Abstract
In this work we present an attempt to replace an a posteriori MOOD loop used in a high accurate Finite Volume (FV) scheme by a trained artificial Neural Network (NN). The MOOD loop, by decrementing the reconstruction polynomial degrees, ensures accuracy, essentially non-oscillatory, robustness properties and preserves physical features. Indeed it replaces the classical a priori limiting strategy by an a posteriori troubled cell detection, supplemented with a local time-step re-computation using a lower order FV scheme (ie lower polynomial degree reconstructions). We have trained shallow NNs made of only two so-called hidden layers and few perceptrons which a priori produces an educated guess (classification) of the appropriate polynomial degree to be used in a given cell knowing the physical and numerical states in its vicinity. We present a proof of concept in 1D. The strategy to train and use such NNs is described on several 1D toy models: scalar advection and Burgers’ equation, the isentropic Euler and radiative M1 systems. Each toy model brings new difficulties which are enlightened on the obtained numerical solutions. On these toy models, and for the proposed test cases, we observe that an artificial NN can be trained and substituted to the a posteriori MOOD loop in mimicking the numerical admissibility criteria and predicting the appropriate polynomial degree to be employed safely. The physical admissibility criteria is however still dealt with the a posteriori MOOD loop. Constructing a valid training data set is of paramount importance, but once available, the numerical scheme supplemented with NN produces promising results in this 1D setting.
Similar content being viewed by others
Notes
We must be careful that the NN may be certain of its prediction, even if it is a wrong one.
Notice that without the a posteriori MOOD loop the 2nd and 4th order schemes crash due to the occurrence of negative densities.
References
Berg, J., Nyström, K.: A unified deep artificial neural network approach to partial differential equations in complex geometries. Neurocomputing 317, 28–41 (2018)
Blachère, F., Turpault, R.: An admissibility and asymptotic preserving scheme for systems of conservation laws with source term on 2d unstructured meshes with high-order mood reconstruction. Comput. Methods Appl. Mech. Eng. 317, 836–867 (2017)
Blanchard, G., Loubère, R.: High order accurate conservative remapping scheme on polygonal meshes using a posteriori mood limiting. Comput. Fluids 136, 83–103 (2016)
Boscheri, W., Dumbser, M., Loubère, R., Maire, P.-H.: A second-order cell-centered lagrangian ader-mood finite volume scheme on multidimensional unstructured meshes for hydrodynamics. J. Comput. Phys. 358, 103–129 (2018)
Boscheri, W., Loubère, R., Dumbser, M.: Direct arbitrary-lagrangian-eulerian ader-mood finite volume schemes for multidimensional hyperbolic conservation laws. J. Comput. Phys. 292, 56–87 (2015)
Braeunig, J.-P., Loubère, R., Motte, R., Peybernes, M., Poncet, R.: A posteriori limiting for 2d lagrange plus remap schemes solving the hydrodynamics system of equations. Comput. Fluids Recent Prog. Nonlinear Numer. Methods Time-dependent Flow Trans. Prob. 169, 249–262 (2018)
Burgers, J.M.: The Nonlinear Diffusion Equation: Asymptotic Solutions and Statistical Problems. Springer, Berlin (1974)
Clain, S., Diot, S., Loubère, R.: A high-order finite volume method for systems of conservation laws–multi-dimensional optimal order detection (MOOD). J. Comput. Phys. 230(10), 4028–4050 (2011)
Clain, S., Figueiredo, J.: The mood method for the non-conservative shallow-water system. Comput. Fluids 145, 99–128 (2017)
Clain, S., Loubère, R., Machado, G.J.: A posteriori stabilized sixth-order finite volume scheme for one-dimensional steady-state hyperbolic equations. Adv. Comput. Math. 44(2), 571–607 (2018)
Coudière, Y., Turpault, R.: A domain decomposition strategy for a very high-order finite volumes scheme applied to cardiac electrophysiology. J. Comput. Sci. 37, 101025 (2019)
Diot, S., Clain, S., Loubère, R.: Improved detection criteria for the multi-dimensional optimal order detection (MOOD) on unstructured meshes with very high-order polynomials. Comput. Fluids 64(Supplement C), 43–63 (2012)
Diot, S., Loubère, R., Clain, S.: The multidimensional optimal order detection method in the three-dimensional case: very high-order finite volume method for hyperbolic systems. Int. J. Numer. Meth. Fluids 73(4), 362–392 (2013)
Dubroca, B., Feugeas, J.L.: Etude théorique et numérique d’une hiérarchie de modèles aux moments pour le transfert radiatif. Comptes Rendus l’Académie des Sci. Ser.I Math. 329(10), 915–920 (1999)
Dumbser, M., Loubère, R.: A simple robust and accurate a posteriori sub-cell finite volume limiter for the discontinuous Galerkin method on unstructured meshes. J. Comput. Phys. 319(Supplement C), 163–199 (2016)
Dumbser, M., Zanotti, O., Loubère, R., Diot, S.: A posteriori subcell limiting of the discontinuous galerkin finite element method for hyperbolic conservation laws. J. Comput. Phys. 278, 47–75 (2014)
