Modern methods for the iterative computation of eigenpairs of matrices of high dimension (Q4812403)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 2093908
Language Label Description Also known as
default for all languages
No label defined
    English
    Modern methods for the iterative computation of eigenpairs of matrices of high dimension
    scientific article; zbMATH DE number 2093908

      Statements

      Modern methods for the iterative computation of eigenpairs of matrices of high dimension (English)
      0 references
      0 references
      20 August 2004
      0 references
      eigenvalue problem
      0 references
      Krylov subspace method
      0 references
      Jacobi-Davidson method
      0 references
      numerical examples
      0 references
      Arnoldi method
      0 references
      quadratic eigenproblems
      0 references
      iterative methods
      0 references
      algorithms
      0 references
      Newton type methods
      0 references
      0 references
      0 references
      The author gives a short review on the main methods of numerical linear algebra used in the computation of eigenvalues and eigenvectors of large matrices. In general these problems arise from the discretization of differential equations where the order of the matrix could reach from few thousands up to 640,000. This makes standard direct methods inapplicable and thus the need for iterative methods. The author points out that the QR and QZ algorithms are derived from the power method and are in general expensive. To overcome the difficulty caused by the size, the strategy is to construct an appropriate subspace with which the sought eigenvector makes a small angle and then solve the eigenvalue problem restricted to these subspaces. This is the fundamental idea used in the Krylov subspace method and its alternatives such as the Jacobi-Davidson method. The author wraps up his survey by unifying these algorithms under Newton type methods. At the end he gives specific examples of the computing time it takes for the CRAY T3d processor to solve a quadratic eigenvalue problem when the size is 274,625.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references