The lowest-order Neural Approximated Virtual Element Method on polygonal elements

Stefano Berrone,Moreno Pintore,Gioana Teora

2024-09-24

Abstract:The lowest-order Neural Approximated Virtual Element Method on polygonal elements is proposed here. This method employs a neural network to locally approximate the Virtual Element basis functions, thereby eliminating issues concerning stabilization and projection operators, which are the key components of the standard Virtual Element Method. We propose different training strategies for the neural network training, each correlated by the theoretical justification and with a different level of accuracy. Several numerical experiments are proposed to validate our procedure on general polygonal meshes and demonstrate the advantages of the proposed method across different problem formulations, particularly in cases where the heavy usage of projection and stabilization terms may represent challenges for the standard version of the method. Particular attention is reserved to triangular meshes with hanging nodes which assume a central role in many virtual element applications.

Numerical Analysis

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are the problems of stability and projection operators in the traditional Virtual Element Method (VEM). Specifically, the standard Virtual Element Method relies on stabilization terms and projection operators to handle non - polynomial basis functions, which may lead to a decrease in accuracy in strongly anisotropic problems and introduce many challenges in complex nonlinear problems.

To solve these problems, the author proposes a new method - the Neural - network - Approximated Virtual Element Method of the lowest - order (NA VEM), which uses neural networks to locally approximate virtual element basis functions. Through this method, problems related to stabilization terms and projection operators can be eliminated, thereby improving the accuracy and stability of the method. In addition, this method pays special attention to the hanging - node problems in triangular meshes, because these problems play a central role in many virtual element applications.

### Main improvement points

1. **Elimination of stabilization terms**: The traditional Virtual Element Method needs to introduce stabilization terms to ensure numerical stability, but the selection of these stabilization terms is highly dependent on specific problems and may affect the accuracy of the method. NA VEM avoids the need for stabilization terms by directly approximating basis functions using neural networks.

2. **Simplification of projection operations**: In the traditional method, polynomial projection operators need to be introduced to calculate the point - by - point values of basis functions, which may bring limitations in the post - processing stage. NA VEM reduces the dependence on projection operators by approximating basis functions with neural networks, thus simplifying the calculation process.

3. **Handling of hanging nodes**: For the hanging - node problems in triangular meshes, NA VEM proposes specific coding and training strategies to ensure that high - accuracy results can also be obtained in these cases. In particular, by converting triangular elements into equilateral triangles and rotating them, hanging nodes in different configurations can be effectively handled.

### Numerical experiment verification

To verify the effectiveness of the proposed method, the author conducted multiple numerical experiments, including tests on general polygonal meshes and applications for highly nonlinear problems. The experimental results show that NA VEM performs well in various problem settings, especially when dealing with strongly anisotropic problems, its advantages are particularly obvious.

In conclusion, this paper aims to overcome the limitations of the traditional Virtual Element Method in terms of stability and projection operations by introducing a method of approximating virtual element basis functions with neural networks, thereby providing a more efficient and more accurate numerical solution scheme.