Krylov subspace methods for eigenvalues with special properties and their analysis for normal matrices (Q1808929): Difference between revisions

The author considers a general approach to the problem of finding the eigenvalues \( \mu _i\) of a matrix \(A\) that satisfy \[ \mid \psi (\mu _1)\mid \geq \mid \psi (\mu _2)\mid \geq ... \] where \(\psi \) is a scalar function such that the eigenvalues of \(A\) satisfying some special property also maximize \(\mid \psi \mid \). For instance, if the property is to be largest in modulus, then \(\psi \) is the identity function, while if it is to have largest real part, then the exponential function is the corresponding function. An important ingredient of the approach to the problem is given by the use of any one among the Arnoldi's method, the Lanczos method and the simultaneous iteration method, which produces Ritz values that approximate the eigenvalues of \(A\) that are largest in modulus. The paper gives a general description as to how these methods can be applied to achieve the aim stated above and provides with a thorough convergence analysis for the case in which \(A\) is normal. This is also illustrated with numerical examples. The main conclusions regarding the case analysed in the paper are that if the matrix has only simple eigenvalues, then all three methods produce the same rates of convergence, while if the matrix has eigenvalues of multiplicity \(2\) or more, then Arnoldi and Lanczos methods have better convergence properties.