Abstract:The use of neural networks to approximate partial differential equations (PDEs) has gained significant attention in recent years. However, the approximation of PDEs with localised phenomena, e.g., sharp gradients and singularities, remains a challenge, due to ill-defined cost functions in terms of pointwise residual sampling or poor numerical integration. In this work, we introduce $h$-adaptive finite element interpolated neural networks. The method relies on the interpolation of a neural network onto a finite element space that is gradually adapted to the solution during the training process to equidistribute a posteriori error indicator. The use of adaptive interpolation is essential in preserving the non-linear approximation capabilities of the neural networks to effectively tackle problems with localised features. The training relies on a gradient-based optimisation of a loss function based on the (dual) norm of the finite element residual of the interpolated neural network. Automatic mesh adaptation (i.e., refinement and coarsening) is performed based on a posteriori error indicators till a certain level of accuracy is reached. The proposed methodology can be applied to indefinite and nonsymmetric problems. We carry out a detailed numerical analysis of the scheme and prove several a priori error estimates, depending on the expressiveness of the neural network compared to the interpolation mesh. Our numerical experiments confirm the effectiveness of the method in capturing sharp gradients and singularities for forward PDE problems, both in 2D and 3D scenarios. We also show that the proposed preconditioning strategy (i.e., using a dual residual norm of the residual as a cost function) enhances training robustness and accelerates convergence.

Interpolation in Jacobi-weighted spaces and its application to a posteriori error estimations of the p-version of the finite element method

Raviart-Thomas interpolation in fractional weighted Sobolev spaces

A posteriori error analysis of a Banach spaces-based fully mixed FEM for double-diffusive convection in a fluid-saturated porous medium

Adaptive Finite Element Interpolated Neural Networks

Quasi-interpolators with application to postprocessing in finite element methods

A-posteriori QMC-FEM error estimation for Bayesian inversion and optimal control with entropic risk measure

A posteriori error estimates for the Brinkman-Darcy-Forchheimer problem

A posteriori estimates for errors of functionals on finite volume approximations to solutions of elliptic boundary value problems

A posteriori error estimates for stabilised mixed finite element methods for a nonlinear elliptic problem

A posteriori error estimator for elliptic interface problems in the fictitious formulation

Residual-based a Posteriori Error Estimation for Elliptic Interface Problems Approximated by Immersed Finite Element Methods

A Posteriori Error Estimates for Hp Finite Element Solutions of Convex Optimal Control Problems

Balancing Discretization and Iteration Error in Finite Element A Posteriori Error Analysis

A-posteriori error estimates for systems of hyperbolic conservation laws

A posteriori error analysis for a continuous space-time finite element method for a hyperbolic integro-differential equation

Optimal $L^\infty(L^2)$ and $L^1(L^2)$ a posteriori error estimates for the fully discrete approximations of time fractional parabolic differential equations

A Posteriori Error Estimates for Finite Element Approximation of Unsteady Incompressible Stochastic Navier-Stokes Equations

A posteriori error estimates and space--time adaptivity for parabolic partial differential equations on stationary surfaces

An accurate a posteriori error estimator for semilinear Neumann problem and its applications.

An improved a posteriori error estimate for the Galerkin spectral method in one dimension

An interpolation result for $A_1$ weights with applications to fractional Poincaré inequalities