Quantitative non-vanishing of Dirichlet \(L\)-values modulo \(p\) (Q2200768)

Let \(p\) be an odd prime. Let \(\lambda\) and \(\chi\) be two (not necessarily primitive) Dirichlet characters of modulus \(N\) and \(F\), respectively. Suppose that \(p\) does not divide \(FN\). Then for each non-negative integer \(k\) the value \(L(-k, \lambda\chi)\) is algebraic and \(p\)-integral. The aim of this paper is to estimate the cardinality of the set \[ \mathfrak{X}_{\lambda,k}(F) := \left\{\chi \in \mathrm{Hom}((\mathbb Z / F \mathbb{Z})^{\times}, \overline{\mathbb{Q}}^{\times}) \mid L(-k, \lambda\chi) \not\equiv 0 \mod \mathfrak{p}\right\} \] as a function of \(F\). Here \(\mathfrak{p}\) denotes a chosen prime in \(\overline{\mathbb{Q}}\) above \(p\).\par Note that this question is closely related to the \(p\)-(in)-divisibility of class groups if \(k=0\) (and of certain étale cohomology groups if \(k>0\)): If \(\psi\) is an odd Dirichlet character of conductor \(N_{\psi}\) and \(p \nmid \varphi(N_{\psi})\), then \(L(0, \psi^{-1})\) is \(p\)-indivisible if and only if the `\(\psi\)-part' of the class group of \(\mathbb{Q}(\zeta_{N_{\psi}})\) is of cardinality coprime to \(p\).\par The authors provide lower bounds for \(\# \mathfrak{X}_{\lambda,k}(F)\) whenever \(F\) is sufficiently large by two different approaches: a `homological' and an `algebraic' approach. The homological approach yields the better bounds and a special case of the corresponding main result is the following (which is taken from the introduction of the authors). Suppose that \(\lambda\) is non-trivial and that \(F\) is a prime such that \(F \nmid N\). Moreover, assume that \(p \nmid \varphi(F)FN\). Then for \(F>N\) one has that \[ \# \mathfrak{X}_{\lambda,0}(F) \geq \left\lfloor\left(\frac{F}{9N}\right)^{1/2} \right\rfloor. \] The proof strategy (in both cases) is as follows. The authors consider sums \[ R_{\lambda}(r,F) = \sum_{\chi} \chi(r) c(\chi) L(0, \chi\lambda), \] where \(r \in (\mathbb{Z}/F\mathbb{Z})^{\times}\) and \(c(\chi) \in \overline{\mathbb{Z}}_p\). Let \(\chi_1, \dots, \chi_t\) be the characters in \(\mathfrak{X}_{\lambda,0}(F)\). Let \(Q>0\) be an integer and choose vectors \((r_i), (s_j) \in (\mathbb{Z}/F\mathbb{Z})^{\times Q}\). It is not hard to see that \[ \mathrm{det}(R_{\lambda}(r_is_j,F)) \equiv 0 \mod \mathfrak{p} \] whenever \(t<Q\). So in other words, if this determinant is non-zero modulo \(\mathfrak{p}\), then \(\# \mathfrak{X}_{\lambda,0}(F)\) is at least \(Q\). \par In the algebraic approach \(c(\chi)\) is taken to be the Gauss sum \(G(\overline{\chi})\), whereas in the homological approach it is \(\overline{\chi}(N)\). \par Finally, we mention a further nice consequence of their results. Suppose \(p\) is inert in \(\mathbb{Q}(\zeta_{\varphi(F)})\), where \(F\) is a sufficiently large prime. Then \(L(0,\chi) \not\equiv 0 \mod \mathfrak{p}\) for all odd primitive characters \(\chi\) of conductor \(F\).