This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

The Brauer-Manin obstruction on algebraic stacks

Chang Lv State Key Laboratory of Information Security
Institute of Information Engineering
Chinese Academy of Sciences
Beijing 100093, P.R. China
lvchang@amss.ac.cn
 and  Han Wu Hubei Key Laboratory of Applied Mathematics, Faculty of Mathematics and Statistics, Hubei University, No. 368, Friendship Avenue, Wuchang District, Wuhan, Hubei, 430062, P.R.China. wuhan90@mail.ustc.edu.cn
Abstract.

For algebraic stacks over a number field, we define their Brauer-Manin pairings, Brauer-Manin sets, and extend the descent theory of Colliot-Thélène and Sansuc. By extending Sansuc’s exact sequence, we show the torsionness of Brauer groups of stacks that are locally quotients of varieties by connected groups. With mild assumptions, for stacks that are locally quotients or Deligne-Mumford, we show that the Brauer-Manin obstruction coincides with some other cohomological obstructions such as obstructions given by torsors under connected groups or abelian gerbes, generalizing Harari’s original results. For Brauer-Manin sets on these stacks, we show the properties such as descent along a torsor, product preservation are still correct. These results extend classical theories of those on varieties.

Key words and phrases:
Brauer-Manin obstruction, algebraic stacks
2000 Mathematics Subject Classification:
Primary 14F22, 14A20; secondary 14F20
C. Lv is partially supported by National Natural Science Foundation of China NSFC Grant No. 11701552. H. Wu is partially supported by NSFC Grant No. 12071448.

1. Introduction

Suppose that XX is a variety over a number field kk, and BrX=Hét2(X,𝐆m)\operatorname{Br}X=H_{\text{\'{e}t}}^{2}(X,\mathbf{G}_{m}) is the (cohomological) Brauer-Grothendieck group of XX [Gro68]. We have the Brauer-Manin pairing (see, for example, [Poo17, 8.2])

X(𝐀k)×BrX\displaystyle X(\mathbf{A}_{k})\times\operatorname{Br}X /,\displaystyle\rightarrow\mathbb{Q}/\mathbb{Z},
((xv)v,A)\displaystyle((x_{v})_{v},A) vΩkinvvA(xv),\displaystyle\mapsto\sum_{v\in\Omega_{k}}\operatorname{inv}_{v}A(x_{v}),

where 𝐀k\mathbf{A}_{k} (resp. Ωk\Omega_{k}) is the adèle (resp. the set of all places) of kk, invv\operatorname{inv}_{v} is the invariant map at vv, and (xv)-(x_{v}) is the pullback functor xv:BrXBrkvx_{v}^{*}:\operatorname{Br}X\rightarrow\operatorname{Br}k_{v}. Under this pairing, the Brauer-Manin set (obstruction) is defined to be the subset X(𝐀k)BrX(𝐀k)X(\mathbf{A}_{k})^{\operatorname{Br}}\subseteq X(\mathbf{A}_{k}) orthogonal to the whole BrX\operatorname{Br}X, and plays an important role in judging the local-global principle.

As the development of moduli theory, some geometric objects, other than schemes were used to enlarge the category of schemes. Artin [Art70], Deligne and Mumford [DM69] defined algebraic stacks, and used them to explain some moduli problems. The arithmetic of these objects are of their own interest. Given an algebraic stack 𝒳\mathcal{X} over 𝒪K\mathcal{O}_{K}, we say 𝒳\mathcal{X} violates the local-global principle for integral points if 𝒳(𝒪v)\mathcal{X}(\mathcal{O}_{v})\neq\emptyset for all vΩKv\in\Omega_{K}, whereas 𝒳(𝒪K)=\mathcal{X}(\mathcal{O}_{K})=\emptyset. Counterexamples to local-global principle for integral points are given in [DG95], [BP20], [WL22]. Bhargava and Poonen [BP20] proved that for any stacky curve over 𝒪K\mathcal{O}_{K} of genus less than 1/21/2, it satisfies local-global principle for integral points. Furthermore, Christensen [Chr20] extended the adèlic topology to the set 𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k}), and proved that for any stacky curve over 𝒪K\mathcal{O}_{K} of genus less than 1/21/2, it satisfies strong approximation. More recently, Santens [San22] showed that the Brauer-Manin obstruction to strong approximation is the only one for every stacky curve over a global field with finite abelian fundamental group.

In this paper, we consider the Brauer-Manin obstruction on algebraic stacks over a number field. In Section 2, we introduce points, cohomological obstructions and torsors on algebraic stacks. In Section 3, by extending Sansuc’s exact sequence, we calculate the Brauer group of stacks that are quotients of varieties by connected groups, which contains almost all geometric objects appearing in moduli spaces. As a consequence, we also show that the Brauer groups of some locally quotient stacks are still torsion, extending the classical results on smooth varieties. In Section 4, for general algebraic stack we establish the descent theory of Colliot-Thélène and Sansuc [CTS87] and define the Brauer-Manin pairing for stacks. In Section 5, we show that with mild assumptions, for stacks that are locally quotients or Deligne-Mumford, the Brauer-Manin obstruction coincides with some other cohomological obstructions such as obstructions given by torsors under connected groups or abelian gerbes, which was originally proved by Harari [Har02] for varieties. In Section 6, we generalize some results on descent for Brauer-Manin set along a torsor under a connected group. In Section 7, we prove that in the same cases as in Section 5, the Brauer-Manin set of product stacks equals the product of the Brauer-Manin sets of stacks. These results extend classical theories of those on varieties. The main technique for extending them is cohomological descent.

2. Points, obstructions and torsors on algebraic stacks

Let us briefly introduce basic notions and facts used in this paper, mainly for cohomological obstructions to local-global principle on algebraic stacks, under the framework [Lv21, Sec. 2] but in a slightly modified way.

For a commutative ring AA, we also write AA for SpecA\operatorname{Spec}A. Let SS be a scheme, 𝒳\mathcal{X} an algebraic stack over SS, in the sense of [Sta22, Tag 026O]. Denote by hp/S{\mathbb{C}\text{hp}}/S the 22-category of algebraic stacks over SS.

2.1. Points and obstructions

Let kk be a number field, RR a commutative kk-algebra, 𝒳\mathcal{X} an algebraic stack over kk. The RR-points of 𝒳\mathcal{X} is the set of isomorphism classes of objects in the groupoid Homhp/k(R,𝒳)\operatorname{Hom}_{{\mathbb{C}\text{hp}}/k}(R,\mathcal{X}) denoted by 𝒳(R)\mathcal{X}(R). Let 𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k}) (resp. 𝒳(k)\mathcal{X}(k)) be the adèlic points (resp. rational points) of 𝒳\mathcal{X}.

Definition 2.1.

Let F:𝒟F:\mathbb{C}\rightarrow\mathcal{D} be a functor from a 22-category to an ordinary category. We say that FF is stable if FF forgets 22-morphisms, and for any 11-morphisms ff and gg of \mathbb{C} that is 22-isomorphic, we have F(f)=F(g)F(f)=F(g). When 𝒟\mathcal{D} is viewed as a 22-category, this is equivalent to say that, FF is a 22-functor.

For a stable functor F:(hp/k)opSetF:({\mathbb{C}\text{hp}}/k)^{\text{op}}\rightarrow{Set} and AF(𝒳)A\in F(\mathcal{X}), as the classical case for varieties in Poonen [Poo17, 8.1.1], one may also define the obstruction given by AA to be the subset 𝒳(𝐀k)A\mathcal{X}(\mathbf{A}_{k})^{A} of 𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k}) whose elements are characterized by

(2.2) 𝒳(𝐀k)A={x𝒳(𝐀k)A(x)imF(q)},\mathcal{X}(\mathbf{A}_{k})^{A}=\{x\in\mathcal{X}(\mathbf{A}_{k})\mid A(x)\in\operatorname{im}F(q)\},

(which is well-defined by stability of FF), and the FF-set (or FF-obstruction) to be the subset 𝒳(𝐀k)F\mathcal{X}(\mathbf{A}_{k})^{F} whose elements are characterized by

𝒳(𝐀k)F=AF(𝒳)𝒳(𝐀k)A={x𝒳(𝐀k)imF(x)imF(q)}.\mathcal{X}(\mathbf{A}_{k})^{F}=\bigcap_{A\in F(\mathcal{X})}\mathcal{X}(\mathbf{A}_{k})^{A}=\{x\in\mathcal{X}(\mathbf{A}_{k})\mid\operatorname{im}F(x)\subseteq\operatorname{im}F(q)\}.

Then we have the map induced by qq,

𝒳(k)𝒳(𝐀k)F𝒳(𝐀k)A𝒳(𝐀k).\mathcal{X}(k)\to\mathcal{X}(\mathbf{A}_{k})^{F}\subseteq\mathcal{X}(\mathbf{A}_{k})^{A}\subseteq\mathcal{X}(\mathbf{A}_{k}).
Remark 2.3.

If 𝒳\mathcal{X} is represented by a scheme, then these definitions coincide with the classical ones. But in general, the map 𝒳(k)𝒳(𝐀k)\mathcal{X}(k)\rightarrow\mathcal{X}(\mathbf{A}_{k}) is not necessary injective. For example, let GG be a linear kk-group. Then

ker(BG(k)BG(𝐀k))=ker(H1(k,G)Hˇfppf1(𝐀k,G))=X1(G/k)\ker(BG(k)\rightarrow BG(\mathbf{A}_{k}))=\ker(H^{1}(k,G)\rightarrow\check{H}_{\text{fppf}}^{1}(\mathbf{A}_{k},G))=\cyrille X^{1}(G/k)

does not necessarily vanish.

Anyway, by abusing notation we may let 𝒳(k)\mathcal{X}(k) be its image in 𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k}). Thus we may write 𝒳(k)𝒳(𝐀k)F\mathcal{X}(k)\subset\mathcal{X}(\mathbf{A}_{k})^{F}.

2.2. Cohomology on algebraic stacks and cohomological obstructions

For any 𝒯hp/S\mathcal{T}\in{\mathbb{C}\text{hp}}/S where SS is a scheme, denote by 𝒯ét\mathcal{T}_{\text{\'{e}t}} (resp. 𝒯fppf\mathcal{T}_{\text{fppf}}) the site with topology inherited from the big étale site (Sch/S)ét({Sch}/S)_{\text{\'{e}t}} (resp. the big fppf site (Sch/S)fppf({Sch}/S)_{\text{fppf}}), cf. [Sta22, Tags 067N and 06TP].

Let τ{ét,fppf}\tau\in\{\text{\'{e}t},\text{fppf}\} and f:𝒯𝒯f:{\mathcal{T}^{\prime}}\rightarrow\mathcal{T} be a 11-morphism in hp/S{\mathbb{C}\text{hp}}/S. Then compositing with ff gives the functor f:PSh(𝒯)PSh(𝒯)f^{*}:{PSh}(\mathcal{T})\rightarrow{PSh}(\mathcal{T}^{\prime}), which has a right adjoint ff_{*}. Both ff^{*} and ff_{*} send τ\tau-sheaves to τ\tau-sheaves, and define a morphism of topoi

f=(f,f):Sh(𝒯τ)Sh(𝒯τ),f=(f_{*},f^{*}):{Sh}(\mathcal{T}^{\prime}_{\tau})\rightarrow{Sh}(\mathcal{T}_{\tau}),

which is also a pair of adjoint functors with Sh{Sh} replaced by Ab{Ab} (or SGrp{SGrp} the category of sheaves of groups) and is compatible with the classical case when 𝒯\mathcal{T} and 𝒯\mathcal{T}^{\prime} are represented by SS-schemes, cf. [Sta22, Tags 06TS and 075J].

