A Simpler Approach to a Descent Conjecture of Wittenberg Yisheng TIAN Abstract A descent conjecture of Wittenberg [ Wit24 ] *Conjecture 3.7.4 predicts that if all the twists of a rationally connected torsor over a smooth base satisfy weak approximation with Brauer–Manin obstruction, then so does the base. We give an alternative proof of Wittenberg’s conjecture for certain torsors under connected linear groups via Cao’s descent formula. 1 Introduction The inverse Galois problem, asking whether any finite group G is a quotient of \gal(\overline{k}|k) for some number field k , is a fundamental open question in number theory. It has a positive answer when G is symmetric or alternating (Hilbert 1892 ) and G is solvable (Shafarevich 1954 ). Other classical results for sporadic groups or non-abelian simple groups of Lie type are summarized in [ Wit24 ] *Section 1.1. As of today, there are many approaches to the inverse Galois problem as mentioned in loc. cit. . In what follows, we shall proceed by a descent method developed by Colliot-Thélène–Sansuc, Harpaz–Wittenberg and others. Throughout, k is a number field. Let \Omega be the set of all places of k and let k_{v} be the completion of k at any v\in\Omega . By a k -variety X , we always mean a separated k -scheme of finite type. Manin [ Man71 ] introduced a pairing (see [ Sko01 ] *§5.2 for more information) \textstyle\prod\limits_{v\in\Omega}X(k_{v})\times\Br_{{\mathrm{nr}}}(X)\to\mathbb{Q}/\mathbb{Z},\ ((x_{v}),\alpha)\mapsto\textstyle\sum\limits_{v\in\Omega}j_{v}\circ x_{v}^{*}(\alpha), where \Br_{{\mathrm{nr}}}(X) is the unramified Brauer group of X (see 2.1 below), x_{v}^{*}:\Br_{{\mathrm{nr}}}(X)\to\Br(k_{v}) is the map induced by x_{v}\in X(k_{v}) and j_{v}:\Br(k_{v})\to\mathbb{Q}/\mathbb{Z} is the local invariant. Let X(k_{\Omega})^{\Br_{{\mathrm{nr}}}} be its left kernel which is a closed subset of X(k_{\Omega})\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\prod_{v\in\Omega}X(k_{v}) with respect to the product of v -adic topologies. A programmatic conjecture is the following Conjecture 1.1 (Colliot-Thélène, [ CT03 ] *p. 174) . Let X be a rationally connected smooth variety over k . Then X(k) is dense in X(k_{\Omega})^{\Br_{{\mathrm{nr}}}} . This conjecture is interesting in its own right as it predicts the geometry of X controls its arithmetic behavior. Moreover, a significant consequence of this conjecture is the inverse Galois problem . More precisely, by embedding a finite group G into the symmetric group \mathfrak{S}_{n} for some n\geq 1 , we may let G act on \mathbb{A}^{n}_{k} via the \mathfrak{S}_{n} -action. Subsequently, taking the open subset Y\subset\mathbb{A}_{k}^{n} consisting of points with pairwise distinct coordinates, we obtain a so-called versal G -torsor Y\to Y/G . In applying 1.1 to X\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}Y/G , we conclude a positive answer to the inverse Galois problem for G (see [ Wit24 ] *Section 3.4 for a detailed discussion). From this perspective, the investigation of the inverse Galois problem may be transferred into that of 1.1 , in which the descent method plays a key role. (1.2). Twisting of torsors. Let G be a linear algebraic group over k and let Y\to X be a left G -torsor. For any [\sigma]\in H^{1}(k,G) , let P{{}_{\sigma}}\to X be a right G -torsor representing [\sigma] . We write {{}_{\sigma}}Y\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}P{{}_{\sigma}}\times^{G}Y for the contracted product. Then the induced morphism {{}_{\sigma}}f:{{}_{\sigma}}Y\to X is a {{}_{\sigma}}G -torsor. Moreover, there is a bijection H^{1}(k,{{}_{\sigma}}G)\to H^{1}(k,G) of sets that sends the neutral class of H^{1}(k,{{}_{\sigma}}G) to [\sigma]\in H^{1}(k,G) . After introducing the twisting technique, the descent method leads us to the following Conjecture 1.3 (Wittenberg, [ Wit24 ] *Conjecture 3.7.4) . Let X be a smooth k -variety and let G be a linear algebraic k -group. Let f:Y\to X be a G -torsor with Y rationally connected. Assume that {{}_{\sigma}}Y(k) is dense in {{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)} for any [\sigma]\in H^{1}(k,G) . Then X(k) is dense in X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} . If G is a torus, the conjecture is known by the works [ CTS87b , HW20 ] . In general, Linh [ Lin26 ] proved it for any connected linear group G and any rationally connected k -variety X using the descent method developed by Harpaz–Wittenberg [ HW20 ] and the abelianization machinery of Borovoi [ Bor98 ] . In the present article, we give an alternative proof under weaker assumption based on the invariant Brauer subgroup introduced by Cao [ Cao18AF ] . Theorem 1.4 . Let G be a connected linear group over k . Let X be a smooth geometrically integral k -variety. Let f:Y\to X be a G -torsor and let Y^{c} be a smooth compactification of Y . Assume that \pi_{1}(Y^{c}_{\overline{k}})^{\ab}=0 and \Br(Y^{c})/\Image\Br(k) is finite. Then we have X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)}=\overline{\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)})}, where \overline{\Lambda} denotes the closure of \Lambda\subset X(k_{\Omega}) with respect to the product topology. In particular, if Y is rationally connected, then the assumption of 1.4 on Y is fulfilled (see the proof of 4.4 ). The main ingredients of the proof is the following descent formula of Cao [ Cao18AF ] *Théorème 5.9 X(\mathbf{A})^{A}=\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(\mathbf{A})^{B_{\sigma}+{{}_{\sigma}}f^{*}(A)}), where \mathbf{A} is the ring of adèles of k , A\subset\Br(X) is a subgroup and B_{\sigma}\subset\Br({{}_{\sigma}}Y) is a subgroup containing \Image\Br(k) . Subsequently, take B_{\sigma}=\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y) and A=(f^{*})^{-1}(B_{e}) , where [e]\in H^{1}(k,G) is the neutral class. In applying Harari’s formal lemma, we may identify X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} with \overline{X(\mathbf{A})^{A}} which yields the desired formula. (1.5). Acknowledgement. The author is grateful to Yang CAO for many insightful discussions and helpful comments. This article is supported by the grant of National Natural Science Foundation of China (no. 12401014). 