Subspace decomposition based DNN algorithm for elliptic type multi-scale PDEs

DOI10.1016/J.JCP.2023.112242arXiv2112.06660MaRDI QIDQ6162913FDOQ6162913

Lei Zhang, Xian Li, Zhi-Qin John Xu

Publication date: 16 June 2023

Published in: Journal of Computational Physics (Search for Journal in Brave)

Abstract: While deep learning algorithms demonstrate a great potential in scientific computing, its application to multi-scale problems remains to be a big challenge. This is manifested by the "frequency principle" that neural networks tend to learn low frequency components first. Novel architectures such as multi-scale deep neural network (MscaleDNN) were proposed to alleviate this problem to some extent. In this paper, we construct a subspace decomposition based DNN (dubbed SD${}^2$NN) architecture for a class of multi-scale problems by combining traditional numerical analysis ideas and MscaleDNN algorithms. The proposed architecture includes one low frequency normal DNN submodule, and one (or a few) high frequency MscaleDNN submodule(s), which are designed to capture the smooth part and the oscillatory part of the multi-scale solutions, respectively. In addition, a novel trigonometric activation function is incorporated in the SD${}^2$NN model. We demonstrate the performance of the SD${}^2$NN architecture through several benchmark multi-scale problems in regular or irregular geometric domains. Numerical results show that the SD${}^2$NN model is superior to existing models such as MscaleDNN.

Full work available at URL: https://arxiv.org/abs/2112.06660

Recommendations

• A multi-scale DNN algorithm for nonlinear elliptic equations with multiple scales
• Deep neural network approximation for high-dimensional elliptic PDEs with boundary conditions
• A non-gradient method for solving elliptic partial differential equations with deep neural networks
• A derivative-free method for solving elliptic partial differential equations with deep neural networks
• Numerical Homogenization of Elliptic Multiscale Problems by Subspace Decomposition
• Solving multiscale elliptic problems by sparse radial basis function neural networks
• Multi-scale fusion network: a new deep learning structure for elliptic interface problems
• DNN-HDG: a deep learning hybridized discontinuous Galerkin method for solving some elliptic problems
• scientific article; zbMATH DE number 1047492
• Improved deep neural networks with domain decomposition in solving partial differential equations

zbMATH Keywords

activation functionmulti-scaleFourierDNNsubspace-decomposed

Mathematics Subject Classification ID

Artificial intelligence (68Txx) Numerical methods for partial differential equations, boundary value problems (65Nxx) Elliptic equations and elliptic systems (35Jxx)

Cites Work

• DGM: a deep learning algorithm for solving partial differential equations
• Polyharmonic homogenization, rough polyharmonic splines and sparse super-localization
• Multigrid with Rough Coefficients and Multiresolution Operator Decomposition from Hierarchical Information Games
• Localization of elliptic multiscale problems
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• Title not available (Why is that?)
• A multiscale finite element method for elliptic problems in composite materials and porous media
• Introduction to the network approximation method for materials modeling.
• Multiscale Finite Element Methods
• Iterative Methods by Space Decomposition and Subspace Correction
• Bayesian Numerical Homogenization
• Flux norm approach to finite-dimensional homogenization approximations with non-separated scales and high contrast
• The variational multiscale method -- a paradigm for computational mechanics
• Generalized multiscale finite element methods (GMsFEM)
• Variational multiscale residual-based turbulence modeling for large eddy simulation of incompressible flows
• Numerical homogenization of the acoustic wave equations with a continuum of scales
• Adaptivity and variational stabilization for convection-diffusion equations
• The heterogeneous multiscale methods
• A localized orthogonal decomposition method for semi-linear elliptic problems
• Analysis of the heterogeneous multiscale method for elliptic homogenization problems
• Homogenization of Parabolic Equations with a Continuum of Space and Time Scales
• Anomalous slow diffusion from perpetual homogenization.
• An Adaptive Generalized Multiscale Discontinuous Galerkin Method for High-Contrast Flow Problems
• Deep learning-based numerical methods for high-dimensional parabolic partial differential equations and backward stochastic differential equations
• Inferring solutions of differential equations using noisy multi-fidelity data
• The Deep Ritz Method: a deep learning-based numerical algorithm for solving variational problems
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Weak adversarial networks for high-dimensional partial differential equations
• Fast Eigenpairs Computation with Operator Adapted Wavelets and Hierarchical Subspace Correction
• A proposal on machine learning via dynamical systems
• Machine learning from a continuous viewpoint. I
• A Phase Shift Deep Neural Network for High Frequency Approximation and Wave Problems
• Deep Potential: A General Representation of a Many-Body Potential Energy Surface
• Machine Learning and Computational Mathematics
• Kernel flows: from learning kernels from data into the abyss
• Adaptive activation functions accelerate convergence in deep and physics-informed neural networks
• On the eigenvector bias of Fourier feature networks: from regression to solving multi-scale PDEs with physics-informed neural networks
• Multi-Scale Deep Neural Network (MscaleDNN) for Solving Poisson-Boltzmann Equation in Complex Domains
• Frequency Principle: Fourier Analysis Sheds Light on Deep Neural Networks
• A Multi-Scale DNN Algorithm for Nonlinear Elliptic Equations with Multiple Scales
• Multi-Scale Deep Neural Network (MscaleDNN) Methods for Oscillatory Stokes Flows in Complex Domains
• Generalized Rough Polyharmonic Splines for Multiscale PDEs with Rough Coefficients

Cited In (2)

• Higher-order multi-scale physics-informed neural network (HOMS-PINN) method and its convergence analysis for solving elastic problems of authentic composite materials
• Loss jump during loss switch in solving PDEs with neural networks

This page was built for publication: Subspace decomposition based DNN algorithm for elliptic type multi-scale PDEs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6162913)