In this section, if F is an abelian group and g an automorphism of F, then we write g⋅f for g(f).
1. τ-Extensions of Groups
In this subsection, we fix a group G, an abelian group F written multiplicatively, and a group homomorphism τ from G to the automorphism group Aut(F) of F. We denote the identity element of G by e and the identity element of F by 1.
Recall (I, §6, No. 1, p. 65) that an extension E of G by F is a triple (Γ,ι,π), where Γ is a group, π:Γ→G is a surjective homomorphism, and ι is an injective homomorphism from F to Γ such that Im(ι ) = Ker(π ). Let E=(Γ,ι,π) be such an extension. For every γ∈Γ, the mapping φγ:F→F defined by
ι(φγ(f))=γι(f)γ−1
for f∈F is an automorphism of F. Since F is commutative, for every f∈F, the automorphism defined by ι(f) is the identity mapping on F. By passing to the quotient, we obtain a homomorphism IntE from G to Aut(F) characterized by
γι(f)γ−1=ι(IntE(π(γ))⋅f)
for γ∈Γ and f∈F.
Aτ-extension of G by F is an extension E=(Γ,ι,π) such that IntE is equal to τ, in other words, that satisfies the relation
γι(f)γ−1=ι(τ(π(γ))⋅f)(1)
for γ∈Γ and f∈F. If E=(Γ,ι,π) and E′=(Γ′,ι′,π′) are τ-extensions, then a morphism of τ-extensions from E to E′ is a morphism of extensions from E to E′ (I, §6, No. 1, p. 65), that is, a group homomorphism u:Γ→Γ′ such that π′∘u=π and ι′=u∘ι. Note that the τ-extensions form a species of structures for which every morphism is an isomorphism (Set Theory, IV, §1, No. 5, p. 264 and I, §6, No. 1, p. 66, Proposition 1). Denote by Isoτ(E,F) the relation
“E and F are isomorphic τ-extensions.”
This is an equivalence relation; the class of the extension E is the class of objects equivalent to E for Isoτ(Set Theory, II, §6, No. 6, p. 122).
Set E0=F×G, and consider the mappings ι0:F→E0 and π0:E0→G defined by ι0(f)=(f,e) for f∈F and π0(f,g)=g for (f,g)∈F×G. Let E=(Γ,ι,π) be a τ-extension of G by F. The mapping π is surjective, so has a section σ:G→Γ such that σ(e)=e. The mapping u:(f,g)↦ι(f)σ(g) from F×G to Γ is bijective. We endow F×G with the group law obtained by transfer of structure. The triple (E0,ι0,π0) is then a τ-extension isomorphic to E. Lemma 1 now follows from Set Theory, II, §6, No. 6, p. 122.
We denote by Exτ(G,F) the set of classes of τ-extensions for the relation Isoτ.
The external semidirect product (F×τG,i,p) (I, §6, No. 1, p. 66, Proposition 3) is a τ-extension; we denote it by Iτ. We say that a τ-extension is semitrivial if it is isomorphic to the τ-extension Iτ.
For i∈{1,2}, take a group Gi, an abelian group Fi, and a group homomorphism τi from Gi to the automorphism group of Fi. We denote by τ1×τ2:G1×G2→ Aut(F1×F2) the homomorphism defined by
for g1∈G1,g2∈G2,f1∈F1, and f2∈F2. Let E1=(Γ1,ι1,π1) be a τ1-extension of G1 by F1 and E2=(Γ2,ι2,π2) a τ2-extension of G2 by F2. Then the triple (Γ1×Γ2,ι1×ι2,π1×π2) is a τ1×τ2-extension of G1×G2 by F1×F2; it is called the exterior product of the extensions E1 and E2 and denoted by E1×E2.
2. Inverse Image of a τ-Extension
Let G and G′ be groups. Let F be an abelian group, and let τ:G→ Aut(F) and u:G′→G be group homomorphisms. Consider the group homomorphism τ′=τ∘u:G′→ Aut(F), and denote by Γ′ the fiber product Γ×GG′ with respect to π and u (I, §4, No. 8, p. 46). It is the subgroup of Γ×G′ consisting of the pairs (γ,g′) such that π(γ)=u(g′). Let ι′:F→Γ′ be the group homomorphism given by ι′(α)=(ι(α),e) for α∈F, and denote the second projection by π′:Γ′→G′. Then the morphism ι′ is injective, the morphism π′ is surjective because π is, and the image of ι′ coincides with the kernel of π′. Moreover, for every α∈F and γ′=(γ,g)∈Γ′, we have the relations
Consequently, (Γ′,ι′,π′) is a τ′-extension of G′ by F; we call it the inverse image of E by u and denote it by u∗(E). The first projection is a group homomorphism φ:Γ′→Γ that we call canonical.
Let F1′ be an abelian group, and let w:F1′→ F and τ1:G′→ Aut(F1′) be group homomorphisms such that
w(τ1(g)(f))=τ(u(g))(w(f))
for every f∈F1′ and g∈G′. Let E1′=(Γ1′,ι1′,π1′) be a τ1-extension of G′ by F1′ and φ1:Γ1′→Γ a group homomorphism such that the following diagram commutes:
F1′ι1′ // Γ1′π1′ // G′
wφ1u
F ι // Γ π // G.
Then there exists a unique group homomorphism ψ:Γ1′→Γ′ such that the diagram
F1′ι1′ // Γ1′π1′ // G′
wψ
F ι′ // Γ′π′ // G′
commutes and φ1=φ∘ψ.
If the group homomorphism ψ:Γ1′→Γ′ has the desired properties, then it satisfies the relations
ψ(γ)=(φ∘ψ(γ),π′∘ψ(γ))=(φ1(γ),π1′(γ))
for every γ∈Γ1′. Conversely, the group homomorphism from Γ1′ to Γ×G′ defined by γ↦(φ1(γ),π1′(γ)) has values in the fiber product Γ×GG′ because π∘φ1(γ)=u∘π1′(γ) for every γ∈Γ1′.
Let E1 and E2 be τ-extensions of G by F, and let ψ be a morphism of τ-extensions from E1 to E2. Denote by φ1(resp. φ2) the canonical homomorphism for u∗(E1)(resp. u∗(E2)). Then there exists a unique morphism of τ′-extensions from u∗(E1) to u∗(E2), denoted by u∗(ψ), such that φ2∘u∗(ψ)=ψ∘φ1.
It suffices to apply Proposition 1 to ψ∘φ1.
The class of the τ′-extension u∗(E) therefore depends only on the class of E. We also denote by u∗: Exτ(G,F)→ Exτ′(G′,F) the mapping that sends the class of a τ-extension E to the class of the τ′-extension u∗(E).
Let u′:G′′→G′ be a group homomorphism, and let E be a τ-extension of G by F. Set τ′′=τ′∘u′, and denote by φ(resp. φ′,φ′′) the canonical homomorphism associated with the τ′-extension u∗(E)(resp. the τ′′-extension u′∗(u∗(E)), the τ′′-extension (u∘u′)∗(E)). Then there exists a unique morphism ψ from the τ′′-extension u′∗(u∗(E)) to the τ′′-extension (u∘u′)∗(E) such that φ′′∘ψ=φ∘φ′.
Let H be a subgroup of G and j:H→G the canonical injection. Then for every τ-extension E=(Γ,ι,π), the τ∘j-extension j∗(E) is isomorphic to (π−1(H),ι′,π′), where ι′:F→π−1(H) (resp. π′:π−1(H)→H) is the group homomorphism f↦ι(f) (resp. γ↦π(γ)). More generally, if the group homomorphism u:G′→G is injective, then the canonical homomorphism φ is injective with image π−1(u(G′)).
3. Direct Image of a τ-Extension
Let G be a group, let F and F′ be abelian groups, let τ (resp. τ′) be a group homomorphism from G to the automorphism group of F (resp. F′), and let v:F→F′ be a group homomorphism such that
v(τ(g)⋅f)=τ′(g)⋅v(f)(3)
for every g∈G and f∈F. Let E=(Γ,ι,π) be a τ-extension of G by F. Let Γ be the external semidirect product F′×τ′∘πΓ. We denote by :F′→Γ the homomorphism f↦(f,e) and by p:Γ→Γ the first projection. Let j:F→Γ be the mapping defined by f↦(v(f),ι(f)−1). Since the image of ι is contained in the kernel of τ′∘π, the mapping j is a group homomorphism; it is injective because ι is injective. We have the relations
(f′,γ)j(f)(f′,γ)−1=(f′,γ)(v(f),ι(f)−1)(f′,γ)−1
=(τ′(π(γ))⋅v(f),ι(τ(π(γ))⋅f)−1)(4)
=j(τ(π(γ))⋅f)
for f∈F,γ∈Γ, and f′∈F′. Consequently, the image of j is a normal subgroup of Γ. We denote by Γ′ the quotient of Γ by the image of j. We denote by ι′ the composition of the canonical surjection from Γ to Γ′ and . The kernel of the homomorphism π∘p is the product F′×ι(F), which contains the image of j. We define π′:Γ′→G as the group homomorphism deduced from π∘p by passing to the quotient. Since ι is injective, the intersection of j(F) and (F′) is reduced to the identity element of Γ; it follows that ι′ is injective. The mapping from F′×F to F′×ι(F)= Ker(π∘p) given by
(f′,f)⟼(f′v(f),ι(f)−1)
is a group isomorphism. The image of ι′ therefore coincides with the kernel of π′. Since π and p are surjective, so is π′. This proves that E′=(Γ′,ι,π′) is a τ′-extension of G by F′; we call it the direct image of E by v and denote it by v∗(E). The composition of the canonical surjection from Γ to Γ′ and the group homomorphism from Γ to Γ given by γ↦(1,γ) is a group homomorphism φ:Γ→Γ′ that we call canonical.