We have an adjoint pair

(2.4) (f,Rf):Dcart(𝒯τ)Dcart(𝒯τ)(f^{*},Rf_{*}):D_{\text{cart}}(\mathcal{T}^{\prime}_{\tau})\rightleftharpoons D_{\text{cart}}(\mathcal{T}_{\tau})

between derived categories of complexes of Cartesianτ\tau-sheaves, c.f. [Sta22, Tag 073S].

Remark 2.5.

By [LZ17, Rmk. 5.3.10], for τ=ét\tau=\text{\'{e}t}, (2.4) is equivalent to (f,f):Dcart(𝒯lis-ét)Dcart(𝒯lis-ét)(f^{*},f_{*}):D_{\text{cart}}(\mathcal{T}^{\prime}_{\text{lis-\'{e}t}})\rightleftharpoons D_{\text{cart}}(\mathcal{T}_{\text{lis-\'{e}t}}).

We also have

RΓ(𝒯,):Dcart(𝒯τ)D(Ab)R\Gamma(\mathcal{T},-):D_{\text{cart}}(\mathcal{T}_{\tau})\rightarrow D({Ab})

Then we have the cohomology functors

Hi(𝒯τ,):Ab(𝒯τ)Ab.H^{i}(\mathcal{T}_{\tau},-):{Ab}(\mathcal{T}_{\tau})\rightarrow{Ab}.

We will write Hτi(𝒯,)=Hi(𝒯τ,)H_{\tau}^{i}(\mathcal{T},-)=H^{i}(\mathcal{T}_{\tau},-). For TSch/ST\in{Sch}/S, Hτi(T,)H_{\tau}^{i}(T,-) recovers the τ\tau-cohomology for schemes over SS. See [Sta22, Tag 075F].

Now we consider the case where S=kS=k a number field, and fix 𝒢Ab((Sch/k)τ)\mathscr{G}\in{Ab}(({Sch}/k)_{\tau}) and let 𝒢𝒯\mathscr{G}_{\mathcal{T}} be the pullback of 𝒢\mathscr{G} to 𝒯\mathcal{T}. Then the functor Hi(,𝒢):(hp/k)opSetH^{i}(-,\mathscr{G}_{-}):({\mathbb{C}\text{hp}}/k)^{\text{op}}\rightarrow{Set} is stable (c.f. [Lv21, 2.29]). If there is no other confusion, we simply write 𝒢Ab(𝒯τ)\mathscr{G}\in{Ab}(\mathcal{T}_{\tau}) for 𝒢𝒯\mathscr{G}_{\mathcal{T}}. Thus we can take F=Hi(,𝒢)F=H^{i}(-,\mathscr{G}) in 2.1 and obtain 𝒳(𝐀k)Hi(𝒳,𝒢)\mathcal{X}(\mathbf{A}_{k})^{H^{i}(\mathcal{X},\mathscr{G})}, the Hi(,𝒢)H^{i}(-,\mathscr{G})-set.

Example 2.6 (The Brauer-Manin obstruction).

The cohomological Brauer-Grothendieck group is also defined for any algebraic stack 𝒯\mathcal{T} over SS. It is the étale cohomology group Br𝒯:=Hét2(𝒯,𝐆m,𝒯)\operatorname{Br}\mathcal{T}:=H_{\text{\'{e}t}}^{2}(\mathcal{T},\mathbf{G}_{m,\mathcal{T}}), where 𝐆m,𝒯\mathbf{G}_{m,\mathcal{T}} is the pullback sheaf of 𝐆m,S\mathbf{G}_{m,S} to 𝒯\mathcal{T}, represented by 𝐆m,S×S𝒯\mathbf{G}_{m,S}\times_{S}\mathcal{T}. When 𝒯\mathcal{T} is a scheme, this agrees with the cohomological Brauer-Grothendieck group. The Brauer group of a stack is considered, for example, by Bertolin and Galluzzi [BG19, Thm. 3.4], and Zahnd [Zah03, 4.3]. Let S=kS=k and F=Br=Hét2(,𝐆m)F=\operatorname{Br}=H_{\text{\'{e}t}}^{2}(-,\mathbf{G}_{m}), we obtain the Brauer-Manin set 𝒳(𝐀k)Br\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}. If 𝒳\mathcal{X} is represented by a scheme, it coincides with the usual Brauer-Manin set.

2.3. Torsors over algebraic stacks

Fix an base scheme SS. Let 𝒳hp/S\mathcal{X}\in{\mathbb{C}\text{hp}}/S and 𝒢\mathscr{G} an sheaf of group on 𝒳τ\mathcal{X}_{\tau}, where τ{ét,fppf}\tau\in\{\text{\'{e}t},\text{fppf}\}.

Definition 2.7.

A 𝒢\mathscr{G}-torsor over 𝒳τ\mathcal{X}_{\tau} is a sheaf 𝒴\mathscr{Y} on 𝒳τ\mathcal{X}_{\tau} acted on by 𝒢\mathscr{G}, such that

𝒢×𝒴𝒴×𝒴,\mathscr{G}\times\mathscr{Y}\cong\mathscr{Y}\times\mathscr{Y},

denoted by 𝒴𝒢𝒳τ\mathscr{Y}\xrightarrow{\mathscr{G}}\mathcal{X}_{\tau}. A morphisms between 𝒢\mathscr{G}-torsors over 𝒳τ\mathcal{X}_{\tau} is a 𝒢\mathscr{G}-equivariant morphism of sheaves, which is necessarily an isomorphism.

We call 𝒴\mathscr{Y} trivial if it has a global section (or equivalently, 𝒴𝒢\mathscr{Y}\cong\mathscr{G} as 𝒢\mathscr{G}-torsor over 𝒳τ\mathcal{X}_{\tau}).

Denote the groupoid of all 𝒢\mathscr{G}-torsor over 𝒳τ\mathcal{X}_{\tau} by 𝖳𝗈𝗋𝗌(𝒳τ,𝒢)\mathsf{Tors}(\mathcal{X}_{\tau},\mathscr{G}). The isomorphism class of torsors makes 𝖳𝗈𝗋𝗌(𝒳τ,𝒢)/\mathsf{Tors}(\mathcal{X}_{\tau},\mathscr{G})_{/\cong} a pointed set whose neutral element correspondences to the class of trivial torsors.

By [Ols16, Ex. 2.H], we may assume 𝒳\mathcal{X} is an object of 𝒳τ\mathcal{X}_{\tau}. Then there is a one-to-one correspondence (c.f. Giraud [Gir71, III.3.6.5 (5)]) of pointed sets

(2.8) 𝖳𝗈𝗋𝗌(𝒳τ,𝒢)/Hˇτ1(𝒳,𝒢).\mathsf{Tors}(\mathcal{X}_{\tau},\mathscr{G})_{/\cong}\overset{\sim}{\rightarrow}\check{H}_{\tau}^{1}(\mathcal{X},\mathscr{G}).

In the case of 𝒢Ab(𝒳τ)\mathscr{G}\in{Ab}(\mathcal{X}_{\tau}), we have a bijection (c.f. [Gir71, IV.3.4.2 (i)])

(2.9) Hˇτ1(𝒳,𝒢)Hτ1(𝒳,𝒢),\check{H}_{\tau}^{1}(\mathcal{X},\mathscr{G})\overset{\sim}{\rightarrow}H_{\tau}^{1}(\mathcal{X},\mathscr{G}),

which maps the neutral element to 0 and is functorial both in 𝒳\mathcal{X} and 𝒢\mathscr{G}.

Since in particular a sheaf 𝒴\mathscr{Y} on 𝒳τ\mathcal{X}_{\tau} is a stack over 𝒳τ\mathcal{X}_{\tau}, one obtain a morphism of stacks 𝒴𝒳\mathscr{Y}\rightarrow\mathcal{X} over SS (c.f. [Sta22, Tag 09WX]). If f:𝒵𝒳f:\mathcal{Z}\rightarrow\mathcal{X} is a morphism of algebraic SS-stacks, then obviously the pullback 𝒴𝒵=𝒵×𝒳𝒴\mathscr{Y}_{\mathcal{Z}}=\mathcal{Z}\times_{\mathcal{X}}\mathscr{Y} is also a 𝒢\mathscr{G}-torsor. It is the pullback sheaf of 𝒴\mathscr{Y} under f:Sh(𝒳τ)Sh(𝒵τ)f^{*}:{Sh}(\mathcal{X}_{\tau})\rightarrow{Sh}(\mathcal{Z}_{\tau}). This operation corresponds to the cohomological pullback

f:Hˇτ1(𝒳,𝒢)Hˇτ1(𝒵,𝒢).f^{*}:\check{H}_{\tau}^{1}(\mathcal{X},\mathscr{G})\rightarrow\check{H}_{\tau}^{1}(\mathcal{Z},\mathscr{G}).
Lemma 2.10.

Let GG be an SS-group scheme (viewed as a sheaf of group on 𝒳τ\mathcal{X}_{\tau} by pullback). Then any 𝒴𝖳𝗈𝗋𝗌(𝒳τ,G)\mathscr{Y}\in\mathsf{Tors}(\mathcal{X}_{\tau},G) is in hp/S{\mathbb{C}\text{hp}}/S

In particular, 𝒴𝖳𝗈𝗋𝗌(𝒳τ,G)\mathcal{Y}\in\mathsf{Tors}(\mathcal{X}_{\tau},G) if and only if 𝒳[𝒴/G]\mathcal{X}\xrightarrow{\sim}[\mathcal{Y}/G] is the quotient stack.

Proof.

We imitate the proof from a note of M. Olsson. To verify the representability of the diagonal 𝒴𝒴×S𝒴\mathscr{Y}\rightarrow\mathscr{Y}\times_{S}\mathscr{Y}, we first show the representability of 𝒴𝒴×𝒳𝒴\mathscr{Y}\rightarrow\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y}. Let p1×p2:T𝒴×𝒳𝒴p_{1}\times p_{2}:T\rightarrow\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y} be a 11-morphism from an SS-scheme. One checks that the fiber product 𝒵=𝒴×𝒴×𝒳𝒴T\mathcal{Z}=\mathscr{Y}\times_{\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y}}T fits into the 22-Cartesian diagram

𝒵\textstyle{\mathcal{Z}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}T\textstyle{T\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(p1×idT)×(p2×idT)\scriptstyle{(p_{1}\times\operatorname{id}_{T})\times(p_{2}\times\operatorname{id}_{T})}𝒴×𝒳T\textstyle{\mathscr{Y}\times_{\mathcal{X}}T\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Δ\scriptstyle{\Delta}(𝒴×𝒳T)×𝒳(𝒴×𝒳T).\textstyle{(\mathscr{Y}\times_{\mathcal{X}}T)\times_{\mathcal{X}}(\mathscr{Y}\times_{\mathcal{X}}T).}

Note that 𝒴×𝒳TT\mathscr{Y}\times_{\mathcal{X}}T\rightarrow T has a section (for example, p1×idTp_{1}\times\operatorname{id}_{T}) as a GG-torsor over TτT_{\tau}. Thus it is a trivial torsor in 𝖳𝗈𝗋𝗌(Tτ,G)\mathsf{Tors}(T_{\tau},G) and in particular 𝒴×𝒳TG×ST\mathscr{Y}\times_{\mathcal{X}}T\cong G\times_{S}T is a scheme. It follows that 𝒵\mathcal{Z} is in Esp/S{Esp}/S (the category of algebraic SS-spaces) and 𝒴𝒴×𝒳𝒴\mathscr{Y}\rightarrow\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y} is representable. Also 𝒴×𝒳𝒴𝒴×S𝒴\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y}\rightarrow\mathscr{Y}\times_{S}\mathscr{Y} is representable since it is a base change of the diagonal 𝒳𝒳×S𝒳\mathcal{X}\rightarrow\mathcal{X}\times_{S}\mathcal{X}. It follows that that the diagonal 𝒴𝒴×S𝒴\mathscr{Y}\rightarrow\mathscr{Y}\times_{S}\mathscr{Y} is representable since it is the composition 𝒴𝒴×𝒳𝒴𝒴×S𝒴\mathscr{Y}\rightarrow\mathscr{Y}\times_{\mathcal{X}}\mathscr{Y}\rightarrow\mathscr{Y}\times_{S}\mathscr{Y}.

