Rational Krylov approximation of matrix functions: numerical methods and optimal pole selection (Q2864805)

The paper provides a review on the numerical computation of \(f(A){\mathbf b}\) by rational Krylov methods, where \(A\) is a square matrix, possibly large and sparse or structured, \({\mathbf b}\) is a vector and \(f(A)\) is a matrix function. Most of the material is taken from the doctoral dissertation of the author, with updates covering recent developments. Important rational Krylov methods such as the rational Arnoldi method, the extended Krylov subspace method, the shift-and-invert Arnoldi method and the generalized Leja point method are revisited. Strategies for optimal or near-optimal pole selection are discussed, with emphasis on some particular matrix functions: the resolvent, exponential and functions of Markov type.