Let G1′ be a group, and let w:G→G1′ and τ1:G1′→ Aut(F′) be group homomorphisms such that τ′=τ1∘w. Let E1′=(Γ1′,ι1′,π1′) be a τ1-extension of G1′ by F′, and let φ1:Γ→Γ1′ be a group homomorphism such that the following diagram commutes:
F ι // Γ π // G
vφ1w
F′ι1′ // Γ1′π1′ // G1′.
Then there exists a unique group homomorphism ψ:Γ′→Γ1′ such that the diagram
F′ι′ // Γ′π′ // G
ψw
F′ι1′ // Γ1′π1′ // G1′ commutes and φ1=ψ∘φ.
For any (f,γ)∈F′×Γ, denote the class of (f,γ) in Γ′ by (f,γ). If the group homomorphism ψ:Γ′→Γ1′ has the desired properties, then it satisfies the relations
ψ(f′,γ)=ψ(ι′(f′)φ(γ))=ι1′(f′)φ1(γ)
for every f′∈F′ and γ∈Γ. Conversely, the mapping ψ from F′×τ′∘πΓ to Γ1′ given by (f,γ)↦ι1′(f)φ1(γ) is a group homomorphism. Indeed, we have the relations
for f,f′∈F′ and γ,γ′∈Γ. The kernel of ψ contains the image of j because ι1′(v(f))=φ1(ι(f)) for f∈F, and the morphism ψ deduced from ψ by passing to the quotient has the desired properties.
Denote by Σ the species of structures of τ′-extensions, and define α-mappings to be mappings from Γ to a group Γ′ underlying a τ′-extension that are group homomorphisms and that make the following diagram commute:
F ι // Γ π // G
vφ(6)
F′ι′ // Γ′π′ // G .
Proposition 2 expresses that v∗(E) is a solution of the corresponding universal mapping problem (Set Theory, IV, §3, No. 1, p. 284).
Let E1 and E2 be τ-extensions of G by F, and let ψ be a morphism of τ-extensions from E1 to E2. Denote by φ1(resp. φ2) the canonical homomorphism for v∗(E1)(resp. v∗(E2)). Then there exists a unique morphism of τ′-extensions from v∗(E1) to v∗(E2), denoted by v∗(ψ), such that φ2∘ψ=v∗(ψ)∘φ1.
It suffices to apply Proposition 2 to φ2∘ψ.
The class of the τ′-extension v∗(E) therefore depends only on the class of E. We also denote by v∗: Exτ(G,F)→ Exτ′(G,F′) the mapping that sends the class of a τ-extension E to the class of the τ′-extension v∗(E).
We keep the notation of the proposition. Let F′′ be an abelian group, and let τ′′:G→ Aut(F′′) and v′:F′→F′′ be group homomorphisms such that
τ′′(g)⋅v′(f)=v′(τ′(g)⋅f)
for g∈G and f∈F′. Let E be a τ-extension of G by F, and denote by φ(resp. φ′,φ′′) the canonical homomorphism associated with v∗(E)(resp. v∗′(v∗(E)), (v′∘v)∗(E)). Then there exists a unique morphism ψ from the τ′′-extension v∗′(v∗(E)) of G by F′′ to the τ′′-extension (v′∘v)∗(E) such that φ′′=ψ∘φ′∘φ.
Let j:{1}→F be the canonical injection. The semitrivial extension Iτ is isomorphic to j∗((G,i, IdG)), where i:{e}→G is the canonical injection. Let c:F→F be the constant homomorphism f↦1, and let E be a τ-extension. The τ-extension c∗(E) is also isomorphic to Iτ.
Let E be a subgroup of F stable under the action of G. Denote by F′ the quotient of F by E, by v:F→F′ the canonical homomorphism, and by τ′:G→ Aut(F′) the action of G on F′ characterized by
τ′(g)⋅v(f)=v(τ(g)⋅f)
for g∈G and f∈F. Let E=(Γ,ι,π) be a τ-extension of G by F. Then ι(E) is a normal subgroup of Γ, and the τ′-extension v∗(E) of G by F′ is isomorphic to the extension (Γ/ι(E),ι,π), where ι and π are the group homomorphisms deduced from ι and π by passing to the quotients. By this isomorphism, the canonical homomorphism φ associated with v∗(E) corresponds to the canonical homomorphism from Γ to Γ/ι(E).
Let G and G′ be groups, and let F and F′ be abelian groups. Let τ:G→ Aut(F),τ′:G→ Aut(F′),u:G′→G, and v:F→F′ be group homomorphisms such that
τ′(g)⋅v(f)=v(τ(g)⋅f)
for g∈G and f∈F. We write τ′′=τ′∘u. Let E be a τ-extension of G by F. We denote by φu(resp. φv,φu′,φv′) the canonical homomorphism corresponding to the τ∘u-extension u∗(E)(resp. to the τ′-extension v∗(E), to the τ′′-extensions u∗(v∗(E)) and v∗(u∗(E))). Then there exists a unique morphism ψ of τ′′-extensions from v∗(u∗(E)) to u∗(v∗(E)) such that φv∘φu=φu′∘ψ∘φv′.
We denote the τ∘u-extension u∗(E) (resp. the τ′′-extension u∗(v∗(E))) by (Γu,ιu,πu) (resp. (Γu′,ιu′,πu′)). Applying Lemma 2 of VIII, p. 287 to φv∘φu, we find that there exists a group homomorphism ψ1:Γu→Γu′ such that the diagram
F ιu // Γuπu // G′
vψ1
F′ιu′ // Γu′πu′ // G′
commutes and φu′∘ψ1=φv∘φu. The existence of ψ follows from Proposition 2 applied to ψ1. Conversely, if ψ′ also has the desired properties, then we have ψ′∘φv′=ψ1 by Lemma 2, so ψ′=ψ (Proposition 2).
4. Group Law on the Classes of τ-Extensions
Let G be a group, F be an abelian group, and τ:G→ Aut(F) be a group homomorphism. We denote by δ:G→G×G the diagonal mapping g↦(g,g) and by m:F×F→F the multiplication homomorphism (f1,f2)↦f1f2. We denote by s:F→F the group homomorphism given by f↦f−1. Let E1=(Γ1,ι1,π1) and E2=(Γ2,ι2,π2) be τ-extensions of G by F. Since we have the relation
m(((τ×τ)∘δ)(g)⋅(f1,f2))=τ(g)⋅m(f1,f2)
for every g∈G and all f1,f2∈F, the extension m∗(δ∗(E1×E2)) is a τ-extension; we call it the product of the τ-extensions E1 and E2 and denote it by E1E2. The class of this extension depends only on the classes of the extensions E1 and E2 (VIII, p. 288, Corollary 1 and VIII, p. 291, Corollary 1). So this gives a law of composition on Exτ(G,F).
Let E1=(Γ1,ι1,π1) and E2=(Γ2,ι2,π2) be τ-extensions of G by F. Let E1E2=(Γ,ι,π) be the product of these extensions. Following the example of VIII, p. 289, the construction gives a surjective group homomorphism from the fiber product Γ1×GΓ2 to Γ whose kernel is the image of the group homomorphism f↦(ι1(f),ι2(f)−1) from F to Γ1×GΓ2.
The product of τ-extensions endows the set Exτ(G,F) with the structure of an abelian group. Its identity element is the class of the semitrivial extension Iτ. The inverse of the class of a τ-extension E is the class of s∗(E).
The associativity of the law follows from the commutativity of the diagrams
G↓δG×Gδδ×IdG×G↓Id×δG×G×G
and
F×F×F↓Id×mF×Fm×IdmF×F↓mF
and from Corollary 2 of VIII, p. 288; Corollary 2 of VIII, p. 291; and Proposition 3 of VIII, p. 292.
