This paper introduces a natural homomorphism linking the relative Brauer group to étale cohomology for affine maps of schemes, and explores its properties and related sequences.
Contribution
It constructs a natural homomorphism from the relative Brauer group to étale cohomology and proves a relative Kummer sequence, advancing understanding of Brauer groups in algebraic geometry.
Findings
01
Constructed a natural homomorphism from Br(f) to étale cohomology.
02
Analyzed the behavior of Brauer groups under subintegral maps.
03
Proved a relative version of Kummer's exact sequence.
Abstract
In this article, we construct a natural group homomorphism ψ:Br(f)→Het1(S,f∗OX×/OS×) for a faithful affine map f:X→S of noetherian schemes. Here Br(f) denotes the relative Brauer group of f. We also discuss the behavior of Brauer groups for a subintegral map. Furthermore, we prove a relative version of Kummer's exact sequence.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Full text
Relative Brauer Groups and étale cohomology
Vivek Sadhu
Department of Mathematics, Indian Institute of Science Education and Research, Bhopal, Bhopal Bypass Road, Bhauri, Bhopal-462066, Madhya Pradesh, India
In this article, we construct a natural group homomorphism
[TABLE]
for a faithful affine map f:X→S of noetherian schemes. Here Br(f) denotes the relative Brauer group of f. We also prove Br(f)=0 whenever f:A↪B is a subintegral extension of noetherian Q-algebras.
Furthermore, we prove a relative version of Kummer’s exact sequence.
Author was supported by SERB-DST MATRICS grant MTR/2018/000283
1. Introduction
For a map f:X→S of schemes, we define the relative Brauer group Br(f) as the Grothendieck group of a certain relative category Az(f∗), i.e., Br(f):=K0(Az(f∗))(see section 2). This Br(f) fits into a natural exact sequence
[TABLE]
Classically, the relative Brauer group is just the kernel of the natural map Br(S)→f∗Br(X) and is denoted by Br(X/S). Note that the relative Brauer group Br(f) is different from the classical one Br(X/S). But in the case of field extension, i.e., if f:L↪K is a field extension then Br(f) is isomorphic to Br(K∣L). The details related to Br(K∣L) can be found in [7], [8], [9]. In general, we have a natural map Br(f)→Br(X/S) and, this map is an isomorphism if and only if Pic(S)→Pic(X) is surjective. .
Now assume that f:X→S is a faithful affine map of schemes, i.e., affine and the structure map OS→f∗OX is injective. We define
[TABLE]
If f:L↪K is a finite field extension then Br(f)≅Br′(f) (see Lemma 3.1). But Br(f) and Br′(f) are different in general (see Example 3.3). This is why we prefer to call the latter group as a relative cohomological Brauer group. One of the goals of this article is to relate Br(f) and Br′(f). We prove the following (see Theorem 4.12):
Theorem 1.1**.**
Let f:X→S be a faithful affine map of noetherian schemes. Then there is a natural group homomorphism θ:Br(f)→Br′(f):=Het1(S,f∗OX×/OS×).
We give an outline of the proof of Theorem 1.1. It is known that there is a natural isomorphism Hˇet1(S,F)→≅Het1(S,F) for every abelian sheaf F. Here Hˇet1(S,F) denotes the first étale Čech cohomology group associated to F. So we can restrict the target of the map θ to the first étale Čech cohomology group associated to (f∗OX×/OS×)et. Let Picetf be the étale sheaf associated to the presheaf Picf on Set, defined as Picf(U)=Pic(fU), where fU:X×SU→U and the group Pic(fU) is generated by pairs (L1,L2) of line bundles on U together with an isomorphism fU∗L1≅fU∗L2 of OX×SU-modules with suitable relations (for details, see [19]). There is a natural isomorphism (f∗OX×/OS×)et≅Picetf (see Lemma 4.11). Some well-known facts pertaining to (pre)-sheaf torsors and the first étale Čech cohomology group allow us to restrict further the target of the map θ to the set of isomorphism classes of Picf-torsors. Therefore, the problem boils down to defining a Picf-torsor associated to each element of Br(f). Given an Azumaya algebra A over a scheme X, one can associate a fibered category FA over Xet which in fact defines an element in Het2(X,OX×). More explicitly, A↦FA defines a natural map Br(X)→Het2(X,OX×) (see [10], [12]). We follow a similar approach to prove our desired assertion. Note that the relative Br(f) is the abelian group generated by isomorphism classes of objects in Az(f∗) modulo certain relations. An object of Az(f∗) is a triple (A1,α,A2), where A1,A2 are Azumaya algebras over S and α:f∗A1≅f∗A2 is an isomorphism in a suitable category. So we have the fibered categories FA1 and FA2 associated to A1 and A2. By using the categories FA1 and FA2, we define a relative category GA over Set associated to A:=(A1,α,A2). Further, using the category GA, we construct a Picf-torsor associated to A:=(A1,α,A2) (see (4.8) and (4.9)). A significant portion of section 4 is dedicated to the construction of such a presheaf torsor and proving its required properties.
Next, we discuss the relative Brauer group of subintegral extensions. We say that an extension A↪B is subintegral if B is integral over A and Spec(B)→Spec(A) is a bijection inducing isomorphisms on all residue fields. For example, C[t2,t3]↪C[t] is a subintegral extension. We show that if f:A↪B is a subintegral extension of noetherian Q-algebras then the induced map Br(A)→f∗Br(B) is an isomorphism and Br(f)=0 (see Theorem 5.1).
