Bourbaki

§ 5. Principal fibre spaces

1. Principal fibre spaces

Definition 1 01Q5

Let G be a topological group and B a topological space. A (right) principal fibre space with base B and group G is a B-space (E,p)(E, p) endowed with a right operation of G on E (A, I, p. 50) having the following property:

(FP) For every point bb of B, there exists a neighbourhood U of bb and a

U-isomorphism f:U×G→p−1(U)f: U\times G\rightarrow \overset{-1}{p}(U) such that for all u∈Uu\in U

and g,g′∈Gg,g'\in G, one has f(u,gg′)=f(u,g)⋅g′f(u, gg') =f(u, g)\cdot g'.

Instead of saying that (E,p)(E, p) is a principal fibre space with base B and group G, one sometimes says that the quadruple (E,G,B,p)(E,G,B, p) is a principal fibration (cf. VAR, R, § 6), or, by abuse, that the mapping p:E→Bp: E\rightarrow B is a principal fibration with base B and group G.

When no doubt is possible as to the base B, the group G and the operation of G on E, we shall simply say that (E,p)(E, p) is a principal fibre space.

Let G be a topological group. Let (E,p)(E, p) be a B-space endowed with a left operation of G. If, for the right operation of the group G∘G^{\circ} opposite to the given operation (A, I, p. 50), (E,p)(E, p) is a right principal fibre space with group G∘G^{\circ}, one says that (E,p)(E, p) is a left principal fibre space with group G.

A principal fibre space is a locally trivial fibre space (I, p. 68, def. 1).

It follows from property (FP) that the group G acts continuously (TG, III, p. 9) and freely on E, and that the orbits of this operation are the fibres of the B-space (E,p)(E, p). The mapping p′:E/G→Bp': E/G\rightarrow B deduced from pp is continuous and bijective. Again from property (FP), the mapping pp is open (TG, I, p. 30, prop. 2 and p. 26, prop. 5); hence p′p' is a homeomorphism (TG, I, p. 32, prop. 3).

Let (E,p)(E, p) and (E′,p′)(E', p') be principal fibre spaces with base B and group G. A mapping f:E→E′f: E\rightarrow E' is said to be a morphism of principal fibre spaces (with base B and group G) if ff is a B-morphism and if one has f(x⋅g)=f(x)⋅gf(x\cdot g) =f(x)\cdot g for every x∈Ex\in E and every g∈Gg\in G. Let (E′′,p′′)(E'', p'') be a principal fibre space with base B and group G. If f:E→E′f: E\rightarrow E' and g:E′→E′′g: E'\rightarrow E'' are morphisms of principal fibre spaces, the mapping g∘f:E→E′′g\circ f: E\rightarrow E'' is a morphism of principal fibre spaces. In accordance with the general definitions (E, IV, p. 11), one may take as morphisms of the structure of principal fibre space with base B and group G those defined above. We denote by CBG(E;E′)\mathscr{C}_B^G(E; E') the set of morphisms of principal fibre spaces from (E,p)(E, p) into (E′,p′)(E', p').

A trivial principal fibre space with base B and group G is the B-space (B×G(B\times G, pr1)_1) endowed with the law of right operation of G on B×GB\times G defined by (b,g)⋅h=(b,gh)(b, g)\cdot h= (b, gh).

A principal fibre space (E,p)(E, p) with base B and group G is said to be trivializable if there exists an isomorphism of (E,p)(E, p) onto the trivial principal fibre space (B×G(B\times G, pr1)_1); such an isomorphism is called a trivialization of the principal fibre space (E,p)(E, p). Property (FP) expresses that there exists an open covering (Ui)i∈I(U_i)_{i\in I} of B such that, for every i∈Ii\in I, the UiU_i-space (p−1(Ui),p∣Ui)(\overset{-1}{p}(U_i), p|U_i), endowed with the right operation of G deduced from that of G on E, is a trivializable principal fibre space.

Example 1 01Q6

Let (E,p)(E, p) be a principal fibre space with base B and group G, and let A be a subspace of B. The subspace EA=p−1(A)E_A=\overset{-1}{p}(A) of E is stable under the operation of G, and the mapping pA:EA→Ap_A: E_A\rightarrow A makes it into a principal fibre space with base A and group G. One says that (EA,pA)(E_A, p_A) is the principal fibre space induced by (E,p)(E, p) over A.

Example 2 01Q7

Let (E,p)(E, p) be a principal fibre space with base B and group G; let (E′,p′)(E', p') be a principal fibre space with base B′B' and group G′G'. We define a law of operation on the right of the group G×G′G\times G' on the space E×E′E\times E' by putting (x,x′)⋅(g,g′)=(x⋅g,x′⋅g′)(x, x')\cdot (g, g') = (x\cdot g, x'\cdot g') for x∈E,x′∈E′,g∈Gx\in E,x'\in E',g\in G and g′∈G′g'\in G'; this operation is continuous. Let U be an open subset of B and f:U×G→p−1(U)f: U\times G\rightarrow \overset{-1}{p}(U) a trivialization of the U-space (p−1(U),pU)(\overset{-1}{p}(U), p_U);

let U′U' be an open subset of B′B' and f′:U′×G′→(−p′1)(U′)f': U'\times G'\rightarrow (^-{p'}^1)(U') a trivialization of the U′U'-space ((−p′1)(U′),pU′(^-{p'}^1)(U'), p'_U). The mapping ((b,b′),(g,g′))↦((b, b'),(g, g'))\mapsto (f(b,g),f′(b′,g′))(f(b, g), f'(b', g')) is a (U×U′)(U\times U')-isomorphism f′′f'' of (U×U′)×(G×G′)(U\times U')\times (G\times G')

onto p−1(U)×(−p′1)(U′)\overset{-1}{p}(U)\times (^-{p'}^1)(U') and one has

f′′((b,b′),(gh,g′h′))=(f(b,gh),f′(b′,g′h′))=f′′((b,b′),(g,g′))⋅(h,h′) f''((b, b'),(gh, g'h')) = (f(b, gh), f'(b', g'h')) =f''((b, b'),(g, g'))\cdot (h, h')

It follows that the (B×B′)(B\times B')-space E×E′E\times E', endowed with the law of operation of G×G′G\times G' defined above, is a principal fibre space, called the product of the principal fibre spaces E and E′E'.