Let Δ:F→F×F be the diagonal mapping f↦(f,f). Let E=(Γ,ι,π) be a τ-extension. Let Δ~:Γ→Γ×GΓ be the group homomorphism given by γ↦(γ,γ). The following diagram commutes:
F↓ΔF×Fι⟶Γ↓Δ~Γ×GΓπ⟶G∥G.
By Proposition 2 of VIII, p. 290, it follows that the (τ×τ)∘δ-extension δ∗(E×E) is isomorphic to Δ∗(E).
Denote by c:F→F the constant homomorphism f↦1. by Example 1 of VIII, p. 292, the fact that Iτ is an identity element for this law of composition follows from the isomorphism from δ∗(E×E) to Δ∗(E) and the commutative diagram
F(c×IdF)∘ΔF×F(IdF×c)∘ΔmF×F↓IdF↓mF.
The last assertion follows from the commutative diagram
F(s×IdF)∘ΔF×F(IdF×s)∘ΔmF×F↓c↓mF.
Let E1=(Γ1,ι1,π1) and E2=(Γ2,ι2,π2) be τ-extensions. The group isomorphism Γ1×Γ2→Γ2×Γ1 given by (γ1,γ2)↦(γ2,γ1) restricts to a group isomorphism σ:Γ1×GΓ2→Γ2×GΓ1. Because of the relations
σ(ι1(f),ι2(f)−1)=(ι2(f−1),ι1(f−1)−1)
for f∈F, the group homomorphism σ induces, by passing to the quotients, a morphism of τ-extensions from E1E2 to E2E1. So the law of composition is commutative.
a) Let G and G′ be groups, and let F be an abelian group. Let τ:G→ Aut(F) and u:G′→G be group homomorphisms. The mapping u∗: Exτ(G,F)→ Exτ∘u(G′,F) is a group homomorphism.
b) Let G be a group, and let F and F′ be abelian groups. Let τ:G→ Aut(F),τ′:G→ Aut(F′), and v:F→F′ be group homomorphisms such that
τ′(g)⋅v(f)=v(τ(g)⋅f)
for g∈G and f∈F. The mapping v∗: Exτ(G,F)→ Exτ′(G,F′) is a group homomorphism.
This follows from the commutativity of the diagrams
G′δ // G′×G′F×Fm // F
uu×u and v×vv
G δ // G×GF′×F′m // F′.
5. Cohomological Description
Let G be a group, let F be an abelian group, and let τ:G→ Aut(F) be a group homomorphism. For any g∈G and f∈F, we also write gf for τ(g)⋅f. A 2-cocycle of G with values in F is a mapping c:G×G→F such that for every (g1,g2,g3)∈G×G×G, we have
Since F is commutative, the set of 2-cocycles is a subgroup of the group of mappings from G×G to F; we denote it by Z2(G,F). We denote the group of mappings from G to F by C1(G,F). For every h∈C1(G,F) and (g1,g2)∈G×G, set
(∂h)(g1,g2)=g1h(g2)h(g1g2)−1h(g1)(8)
A simple calculation shows that the mapping ∂h:G×G→F is a 2-cocycle. The mapping ∂:C1(G,F)→Z2(G,F) is a group homomorphism. For any h∈C1(G,F), the 2-cocycle ∂h is called a 2-coboundary. The group Z2(G,F)/∂(C1(G,F)) is denoted by H2(G,F) and is called the second cohomology group of G with coefficients in F.
∗The notation above agrees with that of X, §6, no8, p. 112 concerning group cohomology.∗
Let E=(Γ,ι,π) be a τ-extension. Let σ be a section of the surjective mapping π(Set Theory, II, §3, No. 8, p. 86), that is, a mapping from G to Γ such that π(σ(g))=g for every g in G. For every g1,g2∈G, σ(g1)σ(g2)σ(g1g2)−1 belongs to Ker(π ), so that there exists a unique mapping cσ:G×G→F such that
ι(cσ(g1,g2))=σ(g1)σ(g2)σ(g1g2)−1(9)
for all g1,g2∈G. The mapping cσ is constant with value 1 if and only if σ is a group homomorphism.
The mapping cσ is an element of the group Z2(G,F), and its class in the cohomology group H2(G,F) depends only on the class of the τ-extension E.
We say that the mapping cσ is the 2-cocycle associated with σ and that its cohomology class in H2(G,F) is the cohomology class of the τ-extension E.
Let us first verify that cσ satisfies the cocycle condition. Let g1,g2, and g3 be elements of G. Using formula (1) of VIII, p. 285 and (9), we obtain the relations
The mapping cσ is therefore an element of Z2(G,F).
Let us now prove that the image of cσ in H2(G,F) is independent of the chosen section σ. Let σ and σ′ be such sections. For every element g of G, there exists a unique element ag of F such that σ′(g)=ι(ag)σ(g). Let g1 and g2 be elements of G. Using the definition (9), we obtain the equalities
as follows from (8). This proves that the classes of cσ′ and cσ in H2(G,F) coincide.
Finally, let E=(Γ,ι,π) and E′=(Γ′,ι′,π′) be isomorphic τ-extensions, let u:E→E′ be a morphism of τ-extensions, and let σ be a section of the mapping π. The mapping u∘σ is a section of the mapping π′, and we have cσ=cu∘σ. The class of cσ in H2(G,F) therefore depends only on the class of E in Exτ(G,F).
We denote by Θτ : Exτ(G,F)→H2(G,F), or more simply Θ, the mapping that sends the class of a τ-extension (Γ,ι,π) to the class of the 2-cocycle cσ, where σ is a section of the surjection π.
The mapping Θ is a group isomorphism from Exτ(G,F) to H2(G,F).
Let us first prove that Θ is a group homomorphism. Let E=(Γ,ι,π) and E′=(Γ′,ι′,π′) be τ-extensions, and let σ (resp. σ′) be a section of the mapping π (resp. π′). We denote by EE′=(Γ′′,ι′′,π′′) the product of the τ-extensions E and E′. We denote by [γ,γ′] the image in Γ′′ of an element (γ,γ′) of Γ×GΓ′ by the surjective homomorphism from the remark of VIII, p. 293. The mapping from G to Γ′′ that sends an element g to [σ(g),σ′(g)] is a section σ′′ of the mapping π′′. Let g1 and g2 be elements of G. We have the relations
ι′′(cσ′′(g1,g2))=σ′′(g1)σ′′(g2)σ′′(g1g2)−1
=[ι(cσ(g1,g2)),ι′(cσ′(g1,g2)]
=ι′′(cσ(g1,g2)cσ′(g1,g2))
We therefore have cσ′′(g1,g2)=cσ(g1,g2)cσ′(g1,g2).
Let us prove that the mapping Θ is injective. Let E=(Γ,ι,π) be a τ-extension, and let σ be a section of the mapping π such that the image of cσ in H2(G,F) is trivial. There exists a mapping a:G→F such that
cσ(g1,g2)=(g1a(g2))a(g1g2)−1a(g1)
for all g1,g2∈G. We then define σ′:G→Γ by
σ′(g)=ι(a(g)−1)σ(g)
for g∈G. Using (11), we obtain that cσ′ is constant with value 1; consequently, σ′ is a group homomorphism, which proves that the τ-extension E is semitrivial (I, §6, No. 1, p. 67, Proposition 4).
Let us prove that the mapping Θ is surjective. Let c be an element of Z2(G,F). We endow the set Γ=F×G with the following law of composition: (12) (f1,g1)(f2,g2)=(f1(g1f2)c(g1,g2),g1g2)
for all f1,f2∈F and g1,g2∈G. We have the relations
for all f1,f2,f3∈F and g1,g2,g3∈G. It follows that this law is associative because c is a 2-cocycle. For every (f,g)∈Γ, we have
(f,g)(c(e,e)−1,e)=(f(gc(e,e)−1)c(g,e),g)
Now, it follows from the definition of a 2-cocycle that c(g,e)=gc(e,e), from which we deduce that (f,g)(c(e,e)−1,e)=(f,g). We prove likewise that
(c(e,e)−1,e)(f,g)=(fc(e,e)−1c(e,g),g)=(f,g)
The law of composition defined by (12) therefore has identity element (c(e,e)−1,e), and every element (f,g) of Γ is invertible, with inverse
(g−1(f−1c(e,e)−1c(g,g−1)−1),g−1)
So we have endowed Γ with a group structure. Let us denote by ι:F→G the mapping that sends f to (c(e,e)−1f,e), by π:Γ→G the second projection, and by σ the mapping g↦(1,g). The mappings ι and π are group homomorphisms, so the triple E=(Γ,ι,π) is a τ-extension, σ is a section of the mapping π, and the associated cocycle cσ is equal to c because
Let G be a group, let F and F′ be abelian groups, let τ (resp. τ′) be a group homomorphism from G to the automorphism group of F (resp. F′), and let v:F→F′ be a group morphism such that
v(τ(g)⋅f)=τ′(g)⋅v(f)(13)
for g∈G and f∈F. Let α be an element of Exτ(G,F). If the cocycle c represents Θ(α), then v∘c represents Θ(v∗(α))∈H2(G,F′).