We further study the Kummer exact sequence in the relative setting. Write Iet for the étale sheaf f∗OX×/OS×. Let μnf denotes the kernel of Iet→nIet. We prove that if f:X→S is a faithful finite map of schemes (i.e. finite and the structure map OS→f∗OX is injective) and characteristic of k(s) does not divide n for any s∈S then the sequence
[TABLE]
of étale sheaves is exact (see Proposition 6.2). As an application, we obtain Het0(S,μnf)≅nPic(f) and the following short exact sequence (see Theorem 6.3)
[TABLE]
where Pic(f) is the relative Picard group of f (see [19]). We also show that if f:A↪B is a finite subintegral extension of noetherian Q-algebras then Heti(Spec(A),μnf)=0 for i≥0 (see Theorem 6.4).
Acknowledgements: The author is grateful to Charles Weibel for his helpful comments during the preparation of this article. He would also like to thank the referee for valuable comments and suggestions.
2. Relative Brauer groups
In this section, we define the notion of relative Brauer groups. The most of the material in this section was developed in [1], [2], [3], [4], [6] by various authors.
Recall that two Azumaya algebras A and A′ over a scheme X are said to be similar if there
exist locally free OX-modules E and E′, of finite rank over OX, such that
[TABLE]
The set of similar classes of Azumaya algebras on X forms an abelian group under ⊗OX, which is known as the Brauer group Br(X) of X. If [A]∈Br(X) then [A]−1=[Aop], where Aop denotes the opposite algebra of A. The cohomological Brauer group of X is Het2(X,OX×) and is denoted by Br′(X). The Picard group of X is denoted by Pic(X).
Given two Azumaya algebras A and B over X, let
Δ(A,B) (resp. Δ~(A,B)) be the set consisting of all triples (P,u,Q) with P, Q are (resp. self dual) locally free OX-modules of finite rank and
[TABLE]
is an isomorphism of algebras. We define an equivalence relation ∼ on Δ(A,B) by (P,u,Q)∼(P′,u′,Q′) if and only if there exist locally free OX-modules E and E′ of finite rank over X and P⊗E≅P′⊗E′,Q⊗E≅Q′⊗E′. By considering E and E′ are to be self dual locally free OX-modules, one can check that ∼ is an equivalence relation on Δ~(A,B) as well.
Remark 2.1**.**
Note that u and u′ do not play any role for the equivalence relation ∼ on Δ(A,B) (resp. Δ~(A,B)). In fact (P,u,Q) and (P,v,Q) both lie in a same class whenever u,v are possibly distinct OX-algebra isomorphisms between A⊗End(P) and B⊗End(Q).**
The category Az(X)
Az(X) is the category whose objects are the Azumaya algebras over a scheme X and the set of morphisms between two objects A and B is defined by
[TABLE]
Let ϕ1=[(P1,u1,Q1)]∈HomAz(X)(A,B),ϕ2=[(P2,u2,Q2)]∈HomAz(X)(B,C). Then we define the composition ϕ2ϕ1:=[(P1⊗P2,u,Q1⊗Q2)] with u:A⊗End(P1⊗P2)≅A⊗End(P1)⊗End(P2)≅u1B⊗End(Q1)⊗End(P2)≅u2C⊗End(Q2)⊗End(Q1)≅C⊗End(Q1⊗Q2). One can easily check that this composition is independent of the chosen representatives of ϕ1 and ϕ2. If ϕ=[(P,u,Q)]∈HomAz(X)(A,B) then ϕ is an isomorphism with inverse ϕ−1=[(Q,u−1,P)] (see Remark 5.7 of [6]). The tensor product ⊗ on OX-algebras defines a product on Az(X).
The category ΩAz(X)
ΩAz(X) is the category consisting of all couples (A,ϕ) with A∈Az(X) and ϕ is an automorphism in Az(X). A morphism (A,ϕ)→(B,φ) is a morphism h:A→B in Az(X) such that φh=hϕ. Note that ΩAz(X) is a category with product and composition in the sense of [3]. We refer to 5.12 of [6] for product and composition rules in ΩAz(X).
Let X be a scheme. Then there are natural isomorphisms of abelian groups
(1)
K0(Az(X))→≅Br(X);**
2. (2)
K0(ΩAz(X))→≅Pic(X).**
Proof.
We get the assertions by applying Theorem 5.9 and Theorem 5.13 of [6] to the category of OX-modules.
∎
For a map f:X→S of schemes, we have the base change functor f∗:Az(S)→Az(X).
The category Az(f∗)
Az(f∗) is the category consisting of all triples (A,α,B) where A,B∈Az(S) and α:f∗(A)→f∗(B) is an isomorphism in Az(X). Here α=[(P,u,Q)] with P, Q are locally free OX-modules of finite rank and u:f∗A⊗End(P)≅f∗B⊗End(Q)(i.e. f∗A is similar to f∗B). A morphism (A,α,B)→(A′,α′,B′) is a pair (u,v), where u:A→A′ and v:B→B′ are morphisms in Az(S) such that α′f∗u=f∗vα. We refer to 5.17 of [6] for product and composition rules in Az(f∗).
Following Bass [3], we define the relative Brauer groupBr(f) of f to be the Grothendieck group K0(Az(f∗)), i.e., the abelian group generated by [(A1,α,A2)], where (A1,α,A2)∈Az(f∗), and with the following relations:
By (2), [(A,1,A)]=0 for any A∈Az(S). Then by using (1), we get that every element of Br(f) has the form [(A,α,EndOS(B))], where A and B are in Az(S). For general details, see Remark 1.3 in Chapter 1 of [3].
