On eigenvalues and boundary curvature of the numerical range (Q1595121): Difference between revisions

For an \(n\times n\) matrix \(A\) let \(W(A)\) be its numerical range, and \(\sigma (A)\) be its spectrum. By the theorem of \textit{W. F. Donoghue jun.} [Michigan Math. J. 4, 261-263 (1958; Zbl 0082.11601)], all corner points of \(W(A)\) belong to~\(\sigma (A)\). The authors obtain bounds for the distance \(D_p(A)\) from a point~\(p\) on the boundary \(\partial W(A)\) to \( \sigma (A)\). Let \(M(A)\) be the smallest constant such that \(D_p(A)\leqslant M(A)R_p(A)\) for all \(p\in \partial W(A)\), where \(R_p(A)\) is the radius of curvature of~\(\partial W(A)\), being defined for all but finitely many points, and let \(M_n=\sup \{M(A) : A\in {\mathbb C}^{n\times n}\}\). The main result is the following inequality: \(\frac n2\sin \frac \pi n\leqslant M_n\leqslant \frac n2\), which becomes an identity for \(n=2\). A description of some matrices \(A\) where \(M(A)=1\), as well as some computations of exact values of \(M(A)\) for \(2\times 2\) matrices \(A\) is given.