To obtain an atlas of 𝒴\mathscr{Y}, one note that 𝒴\mathscr{Y} is locally trivial (c.f. [Gir71, 1.4.1.1]). Then there is a covering {Ui𝒳}\{U_{i}\rightarrow\mathcal{X}\} in 𝒳τ\mathcal{X}_{\tau} (where we may assume UiU_{i} are schemes) such that 𝒴×𝒳UiG×SUi\mathscr{Y}\times_{\mathcal{X}}U_{i}\cong G\times_{S}U_{i}. Thus Y=iG×SUiY=\coprod_{i}G\times_{S}U_{i} is a scheme and Yi𝒴×𝒳Ui𝒴Y\cong\coprod_{i}\mathscr{Y}\times_{\mathcal{X}}U_{i}\rightarrow\mathscr{Y} is a fppf atlas for 𝒴\mathscr{Y}. It follows by [Sta22, Tag 06DC] that 𝒴hp/S\mathscr{Y}\in{\mathbb{C}\text{hp}}/S. ∎

In the following, if not otherwise explained, a 𝒢\mathscr{G}-torsor over 𝒳\mathcal{X} means a 𝒢\mathscr{G}-torsor over 𝒳fppf\mathcal{X}_{\text{fppf}}.

For an algebraic stack 𝒳\mathcal{X} over kk and a linear kk-group GG, contracted products (resp. twists) of GG-torsors over 𝒳\mathcal{X} can also be defined. See [Lv21, 3.24 (resp. 3.30)].

Let f:𝒴𝐺𝒳f:\mathcal{Y}\xrightarrow{G}\mathcal{X} be a torsor. Since the functor Hˇfppf1(,G):(hp/k)opSet\check{H}_{\text{fppf}}^{1}(-,G):({\mathbb{C}\text{hp}}/k)^{\text{op}}\rightarrow{Set} is also stable by [Lv21, 3.14], we may define 𝒳(𝐀k)f𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k})^{f}\subseteq\mathcal{X}(\mathbf{A}_{k}) to be the obstruction set given by [𝒴][\mathcal{Y}] (c.f. (2.2)). It contains 𝒳(k)\mathcal{X}(k) and satisfies

𝒳(𝐀k)f=σH1(k,G)fσ(𝒴σ(𝐀k)),\mathcal{X}(\mathbf{A}_{k})^{f}=\bigcup_{\sigma\in H^{1}(k,G)}f_{\sigma}(\mathcal{Y}_{\sigma}(\mathbf{A}_{k})),

where fσ:𝒴σ𝒳𝖳𝗈𝗋𝗌(𝒳,Gσ)f_{\sigma}:\mathcal{Y}_{\sigma}\rightarrow\mathcal{X}\in\mathsf{Tors}(\mathcal{X},G_{\sigma}) is the twist of 𝒴\mathcal{Y} by σ\sigma. See [Lv21, 3.20] for a proof.

Again, if 𝒳\mathcal{X} is a kk-scheme, the definitions and facts of this subsection coincide with classical ones.

3. Calculation of Brauer groups

Many important algebraic stacks can be realized as quotients of schemes by group actions. In this section, we calculate their Brauer groups.

Let kk be a field of characterized zero. Then any algebraic kk-group is smooth.

3.1. Sansuc’s exact sequence

Sansuc [San81, Prop. 6.10] gave an exact sequence, and now we generalize it to quotient stacks.

Theorem 3.1.

Let XX be a kk-variety and GG a connected kk-group acting on XX. Let 𝒴=[X/G]\mathcal{Y}=[X/G] be the quotient stack, i.e., f:X𝒴f:X\rightarrow\mathcal{Y} is a GG-torosr over an algebraic stack 𝒴\mathcal{Y} (see Lemma 2.10). Let U=𝐆m/k×,Pic=Hét1(,𝐆m)PSh(Sch/k)\operatorname{U}=\mathbf{G}_{m}/k^{\times},\ \operatorname{Pic}=H_{\text{\'{e}t}}^{1}(-,\mathbf{G}_{m})\in{PSh}({Sch}/k), where k×k^{\times} is the constant presheaf. Then we have an exact sequence

(3.2) 0U𝒴fUXUGPic𝒴fPicXPicGBr𝒴fBrXρp2Br(G×kX),0\rightarrow\operatorname{U}\mathcal{Y}\xrightarrow{f^{*}}\operatorname{U}X\rightarrow\operatorname{U}G\rightarrow\operatorname{Pic}\mathcal{Y}\xrightarrow{f^{*}}\operatorname{Pic}X\rightarrow\operatorname{Pic}G\rightarrow\operatorname{Br}\mathcal{Y}\xrightarrow{f^{*}}\operatorname{Br}X\xrightarrow{\rho^{*}-p_{2}^{*}}\operatorname{Br}(G\times_{k}X),

where ρ,p2:G×kXX\rho,\ p_{2}:G\times_{k}X\rightarrow X is the action of GG and the projection to XX, respectively.

Proof.

Let XX_{\bullet} be a Čech nerve of X/𝒴X/\mathcal{Y}. Note that for n0n\geq 0, XnX_{n} is equivalent to the (n+1)(n+1)-fold 22-fibred product X×𝒴X×𝒴×𝒴X=X×kGnX\times_{\mathcal{Y}}X\times_{\mathcal{Y}}\cdots\times_{\mathcal{Y}}X=X\times_{k}G^{n}, which is a kk-variety. Then [San81, Lem. 6.12] still works in our case and its assumptions verify for F=UF=U and Pic\operatorname{Pic}. Now for any Sh(𝒴ét)\mathscr{F}\in{Sh}(\mathcal{Y}_{\text{\'{e}t}}), since X𝒴X\rightarrow\mathcal{Y} is smooth surjective, by a consequence of cohomological descent, there is a spectral sequence

E1p,q=Hétq(Xp,)Hétp+q(𝒴,),E^{p,q}_{1}=H^{q}_{\text{\'{e}t}}(X_{p},\mathscr{F})\Rightarrow H_{\text{\'{e}t}}^{p+q}(\mathcal{Y},\mathscr{F}),

whose E2E_{2}-page is

(3.3) E2p,q=Hˇp(X/𝒴,q())Hétp+q(𝒴,).E^{p,q}_{2}=\check{H}^{p}(X/\mathcal{Y},\mathscr{H}^{q}(\mathscr{F}))\Rightarrow H_{\text{\'{e}t}}^{p+q}(\mathcal{Y},\mathscr{F}).

See (1) of [Sta22, Tag 06XJ]. Taking =𝐆m\mathscr{F}=\mathbf{G}_{m} we obtain the corresponding seven-term exact sequence.

Then the remaining argument is the same as Sansuc’s original one (c.f. [San81, p. 45]). ∎

Remark 3.4.

If in addition GG is linear, we have U(G)=G^(k)U(G)=\hat{G}(k) as in [San81, Prop. 6.10].

Corollary 3.5.

Let GG be a connected kk-group. Then we have UBG=0\operatorname{U}BG=0, PicBG=UG\operatorname{Pic}BG=\operatorname{U}G and a splitting short exact sequence 0PicGBrBGBrk00\rightarrow\operatorname{Pic}G\rightarrow\operatorname{Br}BG\rightarrow\operatorname{Br}k\rightarrow 0.

Proof.

Take X=kX=k in Theorem 3.1 and note that the atlas kBGk\rightarrow BG is a section of the structure morphism BGkBG\rightarrow k. ∎

In general for a torsor 𝒳𝐺𝒴\mathcal{X}\xrightarrow{G}\mathcal{Y} where 𝒴hp/k\mathcal{Y}\in{\mathbb{C}\text{hp}}/k and GG a smooth kk-group, let ρ,p2:𝒳×kG𝒳\rho,\ p_{2}:\mathcal{X}\times_{k}G\rightarrow\mathcal{X} be as in Theorem 3.1, and 𝔭1:𝒳×kGG\mathfrak{p}_{1}:\mathcal{X}\times_{k}G\rightarrow G be the projection to GG. Following [Cao18, Def. 3.1], we may also define the invariant Brauer subgroup to be

BrG𝒳={bBr𝒳ρbp2bp1BrG}\operatorname{Br}_{G}\mathcal{X}=\{b\in\operatorname{Br}\mathcal{X}\mid\rho^{*}b-p_{2}^{*}b\in p_{1}^{*}\operatorname{Br}G\}

The next statement extends [Cao18, Thm. 3.10].

Corollary 3.6.

Notation as in Theorem 3.1, we have a exact subsequence of (3.2)

\textstyle{\cdots\ignorespaces\ignorespaces\ignorespaces\ignorespaces}PicX\textstyle{\operatorname{Pic}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}PicG\textstyle{\operatorname{Pic}G\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Br𝒴\textstyle{\operatorname{Br}\mathcal{Y}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}BrGX\textstyle{\operatorname{Br}_{G}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}\scriptstyle{\square}Bre(G)\textstyle{\operatorname{Br}_{e}(G)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}p1\scriptstyle{p_{1}^{*}}\scriptstyle{\sim}BraG\textstyle{\operatorname{Br}_{a}G}\textstyle{\cdots\ignorespaces\ignorespaces\ignorespaces\ignorespaces}PicX\textstyle{\operatorname{Pic}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}PicG\textstyle{\operatorname{Pic}G\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Br𝒴\textstyle{\operatorname{Br}\mathcal{Y}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}f\scriptstyle{f^{*}}BrX\textstyle{\operatorname{Br}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}ρp2\scriptstyle{\rho^{*}-p_{2}^{*}}Br(G×kX),\textstyle{\operatorname{Br}(G\times_{k}X),}

where

Br1()=ker(Br()Br(×kk¯),\displaystyle\operatorname{Br}_{1}(-)=\ker(\operatorname{Br}(-)\rightarrow\operatorname{Br}(-\times_{k}\overline{k}),
Bra()=coker(BrkBr1()),\displaystyle\operatorname{Br}_{a}(-)=\operatorname{coker}(\operatorname{Br}k\rightarrow\operatorname{Br}_{1}(-)),
Bre(G)=ker(Br1GeGBrk),\displaystyle\operatorname{Br}_{e}(G)=\ker(\operatorname{Br}_{1}G\xrightarrow{e_{G}^{*}}\operatorname{Br}k),

and the right square is Cartesian.

