On the existence of Schur-like forms for matrices with symmetry structures (Q2022388)

Let \(H\) be a fixed \(n\times n\) invertible Hermitian matrix. Then we can define a (generally indefinite) inner product \([~,~]\) on \(\mathbb{C}^{n}\) by \([x,y]:=x^{H}Hy,\) where \(x^{H}\) is the Hermitian transpose of \(x\). The \(H\)-adjoint \(A^{(\ast )}\) of \(A\in \mathbb{C}^{n\times n}\) is the unique matrix such that \([Ax,y]=[x,A^{(\ast )}y]\) and equals \(H^{-1}A^{H}H\). In particular, \(A\) is called \(H\)-selfadjoint, \(H\)-skewadjoint or \(H\)-unitary when \(A^{(\ast )}=A,-A\) or \(A^{-1}\), respectively, and these classes of matrices have been largely classified (see, for example, [\textit{I. Gohberg} et al., Indefinite linear algebra and applications. Basel: Birkhäuser (2005; Zbl 1084.15005)]). For example, in the special case where \(n=2m\) and \[J=\left[ \begin{array}{cc} 0 & I_{m} \\ -I_{m} & 0 \end{array} \right], \] a \(J\)-skewadjoint matrix \(A\) is often called a Hamiltonian matrix. It is known that in general the class of \(H\)-normal matrices (\( AA^{(\ast )}=A^{(\ast )}A\)) is wild (see [\textit{I. Gohberg} and \textit{B. Reichstein}, Integral Equations Oper. Theory 13, No. 3, 364--394 (1990; Zbl 0709.47036)]) and we cannot then expect a useful classification of \(H\)-normal matrices. The author studies the smaller class of matrices called polynomially \(H\)-normal matrices (\(A^{(\ast )}\) is a polynomial in \(A\)) which is a subclass of the \(H\)-normal matrices and equal to the latter when \(H=I_{n}\). A canonical form for a polynomially \(H\)-normal matrix \(A\) was given in [the author, Electron. J. Linear Algebra 15, 50--83 (2006; Zbl 1095.15010)]; it depends on both \(A\) and \(H\) and the polynomial \(p(\lambda )\) such that \(A^{(\ast )}=p(A)\). Let \(H\) be an \(n\times n\) unitary matrix which is Hermitian of inertia index \((\pi ,\nu ,0)\). Suppose that \(A\) is polynomially \(H\)-normal with \(A^{(\ast )}=p(A)\). Then \(m:=\left\vert \pi -\nu \right\vert \) is even and we put \(k:= \frac{1}{2}(n-m)\). The main theorem of the present paper (Theorem 2) provides conditions for the existence of an invertible matrix \(S\) such that \[ S^{-1}AS=\left[ \begin{array}{ccc} B_{11} & B_{12} & B_{13} \\ 0 & p(B_{11})^{H} & 0 \\ 0 & B_{32} & B_{33} \end{array} \right] \text{ and }S^{H}HS=\left[ \begin{array}{ccc} 0 & I_{k} & 0 \\ I_{k} & 0 & 0 \\ 0 & 0 & sI_{m} \end{array} \right], \] where \(B_{11}\) is \(k\times k\) upper triangular, \(B_{33}\) is diagonal and \( s=1 \) if \(\pi \geq \nu \) and \(s=-1\) otherwise. One necessary and sufficient condition for such an \(S\) to exist is the existence of an \(A\)-invariant subspace \(\mathcal{V}\subseteq \mathbb{C}^{n}\) of dimension \(k\) such that \( [x,y]=0\) for all \(x,y\in \mathcal{V}\). Another necessary and sufficient condition for \(S\) is expressed in terms of Mehl's canonical form for \(A\) as a polynomially \(H\)-normal matrix, but is too complicated to describe here. It is also shown that if such an \(S\) exists then it can be chosen to be unitary. Theorem 2 extends known results about Hamiltonian matrices [the author, in: Operator theory. In 2 volumes. Basel: Springer. 431--449 (2015; Zbl 1335.47024)].