Example 3 01Q8

In particular, when B=B′B = B', the B-space E×BE′E\times_BE' is identified with the principal fibre space with group G×G′G\times G' induced by E×E′E\times E' over the diagonal ΔB\Delta_B of B×BB\times B. Endowed with this structure of principal fibre space, the B-space E×BE′E\times_BE' is called the fibre product of the principal fibre spaces E and E′E'.

Example 4 01Q9

Let E be a principal fibre space with base B and group G and let F be a topological space. The (B×F)(B\times F)-space E×FE\times F endowed with the law of operation (x,y)⋅g=(x⋅g,y)(x, y)\cdot g= (x\cdot g, y), for x∈E,y∈Fx\in E,y\in F and g∈Gg\in G, is a principal fibre space with group G. This is the particular case of Example 2, where p′p' is the identity mapping of F and G′G' is a group reduced to the identity element.

Example 5 01QA

Let B be a topological space and (E,p)(E, p) a principal fibre space with base B and group G. Let B′B' be a topological space and f:B′→Bf: B'\rightarrow B a continuous mapping. The group G operates on the right in the fibre product B′×BEB'\times_BE by the law (b′,x)⋅g=(b′,x⋅g)(b', x)\cdot g= (b', x\cdot g), for b′∈B′,x∈Eb'\in B',x\in E and g∈Gg\in G. This operation is continuous and free; the orbits are the fibres of the mapping pr1:B′×BE→B′_1: B'\times_BE\rightarrow B'. It follows from Examples 1 and 4 above and Remark 2 of I, p. 16 that the B′B'-space B′×BEB'\times_BE is a principal fibre space with group G. It is called the principal fibre space deduced from (E,p)(E, p) by the base change ff, or again the inverse image of (E,p)(E, p) by ff.

Proposition 1 01QB

Every morphism of principal fibre spaces is an isomorphism.

Let f:B×G→B×Gf: B\times G\rightarrow B\times G be a morphism of the trivial principal fibre space (B×G(B\times G, pr1)_1) into itself and let φ:B×G→G\varphi : B\times G\rightarrow G be the continuous mapping pr2∘f_2\circ f, so that f(b,g)=(b,φ(b,g))f(b, g) = (b, \varphi (b, g)). For every b∈Bb\in B and all g,h∈Gg,h\in G, one has φ(b,gh)=φ(b,g)⋅h\varphi (b, gh) =\varphi (b, g)\cdot h, whence φ(b,g)=φ(b,e)⋅g\varphi (b, g) =\varphi (b, e)\cdot g, where ee denotes the identity element of G. The morphism ff is therefore an isomorphism whose inverse morphism f−1f^{-1} is defined by f−1(b,g)=f^{-1}(b, g) = (b,φ(b,e)−1g)(b, \varphi (b, e)^{-1}g) for (b,g)∈B×G(b, g)\in B\times G.

Now let E and E′E' be principal fibre spaces and let f:E→E′f: E\rightarrow E' be a morphism of principal fibre spaces. In view of property (FP), for every point b∈Bb\in B, there exists an open neighbourhood U of bb such that the principal fibre spaces EUE_U and EU′E'_U are trivializable. By passing to subspaces, ff induces a morphism fU:EU→EU′f_U: E_U\rightarrow E'_U of principal fibre spaces; by what precedes, this morphism fUf_U is an isomorphism. In particular, fb:Eb→Eb′f_b: E_b\rightarrow E'_b is a bijection. There then exists a unique mapping g:E′→Eg: E'\rightarrow E which induces the bijection fb−1f_b^{-1} by passing to subspaces. By Proposition 4 of I, p. 19, the mapping gg is continuous, hence ff is an isomorphism of principal fibre spaces.

Corollary 01QC

Let G be a topological group, (E,p)(E, p) and (E′,p′)(E', p') principal fibre spaces of group G and with bases B and B′B' respectively. Let f:B′→f: B'\rightarrow B and f′:E′→f': E'\rightarrow E be continuous mappings such that p∘f′=f∘p′p\circ f'=f\circ p' and such that, for every x′∈E′x'\in E' and every g∈Gg\in G, f′(x′⋅g)=f′(x′)⋅gf'(x'\cdot g) =f'(x')\cdot g. Then the square

E′f′{E'}^{f'} E

p′pp'p

B′f{B'}^f B

is a cartesian square.

In fact, the mapping h:E′→B′×BEh: E'\rightarrow B'\times_BE defined by h(x′)=h(x') = (p′(x′),f′(x′))(p'(x'), f'(x')) for x′∈E′x'\in E', is a B′B'-morphism of principal coverings, hence an isomorphism; moreover pr2∘h=f′_2\circ h=f', whence the result (I, p. 8, Prop. 2).

