The numerical radius and bounds for zeros of a polynomial (Q4780409)

For any monic polynomial \(p(t)= t^n+ a_{n-1} t^{n-1}+\cdots+ a_1t+ a_0\), the matrix \[ A= \left[\begin{matrix} 0 &&&&&&-a_0\\ 1 & 0\\ &&\cdot &&&&\cdot\\ &1 &&\cdot &&&\cdot\\ &&\cdot &&\cdot\\ &&&\cdot &&0 &\cdot\\ &&&&\cdot\\ &&&&&1 & -a_{n-1}\end{matrix}\right], \] called the companion matrix of \(p\), is associated. The zeros of \(p\) are exactly the eigenvalues of \(A\). In this paper, the authors obtain two bounds, both in terms of the coefficients of \(p\), for the zeros of \(p\) by different estimates for the eigenvalues of \(A\). One is based on the geometric mean estimate obtained before by the first author for the nonnegative matrix whose entries are the absolute values of those of \(A\). The other is based on a containment region for the numerical range of \(A\) which is the convex hull of the ovals of Cassini. More precisely, it is proved that the numerical range of any \(n\)-by-\(n\) matrix \(A= [a_{ij}]\) is contained in the convex hull of the union of the ovals \[ \{z:|z- a_{ii}||z-a_{jj}|\leq (g_i(A)^2 g_j(A)^2+|a_{ii}- a_{jj}|^2)^{1/2}\} \] with \(i, j= 1,\dots, n\) and \(i\neq j\), where \[ g_i(A)= \Biggl(\sum_{j\neq i}|a_{ij}|+ \sum_{j\neq i}|a_{ji}|\Biggr)/2 \] for each \(i\). This is the numerical range analogue of Brauer's inclusion region for eigenvalues of a matrix.