Algebraic Brauer groups modulo constants of homogeneous spaces and their compactifications (Q6579115)

This article contains an unfortunate negative answer to a question raised by Harari on the Brauer group of homogeneous spaces of linear groups and their (smooth) compactifications. The context of this question is as follows.\N\NIn his influential paper [Duke Math. J. 75, No. 1, 221--260 (1994; Zbl 0847.14001)], \textit{D. Harari} presents a method to study the Brauer-Manin obstruction on a fibration \(f:X\to B\), assuming that this obstruction is the only one for the basis \(B\) and for sufficiently many fibers of \(f\). One of the key points to this method is the fact that one needs to have sufficiently many rational points \(b\in B(k)\) such that the specialization arrows \N\[\N\mathrm{Br}(X_\eta)/\mathrm{Br}(k(\eta))\to \mathrm{Br}(X_b)/\mathrm{Br}(k),\tag{1}\N\]\Nare isomorphisms. Here, \(\eta\) denotes the generic point of \(B\) and \(X_b\) denotes the fiber over \(b\in B\) (in particular, \(X_\eta\) is the generic fiber). This happens for instance if \(f\) admits a section, but it is far from being true in general.\N\NTwo decades later, in [Ann. Sci. Éc. Norm. Supér. (4) 46, No. 4, 651--692 (2013; Zbl 1328.14032)], \textit{M. Borovoi} et al. study the algebraic Brauer group of homogeneous spaces of linear algebraic groups with stabilizers of ``ssumult'' type (this includes connected stabilizers). In this case, one can prove that every class in the Brauer group is \textit{algebraic}, that is, it comes from\N\[\N\mathrm{Br}_1(X):=\ker(\mathrm{Br}(X)\to\mathrm{Br}(\bar X)).\N\]\NThe Hochschild-Serre spectral sequence tells us that this particular group fits in an exact sequence: \N\[\N0\to\mathrm{Br}(k)\to \mathrm{Br}_1(X)\to H^1(k,\mathrm{Pic}(\bar X))\xrightarrow{d} \ker(H^3(k,\mathbb{G}_\mathrm{m})\to H^3_{\text{ét}}(X,\mathbb{G}_{\mathrm{m}})).\tag{2}\N\]\NBorovoi, Demarche and Harari give a similar sequence for the unramified Brauer group of \(X\), that is, the Brauer group of a smooth compactification \(X^c\) of \(X\): \N\[\N0\to\mathrm{Br}(k)\to \mathrm{Br}_1(X^c)\to \Sha^1_\omega(k,\mathrm{Pic}(\bar X))\xrightarrow{d} \ker(H^3(k,\mathbb{G}_\mathrm{m})\to H^3_{\text{ét}}(X,\mathbb{G}_{\mathrm{m}})),\tag{3}\N\]\Nwhere the group \(\Sha^1_\omega(k,\mathrm{Pic}(\bar X))\) corresponds to classes in \(H^1(k,\mathrm{Pic}(\bar X))\) that are trivialized by restriction to every pro-cyclic subgroup of \(\mathrm{Gal}(k)\).\N\NThese sequences are valid over an arbitrary field and, if \(k\) is a number field, then the last term is trivial. Thus, if the generic fiber of the fibration \(f\) is a homogeneous space as above, the original question on the specialization arrows (1) can be answered positively only if the map \(d\) in (2) and (3) is trivial over the field \(k(\eta)\), or more generally over an arbitrary field. This is Harari's original question.\N\NThe author of this article gives a negative answer to both questions. First of all, he gives in Proposition 2 a reinterpretation of the arrow \(d\) (which is a differential in the Hochschild-Serre spectral sequence) as a cup product with the abelianization of the \textit{Springer class} associated to the homogeneous space. In Corollary 3, he gives examples of homogeneous spaces of \(\mathrm{SL}_m\) with finite abelian stabilizers over a very general class of fields for which the last arrow in (2) is nontrivial. This builds essentially on the Bloch-Kato conjecture (now a theorem by Rost and Voevodsky). Finally, building on duality theorems for 2-local fields (see for instance [\textit{D. Izquierdo}, Math. Z. 284, No. 1--2, 615--642 (2016; Zbl 1407.11130)]), he constructs a homogeneous space of \(\mathrm{SL}_m\) over \(\mathbb{C}((t))((x))((y))\) with finite abelian stabilizers for which the arrow \(d\) in (3) is nontrivial.