Abstract:We extend the finite element interpolated neural network (FEINN) framework from partial differential equations (PDEs) with weak solutions in $H^1$ to PDEs with weak solutions in $H(\textbf{curl})$ or $H(\textbf{div})$. To this end, we consider interpolation trial spaces that satisfy the de Rham Hilbert subcomplex, providing stable and structure-preserving neural network discretisations for a wide variety of PDEs. This approach, coined compatible FEINNs, has been used to accurately approximate the $H(\textbf{curl})$ inner product. We numerically observe that the trained network outperforms finite element solutions by several orders of magnitude for smooth analytical solutions. Furthermore, to showcase the versatility of the method, we demonstrate that compatible FEINNs achieve high accuracy in solving surface PDEs such as the Darcy equation on a sphere. Additionally, the framework can integrate adaptive mesh refinements to effectively solve problems with localised features. We use an adaptive training strategy to train the network on a sequence of progressively adapted meshes. Finally, we compare compatible FEINNs with the adjoint neural network method for solving inverse problems. We consider a one-loop algorithm that trains the neural networks for unknowns and missing parameters using a loss function that includes PDE residual and data misfit terms. The algorithm is applied to identify space-varying physical parameters for the $H(\textbf{curl})$ model problem from partial or noisy observations. We find that compatible FEINNs achieve accuracy and robustness comparable to, if not exceeding, the adjoint method in these scenarios.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to extend the Finite Element Interpolation Neural Network (FEINN) framework to deal with partial differential equations (PDEs) whose weak solutions are in $H(\text{curl})$ or $H(\text{div})$ spaces. Specifically, the paper addresses the following problems: 1. **Compatibility and Structure Preservation**: - The paper introduces interpolation trial spaces that satisfy the de Rham Hilbert sub - complex, thus providing stable, structure - preserving neural network discretization methods for various PDEs. These methods are called Compatible FEINNs. 2. **High - Precision Approximation**: - Compatible FEINNs can accurately approximate the $H(\text{curl})$ inner product, and for smooth analytical solutions, the performance of the trained network is several orders of magnitude higher than that of the finite element solution. 3. **Solving Surface PDEs**: - To demonstrate the generality of the method, the paper shows the high - precision performance of Compatible FEINNs in solving surface PDEs such as the Darcy equation on the sphere. 4. **Adaptive Mesh Refinement**: - This framework can integrate adaptive mesh refinement techniques to effectively solve problems with local characteristics. Through an adaptive training strategy, the network is trained on a series of progressively optimized meshes. 5. **Solving Inverse Problems**: - The paper compares the performance of Compatible FEINNs with the adjoint neural network method in solving inverse problems. A single - loop algorithm is proposed, using a loss function containing PDE residual and data error terms to train the neural network to identify unknown and missing parameters. The results show that Compatible FEINNs achieve comparable or even higher accuracy and robustness in these scenarios. 6. **Theoretical Basis and Numerical Experiments**: - The paper provides detailed mathematical notations and theoretical backgrounds, including continuous problems, finite element discretization, neural network structures, etc. And it verifies the performance of Compatible FEINNs in forward and inverse problems through a series of numerical experiments. ### Summary The main objective of the paper is to extend the existing FEINN framework by introducing the Compatible FEINNs method to better handle PDEs in $H(\text{curl})$ and $H(\text{div})$ spaces and demonstrate its advantages in high - precision approximation, solving surface PDEs, adaptive mesh refinement, and solving inverse problems.

Compatible finite element interpolated neural networks

Adaptive Finite Element Interpolated Neural Networks

Finite element interpolated neural networks for solving forward and inverse problems

Finite Element Neural Network Interpolation. Part I: Interpretable and Adaptive Discretization for Solving PDEs

Non-equispaced Fourier Neural Solvers for PDEs

Deep neural network methods for solving forward and inverse problems of time fractional diffusion equations with conformable derivative

Finite Element Neural Network Interpolation. Part II: Hybridisation with the Proper Generalised Decomposition for non-linear surrogate modelling

A mixed formulation for physics-informed neural networks as a potential solver for engineering problems in heterogeneous domains: Comparison with finite element method

De Rham compatible Deep Neural Network FEM

MRF-PINN: A Multi-Receptive-Field convolutional physics-informed neural network for solving partial differential equations

Error assessment of an adaptive finite elements—neural networks method for an elliptic parametric PDE

HiDeNN-FEM: a seamless machine learning approach to nonlinear finite element analysis

Adaptive Multilevel Neural Networks for Parametric PDEs with Error Estimation

Composing Partial Differential Equations with Physics-Aware Neural Networks

Physical informed neural networks with soft and hard boundary constraints for solving advection-diffusion equations using Fourier expansions

Regularity-Conforming Neural Networks (ReCoNNs) for solving Partial Differential Equations

Finite Operator Learning: Bridging Neural Operators and Numerical Methods for Efficient Parametric Solution and Optimization of PDEs

Hybrid Finite Difference with the Physics-informed Neural Network for solving PDE in complex geometries

Finite element-integrated neural network framework for elastic and elastoplastic solids

Hybrid FEM-NN models: Combining artificial neural networks with the finite element method