On skew-Hamiltonian matrices and their Krylov-Lagrangian subspaces (Q2228485)

When the Krylov space \(\mathcal{K}_j(H,x)\) for a skew-Hamiltonian \(H\in {\mathbb R}^{2n\times 2n}\) and some \(x\in {\mathbb R}^{2n}\) is isotropic for any \(j\in {\mathbb N}\), and its dimension is \(n\), \(\mathcal{K}_j(H,x)\) is called a Lagrangian subspace. The authors give a positive answer to the question whether every isotropic subspace is a Krylov space of some skew-Hamiltonian matrix. In the study, they focus on Lagrangian subspaces. The set of skew-Hamiltonian matrices \(H\in {\mathbb R}^{2n\times 2n}\) realizing a Lagrangian subspace \(\mathcal{L} \subset {\mathbb R}^{2n}\) with a basis \(\mathbf{B}=(x_1,\dots,x_n)\) as a Krylov subspace is denoted either by \({\mathbb H}{\mathbb K}(x_1,\dots, x_n)\) or \({\mathbb H}{\mathbb K}(\mathbf{B})\). Let \(\Lambda=\{\lambda_1,\dots,\lambda_n\}\subset {\mathbb C}\) be a given set of \(n\) complex numbers closed under conjugation, and \(p(x):=\Pi_{k=1}^{n}(x-\lambda_k)=x^n+a_{n-1}x^{n-1}+\cdots+a_1x+a_0\in {\mathbb R}[x].\) Then the matrix \[C=\begin{bmatrix}0 &\cdots &0&-a_0\\ a& &\vdots &-a_1\\ & \ddots &0& \vdots\\ 0& & 1& -a_{n-1} \end{bmatrix}\in {\mathbb R}^{n\times n} \] is called the companion matrix of \(p(x)\). Let \(A^{+}\) be the Moore-Penrose pseudoinverse of \(A\in {\mathbb R}^{m\times n}\), \(S\in {\mathbb R}^{(n+1)\times (n+1)}\) be some skew-symmetric matrix, and \[J_{2n}=\begin{bmatrix}0& I_n\\ -I_n & 0 \end{bmatrix}\in {\mathbb R}^{2n\times 2n}. \] Given some Lagrangian subspace \(\mathcal{L} \subset {\mathbb R}^{2n}\) with a basis \(\mathbf{B}=(x_1,...,x_n)\), the authors construct a particular skew-Hamiltonian matrix \[\hat{H}=X_R X_L^{+}+J_{2n}^{T}(X_RX_L^{+})^T J_{2n}\in {\mathbb R}^{2n\times 2n}\] such that \[\hat{H}\in {\mathbb H}{\mathbb K}(\mathbf{B}),\] where \(X_L=\begin{bmatrix} x_1& x_2& \cdots& x_{n-1} \end{bmatrix}\in {\mathbb R}^{2n\times (n-1)}\) and \(X_R=\begin{bmatrix} x_2& x_3& \cdots& x_{n} \end{bmatrix}\in {\mathbb R}^{2n\times (n-1)}\). Then they give a characterization of \({\mathbb H}{\mathbb K}(\mathbf{B})\): \(H\in {\mathbb H}{\mathbb K}(\mathbf{B})\) if and only if \[H=\hat{H}+J_{2n}^{T}X_L^{\bot}S(X_L^{\bot})^{T},\] where \(X_L^{\bot}\in {\mathbb R}^{2n\times (n+1)}\) is a matrix whose columns are an orthonormal basis of \(\mbox{span} \{x_1,\cdots,x_{n-1}\}^{\bot}\). Moreover, they prove that for any \(H\in {\mathbb H}{\mathbb K}(\mathbf{B})\), \[\|\hat{H}\|_{F}\leq \|H\|_{F} \mbox{ and } \|\hat{H}\|_{2}\leq \|H\|_{2},\] where \(\|\cdot\|_F\) denotes the Frobenius norm and \(\|\cdot\|_2\) denotes the spectral norm. They identify special elements in \({\mathbb H}{\mathbb K}(\mathbf{B})\), that is, any skew-Hamiltonian matrix \(H \in {\mathbb H}{\mathbb K}(\mathbf{B})\) such that the eigenvalues of \(H|_{\mathcal{L}}\) are given by \(\Lambda\) can be expressed as \[ H=\widetilde{H}+J_{2n}^{T}X^{\bot}S(X^{\bot})^T,\] where \(\widetilde{H}=XCX^{+}+J_{2n}^{T}(XCX^{+})^{T}J_{2n}\in {\mathbb R}^{2n\times 2n}\), and \(X^{\bot}\in{\mathbb R}^{2n\times n}\) is a matrix whose columns form an orthonormal basis for \(\mathcal{L}^{\bot}\). Finally, given a perturbed skew-Hamiltonian matrix \(A\in {\mathbb R}^{2n\times 2n}\), they apply the above results to find the best approximation of \(A\) by a skew-Hamiltonian matrix.