By Theorem 5.18 of [6] (more precisely, Example 5.19(2) of [6]), there is a natural exact sequence of abelian groups for each f:X→S,
[TABLE]
Here ϱ([(A1,α,A2)])=A1⊗A2op and the map ∂ is defined as follows. By Lemma 2.2(2), we can identify Pic(X) with K0(ΩAz(X)). Let Az′(X) denote the full subcategory of Az(X) whose objects are all f∗A,A∈Az(S). The functor f∗:Az(S)→Az(X) is cofinal (see Proposition 5.15 of [6]). So, Az′(X) is a cofinal subcategory of Az(X). We refer to p. 19 of [3] for general details related to cofinal functors and cofinal subcategories. Now the Theorem 3.1(b) of [3] says that there is an isomorphism κ:K0(ΩAz′(X))→K0(ΩAz(X)). We also have a natural homomorphism ∂1:K0(ΩAz′(X))→K0(Az(f∗)), sending [(f∗A,α)] to [(A,α,A)]. Hence we define ∂:K0(ΩAz(X))→K0(Az(f∗)) as ∂1κ−1.
Remark 2.4**.**
Suppose that S=Spec(Z), i.e., f:X→Spec(Z). By Proposition 4.2 of [20], Br(Z)=0. Then the sequence (2.1) implies that Br(f)≅Pic(X).
Next, we define a slight variant of the group Br(f) which will be used in the section 4.
The categories A~z(X) and A~z(f∗)
A~z(X) is the category whose objects are the Azumaya algebras over a scheme X and the set of morphisms between two objects A and B is defined by
[TABLE]
A~z(f∗) is the category consisting of all triples (A,α,B) where A,B∈A~z(S) and α:f∗(A)→f∗(B) is an isomorphism in A~z(X). A morphism (A,α,B)→(A′,α′,B′) is a pair (u,v), where u:A→A′ and v:B→B′ are morphisms in A~z(S) such that α′f∗u=f∗vα. One can define a product and composition on A~z(X) (resp. A~z(f∗)) in a similar way as it is defined on Az(X) (resp. Az(f∗)).
We define Br~(f):=K0(A~z(f∗)). Given A:=(A1,α,A2)∈Az(f∗) with α=[(P,u,Q)], let End(A):=(EndOS(A1),End(α),EndOS(A2)), where
[TABLE]
with
[TABLE]
Observe that (EndOX(P),End(u),EndOX(Q))∈Δ~(f∗EndOS(A1),f∗EndOS(A2)). So End(A)∈A~z(f∗).
We define a map End:ob(Az(f∗))→Br~(f) by A↦[End(A)]. The following facts are easy to verify:
(1)
If A:=(A1,α,A2)≅B:=(B1,β,B2) in Az(f∗) then [End(A)]≅[End(B)] in Br~(f).
2. (2)
[End(A1⊗A1′,α⊗α′,A2⊗A2′)]=[End(A1,α,A2)]+[End(A1′,α′,A2′)], for every pair of objects (A1,α,A2),(A1′,α′,A2′) in Az(f∗).
3. (3)
[End(A1,βα,A3)]=[End(A1,α,A2)]+[End(A2,β,A3)], for every pair of objects (A1,α,A2),(A2,β,A3) in Az(f∗).
Therefore, the map End induces a natural group homomorphism
[TABLE]
sending [(A1,α,A2)] to [End((A1,α,A2))].
3. Relative Cohomological Brauer Groups
Let f:L↪K be a field extension. Then the relative Brauer group in the classical sense is just the kernel of the natural map Br(L)→Br(K) and it was denoted by Br(K∣L)(for details, see [7], [8], [9]). We begin with the following observation.
Lemma 3.1**.**
If f:L↪K is a finite field extension then
Het1(Spec(L),f∗OK×/OL×)≅Br(K∣L).
Proof.
For a field F, Het2(F,OF×)≅Br(F) and Het1(F,OF×)≅Pic(F)=0. Then (by using the exact sequence (2.1))
[TABLE]
∎
Motivated by the above lemma, we define
[TABLE]
for a faithful affine map f:X→S of schemes. We call it relative cohomological Brauer group of f.
Example 3.2**.**
Suppose that f:X→S is faithful finite, i.e., finite and the structure map OS→f∗OX is injective and S=Spec(A), where A is a hensel local ring. Then X=Spec(B) and B is a finite product of hensel local rings. In this situation, Br(S)≅Het2(S,OS×) and Br(X)↪Het2(X,OX×) by Corollary 2.12 of [12]. Therefore, by a similar argument as Lemma 3.1, Br(f)≅Br′(f).
Example 3.3**.**
In general, Br(f) and Br′(f) are not isomorphic for any faithful affine f:X→S. For example, let S=Spec(A), where A is a strictly hensel local ring. Then X=Spec(B), and Br′(f)=0 because the higher cohomology vanishes for a strictly henselian ring. We also have Br(S)=0. Therefore Br(f)≅Pic(B) by (2.1). In particular, one can consider C⊂C[t2,t3].
Example 3.4**.**
Suppose that f:X→S is faithful finite and S=Spec(A), where A is a local ring. Then X=Spec(B) and B is a semilocal ring because B is finite over a local ring. By (2.1),
[TABLE]
is exact and Br(f)↪Br′(f). We know that if a scheme X has only finitely many connected components then Br(X) is torsion (see Proposition 2.7 of [12]). Therefore Br(f) is torsion. By Theorem 1.1 of [5], Br(A)≅Het2(A,OA×)tor and same holds for B. In fact, we get Br(f)≅Het1(S,f∗OX×/OS×)tor.
Example 3.5**.**
The group Br(f) is not always torsion. For example, consider f:X=Spec(Q[t2−t,t3−t2])→S=Spec(Q). Then Pic(X)≅Q×. By (2.1), we get that Br(f) contains the torsion free subgroup Q×.
4. Main Theorem
The main goal of this section is to construct a natural group homomorphism Br(f)→Het1(S,f∗OX×/OS×). Throughout this section, f:X→S assumed to be a faithful affine map of schemes and fU always denotes the map X×SU→U.
