• DocumentCode
    2644188
  • Title

    A state-of-the-art linear solver IDR(s) method for large scale electromagnetic multiple scattering simulations

  • Author

    Nakashima, N. ; Fujin, S. ; Tateiba, M. ; Onoue, Y.

  • Author_Institution
    Kyushu Univ., Fukuoka, Japan
  • fYear
    2009
  • fDate
    1-5 June 2009
  • Firstpage
    1
  • Lastpage
    4
  • Abstract
    This paper remarks IDR(s) method as an alternative solver for GMRES(m) method. A preconditioned IDR(s) algorithm is presented. Performance evaluations are done for the computation of the dense linear system of equations of order 105 followed by BEM analysis of EM multiple scattering. IDR(s) method with the optimal parameters (5 < s < 10) converges faster than GMRES(mmax) method, using small amount of memory rather than GMRES(50) method. Moreover, as the problem scale becomes larger, the bigger difference of convergence between IDR(s) method and GMRES (mmax) happens. We conclude therefore that IDR(s) method is better than the GMRES method with large restart cycles in terms of not only memory efficiency but also convergence We use the additional operations, but the accuracy fails for large s. The modifications of them may be an important subject.
  • Keywords
    boundary-elements methods; computational electromagnetics; electric field integral equations; electromagnetic wave scattering; BEM analysis; boundary element method; electrical field integral equations; generalized minimal residual method; large scale electromagnetic multiple scattering simulation; memory efficiency; state-of-the-art linear solver IDR method; Boundary element methods; Computational modeling; Convergence; Dielectrics; Electromagnetic scattering; Integral equations; Iterative methods; Large-scale systems; Linear systems; Shape;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Antennas and Propagation Society International Symposium, 2009. APSURSI '09. IEEE
  • Conference_Location
    Charleston, SC
  • ISSN
    1522-3965
  • Print_ISBN
    978-1-4244-3647-7
  • Type

    conf

  • DOI
    10.1109/APS.2009.5171533
  • Filename
    5171533