Title :
Domain decomposition eigenvalue solver for finite element mode computation in air-inlet
Author :
Barka, A. ; Cosnuau, A. ; Roux, F.X.
Author_Institution :
ONERA, Chatillon, France
Abstract :
The method which has been developed solves the Maxwell equations in cylindrical waveguides of any plane shape without internal conducting media. Parallel implementation for 2D electromagnetic mode computation on a distributed memory machine is described. The most expensive task of this problem consists of seeking the smallest eigenvalues (and the associated eigenvectors) of large, sparse and real matrices coming from the finite element discretization. A parallel method is introduced based on Lanczos, QR and iterative inverse power algorithms, but also on a domain-decomposition solver for large and sparse linear systems. The computational times on IPSC860, PARAGON and CRAY-YMP computers are presented.
Keywords :
Maxwell equations; computational complexity; distributed memory systems; eigenvalues and eigenfunctions; electrical engineering; electrical engineering computing; iterative methods; parallel machines; sparse matrices; waveguide theory; 2D electromagnetic mode computation; CRAY-YMP computer; IPSC860 computer; Lanczos algorithm; PARAGON computer; QR algorithm; air-inlet; computational times; distributed memory machine; domain decomposition eigenvalue solver; eigenvectors; finite element discretization; finite element mode computation; iterative inverse power algorithm; large sparse matrices; parallel implementation; parallel method; real matrices; sparse linear systems; Concurrent computing; Distributed computing; Eigenvalues and eigenfunctions; Electromagnetic waveguides; Finite element methods; Maxwell equations; Planar waveguides; Shape; Sparse matrices; Transmission line matrix methods;
Conference_Titel :
Antennas and Propagation Society International Symposium, 1995. AP-S. Digest
Conference_Location :
Newport Beach, CA, USA
Print_ISBN :
0-7803-2719-5
DOI :
10.1109/APS.1995.529963
