Parametric Galois extensions (Q471880)

Given a field \(k\) and a finite group \(H\), an \(H\)-parametric extension over \(k\) is a finite Galois extension \(E\) of \(k(T)\), of Galois group \(G\) containing \(H\), which is \(k\)-regular (i.e. \(E\cap\overline{k} = k\)) and such that any Galois extension of \(k\) of group \(H\) occurs as a specialization of \(E/k(T)\). The existence or not of parametric extensions, which is intermediate between two classical questions of inverse Galois theory, the Beckmann-Black problem and the existence of one parameter generic polynomials, seems to be quite challenging, the answer depending apparently on the base field \(k\) and the finite group \(G\). Here the author is mainly interested in producing non-\(H\)-parametric extensions over \(k\) of group \(G\) containing \(H \neq (1)\). Take \(k\) to be the quotient field of a Dedekind domain \(A\) of characteristic zero with infinitely many distinct primes, additionally assumed to be Hilbertian. The author's systematic approach consists in taking two \(k\)-regular extensions, \(E_1 /k(T)\) of group \(H\) and \(E_2 /k( T)\) of group \(G\), and giving two sufficient conditions which each guarantees that there exist some specializations of \(E_1 /k(T)\) of group \(H\) which cannot be specializations of \(E_2 /k( T)\) (and so \(E_2 /k( T)\) is not \(H\)-parametric over \(k\)). The first condition, called \textit{Branch Point Hypothesis} (BPH), involves the arithmetic of the branch point sets \(\{t_{1,}j , \dots, t_{r j,} j \}\) of the extensions \(E_j /k(T)\). Denote the irreducible polynomial of \(t_{i,j}\) (resp. of \(1/t_{ i,j}\)) over \(k\) by \(m_{ i,j} ( T)\) (resp. \(m_{ i,j}{*} ( T))\). Set \(m_{ i,j }( T)= 1\) if \(t_{ i,j} =\infty\) and \(m_{ i,j }{*} ( T) = 1\) if \(t_{ i,j} = 0\). Finally set \(m_{ E j }(T) = \prod^{r_j}_{ 1} m_{ i,j} (T)\) and \(m^{\ast}_{E_j} (T) = \prod^{r_j}_{1} m_{ i,j}{*} (T)\). A non-zero prime \(\mathcal P\) of \(A\) is called a prime divisor of \(P(T) \in k[T]\) if there exists \(t_0 \in k\) such that \(P ( t_0 )\) belongs to the maximal ideal of the localization of \(A\) at \(\mathcal P\). Then the (BPH) reads: there exist infinitely many distinct primes of \(A\), each of which being a prime divisor of \(m_{E_1} (T) m^{\ast}_{E_1} (T)\) but not of \(m_{E_2} (T) m^{\ast}_{E_2} ( T)\). The second condition, called Inertia Hypothesis (IH), concerns the geometry of the inertia of the extension \(E_1 /k(T)\). At any branch point \(t_ i , 1 \leq i \leq r\), of an extension \(E/k(T)\) of group \(G\), the inertia groups of \(E\overline{k}/\overline{k}( T)\) at \(t_ i\) are cyclic conjugate groups of order equal to the ramification index \(e_ i\), each one having a distinguished generator corresponding to the automorphism \(( T - t_ i )^{\frac{1}{ e_ i}} \to \zeta_{e_ i} ( T - t _i )^{\frac{1}{ e_ i}}\) (the \(\zeta_{ e_ i}\) form a coherent system of primitive\(e_i\) -th roots of unity). The inertia canonical conjugacy class \(C_i\) is defined as the conjugacy class of all the distinguished inertia generators at \(t_i\) , and the unordered \(r\)-tuple \(( C_1 , \dots , C_r )\) is called the inertia canonical invariant of \(E/k(T)\). Denote then by \((C_{1,1} , \dots, C_{ r_1} ,1)\) the inertia canonical invariant of \(\{t_{1,1} , \dots , t_{ r_1} ,1 \}\). The (IH) states the existence of some index \(i \in \{1, \dots , r_1 \}\) satisfying the following two conditions: (a) \(m_{ i,1} ( T) m_{ i,1}^{\ast} ( T)\) has infinitely many distinct primes , (b) the set \(\{C^a_{1,2} , \dots , C^a_{ r_2,{}^2} /a \in \mathbb N \}\) does not contain the \(G\)-conjugacy class of \(C_{ i,1}\). The author's main criterion, called non-\(H\)-parametricity, states that under either one of the conditions (BPH) or (IH), there exist infinitely many distinct finite Galois extensions of \(k\) which are all obtained by specializing \(E_1 /k(T)\), but none of them is a specialization of \(E_2 /k( T)\). Strengthening the (BPH) or the (IH) allows to give more practical forms of the criterion. Computational applications are given in the two last sections of the paper. One key problem is the construction of the two \(k\)-regular extensions \(E_ i /k(T)\) with incompatible ramification data.