2 The invariant Brauer subgroup Let k be a number field. Let k_{\Omega}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\prod_{v\in\Omega}k_{v} and let \mathbf{A} be the ring of adèles of k . Let G be a connected linear group over k . A k -variety is a separated k -scheme of finite type. Definition 2.1 . Let X be a smooth geometrically integral k -variety. (1) The Brauer group of X is defined as \Br(X)\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}H_{\mathrm{\acute{e}t}}^{2}(X,\mathbb{G}_{m}) . The arithmetic Brauer group of X is \Br_{a}(X)\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\frac{\Ker\big(\Br(X)\to\Br(X\times_{k}\overline{k})\big)}{\Image\big(\Br(k)\to\Br(X)\big)}. (2) Suppose X(k)\neq\varnothing . Take any x\in X(k) and let x^{*}:\Br(X)\to\Br(k) be the induced homomorphism. We put \Br_{x}(X)\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\Ker\big(\Br(X)\stackrel{{\scriptstyle x^{*}}}{{\to}}\Br(k)\big). In particular, if we denote by e the neutral element of G , then \Br_{e}(G) is defined. (3) Let \rho:G\times_{k}X\to X be a left G -action on X . After [ Cao18AF , Cao20 ] , the invariant Brauer subgroup of \Br(X) is defined to be \Br_{G}(X)\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\big\{b\in\Br(X)\,|\,(\rho^{*}(b)-p_{2}^{*}(b))\in p_{1}^{*}\Br(G)\big\}, where p_{1}:G\times_{k}X\to G and p_{2}:G\times_{k}X\to X are the canonical projections. (4) The Brauer group is a birational invariant for smooth proper k -varieties by [ CTS21 ] *Corollary 5.2.6. So we are allowed to define the unramified Brauer group \Br_{{\mathrm{nr}}}(X) of X to be the Brauer group of any smooth proper k -variety birationally equivalent to X . Proposition 2.2 . Let X be a smooth geometrically integral k -variety endowed with a left G -action. If \pi_{1}(X_{\overline{k}})^{\ab}=0 , then \Br_{G}(X)=\Br(X) . Proof. Since \pi_{1}(X_{\overline{k}})^{\ab}=0 , we deduce H^{1}(X_{\overline{k}},\mu_{n})=0 by [ SGA1 ] *Exposé XI, §5, ( \ast ). By [ Cao23 ] *Théorème 2.1, we obtain a canonical isomorphism of abelian groups (p_{1}^{*},p_{2}^{*}):H^{2}(G_{\overline{k}},\mu_{n})\ops H^{2}(X_{\overline{k}},\mu_{n})\to H^{2}(G_{\overline{k}}\times_{\overline{k}}X_{\overline{k}},\mu_{n}). The Hochschild–Serre spectral sequence H^{i}(k,H^{j}(-_{\overline{k}},\mu_{n}))\Rightarrow H^{i+j}(-,\mu_{n}) together with the 7 -term exact sequence in low degrees yields an isomorphism of abelian groups (p_{1}^{*},p_{2}^{*}):H^{2}_{e}(G,\mu_{n})\ops H^{2}(X,\mu_{n})\to H^{2}(G\times_{k}X,\mu_{n}). The Kummer exact sequence 0\to\mu_{n}\to\mathbb{G}_{m}\to\mathbb{G}_{m}\to 0 then implies the surjectivity of the induced map (p_{1}^{*},p_{2}^{*}):\Br_{e}(G)\ops\Br(X)\to\Br(G\times_{k}X). Observe that i=(e,\id):\e k\times_{k}X\to G\times_{k}X induces a map i^{*}:\Br(G\times_{k}X)\to\Br(X) such that i^{*}\circ p_{1}^{*}(\Br_{e}(G))=0 and i^{*}\circ p_{2}^{*}=\id . So we have \Image p_{1}^{*}\subset\Ker i^{*} . Conversely, take any \alpha\in\Ker i^{*}\subset\Br(G\times_{k}X) and suppose \alpha=p_{1}^{*}g+p_{2}^{*}x for some (g,x)\in\Br_{e}(G)\oplus\Br(X) . Then x=i^{*}\circ p_{2}^{*}(x)=i^{*}\circ(p_{1}^{*},p_{2}^{*})(g,x)=i^{*}(\alpha)=0 implies \alpha=p_{1}^{*}g , i.e., \Ker i^{*}\subset\Image p_{1}^{*} . Now take any b\in\Br(X) . Since \rho\circ i=p_{2}\circ i=\id_{X} , we conclude i^{*}(\rho^{*}b-p_{2}^{*}b)=0 which implies \rho^{*}b-p_{2}^{*}b\in\Ker i^{*}=\Image p_{1}^{*} . This shows b\in\Br_{G}(X) , i.e., \Br(X)\subset\Br_{G}(X) . ∎ 3 A descent formula of Cao (3.1). Let G be a connected