The category FA
Given an Azumaya algebra A on X one can associate a category FA over Xet as follows (see p. 145 of [12]). For an étale map j:U→X, an object of FA(U) is a pair (E,τ), where E is a locally free OU-module of finite rank and τ is an isomorphism End(E)≅j∗A; a morphism (E,τ)→(E′,τ′) is an isomorphism E→E′ such that the obvious diagram
[TABLE]
commutes.
Notation: Given a sheaf F of OX-modules, we write F∨ for the dual of F.
The category GA
Given an object A:=(A1,α,A2) in A~z(f∗) (or in Az(f∗)), we can associate a category GA over Set as follows. For an étale map j:U→S, an object of GA(U) is a triple ((E1,τ1),ν,(E2,τ2)), where (Ei,τi)∈FAi(U) for i=1,2 and ν is an isomorphism fU∗E1⊗(fU∗E2)∨≅(fU∗E1)∨⊗fU∗E2 of OX×SU-modules. Here FAi are the categories associated to Ai. Similarly, a morphism ((E1,τ1),ν,(E2,τ2))→((E1′,τ1′),ν′,(E2′,τ2′)) is a pair of isomorphisms E1→E1′ and E2→E2′ such that the diagram like (4.1) and
[TABLE]
commute.
Let [GA(U)] be the set of all isomorphism classes of objects of GA(U). We write [((E1,τ1),ν,(E2,τ2))] for the isomorphism class of ((E1,τ1),ν,(E2,τ2)). Then
the assignment
[TABLE]
is a presheaf of sets on Set, where the restriction maps are given by pullbacks. It is denoted by [GA].
Presheaf torsors
Let C be a site. Let G be a presheaf of groups on C. A G-torsor is a presheaf of sets F on C equipped with an action ρ:G×F→F such that
(1)
the action ρ(U):G(U)×F(U)→F(U) is simply transitive provided F(U) is nonempty.
2. (2)
for every U∈ob(C) there exists a covering {Ui→U}i∈I of U such that F(Ui) is nonempty for all i.
A morphism of G-torsors F→F′ is just a morphism of presheaves of sets compatible with the G-action. A trivial G-torsor is the presheaf G with obvious left G-action. Note that a morphism between G-torsors is always an isomorphism. Moreover, a G-torsor F is trivial if and only if Γ(C,F)=∅ (see chapter 21 of [16]).
Relative Picard groups
The relative Pic(f) is the abelian group generated by
[L1,α,L2], where the Li are
line bundles on S and α:f∗L1→f∗L2 is an isomorphism.
The relations are:
This relative Picard group Pic(f) fits into the following exact sequence
[TABLE]
Some relevant details and basic properties can be found in [18], [19].
Next, we define a group which is a slight variant of the group Pic(f).
The group ~Pic(f)
The relative ~Pic(f) is the abelian group generated by [L1,α,L2], where the Li are line bundles on S and α:f∗L1⊗f∗L2−1→f∗L1−1⊗f∗L2 is an isomorphism. The relations are similar to Pic(f).
There is a natural group homomorphism
[TABLE]
sending [L1,α,L2] to [L1⊗L2−1,α,L1−1⊗L2].
Let Picf be the étale presheaf on Set, defined as Picf(U)=Pic(fU). Similarly, we can define the étale presheaf ~Picf on Set.
Lemma 4.1**.**
Let V1,V2 and V3 be three locally free OX-modules of finite rank. Suppose that ι:V1⊗V3→V2⊗V3 is an isomorphism as an OX-module. Then there is an OX-module isomorphism V1→V2.
Proof.
We know that F⊗End(F)F∨≅OX for any locally free OX-module F of finite rank. Then we get
V1→≅V1⊗V3⊗End(V3)V3∨→ι⊗idV2⊗V3⊗End(V3)V3∨→≅V2.∎
Lemma 4.2**.**
Let A:=(A1,α,A2) be in A~z(f∗)(or in Az(f∗)). Let ((E1,τ1),ν,(E2,τ2)),((E1′,τ1′),ν′,(E2′,τ2′)) be any two objects in GA(U), where U∈ob(Set). Then
(1)
we can construct a unique element of ~Picf(U). Call it [L1,β,L2].
2. (2)
((L1⊗E1,τ1a1),β⊗ν,(L2⊗E2,τ2a2))* defines an object in GA(U), where ak:End(Ek⊗Lk)≅End(Ek) for k=1,2. Moreover, ((L1⊗E1,τ1a1),β⊗ν,(L2⊗E2,τ2a2)) is isomorphic to ((E1′,τ1′),ν′,(E2′,τ2′)) in GA(U).*
Proof.
For simplicity, we write E(Ek) instead of End(Ek) throughout the proof.
(1) By definition, for k=1,2, we have
[TABLE]
where j denotes the étale map U→S. Then τk′−1τk:E(Ek)≅E(Ek′). So, we can find unique invertible sheaves Lk=Ek′⊗E(Ek)Ek∨ such that (see Lemma 4.3 of [6])
[TABLE]
Note that vk’s are (E(Ek),OU)-linear isomorphisms. By using ν,ν′ and vk, we obtain an OX×SU-module isomorphism
[TABLE]
Set
[TABLE]
[TABLE]
[TABLE]
Thus, we have
[TABLE]
where 1L∨ denotes the identity map on L∨.
Now the Lemma 4.1 implies that there is an OX×SU-module isomorphism β:L≅L∨. More precisely, we get β as follows:
[TABLE]
Here δ=Γ∨ν′Γ. Therefore, [L1,β,L2]∈~Picf(U). This proves (1).
(2) Note that β⊗ν gives an isomorphism between L⊗E and L∨⊗E∨. Hence, clearly ((L1⊗E1,τ1a1),β⊗ν,(L2⊗E2,τ2a2)) defines an object in GA(U), where ak:E(Ek⊗Lk)≅E(Ek) for k=1,2.