Under the hypotheses of the preceding corollary, one sometimes says that f′f' is an ff-morphism of principal fibre spaces.

Proposition 2 01QD

Let (E,p)(E, p) be a principal fibre space with base B and group G, and let ε\varepsilon be the section b↦(b,e)b\mapsto (b, e) of the trivial principal fibre space (B×G(B\times G, pr1)_1), where ee denotes the identity element of G. The mapping h↦h∘εh\mapsto h\circ \varepsilon is a bijection of CBG(B×G;E)\mathscr{C}_B^G(B\times G; E) onto the set S(B;E)\mathscr{S}(B; E) of continuous sections of pp. The inverse bijection associates to a continuous section ss of pp the B-morphism (b,g)↦s(b)⋅g(b, g)\mapsto s(b)\cdot g.

Corollary 01QE

A principal fibre space is trivializable if and only if it admits a continuous section.

Proposition 3 01QF

Let E and B be topological spaces, p:E→Bp: E\rightarrow B a continuous mapping and G a topological group operating on the right on E. The following conditions are equivalent:

(i) Endowed with the mapping pp, E is a principal fibre space with base B and group G;

(ii) For every x∈Ex\in E and every g∈Gg\in G, one has p(x⋅g)=p(x)p(x\cdot g) =p(x), the mapping θ:(x,g)↦(x,x⋅g)\theta : (x, g)\mapsto (x, x\cdot g) is a homeomorphism of E×GE\times G onto E×BEE\times_BE and every point of B possesses a neighbourhood over which there exists a continuous section of the mapping pp.

(i)⇒\Rightarrow(ii) : Suppose that (E,p)(E, p) is a principal fibre space. The group G operates continuously and freely in E, with the fibres of pp as orbits. Hence the mapping θ\theta is bijective and continuous. More precisely, if E×BEE\times_BE is endowed with the operation of G defined by (x,y)⋅g=(x,y⋅g)(x, y)\cdot g= (x, y\cdot g), the mapping θ\theta is an E-morphism of the trivial principal fibre space E×GE\times G into the principal fibre space (E×BE→E(E\times_BE\rightarrow E, pr1)_1) (example 5), hence an isomorphism (prop. 1). The other assertions of (ii) result immediately from the definition.

(ii)⇒\Rightarrow(i) : Put φ=\varphi = pr2∘θ−1_2\circ \theta^{-1}, so that, for every (x,y)∈(x, y)\in E×BEE\times_BE, one has

θ−1(x,y)=(x,φ(x,y))(1) \theta^{-1}(x, y) = (x, \varphi (x, y)) \tag{1}

The mapping φ:E×BE→G\varphi : E\times_BE\rightarrow G is continuous and, for (x,y)∈E×BE(x, y)\in E\times_BE, the element φ(x,y)\varphi (x, y) is the unique element gg of G such that y=x⋅gy=x\cdot g. Let bb be a point of B; let U be a neighbourhood of bb and let ss be a continuous section of pp over U. For (u,g)∈U×G(u, g)\in U\times G, put f(u,g)=s(u)⋅f(u, g) =s(u)\cdot gg; the mapping ff is a U-morphism of U×GU\times G into p−1(U)\overset{-1}{p}(U). For y∈p−1(U)y\in \overset{-1}{p}(U), put f′(y)=(p(y),φ(s(p(y)),y))f'(y) = (p(y), \varphi (s(p(y)), y)). The mapping f′f' is a U-morphism of p−1(U)\overset{-1}{p}(U) into U×GU\times G. One has

(f∘f′)(y)=s(p(y))⋅φ(s(p(y)),y)=y (f\circ f')(y) =s(p(y))\cdot \varphi (s(p(y)), y) =y

On the other hand, for (u,g)∈U×G(u, g)\in U\times G, one has

(f′∘f)(u,g)=f′(s(u)⋅g)=(u,φ(s(u),s(u)⋅g))=(u,g) (f'\circ f)(u, g) =f'(s(u)\cdot g) = (u, \varphi (s(u), s(u)\cdot g)) = (u, g)

Hence ff is a U-isomorphism. It follows that (E,p)(E, p) is a principal fibre space with base B and group G.

Remark 1 01QG

Let (E,p)(E, p) be a principal fibre space with base B and group G. With the notation of proposition 3, the mapping θ−1:E×BE→E×G\theta^{-1}: E\times_BE\rightarrow E\times G is a trivialisation of the principal fibre space pr1:E×BE→E_1: E\times_BE\rightarrow E. One says that θ−1\theta^{-1} is the canonical trivialisation of this principal fibre space.

Corollary 1 01QH

Let G be a topological group operating continuously on the right in a topological space E. The following conditions are equivalent:

(i) The orbit space E/GE/G is Hausdorff and the space E, endowed with the canonical mapping p:E→E/Gp: E\rightarrow E/G, is a principal fibre space with group G;

(ii) The group G operates properly (TG, III, p. 27) and freely in E, and every point of E/GE/G possesses a neighbourhood over which there exists a continuous section of the mapping pp.

