On a question of Hayes concerning integrality of Brumer elements (Q2398250)

Let \(L/K\) be an abelian \(CM\)-extension with Galois group \(G\), denote by \(w_K\) the number of roots of unity contained in \(K\), and let \(S\) be a finite set of places of \(K\) containing all places ramified in \(L/K\) as well as all Archimedean places. For \(\sigma\in G\) and \(\mathrm{Re} (s)>1\) put \[ \zeta_{K,S}(\sigma,s) = \sum_{(I,S)=1,(I,L/K)=\sigma}N(I)^{-s}, \] and define the Brumer element corresponding to \(S\) by \[ \theta_{L/K,S} = \sum_{\sigma\in G}\zeta_{K,S}(\sigma,0)\sigma^{-1}\in \mathbb Q[G]. \] It has been conjectured by \textit{D. R. Hayes} [Contemp. Math. 358, 193--206 (2004; Zbl 1076.11057)] (see also Conjecture 4 in the paper of \textit{B. R. Smith} [Math. Z. 270, No. 3--4, 645--657 (2012; Zbl 1292.11129)]) that \(w_K\theta_{L/K,S}\) has all its coefficients divisible by an odd prime \(p\mid w_K\), provided the following conditions are satisfied: a) At least one prime divisor of \(p\) in \(K\) is unramified in \(L/K\), b) There is a unit \(u\) of the maximal real subfield of \(L\) such that the extension \(L(\root p\of u)/L\) is cyclic and unramified. The author provides a family of counterexamples to this conjecture. Let \(\alpha\) be the generator of a cyclic field of degree \(9\) over \(\mathbb Q\) and prime conductor \(r\equiv10,19\) mod \(27\), and choose \(m>1\) divisible by \(3\) such that the class-number of \(K=\mathbb Q(\sqrt{-m})\) is not divisible by \(3\) and \(r\mathbb Z_K\) is a prime ideal. Theorem 3.3 shows that if \(K=\mathbb Q(\sqrt{3m})\) and \(L=K(\alpha,\zeta_3)\), then the extension \(L/K\) is the desired counterexample. From the results of \textit{M. Bhargava} and \textit{I. Varma} [Proc. Lond. Math. Soc. (3) 112, No. 2, 235--266 (2016; Zbl 1407.11126)] it follows that the described situation arises infinitely often.