6. Restriction and Corestriction
Let G and G′ be groups, F be an abelian group, and τ be a homomorphism from G to the automorphism group of F. Let u:G′→G be a group homomorphism. We write τ′=τ∘u.
If ψ:G×G→F is a 2-cocycle of G with values in F, then the mapping ψ∘(u×u) from G′×G′ to F given by (g1,g2)↦ψ(u(g1),u(g2)) is a 2-cocycle of G′ with values in F, and the mapping ψ↦ψ∘(u×u) from Z2(G,F) to Z2(G′,F) induces a group homomorphism u∗:H2(G,F)→H2(G′,F). For any λ∈H2(G,F), the element u∗(λ) is called the inverse image of λ by u. When G′ is a subgroup of G and u:G′→G is the canonical injection, the homomorphism u∗ is called the restriction homomorphism from G to G′; we denote it by ResG′G. When G is a quotient of G′ and u:G′→G is the canonical surjection, the homomorphism u∗ is called the inflation homomorphism from G to G′.
In the notation from above, the following diagram commutes:
Exτ(G,F)u∗ // Exτ(G′,F)
ΘτΘτ′
H2(G,F)u∗ // H2(G′,F).
Let H be a subgroup of G of finite index. Let s be a section of the canonical surjection from G to H\G. We denote by (g,x)↦x⋅g the right action of G on H\G induced by the right action of G on itself by multiplication. Note that for every x∈H\G and g∈G, the element s(x)gs(x⋅g)−1 belongs to H. For any mapping c:H×H→F, we can therefore define a mapping cs:G×G→F by the relation
for all g1,g2∈G. The first equality comes from the cocycle relation (VIII, p. 295, formula (7)) applied to the elements
h(x)−1,h(x)s(x)g1s′(x⋅g1)−1, and s′(x⋅g1)g2s′(x⋅g1g2)−1;
the second is obtained by applying the cocycle relation to the elements
s(x)g1s(x⋅g1)−1,h(x⋅g1)−1, and h(x⋅g1)s(x⋅g1)g2s′(x⋅g1g2)−1; and the last simply uses the fact that the mapping x↦x⋅g1 is a permutation of H\G.
The last two lines of the obtained expression correspond to a 2-coboundary. We find that cs′ has the same class in H2(G,F) as the cocycle whose value in (g1,g2)∈G2 is given by the expression
where the first equality follows from the cocycle relation applied to the elements
s(x)g1s(x⋅g1)−1,s(x⋅g1)g2s(x⋅g1g2)−1, and h(x⋅g1g2)−1.
The last two lines of the obtained expression correspond to a 2-coboundary. We find that the class of cs′ coincides with that of cs.
We have constructed a homomorphism from the group H2(H,F) to the group H2(G,F), which we call the corestriction homomorphism from H to G and denote by CorHG.
Let H be a subgroup of G of finite index. The endomorphism CorHG∘ ResHG of the group H2(G,F) coincides with multiplication by the index (G : H).
Let α be an element of H2(G,F), and let c be an element of Z2(G,F) representing α. The element CorHG∘ ResHG(α) is the class of the cocycle whose value in (g1,g2)∈G2 is given by the expression
Let K be a commutative field. If E is a K-algebra, then we denote its automorphism group by AutK(E). If E is a Galois extension of the field K, then the group AutK(E) is simply the Galois group Gal(E/K) (V, §10, No. 2, p. 58).
Let G be a group. A(K,G)-algebra is a K-algebra E endowed with a group homomorphism λ:G→ AutK(E). The homomorphism λ then endows E with the structure of a group with operators in G as well as the structure of a left K[G]-module with external law given by
((g∈G∑μgg)x=g∈G∑μgλ(g).x(14)
for every x∈L and every element (μg)g∈G of K[G]. A morphism of (K,G)-algebras is a morphism of algebras that is also a morphism of groups with operators.
For every family (Ei)i∈I of (K, G)-algebras, the product K-algebra ∏i∈IEi endowed with its structure of a product group with operators is a (K, G)-algebra.
If E is a (K, G)-algebra, then the set EG of elements of E invariant under G is a subalgebra of E.
Let E be a (K, G)-algebra, where G acts by λ:G→ AutK(E). If K′ is an extension of K, then for any g∈G, let λ′(g) be the automorphism IdK′⊗λ(g) of the K′-algebra L(K′). Then λ′:g↦λ′(g) endows L(K′) with the structure of a (K′, G)-algebra.
Given a group H and H-sets X and Y, we denote by FH(X,Y) the set of morphisms of H-sets from X to Y. It is the set of mappings f:X→Y such that f(hx)=hf(x) for every h∈H and x∈X.
Let G be a group with identity element e, let H be a subgroup of G, and let E be a (K, H)-algebra. The (K,G)-algebra deduced from E by coinduction from H to G, denoted by CoindHG(E), is the K-algebra FH(G,E) endowed with the action of G given by the homomorphism λ from G to AutK(CoindHG(E)) defined by
a) Let S be a subset of G such that every element of G can be written uniquely as hs with h∈H and s∈S. Then the mapping that sends f to (f(s))s∈S is an isomorphism from the K-algebra CoindHG(E) to the K-algebra F(S,E) of mappings from S to E.
b) The algebra CoindHG(E) has finite degree over K if and only if E has finite degree over K and the index of H in G is finite. In this case, we have the formula
[CoindHG(E):K]=(G:H)[E:K].
c) Let EH be the algebra of invariants of the group H in E and CoindHG(E)G that of the group G in CoindHG(E). The mapping f↦f(e) from CoindHG(E) to E restricts to an algebra isomorphism from CoindHG(E)G to EH.
Assertion a) follows from the definitions and implies b). By (15), a mapping from G to E is an element of CoindHG(E) invariant under G if and only if it is constant with value an element of E invariant under H.
Let G be a group, H a subgroup of G, and N a subgroup of H. Let E be a (K, N)-algebra. Let α be an element of FH(G,FN(H,E)). We have the relation
α(g)(nh)=n(α(g)(h))(16)
for every g∈G,h∈H, and n∈N and the relations
α(hg)(h′)=(hα(g))(h′)=α(g)(h′h)(17)
for all h,h′∈H and every g∈G. Consequently, α(ng)(e)=α(g)(n)=n(α(g)(e)) for every g∈G and n∈N. We can therefore consider the mapping
ψ:FH(G,FN(H,E))→FN(G,E)
defined by the relation ψ(α)(g)=α(g)(e) for α in FH(G,FN(H,E)) and g in G. The mapping ψ is an algebra isomorphism from CoindHG(CoindNH(E)) to CoindNG(E) whose inverse sends an element β from FN(G,E) to the mapping α defined by the relation α(g)(h)=β(hg) for g∈G and h∈H.
We now suppose given a finite group G and a reduced (V, §6, No. 6, p. 32) commutative K-algebra L of finite degree endowed with an action of G given by a homomorphism λ from G to AutK(L). For x∈L and g∈G, we denote by g⋅x the transform of x under the automorphism λ(g) of L. Let S be the set of maximal ideals of L; we denote by g⋅m the transform of an element m of S under the automorphism λ(g) of L. It is an element of S. For every m in S, the field L/m is a finite extension of K. We write πm:L→L/m for the projection and denote by Gm the stabilizer of m in G, that is, the set of g∈G such that g⋅m=m. The K-algebra L/m is endowed with an action of Gm through the homomorphism λm from Gm to AutK(L/m) that sends an element h of Gm to the automorphism of L/m deduced from λ(h) when passing to the quotients.
Let O be the set of orbits of G in S. Given an orbit σ∈O, set aσ=⋂m∈σm and Lσ=L/aσ. Since aσ is invariant under G, by passing to the quotients, the action of G on L defines a homomorphism λσ from G to AutK(Lσ). Finally, denote by πσ the canonical mapping from L to Lσ.
b) The mapping π:x↦(πσ(x))σ∈O is an isomorphism of (K,G)-algebras from L to ∏σ∈OLσ.
c) Denote by LG(resp. LσG) the subalgebra of L (resp. Lσ) of elements invariant under the action of G. Then π induces an isomorphism from LG to ∏σ∈OLσG.
Since the algebra L is reduced and of finite degree, the intersection of the maximal ideals of L is reduced to 0 (VIII, p. 173, Corollary 2). Moreover, if m and m′ are two distinct maximal ideals of L, then we have m+m′=L. By Proposition 10 of I, §8, No. 11, p. 110, the canonical mapping from L to ∏m∈SL/m is an isomorphism, as is the canonical mapping from L/aσ to ∏m∈σL/m for every σ∈O. Assertion a) follows. Since O is a partition of S, assertion b) follows; assertion c) is an immediate consequence of b).
