On the elementary obstruction to the existence of rational points (Q2460513)

Let \(\mathcal Y\) and \(\mathcal X\) be abelian categories with enough injectives and let \(f_{*}: \mathcal Y\to\mathcal X\) be a left-exact additive functor having a left adjoint functor \(f^{*}\). For \((\mathcal E,\mathcal M)\in\text{Ob}(\mathcal Y)\times\text{Ob}(\mathcal X)\), there exists a Grothendieck spectral sequence \[ E_{2}=\text{Ext}_{\mathcal X}^{p}(\mathcal M,R^{q}f_{*}\mathcal E)\implies\text{Ext}_{\mathcal Y}^{p+q}(f^{*}\mathcal M,\mathcal E) \] with associated differentials \[ d_{2}^{\,i,j}: \text{Ext}_{\mathcal X}^{i}(\mathcal M,R^{j}f_{*}\mathcal E)\to\text{Ext}_{\mathcal Y}^{i+2}(\mathcal M,R^{j-1}f_{*}\mathcal E). \] Now, setting \(\mathcal M=R^{j}f_{*}\mathcal E\) in (1), we obtain a spectral sequence \[ \text{Ext}_{\mathcal X}^{p}(R^{j}f_{*}\mathcal E,R^{q}f_{*}\mathcal E)\implies\text{Ext}_{\mathcal Y}^{p+q}(f^{*}(R^{j}f_{*}\mathcal E),\mathcal E) \] with associated differentials \[ \partial=\partial_{2}^{0,j}:\text{Hom}_{\mathcal X}(R^{j}f_{*}\mathcal E,R^{j}f_{*}\mathcal E)\to\text{Ext}_{\mathcal X}^{2}(R^{j}f_{*}\mathcal E,R^{j-1}f_{*}\mathcal E). \] The first result of the paper is a description of \(d_{2}^{\,i,j}\) in terms of \(\partial\), namely, for any \(i\geq 0\), \(j\geq 1\) and \(\alpha\in\text{Ext}_{\mathcal X}^{i}(\mathcal M,R^{j}f_{*}\mathcal E)\), one has \(d_{2}^{\,i,j}(\alpha)=(-1)^{i}\alpha\cup\partial(\text{Id}_{j}), \) where \(\text{Id}_{j}\in\text{Hom}_{\mathcal X}(R^{j}f_{*}\mathcal E,R^{j}f_{*}\mathcal E)\) is the identity morphism and \(\cup\) is the Yoneda pairing \[ \text{Ext}_{\mathcal X}^{i}(\mathcal M,R^{j}f_{*}\mathcal E)\times\text{Ext}_{\mathcal X}^{2}(R^{j}f_{*}\mathcal E, R^{j-1}f_{*}\mathcal E)\to \text{Ext}_{\mathcal X}^{i+2}(\mathcal M,R^{j-1}f_{*}\mathcal E). \] Now let \(k\) be a field of characteristic 0 with algebraic closure \(\overline{k}\), let \(\Gamma=\text{Gal}(\overline{k}/k)\) and let \(X\) be a smooth and geometrically integral variety such that \(\overline{k}[X]^{*}=\overline{k}^{\,*}\). Setting above \(\mathcal Y\) equal to the category of étale sheaves on \(X\), \(\mathcal X\) equal to the category of discrete \(\Gamma\)-modules, \(f_{*}:\mathcal Y\to\mathcal X\) equal to the functor induced by the structure morphism \(p: X\to\text{Spec}(k)\), \(\mathcal E=\mathbb G_{m}\) and \(j=1\), one associates to \(X\) a canonical ``elementary obstruction'' \(e(X)=\partial^{\,0,1}(\text{Id}_{1})\in\text{Ext}^{2}_{k} \big(\text{Pic}\,\overline{X},\overline{k}^{\,*}\big)\). The negative of this class coincides with that of the well-known 2-extension of \(\Gamma\)-modules \[ 1\to \overline{k}^{\,*}\to \overline{k}(X)^{*}\to\text{Div}\,\overline{X}\to \text{Pic}\,\overline{X}\to 0 \] and it vanishes if \(X\) has a 0-cycle of degree one (since in that case \(\overline{k}^{\,*}\to \overline{k}(X)^{*}\) has a \(\Gamma\)-equivariant section). Then the preceding discussion yields the following result: \(\text{Br}_{1}X/\,\text{Br}_{0}\,X\) is canonically isomorphic to the subgroup of \(H^{1}\big(k,\text{Pic}\,\overline{X}\big)\) of all classes \(\xi\) such that \(\xi\cup e(X)=0\). Here \(\text{Br}_{1}\,X=\text{Ker}\big[\text{Br}\,X\to \text{Br}\,\overline{X}\,\big]\) and \(\text{Br}_{0}\,X\) is the image of \(\text{Br}\,k\) in \(\text{Br}\,X\) under the natural map. Assume now that \(X\) is complete and let \(J\) (resp. \(A\)) be its Picard (resp. Albanese) variety. By classical work of \textit{A. Weil} [Am. Math. J. 77, 493--512 (1955; Zbl 0065.14202)], there exists an \(A\)-torsor \(D\) over \(k\) and a map \(\text{Alb}: X\to D\) whose extension to \(\overline{k}\) is the classical Albanese map. Let \(\delta(X)=[D]\in H^{1}(K,A)=\text{Ext}^{2}_{k-\text{gps.}}(J,\mathbb G_{m})\). Another key result of this paper is the following: if \(i: J\big(\overline{k}\big)\hookrightarrow\text{Pic}\,\overline{X}\) is the natural embedding, then \(-i^{*}(e(X))\in\text{Ext}^{2}_{k}\big(J\big(\overline{k}\big), \overline{k}^{\,*}\big)\) is obtained from \(\delta(X)\) by applying the forgetful functor \(J\to J\big(\overline{k}\big)\) (the proof is interesting and not difficult to follow). Several interesting remarks follow the proof which discuss the question ``\(e(X)=0\Leftrightarrow\delta(X)=0\)?'' Another result of this paper is the following: assume in addition that \(X\) is a curve. Then \(\text{Br}\,X/\,\text{Br}_{0}\,X\) is canonically isomorphic to \(\delta(X)^{\bot}/\langle\delta(X)\rangle\), where \(\delta(X)^{\bot}\) is the orthogonal complement of \(\delta(X)\in H^{1}(k,J)\) under a certain canonical pairing \(H^{1}(k,J)\times H^{1}(k,A)\to H^{3}\big(k,\overline{k}^{\,*}\big)\). The last section of the paper discusses Brauer groups of torsors under semisimple algebraic groups over \(k\). We should mention that the results of this paper were applied in [\textit{M. Borovoi, J.-L. Colliot-Thélène} and \textit{A. N. Skorobogatov}, Duke Math. J. 141, No. 2, 321--364 (2008; Zbl 1135.14013)] to establish the equivalence of the vanishing of the elementary obstruction \(e(X)\) and the injectivity of the natural map \(\text{Br}\,k\to\text{Br}\,X\) over a local field \(k\). Finally, this paper is well-written. It was a pleasure to review it.