Next, we shall show that (v1,v2) defines an isomorphism between ((L1⊗E1,τ1a1),β⊗ν,(L2⊗E2,τ2a2)) and ((E1′,τ1′),ν′,(E2′,τ2′)) in GA(U). To prove (v1,v2) is an isomorphism in GA(U), we have to check that the diagrams like (4.1) and (4.2) commute. To check these commutativity, we may assume that U is affine.
We first check that the following diagram
[TABLE]
commutes.
For s∈E(E1′), we have E(v1)(s)=v1−1sv1. Note that E(v1)−1=E(v1−1). Let τ=τ1′−1τ1. So, it suffices to check that E(v1−1)a1−1=τ, i.e., (E(v1−1)a1−1)(t)=τ(t) for all t∈E(E1). We can write a1−1(t)=1L1⊗t. Moreover, v1 is E(E1)-linear map and E1′ is E(E1)-module via τ. Using these facts, we get ((E(v1−1)a1−1)(t))(e′)=τ(t)(e′) for all e′∈E1′. Hence the assertion.
Further, we show that the diagram like (4.2) commutes. By (4.5), we have the following commutative diagram
[TABLE]
We claim that β⊗ν=δ. Recall that β=b(1L∨⊗ν−1⊗1E∨)(δ⊗1E∨)a. Then β⊗ν=(b⊗ν)(1L∨⊗ν−1⊗1E∨⊗1E)(δ⊗1E∨⊗1E)(a⊗1E). We get the following commutative diagram
[TABLE]
Here Γ1=a⊗1E,Γ2=δ⊗1E∨⊗1E,Γ3=(b⊗ν).(1L∨⊗ν−1⊗1E∨⊗1E) and m,n are natural isomorphisms sending x⊗e~ to xe~ for all e~∈E(E). One can easily check that mΓ1 and nΓ3−1 both are identity isomorphisms. Hence β⊗ν=δ. This completes the proof. ∎
Given an étale map U→S, we define a group action ρ(U):~Picf(U)×[GA](U)→[GA](U) by
[TABLE]
Here L1,L2 are line bundles on U with an isomorphism β:fU∗L1⊗fU∗L2−1≅fU∗L1−1⊗fU∗L2 and ak:End(Ek⊗Lk)≅End(Ek) for k=1,2.
Proposition 4.3**.**
Let A:=(A1,α,A2) be in A~z(f∗)(or in Az(f∗)). Assume that S is connected. Then [GA] is a ~Picf-torsor.
Proof.
We need to show that the following are true:
(1)
the action ρ(U) is simply transitive,
2. (2)
for every U∈ob(Set) there exist a étale covering {Ui→U} such that [GA](Ui)=∅ for all i.
Let [(E1,τ1),ν,(E2,τ2)],[(E1′,τ1′),ν′,(E2′,τ2′)]∈[GA](U). Then by Lemma 4.2, we can construct a unique [L1,β,L2]∈~Picf(U) such that
[TABLE]
This proves (1).
Since S is connected, A1 and A2 have constant rank say n2 and k2. Let j:U→S be an étale map. Then j∗A1 and j∗A2 are Azumaya algebras on U of rank n2 and k2. By Proposition IV.2.1(d) of [12], there exist an étale covering {bi:Ui→U} such that
[TABLE]
[TABLE]
for each i. This implies that [(OUi⊕n,τ1),ν,(OUi⊕k,τ2)]∈[GA](Ui) for all i. Here ν is the obvious isomorphism. Hence [GA](Ui)=∅ for all i. This proves (2). ∎
Remark 4.4**.**
Note that the action ρ(U) is simply transitive without the assumption S being connected.**
Lemma 4.5**.**
Let ψ:A:=(A1,α,A2)→B:=(B1,β,B2) be a morphism in A~z(f∗). Then the induced map [GA]→[GB] is an isomorphism as ~Picf-torsors.
Proof.
Since α and β are morphisms in A~z(X),α=[(P,u,Q)] and β=[(R,v,T)]. We have u:f∗A1⊗End(P)≅f∗A2⊗End(Q) and v:f∗B1⊗End(R)≅f∗B2⊗End(T). Write ψ=(ψ1,ψ2), where ψ1:=[(P1,u1,Q1)] and ψ2:=[(R1,v1,T1)] are morphisms in A~z(S). In view of Remark 2.1, we prefer to write α=[(P,Q)],β=[(R,T)],ψ1=[(P1,Q1)] and ψ2=[(R1,T1)]. Since ψ is a morphism, the following diagram
[TABLE]
commutes, i.e.,
[TABLE]
This means that there exist self dual locally free OX-modules H1 and H2 of finite rank over X such that
[TABLE]
Let ϕ1=([(P1,OS)],[(R1,OS)]) and ϕ2=([(Q1,OS)],[(T1,OS)]).
We claim that
[TABLE]
[TABLE]
both are morphisms in A~z(f∗), where
α′:=[(P⊗f∗R1,Q⊗f∗P1)]andβ′:=[(R⊗f∗T1,T⊗f∗Q1)]. Clearly, [(P1,OS)] and [(R1,OS)]) both are morphisms in A~z(S). Note that (P⊗f∗R1⊗f∗P1,Q⊗f∗P1)∼(P⊗f∗R1,Q) in Δ~(f∗A1,f∗A2⊗f∗End(R1)). Thus the following diagram
[TABLE]
commutes. Therefore, ϕ1 is a morphism in A~z(f∗). Similarly for ϕ2. Hence the cliam.
Moreover α′∼β′, because there exist self dual OX-modules f∗T1⊗H1,f∗P1⊗H2 such that (by using (4.7))
[TABLE]
This implies that ([(OS,OS)],[(OS,OS)]) defines a morphism A′→B′ in A~z(f∗).