Let us now fix an orbit σ∈O and an element m of σ. Set Fm = CoindGmG(L/m), and denote by λFm the action of G on Fm. For any x∈L, denote by x the mapping from G to L/m that sends g to πm(gx). By formula (15) of VIII, p. 305 and the definition of the action of Gm on L/m, it is immediate that x belongs to Fm, and the mapping u:x↦x from L to Fm satisfies λFm(g)∘u=u∘λ(g) for g∈G. In other words, u is a morphism of (K, G)-algebras.
The morphism u is surjective with kernel aσ. The mapping deduced from u by passing to the quotient is an isomorphism of (K,G)-algebras from Lσ to Fm.
Since the kernel of πm is equal to m, that of u is equal to ⋂g∈Gg−1.m=aσ. To prove that u is surjective, it suffices to prove that the vector spaces Lσ=L/aσ and Fm have the same dimension over K. Now, all ideals g⋅m have the same codimension in L, and, by Lemma 8, a),
[Lσ:K]= Card(σ ).[L/m:K].
Moreover, by Lemma 7, b), we have
[Fm:K]=(G:Gm)[L/m:K]
Since we have Card(σ ) =(G:Gm), we have proved the equality [Lσ:K]=[Fm:K].
We now suppose, furthermore, that the homomorphism λ from G to AutK(L) is injective. We identify K with its image in L through the mapping ξ↦ξ⋅1. Let Ω be an algebraically closed extension of K. The set F(G,Ω) of mappings from G to Ω coincides with the coinduced (Ω, G)-algebra Coind{e}G(Ω). It is a free Ω[G]-module of rank 1. Let H be the set of K-algebra homomorphisms from L to Ω. We define a right action of G on H by (g,χ)↦χ∘λ(g). The Ω-algebra F(H,Ω) is then endowed with the structure of an (Ω, G)-algebra deduced from the right action of G on H.
Let L be a (K,G)-algebra that is étale over K. The mapping ψ from L(Ω) to F(H,Ω) characterized by the relation
ψ(ξ⊗x)=(ξχ(x))χ∈H
is an isomorphism of (K,G)-algebras.
Since L is étale, the mapping ψ is an isomorphism of Ω-algebras (V, §6, No. 3, p. 30, Proposition 2 and V, §6, No. 3, p. 29, Proposition 1, c)). We have the relations
ψ((Id ⊗λ(g))(ξ⊗x)) =(ξ(χ∘λ(g))(x))χ∈H
for ξ∈Ω,x∈L, and g∈G. So ψ is a morphism of (Ω, G)-algebras.
Let G be a finite group, and let L be a commutative K-algebra of finite degree endowed with an action of G given by an injective homomorphism λ from G to AutK(L). The following properties are then equivalent:
(i) There exist a subgroup H of G, a Galois extension E of K of finite degree, an isomorphism from H to Gal(E/K), and an isomorphism of (K,G)-algebras from L to CoindHG(E).
(ii) The algebra L is étale, and H is a homogeneous principal G-set (I, §5, No. 6, p. 60, Definition 7).
(iii) There exists an isomorphism of (Ω,G)-algebras ψ:L(Ω)→F(G,Ω); in other words, for every g∈G, the automorphism ψ∘λ(g)(Ω)∘ψ−1 of ΩG is equal to the automorphism
(xh)h∈G⟼(xhg)h∈G
(iv) The algebra L is reduced, and L is a free K[G]-module of rank 1.
(v) The algebra L is reduced, we have Card(G) = [L : K], and K is the subring of L of elements invariant under the action of G.
(vi) The algebra L is reduced, the group G acts transitively on the set of maximal ideals of L, and, for every maximal ideal m of L, the stabilizer Gm of m in G acts faithfully on L/m and admits K as subfield of invariants.
(i)⇒(ii): Let E be a Galois extension of K of finite degree and τ an isomorphism from H to AutK(E). Let S be a system of representatives of the right cosets of G modulo H. The K-algebra F = CoindHG(E) is isomorphic to F(S,E) (Lemma 7, a)); it is therefore étale. We denote by λF the action of G on F. Furthermore, let ψ be a K-algebra homomorphism from E to Ω, and let χ0 be the homomorphism f↦ψ(f(e)) from F to Ω. Let g∈G be such that we have χ0∘λF(g)=χ0; since ψ is injective, we then have f(g)=f(e) for every f∈F. In view of Lemma 7, a), we have g∈H, and by the formula
f(h)=τ(h)⋅f(e)(18)
which holds for every h∈H, this can only happen if g=e. On the other hand, by Lemma 7, b) and Theorem 3 of V, §10, No. 6, p. 66, we have
The set K of K-homomorphisms from F to Ω has cardinal [F : K] because F is étale (V, §6, No. 5, p. 32, Proposition 4), so Card(K) = Card(G). Since the stabilizer of χ0 in G is equal to {e} by the above, K is a homogenous principal G-set.
(ii)⇒(iii): Suppose that L is étale and that H is a homogenous principal G-set. By Lemma 10, the (Ω, G)-algebras L(Ω) and F(H,Ω) are isomorphic. Since H is a homogenous principal G-set, the (Ω, G)-algebras F(H,Ω) and F(G,Ω) are isomorphic.
(iii)⇒(iv): Suppose that property (iii) holds. Then L(Ω) is a free module of rank 1 over the algebra Ω[G]; the latter can be canonically identified with K[G](Ω). We then apply Theorem 3 of VIII, p. 37.
The implication (iv)⇒(v) is immediate.
Let us prove the implication (v)⇒(vi). The algebra L is reduced. By Lemma 8, c), the group G acts transitively on the set S of maximal ideals of L. Let m be an element of S. By Lemma 9, since ⋂n∈Sn={0}, the algebra L is isomorphic to the algebra CoindGmG(L/m). The algebra of invariants of Gm in L/m therefore coincides with K by Lemma 7, c). Hence the homomorphism λm from Gm to Gal((L/m)/K) is surjective. By Lemma 7, we moreover have
Card(G) =[L:K]=(G:Gm)[L/m:K]. So Card(Gm) =[L/m:K], and the homomorphism λm is injective.
It remains to prove the implication (vi)⇒(i). Let m be a maximal ideal of L. By Lemma 9, the algebra L is isomorphic to the algebra CoindGmG(L/m) as a (K, G)-algebra. Since Gm acts faithfully in L/m and admits K as subfield of invariants, the group homomorphism λm defines an isomorphism from Gm to Gal((L/m)/K).
In Theorem 2, we can replace the assumption that Ω is algebraically closed with the assumption that Ω is separably closed. Indeed, if L is étale, then the image of every K-algebra homomorphism from L to Ω is a separable extension of K.
Let G be a finite group, and let L be a nonzero commutative algebra of finite degree over K endowed with the structure of a (K,G)-algebra, where the action of G is given by an injective homomorphism from G to the group AutK(L). We say that L is a Galois algebra with group G if it has the equivalent properties (i) through (vi) of Theorem 2.
Suppose that L is an extension of K endowed with an action λ of G. Then L is a Galois algebra over K if and only if L is a Galois extension of K and λ is an isomorphism from G to the Galois group of L over K.
Suppose that the group G is abelian. If G acts faithfully and transitively on a set X, then the stabilizer of any point of X is reduced to the identity (because the stabilizers of the points of X are all equal and their intersection is reduced to the identity element of G). Consequently, properties (i) through (vi) of Theorem 2 are also equivalent to the following:
(vii) The algebra L is étale, and G acts faithfully and transitively on H.
By V, §10, No. 1, p. 75, Lemma 5; V, §10, No. 1, p. 76, Proposition 12; and VIII, p. 137, Proposition 3, a Galois algebra is commutative and semisimple.
Let n be a strictly positive integer, prime to the characteristic exponent of K. Suppose that the group μn of n-th roots of unity in K has order n. Then for every divisor d of n, the group μd of d-th roots of unity has order d. Let a be a nonzero element of K, let L be the algebra K[X]/(Xn−a), and let x be the class of X in L. The sequence (1,x,...,xn−1) is a basis of the vector space L over the field K, and we have xn=a. Moreover, the polynomial Xn−a is prime to its derivative nXn−1, so the algebra L is étale (V, §7, No. 2, p. 37, Proposition 3). For every ζ in μn, the automorphism P(X)↦P(ζX) of the ring K[X] defines, by passing to the quotients, an endomorphism λ(ζ) of L because we have (ζX)n−a=Xn−a; it is an automorphism. We have
λ(ζ)xi=ζixi(19)
for 0⩽i<n. The mapping λ:ζ↦λ(ζ) is an injective homomorphism from μn to AutK(L), and the ring of invariants of the group λ(μn) in L is equal to K⋅1. Since the cardinal of μn is equal to n=[L:K], the algebra L endowed with the action of λ is a Galois algebra (VIII, p. 308, Theorem 2, (v)).
