1126/sciadv.1602614
[10]	Z. Wu, R. Zhang, Learning physics by data for the motion of a sphere falling in a non-Newtonian fluid, Commun Nonlinear Sci Numer Simul, 67 (2019), 577–593. https://doi.org/10.1016/j.cnsns.2018.05.007 doi: 10.1016/j.cnsns.2018.05.007
[11]	M. Raissi, P. Perdikaris, G.E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, J. Comput. Phys., 378 (2019), 686–707. https://doi.org/10.1016/j.jcp.2018.10.045 doi: 10.1016/j.jcp.2018.10.045
[12]	O Fuks, H.A. Tchelepi, Limitations of physics informed machine learning for nonlinear two-phase transport in porous media, J. Mach. Learn. Model. Comput., 1 (2020), 19–37. https://doi.org/10.1615/JMachLearnModelComput.2020033905 doi: 10.1615/JMachLearnModelComput.2020033905
[13]	Z. Long, Y. Lu, X. Ma, B. Dong, PDE-net: Learning PDEs from data, Proceedings of the 35th International Conference on Machine Learning, 80 (2018), 3208–3216.
[14]	Z. Long, Y. Lu, B. Dong, PDE-Net 2.0: Learning PDEs from data with a numeric-symbolic hybrid deep network, J. Comput. Phys., 399 (2019), 108925. https://doi.org/10.1016/j.jcp.2019.108925 doi: 10.1016/j.jcp.2019.108925
[15]	R.J. LeVeque, Finite volume methods for hyperbolic problems, Cambridge: Cambridge Texts in Applied Mathematics, 2007.
[16]	H. Holden, N.H. Risebro, Front tracking for hyperbolic conservation laws, Berlin: Springer, 2011.
[17]	G. Martius, C.H. Lampert, Extrapolation and learning equations, arXiv: 1610.02995, [Preprint], (2016) [cited 2022 Oct 18]. Available form: https://arXiv.53yu.com/abs/1610.02995
[18]	S. Sahoo, C. Lampert, G. Martius, Learning equations for extrapolation and control, Proceedings of the 35th International Conference on Machine Learning, 80 (2018), 4442–4450.
[19]	F. James, M. Sepúlveda, Convergence results for the flux identification in a scalar conservation law, SIAM J. Control. Optim., 37 (1999), 869–891. https://doi.org/10.1137/S0363012996272722 doi: 10.1137/S0363012996272722
[20]	H. Holden, F.S. Priuli, N.H. Risebro, On an inverse problem for scalar conservation laws, Inverse Probl, 30 (2014), 035015. https://doi.org/10.1088/0266-5611/30/3/035015 doi: 10.1088/0266-5611/30/3/035015
[21]	M. C. Bustos, F. Concha, R. Bürger, E.M. Tory, Sedimentation and thickening–Phenomenological Foundation and Mathematical Theory. Dordrecht: Kluwer Academic Publishers, 1999.
[22]	S. Diehl, Estimation of the batch-settling flux function for an ideal suspension from only two experiments, Chem. Eng. Sci., 62 (2007), 4589–4601. https://doi.org/10.1016/j.ces.2007.05.025 doi: 10.1016/j.ces.2007.05.025
[23]	R. Bürger, S. Diehl, Convexity-preserving flux identification for scalar conservation laws modelling sedimentation, Inverse Probl, 29 (2013), 045008. https://doi.org/10.1088/0266-5611/29/4/045008 doi: 10.1088/0266-5611/29/4/045008
[24]	R. Bürger, J. Careaga, S. Diehl, Flux identification of scalar conservation laws from sedimentation in a cone, IMA J Appl Math, 83 (2018), 526–552. https://doi.org/10.1093/imamat/hxy018 doi: 10.1093/imamat/hxy018
[25]	S. Diehl, Numerical identification of constitutive functions in scalar nonlinear convection–diffusion equations with application to batch sedimentation, Appl Numer Math, 95 (2015), 154–172. https://doi.org/10.1016/j.apnum.2014.04.002 doi: 10.1016/j.apnum.2014.04.002
[26]	M. Mishra, Machine learning framework for data driven acceleration of computations of differential equations, Math. eng., 1 (2018), 118–146. https://doi.org/10.3934/Mine.2018.1.118 doi: 10.3934/Mine.2018.1.118
[27]	J.W. Thomas, Numerical partial differential equations–Conservation laws and elliptic equations, Texts in Applied Mathematics, New York: Springer, 1999.
[28]	J.S. Hesthaven, Numerical methods for conservation laws. from analysis to algorithms, Philadelphia: Society for Industrial and Applied Mathematics, 2017.
[29]	D. Kröener, Numerical schemes for conservation laws, New York: John Wiley & Sons, 1997.
[30]	M. Mishra, U.S. Fjordholm, R. Abgrall, Numerical methods for conservation laws and related equations. Lecture notes for Numerical Methods for Partial Differential Equations 57 (2019), 58.
[31]	Valerii Iakovlev, Markus Heinonen, Harri Lähdesmäki, Learning continuous-time pdes from sparse data with graph neural networks, arXiv: 2006.08956, [Preprint], (2020) [cited 2022 Oct 18]. Available form: https://arXiv.53yu.com/abs/2006.08956
[32]	C. Zhu, R.H. Byrd, P. Lu, J. Nocedal, Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization, Acm T math software, 23 (1997), 550–560. https://doi.org/10.1145/279232.279236 doi: 10.1145/279232.279236
[33]	Sebastian R, An overview of gradient descent optimization algorithms, arXiv: 1609.04747, [Preprint], (2016) [cited 2022 Oct 18]. Available form: https://arXiv.org/abs/1609.04747
[34]	H.J. Skadsem, S. Kragset, A numerical study of density-unstable reverse circulation displacement for primary cementing, J. Energy. Resour. Technol., 144 (2022), 123008. https://doi.org/10.1115/1.4054367 doi: 10.1115/1.4054367

Reader Comments

Your name:*

Email:*
© 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)