An exploratory study of a subspace iteration method as an alternative to the QR method for Floquet eigenanalysis (Q1324703)

The authors consider non-symmetric eigenvalue problems of the form \(Ax= \lambda x\), \(x\in \mathbb{C}^ n\), \(x\neq 0\), \(A\neq A^ T\), preferably arising from Floquet eigenanalysis in the investigation of rotorcraft stability, and compare the subspace iteration method of Arnoldi-Saad (AS) [\textit{Y. Saad}, Linear Algebra Appl. 34, 269-295 (1980; Zbl 0456.65017)] with the generic QR method that is almost exclusively used up to now. The AS-method basically requires the construction of an orthonormal basis \(\{v_ 1,v_ 2,\dots,v_ m\}\) of the Krylov subspace \(\text{span}\{v_ 1, Av_ 1,\dots,A^{m-1} v_ 1\}\) of dimension \(m\) \((m\ll n)\) by the Gram-Schmidt process and the solution of an eigenvalue problem with the Hessenberg matrix \(H_ m= B^ T AB\), with \(B= (v_ 1,v_ 2,\dots,v_ m)\). The AS-method becomes more effective than the QR-method if the eigenanalysis requires the computation of only a few dominant eigenvalues (say \(p\), with \(p<m\)) together with the eigenvector pairs. This is confirmed by the numerical results presented.