Let r be the least strictly positive integer such that ar belongs to K∗n; it divides n, and there exists an element b of K∗ such that a=bn/r. Then (V, §11, No. 8, p. 91, Remark) the polynomial Xr−b is irreducible, and we have Xn−a=∏ζ∈μn/r(Xr−ζb). Let E be the field K[Y]/(Yr−b), and let y be the class of Y in E. There exists an isomorphism θ from μn/r to Gal(E/K), characterized by the relation θ(ξ)(y)=ξy (V, §11, No. 8, p. 91, Example 3). We then verify that the Galois algebra L is isomorphic to the (K,μn)-algebra Coindμn/rμn(E).
Now suppose that the field K has characteristic p=0. Let c be an element of K. The polynomial f=Xp−X−c is prime to its derivative f′=−1, so the algebra L=K[X]/(f) is étale (V, §7, No. 2, p. 37, Proposition 3). We denote the image of X in L by x; we have the relation xp=x+c. The sequence (1,x,...,xp−1) is a basis of L viewed as a vector space over K.
Let P be the additive group of the prime subfield of K; it is a cyclic group of order p, generated by the unit element 1 of K. For every j in P, we have jp=j (V, §1, No. 3, p. 4, formula (4)) and therefore f(X+j)=f(X). Hence there exists an automorphism γ(j) of the algebra L characterized by the relation γ(j)(x)=x+j; moreover, the resulting mapping γ is an injective homomorphism from P to AutK(L).
Let Ω be an algebraically closed extension of K, and let ξ be a root of the polynomial f in Ω. We have ξp=ξ+c, hence
Xp−X−c=(Xp−ξp)−(X−ξ)=(X−ξ)p−(X−ξ)=j∈P∏(X−ξ−j)
by V, §12, No. 1, p. 94, formula (1). For every j in P, there exists a unique algebra homomorphism χj:L→Ω that sends x to ξ+j; moreover, every homomorphism from L to Ω is one of the χj, and we have the relation χj=χ0∘γ(j). The algebra L endowed with γ has property (ii) of Theorem 2 of VIII, p. 308 and is therefore a Galois algebra over K.
To describe the structure of L, we must distinguish between two cases:
a) We have ξ∈/K. Then the polynomial f(X) is irreducible in K[X] (V, §11, No. 9, p. 93, Example 3). In this case, L is a cyclic extension of K of degree p, and γ is an isomorphism from P to Gal(L/K).
b) We have ξ∈K. Then the mapping ψ:y↦(χj(y))j∈P is an isomorphism from the algebra L to the product algebra KP; moreover, ψ∘γ(k)∘ψ−1 is the automorphism (xj)j∈P↦(xj+k)j∈P of KP for every k∈P.
Let G be a finite group, H be a subgroup of G, and E be a Galois algebra over the field K with group H. Then the (K,G)-algebra CoindHG(E) deduced from E by coinduction from H to G is a Galois algebra over the field K with group G.
Since E is a Galois algebra over K, by property (v) of Theorem 2 of VIII, p. 308, it is reduced, we have Card(H) = [E : K], and K is the ring of invariants of H in E. But by Lemma 7 of VIII, p. 305, the algebra F = CoindHG(E) is reduced, and we have
Let G be a finite group. Let L be a Galois algebra over the field K with group G, and let K′ be an extension of K. Then the (K′,G)-algebra L(K′) is a Galois algebra over K′.
We use property (v) in Theorem 2 by observing that if the K-algebra L is étale, then the K′-algebra L(K′) is too (V, §6, No. 5, p. 32, Corollary 2), that we have the equality [L(K′):K′]=[L:K], and that the ring of invariants of G in L(K′) is (LG)(K′), where LG is the ring of invariants of G in L.
Let G1 and G2 be groups. Let L1 and L2 be Galois algebras over K with respective actions λ1:G1→ AutK(L1) and λ2:G2→ AutK(L2). Set L=L1⊗KL2,G=G1×G2, and λ(g1,g2)=λ1(g1)⊗λ2(g2) for (g1,g2)∈G. Then the K-algebra L endowed with the action λ is a Galois algebra over K.
We reason as before, taking the following into account: If L1 and L2 are étale, then so is the algebra L=L1⊗KL2 (V, §6, No. 5, p. 32, Corollary 1), and we have the equalities
[L:K]=[L1:K][L2:K] and Card(G) = Card(G1) Card(G2).
Moreover, if LiGi denotes the ring of invariants of Gi in Li, then it follows from Lemma 11 below that L1G1⊗KL2G2 is the ring of invariants of G1×G2 in L1⊗KL2.
Let G1 and G2 be groups, and let W1 and W2 be K-vector spaces. We endow W1(resp. W2) with an action of G1(resp. G2) given by a group homomorphism ρ1:G1→ AutK(W1)(resp. ρ2:G2→ AutK(W2)). We consider the group homomorphism
ρ1⊗ρ2:G1×G2⟶ AutK(W1⊗KW2)
defined by the relation (ρ1⊗ρ2)(g1,g2)(w1⊗w2)=ρ1(g1)(w1)⊗ρ2(g2)(w2) for g1∈G1,g2∈G2,w1∈W1, and w2∈W2. Then the linear mapping from W1G1⊗KW2G2 to W1⊗KW2 given by the tensor product of the canonical injections induces an isomorphism of K-vector spaces from W1G1⊗KW2G2 to (W1⊗KW2)G1×G2.
This follows from Lemma 1 of VIII, p. 213 applied to the K[G]-modules M1=M2=K endowed with the trivial action of G,N1=W1, and N2=W2.
Let L be a Galois extension of the field K of finite degree, and let G be its Galois group; we denote the identity mapping on G by λ. Then L endowed with λ is a Galois algebra over K with group G. Let K′ be an extension of K. By Proposition 10, the algebra L(K′) is Galois over K′, but it is not in general an extension of K′. Analogously, by Proposition 11, the tensor product of Galois extensions E and F of K of finite degree can be viewed as a Galois algebra; in general, it is not a Galois extension of K. It is one, however, if E and F are moreover linearly disjoint subextensions of an extension of K (V, §2, No. 5, p. 13 and V, §10, No. 1, p. 57, Proposition 1).
9. Cross Products
Let K be a field, and let G be a group whose identity element we denote by e. Let L be a commutative K-algebra, and let λ be a homomorphism from G to the automorphism group of the K-algebra L. For any g in G, let τ(g) be the automorphism of the multiplicative group L∗ of L induced by λ(g).
Let E=(Γ,ι,π) be a τ-extension of the group G by L∗. Define a right action of the group L∗ on the set L×Γ by
(β,γ).α=(βα,ι(α)−1γ)(20)
for α∈L∗,β∈L, and γ∈Γ. We denote by E the set of orbits of L∗ in L×Γ and by [β;γ] the orbit of the pair (β,γ). We therefore have, by construction, the relation
[βα;γ]=[β;ι(α)γ](21)
for α∈L∗,β∈L, and γ∈Γ.
Given β in L and γ in Γ, we denote by γβ the transform of β under the automorphism λ∘π(γ) of L; we have the relations
(22) γ(γ′β)=γγ′β,γ(β+β′)=γβ+γβ′, and γ(ββ′)=γβγβ′
for γ,γ′ in Γ and β,β′ in L. There exists a law of composition on E characterized by the relation
[β;γ][β′;γ′]=[βγβ′;γγ′](23)
Indeed, it suffices to verify that the right-hand side does not change if we replace, respectively, β,γ,β′, and γ′ with βα,ι(α)−1γ,β′α′, and ι(α′)−1γ′ for all α,α′ in L∗. This follows immediately by applying to E formula (1) of VIII, p. 285, which can also be written as γι(α)=ι(γα)γ for γ∈Γ and α∈L∗. By the formulas in (22), the set E endowed with this law is a monoid with identity element [1;e].
Since π∘ι is constant with value e, there exists a mapping π from E to G such that we have
π([β;γ])=π(γ)(24)
for β∈L and γ∈Γ.
Let g be an element of G, and let Eg=π−1(g). If γ0 is a fixed element of π−1(g), then the mapping β↦[β;γ0] is a bijection from L to Eg, thanks to which we will transfer to Eg the K-vector space structure on L. Since π−1(g) is composed of elements of the form ι(α)γ0, where α runs through L∗, and we have
[β;ι(α)γ0]=[βα;γ0]
the vector space structure on Eg does not depend on the choice of γ0.
Let g and g′ be elements of G; by formulas (22) and (23), the law of composition on E induces by restriction a K-bilinear mapping from Eg×Eg′ to Egg′. Consequently, the vector space P=⨁g∈GEg is endowed with the structure of an associative and unital algebra, whose multiplication induces the previous bilinear mapping from Eg×Eg′ to Egg′ for all g and g′ in G. The algebra P is called the cross product of L and E and is denoted by A[E;L]; its unit element is the element [1;e] of Ee.
