Abstract
In this paper, we study the inverse problem of approximating a space-dependent wave source in a one-dimensional wave equation. The problem is converted to a nonclassical second-order hyperbolic equation and is later reduced to the solution of a linear system of algebraic equations by employing the Ritz collocation method. The obtained Ritz approximation is used to train a physics-informed neural network (PINN). However, unlike the conventional training in PINN, the loss function of the proposed method is not additive and the loss terms are decoupled in the sense that the neural network is re-trained separately according to each loss term. Hence, we name the proposed method decoupled PINN or D-PINN for short. Since the function of the initial condition is nonlinear and continuously differentiable, it is used, up to a scalar, as the activation function as opposed to using an off-the-shelf activation function such as hyperbolic tangent. Experiments indicate that D-PINN is capable of approximating the solution more accurately than Ritz and PINN.
Similar content being viewed by others
Availability of data and materials
The data is generated by a numerical method and it is available.
Notes
In this work we represent the transpose operator by notation tr.
References
Baydin AG, Pearlmutter BA, Radul AA, Siskind JM (2017) Automatic differentiation in machine learning: a survey. J Mach Learn Res 18(1):5595–5637
Bell WW (2004) Special functions for scientists and engineers. Dover Publications, New York
Blechschmidt J, Ernst OG (2021) Three ways to solve partial differential equations with neural networks: a review. GAMM Mitt 44(2):e202100006
Cannon JR, Dunninger DR (1970) Determination of an unknown forcing function in a hyperbolic equation from overspecified data. Ann Mat Pura Appl 1:49–62
Chen P, Liu, Aihara K, Chen L (2020) Autoreservoir computing for multistep ahead prediction based on the spatiotemporal information transformation. Nat Commun 11(1):1–15
Cuomo S, Di Cola VS, Giampaolo F, Rozza G, Raissi M, Piccialli F (2022)Scientific machine learning through physics-informed neural networks: where we are and what’s next. arXiv preprint arXiv:2201.05624
Engl HW, Scherzer O, Yamamoto M (1994) Uniqueness and stable determination of forcing terms in linear partial differential equations with overspecified boundary data. Inverse Probl 10:1253–1276
Gao H, Sun L, Wang J-X (2021) PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain. J Comput Phys 428:110079
Gao H, Zahr MJ, Wang J-X (2022) Physics-informed graph neural Galerkin networks: a unified framework for solving PDE-governed forward and inverse problems. Comput Methods Appl Mech Eng 390:114502
Hajimohammadi Z, Parand K (2021) Numerical learning approximation of time-fractional sub diffusion model on a semi-infinite domain. Chaos Solitons Fractals 142:110435
Hochstadt H (1986) The functions of mathematical physics. Dover Publications Inc., New York
Hussein SO, Lesnic D (2014) Determination of a space-dependent source function in the one-dimensional wave equation. Electron J Bound Elem 12:1–26
Hussein SO, Lesnic D (2016) Determination of forcing functions in the wave equation. Part I: the space-dependent case. J Eng Math 96:115–133
Isakov V (2006) Inverse problems for partial differential equations. Springer, New York
Jagtap AD, Kawaguchi K, Karniadakis GE (2020a) Adaptive activation functions accelerate convergence in deep and physics-informed neural networks. J Comput Phys 404:109136
Jagtap AD, Kawaguchi K, Karniadakis GE (2020b) Locally adaptive activation functions with slope recovery for deep and physics-informed neural networks. Proc R Soc A 476:20200334
Jagtap AD, Kharazmi E, Karniadakis GE (2020c) Conservative physics-informed neural networks on discrete domains for conservation laws: applications to forward and inverse problems. Comput Methods Appl Mech Eng 365:113028
Jagtap AD, Mao Z, Adams N, Karniadakis GE (2022) Physics-informed neural networks for inverse problems in supersonic flows. arXiv preprint arXiv:2202.11821
Jin X, Cai S, Li H, Karniadakis GE (2021) NSFnets (Navier–Stokes flow nets): Physics-informed neural networks for the incompressible Navier–Stokes equations. J Comput Phys 426:109951
Lesnic D, Hussein SO, Johansson BT (2016) Inverse space-dependent force problems for the wave equation. J Comput Appl Math 306:10–39
Mao Z, Jagtap AD, Karniadakis GE (2020) Physics-informed neural networks for high-speed flows. Comput Methods Appl Mech Eng 360:112789
