Recursion relations for the extended Krylov subspace method (Q630517)

Many problems in applied mathematics require the computation of expressions of the form \(w:=f(A)v\), where \(A\in{\mathbb R}^{n\times n}\) is a large, sparse or structured, symmetric matrix, \(f\) is a nonlinear function defined on the spectrum of \(A\), and \(v\in{\mathbb R}^n\). The authors implement rational Lanczos methods for generating orthonormal bases of extended rational Krylov subspaces \({\mathbb K}^{m,mi+1}(A,v)=\text{span}\{A^{-m+1},\dots,A^{-1},v,Av,\dots,A^{mi}v\}\), \(m=1,2,\dots\) with \(i\) a positive integer, with short recursion relations. For all integers \(i\geq 1\), the projected matrix \(H_{m(i+1)}\) is pentadiagonal with a block-tridiagonal structure. Computed examples show that the above Krylov subspaces yield more accurate approximants for \(i>1\) than with \(i=1\) or standard Krylov subspaces \({\mathbb K}^m(A,v)=\text{span}\{v,Av,\dots,A^{m-1}v\}\) of the same dimension.