On numerical ranges of the compressions of normal matrices (Q621019)

Let \(A\) be an \(n\times n\) normal matrix whose numerical range \(W(A)\) is a \(k\)-polygon \((k\leq n)\) with vertices \(\lambda_1,\dots,\lambda_k\). It is known that the \(\lambda_i\)'s are eigenvalues of \(A\). Let \(x_1,\dots, x_k\) be a corresponding system of orthonormal eigenvectors of \(A\). For a unit vector \(v=\sum^k_{i=1} v_ix_i\) with \(v_i\neq 0\) for all \(i\), an \(n\times (k-1)\) isometry \(P\) may be constructed which maps the orthogonal complement of \(v\) in the \(k\)-dimensional subspace spanned by \(x_1,\dots,x_k\) isometrically into \(\mathbb{C}^n\). If \(C= P^*AP\), then \(W(C)\) is contained in \(W(A)\) and \(\partial W(A)\cap\partial W(C)\) consists of \(k\) points \(c_1,\dots, c_k\) with \(c_i\) on the open line segment \((\lambda_i, \lambda_{i+1})\) \((\lambda_{k+1}\equiv\lambda_1)\) for all \(i\). The above constructions are known before by Adam-Maroulas, Gau-Wu and Mirman. The present paper elaborates on these by considering another \(n\times n\) normal matrix \[ B= \text{diag}(\lambda_1,\dots,\lambda_{k-2}, \lambda_s, u,\lambda_{k+1},\dots, \lambda_n), \] where \(\lambda_s\) is the intersection point of the lines of \(\lambda_{k-2}\lambda_{k-1}\) and \(\lambda_k\lambda_1\), and \(u\) is an interior point of \(W(A)\). Then \(W(B)\) is the \((k-1)\)-polygon with vertices \(\lambda_1,\dots, \lambda_{k-2},\lambda_s\). Proceeding as before, consider \(e_1,\dots, e_{k-1}\), where the \(e_i\)'s are (part of) the standard basis of \(\mathbb{C}^n\), instead of \(x_1,\dots,x_k\), and \(z= \sum^{k-1}_{i=1} z_ie_i\) with \(z_i\neq 0\) for all \(i\), and construct the \(n\times (k-2)\) isometry \(Q\) and \(D= Q^*BQ\). Then \(\partial W(B)\cap\partial W(D)\) consists of \(k-1\) points \(d_1,\dots, d_{k-1}\). The main results in Section 2 relate the vectors \(v\) and \(z\) by expressing the \(z_i\)'s in terms of the \(c_i\)'s and \(\lambda_i\)'s and, conversely, the \(v_i\)'s in terms of the \(d_i\)'s and \(\lambda_i\)'s. Section 3 proves some properties of \(C\) and \(W(C)\), namely, (1) \(\partial W(C)\) contains no line segment, (2) \(C\) is irreducible, that is, \(C\) is not unitarily equivalent to the direct sum of two other matrices, and (3) \(\partial W(C)\) is a differentiable curve, generalizing the corresponding result by Gau-Wu for \(A\) with \(n\) distinct eigenvalues.