Put B=E/GB = E/G. Under each of the hypotheses (i) and (ii), the group G operates continuously and freely in E, so that the mapping θ:(x,g)↦(x,x⋅g)\theta : (x, g)\mapsto (x, x\cdot g) is a continuous bijection of E×GE\times G onto E×BEE\times_BE. Let us place ourselves under this hypothesis and denote by φ:E×BE→G\varphi : E\times_BE\rightarrow G the mapping pr2∘θ−1_2\circ \theta^{-1}. Having regard to formula (1), in order that θ\theta be a homeomorphism, it is necessary and sufficient that the mapping φ\varphi be continuous. On the other hand, since the equivalence relation defined by G is open (TG, III, p. 10, lemma 2), in order that the space E/GE/G be Hausdorff, it is necessary and sufficient that the graph E×BEE\times_BE of this equivalence relation be closed in E×EE\times E (TG, I, p. 55, prop. 8). Finally (TG, III, p. 31, prop. 6), in order that G operate properly in E, it is necessary and sufficient that E×BEE\times_BE be closed in E×EE\times E and that the mapping φ\varphi be continuous. The equivalence of conditions (i) and (ii) then follows from prop. 3.

Corollary 2 01QI

Let G be a topological group, let H be a subgroup of G, and let p:G→G/Hp: G\rightarrow G/H be the canonical mapping. If the mapping pp possesses a continuous section over a nonempty open set of G/HG/H, the mapping pp makes G into a principal fibre space with base G/HG/H and group H.

By left translations, every point of G/HG/H possesses a neighbourhood over which the mapping pp possesses a continuous section. On the other hand, for (g,g′)(g, g') belonging to G×GG\times G, put φ(g,g′)=\varphi (g, g') = g−1g′g^{-1}g'. If p(g)=p(g′),φ(g,g′)p(g) =p(g'),\varphi (g, g') belongs to H. The mapping (g,g′)↦(g, g')\mapsto (g,φ(g,g′))(g, \varphi (g, g')) of G×G/HGG\times_{G/H}G into G×HG\times H is continuous and is the inverse of the continuous mapping θ:G×H→G×G/HG\theta : G\times H\rightarrow G\times_{G/H}G defined by θ(g,h)=(g,gh)\theta (g, h) = (g, gh), whence the corollary.

Remark 2 01QJ

This situation presents itself in particular when G is a real Lie group, of finite dimension, countable at infinity, operating transitively and analytically on an analytic manifold X. If one takes for H the stabiliser of a point of X, the mapping pp is a submersion of G onto the homogeneous Lie space G/HG/H, isomorphic to X (LIE, III, p. 109, corollary). It therefore possesses local sections (VAR, p. 50).

2. Principal Coverings

Definition 2 01QK

Let B be a topological space and let G be a group. Endow G with the discrete topology. A principal fibre space with base B and group G is called a principal covering of B with group G.

This terminology is legitimate, for such a B-space is a covering of B.

A principal covering with group G of a nonempty topological space possesses a degree, equal to Card G.

Proposition 4 01QL

Let B be a topological space, (E,p)(E, p) a B-space, and G a discrete topological group operating on the right on E. The following assertions are equivalent:

(i) The B-space E is a principal covering with group G;

(ii) For every g∈Gg\in G and every x∈Ex\in E, one has p(x⋅g)=p(x)p(x\cdot g) =p(x), the mapping pp induces a homeomorphism of E/GE/G onto B, and the mapping θ:(x,g)↦(x,x⋅g)\theta : (x, g)\mapsto (x, x\cdot g) is a homeomorphism of E×GE\times G onto E×BEE\times_BE;

(iii) The group G operates continuously and freely in E, the mapping pp is etale, and its fibres are the orbits of G.

(i)⇒\Rightarrow(ii) : This follows from I, p. 91 and from proposition 3 of I, p. 94.

(ii)⇒\Rightarrow(iii): Under assumption (ii), the action law of G is continuous, and the mapping pp is surjective and open (TG, III, p. 10, lemma 2). Let x∈Ex\in E; the mapping θ\theta induces a bijection of {x}×G\{x\} \times G onto {x}×\{x\} \times p−1(p(x))\overset{-1}{p}(p(x)), hence the group G operates freely and its orbits are the fibres of pp. If ee denotes the identity element of G, the diagonal of E×BEE\times_BE is the image of E×{e}E\times \{e\} under the homeomorphism θ\theta. Since the group G is discrete, the set {e}\{e\} is open in G, therefore the diagonal ΔE\Delta_E is open in E×BEE\times_BE. By proposition 7 of I, p. 31, the mapping pp is étale.

(iii)⇒\Rightarrow(i): Every fibre of pp being an orbit of the operation of G in E, the mapping pp is surjective. Since the mapping pp is étale, for every point bb of B, there exists an open neighbourhood U of bb and a continuous section ss of pp over U. The set s(U)s(U) is open in E and ss induces a homeomorphism of U onto s(U)s(U) (I, p. 30, cor. 3). For every g∈Gg\in G, the set s(U)⋅gs(U)\cdot g is open in E and the union of the sets s(U)⋅gs(U)\cdot g, for all g∈Gg\in G, is equal to p−1(U)\overset{-1}{p}(U). If gg and g′g' are two distinct elements of G, the sets s(U)⋅gs(U)\cdot g and s(U)⋅g′s(U)\cdot g' are disjoint, since G operates freely in E. The mapping f:U×G→f: U\times G\rightarrow p−1(U)\overset{-1}{p}(U) defined by f(u,g)=s(u)⋅gf(u, g) =s(u)\cdot g for (u,g)∈U×G(u, g)\in U\times G is therefore a homeomorphism of U×GU\times G onto p−1(U)\overset{-1}{p}(U), compatible with the right operations of G. By definition, E is therefore a principal covering of B with group G.

