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

We present a new approach to using neural networks to approximate 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 can be used to define a standard Galerkin approximation of the variational equation. This approach enjoys advantages including the following: 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 to some extent 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.