Cyclic weighted shift matrix with reversible weights (Q1738212)

The numerical range of \(A\in\mathbb C_{n\times n}\) is the compact convex set \(W(A)= \{x^*Ax: x\in {\mathbb C}^n, \,x^*x=1\}\) in \(\mathbb C\). For arbitrary \(A\in \mathbb C_{n\times n}\), there exists symmetric matrix \(S\in\mathbb C_{n\times n}\) satisfying \(W(A)=W(S)\) [\textit{J. W. Helton} and \textit{I. M. Spitkovsky}, Oper. Matrices 6, No. 3, 607--611 (2012; Zbl 1270.15014)]. However, not every \(n \times n\) matrix is unitarily similar to a symmetric matrix. Matrices that are unitarily similar to complex symmetric matrices or operators have been studied, for example, in [\textit{S. R. Garcia} and \textit{M. Putinar}, Trans. Am. Math. Soc. 358, No. 3, 1285--1315 (2006; Zbl 1087.30031); \textit{M.-T. Chien} et al., Linear Multilinear Algebra 65, No. 10, 2131--2144 (2017; Zbl 1387.15030)]. In the present paper, it is proved that \[ A= \left[ \begin{array}{@{}c|ccc@{}} a_{11} & a_{12}&\cdots & a_{1n} \\ \hline a_{1n} & \quad&\quad& \quad \\ \vdots & \quad&\tilde{A}& \quad \\ a_{12} & \quad&\quad& \quad \end{array} \right]\in\mathbb C_{n\times n} \] is unitarily similar to a symmetric matrix via Fourier transform if and only if \(a_{1,k+1}=a_{n-k+1,1}\) and \(a_{1+k,1+l}=a_{n+1-l,n+1-k}\), i.e., \(\tilde{A}\) is symmetric with respect to the main diagonal for all \(k,l=1,2,\dots,n-1\); see Theorem 2.1. It follows that a weighted shift matrix \[ \left[ \begin{array}{@{}c|ccccc@{}} \quad & w_1&\quad & \quad&\quad&\quad \\ \quad & \quad&w_2 & \quad&\quad&\quad \\ 0 & \quad&\quad & w_3&\quad&\quad \\ \quad &\quad&\quad & \quad&\ddots&\quad \\ \quad & \quad&\quad & \quad&\quad&w_{n-1}\\ \hline w_n & \quad&\quad&0&\quad& \quad \end{array} \right]\in\mathbb C_{n\times n} \] with reversible weights, i.e., \(w_{n-k+1}=w_k\) for \(k=1,2,\dots,n\), is unitarily similar to a symmetric matrix; see Corollary 2.2. The authors use the unitary matrix from the Fourier transform to realize the unitary similarity. The unitary matrix \(\frac{1}{\sqrt{2}}(I_n+i(1\oplus J_{n-1})) \in \mathbb C_{n\times n}\), where \(J_{n-1}\in\mathbb C_{(n-1)\times (n-1)}\) is the backward identity, can also be used to prove Theorem 2.1.