Example 1 01QM

Let E be a topological space, G a discrete topological group operating continuously and freely in E on the right. In order that the canonical mapping p:E→E/Gp: E\rightarrow E/G make E a principal covering of E/GE/G, it is necessary and sufficient that it be étale (condition (iii) of proposition 4). For example, suppose there exists a topological space X and an étale mapping q:E→Xq: E\rightarrow X compatible with the operation of G in E; then E is a principal covering of E/GE/G. In fact, let us denote by q′:E/G→Xq': E/G\rightarrow X the mapping deduced from qq; one has q=q′∘pq=q'\circ p, therefore pp is étale by proposition 6, d) of I, p. 29.

Example 2 01QN

Let q:E→q: E\rightarrow X be an étale and separated mapping. The group AutX(E)_X(E), endowed with the discrete topology, operates continuously on the left in E. If the space E is connected, this operation is free (I, p. 34, corollary 2). Let us denote by G the group AutX(E)∘_X(E)^{\circ} and endow E with the right operation opposite to that of AutX(E)_X(E). By the preceding example, the space E, endowed with the canonical mapping p:E→E/Gp: E\rightarrow E/G, is a principal covering with group G.

3. Proper and free operations of discrete groups

Theorem 1 01QO

Let G be a discrete group operating continuously on the right in a topological space E. Suppose that every point of E possesses a neighbourhood U such that U∩(U⋅g)=∅U\cap (U\cdot g) =\emptyset for every element gg of G other than the identity element. Then the canonical mapping p:E→E/Gp: E\rightarrow E/G makes E a principal covering of E/GE/G with group G.

The operation of G in E is free by assumption, so it is enough to prove that the mapping pp is étale (I, p. 98, example 1). Let xx be a point of E and let U be an open neighbourhood of xx such that U∩U⋅g=∅U\cap U\cdot g=\emptyset for every element gg of G distinct from the identity element. The mapping pp is open (TG, III, p. 10, lemma 2) and induces an injective, continuous and open mapping of U onto p(U)p(U), hence a homeomorphism, which proves that the mapping pp is étale.

Example 1 01QP

Let nn be an integer ⩾0\geqslant 0, let Pn(R)\mathbf{P}_n(\mathbf{R}) be the projective space of dimension nn (TG, VI, p. 13) and let Sn\mathbf{S}_n be the unit sphere of Rn+1\mathbf{R}^{n+1} (TG, VI, p. 9). The canonical mapping of Sn\mathbf{S}_n onto Pn(R)\mathbf{P}_n(\mathbf{R}) makes Sn\mathbf{S}_n into a principal covering of group {1,−1}\{1,-1\}, this group operating by homotheties; the orbits are the pairs of diametrically opposite points.

Example 2 01QQ

Every covering of degree 2 possesses a unique structure of principal covering of group Z/2Z\mathbf{Z}/2\mathbf{Z}.

Example 3 01QR

Let X be a separated differentiable manifold locally of finite dimension over R\mathbf{R}, and let X~\widetilde{X} be the manifold of orientations of the tangent bundle of X (VAR, R, 10.2.4). The space X~\widetilde{X}, endowed with its canonical projection onto X, is a principal covering of X of group {1,−1}\{1,-1\}.

Corollary 1 01QS

Let G be a discrete topological group operating continuously on the right in a topological space E. The following conditions are equivalent:

(i) The group G operates properly and freely in E;

(ii) The space E/GE/G is separated and the space E is a principal covering with base E/GE/G and group G.

Moreover, under these conditions, the space E is separated.

Suppose condition (i) satisfied. Then the spaces E and E/GE/G are separated (TG, III, p. 29, prop. 3) and for every x∈Ex\in E there exists an open neighbourhood U of xx in E such that U∩(U⋅g)=∅U\cap (U\cdot g) =\emptyset for every element gg of G other than the identity element (TG, III, p. 32, prop. 8). By Theorem 1, condition (ii) is then satisfied.

The implication (ii)⇒\Rightarrow(i) follows from corollary 1 of I, p. 96.

Corollary 2 01QT

Let G be a topological group and let H be a discrete subgroup of G. Let H operate on G by right translations. Then the canonical mapping of G into the space G/HG/H of left classes modulo H endows G with a structure of principal covering of G/HG/H of group H.

Let V be a neighbourhood of the identity element ee of G such that H∩V={e}H\cap V =\{e\}. There exists an open neighbourhood U of ee in G such that U−1⋅U⊂VU^{-1}\cdot U\subset V (TG, III, p. 3) and consequently U∩U⋅h=∅U\cap U\cdot h=\emptyset for every h∈H,h=e̸h\in H,h=\not e. Corollary 2 of I, p. 100 therefore follows from theorem 1.

Example 4 01QU

Endowed with the canonical mapping of R\mathbf{R} onto T=R/Z\mathbf{T}=\mathbf{R}/\mathbf{Z} (TG, V, p. 22),R,\mathbf{R} is a principal covering of R/Z\mathbf{R}/\mathbf{Z} of group Z\mathbf{Z}.

Remark 1 01QV

Let G be a separated topological group, K a compact subgroup of G and H a discrete subgroup of G. The group G operates continuously and properly on the right in the space K\GK\backslash G of right classes modulo K (TG, III, p. 30, corollary). Since the group H is closed in G, it also operates properly in K\GK\backslash G (TG, III, p. 27, example 1). If, moreover, one has H∩gKg−1={e}H\cap gKg^{-1}=\{e\} for every g∈Gg\in G, the group H operates freely in K\GK\backslash G. The space K\GK\backslash G is then a principal covering of group H of K\G/HK\backslash G/H by corollary 1.

Corollary 3 01QW

Let G and G′G' be topological groups, and let φ:G→G′\varphi : G\rightarrow G' be a continuous, open and surjective homomorphism. Suppose that the kernel H of φ\varphi is discrete. Then, for the operation of H in G by right translations, φ\varphi makes G into a principal covering of G′G' of group H.

