A note on convexity of sections of quaternionic numerical range (Q2419024)

In analogy to the usual numerical range of linear operators on a Hilbert space, the numerical range for \(n\times n\) quaternionic matrices is defined as \(W_{\mathbb H}(A) = \{ \langle AX, X\rangle : X\in \mathbb H^n, \|X\| = 1 \}\). By the well-known Toeplitz-Hausdorff theorem, the numerical range of a complex matrix is convex. However, this is not true in the quaternionic case and the author gives examples of \(A\in M(n\times n, \mathbb H)\) whose numerical range is not convex. He proves that for every imaginary unit \(m\) the section \(W_{\mathbb H}(A)\cap \mathbb C^+_m\) is convex, where \(\mathbb C_m^+ = \{ \alpha + \beta m : \alpha\in\mathbb R, \beta > 0\}\). He also shows that \(W_{\mathbb H}(A)\) is convex if and only if its projection on \(\mathbb C\) coincides with its intersection with \(\mathbb C\). As a corollary it follows that the numerical range of a selfadjoint quaternionic matrix is convex. Finally, the author gives upper and lower bounds for the numerical radius \(w_{\mathbb H}(A) := \sup \{ |q| : q\in W_{\mathbb H}(A)\}\). In particular he shows that \(w_{\mathbb H}\) is a norm on \(M(n\times n,\mathbb H)\) which is equivalent to the operator norm.