Fernández-Fidalgo, J., Nogueira, X., Ramírez, L., Colominas, I.: An a posteriori, efficient, high-spectral resolution hybrid finite-difference method for compressible flows. Comput. Methods Appl. Mech. Eng. 335, 91–127 (2018)
Giri, P., Qiu, Q.J.J.: A high-order runge-kutta discontinuous galerkin method with a subcell limiter on adaptive unstructured grids for two-dimensional compressible inviscid flows. Int. J. Numer. Meth. Fluids 91(8), 367–394 (2019)
Gottlieb, S., Shu, C.W.: Total variation diminishing Runge-Kutta schemes. Math. Comput. 67, 73–85 (1998)
Gottlieb, S., Shu, C.-W.: Total variation diminishing Runge-Kutta schemes. Math. Comput. 6(221), 73–85 (1998)
Gyrya, V., Shashkov, M., Skurikhin, A., Tokareva, S.: Machine learning approaches for the solution of the riemann problem in fluid dynamics: a case study. preprint, (2020)
Hagan, M.T., Demuth, H.B., Beale, M.H.: Neural Network Design. PWS Publishing, India (1996)
Haykin, S.S.: Neural Networks and Learning Machines, 3rd edn. Pearson Education, London (2009)
Hebb, D.O.: The Organization of Behavior, vol. 65. Wiley, New York (1949)
Shu, C.-W., Qiu, J.: A comparison of troubled-cell indicators for Runge–Kutta discontinuous Galerkin methods using weighted essentially nonoscillatory limiters. J. Sci. Comput. 27, 995–1013 (2005)
Jiang, G.-S., Shu, C.-W.: Efficient implementation of weighted eno schemes. J. Comput. Phys. 126(1), 202–228 (1996)
Jiang, Z.-H., Yan, C., Jian, Y.: Efficient methods with higher order interpolation and mood strategy for compressible turbulence simulations. J. Comput. Phys. 371, 528–550 (2018)
Kriesel, D.: A Brief Introduction to Neural Networks. (2007)
Loubère, R., Dumbser, M., Diot, S.: A new family of high order unstructured MOOD and ADER finite volume schemes for multidimensional systems of hyperbolic conservation laws. Commun. Comput. Phys. 16(3), 718–763 (2014)
Marquardt, D.: An algorithm for least-squares estimation of nonlinear parameters. SIAM J. Appl. Math. 11(2), 431–441 (1963)
McCulloch, W.S., Pitts, W.: A logical calculus of the ideas immanent in nervous activity. Bull. Math. Biophys. 5, 115–133 (1943)
Minsky, M., Papert, S.: Perceptrons: An Introduction to Computational Geometry. MIT Press, Cambridge (1969)
Nogueira, X., Ramírez, L., Clain, S., Loubère, R., Cueto-Felgueroso, L., Colominas, I.: High-accurate sph method with multidimensional optimal order detection limiting. Comput. Methods Appl. Mech. Eng. 310, 134–155 (2016)
Maziar Raissi and George Em Karniadakis: Hidden physics models: machine learning of nonlinear partial differential equations. J. Comput. Phys. 357, 125–141 (2018)
Ray, D., Hesthaven, J.S.: An artificial neural network as a troubled-cell indicator. J. Comput. Phys. 367, 166–191 (2018)
Ray, D., Hesthaven, J.S.: Detecting troubled-cells on two-dimensional unstructured grids using a neural network. J. Comput. Phys. 397, 108845 (2019)
Rosenblatt, F.: The perceptron: a probabilistic model for information storage and organization in the brain. Psychological Review, pp. 65–386, (1958)
Rosenblatt, F.: Principles of Neurodynamics. Spartan, New York (1962)
Sarazin-Desbois, C.: Méthodes numériques pour des systèmes hyperboliques avec terme source provenant de physiques complexes autour du rayonnement. Ph.D. thesis, Université de Nantes, (2013)
Schmidhuber, J.: Deep learning in neural networks: an overview. Neural Networks 61, 85–117 (2015)
Semplice, M., Loubère, R.: Adaptive-mesh-refinement for hyperbolic systems of conservation laws based on a posteriori stabilized high order polynomial reconstructions. J. Comput. Phys. 354, 86–110 (2018)
Suresh, A., Huynh, H.T.: Accurate monotonicity-preserving schemes with runge-kutta time stepping. J. Comput. Phys. 136, 83–99 (1997)
Tann, S., Deng, X., Shimizu, Y., Loubère, R., Xiao, F.: Solution property preserving reconstruction for finite volume scheme: a bvd+mood framework. International Journal for Numerical Methods in Fluids, n/a(n/a), (2019)
Toro, E.F.: Riemann Sovlers and Numerical Methods for Fluid Dynamics. Springer, Berlin (2009)
Veiga, M.H., Abgrall, R.: Towards a general stabilisation method for conservation laws using a multilayer perceptron neural network: 1d scalar and system of equations. In European Conference on Computational Mechanics and VII European Conference on Computational Fluid Dynamics, No. 1, pp. 2525–2550. ECCM, (June 2018)
Vilar, F.: A posteriori correction of high-order discontinuous galerkin scheme through subcell finite volume formulation and flux reconstruction. J. Comput. Phys. 387, 245–279 (2019)
Werbos, P.J.: Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. Harvard University, Cambridge (1975)
Acknowledgements
RL would like to thank M. Han Veiga (University of Zürich) and S. Clain (Universidade do Minho) for sharing fruitful discussions on neural networks. The authors thank the anonymous reviewers who have led to an improved version of this paper. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Bourriaud, A., Loubère, R. & Turpault, R. A Priori Neural Networks Versus A Posteriori MOOD Loop: A High Accurate 1D FV Scheme Testing Bed. J Sci Comput 84, 31 (2020). https://doi.org/10.1007/s10915-020-01282-1
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10915-020-01282-1