linear group over k . Let Z be a smooth geometrically integral k -variety endowed with a left G -action. For any [\sigma]\in H^{1}(k,G) , let P_{\sigma} be a right G -torsor over k representing [\sigma] . Consider the contracted product and the projection {{}_{\sigma}}Z\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}P_{\sigma}\times_{k}^{G}Z\hskip 10.00002pt\ \textup{and}\hskip 10.00002pt\ \theta^{\sigma}_{Z}:P_{\sigma}\times_{k}Z\to{{}_{\sigma}}Z. By [ Cao18AF ] *Lemme 3.12, the projections p_{1}:{{}_{\sigma}}Z\times_{k}Z\to{{}_{\sigma}}Z and p_{2}:{{}_{\sigma}}Z\times_{k}Z\to Z induce a canonical isomorphism of abelian groups (p_{1}^{*},p_{2}^{*}):\Br_{a}({{}_{\sigma}}Z)\ops\frac{\Br_{G}(Z)}{\Image\Br(k)}\to\frac{\Br_{{{}_{\sigma}}G\times_{k}G}({{}_{\sigma}}Z\times_{k}Z)}{\Image\Br(k)}, which induces a further canonical homomorphism of abelian groups \textstyle{\displaystyle\Theta^{\sigma}_{Z}:\frac{\Br_{{}_{\sigma}G}({{}_{\sigma}}Z)}{\Image\Br(k)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{(\theta^{\sigma}_{Z})^{*}} \textstyle{\displaystyle\frac{\Br_{{{}_{\sigma}}G\times_{k}G}({{}_{\sigma}}P\times_{k}Z)}{\Image\Br(k)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\pr} \textstyle{\displaystyle\frac{\Br_{G}(Z)}{\Image\Br(k)}.} Lemma 3.2 ( [ Cao18AF ] *Lemme 5.8) . Let X be a smooth geometrically integral k -variety and let f:Y\to X be a left G -torsor. The map \Theta^{\sigma}_{Y} is an isomorphism for any [\sigma]\in H^{1}(k,G) . Similarly, for any [\tau]\in H^{1}(k,{{}_{\sigma}}G) , there is an isomorphism of abelian groups \Theta_{{{}_{\sigma}}Y}^{\tau}:\frac{\Br_{{{}_{\tau}}(_{\sigma}G)}({{}_{\tau}}(_{\sigma}Y))}{\Image\Br(k)}\to\frac{\Br_{{}_{\sigma}G}({{}_{\sigma}}Y)}{\Image\Br(k)}. (3.3). For any subgroup B_{\sigma}\subset\Br_{{{}_{\sigma}}G}({{}_{\sigma}}Y) containing \Image\Br(k) , let \overline{B_{\sigma}}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}B_{\sigma}\bmod{\Image\Br(k)} and denote by \widetilde{\Theta}_{Y}^{\sigma}(B_{\sigma})\subset\Br_{G}(Y) the preimage of \Theta^{\sigma}_{Y}(\overline{B_{\sigma}})\subset\Br_{G}(Y)/\Image\Br(k) . If we denote by [\sigma^{\prime}]\in H^{1}(k,{{}_{\sigma}}G) the class of the inverse torsor of P_{\sigma} (see [ Sko01 ] *p. 20, Example 2), then \widetilde{\Theta}_{{{}_{\sigma}}Y}^{\sigma^{\prime}}\circ\widetilde{\Theta}_{Y}^{\sigma}(B_{\sigma})=B_{\sigma} by [ Cao18AF ] *Lemme 5.8. Theorem 3.4 ( [ Cao18AF ] *Théorème 5.9) . Let X be a smooth geometrically integral k -variety and let f:Y\to X be a left G -torsor. For any [\sigma]\in H^{1}(k,G) , let B_{\sigma}\subset\Br_{{{}_{\sigma}}G}({{}_{\sigma}}Y) be a subgroup containing \Image\Br(k) . Let A\subset\Br(X) be a subgroup such that for any [\sigma]\in H^{1}(k,G) , ({{}_{\sigma}}f^{*})^{-1}\Big(\textstyle\sum\limits_{\tau\in\Sha^{1}({{}_{\sigma}}G)}\widetilde{\Theta}_{{{}_{\sigma}}Y}^{\tau}(B_{\sigma+\tau})\Big)\subset A, ( \dagger ) where B_{\sigma+\tau}\subset\Br_{{{}_{\tau}}({{}_{\sigma}}G)}({{}_{\tau}}({{}_{\sigma}}Y)) . Then the following descent formula holds X(\mathbf{A})^{A}=\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(\mathbf{A})^{B_{\sigma}+{{}_{\sigma}}f^{*}(A)}). In the sequel, we only need the following case where the condition ( † ‣ 3.4 ) is simplified. Lemma 