Set
u(β)=[β;e](25)
for β in L. Then u:L→A[E;L] is an injective homomorphism of K-algebras. By (23), for every γ∈Γ, the element [1;γ] is invertible in A[E,L], and the mapping v:Γ→A[E,L]∗ that sends γ to [1;γ] is an injective group homomorphism. The homomorphisms u and v are called canonical. We have the relations
u(α)=v(ι(α))(26)
u(γβ)=v(γ)u(β)v(γ)−1(27)
[β;γ]=u(β)v(γ)(28)
for α∈L∗,β∈L, and γ∈Γ.
Conversely, we have the following universal property of the algebra A[E;L].
Let B be a K-algebra, u′:L→B a K-algebra homomorphism, and v′:Γ→B∗ a group homomorphism. We assume satisfied the relations
u′(α)=v′(ι(α))(29)
u′(γβ)=v′(γ)u′(β)v′(γ)−1(30)
for α∈L∗,β∈L, and γ∈Γ. Then there exists a unique algebra homomorphism f from A[E;L] to B such that we have u′=f∘u and v′=f∘v.
To prove the uniqueness of f, observe that the vector space A[E;L] over the field K is generated by the set of elements of the form [β;γ]=u(β)v(γ). If the homomorphism f′:A[E;L]→B satisfies f′∘u=u′ and f′∘v=v′, then it sends [β;γ] to u′(β)v′(γ) and therefore coincides with f.
By formula (29), we have
u′(βα)v′(ι(α)−1γ)=u′(β)v′(γ)(31)
for α in L∗,β in L, and γ in Γ. By the definition of E, there consequently exists a mapping f0:E→B such that f0([β;γ])=u′(β)v′(γ). By formulas (23) and (30), we have f0(xx′)=f0(x)f0(x′) for x and x′ in E. The restriction of f0 to Eg is K-linear for every element g of G; consequently, there exists a unique K-linear mapping f from A[E;L] to B that coincides with f0 on E. The mapping f is an algebra morphism that satisfies u′=f∘u and v′=f∘v.
Let σ:G→Γ be a section of the mapping π. Set εg=v(σ(g)) for every g in G, and denote by cσ the 2-cocycle associated with σ (VIII, p. 296). In particular, we have
εgεg′=u(cσ(g,g′))εgg′(32)
for all g,g′∈G. Let us, moreover, identify L with a subalgebra of A[E;L] via the homomorphism u. Then every element of A[E;L] can be written uniquely as ∑g∈Gagεg, where (ag) is a family of elements of L with finite support. The multiplication in A[E;L] can be expressed by the formula
(g∑agεg)(g∑bgεg)=g∑dgεg(33)
with
dg=hh′=g∑ah(λ(h)⋅bh′)cσ(h,h′)(34)
If the extension E is semitrivial, then we can choose a section σ of π that is a group morphism from G to Γ; the cocycle cσ is therefore constant with value 1, and formula (34) simplifies to
dg=hh′=g∑ahλ(h)⋅bh′(35)
Let K′ be an extension of the field K. Denote by L′ the K′-algebra L(K′), and for any g in G, denote by λ′(g) the automorphism λ(g)(K′) induced by λ(g) on L(K′). Moreover, let us denote by τ′(g) the automorphism of L′∗ induced by λ′(g). Finally, let h be the homomorphism from L∗ to L′∗ that sends x to x⊗1. Let E′=(Γ′,ι′,π′) be the direct image of the extension of E by h (VIII, p. 289). Let A[E′,L′] be the cross product K′-algebra of E′ and L′, and let u′:L′→A[E′,L′] and v′:Γ′→A[E′,L′]∗ be the canonical homomorphisms.
There exists a unique isomorphism of K′-algebras φ from A[E;L](K′) to A[E′;L′] that satisfies the relations
u′(h(β))=φ(1⊗u(β))(36)
for β∈L and
v′(h(γ))=φ(1⊗v(γ))(37)
for γ∈Γ.
The proof follows immediately from the constructions.
The isomorphism φ is called canonical.
10. Application to the Brauer Group
In this subsection, K is a field, and L is a Galois algebra over K with action λ:G→ AutK(L). We denote the degree of L over K by n. We have n= Card(G).
Let E=(Γ,ι,π) be a τ-extension of G by L∗. The algebra A[E;L] is central simple and of degree n2 over K. Moreover, the canonical homomorphism u from L to A[E;L] is an isomorphism from L to a maximal commutative subalgebra of A[E;L].
a) There does not exist any ideal of L, other than {0} and L that is invariant under G.
b) Let g be an element of G distinct from the identity element, and let ag be the ideal of L generated by the elements of the form x−λ(g)x, where x runs through L. We have ag=L.
Let S be the set of maximal ideals of L; for any subset S of S, set a(S)=⋂m∈Sm. Since the ring L is commutative and semisimple (VIII, p. 310, Remark 6), the map W↦a(W) is a bijection from the set of subsets of S to the set of ideals of L (VIII, p. 142, Proposition 9). The ideal a(S) is invariant under G if and only if S is. Since G acts transitively on S (VIII, p. 308, Theorem 2, vi)), the only subsets of S invariant under G are ∅ and S. Since a(∅)=L and a(S)={0}, we have proved assertion a).
Let us prove b) by contradiction; suppose that we have ag=L. Let m be a maximal ideal of L containing ag. We then have λ(g)x≡x mod m for every x in L. In particular, m is invariant under g, and λ(g) is induced by the identity on the field L/m. By property (vi) in Theorem 2 of VIII, p. 308, the element g of Gm is trivial, which contradicts the assumption of b).
Let us now prove Theorem 3. The vector space A[E;L] has finite dimension Card(G)[L:K]=n2 over K, by VIII, p. 316 and the assumed equality Card(G) =[L:K]=n.
Let a be a nonzero two-sided ideal of the algebra A[E;L]. We use the notation of VIII, p. 316. Every element a of A[E;L] can be written uniquely as a=∑g∈Gagεg with ag∈L for every g∈G; we denote by Φ(a) the set of elements g of G such that ag=0. By formula (32) of VIII, p. 316, we have the relation
εgεg′=cσ(g,g′)εgg′(38)
for all g,g′∈G and consequently
Φ(aεg)=Φ(a).g(39)
for every g∈G and a∈A[E;L].
Let a be a nonzero element of a for which Φ(a) is minimal for the inclusion; by formula (39), if necessary replacing a with an element of the form aεg−1 with g∈Φ(a), we may assume that we have e∈Φ(a). Let s be an element of Φ(a) distinct from e. By Lemma 12, b), there exists an element x of L such that as(x−λ(s)⋅x)=0. But we have the relation
xa−ax=g∑ag(x−λ(g)⋅x)εg(40)
where the sum is taken over the elements g of Φ(a) distinct from e. Because of the minimality of Φ(a), we then have xa−ax=0, but this contradicts the assumption on x. We have, therefore, proved by contradiction that Φ(a) contains only the identity element e of G, so we have a∈L.
Hence L∩a is an ideal of L not reduced to {0}. Moreover, for every x in L, we have εgxεg−1=λ(g)⋅x, so L∩a is invariant under G. By Lemma 12, we therefore have L∩a=L, that is, L⊂a. Since L contains the identity element of A[E;L], we have a=A[E;L].
Since the algebra A[E;L] has nonzero finite dimension over K and its only two-sided ideals are {0} and A[E;L], it is simple.
Let us prove that every element a of A[E;L] that commutes with L belongs to L. We write a as ∑g∈Gagεg with coefficients ag in L. For every x in L, we have xa−ax=0, and relation (40) shows that we have
ag(x−λ(g)⋅x)=0(41)
for every g∈G and x∈L. By Lemma 12, we therefore have ag=0 for g=e; consequently, a∈L.
Let us, finally, determine the center of A[E;L]. If z belongs to the center, then it commutes with L, so belongs to L. But we then have
0=εgz−zεg=(λ(g)⋅z−z)εg
for every g of G. Hence z is invariant under the automorphism group λ(G) of L. We therefore have z∈K by assumption.
Let A be a central simple algebra of finite degree over K, and let L be a maximal commutative subalgebra of A. Then there exists a τ-extension E of G by L∗ such that A is isomorphic to A[E;L].
Without loss of generality, we may assume that L is a maximal commutative subalgebra of A. Let Γ be the multiplicative group consisting of the invertible elements γ of A such that there exists a g in G with
γxγ−1=λ(g).x(42)
for every element x of L. If γ∈Γ is given, then the element g satisfying this relation is unique; we denote it by π(γ). It is immediate that π is a homomorphism from Γ to G and that its kernel is equal to L∗. The surjectivity of π follows from the Skolem–Noether theorem (VIII, p. 263, Corollary).
