Electromagnetic devices can be analysed by the coupled BE-FE method, where the conducting and magnetic parts are discretized by finite elements. In contrast, the surrounding space is described with the help of the boundary element method (BEM). This discretization scheme is well suited especially for problems including moving parts (see [12]). The BEM discretization of the boundary integral operators usually leads to dense matrices without any structure. A naive strategy for the solution of the corresponding linear system would need at least OðN 2 Þ operations and memory, where N ist the number of unknowns. Methods such as fast multipole [6] and panel clustering [9] provide an approximation to the matrix in almost linear complexity. These methods are based on explicitly given kernel approximations by degenerate kernels, i.e. a finite sum of separable functions, which may be seen as a blockwise low-rank approximation of the system matrix. The blockwise approximant permits a fast matrix-vector multiplication, which can be exploited in iterative solvers, and can be stored efficiently. In contrast to the methods mentioned we will generate [2] the low-rank approximant from the matrix itself using only few entries and without using any explicit a priori known degenerate-kernel approximation. Special emphasis is put on the handling of symmetry conditions in connection with ACA. The feasibility of the proposed method is demonstrated by means of a numerical example.
IntroductionThe application of the boundary element method for the solution of linear electromagnetic problems has many advantages. Only the boundaries of the considered domains need to be discretized, open boundary problems pose no additional difficulties, and problems including motion can be treated elegantly. The BE-FE coupling procedure is elaborated in [12]. However, application of the BEM leads to dense matrices. The storage requirements and computational costs are of OðN 2 Þ, where N is the number of unknowns. One remedy could be the use of an approximation algorithm with almost linear complexity. Section 2 presents the adaptive cross approximation (ACA) algorithm [13] which generates blockwise low rank approximants for the BEM matrices without using an explicit kernel expansion. The exploitation of symmetry is another possibility to reduce computational costs and has been presented in [1, 4, 5] using linear representation theory for finite groups. The aim is a decomposition of function spaces into orthogonal subspaces of symmetric functions, such that each subproblem is defined on a so called symmetry cell. The global solution can then be reconstructed from these components. Sections 3.1-3.2 give an overview of using symmetry in FE and BE methods. Next, in Sect. 3.3 the major contribution of this paper, a symmetry-exploiting ACA algorithm is discussed, which not only reduces the problem size due to symmetry but also yields additional benefit by compression of BEM matrices and possesses an almost linear complexity w.r.t. the number of unknowns. Numerica...