Convex domains and K-spectral sets (Q811846): Difference between revisions

Let \(\Omega\) be a nonempty proper open convex subset of the complex plane. The present paper studies its associated constant \(C (\Omega)=\sup\{\|r(A)\|: \overline{W(A)}\subseteq\Omega\), \(r\) a rational function, \(\|r(z)\|\leq 1\) \(\forall z\in\Omega\}\), where the supremum is taken over all bounded linear operators \(A\) on Hilbert spaces and all rational functions \(r\) satisfying the specified conditions, and its completely bounded analogue \(C_{cb}(\Omega)\) with \(r\) replaced by a matrix-valued function \(R=[r_{ij}]^{m,n}_{i=1,j=1}\), where \(r_{ij}\) is rational for all \(i\) and \(j\), satisfying \(\|R(z)\|\leq 1\) for all \(z\) in \(\Omega\). It is known, for example, that \(1\leq C(\Omega)\leq C_{cb}(\Omega)< 57\) and \(C(D)=2\) for any open circular disc \(D\). The present paper uses various different approaches to estimate \(C(\Omega)\) and \(C_{cb}(\Omega)\), which improve or recover these known cases. The first approach is based on an appropriate integral representation of \(r(A)\) or \(R(A)\) and makes use of the positivity of the double layer potential of \(\Omega\). This yields, in Section 2, the estimate \(C_{cb} (\Omega)\leq 2+\pi+\inf_{w\in\Omega}TV(\log\|\sigma-w|)\) for every bounded convex domain \(\Omega\), where \(TV(\log|\sigma-w|)\) denotes the total variation of the function \(\log|\sigma-w|\) as \(\sigma\) runs around \(\partial\Omega\). If \(\Omega\) contains a sector of angle \(2 \alpha\), \(0<\alpha<\pi/2\), then the estimate, given in Section 3, is \[ C_{cb}(\Omega)\leq 1+\frac{2}{\pi} \int_\alpha^{\pi /2}\frac{\pi-x+\sin x}{\sin x}dx. \] The second approach as developed in Section 4 is through another integral representation for \(r(A)\) or \(R(A)\) based on the solution of the C. Neumann problem for the double layer potential on \(\Omega\). For the unbounded \(\Omega\) which contains a sector as above, this yields \(C_{cb}(\Omega)\leq(\pi-\alpha)/\alpha\). Connections with the dilation theorem and Kato's spectral mapping theorem are also indicated here. In the final section, Section 5, the authors prove that \(\sup_\Omega C(\Omega,2)=\sup_\Omega C_{cb} (\Omega,2)=2\). Here \(C(\Omega,2)\) (resp., \(C_{cb} (\Omega,2))\) denotes the supremum of \(\|r(A)\|\) (resp., \(\|R(A)\|)\) when \(A\) is restricted to 2-by-2 matrices with \(\overline{W(A)}\subseteq \Omega\) and \(r\) is as before. We remark that the final theorem, Theorem 5.4, has already been known before; cf. [\textit{M.-D. Choi} and \textit{C.-K. Li}, Linear Multilinear Algebra 47, No.~1, 35--48 (2000; Zbl 0952.47005) Theorem 2.1].