If we denote the canonical injection of L∗ into Γ by ι, then it follows from the constructions that E=(Γ,ι,π) is a τ-extension of G by L∗. Let
u:L→A[E;L] and v:Γ→A[E;L]
be the canonical homomorphisms. By the universal property of A[E;L] (VIII, p. 315, Proposition 12), there exists a unique algebra homomorphism f:A[E;L]→A such that f∘u(x)=x and f∘v(γ)=γ for x∈L and γ∈Γ. Since the algebra A[E;L] is simple, the homomorphism f is injective. Now, the algebra A[E;L] has degree n2 over K and so does A because L is a maximal commutative semisimple subalgebra of A and we have n=[L:K] (VIII, p. 262, Proposition 3, (ii)). Hence f is bijective.
that sends the class of a τ-extension E of G by L∗ to the class in Br(L/K) of the algebra A[E;L].
To define Ψ and verify that it is a bijection, we must establish the following points:
a) If E and E′ are isomorphic τ-extensions of G by L∗, then the algebras A[E;L] and A[E′;L] are isomorphic.
b) Conversely, if the algebras A[E;L] and A[E′;L] are isomorphic, then the τ-extensions E and E′ of G by L∗ are isomorphic.
c) In every class in Br(L/K), there is an algebra E containing L as a maximal commutative subalgebra.
d) If E is a central simple algebra of finite degree over K containing L as a maximal commutative subalgebra, then there exists a τ-extension E of G by L∗ such that E is isomorphic to A[E;L].
Assertion a) follows from the construction of the cross product; b) follows from VIII, p. 263, Corollary. Assertion c) follows from Proposition 5 of VIII, p. 281, and d) is simply Theorem 4 above.
It remains to verify that Ψ is a group homomorphism; for this, it suffices to prove that if E1=(Γ1,ι1,π1) and E2=(Γ2,ι2,π2) are τ-extensions, then the algebra A[E1E2;L] is equivalent to the algebra A[E1;L]⊗A[E2;L]. We denote the product extension E1E2 by E=(Γ,ι,π). The group Γ is isomorphic to the cokernel of the homomorphism ρ from L∗ to the fiber product Γ1×GΓ2 that sends μ to (ι1(μ),ι2(μ)−1). For i∈{1,2}, write Ai=A[Ei;L]. Let ui:L→Ai and vi:Γi→Ai∗ be the canonical homomorphisms. We identify L with its images by the homomorphisms ui, which make L into maximal commutative subalgebras of the Ai. We denote by Vi the L-vector space defined by left multiplication in Ai. The ring L⊗KAio acts on Vi. Since it is simple and has dimension n2 over L, we obtain an isomorphism from L⊗KAio to EndL(Vi). Hence the ring L⊗KA1o⊗KA2o, which we can identify with (L⊗KA1o)⊗L(L⊗KA2o), is isomorphic to EndL(V1⊗LV2). Set C = EndA1o⊗KA2o(V1⊗LV2). By Lemma 3 of VIII, p. 282, the ring C is similar to A1⊗KA2, and L⊗1⊗1 is a maximal commutative subalgebra of C. For every pair (γ1,γ2)∈Γ1×Γ2 satisfying π1(γ1)=π2(γ2), we denote by w(γ1,γ2) the unique λ(π1(γ1))-semilinear (II, §1, No. 13, p. 223) endomorphism such that
w(γ1,γ2)(x1⊗x2)=v1(γ1)x1⊗v2(γ2)x2
for x1∈V1 and x2∈V2. We have w(γ1,γ2)∈C∗, and w is a group homomorphism from the fiber product Γ1×GΓ2 to C∗. This homomorphism is trivial on the image of ρ and induces a homomorphism v from Γ to C∗. Denote by u:L→C the morphism given by u:ℓ↦ℓ⊗1⊗1. We can verify the relations
u(α)=v(ι(α)) and u(γβ)=v(γ)u(β)v(γ)−1
for α∈L∗,β∈L, and γ∈Γ. Proposition 12 of VIII, p. 315 gives a homomorphism f from the algebra A[E;L] to C. Since the algebra A[E;L] is simple, the homomorphism f is injective. The algebras C and A[E;L] have the same dimension over K, so f is an isomorphism.
If L is an étale algebra over K and G is the automorphism group of L, then it is not always true that the algebra A[E;L] is central simple (for example, we can take L=Kn and G=Sn).
We can calculate a 2-cocycle c associated with an algebra A split by a finite Galois extension L with group G as follows. First, there exists a K-homomorphism φ:A→Mm(L), where [A:K]=m2. For g∈G, let φg be the homomorphism from A to Mm(L) given by a↦φ(g−1ag). By the Skolem–Noether theorem (VIII, p. 256, Theorem 3), for every g∈G, there exists an element ug of GLm(L) such that
φg(a)=ugφ(a)ug−1
for a∈A. We then set
c(g,g′)=ugug′ugg′−1
We can also define an extension of G by L∗ using φ: we consider the group Γ⊂GLm(L) consisting of the γ for which there exists a g∈G with
φg(a)=γφ(a)γ−1
for every a∈A. The class of this extension is the inverse image by Ψ of the class of A in Br(L/K).
The mapping ΦL/K=Θ∘Ψ−1 defines a group isomorphism from Br(L/K) to H2(G,L∗).
Let K′ be an extension of K and φ:K′→L a morphism of K-algebras. The set H of elements h of G such that λ(h)∘φ=φ is a subgroup of G, and the K′-algebra L endowed with the restriction of λ to H is a Galois algebra over K′.
Let K be a commutative field, and let A be a central simple A-algebra of finite degree over K. Let L be a separable extension of K of finite degree that is a splitting field for the algebra A. Then [L : K][A] is zero in Br(K).
There exists an extension M of L that is a Galois extension of K of finite degree (V, §10, No. 1, p. 57, Proposition 2). The class [A] of A in the Brauer group of K belongs to the subgroup Br(M/K). Let G be the Galois group of M over K, and let α be the image of [A] in H2(G,M∗) (VIII, p. 321, Corollary). Let H be the Galois group of M over L. Then H is a subgroup of index [L : K] in G (V, §10, No. 7, p. 68, Corollary 5). Since ResHG(α)=ΦM/L(A(L)) (Proposition 14), we have ResHG(α)=0. By Proposition 8 of VIII, p. 303, it follows that [L:K]α=0, and consequently [L : K][A] = 0.
Let K be a commutative field, and let A be a central simple algebra of finite degree over K. Then A is isomorphic to an algebra of the form Mn(D), where D is a field with center K, and [A] = [D] in Br(K). The reduced degree of D depends only on A. We call this reduced degree the index of A. The index of A divides the reduced degree of A. The exponent of A is the order of the class of A in the Brauer group of K.
The exponent of a central simple algebra of finite degree over a commutative field divides its index.
It suffices to prove this for a field D of finite degree over its center K. Let L be a maximal commutative subfield of D that is a separable extension of K; we have [D:K]=[L:K]2 (VIII, p. 265, Corollary 2, b) and c)). Then the extension L of K is a splitting field for the algebra D (VIII, p. 281, Proposition 5), and [L : K] coincides with the reduced degree of D. We then apply Theorem 6.
Let K be a commutative field, and let A be a central simple algebra of finite degree over K. Let p be a prime number. If p divides the index of A, then it divides the exponent of A.
Let us suppose that the prime number p does not divide the exponent of A and prove that it does not divide its index. It suffices to prove this result in the case when A is a field. Let L be a Galois extension of K of finite degree that splits A. Let G be the Galois group of L over K, and let H be a Sylow p-subgroup of G (I, §6, No. 6, p. 78). We denote by L′=LH the subfield of L of invariants of H. The exponent of A(L′) divides that of A and is therefore prime to p. By Theorem 6, we have [L:L′][A(L′)]=0 in the Brauer group of L′. It follows that [A(L′)]=0 and that the field L′ splits A. We then apply Corollary 2 of VIII, p. 283; it follows that the index of A divides [L′:K] and is therefore not divisible by p.
Let p be a prime number and let K be a perfect field of characteristic p. Let A be a central simple algebra of finite degree over K. Then p does not divide the index of A.
Let us prove that the Brauer group of K does not contain any element of order p. Every Galois extension M of K of finite degree is a perfect field (V, §7, No. 1, p. 36, Proposition 2). Hence taking the p-th power is an automorphism of the group M∗. It follows that multiplication by p is an automorphism of the group H2(Gal(M/K),M∗) that is isomorphic to Br(M/K).
Consequently, the order of [A] is prime to p and, by Corollary 2, its index is not divisible by p.
By considering tensor products of quaternion algebras, it is possible to construct fields with center K, exponent 2, and arbitrarily large index (cf. VIII, p. 371, Exercises 7 and 8).
∗Conversely, if K is a finite Galois extension of a p-adic field or a field of formal power series over a finite field, then the exponent of a central simple algebra of finite degree over K is equal to its index (VIII, p. 332, Exercise 17, e)).∗