Efficient parallel solutions to 3D electromagnetic problems using potentials

Abstract

We present a massively parallel finite-element method for 3D electromagnetic forward modelling using vector and scalar potentials. The numerical scheme employs parallel Krylov subspace solvers with efficient preconditioning and hence displays good performance in terms of convergence and scalability for large parallel applications. The method supports completely unstructured tetrahedral meshes and electric anisotropy. A realistic synthetic test case demonstrates the necessity of accurate representation of bathymetry and complex subsurface structures. Another focus of this study is to investigate the performance of different approximate inverse-based preconditioners, which are well suited for parallel implementation, but can lack efficiency when applied without regard for specific matrix properties.