We describe a parallel software tool intended for the quantification of the sensitivity of eigenvalues to matrix perturbations of varying size: the spectral portrait shows a representation of the normwise epsilon-pseudospectra of a matrix A. This requires to compute || (A - zI)~(-1) ||_2 for all z in a discretized region of the complex plane. The method used for computing || (A - zI)~(-1) ||_2 has to be reliable and optimized in terms of both the computation time and the storage requirements in order to treat large sparse matrices. The main features of the code are described. Different parallel computation strategies are compared in terms of speed-up. We will show an application on the analysis of an iterative scheme used for the solutionof an electromagnetism problem.
展开▼
