Deep learning-based schemes for singularly perturbed convection-diffusion problems

Adrien Beguinet; Virginie Ehrlacher; Roberta Flenghi; Maria Fuente; Olga Mula; Agustin Somacal

doi:10.1051/proc/202373048

Geodesic

Parcourir par

Deep learning-based schemes for singularly perturbed convection-diffusion problems

Adrien Beguinet ¹ ; Virginie Ehrlacher ² ; Roberta Flenghi ³ ; Maria Fuente ⁴ ; Olga Mula ⁵ ; Agustin Somacal ⁶

¹ Fédération de Mathématiques, CentraleSupélec FR3487 CNRS, Paris-Sacla University, France
² Ecole des Ponts ParisTech & MATHERIALS INRIA team-project, Marne-la-Vallée, France
³ Cermics, École des Ponts, INRIA, Marne-la-Vallée, France
⁴ COMMEDIA, INRIA Paris, France & LJLL, Sorbonne Université
⁵ Department of Mathematics and Computer Science, Eindhoven University of Technology, 5600 MB, Eindhoven, Netherlands
⁶ LJLL, Sorbonne Université, Paris, France

ESAIM. Proceedings, Tome 73 (2023), pp. 48-67.

Voir la notice de l'article provenant de la source EDP Sciences

Résumé

Deep learning-based numerical schemes such as Physically Informed Neural Networks (PINNs) have recently emerged as an alternative to classical numerical schemes for solving Partial Differential Equations (PDEs). They are very appealing at first sight because implementing vanilla versions of PINNs based on strong residual forms is easy, and neural networks offer very high approximation capabilities. However, when the PDE solutions are low regular, an expert insight is required to build deep learning formulations that do not incur in variational crimes. Optimization solvers are also significantly challenged, and can potentially spoil the final quality of the approximated solution due to the convergence to bad local minima, and bad generalization capabilities. In this paper, we present an exhaustive numerical study of the merits and limitations of these schemes when solutions exhibit low-regularity, and compare performance with respect to more benign cases when solutions are very smooth. As a support for our study, we consider singularly perturbed convection-diffusion problems where the regularity of solutions typically degrades as certain multiscale parameters go to zero.

DOI : 10.1051/proc/202373048

Affiliations des auteurs :

Adrien Beguinet ¹ ; Virginie Ehrlacher ² ; Roberta Flenghi ³ ; Maria Fuente ⁴ ; Olga Mula ⁵ ; Agustin Somacal ⁶

@article{EP_2023_73_a3,
     author = {Adrien Beguinet and Virginie Ehrlacher and Roberta Flenghi and Maria Fuente and Olga Mula and Agustin Somacal},
     title = {Deep learning-based schemes for singularly perturbed convection-diffusion problems},
     journal = {ESAIM. Proceedings},
     pages = {48--67},
     publisher = {mathdoc},
     volume = {73},
     year = {2023},
     doi = {10.1051/proc/202373048},
     language = {en},
     url = {http://geodesic.mathdoc.fr/articles/10.1051/proc/202373048/}
}

TY  - JOUR
AU  - Adrien Beguinet
AU  - Virginie Ehrlacher
AU  - Roberta Flenghi
AU  - Maria Fuente
AU  - Olga Mula
AU  - Agustin Somacal
TI  - Deep learning-based schemes for singularly perturbed convection-diffusion problems
JO  - ESAIM. Proceedings
PY  - 2023
SP  - 48
EP  - 67
VL  - 73
PB  - mathdoc
UR  - http://geodesic.mathdoc.fr/articles/10.1051/proc/202373048/
DO  - 10.1051/proc/202373048
LA  - en
ID  - EP_2023_73_a3
ER  -

%0 Journal Article
%A Adrien Beguinet
%A Virginie Ehrlacher
%A Roberta Flenghi
%A Maria Fuente
%A Olga Mula
%A Agustin Somacal
%T Deep learning-based schemes for singularly perturbed convection-diffusion problems
%J ESAIM. Proceedings
%D 2023
%P 48-67
%V 73
%I mathdoc
%U http://geodesic.mathdoc.fr/articles/10.1051/proc/202373048/
%R 10.1051/proc/202373048
%G en
%F EP_2023_73_a3

Adrien Beguinet; Virginie Ehrlacher; Roberta Flenghi; Maria Fuente; Olga Mula; Agustin Somacal. Deep learning-based schemes for singularly perturbed convection-diffusion problems. ESAIM. Proceedings, Tome 73 (2023), pp. 48-67. doi : 10.1051/proc/202373048. http://geodesic.mathdoc.fr/articles/10.1051/proc/202373048/

Cité par Sources :