Matrices and spectra satisfying the Newton inequalities (Q1020936)

A sequence of real numbers \((c_n)_{n \geq 1}\) is called a Newton sequence if \( c_k^2 - c_{k - 1} c_{k + 1} \geq 0 \) for every \(k \geq 1\) (considering that \(c_0 = 1\)). If \(A\) is a matrix of order \(n\) with eigenvalues \(\{ \lambda_1, \lambda_2, \dots , \lambda_n \}\) then we define \[ S_k (A) = \sum_{1 \leq j_1 < \dots < j_k \leq n} \lambda_{j_1} \lambda_{j_2} \dots \lambda_{j_k} \] and \[ c_k (A) = \frac{1}{{n \choose k}} S_k (A). \] These numbers \(c_k(A)\) are called the Newton coefficients of the matrix. A matrix \(A\) is called a Newton matrix (and its spectrum a Newton spectrum) if the Newton coefficients are real and form a Newton sequence. It is known that if all eigenvalues of a matrix are real then the matrix is a Newton matrix. The authors discuss possible Newton matrices with complex eigenvalues. First, in Section 2, all such matrices of order 2, 3 and 4 are completely characterized. Section 3 is concerned with some general properties: similarity invariance of the property, the fact that \(A\) is a Newton matrix if and only if \(\alpha A\) is for every real \(\alpha\), the fact that an invertible matrix \(A\) is a Newton matrix if and only if its inverse is a Newton matrix and that \(A\) is a Newton matrix if and only if \(A \oplus 0\) is a Newton matrix. The last two sections discuss the behavior of the property of being a Newton matrix under translations.