Proof.

Since XX and GG are schemes, we have Bra(G×kX)BraGBra(X)\operatorname{Br}_{a}(G\times_{k}X)\cong\operatorname{Br}_{a}G\oplus\operatorname{Br}_{a}(X) and BraGBreG\operatorname{Br}_{a}G\cong\operatorname{Br}_{e}G. It follows that p1BreGBra(G×kX)p_{1}^{*}\operatorname{Br}_{e}G\rightarrow\operatorname{Br}_{a}(G\times_{k}X) is injective ( c.f. [Cao18, 3.2]. Then we use [Cao18, Prop. 3.7] to modify (3.2). ∎

3.2. Torsionness of the Brauer group

Lemma 3.7 (Restriction-Corestriction).

Let f:𝒳𝒴f:\mathcal{X}\rightarrow\mathcal{Y} be a finite étale morphism of degree nn between algebraic stacks and Ab(𝒴ét)\mathscr{F}\in{Ab}(\mathcal{Y}_{\text{\'{e}t}}). Then the composition

(3.8) Héti(𝒴,)fHéti(𝒳,f)fHéti(𝒴,)H_{\text{\'{e}t}}^{i}(\mathcal{Y},\mathscr{F})\xrightarrow{f^{*}}H_{\text{\'{e}t}}^{i}(\mathcal{X},f^{*}\mathscr{F})\xrightarrow{f_{*}}H_{\text{\'{e}t}}^{i}(\mathcal{Y},\mathscr{F})

is the map of multiplication by nn.

Proof.

The first map of (3.8) is induced by adjunction Rff\mathscr{F}\rightarrow Rf_{*}f^{*}\mathscr{F}, and the second one is induced by the trace map Rff=f!fRf_{*}f^{*}\mathscr{F}=f_{!}f^{*}\mathscr{F}\rightarrow\mathscr{F}. The composition of them is the multiplication by nn map. The result follows by taking cohomology. ∎

Lemma 3.9.

Assume kk is algebraically closed. Let XX be a smooth kk-variety and GG a linear kk-group acting on XX, 𝒴=[X/G]\mathcal{Y}=[X/G] the quotient stack. Then Br𝒴\operatorname{Br}\mathcal{Y} is torsion.

Proof.

By Lemma 3.7, we may assume GG is connected. Since XX is smooth, BrX\operatorname{Br}X is torsion (c.f. [Gro68, II, Prop. 1.4]). Also GG is linear over algebraically closed field, so PicG=0\operatorname{Pic}G=0. Note that GG is connected, the result follows from Theorem 3.1. ∎

Lemma 3.10.

Let p:𝒳kp:\mathcal{X}\rightarrow k be in hp/k{\mathbb{C}\text{hp}}/k. Denote 𝒳¯=𝒳×kk¯\overline{\mathcal{X}}=\mathcal{X}\times_{k}\overline{k}. Then RΓ(𝒳¯,K)RpKR\Gamma(\overline{\mathcal{X}},K)\cong Rp_{*}K for any KDcart(𝒳ét)K\in D_{\text{cart}}(\mathcal{X}_{\text{\'{e}t}}).

Proof.

By cohomological descent, we may assume 𝒳\mathcal{X} is a scheme. Then it is classical. ∎

Lemma 3.11.

Let p:𝒳kp:\mathcal{X}\rightarrow k be in hp/k{\mathbb{C}\text{hp}}/k. If Br𝒳¯\operatorname{Br}\overline{\mathcal{X}} is torsion, then so is Br𝒳\operatorname{Br}\mathcal{X}.

Proof.

By Lemma 3.10, the spectral sequence

E2p,q=Hp(k,Rqp𝐆m))Hp(k,Hétq(𝒳¯,𝐆m))Hétp+q(𝒳,𝐆m)E_{2}^{p,q}=H^{p}(k,R^{q}p_{*}\mathbf{G}_{m}))\xrightarrow{\sim}H^{p}(k,H_{\text{\'{e}t}}^{q}(\overline{\mathcal{X}},\mathbf{G}_{m}))\Rightarrow H_{\text{\'{e}t}}^{p+q}(\mathcal{X},\mathbf{G}_{m})

yields exact sequences

H2(k,Hét0(𝒳¯,𝐆m))Br1𝒳H1(k,Pic𝒳¯),\displaystyle H^{2}(k,H_{\text{\'{e}t}}^{0}(\overline{\mathcal{X}},\mathbf{G}_{m}))\rightarrow\operatorname{Br}_{1}\mathcal{X}\rightarrow H^{1}(k,\operatorname{Pic}\overline{\mathcal{X}}),
0Br1𝒳Br𝒳(Br𝒳¯)Γk,\displaystyle 0\rightarrow\operatorname{Br}_{1}\mathcal{X}\rightarrow\operatorname{Br}\mathcal{X}\rightarrow(\operatorname{Br}\overline{\mathcal{X}})^{\Gamma_{k}},

where 𝒳¯=𝒳×kk¯\overline{\mathcal{X}}=\mathcal{X}\times_{k}\overline{k} and Γk=Gal(k¯/k)\Gamma_{k}=\operatorname{Gal}(\overline{k}/k). Since all Galois cohomology groups are torsion, that Br𝒳¯\operatorname{Br}\overline{\mathcal{X}} is torsion implies Br𝒳\operatorname{Br}\mathcal{X} is torsion. ∎

Corollary 3.12.

Let 𝒳\mathcal{X} be a regular Noetherian Deligne-Mumford stack, or an algebraic stack over kk which can be covered by finitely many open substacks [Xi/Gi][X_{i}/G_{i}] where XiX_{i} is a smooth kk-variety and GiG_{i} a linear kk-group acting on XiX_{i}. Then Br𝒳\operatorname{Br}\mathcal{X} is torsion.

Proof.

The case that 𝒳\mathcal{X} is regular Noetherian Deligne-Mumford stack follows from [AM20, Prop 2.5 (iii)]. We now show the other case. By Mayer-Vietoris sequence for étale cohomology (c.f. [Sta22, Tag 0A50], which is also correct for 𝒳\mathcal{X} being an algebraic stack) it suffices to assume that 𝒳=[X/G]\mathcal{X}=[X/G]. Also by Lemma 3.11, we may assume kk is algebraically closed. Then the result follows from Lemma 3.9. ∎

4. Descent theory and the Brauer-Manin pairing

Throughout this section, let p:𝒳kp:\mathcal{X}\rightarrow k be an algebraic stack over a number field kk. The following results extend the ones of [HS13, Thm. 8.1], which is originally the descent theory of Colliot-Thélène and Sansuc in [CTS87].

4.1. Fundamental exact sequence for groups of multiplicative type

Consider

Rp:Dcart(𝒳ét)D(két)D(k)Rp_{*}:D_{\text{cart}}(\mathcal{X}_{\text{\'{e}t}})\rightarrow D(k_{\text{\'{e}t}})\cong D(k)

where D(k)D(k) is the derived category of complexes of kk-modules. Let SS be a kk-group of multiplicative type. Then its Cartier dual S^\hat{S} is a finitely generated kk-module. Let KD(𝒳)\operatorname{KD}^{\prime}(\mathcal{X}) be the cone of 𝐆m[1]Rp𝐆m[1]\mathbf{G}_{m}[1]\rightarrow Rp_{*}\mathbf{G}_{m}[1] in D(k)D(k).

Theorem 4.1.

Then we have the fundamental exact sequence

(4.2) 0H1(k,S)pHfppf1(𝒳,S)𝜒HomD(k)(S^,KD(𝒳))H2(k,S)pHfppf2(𝒳,S),0\rightarrow H^{1}(k,S)\xrightarrow{p^{*}}H_{\text{fppf}}^{1}(\mathcal{X},S)\xrightarrow{\chi}\operatorname{Hom}_{D(k)}(\hat{S},\operatorname{KD}^{\prime}(\mathcal{X}))\xrightarrow{\partial}H^{2}(k,S)\xrightarrow{p^{*}}H_{\text{fppf}}^{2}(\mathcal{X},S),

where the map χ\chi is the extended type.

Proof.

The proof is the same as that of [HS13, Thm. 8.1], except that we must show that Hétp(𝒳,𝐆m)=Hfppfp(𝒳,𝐆m)H^{p}_{\text{\'{e}t}}(\mathcal{X},\mathbf{G}_{m})=H^{p}_{\text{fppf}}(\mathcal{X},\mathbf{G}_{m}) is still valid for 𝒳\mathcal{X} being an algebraic stack. But this is standard. By cohomological descent (for example, using spectral sequence (3.3) in both étale and fppf topology for an altas Y𝒳Y\rightarrow\mathcal{X} of 𝒳\mathcal{X}), we reduce this to the case where 𝒳\mathcal{X} is a scheme, which is clear since 𝐆m\mathbf{G}_{m} is smooth as group scheme. See [Mil80, III, Thm. 3.9]. ∎

Remark 4.3.

By (2.8) and (2.9) Hfppf1(𝒳,S)H_{\text{fppf}}^{1}(\mathcal{X},S) classifies SS-torsor over 𝒳\mathcal{X}, and the map χ\chi associate a torsor class [Y][Y] to its extended type. The first three termes of the fundamental exact sequence shows that two tosors have the same extended type if and only if they are isomorphic up to a twist.

4.2. Descent

Let aH1(k,S^)a\in H^{1}(k,\hat{S}). As shown in the above proof, for 𝒳\mathcal{X} being an algebraic stack over kk, the diagram in [Lv22, Rkm. 2.7 (2)] is still correct. We shall use a part of this diagram

(4.4) H1(𝒳,S)\textstyle{H^{1}(\mathcal{X},S)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}χ\scriptstyle{\chi}p(a)\scriptstyle{p^{*}(a)\cup-}HomD(k)(S^,KD(𝒳))\textstyle{\operatorname{Hom}_{D(k)}(\hat{S},\operatorname{KD}^{\prime}(\mathcal{X}))\ignorespaces\ignorespaces\ignorespaces\ignorespaces}a\scriptstyle{a\cup-}Br1𝒳\textstyle{\operatorname{Br}_{1}\mathcal{X}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}r\scriptstyle{r}H1(k,KD(𝒳)).\textstyle{H^{1}(k,\operatorname{KD}^{\prime}(\mathcal{X})).}

Let f:𝒴𝒳f:\mathcal{Y}\rightarrow\mathcal{X} be an SS-torsor. Define λ=χ([f])\lambda=\chi([f]), and

Brλ𝒳=r1(λ(H1(k,S^)))Br1𝒳.\operatorname{Br}_{\lambda}\mathcal{X}=r^{-1}(\lambda_{*}(H^{1}(k,\hat{S})))\subseteq\operatorname{Br}_{1}\mathcal{X}.
Proposition 4.5.

We have 𝒳(𝐀k)f=𝒳(𝐀k)Brλ\mathcal{X}(\mathbf{A}_{k})^{f}=\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}_{\lambda}}.

Proof.

