Poitou-Tate without restrictions on the order (Q530616)

For a global field \(K\) and a non empty (possibly infinite) set \(\Sigma\) of places of \(K\) containing the Archimedean places, the 9-term Poitou-Tate exact sequence (say PT) relates the Galois cohomology of a finite discrete \(\mathrm{Gal}(K_\Sigma/K)\)-module \(M\) to that of the dual module \(M^D:=\mathrm{Hom}(M,\mathbb{Q/Z} (1))\) via restricted products \(\prod'_{v\in\Sigma}(.)\) of local cohomology groups, under the condition that \(\#M\) is a unit at all places outside \(\Sigma\). The purpose of the present article is to get rid of this requirement on \(\#M\) by constructing a generalized Poitou-Tate exact sequence (say GPT) using fppf cohomology of adèles. More precisely, for a global field \(K\), let \(S\) be \(\mathrm{Spec }\mathcal O_K\) if \(\mathrm{char}(K )=0\), and if \(\mathrm{char}(K )\neq 0\), let \(S\) be the proper smooth curve over a finite field such that the function field of \(S\) is \(K\). Fix \(\Sigma\) as above. Write \((S -\Sigma)\) for the Dedekind scheme which is the localization of \(S\) away from \(\Sigma\), i.e., \((S-\Sigma)=\displaystyle\lim_{\leftarrow}U\) where \(U\) ranges over the non empty open sets \(S\) with \(S\backslash U\subset\Sigma\). For a finite subset \(\Sigma_0\subset\Sigma\), set \(\mathbb A^{\in\Sigma}_{K,\Sigma_0}:=\prod_{\Sigma_0} K_v\times\prod_{\Sigma\backslash\Sigma_0}\mathcal O_v\), define the ring of \(\Sigma\)-adèles \(\mathbb A^{\in\Sigma}_K:=\displaystyle\lim_{\rightarrow}\mathbb A^{\in\Sigma}_{K,\Sigma_0}\), and denote by \(H^n(\mathbb A_K^{\in\Sigma},.)\) its fppf cohomology. Write \((.)^\ast\) for the Pontryagin dual. Then for Cartier dual commutative finite flat \((S-\Sigma)\)-group schemes \(G\) and \(H\), the new GPT exact sequence reads: \[ H^0(S-\Sigma,G)\overset{\mathrm{loc}^0(G)}{\longrightarrow}\widehat H^0(\mathbb A^{\in\Sigma}_K,G)\overset{\mathrm{loc}^2(H)^\ast}{\longrightarrow}H^2(S-\Sigma,H)^\ast\overset{\alpha}{\rightarrow}H^1(S-\Sigma,G)\overset{\mathrm{loc}^1(G)}{\longrightarrow}H^1(\mathbb A^{\in\Sigma}_K,G) \] \[ \overset{\mathrm{loc}^1(H)^\ast}{\longrightarrow}H^1(S-\Sigma,H)^\ast\overset{\beta}{\rightarrow}H^2(S-\Sigma,G)\overset{\mathrm{loc}^2(G)}{\longrightarrow}H^2(\mathbb A^{\in\Sigma}_K,G)\overset{\mathrm{loc}^0(H)^\ast}{\longrightarrow}H^0(S-\Sigma,H)^\ast. \] Here a comparison theorem allows to interpret the cohomology of \(\mathbb A^{\in\Sigma}_K\) in terms of restricted products: for \(n\geq 0\), \(H^n(A^{\in\Sigma}_K,G)\cong\prod'_{v\in\Sigma}H^n(K_v,G)\), the restricted product being w.r.t. the open subgroups \(H^n(\mathcal O_v,\mathcal G)\subset H^n(K_v,G)\) where \(\mathcal G\to U\) is a model of \(G\) over a non empty open \(U\subset S\) and \(v\in U\cap\Sigma\). For \(n=0\), the Tate modified group \(\widehat H^0(\mathbb A^{\in\Sigma}_K,G)\) means that \(H^0(K_v,.)\) is replaced by \(\widehat H^0(K_v,.)\) when \(v\) is Archimedean. As for the connecting maps: the \(\mathrm{loc}^n(H)^\ast\) are defined using a perfect pairing \(H^1(\mathbb A^{\in\Sigma}_K,G)\times\overset{\mathrm{inv}}{\rightarrow}\mathbb{Q/Z}\), \(\alpha\) and \(\beta\) are obtained similarly from perfect \(\widehat H^0(\mathbb A^{\in\Sigma}_K,G)\times H^2(\mathbb A^{\in\Sigma}_K,H)\to\mathbb{Q/Z}\) and \(H^2(\mathbb A^{\in\Sigma}_K,G)\times\widehat H^0(\mathbb A^{\in\Sigma}_K,H)\to\mathbb{Q/Z}\); all the maps are continuous when \(H^n(S-\Sigma,G)\) and \(H^n(S-\Sigma,H)\) are endowed with the discrete topology. Appendix A explains how to recover (PT) from (GPT) : if \(\#M\) is a unit at all places outside \(\Sigma\), then \(M\) and \(M^D\) can be extended to Cartier dual commutative finite étale \((S-\Sigma)\)-group schemes whose fppf cohomology groups can be identified with Galois cohomology groups. In the generality permitted by (GPT), the author can also prove a Cassels-Poitou-Tate exact sequence which relates certain Selmer subgroups of \(H^1(S-\Sigma,G)\) to those of \(H^1(S-\Sigma,H)\). In the setting of (GPT), if each of \(\mathrm{Sel}(G_{\mathbb A^{\in\Sigma}_K})\) and \(\mathrm{Sel}(H_{\mathbb A_K^{\in\Sigma}})\) is either open or compact (when it is endowed with the subspace topology), one has a 7-term sequence: \(0\to\mathrm{Sel}(G)\to H^1(S-\Sigma,G)\to H^1(\mathbb A^{\in\Sigma}_K,G)/\mathrm{Sel}(G_ {\mathbb A^{\in\Sigma}_K})\overset{y(G)}{\longrightarrow}\mathrm{Sel}(H)^\ast\overset{x(G)}{\longrightarrow}H^2(S-\Sigma,G)\overset{\mathrm{loc}^2(G)}{\longrightarrow}H^2(\mathbb A^{\in\Sigma}_K,G)\overset{\mathrm{loc}^0(H)^\ast}{\longrightarrow}H^0(S-\Sigma,H)^\ast\), where \(y(G)\) is the dual of the restriction of \(\mathrm{loc}^1(H):\mathrm{Sel}(H)\to\mathrm{Sel}(H_{\mathbb A^{\in\Sigma}_K})\) and \(x(G)\) factors the map \(\beta\) above. The particular case of a finite \(\Sigma\) had previously been settled by the author in [``\(p\)-Selmer growth in extensions of degree \(p\)'', Preprint, \url{arXiv:1408.1151}].