The generic irreducibility of the numerator of the zeta function in a family of curves with large monodromy (Q1355295)

N. Katz made the following conjecture. Let \(U/\mathbb F_q\) be an open subset of the affine line \(A_{\mathbb F_q}^1\) and let \(\psi:X\to U\) be a proper, smooth family of curves of genus \(g\). Suppose that the family has large monodromy (see below). Denote by \(p_n\) the proportion of points \(u\in U(\mathbb F_{q^n})\) such that the polynomial \(P(T)\), the numerator of the zeta function \(Z(X_u/ \mathbb F_{q^n},T)\), is irreducible over \(\mathbb Q\). Then \(\lim_{n\to\infty} p_n=1\). The author proves Katz's conjecture. To illustrate the argument and the conjecture, consider the set \(M_R\) of monic polynomials of degree \(d\), where \(d\) is a fixed integer, whose coefficients are integers in the interval \([1,R]\), \(R\) a positive integer. Define the density by \[ \lim_{R\to\infty} \frac {\operatorname {Card}\{\text{ irreducible polynomials in } M_R\}} {\operatorname {Card}M_R}. \] Then the density is 1; that is, `most' polynomials are irreducible. That result is implied by the stronger: \[ \lim_{R\to\infty} \frac {\operatorname {Card}\{ \text{ polynomials in } M_R \text{ that are reducible}, \bmod \ell, \text{some prime }\ell\}} {\operatorname {Card}M_R} =0. \] To prove the latter, begin with the fact that, if \(\ell\) is a prime, there are approximately \(1-\frac{1}{d}\) monic reducible polynomials of degree \(d\) in \(\mathbb F_q[T]\). If one reduces \(\bmod \ell_1,\dots, \ell_r\) (distinct primes), then the Chinese remainder theorem implies that if \(R\) is divisible by the product \(\ell_1,\dots, \ell_r\), then the values of the reductions, \(\operatorname {mod} \ell_1,\dots, \ell_r\), of the polynomials in \(M_R\) are independent random variables. So the probability of the event that a polynomial is reducible \(\operatorname {mod}\ell_j\), \(j=1,\dots, r\), is \((1-\frac{1}{d})^r\) which can be made \(<\varepsilon\) by choosing \(r\) large enough, whence the result. The author's idea is that one can apply a similar argument to Katz's conjecture if the \(\operatorname {mod}\ell\) monodromy of the family of curves is large for \(\ell\) large; the reductions, \(\operatorname {mod}\ell\), of the numerators of the zeta functions of curves in the family turn out to be the characteristic polynomials of certain Frobenius elements in the symplectic similitude group, \(SSp(2g, \mathbb{F}_\ell)\). Knowledge of the \(\operatorname {mod}\ell\) monodromy permits the application of results concerning the equidistribution of the Frobenius elements in \(SSp(2g, \mathbb{F}_\ell)\). The idea then is to show that \[ \begin{aligned} &\text{Prob} \{u \in U(\mathbb{F}_{q^n}) \text{ such that the fibre over \(u\) has a reducible numerator of the zeta function}\}\\ &\leq \text{Prob} \{u\in U(\mathbb{F}_{q^n}) \text{ such that the numerator of } Z(X_u) \text{ is reducible, }\operatorname {mod}\ell_i,\;i=1,\ldots,r\}\\ \sim &(1-\tfrac{1}{2g})^r< \varepsilon, \end{aligned} \] if \(r\) is large enough. Those estimates depend on the \(\ell_i\), \(i=1,\dots, r\) being large to ensure that the \(\operatorname {mod}\ell_i\) monodromy is large for \(i=1,\dots, r\) and that in turn leads to a proof of Katz's conjecture.