Regularity-conforming neural networks (ReCoNNs) for solving partial differential equations
Year: 2025
Type: article
Abstract: Whilst the Universal Approximation Theorem guarantees the existence of approximations to Sobolev functions –the natural function spaces for PDEs– by Neural Networks (NNs) of sufficient size, low-regularity solutions may lead to poor approximations in practice. For example, classical fully-connected feed-forward NNs fail to approximate continuous functions whose gradient is discontinuous when employing strong formulations like in Physics Informed Neural Networks (PINNs). In this article, we propose the use of regularity-conforming neural networks, where a priori information on the regularity of solutions to PDEs can be employed to construct proper architectures. We illustrate the potential of such architectures via a two-dimensional (2D) transmission problem, where the solution may admit discontinuities in the gradient across interfaces, as well as power-like singularities at certain points. In particular, we formulate the weak transmission problem in a PINNs-like strong formulation with interface and continuity conditions. Such architectures are partially explainable; discontinuities are explicitly described, allowing the introduction of novel terms into the loss function. We demonstrate via several model problems in one and two dimensions the advantages of using regularity-conforming architectures in contrast to classical architectures. The ideas presented in this article easily extend to problems in higher dimensions.
Source: Journal of Computational Physics
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,980
Funders: Ministerio de Ciencia, Innovación y Universidades, European Commission, Federación Española de Enfermedades Raras, Eusko Jaurlaritza, Ministerio de Ciencia e Innovación, Ministerio de Asuntos Económicos y Transformación Digital, Gobierno de España, Basque Center for Applied Mathematics, Horizon 2020 Framework Programme, HORIZON EUROPE Framework Programme, Horizon 2020, Agencia Estatal de Investigación, H2020 Marie Skłodowska-Curie Actions, HORIZON EUROPE Marie Sklodowska-Curie Actions
Awards: G1480809515, G1612158077, G1633674249, G1855990336, G2262748287, G2269550651, G2319203330, G3170287466, G3480869486, G3607574089, G3681454997, G3690652358, G3690828678, G4042783231, G4165417455, G4241082588, G425668148, G434337789, G438088396, G542055026, G5421974903, G55232499, G5867013586, G6072587662, G6294024929, G7436716198, Space-time DPG methods for partial-differential equations with geophysical applications, G7748812462, G7951220496, G8305580143, G8684520763, G8939903727, G95601984
OpenAlex ID: W4408950304