English

Galerkin Neural Networks: A Framework for Approximating Variational Equations with Error Control

Machine Learning 2021-06-01 v1 Numerical Analysis Numerical Analysis

Abstract

We present a new approach to using neural networks to approximate the solutions of variational equations, based on the adaptive construction of a sequence of finite-dimensional subspaces whose basis functions are realizations of a sequence of neural networks. The finite-dimensional subspaces are then used to define a standard Galerkin approximation of the variational equation. This approach enjoys a number of advantages, including: the sequential nature of the algorithm offers a systematic approach to enhancing the accuracy of a given approximation; the sequential enhancements provide a useful indicator for the error that can be used as a criterion for terminating the sequential updates; the basic approach is largely oblivious to the nature of the partial differential equation under consideration; and, some basic theoretical results are presented regarding the convergence (or otherwise) of the method which are used to formulate basic guidelines for applying the method.

Keywords

Cite

@article{arxiv.2105.14094,
  title  = {Galerkin Neural Networks: A Framework for Approximating Variational Equations with Error Control},
  author = {Mark Ainsworth and Justin Dong},
  journal= {arXiv preprint arXiv:2105.14094},
  year   = {2021}
}
R2 v1 2026-06-24T02:35:19.047Z