Use (4.4) for the stack 𝒳\mathcal{X}. Then the original proof of [HS13, 8.12] is still valid. ∎

4.3. The Brauer-Manin pairing

Suppose that 𝒳\mathcal{X} is of finite type over kk. Then there exists a finite set of places SS, such that 𝒳\mathcal{X} extends to an integral model 𝒳\mathscr{X} of 𝒳\mathcal{X} over 𝒪k,S\mathscr{O}_{k,S} (that is, a 𝒪k,S\mathscr{O}_{k,S}-stack 𝒳\mathscr{X} such that 𝒳×𝒪k,Sk𝒳\mathscr{X}\times_{\mathscr{O}_{k,S}}k\xrightarrow{\approx}\mathcal{X}, and this can be shown by looking at a smooth groupoid presentation of 𝒳\mathcal{X} ). Moreover, by [Chr20, Sec. 13], we have

𝒳(𝐀k)colimfinite SS(vS𝒳(kv)×vS𝒳(𝒪v)).\mathcal{X}(\mathbf{A}_{k})\cong\operatorname{colim}_{\text{finite }S^{\prime}\supseteq S}(\prod_{v\in S^{\prime}}\mathscr{X}(k_{v})\times\prod_{v\not\in S^{\prime}}\mathscr{X}(\mathscr{O}_{v})).

In particular, any x𝒳(𝐀k)x\in\mathcal{X}(\mathbf{A}_{k}) can be viewed as (xv)vvS𝒳(kv)×vS𝒳(𝒪v)(x_{v})_{v}\in\prod_{v\in S^{\prime}}\mathscr{X}(k_{v})\times\prod_{v\not\in S^{\prime}}\mathscr{X}(\mathscr{O}_{v}) for some finite SSS^{\prime}\supseteq S. Then one can imitate the proof of [Poo17, 8.2.1], to show that the Brauer-Manin pairing for 𝒳\mathcal{X}

,BM:𝒳(𝐀k)×Br𝒳\displaystyle\langle-,-\rangle_{\text{BM}}:\mathcal{X}(\mathbf{A}_{k})\times\operatorname{Br}\mathcal{X} /,\displaystyle\rightarrow\mathbb{Q}/\mathbb{Z},
((xv)v,A)\displaystyle((x_{v})_{v},A) vΩkinvvA(xv),\displaystyle\mapsto\sum_{v\in\Omega_{k}}\operatorname{inv}_{v}A(x_{v}),

is well-defined, and the Brauer-Manin set 𝒳(𝐀k)Br\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}} coincides with the classical definition using ,BM\langle-,-\rangle_{\text{BM}} since Br𝐀kvBrkv\operatorname{Br}\mathbf{A}_{k}\rightarrow\bigoplus_{v}\operatorname{Br}k_{v} is an isomorphism.

Define

B𝒳=ker(Bra𝒳vΩkBra𝒳v)\cyrille B\mathcal{X}=\ker(\operatorname{Br}_{a}\mathcal{X}\rightarrow\prod_{v\in\Omega_{k}}\operatorname{Br}_{a}\mathcal{X}_{v})

where 𝒳v=𝒳×kkv\mathcal{X}_{v}=\mathcal{X}\times_{k}k_{v}. For AB𝒳A\in\cyrille B\mathcal{X} and (xv)𝒳(𝐀k)(x_{v})\in\mathcal{X}(\mathbf{A}_{k}), one checks that

(xv),ABM=vΩkinvvA(xv)\langle(x_{v}),A\rangle_{\text{BM}}=\sum_{v\in\Omega_{k}}\operatorname{inv}_{v}A(x_{v})

does not depend on the choice of (xv)(x_{v}). Assuming 𝒳(𝐀k)\mathcal{X}(\mathbf{A}_{k})\neq\emptyset, we obtain a well-defined map i=(xv),BM:B𝒳/i=\langle(x_{v}),-\rangle_{\text{BM}}:\cyrille B\mathcal{X}\rightarrow\mathbb{Q}/\mathbb{Z}. This notation will be used in Subsection 7.2.

5. Comparison to some other cohomological obstructions

Let kk be a number field. Harari [Har02] compared the Brauer-Manin obstruction of varieties with some other cohomological obstructions such as obstructions given by torsors under connected groups or abelian gerbes. In this section, we extend them to a class of algebraic stacks.

5.1. Descent and second descent obstruction on algebraic stacks

Since for a (resp. Commutative) kk group GG, Hˇfppf1(,G)\check{H}_{\text{fppf}}^{1}(-,G) (resp. Hét2(,G)H_{\text{\'{e}t}}^{2}(-,G)) is stable, the following definitions also make sense.

Definition 5.1.

The descent obstruction is

𝒳(𝐀k)desc=linear k-group G𝒳(𝐀k)Hˇfppf1(,G).\mathcal{X}(\mathbf{A}_{k})^{\text{desc}}=\bigcap_{\text{linear $k$-group $G$}}\mathcal{X}(\mathbf{A}_{k})^{\check{H}_{\text{fppf}}^{1}(-,G)}.

We also define

𝒳(𝐀k)conn=connected linear k-group G𝒳(𝐀k)Hˇfppf1(,G),\mathcal{X}(\mathbf{A}_{k})^{\text{conn}}=\bigcap_{\text{connected linear $k$-group $G$}}\mathcal{X}(\mathbf{A}_{k})^{\check{H}_{\text{fppf}}^{1}(-,G)},

and the second descent obstruction (c.f. [Lv21, 4.2], slightly modified here) is

𝒳(𝐀k)2-desc=commutative linear k-group G𝒳(𝐀k)Hét2(,G).\mathcal{X}(\mathbf{A}_{k})^{2\text{-}\text{desc}}=\bigcap_{\text{commutative linear $k$-group $G$}}\mathcal{X}(\mathbf{A}_{k})^{H_{\text{\'{e}t}}^{2}(-,G)}.

5.2. Comparison of obstructions

Theorem 5.2.

Let XX be a smooth algebraic kk-stack of finite type that is either Deligne-Mumford or Zariski-locally quotient of kk-variety by a linear kk-group. Then

𝒳(𝐀k)Br=𝒳(𝐀k)2-desc.\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}=\mathcal{X}(\mathbf{A}_{k})^{{2\text{-}\text{desc}}}.
Proof.

Since 𝔾m\mathbb{G}_{m} is a commutative group, we have 𝒳(𝐀k)2-desc𝒳(𝐀k)Br.\mathcal{X}(\mathbf{A}_{k})^{{2\text{-}\text{desc}}}\subseteq\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}. We need to prove the other side 𝒳(𝐀k)Br𝒳(𝐀k)2-desc.\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}\subseteq\mathcal{X}(\mathbf{A}_{k})^{{2\text{-}\text{desc}}}.

Since every commutative group is the product of groups of multiple type and some 𝔾a\mathbb{G}_{a} factors, and H2(k,𝔾a)=H2(kv,𝔾a)=0H^{2}(k,\mathbb{G}_{a})=H^{2}(k_{v},\mathbb{G}_{a})=0 for all places vΩk,v\in\Omega_{k}, we have

𝒳(𝐀k)2-desc=commutative linear k-group G𝒳(𝐀k)Hét2(,G)=k-group G of multiple type𝒳(𝐀k)Hét2(,G).\mathcal{X}(\mathbf{A}_{k})^{2\text{-}\text{desc}}=\bigcap_{\text{commutative linear $k$-group $G$}}\mathcal{X}(\mathbf{A}_{k})^{H_{\text{\'{e}t}}^{2}(-,G)}=\bigcap_{\text{$k$-group $G$ of multiple type}}\mathcal{X}(\mathbf{A}_{k})^{H_{\text{\'{e}t}}^{2}(-,G)}.

For a divisible kk-group SS of multiple type, By Corollary 3.12, the group H2(𝒳,S)H^{2}(\mathcal{X},S) is torsion. Furthermore, for any fixed fH2(𝒳,S),f\in H^{2}(\mathcal{X},S), there exists some positive integer nn such that ff lies in the image of H2(𝒳,nS)H2(𝒳,S).H^{2}(\mathcal{X},_{n}S)\to H^{2}(\mathcal{X},S). Since every group of multiple type is the product of a finite group and a divisible group, we have

𝒳(𝐀k)2-desc=k-group G of multiple type𝒳(𝐀k)Hét2(,G)=finite commutative k-group G𝒳(𝐀k)Hét2(,G).\mathcal{X}(\mathbf{A}_{k})^{2\text{-}\text{desc}}=\bigcap_{\text{$k$-group $G$ of multiple type}}\mathcal{X}(\mathbf{A}_{k})^{H_{\text{\'{e}t}}^{2}(-,G)}=\bigcap_{\text{finite commutative $k$-group $G$}}\mathcal{X}(\mathbf{A}_{k})^{H_{\text{\'{e}t}}^{2}(-,G)}.

We assume that GG is a finite abelian group over kk. Let G^\hat{G} be the dual of GG, and let F^=H0(k,G^).\hat{F}=H^{0}(k,\hat{G}). Then F^i=1r/ni\hat{F}\cong\bigoplus\limits_{i=1}^{r}\mathbb{Z}/n_{i}\mathbb{Z} with integers r>0r>0 and nin_{i}\in\mathbb{N}^{*}. Then Fi=1rμni.F\cong\bigoplus\limits_{i=1}^{r}\mu_{n_{i}}. We have the following commutative diagram

H2(k,G)\textstyle{H^{2}(k,G)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H2(k,F)\textstyle{H^{2}(k,F)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H2(k,𝔾mr)\textstyle{H^{2}(k,\mathbb{G}_{m}^{r})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}vΩkH2(kv,G)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},G)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}jvG\scriptstyle{\sum j_{v}^{G}}vΩkH2(kv,F)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},F)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}jvF\scriptstyle{\sum j_{v}^{F}}vΩkH2(kv,𝔾mr)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},\mathbb{G}_{m}^{r})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(jv)r\scriptstyle{(\sum j_{v})^{r}}H0(k,G^)=F^\textstyle{H^{0}(k,\hat{G})^{*}=\hat{F}^{*}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H0(k,F^)=F^\textstyle{H^{0}(k,\hat{F})^{*}=\hat{F}^{*}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H0(k,r)=(/)r\textstyle{H^{0}(k,\mathbb{Z}^{r})^{*}=(\mathbb{Q}/\mathbb{Z})^{r}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}0\textstyle{0}0\textstyle{0}0.\textstyle{0.}

By the Poitou-Tate exact sequence ([Ser94] or [Mil06, Chapt. I 4.10]), every column of this diagram is exact. By taking a point (pv)𝒳(𝐀k)Br(p_{v})\in\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}, we have the following commutative diagram

H2(𝒳,G)\textstyle{H^{2}(\mathcal{X},G)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(Pv)\scriptstyle{(P_{v})}H2(𝒳,F)\textstyle{H^{2}(\mathcal{X},F)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(Pv)\scriptstyle{(P_{v})}H2(𝒳,𝔾mr)\textstyle{H^{2}(\mathcal{X},\mathbb{G}_{m}^{r})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(Pv)\scriptstyle{(P_{v})}vΩkH2(kv,G)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},G)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}jvG\scriptstyle{\sum j_{v}^{G}}vΩkH2(kv,F)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},F)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}jvF\scriptstyle{\sum j_{v}^{F}}vΩkH2(kv,𝔾mr)\textstyle{\bigoplus\limits_{v\in\Omega_{k}}H^{2}(k_{v},\mathbb{G}_{m}^{r})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(jv)r\scriptstyle{(\sum j_{v})^{r}}H0(k,G^)=F^\textstyle{H^{0}(k,\hat{G})^{*}=\hat{F}^{*}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H0(k,F^)=F^\textstyle{H^{0}(k,\hat{F})^{*}=\hat{F}^{*}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}H0(k,r)=(/)r.\textstyle{H^{0}(k,\mathbb{Z}^{r})^{*}=(\mathbb{Q}/\mathbb{Z})^{r}.}

Combining these two diagrams, we have 𝒳(𝐀k)Br𝒳(𝐀k)H2(𝒳,G).\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}\subseteq\mathcal{X}(\mathbf{A}_{k})^{H^{2}(\mathcal{X},G)}. Hence 𝒳(𝐀k)Br=𝒳(𝐀k)2-desc.\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}=\mathcal{X}(\mathbf{A}_{k})^{{2\text{-}\text{desc}}}.

