Gröbner basis of the alternating Galoisian ideal (Q629348)

An ideal \(I\subseteq k[x_1,\dots,x_n]\) is Galoisian relative to \(f\in k[x]\) if every \(g\in I\) vanishes on a permutation of the \(n\) roots of \(f\). (It is assumed that \(f\) has no multiple roots.) We can associate with each Galoisian ideal a subgroup of \(S_n\) that leaves the ideal invariant; this is the decomposition group \(\mathrm{Gr}(I)\) of the ideal. It is clear that \(\mathrm{Gr}(I)\subseteq L\) for the largest \(L\subseteq S_n\) that defines \(I\); if \(\mathrm{Gr}(I)=L\), then \(I\) is pure. Suppose that the Galois group of \(f\) over \(k\) is a subgroup of the alternating group \(A_n\); if \(\mathrm{Gr}(I)=A_n\), we say that \(I\) is an alternating galoisian ideal associated with \(f\). There are in fact two such ideals, generated by the sets \(F_1=\{C_1,\dots,C_{n-2},G,C_n\}\) and \(F_2=\{C_1,\dots,C_{n-2},H,C_n\}\), where \(G\) and \(H\) are polynomials in \(x_1,\dots,x_{n-1}\), of degree~1 in \(x_{n-1}\), and each \(C_i\) is one of the Cauchy moduli of \(f\), defined as \[ C_1(x_1)=f(x_1)\quad\roman{and}\quad C_{r+1}=\frac{C_r(x_1,\dots,x_{r-1},x_{r+1})-C_r(x_1,\dots,x_{r-1},x_r)}{x_{r+1}-x_r}. \] Previous work by \textit{P. Aubry} and \textit{A. Valibouze} [J.~Symb.~Comp.~30, No.6, 635--651 (2000; Zbl 0989.12004)] has shown that \(F_1\) and \(F_2\) are triangular sets that form a Gröbner basis in the lexicographic order \(x_i<x_{i+1}\). If one can compute \(G\), one has a Gröbner basis of \(I\). Let \(\mathbf a\) be in the variety of an alternating galoisian ideal \(I\); \((a_1,\dots,a_{n-2})\) is in the variety of the ideal generated by \(C_1,\dots,C_{n-2}\), so it suffices to find an expression of \(a_{n-1}\) as a polynomial in \(a_1,\dots,a_{n-2}\) over \(k\). The author briefly describes three known methods, as well as why they are impractical for the general, symbolic case. The author then describes a new method, called ``the descent of the Vandermonde determinants'', from the elegant expression \[ v_r=\frac{v_{r-1}}{c_r(a_{r-1})} \] where \[ v_r=\prod_{r\leq i<j\leq n}(a_i-a_j) \] is the Vandermonde determinant associated to the polynomial \[ c_r(x)=C_r(a_1,\dots,a_{r-1},x). \] One obtains the desired expression for \(a_{n-1}\) by computing the inverse of each \(c_r(a_{r-1})\) in \(k(a_1,\dots,a_{r-1})\), which is accomplished via the extended Euclidean algorithm. A very brief comparison with a Gröbner basis function in Magma follows. The article concludes with an example computation for a quartic polynomial over the field \(\mathbb{Q}(t)\).