magazinelogo

International Journal of Statistics and Data Science

ISSN Online: 3070-6459 CODEN:
Frequency: Quarterly Email: IJSDS@hillpublish.com
Total View: 71780 Downloads: 5115 Citations: 0 (From Dimensions)
ArticleOpen Access http://dx.doi.org/10.26855/ijsds.2026.06.002

Enhanced Training of Physics-Informed Radial Basis Function Networks: Algorithmic Modifications and Applications to Fluid Dynamics

Dmitry Stenkin, Vladimir Gorbachenko*

Department of Computer Technologies, Penza State University, Penza 440026, Russia.

*Corresponding author: Vladimir Gorbachenko

Published: May 14,2026

Abstract

This paper presents a comparative study of physics-informed neural networks (PINNs) for solving boundary value problems governed by partial differential equations. Particular attention is given to physics-informed radial basis function networks (PIRBFNs), which exhibit significantly faster convergence than conventional fully connected PINNs, especially when both network weights and radial basis function parameters are optimized simultaneously. Modified training algorithms for PIRBFNs are developed based on the ADAM, Sophia, and Levenberg–Marquardt methods, extended to enable joint optimization of network weights and basis function parameters. This strategy enhances optimization efficiency by improving both convergence speed and the adaptability of the approximation model. Architectural designs and computational procedures are proposed for applying PIRBFNs to a set of benchmark problems in computational fluid dynamics, including Kovasznay flow, the Taylor-Green vortex, and lid-driven cavity flow. Numerical experiments demonstrate that the proposed training algorithms reduce the number of optimization iterations by a factor of 25–41.3 and decrease the total computational time by 24.3–31.0 times compared with the standard Adam algorithm, while maintaining high solution accuracy.

Keywords

Physics-informed neural networks; radial basis function networks; partial differen-tial equations; boundary value problems; scientific machine learning; Navier–Stokes equations; Levenberg–Marquardt method; optimization algorithms; com-putational hydrodynamics

References

[1] Lagaris IE, Likas A, Fotiadis DI. Artificial neural networks for solving ordinary and partial differential equations. IEEE Trans Neural Netw. 1998;9(5):987-1000. doi:10.1109/72.712178.

[2] Yadav N, Yadav A, Kumar M. An introduction to neural network methods for differential equations. Dordrecht: Springer; 2015. doi:10.1007/978-94-017-9816-7.

[3] Raissi M, Perdikaris P, Karniadakis GE. Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations. arXiv. 2017. Preprint. doi:10.48550/arXiv.1711.10561.

[4] Raissi M, Perdikaris P, Karniadakis GE. Physics Informed Deep Learning (Part II): Data-driven Discovery of Nonlinear Partial Differential Equations. arXiv. 2017. Preprint. doi:10.48550/arXiv.1711.10566.

[5] Raissi M, Perdikaris P, Karniadakis GE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys. 2019;378:686-707. doi:10.1016/j.jcp.2018.10.045.

[6] Lakshmanan V, Robinson S, Munn M. Machine Learning Design Patterns: Solutions to Common Challenges in Data Preparation, Model Building, and MLOps. 1st ed. Sebastopol: O'Reilly Media; 2020.

[7] Jin X, Cai S, Li H, Karniadakis GE. NSFnets (Navier-Stokes Flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations. J Comput Phys. 2021;426:109951. doi:10.1016/j.jcp.2020.109951.

[8] Averweg S, Schwarz A, Schwarz C, Schröder J. 3D modeling of generalized Newtonian fluid flow with data assimilation using the least-squares finite element method. Comput Methods Appl Mech Eng. 2022;392:114668. doi:10.1016/j.cma.2022.114668.

[9] Zhang L-C, Lee D. Design-based individual prediction. arXiv. 2023. Preprint. doi:10.48550/arXiv.2301.09117.

[10] Karniadakis GE, Kevrekidis IG, Lu L, Perdikaris P, Wang S, Yang L. Physics-informed machine learning. Nat Rev Phys. 2021;3:422-440. doi:10.1038/s42254-021-00314-5.

[11] Zubov K, McCarthy Z, Ma Y, et al. NeuralPDE: Automating Physics-Informed Neural Networks (PINNs) with Error Approximations. arXiv. 2021. Preprint. doi:10.48550/arXiv.2107.09443.

[12] Taha WM, Taha AE, Thunberg J. Cyber-Physical Systems: A Model-Based Approach. Cham: Springer; 2021. doi:10.1007/978-3-030-36071-9.

[13] Lu Y, Liu C, Wang K, Huang H, Xu X. Digital Twin-driven smart manufacturing: Connotation, reference model, applications and research issues. Robot Comput Integr Manuf. 2020;61:101837. doi:10.1016/j.rcim.2019.101837.