3.5 . Let A\subset\Br(X) be a subgroup such that for each [\sigma]\in H^{1}(k,G) (f^{*})^{-1}(B)\subset A\hskip 10.00002pt\ \textup{and}\hskip 10.00002pt\ \Theta^{\sigma}_{Y}(\overline{B_{\sigma}})\subset\overline{B}, where B\subset\Br_{G}(Y) is the subgroup for the neutral class [e]\in H^{1}(k,G) . Then ( † ‣ 3.4 ) holds. Proof. For any [\sigma]\in H^{1}(k,G) and any [\tau]\in H^{1}(k,{{}_{\sigma}}G) , let [\sigma+\tau]\in H^{1}(k,G) denote the class corresponding to the neutral class of H^{1}(k,{{}_{\tau}}({{}_{\sigma}}G)) . By assumption, we conclude \widetilde{\Theta}_{Y}^{\sigma+\tau}(B_{\sigma+\tau})\subset B\subset f^{*}(A) . Then applying (\widetilde{\Theta}_{Y}^{\sigma})^{-1} yields \widetilde{\Theta}_{{}_{\sigma}Y}^{\tau}(B_{\sigma+\tau})\subset{{}_{\sigma}}f^{*}(A) . ∎ 4 Wittenberg’s descent conjecture (4.1). Throughout this section, let X be a smooth geometrically integral k -variety. Let G be a connected linear k -group and let f:Y\to X be a G -torsor. The next lemma is probably well-known. We still write it here for the lack of a reference. Lemma 4.2 . Keep the same notation as in Paragraph ( 4.1 ) . The Picard group \Pic(G) and the kernel \Ker(\Br(X)\to\Br(Y)) are finite. Proof. Let \rad^{u}(G) be the unipotent radical of G and let G^{\mathrm{red}}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}G/\rad^{u}(G) . The exact sequence 1\to\rad^{u}(G)\to G\to G^{\mathrm{red}}\to 1 induces an exact sequence \Pic(G^{\mathrm{red}})\to\Pic(G)\to\Pic(\rad^{u}(G)) by [ San81 ] *Corollaire 6.11. Since the underlying variety of \rad^{u}(G) is affine, the map \Pic(G^{\mathrm{red}})\to\Pic(G) is surjective. Thus it suffices to show that \Pic(G^{\mathrm{red}}) is finite. Let G^{\mathrm{ss}} be the derived subgroup of G^{\mathrm{red}} (which is semi-simple) and let G^{\mathrm{tor}}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}G^{\mathrm{red}}/G^{\mathrm{ss}} (which is a torus). There is an exact sequence \Pic(G^{\mathrm{tor}})\to\Pic(G^{\mathrm{red}})\to\Pic(G^{\mathrm{ss}}) by loc. cit. . By [ San81 ] *Lemme 6.9, we see that \Pic(G^{\mathrm{tor}})\simeq H^{1}(k,\mathbf{X}^{*}(G^{\mathrm{tor}})) (where \mathbf{X}^{*}(G^{\mathrm{tor}}) is the module of characters of G^{\mathrm{tor}} ) and that \Pic(G^{\mathrm{ss}}) is finite. But H^{1}(k,\mathbf{X}^{*}(G^{\mathrm{tor}})) is a finitely generated torsion abelian group, so it is finite. Consequently, \Pic(G^{\mathrm{red}}) is finite as well. Finally, due to [ San81 ] *(6.10.1) there is an exact sequence \Pic(G)\to\Br(X)\to\Br(Y) of abelian groups. Thus the finiteness of \Pic(G) yields that of \Ker(\Br(X)\to\Br(Y)) . ∎ Theorem 4.3 . Keep the same notation as in Paragraph ( 4.1 ) . Let Y^{c} be a smooth compactification of Y . If \pi_{1}(Y^{c}_{\overline{k}})^{\ab}=0 and \Br(Y^{c})/\Image\Br(k) is finite, then we have X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)}=\overline{\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)})}. (1) If {{}_{\sigma}}Y(k) is dense in {{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)} for any [\sigma] , then X(k) is dense in X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} . Proof. Recall that \pi_{1}(Y^{c}_{\overline{k}}) and \Br(Y^{c}) are birational invariants for smooth proper k -varieties (see [ SGA1 ] *Exposé X, Corollaire 3.4 and [ CTS21 ] *Corollary 5.2.6 respectively), so the assumptions on Y^{c} are independent of the choice of it. According to [ Bri22 ] *Theorem 2, we may choose Y^{c} such that the G -action on Y extends to it G -equivariantly. Subsequently, the twist {{}_{\sigma}}Y^{c} is a {{}_{\sigma}}G -equivariant smooth compactification of {{}_{\sigma}}Y for each [\sigma]\in H^{1}(k,G) . Let \Theta_{Y^{c}}^{\sigma} be the canonical homomorphism defined in ( 3.1 ) which makes the diagram commutative \begin{array}[]{c}\lx@xy@svg{\hbox{\raise 2.5pt\hbox{\kern 50.16339pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\halign{\entry@#!@&&\entry@@#!@\cr&\cr&\crcr}}}\ignorespaces{\hbox{\kern-50.16339pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{{\Br_{{{}_{\sigma}}G}({{}_{\sigma}}Y^{c})}/{\Image\Br(k)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 74.16339pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern-21.89653pt\raise-16.5pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.99225pt\hbox{$\scriptstyle{\Theta_{Y^{c}}^{\sigma}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 0.0pt\raise-24.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 74.16339pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{{\Br_{{{}_{\sigma}}G}({{}_{\sigma}}Y)}/{\Image\Br(k)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 122.28992pt\raise-16.5pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.08252pt\hbox{$\scriptstyle{\Theta_{Y}^{\sigma}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 122.28992pt\raise-24.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern-45.3085pt\raise-32.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{{\Br_{G}(Y^{c})}/{\Image\Br(k)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 77.6294pt\raise-32.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 77.6294pt\raise-32.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{{\Br_{G}(Y)}/{\Image\Br(k)}.}$}}}}}}}\ignorespaces}}}}\end{array} (2) For each [\sigma]\in H^{1}(k,G) , let B_{\sigma}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y) and B\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\Br_{{\mathrm{nr}}}(Y) . Let f^{*}:\Br(X)\to\Br(Y) be the induced map and let A\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}(f^{*})^{-1}(B)\subset\Br(X) . By assumption, we obtain \pi_{1}({{}_{\sigma}}Y^{c}_{\overline{k}})^{\ab}=0 and hence \Br({{}_{\sigma}}Y^{c})=\Br_{{{}_{\sigma}}G}({{}_{\sigma}}Y^{c}) by 2.2 . So we have B_{\sigma}\mathrel{\mathchoice{\vbox{\hbox{$\displaystyle:$}}}{\vbox{\hbox{$\textstyle:$}}}{\vbox{\hbox{$\scriptstyle:$}}}{\vbox{\hbox{$\scriptscriptstyle:$}}}{=}}\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)\simeq\Br({{}_{\sigma}}Y^{c})=\Br_{{}_{\sigma}G}({{}_{\sigma}}Y^{c})\subset\Br_{{}_{\sigma}G}({{}_{\sigma}}Y). It follows that \Theta^{\sigma}_{Y}(\overline{B_{\sigma}})=\Theta^{\sigma}_{Y^{c}}(\overline{B_{\sigma}})\subset\Br(Y^{c})/\Image\Br(k)=\overline{B} by ( 2 ). Thus \Theta^{\sigma}_{Y}(\overline{B_{\sigma}})=\overline{B} by ( 3.3 ) and A=(f^{*})^{-1}(B)=({{}_{\sigma}}f^{*})^{-1}(B_{\sigma}) . In particular, the conditions of 3.5 are fulfilled. Consequently, we deduce