Remark 5.3.

The last part of this proof is essentially from [Har02, Prop. 1]. For the convenience of reading, we give the complete details.

Theorem 5.4.

Let XX be an algebraic kk-stack. Then

𝒳(𝐀k)Br𝒳(𝐀k)conn.\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}\subseteq\mathcal{X}(\mathbf{A}_{k})^{\text{conn}}.
Proof.

The proofs of [Har02, Cor. 1 and Sec. 3] remain valid here and the result follows. ∎

6. Descent for Brauer-Manin set along a torsor

In this section we extend the results on descent for Brauer-Manin set along a torsor of Cao [Cao18, Sec. 5] to allow the base space XX to be a quotient stack.

Lemma 6.1.

Let 𝒳=[Y/G]\mathcal{X}=[Y/G] where YY is a kk-variety and GG a connected linear kk-group acting on YY. Let f:Y𝒳f:Y\rightarrow\mathcal{X} be the canonic map making YY a GG-torosr over 𝒳\mathcal{X}, and f:Br𝒳BrYf^{*}:\operatorname{Br}\mathcal{X}\rightarrow\operatorname{Br}Y the pullback. Then we have

𝒳(𝐀k)kerf=σH1(k,G)fσ(Yσ(𝐀k)).\mathcal{X}(\mathbf{A}_{k})^{\ker f^{*}}=\bigcup_{\sigma\in H^{1}(k,G)}f_{\sigma}(Y_{\sigma}(\mathbf{A}_{k})).
Proof.

Choosing an embedding ι:GSLn\iota:G\hookrightarrow\operatorname{SL}_{n} we obtain a canonic surjection π:SLnW:=SLn/G\pi:\operatorname{SL}_{n}\rightarrow W:=\operatorname{SL}_{n}/G. The pushforward of YY under the map ι:Hfppf1(𝒳,G)Hfppf1(𝒳,SLn)\iota_{*}:H_{\text{fppf}}^{1}(\mathcal{X},G)\rightarrow H_{\text{fppf}}^{1}(\mathcal{X},\operatorname{SL}_{n}) is an SLn\operatorname{SL}_{n}-torsor g:Z=Y×GSLnSLn𝒳g:Z=Y\times^{G}\operatorname{SL}_{n}\xrightarrow{\operatorname{SL}_{n}}\mathcal{X} with a canonic map

p:ZY×GW=X×kWW,p:Z\rightarrow Y\times^{G}W=X\times_{k}W\rightarrow W,

such that for all wW(k)w\in W(k), p1(w)Y×Gπ1(w)Ywp^{-1}(w)\cong Y\times^{G}\pi^{-1}(w)\cong Y_{\partial w} is a twist of YY, where :W(k)H1(k,G)\partial:W(k)\rightarrow H^{1}(k,G) is the connecting map associated to 1G𝜄SLn𝜋W11\rightarrow G\xrightarrow{\iota}\operatorname{SL}_{n}\xrightarrow{\pi}W\rightarrow 1.

Write A=kerfA=\ker f^{*}. It is obvious that

(6.2) 𝒳(𝐀k)A=g(Z(𝐀k)gA.\mathcal{X}(\mathbf{A}_{k})^{A}=g(Z(\mathbf{A}_{k})^{g^{*}A}.

Thus to complete the proof, we only need to show that each x𝒳(𝐀k)Ax\in\mathcal{X}(\mathbf{A}_{k})^{A} lifts to Yσ(𝐀k)Y_{\sigma}(\mathbf{A}_{k}) for some σH1(k,G)\sigma\in H^{1}(k,G). For this purpose, we use strong approximation of WW with respect to Brauer-Manin obstruction, which states that

π(SLn(𝐀k))W(k)=W(𝐀k)Br.\pi(\operatorname{SL}_{n}(\mathbf{A}_{k}))W(k)=W(\mathbf{A}_{k})^{\operatorname{Br}}.

Claim that

p(Z(𝐀k)gAW(𝐀k)Br.p(Z(\mathbf{A}_{k})^{g^{*}A}\subseteq W(\mathbf{A}_{k})^{\operatorname{Br}}.

Then for any x𝒳(𝐀k)Ax\in\mathcal{X}(\mathbf{A}_{k})^{A}, by (6.2) there is zZ(𝐀k)gAz\in Z(\mathbf{A}_{k})^{g^{*}A} such that x=g(z)x=g(z). Then by the claim we have

p(z)W(𝐀k)Br=π(SLn(𝐀k))W(k).p(z)\in W(\mathbf{A}_{k})^{\operatorname{Br}}=\pi(\operatorname{SL}_{n}(\mathbf{A}_{k}))W(k).

It follows that p(z)=π(s)wp(z)=\pi(s)w with some sSLn(𝐀k)s\in\operatorname{SL}_{n}(\mathbf{A}_{k}) and wW(k)w\in W(k). Letting z=s1zz^{\prime}=s^{-1}z we know that x=g(z)=g(z)x=g(z)=g(z^{\prime}) and p(z)=wp(z^{\prime})=w. Thus there exists yp1(w)(𝐀k)Yw(𝐀k)y\in p^{-1}(w)(\mathbf{A}_{k})\cong Y_{\partial w}(\mathbf{A}_{k}) lifting zz^{\prime} and f(y)=xf(y)=x, which is the demanded result.

To justify the claim, we only need two facts:

  1. (i)

    BrZBr𝒳\operatorname{Br}Z\cong\operatorname{Br}\mathcal{X}, which follows from Theorem 3.1 and Br(SLn×kZ)BrZ\operatorname{Br}(\operatorname{SL}_{n}\times_{k}Z)\cong\operatorname{Br}Z, and

  2. (ii)

    BrWBrZBrY\operatorname{Br}W\rightarrow\operatorname{Br}Z\rightarrow\operatorname{Br}Y factor through Brk\operatorname{Br}k, which follows from YZ×W,1WkY\cong Z\times_{W,1_{W}}k, i.e., p1(1W)Yp^{-1}(1_{W})\cong Y.

The proof is complete. ∎

Lemma 6.3.

Let f:Y𝒳f:Y\rightarrow\mathcal{X} be as in Lemma 6.1. Assume that X1G=0\cyrille X^{1}G=0. Suppose ABr𝒳A\subseteq\operatorname{Br}\mathcal{X} is a subgroup containing kerf\ker f^{*} and for each σH1(k,G)\sigma\in H^{1}(k,G), we are given BσBrGσYσB_{\sigma}\in\operatorname{Br}_{G_{\sigma}}Y_{\sigma} such that fσ1Bσ=A{f_{\sigma}^{*}}^{-1}B_{\sigma}=A. Then

𝒳(𝐀k)A=σH1(k,G)fσ(Yσ(𝐀k)Bσ).\mathcal{X}(\mathbf{A}_{k})^{A}=\bigcup_{\sigma\in H^{1}(k,G)}f_{\sigma}(Y_{\sigma}(\mathbf{A}_{k})^{B_{\sigma}}).
Proof.

Lemma 6.1 together with AkerfA\supseteq\ker f^{*} implies that

𝒳(𝐀k)A=σH1(k,G)fσ(Yσ(𝐀k)fσA).\mathcal{X}(\mathbf{A}_{k})^{A}=\bigcup_{\sigma\in H^{1}(k,G)}f_{\sigma}(Y_{\sigma}(\mathbf{A}_{k})^{f_{\sigma}^{*}A}).

It suffices to show that Yσ(𝐀k)fσAYσ(𝐀k)BσY_{\sigma}(\mathbf{A}_{k})^{f_{\sigma}^{*}A}\subseteq Y_{\sigma}(\mathbf{A}_{k})^{B_{\sigma}} and one may assume that σ\sigma is the neutral element. Using Corollary 3.6 and [Dem11, Thm. 5.1], the remaining argument is the same as [Cao18, Lem. 5.1]. ∎

Theorem 6.4.

Let f:Y𝒳f:Y\rightarrow\mathcal{X} be as in Lemma 6.1. Then

𝒳(𝐀k)Br=σH1(k,G)fσ(Yσ(𝐀k)BrGσ(Yσ)).\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}=\bigcup_{\sigma\in H^{1}(k,G)}f_{\sigma}(Y_{\sigma}(\mathbf{A}_{k})^{\operatorname{Br}_{G_{\sigma}}(Y_{\sigma})}).
Proof.

By [Cao18, Prop. 5.6], one can find a resolution of GG

1G𝜓HϕT01\rightarrow G\xrightarrow{\psi}H\xrightarrow{\phi}T\rightarrow 0

such that X1H=0\cyrille X^{1}H=0 and H3(k,T^)=0H^{3}(k,\hat{T})=0. The pushforward of YY under the map ψ:Hfppf1(𝒳,G)Hfppf1(𝒳,H)\psi_{*}:H_{\text{fppf}}^{1}(\mathcal{X},G)\rightarrow H_{\text{fppf}}^{1}(\mathcal{X},H) is an HH-torsor g:Z=Y×GH𝐻𝒳g:Z=Y\times^{G}H\xrightarrow{H}\mathcal{X} with a canonic map

p:ZY×GT=X×kTT.p:Z\rightarrow Y\times^{G}T=X\times_{k}T\rightarrow T.

Since X1H=0\cyrille X^{1}H=0, applying Lemma 6.3 to g:Z𝐻𝒳g:Z\xrightarrow{H}\mathcal{X} we obtain

(6.5) 𝒳(𝐀k)Br=σH1(k,H)gσ(Zσ(𝐀k)BrHσ(Zσ)).\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}=\bigcup_{\sigma\in H^{1}(k,H)}g_{\sigma}(Z_{\sigma}(\mathbf{A}_{k})^{\operatorname{Br}_{H_{\sigma}}(Z_{\sigma})}).

Also H3(k,T^)=0H^{3}(k,\hat{T})=0 implies that BraHBraG\operatorname{Br}_{a}H\rightarrow\operatorname{Br}_{a}G is surjective. It follows from [Cao18, Prop. 3.13] that BrHZBrGY\operatorname{Br}_{H}Z\rightarrow\operatorname{Br}_{G}Y is also surjective. The remaining argument is similar as in Lemma 6.1. For each x𝒳(𝐀k)Ax\in\mathcal{X}(\mathbf{A}_{k})^{A}, by (6.5), there is some σH1(k,H)\sigma\in H^{1}(k,H) and zZσ(𝐀k)BrHσ(Zσ)z\in Z_{\sigma}(\mathbf{A}_{k})^{\operatorname{Br}_{H_{\sigma}}(Z_{\sigma})} such that x=gσ(z)x=g_{\sigma}(z). We may assume σ\sigma is the neutral element in the sequel. By definition we have p(Z(𝐀k)BrH(Z))T(𝐀k)Br1p(Z(\mathbf{A}_{k})^{\operatorname{Br}_{H}(Z)})\subseteq T(\mathbf{A}_{k})^{\operatorname{Br}_{1}}. Also [CDX19, Prop. 3.3] gives

T(𝐀k)Br1=ψ(H(𝐀k)Br1)T(k).T(\mathbf{A}_{k})^{\operatorname{Br}_{1}}=\psi(H(\mathbf{A}_{k})^{\operatorname{Br}_{1}})T(k).