[14] Uhlemann THJ, Steinhilper CLR, Steinhilper R. The Digital Twin: Realizing the Cyber-Physical Production System for Industry 4.0. Procedia CIRP. 2017;61:335-340. doi:10.1016/j.procir.2016.11.152.

[15] Tarkhov DA, Malykhina GF. Neural network modelling methods for creating digital twins of real objects. J Phys Conf Ser. 2019;1236:012068. doi:10.1088/1742-6596/1236/1/012068.

[16] Tarkhov DA, Vasilyev AN. Semi-Empirical Neural Network Modeling and Digital Twins Development. Cambridge: Academic Press; 2019.

[17] Raissi M, Perdikaris P, Karniadakis GE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys. 2019;378:686-707. doi:10.1016/j.jcp.2018.10.045.

[18] Lu L, Meng X, Mao Z, Karniadakis GE. DeepXDE: A deep learning library for solving differential equations. SIAM Rev. 2021;63(1):208-228. doi:10.1137/19M1274067.

[19] Cai S, Wang Z, Wang S, Perdikaris P, Karniadakis GE. Physics-Informed Neural Networks (PINNs) for Heat Transfer. J Heat Transfer. 2021;143(6):060801. doi:10.1115/1.4050542.

[20] Geron A. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intel-ligent Systems. 3rd ed. Sebastopol: O'Reilly Media; 2022.

[21] Bavdin AG, Pearlmutter BF, Radul AA, Siskind JM. Automatic Differentiation in Machine Learning: a Survey. J Mach Learn Res. 2018;18(153):1-43. doi:10.48550/arXiv.1502.05767.

[22] Strizhak SV, Koshelev KB. The use of a physically based neural network on the example of modeling hydrodynamic processes that allow an analytical solution. Proc Inst Syst Program RAS. 2023;35(5):245-258. doi:10.15514/ISPRAS-2023-35(5)-16.

[23] Sophiya AA, Nair AK, Maleki S, Krishnababu SK. A comprehensive analysis of PINNs: Variants, Applications, and Challenges. arXiv. 2025. Preprint. doi:10.48550/arXiv.2505.22761.

[24] Gorbachenko VI, Stenkin DA. Solving Equations Describing Processes in a Piecewise Homogeneous Medium on Radial Basis Functions Networks. In: Studies in Computational Intelligence. Vol 925. Cham: Springer; 2021:412-419. doi:10.1007/978-3-030-60577-3_49.

[25] Gorbachenko VI, Stenkin DA. Solving of Inverse Coefficient Problems on Networks of Radial Basis Functions. In: Studies in Computational Intelligence. Vol 1008. Cham: Springer; 2022:230-237. doi:10.1007/978-3-030-91581-0_31.

[26] Cybenko G. Approximation by Superposition of a Sigmoidal Function. Math Control Signals Syst. 1989;2:303-314. doi:10.1007/BF02134016.

[27] Maiorov V, Pinkus A. Lower bounds for approximation by MLP neural networks. Neurocomputing. 1999;25(1-3):81-91. doi:10.1016/S0925-2312(98)00111-8.

[28] Hornik K. Approximation capabilities of multilayer feedforward networks. Neural Netw. 1991;4(2):251-257. doi:10.1016/0893-6080(91)90009-T.

[29] Hornik K, Stinchcombe M, White H. Multilayer feedforward networks are universal approximators. Neural Netw. 1989;2(5):359-366. doi:10.1016/0893-6080(89)90020-8.

[30] Leshno M, Lin VY, Pinkus A, Schocken S. Multilayer Feedforward Networks with a Non-Polynomial Activation Function Can Approximate Any Function. Neural Netw. 1993;6(6):861-867. doi:10.1016/S0893-6080(05)80131-5.

[31] Osowski S. Sieci neuronowe do przetwarzania informacji. Warszawa: Oficyna Wydawnicza Politechniki Warszawskiej; 2000.

[32] Haykin S. Neural Networks and Learning Machines. 3rd ed. London: Pearson; 2009.

[33] Park J, Sandberg IW. Approximation and Radial-Basis-Function Networks. Neural Comput. 1993;5(2):305-316. doi:10.1162/neco.1993.5.2.305.

[34] Park J, Sandberg IW. Universal Approximation Using Radial-Basis-Function Networks. Neural Comput. 1991;3(2):246-257. doi:10.1162/neco.1991.3.2.246.

[35] Buhmann MD. Radial Basis Functions: Theory and Implementations. Cambridge: Cambridge University Press; 2003.

