Abstract:We introduce the concept of data-driven finite element methods. These are finite-element discretizations of partial differential equations (PDEs) that resolve quantities of interest with striking accuracy, regardless of the underlying mesh size. The methods are obtained within a machine-learning framework during which the parameters defining the method are tuned against available training data. In particular, we use a stable parametric Petrov-Galerkin method that is equivalent to a minimal-residual formulation using a weighted norm. While the trial space is a standard finite element space, the test space has parameters that are tuned in an off-line stage. Finding the optimal test space therefore amounts to obtaining a goal-oriented discretization that is completely tailored towards the quantity of interest. As is natural in deep learning, we use an artificial neural network to define the parametric family of test spaces. Using numerical examples for the Laplacian and advection equation in one and two dimensions, we demonstrate that the data-driven finite element method has superior approximation of quantities of interest even on very coarse meshes

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to analyze the Quantity of Interest (QoI) with high precision when solving partial differential equations (PDEs) by the finite element method (FEM) on a very coarse grid. Specifically, the authors introduced a data - driven finite element method. This method can significantly improve the approximation accuracy of a specific QoI without relying on grid refinement. The core of this method lies in using a machine - learning framework to adjust the parameters in the test space, thereby achieving goal - oriented discretization, making the method fully optimized for a specific QoI. ### Main Contributions 1. **Data - Driven Acceleration Framework**: Extended the data - driven acceleration framework of the finite - difference scheme for ordinary differential equations (ODEs) and partial differential equations (PDEs) proposed by Mishra and applied it to the finite element method. 2. **Parametric Petrov - Galerkin Method**: Proposed a parametric Petrov - Galerkin method, in which the trial space is fixed on a given grid, while the test space has trainable parameters, which are determined through an offline training process. 3. **Stability Analysis**: Proved the stability of the parametric Petrov - Galerkin method through the equivalent minimum - residual formula and the discrete dual norm. 4. **Application of Neural Networks**: Used artificial neural networks to define the weight function of the test space, and trained the parameters in the neural network by minimizing the loss function, thereby optimizing the test space. ### Method Overview - **Abstract Problem**: Consider a continuous (or infinite - dimensional) problem. The goal is to find a solution \( u_\lambda \) such that \( Bu_\lambda=\ell_\lambda \), where \( B \) is a bounded invertible linear operator, \( \ell_\lambda \) is the right - hand - side functional, and \( q \) is the quantity of interest. - **Main Idea of the Acceleration Method**: Assume that the space \( V \) can be endowed with a set of equivalent weighted inner products. By choosing an appropriate weight \( \omega \), find a discrete solution \( u_{h,\lambda,\omega} \) on a coarse grid such that the error on a specific QoI is minimized. - **Analysis of the Discrete Method**: Proved the well - posedness and stability of the discrete method through the equivalent Petrov - Galerkin formula and the minimum - residual formula. ### Experimental Results - **Numerical Experiments**: Through numerical experiments on one - dimensional and two - dimensional diffusion equations and convection equations, demonstrated the significant improvement of the data - driven finite element method on QoI on very coarse grids. - **Specific Example**: For example, in the one - dimensional diffusion problem, by choosing an appropriate weight function, the relative error of QoI can be significantly reduced even on a very coarse grid. ### Conclusion This paper successfully achieved a high - precision approximation of a specific QoI on a very coarse grid by introducing the data - driven finite element method. This method not only improves computational efficiency but also provides a new and effective tool for solving practical engineering and scientific problems.

Data-Driven Finite Elements Methods: Machine Learning Acceleration of Goal-Oriented Computations

Computational Fluid Dynamics Iteration Driven by Data

A data-driven learned discretization approach in finite volume schemes for hyperbolic conservation laws and varying boundary conditions

Physics-Based Deep Learning for Flow Problems

Element learning: a systematic approach of accelerating finite element-type methods via machine learning, with applications to radiative transfer

An energy approach to the solution of partial differential equations in computational mechanics via machine learning: Concepts, implementation and applications

Linking Machine Learning with Multiscale Numerics: Data-Driven Discovery of Homogenized Equations

Simulation of hyperelastic materials in real-time using Deep Learning

Recovering missing CFD data for high-order discretizations using deep neural networks and dynamics learning

A machine learning framework for data driven acceleration of computations of differential equations

A Machine Learning approach to enhance the SUPG stabilization method for advection-dominated differential problems

Reducing Spatial Discretization Error on Coarse CFD Simulations Using an OpenFOAM-Embedded Deep Learning Framework

Deep-learning of parametric partial differential equations from sparse and noisy data

Learning data-driven discretizations for partial differential equations

DeepFEM: A Novel Element-Based Deep Learning Approach for Solving Nonlinear Partial Differential Equations in Computational Solid Mechanics

Adaptive Finite Element Interpolated Neural Networks

Machine learning accelerated computational fluid dynamics

Meta-PDE: Learning to Solve PDEs Quickly Without a Mesh

Data-Driven Computing Methods for Nonlinear Physics Systems with Geometric Constraints

The Deep Learning Galerkin Method for the General Stokes Equations

Learning a Neural Solver for Parametric PDE to Enhance Physics-Informed Methods