A Generalized SOR Method for Dense Linear Systems of Boundary Element Equations
boundary element equationselastostaticsnumerical examplespotential equationreorderingSORsuccessive overrelaxation
Integral representations, integral operators, integral equations methods in higher dimensions (31B10) Laplace operator, Helmholtz equation (reduced wave equation), Poisson equation (35J05) Iterative numerical methods for linear systems (65F10) Boundary element methods for boundary value problems involving PDEs (65N38) Classical linear elasticity (74B05) Boundary element methods applied to problems in solid mechanics (74S15)
The authors examine an iterative scheme that extends the successive overrelaxation (SOR) method, in order to numerically solve systems of boundary element equations associated to three-dimensional potential and elastostatic problems. Such systems arise in the discretization of the boundary integral formulation of those differential problems. Thus, in order to solve \[ Hx=c, \] where it is supposed that no term vanishes in the diagonal \(D\) of the square matrix \(H\) of order \(n\), the following iterative scheme is proposed: \[ (D+(\Omega H)_{sl})x^{(k+1)}=(D-(\Omega H)_u)x^{(k)}+\Omega c . \] Here \(\Omega \) stands for an upper triangular matrix to be determined whereas the subindices \(sl\) and \(u\) stand for strictly lower and upper, respectively. If \(\Omega = \omega I\), the scheme becomes the standard SOR. It is shown then that it is possible to determine \(\Omega _0\) such that \(D- (\Omega _0 H)_u =0\), provided that the principal square matrices \(H_r\), \(1\leq r \leq n \) are non singular; \(H_r\) is obtained by deleting the first \(r-1\) rows and columns of \(H\). However, since the exact determination of the unique \(\Omega _0\) implies a far higher computational cost than the one involved in solving the original system by means of \(UL\) decomposition, the authors propose two different families of sparse approximants to \(\Omega _0\). This is combined with a heuristic reordering of row and columns in \(H\). The numerical results show that the proposed families yield convergence in most cases and they do so with a significant reduction in computing time, when compared with Gaussian elimination. In one of the cases the reduction is by a factor of \(20\). This suggests that the scheme could be a first choice if the reduction in computing time is one of the issues when solving problems as those treated in the paper.
This page was built for publication: A Generalized SOR Method for Dense Linear Systems of Boundary Element Equations
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4389278)