Now for an étale map j:U→S,ϕ1 induces a ~Picf(U)-linear map [GA](U)→[GA′](U) by sending
[TABLE]
where τ1′:End(E1⊗j∗P1)≅j∗(A1⊗End(P1)),τ2′:End(E2⊗j∗R1)≅j∗(A2⊗End(R1)) and ν′:fU∗(E1⊗j∗P1)⊗(fU∗(E2⊗j∗R1))∨≅(fU∗(E1⊗j∗P1))∨⊗fU∗(E2⊗j∗R1) (for ν′, we use ν and the fact that P1∨≅P1 and R1∨≅R1). Note that ~Picf(U) acts on both [GA](U),[GA′](U) simply transitively (see Remark 4.4). Thus, we get [GA](U)≅[GA′](U) as sets. Therefore [GA]≅[GA′] as ~Picf-torsors. Similarly for ϕ2, i.e., [GB]≅[GB′]. Clearly, [GA′]≅[GB′] because ([(OS,OS)],[(OS,OS)]) defines a morphism A′→B′ in A~z(f∗). Hence the result. ∎
Remark 4.6**.**
We do not know whether the Lemma 4.5 holds or not whenever ψ is a morphism in Az(f∗).**
Presheaf contracted products
Let G be a presheaf of groups. Let F1 and F2 be two presheaves of sets on a site C. Suppose that F1(resp. F2) has a right (resp. left) G action. We can consider the left action of G on the product given for every object X by:
[TABLE]
Then the presheaf contracted product of F1 and F2 to be the presheaf defined as the quotient by the action of G:
[TABLE]
i.e., the quotient by the equivalence relation define by (xg,y)∼(x,gy) for all g∈G(X). It is denoted by F1ΠGF2. We write [x,y] for the element of F1ΠGF2(X), where X∈ob(C),x∈F1(X) and y∈F2(X). For every X∈ob(C), the left action of G on the contracted product is given by
[TABLE]
Assume that G is a presheaf of abelian groups. Then there is no distinction between left and right G-torsors. Let Tors(S,G) denotes the set of isomorphism classes of G-torsors. It is well known that the set Tors(S,G) has an abelian group structure under the operation presheaf contracted product with identity G and the inverse of F is F itself with the G action (g,x)↦g−1.x.
Lemma 4.7**.**
(1)
Let A:=(A1,α,A2), A′:=(A1′,α′,A2′) be in A~z(f∗). Write A⊗A′:=(A1⊗A1′,α⊗α′,A2⊗A2′). Then [GA⊗A′]≅[GA]ΠP~icf[GA′] as ~Picf-torsors.
2. (2)
Let A:=(A1,α,A2), B:=(A2,β,A3) be in A~z(f∗). Write B∘A:=(A1,βα,A3). Then [GB∘A]≅[GA]ΠP~icf[GB] as ~Picf-torsors.
Proof.
(1) For an étale map U→S, we have a ~Picf(U)-linear map
[TABLE]
sending
[TABLE]
Since ~Picf(U) acts simply transitively on [GA](U)×[GA′](U)/∼ and [GA⊗A′](U), we get [GA](U)×[GA′](U)/∼≅[GA⊗A′](U) as sets. Therefore [GA⊗A′]≅[GA]ΠP~icf[GA′] as ~Picf-torsors.
(2) Let [(E1,τ1),ν,(E2,τ2)]∈[GA](U) and [(E2′,τ2′),ν′,(E3,τ3)]∈[GB](U). Since (E2,τ2) and (E2′,τ2′)∈FA2(U), there exists a unique line bundle L such that E2′≅E2⊗L (see Lemma 4.3 of [6]). The map ν⊗ν′ gives an isomorphism
[TABLE]
of OX×SU-modules. Then by applying Lemma 4.1, we get an OX×SU-module isomorphism χ:fU∗(E1⊗L)⊗(fU∗E3)∨≅(fU∗(E1⊗L))∨⊗fU∗E3. This shows that [(E1⊗L,τ1a1),χ,(E3,τ3)]∈[GB∘A](U), where a1:End(E1⊗L)≅End(E1). Now, we have a ~Picf(U)-linear map
[TABLE]
sending [[(E1,τ1),ν,(E2,τ2)],[(E1′,τ1′),ν′,(E3,τ3)]]to[[(E1⊗L,τ1a1),χ,(E3,τ3)]]. The rest of the arguments similar to (1). Hence the result. ∎
In fact, this induces a well-define group homomorphism
[TABLE]
by Lemma 4.7. Moreover, ω is a natural map (see below).
Lemma 4.8**.**
The map ω is natural.
Proof.
Given a commutative diagram
[TABLE]
we want to show that the following diagram
[TABLE]
is commutative. Here h∗([(A1,α,A2)])=[(h∗A1,h∗α,h∗A2)] and Tors(h) is defined as follows. For an étale map V→iS′, (h−1G)(V):=limG(U), where G∈Tors(S,~Picf) and the direct limit is over the commutative diagrams
[TABLE]
with j étale. Now, for such a diagram, we obtain a natural group homomorphism
[TABLE]
Here ε is the unique map X′×S′V→X×SU compatible with the maps g, h and k. By the universal property of direct limits, there is a unique homomorphism h−1~Picf→~Picf′. Then Tors(h)(G)=h−1GΠh−1~Picf~Picf′.
More explicitly, we have to show that there is an isomorphism
[TABLE]
of ~Picf′-torsors. Note that we have a ~Picf(U)-linear map
[TABLE]
sending [(E1,τ1),ν,(E2,τ2)] to [(k∗E1,k∗τ1),ε∗ν,(k∗E2,k∗τ2)]. Therefore, we deduce a ~Picf′(V)-linear map
[TABLE]
by passing to the limit over such all such diagrams. Since ~Picf′(V) acts simply transitively on both sides, we get the result. ∎
Now, by using the maps Υ (see (2.2)) and Ψ (see (4.4)), we define a natural group homomorphism Br(f)→Tors(S,Picf) as follows
[TABLE]
where Tors(S,Ψ) is the natural induced map sending any ~Picf-torsor G to GΠ~PicfPicf.
