An Extremal Formula for Elliptical Width and the Bourin--Lee Conjecture

Abstract

Let \(X\in M_n(\mathbb C)\), \(n\geq2\), and let \[ δ_2(X) = \sup_{\dim S=2}δ(X_S), \] where \(X_S\) denotes the compression of \(X\) to \(S\), and \(δ(X_S)\) is the diameter of the largest disk contained in the numerical range \(W(X_S)\). Bourin and Lee proved that every positive block matrix \[ M= \begin{pmatrix} A&X X^*&B \end{pmatrix} \] satisfies \[ \norm{M}\leq\norm{A+B}+δ_2(X). \] In the same work, they conjectured that the inequality \[ \norm{M}\leq\norm{A+B} \] for every positive block matrix with prescribed off-diagonal block \(X\) characterizes essentially Hermitian matrices. We resolve this conjecture by proving the stronger exact formula \[ \sup_{\left(\begin{smallmatrix}A&X\\X^*&B\end{smallmatrix}\right)\geq0} \left\{ \norm{\begin{pmatrix}A&X\\X^*&B\end{pmatrix}} -\norm{A+B} \right\} = δ_2(X). \] Thus the elliptical width is precisely the optimal norm defect associated with a fixed off-diagonal block. The reverse inequality is obtained by lifting an arbitrary two-dimensional compression of \(X\) to an explicit family of positive block matrices, yielding quantitative two-sided estimates with error of order \(t^{-1}\). As a consequence, the norm inequality above holds universally if and only if \(X\) is essentially Hermitian, proving Conjecture 3.3 of Bourin and Lee. We also construct an explicit \(3\times3\) normal example showing that the radius constant in their normal-off-diagonal eigenvalue estimate is optimal at the leading eigenvalue level. By direct-sum amplification of this construction, we further prove that the same radius constant is sharp at every eigenvalue level \(j\geq0\), thereby resolving their corresponding sharpness question.