Nonlinear reduced-order modeling with a Graph Convolutional Autoencoder for time-domain electromagnetics
Résumé
In the present work, we provide insights into the realm of computational electromagnetics, with a particular focus on time-domain electromagnetism. Numerical modeling plays a crucial role in revealing the behavior of light and matter interactions at the nanoscale, exploiting computational schemes, such as Finite-Differences Time-Domain and, as in our case, Discontinuous Galerkin methods. Since the choice of the basis elements is fundamental to enhance particularly interesting features, in the following we will consider nodal basis, thus leading to the Nodal discontinuous Galerkin. Furthermore, we will introduce Reduced Order Modeling (ROM) strategies, as a consequence of the pressing need for more efficient and accurate models capable of handling parameterized electromagnetic problems. Traditional ROM techniques like Proper Orthogonal Decomposition (POD) and the Greedy algorithm have already been investigated in the literature, along with their inherent limitations in effectively capturing nonlinear phenomena. Indeed, subsequently, we will introduce a particular deep learning-based ROM, the Graph Convolutional Autoencoder (GCA) method. The GCA method serves as a nonlinear extension of POD compression, harnessing the power of Graph Neural Networks (GNNs) to retain geometric structures within unstructured meshes.
Origine | Fichiers produits par l'(les) auteur(s) |
---|