Square roots of \(H\)-nonnegative matrices (Q2020699)

Consider an indefinite inner product \([x,y]:=(Hx,y)\) on the complex space \( \mathbb{C}^{n}\) where \(H\) is a nonsingular (indefinite) Hermitian matrix and \((~,~)\) is the standard inner product. A matrix \(B\) is called \(H\)-selfadjoint if \([Bx,y]=[x,By]\) for all \(x,y\) (equivalently, \(HB=B^{\ast }H\) ) and \(B\) is \(H\)-nonnegative if in addition \([Bx,x]\geq 0\) for each \(x\in \mathbb{C}^{n}\). In the indefinite case, the theory of square roots of \(H\)-selfadjoint matrices is more complicated than that for the positive definite case because \(B\) may be non-diagonalizable (see, for example, [\textit{Y. Bolshakov} et al., Linear Algebra Appl. 261, 91--141 (1997; Zbl 0881.15013); \textit{C. V. M. van der Mee} et al., Linear Algebra Appl. 302--303, 77--104 (1999; Zbl 0954.15008)]). The authors begin their study of square roots of \(H\)-nonnegative matrices from a known result: all eigenvalues of an \(H\)-nonnegative matrix are real and its Jordan blocks have size \(1\) for each nonzero eigenvalue and size at most \(2\) for a zero eigenvalue [\textit{I. Gohberg} et al., Indefinite linear algebra and applications. Basel: Birkhäuser (2005; Zbl 1084.15005)]. This leads to a complete description of the square roots of an \(H\)-nonnegative matrix \(B\) based on its Jordan decomposition, and a criterion which determines whether or not \(B\) has a square root which is \(H\)-nonnegative.