De Rham compatible deep neural network FEM

DOI10.1016/J.NEUNET.2023.06.008MaRDI QIDQ6057971zbMATH OpenOpenAlexFDO

Authors Marcello Longo, Joost A. A. Opschoor, Nico Disch, Christoph Schwab, J. Zech

Publication date 26 October 2023

Published in Neural Networks (Search for Journal in Brave)

Copyright license Creative Commons Attribution 4.0 International

Full work available at URL https://arxiv.org/abs/2201.05395

zbMATH Keywords

de Rham complexfinite elementsLavrentiev gapneural networksPINNs

Mathematics Subject Classification ID

de Rham theory in global analysis (58A12) Finite element, Rayleigh-Ritz and Galerkin methods for initial value and initial-boundary value problems involving PDEs (65M60) Artificial neural networks and deep learning (68T07) Finite element, Galerkin and related methods applied to problems in optics and electromagnetic theory (78M10)

Abstract: On general regular simplicial partitions $mathcalT$ of bounded polytopal domains $Omegasubsetmathbb{R}^d$, $din2,3$, we construct emph{exact neural network (NN) emulations} of all lowest order finite element spaces in the discrete de Rham complex. These include the spaces of piecewise constant functions, continuous piecewise linear (CPwL) functions, the classical ``Raviart-Thomas element, and the ``N'{e}d'{e}lec edge element. For all but the CPwL case, our network architectures employ both ReLU (rectified linear unit) and BiSU (binary step unit) activations to capture discontinuities. In the important case of CPwL functions, we prove that it suffices to work with pure ReLU nets. Our construction and DNN architecture generalizes previous results in that no geometric restrictions on the regular simplicial partitions $mathcalT$ of $Omega$ are required for DNN emulation. In addition, for CPwL functions our DNN construction is valid in any dimension $dgeq2$. Our ``FE-Nets are required in the variationally correct, structure-preserving approximation of boundary value problems of electromagnetism in nonconvex polyhedra $Omegasubsetmathbb{R}^3$. They are thus an essential ingredient in the application of e.g., the methodology of ``physics-informed NNs or ``deep Ritz methods to electromagnetic field simulation via deep learning techniques. We indicate generalizations of our constructions to higher-order compatible spaces and other, non-compatible classes of discretizations, in particular the ``Crouzeix-Raviart elements and Hybridized, Higher Order (HHO) methods.

Recommendations

• ReLU deep neural networks and linear finite elements
• Can neural networks learn finite elements?
• Deep ReLU networks and high-order finite element methods
• A finite element based deep learning solver for parametric PDEs

Cites work

• A coercive bilinear form for Maxwell's equations
• An optimal domain decomposition preconditioner for low-frequency time-harmonic Maxwell equations
• B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data
• Boundary element methods
• Boundary element methods for Maxwell transmission problems in Lipschitz domains
• Crouzeix-Raviart approximation of the total variation on simplicial meshes
• Crouzeix-Raviart finite element method for non-autonomous variational problems with Lavrentiev gap
• Deep learning in high dimension: neural network expression rates for generalized polynomial chaos expansions in UQ
• Deep vs. shallow networks: an approximation theory perspective
• Differential complexes and stability of finite element methods. II: The elasticity complex
• Enforcing exact physics in scientific machine learning: a data-driven exterior calculus on graphs
• Error bounds for approximations with deep ReLU networks
• Exponential ReLU DNN expression of holomorphic maps in high dimension
• Finite element quasi-interpolation and best approximation
• Finite elements I. Approximation and interpolation
• scientific article; zbMATH DE number 3848098 (Why is no real title available?)
• Hybrid high-order methods. A primer with applications to solid mechanics
• Maxwell and Lamé eigenvalues on polyhedra
• Multipatch approximation of the de Rham sequence and its traces in isogeometric analysis
• Nonconforming Finite Elements for the Stokes Problem
• Optimal approximation of piecewise smooth functions using deep ReLU neural networks
• Orientation embedded high order shape functions for the exact sequence elements of all shapes
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• PowerNet: efficient representations of polynomials and smooth functions by deep neural networks with rectified power units
• Regularity of minimizers of integrals of the calculus of variations with non-standard growth conditions
• ReLU deep neural networks and linear finite elements
• ReLU neural network Galerkin BEM
• Singularities and computation of minimizers for variational problems
• Singularities of Maxwell interface problems
• The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems
• Vector potentials in three-dimensional non-smooth domains

Cited in

(7)

• Deep ReLU networks and high-order finite element methods. II: Chebyšev emulation
• Can neural networks learn finite elements?
• Neural and spectral operator surrogates: unified construction and expression rate bounds
• Deep operator network approximation rates for Lipschitz operators
• Exponential expressivity of ReLU^k neural networks on Gevrey classes with point singularities.
• Sparsely connected neural network representation of Lagrange finite element function
• Exact Representation of Piecewise Polynomial Finite Element Functions by Two-Hidden-Layer Multilayer Perceptrons

This page was built for publication: De Rham compatible deep neural network FEM

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6057971)