Krylov subspaces associated with higher-order linear dynamical systems (Q2583154): Difference between revisions

Let \(i=1,2,\dots,\ell\). In case I, the author introduces the matrices \({\mathcal M}=(c\otimes I_{n_0})[M^{(1)}\;M^{(2)}\dots M^{(\ell)}]+ \Sigma\otimes I_{n_0}\in{\mathbb C}^{N\times N}\) and \({\mathcal R}=c\otimes R\in{\mathbb C}^{N\times m}\) where \(M^{(i)}\in {\mathbb C}^{n_0\times n_0}\), \(R\in{\mathbb C}^{n_0\times m}\), \(c=(c_i\neq0)\in{\mathbb C}^\ell\) and \(\Sigma=(\sigma_{ij})\in{\mathbb C}^{\ell\times\ell}\). The \(N\times mN\) matrix \(K=[{\mathcal R}\;{\mathcal M}{\mathcal R}\;{\mathcal M}^2{\mathcal R}\;\cdots{\mathcal M}^{N-1}{\mathcal R}]\) is called the block-Krylov matrix induced by \({\mathcal M}\) and \({\mathcal R}\). Under the above definitions, Theorem~3.1 states that if \({\mathcal V}\in{\mathbb C}^{N\times N_0}\) is a basis matrix of \(K\), then \({\mathcal V}=(WU^{(i)})\), where \(W\in{\mathbb C}^{n_0\times N_0}\) and, for each \(i\), \(U^{(i)}\) is nonsingular and upper triangular. In Case II, \({\mathcal M}=[C^{(1)}\; \cdots\; C^{(\ell)}]^T[M^{(1)}\;\cdots\; M^{(\ell)}]+ \text{ diagonal}[\sigma_1 I_{n_1}\;\cdots\; \sigma_\ell I_{n_\ell}]\in{\mathbb C}^{N\times N}\) and \({\mathcal R}= [C^{(1)}\; \cdots\; C^{(\ell)}]^T R\in{\mathbb C}^{N\times m}\), where \(c^{(i)}\in{\mathbb C}^{n_1\times n_0}\), \(M^{(i)}\in{\mathbb C}^{n_0\times n_i}\), \(\sigma_i\in{\mathbb C}\), and \(R\in{\mathbb C}^{n_0\times m}\). Theorem~3.2 states that the basis of \(K\) is \({\mathcal V}=(C^{(i)}WU^{(i)})\) with \(W\) and \(U^{(i)}\) as in the previous theorem. It follows from these theorems that the associated structured block-Krylov subspaces consist of multiple copies of certain subspaces of the state space of the original higher-order system.