Matrix powers with circular numerical range (Q2197134)

The numerical range of an operator \(T\in B(H)\) on the complex Hilbert space \(H\) is \[W(T):=\{\langle Tx, x\rangle: x\in H, \langle x, x\rangle = 1\}\subset {\mathbb C}.\] The Toeplitz-Hausdorff theorem (1918, 1919) asserts that \(W(T)\) is a convex set. If \(A\) is an \(n\times n\) matrix and \({\mathbb C}^n\) is equipped with the standard inner product, we have \(W(A)= \{x^*Ax: x\in {\mathbb C}^n, x^*x=1\}\), a compact convex set. In this paper, the authors show that if an \(n \times n\) nonzero matrix \(A\) satisfies \(W(A^k) =W(A)\) for all \(1 \le k\le n\), then \(W(A)\) cannot be a (nondegenerate) circular disc. Moreover, they show that \[W(A) =W(A^{n-1}) =\{z\in {\mathbb C}: |z| \le 1\}\] if and only if \(A\) is unitarily similar to \(K_n\), where \[ K_2 = \begin{bmatrix} 0&0\\2&0\end{bmatrix};\quad K_n=\begin{bmatrix} 0 &&&&& \\ \sqrt 2 &0&&&&\\ &1&0&&&\\ &&\ddots &\ddots &&\\ &&&1&0&\\ &&&&\sqrt 2 &0 \end{bmatrix}, \ n\ge 3. \] Therefore, if \(n\ge 4\) and \(A\) is an \(n\times n\) nonzero matrix satisfying \(W(A^k)=\) \(W(A)\) for all \(1\le k\le n-1\), then \(W(A)\) cannot be the closed unit disc. Denote by \(w(T):=\sup \{|z|: z\in W(T)\}\) the numerical radius of \(T\in B(H)\). The operator \(T\) is said to be a numerical contraction if \(w(T)\le 1\). A deep result of \textit{T. Ando} [Acta Sci. Math. 34, 11--15 (1973; Zbl 0258.47001)] characterizes numerical contractions and it asserts that \(T\) is a numerical contraction if and only if \(T= (I-Z)^{1/2}C(I+Z)^{1/2}\) for some contraction \(C\) (i.e., \(\|C\|\le 1\)) and Hermitian contraction \(Z\). Ando's result is used to provide more intuitive proofs of some results of \textit{M. J. Crabb} [Mich. Math. J. 18, 253--256 (1971; Zbl 0198.17204)], namely Proposition 3.1, Theorem 3.2, and Proposition 3.4. In particular, for a numerical contraction \(T\), \(\lim_{n\to \infty} \|T^n x\| = \sqrt 2\) for some unit vector \(x\in H\) if and only if \(T \) is unitarily similar to \(K_\infty \oplus T'\), where \(K_\infty\) is the weighted shift operator with weights \(\sqrt 2, 1, 1, 1,\dots\) and \(T'\) is a numerical contraction. Some improvements of the results of Crabb are given using the new approach.