Families of cyclic polynomials obtained from geometric generalization of Gaussian period relations (Q4671850)

Let \(p = e \cdot f +1\) be a prime and \(\mathbb K = \mathbb Q(\eta) \subset \mathbb Q(\zeta_p)\) be the subfield of degree \(e\) of the \(p\)-th cyclotomic field and \(\eta_i = \text{Tr}_{\mathbb Q/\mathbb K}(\zeta_p^{g^{i}}),\) \(i = 0, 1, \dots, e-1\), with \(\langle g\rangle = \mathbb Z/p\cdot\mathbb Z\), be the \(e\) conjugate traces of \(\zeta\) to \(\mathbb K\); we can assume, \(\eta = \eta_0\). These traces, also called Gauss periods, have been subject of intensive research in various contexts of number theory. One thread of research originated in work of Kummer and the Lehmers, aims to find parametrized families of cyclic polynomials of degree \(e\). Aside from some incidental papers by other authors, this direction has been followed in the last decade intensively by F. Thaine, in a series of 5 papers. The well-written paper under review treats a similar subject and the authors also point out the connection to the inverse Galois problem. They develop a technique for deriving families of cyclic polynomials, which is different from the ones introduced by Thaine. This can be described briefly as follows: The Gauss sums are a normal base for the integers of \(\mathbb K\) and have a multiplication table of the shape \[ \eta_m \eta_{m+i} = \sum_{j=0}^{e-1} c_{i,j} \eta_{m+j},\quad m = 0, 1, \dots, e-1. \] The authors denote this as the axiom of their theory. If \(\mathbf{R} = \text{circ}(\mathbf{y})\) with \(\mathbf{y} = (y_0, y_1, \dots, y_{e-1})\) is the circulant matrix of the vector \(\mathbf{y}\) and \(\mathbf{D} = \text{diag}(\mathbf{y})\), the multiplication table can be rewritten as \[ \mathbf{R}(\mathbf{\eta}) \times \mathbf{D}(\mathbf{\eta}) = \mathbf{C} \times \mathbf{R}(\mathbf{\eta}), \] with the appropriate matrix \(\mathbf{C}\). Starting from this relation, the authors replace the Gauss sums by formal indeterminates and derive an equation in the field \(\mathbb L = \mathbb Q(y_0, y_1, \dots, y_{e-1})\) on which they let the cyclic permutation \(\sigma : y_j \to y_{(j+1) \;\text{rem} \, e}\) act. The main observation (key lemma) is that \(\mathbb Q(\mathbf{C}) = \mathbb L^{\sigma}\) and \(\mathbb L/\mathbb Q(\mathbf{C})\) is cyclic of degree \(e\), generated by the characteristic polynomial of \(\mathbf{C}\); here \(\mathbb Q(\mathbf{C})\) denotes, of course, the extension generated by the entries of the matrix \(\mathbf{C}\). Based upon this result, the authors then derive a general \([\frac{e-2}{2}]\)-parametric family of cyclic polynomials and deduce subsequently, by adequate parameter restriction, some simpler families for small fixed values of \(e\).