It follows that p(z)=ψ(h)tp(z)=\psi(h)t with some hH(𝐀k)Br1h\in H(\mathbf{A}_{k})^{\operatorname{Br}_{1}} and tT(k)t\in T(k). Letting z=s1zZ(𝐀k)BrH(Z)z^{\prime}=s^{-1}z\in Z(\mathbf{A}_{k})^{\operatorname{Br}_{H}(Z)}, we know that x=g(z)=g(z)x=g(z)=g(z^{\prime}) and p(z)=tp(z^{\prime})=t. Thus putting τ=t\tau=\partial t, there exists yp1(t)(𝐀k)Yτ(𝐀k)y\in p^{-1}(t)(\mathbf{A}_{k})\cong Y_{\tau}(\mathbf{A}_{k}) lifting zz^{\prime} and f(y)=xf(y)=x. On the other hand, since zZ(𝐀k)BrH(Z)z^{\prime}\in Z(\mathbf{A}_{k})^{\operatorname{Br}_{H}(Z)} and BrHZBrGY\operatorname{Br}_{H}Z\rightarrow\operatorname{Br}_{G}Y is surjective, we also have yYτ(𝐀k)BrGτYτy\in Y_{\tau}(\mathbf{A}_{k})^{\operatorname{Br}_{G_{\tau}}Y_{\tau}}. This finishes the proof. ∎

7. Brauer-Manin set under a product

Let kk be a number field. The product preservation property of Brauer-Manin set was first established by Skorobogatov and Zarhin [SZ14] for smooth geometrically integral projective kk-varieties, and later by the first author [Lv20] for open ones. In this section we extend this result to a class of algebraic stacks over kk.

Theorem 7.1.

The functor

(𝐀k)Br:hp1/kSet-(\mathbf{A}_{k})^{\operatorname{Br}}:{\mathbb{C}\text{hp}}_{1}/k\rightarrow{Set}

preserves finite product, where hp1/khp/k{\mathbb{C}\text{hp}}_{1}/k\subset{\mathbb{C}\text{hp}}/k is the full sub-22-category spanned by smooth algebraic kk-stacks of finite type admitting separated and geometrically integral atlases XX such that X(𝐀k)BX(\mathbf{A}_{k})^{\cyrille B}\neq\emptyset, and is either Deligne-Mumford or Zariski-locally quotients of kk-varieties by linear kk-groups.

7.1. A variant of Künneth formula

Let Λ\Lambda be a ring killed by some integer. For 𝒳hp/k\mathcal{X}\in{\mathbb{C}\text{hp}}/k, let Dctfb(𝒳ét,Λ)D_{\text{ctf}}^{b}(\mathcal{X}_{\text{\'{e}t}},\Lambda) be the full subcategory of Dcart(𝒳ét,Λ)D_{\text{cart}}(\mathcal{X}_{\text{\'{e}t}},\Lambda) such that fKDctfb(Xét,Λ)f^{*}K\in D_{\text{ctf}}^{b}(X_{\text{\'{e}t}},\Lambda) for every KDcart(𝒳ét,Λ)K\in D_{\text{cart}}(\mathcal{X}_{\text{\'{e}t}},\Lambda) and every atlas X𝒳X\rightarrow\mathcal{X}.

Lemma 7.2 (Künneth formula).

Let p:𝒳kp:\mathcal{X}\rightarrow k and q:𝒴kq:\mathcal{Y}\rightarrow k be smooth algebraic stacks of finite type over a field kk, KDctfb(𝒳ét,Λ)K\in D_{\text{ctf}}^{b}(\mathcal{X}_{\text{\'{e}t}},\Lambda) and LDctfb(𝒴ét,Λ)L\in D_{\text{ctf}}^{b}(\mathcal{Y}_{\text{\'{e}t}},\Lambda). Then we have

RpKΛLRqLR(p×q)(KΛLL).Rp_{*}K\boxtimes_{\Lambda}^{L}Rq_{*}L\cong R(p\times q)_{*}(K\boxtimes_{\Lambda}^{L}L).
Proof.

The proof is standard. By cohomological descent we may assume that 𝒳\mathcal{X} and 𝒴\mathcal{Y} are in Esp/k{Esp}/k and then in Sch/k{Sch}/k, in which case it is classical (see for example, [Fu11, Cor. 9.3.5]). ∎

Lemma 7.3 (Smooth base change).

Let

𝒲\textstyle{\mathcal{W}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}g\scriptstyle{g}q\scriptstyle{q}𝒵\textstyle{\mathcal{Z}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}p\scriptstyle{p}𝒴\textstyle{\mathcal{Y}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}f\scriptstyle{f}𝒳\textstyle{\mathcal{X}}

be a 22-Cartesian diagram in hp/S{\mathbb{C}\text{hp}}/S where pp is smooth. Then

pRfRgq:Dcart(𝒴ét,Λ)Dcart(𝒵ét,Λ)p^{*}Rf_{*}\rightarrow Rg_{*}q^{*}:D_{\text{cart}}(\mathcal{Y}_{\text{\'{e}t}},\Lambda)\rightarrow D_{\text{cart}}(\mathcal{Z}_{\text{\'{e}t}},\Lambda)

is an isomorphism.

Proof.

In view of Remark 2.5, this is [LZ17, Cor. 0.1.5]. ∎

7.2. The universal torsor of nn-torsion

Let 𝒳hp/k\mathcal{X}\in{\mathbb{C}\text{hp}}/k. By cohomological descent, the Gal(k¯/k)\operatorname{Gal}(\overline{k}/k)-module Hét1(𝒳¯,μn)H_{\text{\'{e}t}}^{1}(\overline{\mathcal{X}},\mu_{n}) is finite as abelian group. Let S𝒳S_{\mathcal{X}} be the group of multiplicative type dual to it, that is, S^𝒳=Hét1(𝒳¯,μn)\hat{S}_{\mathcal{X}}=H_{\text{\'{e}t}}^{1}(\overline{\mathcal{X}},\mu_{n}). Then as in [Cao20, p. 2156], we have a similar commutative diagram of distinguished triangles for 𝒳\mathcal{X}. Also we have H0(Δn)S^𝒳H^{0}(\Delta_{n})\cong\hat{S}_{\mathcal{X}} and τ0Δ=KD(𝒳)\tau_{\leq 0}\Delta=\operatorname{KD}^{\prime}(\mathcal{X}). Then one also defines the universal torsor of nn-torsion of 𝒳\mathcal{X} to be an inverse image of τ0(ΔnΔ):S^𝒳τ0Δ\tau_{\leq 0}(\Delta_{n}\rightarrow\Delta):\hat{S}_{\mathcal{X}}\rightarrow\tau_{\leq 0}\Delta under χ\chi (see (4.2) with S=S𝒳S=S_{\mathcal{X}}).

Lemma 7.4.

Suppose that 𝒳/k\mathcal{X}/k admitting an atlas XX which is a smooth geometrically integral kk-variety and satisfies X(𝐀k)BX(\mathbf{A}_{k})^{\cyrille B}\neq\emptyset. Then

χ:Hfppf1(𝒳,S)HomD(k)(S^,KD(𝒳))\chi:H_{\text{fppf}}^{1}(\mathcal{X},S)\rightarrow\operatorname{Hom}_{D(k)}(\hat{S},\operatorname{KD}^{\prime}(\mathcal{X}))

in (4.2) is surjective.

Proof.

We assume first p:𝒳kp:\mathcal{X}\rightarrow k is in Esp/k{Esp}/k. Let X:Δ+opSch/kX_{\bullet}:\Delta_{+}^{\text{op}}\rightarrow{Sch}/k be a Čech nerve of X𝒳X\rightarrow\mathcal{X} and denote by p:Xkp_{\bullet}:X_{\bullet}\rightarrow k the composed map. Since X(𝐀k)BX(\mathbf{A}_{k})^{\cyrille B}\neq\emptyset, by functoriality, we obtain a simplicial set

X(𝐀k)B:ΔopSetX_{\bullet}(\mathbf{A}_{k})^{\cyrille B}:\Delta^{\text{op}}\rightarrow{Set}

which is nonempty. It follows from [HS13, Cor. 8.17] that pn:H2(k,S)Hfppf2(Xn,S)p_{n}^{*}:H^{2}(k,S)\rightarrow H_{\text{fppf}}^{2}(X_{n},S) is injective for all n0n\geq 0. Let q1,q2:X1=X×𝒳XXq_{1},q_{2}:X_{1}=X\times_{\mathcal{X}}X\rightarrow X be the two projections. Since p0q1=p0q2p_{0}q_{1}=p_{0}q_{2}, the injection p1:H2(k,S)Hfppf2(X1,S)p_{1}^{*}:H^{2}(k,S)\rightarrow H_{\text{fppf}}^{2}(X_{1},S) factors through q1q2:Hfppf2(X,S)Hfppf2(X1,S)q_{1}^{*}-q_{2}^{*}:H_{\text{fppf}}^{2}(X,S)\rightarrow H_{\text{fppf}}^{2}(X_{1},S) so that we obtain a injection

H2(k,S)ker(q1q2)=Hˇ0(fppf2(S)).H^{2}(k,S)\rightarrow\ker(q_{1}^{*}-q_{2}^{*})=\check{H}^{0}(\mathscr{H}_{\text{fppf}}^{2}(S)).

But this is the composition

H2(k,S)pHfppf2(𝒳,S)Hˇ0(fppf2(S))H^{2}(k,S)\xrightarrow{p^{*}}H_{\text{fppf}}^{2}(\mathcal{X},S)\rightarrow\check{H}^{0}(\mathscr{H}_{\text{fppf}}^{2}(S))

where the second map is the edge map E2E20,2E^{2}\rightarrow E_{2}^{0,2} of the spectral sequence of cohomological descent (3.3) with =S\mathscr{F}=S. It follows that pp^{*} is also injective, i.e., χ\chi is surjective. Thus we may assume 𝒳\mathcal{X} is an smooth geometrically integral kk variety.

Now let 𝒳\mathcal{X} be in hp/k{\mathbb{C}\text{hp}}/k, repeating the first reduction step we may then assume 𝒳\mathcal{X} is an smooth geometrically integral kk variety. The result follows from the original result for varieties [HS13, Cor. 8.17]. ∎

By Lemma 7.4, if 𝒳\mathcal{X} (resp. 𝒴\mathcal{Y}) satisfies the assumptions, then there exists a universal torsor of nn-torsion 𝒯𝒳S𝒳𝒳\mathcal{T}_{\mathcal{X}}\xrightarrow{S_{\mathcal{X}}}\mathcal{X} of 𝒳\mathcal{X} (resp. 𝒯𝒴S𝒴𝒴\mathcal{T}_{\mathcal{Y}}\xrightarrow{S_{\mathcal{Y}}}\mathcal{Y} of 𝒴\mathcal{Y}). Then the homomorphism introduced by Skorobogatov and Zarhin (c.f. [Cao23, (1-4)]) can also be defined, that is,

ϵ:Homk(S𝒳/nS𝒴,μn)\displaystyle\epsilon^{\prime}:\operatorname{Hom}_{k}(S_{\mathcal{X}}\otimes_{\mathbb{Z}/n\mathbb{Z}}S_{\mathcal{Y}},\mu_{n}) Hét2(𝒳×k𝒴,μn)\displaystyle\rightarrow H_{\text{\'{e}t}}^{2}(\mathcal{X}\times_{k}\mathcal{Y},\mu_{n})
ϕ\displaystyle\phi ϕ(𝒯𝒳𝒯𝒴).\displaystyle\mapsto\phi_{*}(\mathcal{T}_{\mathcal{X}}\cup\mathcal{T}_{\mathcal{Y}}).
Lemma 7.5.

