Cohomology and the Brauer groups of diagonal surfaces (Q2130506)

This paper spectacularly solves an important problem about the Brauer group of diagonal quartic surfaces, introducing methods that are expected to be more widely applicable. Indeed diagonal quartic surfaces have long been a testing ground for techniques in diophantine geometry. Peter Swinnerton-Dyer, to whose memory this paper is dedicated, addressed the subject in both his first and his last published papers, as well as in many others in the seventy three years that separate them. The chief success here is the determination of the \(2\)-primary torsion part of \(\mathrm{Br}(X)\) (and thereby the entire Brauer group) for diagional quartic surfaces with rational coefficients, considered over \(\mathbb Q\) or \(\mathbb Q(i)\). What is most interesting, though, is the method, which is well described in the introduction to the paper. Let \(k\) be a number field, \(\Gamma=\mathrm{Gal}(\bar k/k)\) and \(X\) a projective surface over \(k\) with \(\mathrm{Pic}(\overline X)\) torsion-free and \(\mathrm{Br}(\overline{X})^\Gamma\) finite: these conditions hold in many important cases. Then the authors give a procedure to determine \(\mathrm{Br}(X)/\mathrm{Br}_0(X)\) in three stages. First, determine the action of \(\Gamma\) on \(\mathrm{Br}(\overline{X})\) and compute \(\mathrm{Br}(\overline{X})^\Gamma\). This part has been the subject of much work over many years and is accessible in many cases. Second, determine the image of \(\mathrm{Br}(X)\) in \(\mathrm{Br}(\overline{X})^\Gamma\). Third, compute \(\mathrm{Br}(X)/\mathrm{Br}_0(X)\) as an extension of \(\mathrm{Br}(X)/\mathrm{Br}_1(X)\) by \(\mathrm{Br}_1(X)/\mathrm{Br}_0(X)\cong H^1(k,\mathrm{Pic}(\overline{X})\). The last two steps are not in general practical. The method used here is to compare the map of \(\Gamma\)-modules \(\mathrm{Hom}(\mathrm{Pic}(X_{\mathbb C}),\mathbb{Z})\to T(X_{\mathbb C})\otimes\mathbb{Q}/\mathbb{Z}\), viewed as a complex, with a truncation of the complex \(\mathbf{R}p_*\mathbb{G}_{m_X}\) obtained by pushing forward under the structure morphism \(p\colon X\to \mathrm{Spec}\,k\). This allows them to compute the relevant Brauer groups as Galois hypercohomology of this (truncated) complex. A lot of work, some of it already available and some not, is needed to carry out this programme in the case at hand. The result for the case of diagonal quartics is that \(2\)-primary torsion is zero, except for the surface \(x_0^2+x_1^2+2x_2^2-2x_3^2=0\) (working over \(\mathbb Q\) or \(\mathbb{Q}(i)\)) and, over \(\mathbb{Q}\) only, the surface \(x_0^2+x_1^2+8x_2^2-8x_3^2=0\). This is in contrast to the odd case, where \(3\)- and \(5\)-torsion arise in infinitely many cases. The precise description of the extension that their method gives also allows a more refined description of \(\mathrm{Br}(X)/\mathrm{Br}_0(X)\) in these two cases.