If moreover the group G is connected, H is contained in the center of G.

The homomorphism φ\varphi induces a topological group isomorphism of G/HG/H onto G′G' (TG, III, p. 16, prop. 24), whence the first assertion by Corollary 2. Suppose the group G connected. For every h∈Hh\in H, the continuous mapping of G into H defined by g↦ghg−1g\mapsto ghg^{-1} is constant, with value hh. The group H is therefore contained in the center of G.

Remark 2 01QX

Let G and G′G' be topological groups and let φ:G→G′\varphi : G\rightarrow G' be a continuous open homomorphism. If the group G′G' is connected, the homomorphism φ\varphi is surjective. In fact, φ(G)\varphi (G) is an open, therefore closed, subgroup of G′G', and consequently equal to G′G'.

Remark 3 01QY

Let G be a locally compact group countable at infinity and let G′G' be a separated topological group whose underlying space is a Baire space. Every continuous surjective homomorphism of G into G′G' is open (TG, IX, p. 56, corollary and TG, III, p. 16, prop. 24).

Example 5 01QZ

For every integer n>n > 0, let us denote by μn\mu_n the group of nnth roots of unity in C\mathbf{C} (A, V, p. 75). The mapping z↦znz\mapsto z^n makes C∗\mathbf{C}^* into a principal covering of C∗\mathbf{C}^* with group μn\mu_n, this group operating in C∗\mathbf{C}^* by multiplication.

Example 6 01R0

Endowed with the mapping z↦e2πizz\mapsto e^{2\pi iz} (TG, VIII, p. 8, remark), the space C\mathbf{C} is a principal covering of C∗\mathbf{C}^* with group Z\mathbf{Z}. The mapping x↦e2πix=e(x)x\mapsto e^{2\pi ix}=\mathbf{e}(x) of R\mathbf{R} onto S1\mathbf{S}_1 makes R\mathbf{R} into a principal covering of S1\mathbf{S}_1 with group Z\mathbf{Z}.

4. Galois Coverings

Definition 3 01R1

Let B be a nonempty topological space. A covering E of B is said to be Galois if it is connected and if, for every point bb of B, the operation of the group AutB(E)_B(E) of B-automorphisms of E on the fibre EbE_b is transitive.

Let E be a Galois covering of a topological space B and let pp be its projection. The mapping pp is surjective (A, I, p. 56, def. 6); therefore the space B is connected. Consequently, the covering (E,p)(E, p) has a nonzero degree.

Proposition 5 01R2

Let B be a nonempty topological space and let E be a Galois covering of B. The mapping (h,x)↦h(x)(h, x)\mapsto h(x) of AutB(E)×E_B(E)\times E into E is a right operation law of the group AutB(E)∘_B(E)^{\circ} on E which makes E into a principal covering with group AutB(E)∘_B(E)^{\circ}.

Let us endow the group AutB(E)∘_B(E)^{\circ} with the discrete topology; the operation law (h,x)↦h(x)(h, x)\mapsto h(x) is then continuous. This operation is free (I, p. 34, Corollary 2 of Proposition 11). Since E is a Galois covering of B, its fibres are the orbits of this operation. By Proposition 4 of I, p. 97, E is a principal covering with group AutB(E)∘_B(E)^{\circ}.

Theorem 2 01R3

Let B be a connected topological space, let E be a covering of B, connected and nonempty. The following properties are equivalent:

(i) The covering E is Galois;

(ii) There exists a discrete topological group G and a continuous right operation of G on E making E into a principal covering with group G;

(iii) The covering E×BEE\times_BE of E defined by the projection pr1_1 is trivializable.

When these conditions are satisfied, the canonical mapping G→G\rightarrow AutB(E)∘_B(E)^{\circ} defined by the operation of G is a group isomorphism.

If, moreover, the space B is locally connected, the preceding properties are equivalent to the following property:

(i′') There exists a point bb of B such that the operation of the group AutB(E)_B(E) on the fibre EbE_b is transitive.

The implication (i)⇒\Rightarrow(ii) results from proposition 5 and the implication (ii)⇒\Rightarrow(iii) from remark 1 of I, p. 95. Let us prove (iii)⇒\Rightarrow(i). Let us denote by p:E→Bp: E\rightarrow B the projection of the covering E. Since B is connected, this covering has a degree, and this degree is not zero since E is not empty. Let bb be a point of B. Let us prove that the operation of AutB(E)_B(E) on the fibre EbE_b is transitive. From the above, this fibre is not empty. Let xx and x′x' be points of EbE_b. The B-morphisms h:E→Eh: E\rightarrow E such that h(x)=x′h(x) =x' correspond bijectively to the continuous sections ss of the mapping pr1:E×BE→E_1: E\times_BE\rightarrow E such that s(x)=(x,x′)s(x) = (x, x') (I, p. 9, prop. 3). Under assumption (iii), such a section exists (I, p. 70, cor. 2) and is unique (I, p. 34, cor. 1 of prop. 11) since the space E is connected. Thus, for every pair (x,x′)∈E×BE(x, x')\in E\times_BE, there exists a unique B-morphism h:E→Eh: E\rightarrow E such that h(x)=x′h(x) =x'. If h′:E→Eh': E\rightarrow E is the unique B-morphism such that h′(x′)=xh'(x') =x, one has h′(h(x))=xh'(h(x)) =x and h(h′(x′))=x′h(h'(x')) =x', whence h′∘h=h'\circ h= IdE_E and h∘h′=h\circ h'= IdE′_{E'}. This proves that hh is a B-automorphism of E. Hence the covering E is galoisian.

