A bridge between quaternionic and complex numerical ranges (Q2273892)

Consider an \(n\times n\) matrix \(A\) with entries in \(\mathbb{F}\), where \(\mathbb{F}\) can be the field of reals \(\mathbb{R}\), the field of complex numbers \(\mathbb{C}\) or the skew field of quaternions \(\mathbb{H}\). The numerical range of \(A\) is defined as \[ W_{\mathbb{F}}(A):=\{x^*Ax:x\in S_{\mathbb{F}^n}\}, \] where \(S_{\mathbb{F}^n}\) denotes the unit sphere of \(\mathbb{F}^n\) and \(x^*\) is the conjugate transpose of \(x\). For \(\mathbb{F}=\mathbb{C}\), the numerical range is always convex (Hausdorff-Toeplitz theorem), while this is known to be false in general in the quaternionic case. The authors prove a sufficient criterion for the quaternionic numerical range \(W_{\mathbb{H}}(A)\) of a \textit{complex} \(n\times n\) matrix \(A\) to be convex and they also show by a counterexample that this condition is not necessary. Furthermore, they prove that the quaternionic numerical range of a \textit{real} matrix is always convex. As an application, they give a complete characterization of the geometric shape of the quaternionic numerical range of real \(2\times 2\) matrices.