Geometry-aware domain decomposition preconditioning for hybrid finite element-boundary integral method
Hong-Wei Gao,Zhen Peng,Xin-Qing Sheng
DOI: https://doi.org/10.1109/PIERS.2016.7735732
2016-01-01
Abstract:Summary form only given. The hybrid finite elements-boundary integral (FE-BI) method has been one of the most widely used numerical methods for the full wave solution of time-harmonic electromagnetic (EM) problems. The method first divides the computational domain into interior and exterior regions. The EM fields in the interior region are discretized using the finite-element method (FEM) and the fields in the unbounded exterior region are formulated by the boundary integral equation (BIE) method. The fields in interior and exterior regions are then coupled by the continuity conditions. Therefore, the method preserves FEM's flexibility in modeling complex geometrical features and spatially varying material properties, and also inherits the advantage of BIE for the solution of unbounded homogeneous regions. On the other hand, application of the FE-BI method to time-harmonic Maxwell Equations leads to a partially sparse, partially dense, complex and indefinite matrix equation. A variety of matrix compression techniques have been employed to speed up the dense BIE matrix-vector multiplication. Nevertheless, for many problems of practical interest, the use of Krylov subspace iterative methods exhibits very slow convergence. This has been the main motivation for this work. We proposed a novel geometry-based domain decomposition (DD) preconditioning technique for the iterative solution of hybrid FE-BI formulation. The technique ingredients include a volume-based Schwarz FE DD method and a surface-based interior penalty BIE DD method. Comparing to previously algorithms, the work has two major benefits: (i) it results in a robust and cost-effective preconditioning technique for the solution of the FE-BI linear system of equations; (ii) it provides a much more flexible and natural way to set up the mathematical models, to create the problem geometries and to discretize the computational domain. We will outline the key technical aspects of the new formulation, give the details of the numerical analysis, and report numerical experiments verifying the analysis and display the capabilities of the proposed methods.