We have proved that conditions (i), (ii), (iii) are equivalent. Suppose them satisfied. Let δ:G→\delta : G\rightarrow AutB(E)_B(E) be the mapping which to g∈Gg\in G associates the B-automorphism x↦x⋅gx\mapsto x\cdot g of E. The mapping δ\delta is a group homomorphism of G into AutB(E)∘_B(E)^{\circ}. Since G operates freely in E, the mapping δ\delta is injective. Let h∈h\in AutB(E)_B(E), let x∈x\in E and let gg be the unique element of G such that x⋅g=h(x)x\cdot g=h(x). The B-morphisms hh and δ(g)\delta (g) coincide at xx, hence everywhere, since the space E is connected (I, p. 34, cor. 1 of prop. 11), which proves that δ\delta is an isomorphism.

Now suppose the space B locally connected and let us prove, under assumption (i′'), that the covering E is galoisian. Let E′E' be the quotient space of E by the right operation of AutB(E)∘_B(E)^{\circ} and let us denote by q:E→E′q: E\rightarrow E' the canonical mapping. It makes E into a surjective covering of E′(E'(I, p. 98, example 2); the mapping p:E→Bp: E\rightarrow B defines by passing to the quotient a continuous mapping p′:E′→Bp': E'\rightarrow B such that p′∘q=pp'\circ q=p. Since the space B is locally connected, the B-space (E′,p′)(E', p') is a covering (I, p. 81, prop. 7). Under assumption (i′'), there exists a point bb of B such that the fibre Eb′E'_b has exactly one element; since the space B is connected, the same is then true for every point bb of B (I, p. 74, prop. 4), which proves that E is a galoisian covering of B.

Proposition 6 01R4

Let B be a topological space and G a group. Let (E,p)(E, p) be a principal covering of B with group G. Suppose the space B connected and locally connected. Let E0E_0 be a connected component of E and let G0G_0 be the stabilizer subgroup of G of E0E_0 (A, I, p. 51). The B-space (E0,p∣E0)(E_0, p|E_0) is a principal covering for the right operation of G0G_0 in E0E_0. This covering is Galois.

Since the space B is locally connected, the same is true of E, so that E0E_0 is an open subspace of E (TG, I, p. 85, prop. 11). Since it is also closed (TG, I, p. 83), the space E0E_0 is therefore a covering of B (I, p. 80, cor. 1). Since B is connected, this covering has a degree; since E0E_0 is not empty, all the fibres of p∣E0p|E_0 are therefore non-empty. Let xx be a point of E0E_0, let x′∈E0x'\in E_0 be such that p(x′)=p(x)p(x') =p(x); there exists an element g∈Gg\in G such that x⋅g=x′x\cdot g=x'. Since the connected components E0E_0 and E0⋅gE_0\cdot g then have a common point, they are equal, whence g∈G0g\in G_0. Thus, the group G0G_0 operates transitively in the fibre of the B-space E0E_0 at p(x)p(x). Proposition 6 follows.

Remark 01R5

If, in proposition 6, one does not suppose the space B locally connected, it may happen that the space E0E_0 is not a covering of B (I, p. 147, exerc. 1).

5. Associated fibre spaces

Let E and F be sets and let G be a group operating on the right on E and on the left on F. The group G operates on the right on the product E×FE\times F by the law (x,y)⋅g=(x⋅g,g−1⋅y)(x, y)\cdot g= (x\cdot g, g^{-1}\cdot y), for g∈Gg\in G, (x,y)∈E×F(x, y)\in E\times F. The quotient set of E×FE\times F by this operation is denoted by E×GFE\times^GF.

When E and F are topological spaces, the set E×GFE\times^GF is endowed with the quotient topology of that of E×FE\times F. The canonical mapping of E×FE\times F onto E×GFE\times^GF is continuous. If the group G operates continuously in E and F, it is open.

Moreover, let F′F' be a set on which G operates on the left and let h:F→F′h: F\rightarrow F' be a mapping compatible with the operations of G on F and F′F' (A, I, p. 50). The mapping IdE×h:E×F→E×F′_E\times h: E\times F\rightarrow E\times F' is compatible with the operations of G and defines by passing to the quotients a mapping denoted by IdE×Gh_E\times^Gh of E×GFE\times^GF into E×GF′E\times^GF'.

If h:F→F′h: F\rightarrow F' is a continuous mapping (compatible with the operations of G on F and F′F'), the mapping IdE×Gh_E\times^Gh is continuous (TG, I, p. 21, prop. 6).

Example 1 01R6

Let F be a topological space and let G be a topological group operating continuously on the left in F. If the topological space G is endowed with the operation of G by right translations, the space G×GFG\times^GF is canonically identified with F in the following way. The continuous mappings φ:F→G×F\varphi : F\rightarrow G\times F and ψ:G×F→\psi : G\times F\rightarrow F defined by φ(f)=(e,f)\varphi (f) = (e, f) (where ee denotes the identity element of G) and ψ(g,f)=g⋅f\psi (g, f) =g\cdot f induce continuous mappings φ‾:F→G×GF\overline{\varphi}: F\rightarrow G\times^GF and ψ‾:G×GF→F\overline{\psi}: G\times^GF\rightarrow F which are reciprocal to one another.