X(\mathbf{A})^{A}=\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(\mathbf{A})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)+{{}_{\sigma}}f^{*}(A)})=\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(\mathbf{A})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)}), where the last equality follows from {{}_{\sigma}}f^{*}(A)=\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y) . Thanks to 4.2 , the groups \Pic(G) and \Ker f^{*} are finite. Since \Br(Y^{c})/\Image\Br(k) is finite by assumption, the quotient (f^{*})^{-1}(\Br(Y^{c}))/\Image\Br(k)=A/\Image\Br(k) is also finite. Since \Br_{{\mathrm{nr}}}(X)\subset A by construction, we conclude \overline{X(\mathbf{A})^{A}}=X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} by Harari’s formal lemma [ Har94 ] *Corollaire 2.6.1 (see also [ Lin26 ] *Lemma 3.10(ii) for a detailed argument). Subsequently, we immediately deduce \overline{\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(\mathbf{A})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)})}=\overline{\textstyle\bigcup\limits_{[\sigma]\in H^{1}(k,G)}{{}_{\sigma}}f({{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)})}=X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)}, where the first equality follows from Harari’s formal lemma together with the finiteness of \Br({{}_{\sigma}}Y^{c})/\Image\Br(k) . The continuity of {{}_{\sigma}}f implies the density of {{}_{\sigma}}f({{}_{\sigma}}Y(k)) in {{}_{\sigma}}f({{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)}) . Hence X(k) is dense in \bigcup{{}_{\sigma}}f({{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)}) and we obtain \overline{X(k)}=X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} by ( 1 ). ∎ As an immediate consequence, we conclude the promised conjecture of Wittenberg. The argument is probably well-known, but we still give a complete proof for the convenient of the readers. Corollary 4.4 . let X be a smooth geometrically integral k -variety. Let G be a connected linear k -group and let f:Y\to X be a G -torsor with Y rationally connected. Assume that {{}_{\sigma}}Y(k) is dense in {{}_{\sigma}}Y(k_{\Omega})^{\Br_{{\mathrm{nr}}}({{}_{\sigma}}Y)} for any [\sigma]\in H^{1}(k,G) . Then X(k) is dense in X(k_{\Omega})^{\Br_{{\mathrm{nr}}}(X)} . Proof. Since X is geometrically integral and \tz(k)=0 , the G -torsor Y is also geometrically integral. Let Y^{c} be a smooth compactification of Y which is irreducible. Hence Chow’s lemma [ EGAII ] *Théorème 5.6.1 and Corollaire 5.6.2 yields a birational surjective morphism Y^{\prime}\to Y^{c} with projective Y^{\prime} . Let Y^{\prime\prime}\to Y^{\prime} be a resolution of singularities with projective Y^{\prime\prime} . Now Y^{\prime\prime} is a smooth projective variety that is birationally equivalent to Y^{c} , thus Y^{\prime\prime} is rationally connected. So we deduce \pi_{1}(Y^{c}_{\overline{k}})\simeq\pi_{1}(Y^{\prime\prime}_{\overline{k}})=0 where the vanishing follows from [ Kol96 ] *p. 200, Proposition (3.3.1) and [ Kol01 ] *Theorem 13. Since the group \Br(Y^{c})/\Image\Br(k) is finite by [ CTS21 ] *Corollary 4.4.4, 4.3 implies the density of X(k) , i.e., 1.3 holds. ∎ Yisheng TIAN Institute for Advanced Study in Mathematics Harbin Institute of Technology Harbin 150001, China Email: tysmath@mail.ustc.edu.cn