English

The global extended-rational Arnoldi method for matrix function approximation

Numerical Analysis 2020-04-02 v1 Numerical Analysis

Abstract

The numerical computation of matrix functions such as f(A)Vf(A)V, where AA is an n×nn\times n large and sparse square matrix, VV is an n×pn \times p block with pnp\ll n and ff is a nonlinear matrix function, arises in various applications such as network analysis (f(t)=exp(t)f(t)=exp(t) or f(t)=t3)f(t)=t^3), machine learning (f(t)=log(t))(f(t)=log(t)), theory of quantum chromodynamics (f(t)=t1/2)(f(t)=t^{1/2}), electronic structure computation, and others. In this work, we propose the use of global extended-rational Arnoldi method for computing approximations of such expressions. The derived method projects the initial problem onto an global extended-rational Krylov subspace RKme(A,V)=span({i=1m(AsiIn)1V,,(As1In)1V,V\mathcal{RK}^{e}_m(A,V)=\text{span}(\{\prod\limits_{i=1}^m(A-s_iI_n)^{-1}V,\ldots,(A-s_1I_n)^{-1}V,V ,AV,,Am1V}),AV, \ldots,A^{m-1}V\}) of a low dimension. An adaptive procedure for the selection of shift parameters {s1,,sm}\{s_1,\ldots,s_m\} is given. The proposed method is also applied to solve parameter dependent systems. Numerical examples are presented to show the performance of the global extended-rational Arnoldi for these problems.

Keywords

Cite

@article{arxiv.2004.00059,
  title  = {The global extended-rational Arnoldi method for matrix function approximation},
  author = {A. H. Bentbib and M. El Ghomari and K. Jbilou},
  journal= {arXiv preprint arXiv:2004.00059},
  year   = {2020}
}