[36] Micchelli CA. Interpolation of scattered data: Distance matrices and conditionally positive definite functions. Constr Approx. 1986;2(1):11-22. doi:10.1007/BF01893414.

[37] Eivazi H, Tahani M, Schlatter P, Vinuesa R. Physics-informed neural networks for solving Reynolds-averaged Navier–Stokes equations. Phys Fluids. 2022;34(7):075117. doi:10.1063/5.0095270.

[38] De Ryck T, Jagtap AD, Mishra S. Error estimates for physics-informed neural networks approximating the Navier–Stokes equations. IMA J Numer Anal. 2024;44(1):83-119. doi:10.1093/imanum/drac085.

[39] Demirkaya G, Wafo SC, Ilegbusi OJ. Direct solution of Navier–Stokes equations by radial basis functions. Appl Math Model. 2008;32(9):1848-1858. doi:10.1016/j.apm.2007.06.019.

[40] Chinchapatnam PP, Djidjeli K, Nair PB, Tan M. A compact RBF-FD based meshless method for the incompressible Navier–Stokes equations. Proc Inst Mech Eng M: J Eng Marit Environ. 2009;223(3):275-290. doi:10.1243/14750902JEME151.

[41] Abdukhakimov F, Xiang C, Kamzolov D, Gower R, Takáč M. SANIA: Polyak-type Optimization Framework Leads to Scale In-variant Stochastic Algorithms. arXiv. 2023. Preprint. doi:10.48550/arXiv.2312.17369.

[42] Nagwekar A. Towards Guided Descent: Optimization Algorithms for Training Neural Networks At Scale. arXiv. 2024. Preprint. doi:10.48550/arXiv.2512.18373.

[43] Liu H, Li Z, Hall D, Liang P, Ma T. Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training. arXiv. 2023. Preprint. doi:10.48550/arXiv.2305.14342.

[44] Hutchinson MF. A stochastic estimator of the trace of the influence matrix for laplacian smoothing splines. Commun Stat Simul Comput. 1989;18(3):1059-1076. doi:10.1080/03610919008812866.

[45] Schraudolph NN. Fast curvature matrix-vector products for second-order gradient descent. Neural Comput. 2002;14(7):1723-1738. doi:10.1162/08997660260028683.

[46] Pearlmutter BA. Fast Exact Multiplication by the Hessian. Neural Comput. 1994;6(1):147-160. doi:10.1162/neco.1994.6.1.147.

[47] Kovasznay LSG. Laminar flow behind a two-dimensional grid. Math Proc Camb Philos Soc. 1948;44(1):58-62. doi:10.1017/S0305004100023999.

[48] Slezkin NA. Dynamics of viscous incompressible fluid. Moscow: Gostekhizdat; 1955. Russian.

[49] Taylor GI, Green AE. Mechanism of the Production of Small Eddies from Large Ones. Proc R Soc Lond A. 1937;158(895):499-521. doi:10.1098/rspa.1937.0036.

[50] Botella O, Peyret R. Benchmark spectral results on the lid-driven cavity flow. Comput Fluids. 1998;27(4):421-433. doi:10.1016/S0045-7930(98)00002-4.

[51] Ghia U, Ghia KN, Shin CT. High-Re solutions for incompressible flow using the Navier-Stokes equations and a multigrid method. J Comput Phys. 1982;48(3):387-411. doi:10.1016/0021-9991(82)90058-4.

[52] Xie T, Yu H, Wilamowski BM. Comparison between Traditional Neural Networks and Radial Basis Function Networks. In: 2011 IEEE International Symposium on Industrial Electronics; 2011 Jun 27-30; Gdansk, Poland. IEEE; 2011. p. 1194-1199. doi:10.1109/ISIE.2011.5984334.

[53] Wang S, Teng Y, Perdikaris P. Understanding and mitigating gradient pathologies in physics-informed neural networks. arXiv. 2020. Preprint. doi:10.48550/arXiv.2001.04536. 

Copyright

© 2026 by the author(s).
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution-NonCommercial-NoDerivatives (CC BY-NC-ND) license, which permits non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited and is not modified or adapted.
https://creativecommons.org/licenses/by-nc-nd/4.0/

How to cite this paper

Enhanced Training of Physics-Informed Radial Basis Function Networks: Algorithmic Modifications and Applications to Fluid Dynamics

How to cite this paper: Dmitry Stenkin, Vladimir Gorbachenko. (2026). Enhanced Training of Physics-Informed Radial Basis Function Networks: Algorithmic Modifications and Applications to Fluid Dynamics. International Journal of Statistics and Data Science2(1), 11-29.

DOI: http://dx.doi.org/10.26855/ijsds.2026.06.002