Lemma 4.9**.**
There is a natural group homomorphism ζ:Tors(S,G)→Hˇet1(S,G), where G is a presheaf of abelian groups and Hˇet1(S,G) denotes the first étale Čech cohomology group associated to G.
Proof.
For the map ζ, we refer to section 11 of [13] or Proposition 4.6 of [12]. The only difference is that we are defining the map here for presheaf torsors. However, we observe that the sheaf properties have not been used in [13] to define such a map. Hence, one can define such map in a similar way for presheaf torsors as well.
By definition Hˇet1(S,G)=limUHˇet1(U,G), where the limit is taken over all covering U={Ui→S}. Write Uij for Ui×SUj. Let F1,F2∈Tors(S,G). Then for some covering U={Ui→S},ζ(F1)=(gij)(i,j)∈I×I and ζ(F2)=(hij)(i,j)∈I×I, where gij,hij∈G(Uij). Note that (gij)(i,j)∈I×I and (hij)(i,j)∈I×I satisfy the following properties: gi,j.xi∣Uij=xj∣Uij and hi,j.yi∣Uij=yj∣Uij, where xi∈F1(Ui),yi∈F2(Ui).
Recall that (F1ΠGF2)(Ui) consists of elements of the form [xi,yi] with xi∈F1(Ui),yi∈F2(Ui). We have
[TABLE]
This implies that (gijhij)(i,j)∈I×I defines an element of Hˇet1(U,G). Therefore ζ(F1ΠGF2)=(gijhij)(i,j)∈I×I=ζ(F1)ζ(F2) and hence ζ is a group homomorphism. ∎
Remark 4.10**.**
If G is a sheaf of abelian groups then the map ζ is an isomorphism (see Proposition 11.1 of [13]).
Lemma 4.11**.**
Let Picetf be the étale sheafification of the presheaf Picf on Set. Then
[TABLE]
Proof.
For each étale U→S, there is a natural map (using (4.3))
[TABLE]
Here X~ stands for X×SU. Then by the universal property of sheafification, we get a unique map (f∗OX×/OS×)et→Picetf of sheaves. Since Pic vanishes for a strictly hensel local ring, ((f∗OX×/OS×)et)sˉ≅(Picetf)sˉ for all geometric point sˉ of S by the sequence (4.3). Therefore, (f∗OX×/OS×)et≅Picetf. ∎
Theorem 4.12**.**
Let f:X→S be a faithful affine map of noetherian schemes.
Then there is a natural group homomorphism θ:Br(f)→Br′(f):=Het1(S,f∗OX×/OS×).
Proof.
Since S is noetherian, we can write S=⊔i=1nSi, where Si’s are connected components of S. Then X=⊔i=1nX×SSi and f=⊔i=1nfi, where fi:X×SSi→Si. For each ηi:fi→f, we get an induced map γi:Br(f)→Br(fi). Therefore, γ=(γi) defines a map Br(f)→Br(f1)×⋯×Br(fn). Moreover, the cohomology Hi commute with the finite disjoint union. So, we may assume that S is connected. Now, we note that the following two well-known facts.
(1)
Given a presheaf F of abelian groups on Set and the sheafification F+, there is always a natural map
[TABLE]
2. (2)
For all abelian sheaf F, the natural map
[TABLE]
is an isomorphism.
The above two facts together with Lemmas 4.9 and 4.11 allow us to define the desired map θ as the composition of the following maps (see also (4.9))
[TABLE]
∎
Remark 4.13**.**
The map θ is not injective in general (see Example 3.3).
5. The Relative Brauer group of subintegral extensions
In this section, we study the relative Brauer group Br(f) in the case when f is a subintegral extension. Some details pertaining to subintegral extensions can be found in [21].
Theorem 5.1**.**
Let f:A↪B be a subintegral extension of noetherian Q-algebras. Then the following are true:
(1)
the natural map f∗:Br(A)→Br(B) is an isomorphism.
2. (2)
Br(f)=0.
Proof.
(1) Since f is subintegral, B=∪λBλ where each Bλ can be obtained by finite succession of elementary subintegral extension (i.e. A⊂A[b] such that b2,b3∈A) (see [21]). In other words, f=∪λfλ where fλ:A↪Bλ is a finite map. For each λ, we have the following two exact sequences
[TABLE]
As Br commutes with filtered limit of rings, Br(B)≅limλBr(Bλ). Thus we get ker(f∗)≅limλker(fλ∗) and coker(f∗)≅limλcoker(fλ∗). So it is enough to show that ker(fλ∗)=0=coker(fλ∗) for each λ.
By Proposition 5.4(3) of [17], Heti(Spec(A),OA×)≅Heti(Spec(Bλ),OBλ×) for all i>1 because fλ is a finite subintegral extension of Q-algebras. In particular, we get Br′(A)≅Br′(Bλ). Then the torsion subgroups Br′(A)tor and Br′(Bλ)tor are also isomorphic. Therefore, the assertion follows from the following commutative diagram
[TABLE]
where the vertical maps are isomorphisms by a theorem of Gabber (see [5]).
(2) Since f is subintegral, the map Pic(A)→Pic(B) is surjective by Proposition 7 of [11]. Hence the result by (1) and the sequence (2.1).
∎
Given an extension f:A↪B, write +f for the induced map +A↪B where +A is the subintegral closure of A in B.
Corollary 5.2**.**
Let f:A↪B be an extension of noetherian Q-algebras. Then Br(f)≅Br(+f).
Proof.