Let 𝒳\mathcal{X} and 𝒴\mathcal{Y} be algebraic kk-stacks satisfying the assumptions of Lemma 7.4, Then we have isomorphisms of Γk\Gamma_{k}-modules

(7.6) (p1,p2):Hét1(𝒳¯,μn)Hét1(𝒴¯,μn)Hét1(𝒳¯×k¯𝒴¯,μn),\displaystyle(p_{1}^{*},p_{2}^{*}):H_{\text{\'{e}t}}^{1}(\overline{\mathcal{X}},\mu_{n})\oplus H_{\text{\'{e}t}}^{1}(\overline{\mathcal{Y}},\mu_{n})\xrightarrow{\sim}H_{\text{\'{e}t}}^{1}(\overline{\mathcal{X}}\times_{\overline{k}}\overline{\mathcal{Y}},\mu_{n}),
(7.7) (p1,p2,ϵ):Hét2(𝒳¯,μn)Hét2(𝒴¯,μn)Homk¯(S𝒳¯/nS𝒴¯,μn)Hét2(𝒳¯×k¯𝒴¯,μn).\displaystyle(p_{1}^{*},p_{2}^{*},\epsilon^{\prime}):H_{\text{\'{e}t}}^{2}(\overline{\mathcal{X}},\mu_{n})\oplus H_{\text{\'{e}t}}^{2}(\overline{\mathcal{Y}},\mu_{n})\oplus\operatorname{Hom}_{\overline{k}}(S_{\overline{\mathcal{X}}}\otimes_{\mathbb{Z}/n\mathbb{Z}}S_{\overline{\mathcal{Y}}},\mu_{n})\xrightarrow{\sim}H_{\text{\'{e}t}}^{2}(\overline{\mathcal{X}}\times_{\overline{k}}\overline{\mathcal{Y}},\mu_{n}).
Proof.

The original proof (c.f. [Cao23, Thm. 2.1]) is also available if we use lemmas of stacky version (Lemmas 7.2 and 7.3). ∎

7.3. Proof of the product preserving property

Now we can prove Theorem 7.1. Note hp1/k{\mathbb{C}\text{hp}}_{1}/k is closed under products, we need only to verify that

𝒳(𝐀k)Br×𝒴(𝒜)Br(𝒳×k𝒴)(𝐀k)Br\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}\times\mathcal{Y}(\mathcal{A})^{\operatorname{Br}}\xrightarrow{\sim}(\mathcal{X}\times_{k}\mathcal{Y})(\mathbf{A}_{k})^{\operatorname{Br}}

for any 𝒳,𝒴hp1/k\mathcal{X},\mathcal{Y}\in{\mathbb{C}\text{hp}}_{1}/k. We go through the proofs in [SZ14, Sec. 5]. Since 𝒳\mathcal{X} and 𝒴\mathcal{Y} are smooth, of finite type, either Deligne-Mumford kk-stacks or Zariski-locally quotients of kk-varieties by linear kk-groups, there Brauer groups are torsion by Corollary 3.12. Also, there exists a universal torsors of nn-torsion 𝒯𝒳S𝒳𝒳\mathcal{T}_{\mathcal{X}}\xrightarrow{S_{\mathcal{X}}}\mathcal{X} of 𝒳\mathcal{X} and 𝒯𝒴S𝒴𝒴\mathcal{T}_{\mathcal{Y}}\xrightarrow{S_{\mathcal{Y}}}\mathcal{Y} of 𝒴\mathcal{Y}, since both 𝒳\mathcal{X} and 𝒴\mathcal{Y} satisfy the assumptions of Lemma 7.4. Then we use [SZ14, Lemmas 5.2 and 5.3] to finish the proof. The only non-trivial thing is to replace [SZ14, Cor. 2.8 (resp. Thm. 2.8] by (7.6) (resp. (7.7)). The proof is complete.

Corollary 7.8.

If 𝒳\mathcal{X} and 𝒴\mathcal{Y} are stacks quotients of smooth geometrically integral kk-varieties by connected linear kk-groups. Then

𝒳(𝐀k)Br×𝒴(𝒜)Br(𝒳×k𝒴)(𝐀k)Br\mathcal{X}(\mathbf{A}_{k})^{\operatorname{Br}}\times\mathcal{Y}(\mathcal{A})^{\operatorname{Br}}\xrightarrow{\sim}(\mathcal{X}\times_{k}\mathcal{Y})(\mathbf{A}_{k})^{\operatorname{Br}}
Proof.

Combine Theorems 6.4 and 7.1. ∎

Acknowledgment

The authors would like to thank Ye Tian, Weizhe Zheng, Dasheng Wei, Yang Cao, Shizhang Li and Junchao Shentu for helpful discussions. The work was partially done during the authors’ visits to the Morningside Center of Mathematics, Chinese Academy of Sciences. They thank the Center for its hospitality.

References

  • [AM20] Benjamin Antieau and Lennart Meier, The Brauer group of the moduli stack of elliptic curves, Algebra Number Theory 14 (2020), no. 9, 2295–2333. MR 4172709
  • [Art70] M. Artin, Algebraization of formal moduli. II. Existence of modifications, Ann. of Math. (2) 91 (1970), 88–135. MR 260747
  • [BG19] Cristiana Bertolin and Federica Galluzzi, Gerbes and Brauer groups over stacks, arXiv preprint arXiv:1705.01382v4 (2019).
  • [BP20] M. Bhargava and B. Poonen, The local-global principle for integral points on stacky curves, Preprint, arXiv:2006.00167 [math.NT] (2020).
  • [Cao18] Yang Cao, Approximation forte pour les variétés avec une action d’un groupe linéaire, Compos. Math. 154 (2018), no. 4, 773–819. MR 3778194
  • [Cao20] by same author, Sous-groupe de Brauer invariant et obstruction de descente itérée, Algebra Number Theory 14 (2020), no. 8, 2151–2183. MR 4172704
  • [Cao23] by same author, Correction à l’article Sous-groupe de Brauer invariant et obstruction de descente itérée, Algebra Number Theory 17 (2023), no. 1, 261–266. MR 4549285
  • [CDX19] Yang Cao, Cyril Demarche, and Fei Xu, Comparing descent obstruction and Brauer-Manin obstruction for open varieties, Trans. Amer. Math. Soc. 371 (2019), no. 12, 8625–8650. MR 3955558
  • [Chr20] Atticus Christensen, A topology on points on stacks, arXiv preprint arXiv:2005.10231 (2020).
  • [CTS87] Jean-Louis Colliot-Thélène and Jean-Jacques Sansuc, La descente sur les variétés rationnelles. II, Duke Math. J. 54 (1987), no. 2, 375–492. MR 899402
  • [Dem11] Cyril Demarche, Le défaut d’approximation forte dans les groupes linéaires connexes, Proc. Lond. Math. Soc. (3) 102 (2011), no. 3, 563–597. MR 2783137
  • [DG95] Henri Darmon and Andrew Granville, On the equations zm=F(x,y)z^{m}=F(x,y) and Axp+Byq=CzrAx^{p}+By^{q}=Cz^{r}, Bull. London Math. Soc. 27 (1995), no. 6, 513–543. MR 1348707
  • [DM69] P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus, Inst. Hautes Études Sci. Publ. Math. (1969), no. 36, 75–109. MR 262240
  • [Fu11] Lei Fu, Etale cohomology theory, Nankai Tracts in Mathematics, vol. 13, World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2011. MR 2791606
  • [Gir71] Jean Giraud, Cohomologie non abélienne, Springer-Verlag, Berlin-New York, 1971, Die Grundlehren der mathematischen Wissenschaften, Band 179. MR 0344253
  • [Gro68] Alexander Grothendieck, Le groupe de Brauer. I-III, Dix exposés sur la cohomologie des schémas, Adv. Stud. Pure Math., vol. 3, North-Holland, Amsterdam, 1968, pp. 46–66, 67–87, 88–188.
  • [Har02] David Harari, Groupes algébriques et points rationnels, Math. Ann. 322 (2002), no. 4, 811–826. MR 1905103
  • [HS13] David Harari and Alexei N. Skorobogatov, Descent theory for open varieties, Torsors, étale homotopy and applications to rational points, London Math. Soc. Lecture Note Ser., vol. 405, Cambridge Univ. Press, Cambridge, 2013, pp. 250–279. MR 3077172
  • [Lv20] Chang Lv, A note on the Brauer group and the Brauer-Manin set of a product, Bull. Lond. Math. Soc. 52 (2020), no. 5, 932–941. MR 4171413
  • [Lv21] by same author, The second descent obstruction and gerbes, arXiv preprint arXiv:2009.06915v2 (2021), 1–31.
  • [Lv22] by same author, On commutative diagrams consisting of low term exact sequences, arXiv preprint arXiv:2112.14386v2 (2022), 1–9.
  • [LZ17] Yifeng Liu and Weizhe Zheng, Enhanced six operations and base change theorem for higher Artin stacks, arXiv preprint arXiv:1211.5948v3 (2017).
  • [Mil80] J. S. Milne, Étale cohomology, Princeton Mathematical Series, vol. 33, Princeton Univ. Press, 1980.
  • [Mil06] James Milne, Arithmetic duality theorems, BookSurge, 2006.
  • [Ols16] Martin Olsson, Algebraic spaces and stacks, American Mathematical Society Colloquium Publications, vol. 62, American Mathematical Society, Providence, RI, 2016. MR 3495343
  • [Poo17] Bjorn Poonen, Rational points on varieties, vol. 186, American Mathematical Soc., 2017.
  • [San81] J-J Sansuc, Groupe de Brauer et arithmétique des groupes algébriques linéaires sur un corps de nombres., Journal für die reine und angewandte Mathematik 327 (1981), 12–80.
  • [San22] Tim Santens, The Brauer-Manin obstruction for stacky curves, arXiv preprint arXiv:2210.17184 (2022).
  • [Ser94] Jean-Pierre Serre, Galois cohomology, tanslated from the french by Patrick Ion, Springer-Verlag, 1994.
  • [Sta22] The Stacks project authors, The stacks project, https://stacks.math.columbia.edu, 2022.
  • [SZ14] Alexei N. Skorobogatov and Yuri G. Zarhin, The Brauer group and the Brauer-Manin set of products of varieties, J. Eur. Math. Soc. (JEMS) 16 (2014), no. 4, 749–768. MR 3191975
  • [WL22] Han Wu and Chang Lv, Genus one half stacky curves violating the local-global principle, arXiv preprint arXiv:2203.05810 (2022), 1–5.
  • [Zah03] Stéphane Zahnd, Descente de torseurs, gerbes et points rationnels, Ph.D. thesis, Université des Sciences et Technologie de Lille-Lille I, 2003.