Galois groups of random polynomials over the rational function field (Q7035141)

This paper studies a variation of the classical problem of determining the Galois group of a random polynomial\N\[\Nf(x) = x^n + a_{n-1} x^{n-1} + \cdots + a_1 x + a_0 \in \mathbb{Z}[x]\N\]\Nwhere \(\lvert a_j \rvert \leq H\). The case when the degree \(n\) is fixed and the coefficient bound \(H \to \infty\) had a recent breakthrough due to \textit{M. Bhargava} [Ann. Math. (2) 201, No. 2, 339--377 (2025; Zbl 1573.11349)].\N\NIt is also possible to consider the other limit: fixed \(H\) and increasing degree \(n\). When \(H \geq 2\), it is generally conjectured that a random polynomial should have Galois group \(S_n\) with probability \(1\), but this remains unproven.\N\NThis article studies a version of this latter problem, but where the coefficients are taken from function fields. Let \(\mathbb{F}_q[t]_{\leq d}\) denote the space of polynomials \(h \in \mathbb{F}_q[t]\) of degree at most \(d\), and consider random polynomials of the form\N\[\Nf(x, t) = x^n + a_{n-1}(t) x^{n-1} + \cdots + a_1(t)x + a_0(t) \in \mathbb{F}_q[t][x]\N\]\Nwhere each \(a_{n-1}(t)\) is taken uniformly randomly and independently from \(\mathbb{F}_q[t]_{\leq d}\). For such a polynomial, let \(G_f\) denote the Galois group of its splitting field over \(\mathbb{F}_q(t)\).\N\NWith \(d\) fixed and \(n \to \infty\), what is the distribution of Galois groups? This was previously considered by the author and \textit{A. Popov} in [Finite Fields Appl. 98, Article ID 102466, 27 p. (2024; Zbl 1561.11092)], where they computed the probability that \(f\) has a nontrivial factor in \(\mathbb{F}_q[x]\) and showed that it's positive. Thus the probability that \(G_f = S_n\) cannot be \(1\), and this case is distinct from the classical problems where the coefficients are rational integers.\N\NTo phrase the correct result, write \(f = \sum b_j(x) t^j\) and let \(c_f(x) = \gcd(b_0, \ldots, b_d) \in \mathbb{F}_q[t]\) denote the content of \(f\) in the variable \(t\). Let \(C = \mathrm{Gal}(c(x) / \mathbb{F}_q(t))\). We have \(f(x, t) = c_f(x) g(x, t)\) where \(c_g(x) = 1\). If the degree of \(c_f\) is \(k\), then the degree of \(g\) is \(n - k\) and the Galois group of the splitting field of \(g\) is a subgroup of \(S_{n - k}\). Heuristically, there is no obstruction preventing a generic \(g\) from having full Galois group, so one might conjecture that this occurs with probability \(1\).\N\NIndeed, the primary result of this paper, Theorem 1, proves this by showing that the probability that \(G_f = S_{n - k} \times C\) tends to \(1\) as \(n \to \infty\).\N\NAs with many of these variations, the largest difficulty towards proving these results is distinguishing between the full symmetric group \(S_{n - k}\) and the alternating group \(A_{n - k}\). One way to handle the alternating group is to study the probability that the discriminant of \(f\) is a square. The main brunt of this paper concerns bounding this probability.\N\NThe previous work of Entin and Popov assumed Chowla's conjecture over \(\mathbb{F}_q[t]\). The ideas for the improvement in this paper build on the remarkable work on Chowla's conjecture in [\textit{W. Sawin} and \textit{M. Shusterman}, Ann. Math. (2) 196, No. 2, 457--506 (2022; Zbl 1499.14038)].