By comparing Br-Pic sequences (see (2.1)) for f and +f, we get the result. ∎
We say that a faithful affine map f:X→S is subintegral if OS(U)→f∗OX(U) is subintegral for all affine open subsets U of S (see Definition 5.1 of [17]).
Remark 5.3**.**
*(Cf. Example 6.6 of [19]) *
Proposition 5.1(2) is not true for non-affine schemes. For example, consider S=PQ1 and X=Spec(OB) where OB=OS⊕OS(−2) with OS(−2) being a square zero ideal. Then f:X→S is a subintegral map and H=H1(P1,OS(−2)) is nonzero.
We also have Pic(X)=Pic(S)⊕H. Therefore, Pic(S)→Pic(X) is not surjective. Hence Br(f)=0 by the sequence (2.1).
6. Relative Kummer’s sequence
For a commutative ring A, let μn(A):=ker[A×→nA×] and for a ring extension f:A↪B, let μn(f):=ker[Pic(f)→nPic(f)]. We can identify Pic(f) with I(f), the multiplicative group of invertible A-submodules of B by Lemma 1.2 of [19]. Some details related to I(f) can be found in section 2 of [15].
Lemma 6.1**.**
Let f:A↪B be a finite ring extension. Assume that A is a strictly hensel local ring with residue field k and characteristic of k does not divide n. Then the following sequences
[TABLE]
[TABLE]
are exact.
Proof.
Since B is finite over A,B is finite product of strictly hensel local rings. We know that for a strictly hensel local ring A, the sequence
[TABLE]
is exact.
Now the result follows from the following commutative diagram
[TABLE]
where the left two columns and the bottom two rows are short exact sequences (see (4.3)). ∎
Let f:X→S be a faithful affine map of schemes. Recall that Picetf is the étale sheafification of the presheaf U↦Pic(fU) on Set, where fU:X×SU→U. For notational convenience, we prefer to write Iet instead of Picetf≅(f∗OX×/OS×)et (see Lemma 4.11).
We write μn,X (or simply μn) for the kernel of OX,et×→nOX,et×. Similarly, μnf denotes the kernel of Iet→nIet.
Proposition 6.2**.**
Let f:X→S be a faithful finite map of schemes. Assume that the characteristic of k(s) does not divide n for any s∈S. Then the following sequences
From the Proposition 6.2, we obtain the following two long exact sequences
[TABLE]
[TABLE]
For a group G, we denote the subgroup of elements of order dividing n in G by nG.
Theorem 6.3**.**
Let f:X→S be a faithful finite map of schemes. Assume that the characteristic of k(s) does not divide n for any s∈S.
Then
(1)
Het0(S,μnf)≅nPic(f)* and the sequence*
[TABLE]
is exact.
2. (2)
if f:L↪K is a finite field extensions and characteristic of L does not divide n then the sequence
[TABLE]
is exact.
Proof.
(1) By Lemma 5.4 of [18], Het0(S,Iet)≅Pic(f). The long exact sequence (6.2) implies the assertion.
(2) Note that Pic(f)≅K×/L× by (4.3) and Het1(Spec(L),f∗OK×/OL×)≅Br(K∣L) by Lemma 3.1. Hence the result by the short exact sequence (6.3). ∎
Let Piczarf be the Zariski sheaf associated to the presheaf Picf on Szar, defined as Picf(U)=Pic(fU). A proof similar to Lemma 4.11 shows that Piczarf≅(f∗OX×/OS×)zar. We write Izar for the Zariski sheaf Piczarf≅(f∗OX×/OS×)zar.
Theorem 6.4**.**
Let f:X→S be a faithful finite map of noetherian schemes over Q.
(1)
If f is a subintegral extension of affine schemes then Heti(S,μnf)=0 for i≥0. Moreover, Heti(S,μn)≅Heti(X,μn) for i≥0.
2. (2)
If f is subintegral and S is a projective Q-scheme then Het1(S,μnf)≅nHet1(S,Iet).
Proof.
(1) For a subintegral extension f:A↪B of Q-algebras, we have Pic(f)≅B/A by Theorem 5.6 of [15] and Theorem 2.3 of [14]. This implies that Pic(f) is a Q-vector space, hence it is a divisible group. Therefore, the map Pic(f)→nPic(f) is bijective. For i>0,Heti(S,Iet)≅Hzari(S,Izar)=0 by Proposition 5.4 of [17]. Hence, Heti(S,μnf)=0 for i≥0 by (6.2).
(2) By Proposition 5.3 of [17], Pic(f)≅Hzar0(S,f∗OX/OS). Note that f∗OX/OS is coherent. So, Pic(f) is a Q-vector space. Hence the result by the short exact sequence (6.3). ∎
Bibliography21
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] B. Auslander, The Brauer Group of a Ringed Space , Journal of Algebra, 4 , (1966) 220-273.
2[2] M. Auslander and O. Goldman, The Brauer group of a commutative ring . Trans. Amer. Math. Soc. 97 , (1960) 367-409.
3[3] H. Bass and A. Roy, Lectures on topics in algebraic K-theory, Tata Institute of Fundamental Research Lectures on Mathematics, No. 41, Tata Institute of Fundamental Research, Bombay 1967.
4[4] H. Bass, Algebraic K-theory . Benjamin, Newyork, 1968.
5[5] A.J. De Jong, A result of Gabber , Preprint, 2003.
6[6] J.M Ferncindez Vilaboa, E. Villanueva Novoa and R. Gonzilez Rodriguez, The exact sequence for a monoidal functor between exact categories , Communication in Algebra, 26(9) , (1998) 2727-2743.
7[7] B. Fein and M. Schacher, Relative Brauer groups. I. J. Reine Angew. Math. 321 , (1981) 179-194.
8[8] B. Fein and M. Schacher, Kantor. William, Relative Brauer groups. II. J. Reine Angew. Math. 328 , (1981) 39-57.