Meer R, Oosterlee CW, Borovykh A (2022) Optimally weighted loss functions for solving pdes with neural networks. J Comput Appl Math 405:113887
Meronen L, Trapp M, Solin A (2021) Periodic activation functions induce stationarity. Adv Neural Inf Process Syst 34:1673–1685
Mishra S, Molinaro R (2022) Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for PDEs. IMA J Numer Anal 42(2):981–1022
Moseley B, Markham A, Meyer TN (2020) Solving the wave equation with physics-informed deep learning. arXiv preprint arXiv:2006.11894
Murray M, Abrol V, Tanner J (2022) Activation function design for deep networks: linearity and effective initialisation. Appl Comput Harmon Anal 59:117–154
Nguyen LH (2019) An inverse space-dependent source problem for hyperbolic equations and the Lipschitz-like convergence of the quasi-reversibility method. Inverse Probl 35(35):035007. https://doi.org/10.1088/1361-6420/aafe8f
Prilepko AI, Orlovsky DG, Vasin IA (2000) Methods for solving inverse problems in mathematical physics. Marcel Dekker Inc, New York
Raissi M, Perdikaris P, Karniadakis GE (2019) Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys 378:686–707
Rashedi K (2021) A numerical solution of an inverse diffusion problem based on operational matrices of orthonormal polynomials. Math Method Appl Sci 44:12980–12997
Rashedi K (2022a) Recovery of coefficients of a heat equation by Ritz collocation method. Kuwaut J Sci. https://doi.org/10.48129/kjs.18581
Rashedi K (2022b) A spectral method based on Bernstein orthonormal basis functions for solving an inverse Roseneau equation. Comput Appl Math. https://doi.org/10.1007/s40314-02226401908-0
Rashedi K, Baharifard F, Sarraf A (2022) Stable recovery of a space-dependent force function in a one-dimensional wave equation via Ritz collocation method. J Math Model 10:463–480
Samarskii AA, Vabishchevich AN (2007) Numerical methods for solving inverse problems of mathematical physics. Walter de Gruyter, Berlin
Sitzmann V et al (2020) Implicit neural representations with periodic activation functions. Adv Neural Inf Process Syst 33:7462–7473
Stoer J, Bulirsch R (1980) Introduction to numerical analysis. Springer, New York
Wei T, Li M (2006) High order numerical derivatives for one-dimensional scattered noisy data. Appl Math Comput 175:1744–1759
Wen J, Yamamoto M, Wei T (2013) Simultaneous determination of a time-dependent heat source and the initial temperature in an inverse heat conduction problem. Inverse Probl Sci Eng 21:485–499
Yamamoto M (1995) Stability, reconstruction formula and regularization for an inverse source hyperbolic problem by a control method. Inverse Probl 1:481–496
Yuan L et al (2022) A-PINN: auxiliary physics informed neural networks for forward and inverse problems of nonlinear integro-differential equations. J Comput Phys 462:111260
Zhang D, Lu L, Guo L, Karniadakis GE (2019) Quantifying total uncertainty in physics-informed neural networks for solving forward and inverse stochastic problems. J Comput Phys 397:108850
Acknowledgements
The authors are very grateful to the referees for carefully reading the paper and for their comments and suggestions which have led to improvements of the paper.
Funding
We confirm that the authors did not receive any funding for this research.
Author information
Authors and Affiliations
Contributions
AS designed the numerical algorithm, implemented the algorithm, and conducted numerical experiments for the neural network approximation. FB worked on writing the introduction, and contributed to the main methodology (the neural network aspect). KR worked on the Ritz approximation of the inverse wave problems, and contributed to the main methodology (the Ritz aspect) and conducted numerical experiments for the Ritz approximation. All authors reviewed the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
We confirm that there are no competing interests.
Ethical approval and consent to participate
All authors gave their consent for participation. There are no other participants other than the authors.
Consent for publication
All authors gave their consent for publication.
Human and animal ethics
Not applicable.
Additional information
Communicated by Vinicius Albani.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Sarraf, A., Baharifard, F. & Rashedi, K. A decoupled physics-informed neural network for recovering a space-dependent force function in the wave equation from integral overdetermination data. Comp. Appl. Math. 42, 178 (2023). https://doi.org/10.1007/s40314-023-02323-9
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40314-023-02323-9