Example 2 01R7

Example 1 is generalized as follows. Let B and F be topological spaces and let G be a topological group operating continuously on the left in F. By passing to the quotients, the mapping of B×FB\times F into (B×G)×F(B\times G)\times F given by (b,f)↦((b,e),f)(b, f)\mapsto ((b, e), f) and the mapping of (B×G)×F(B\times G)\times F into B×FB\times F given by ((b,g),f)↦(b,gf)((b, g), f)\mapsto (b, gf) define reciprocal B-isomorphisms B×F→(B×G)×GFB\times F\rightarrow (B\times G)\times^GF and (B×G)×GF→B×F(B\times G)\times^GF\rightarrow B\times F.

Example 3 01R8

Analogously, let E be a topological space and let G be a topological group operating continuously on the right in E. If the topological space G is endowed with the operation of G by left translations, the space E×GGE\times^GG is identified with E.

Let E and F be topological spaces and let G be a topological group operating continuously on the right in E and on the left in F. Let B be a topological space and let p:E→Bp: E\rightarrow B be a continuous mapping such that p(x⋅g)=p(x)p(x\cdot g) =p(x) for x∈Ex\in E and g∈Gg\in G. The mapping p∘p\circ pr1:E×F→B_1: E\times F\rightarrow B defines, by passing to the quotient, a continuous mapping pF:E×GF→Bp^F: E\times^GF\rightarrow B and the canonical mapping π:E×F→E×GF\pi : E\times F\rightarrow E\times^GF is a B-morphism.

Let B′B' be a topological space and let h:B′→Bh: B'\rightarrow B be a continuous mapping. The group G operates continuously on the right in B′×BEB'\times_BE by the operation law ((b′,x),g)↦(b′,x⋅g)((b', x), g)\mapsto (b', x\cdot g). By composing the canonical isomorphism ((b′,x),y)↦(b′,(x,y))((b', x), y)\mapsto (b',(x, y)) of (B′×BE)×F(B'\times_BE)\times F onto B′×B(E×F)B'\times_B(E\times F) and the mapping IdB′×π_{B'}\times \pi of B′×B(E×F)B'\times_B(E\times F) into B′×B(E×GF)B'\times_B(E\times^GF), one obtains a continuous mapping λ0:(B′×BE)×F→B′×B(E×GF)\lambda_0: (B'\times_BE)\times F\rightarrow B'\times_B(E\times^GF). It defines, by passing to the quotient, a continuous mapping

λ:(B′×BE)×GF→B′×B(E×GF) \lambda : (B'\times_BE)\times^GF\rightarrow B'\times_B(E\times^GF)

Lemma 01R9

The mapping λ\lambda is a homeomorphism.

The mapping λ0\lambda_0 is surjective and two elements of (B′×BE)×F(B'\times_BE)\times F have the same image under λ0\lambda_0 if and only if they belong to the same orbit for the operation of G in (B′×BE)×F(B'\times_BE)\times F. It follows that the mapping λ\lambda is bijective. Since the mapping π\pi is open, the same is true of the mapping IdB′×Bπ(_{B'}\times_B\pi (I, p. 17, prop. 8), hence of λ0\lambda_0 and of λ\lambda (TG, I, p. 32, prop. 3), which proves that λ\lambda is a homeomorphism.

Proposition 7 01RA

Let B be a topological space, let G be a topological group, let (E,p)(E, p) be a principal fibred space over B with group G and let F be a topological space in which the group G operates continuously on the left.

a) The topological space E×GFE\times^GF endowed with the continuous mapping pF:E×GF→Bp^F: E\times^GF\rightarrow B deduced from the mapping p∘p\circpr1:E×F→B_1: E\times F\rightarrow B by passing to the quotient is a locally trivial fibred space over B of fibre type F; it is trivializable if the fibred space over B E is trivializable.

b) Let π:E×F→E×GF\pi : E\times F\rightarrow E\times^GF be the canonical surjection. The mapping μ:E×F→E×B(E×GF)\mu: E\times F\rightarrow E\times_B(E\times^GF) which associates to (x,f)(x, f) the element (x,π(x,f))(x, \pi (x, f)) is a homeomorphism whose inverse mapping is a trivialization of the locally trivial fibred space over E (E×B(E×GF)(E\times_B(E\times^GF), pr1)_1).

Suppose first that E is the trivial principal fibred space over B B×GB\times G. By example 2, the space E×GFE\times^GF is then identified with B×FB\times F and the mapping pFp^F with the first projection pr1:B×F→B_1: B\times F\rightarrow B, which proves that (E×GF,pF)(E\times^GF, p^F) is a trivializable fibred space over B in this case. In the general case, every point of B possesses a neighbourhood U such that the mapping pU:p−1(U)→Up_U:\overset{-1}{p}(U)\rightarrow U makes p−1(U)\overset{-1}{p}(U) into a trivializable principal fibred space over U. From what precedes, (p−1(U)×GF→U,(pU)F)(\overset{-1}{p}(U)\times^GF\rightarrow U,(p_U)^F) is a trivializable fibred space over U. By the above lemma, applied to the case where B′=UB'= U and h:U→Bh: U\rightarrow B is the canonical injection, the same is

true of the fibred space over U ((p−F1)(U),(pF)U(p^{-F1})(U),(p^F)_U) deduced from (E×GF,pF)(E\times^GF, p^F) by restriction above U. This proves that (E×GF,pF)(E\times^GF, p^F) is a locally trivial fibred space over B and concludes the proof of assertion a).

The mapping θ:E×G→E×BE\theta : E\times G\rightarrow E\times_BE defined by θ(x,g)=(x,x⋅g)\theta (x, g) = (x, x\cdot g) is a homeomorphism (I, p. 94, prop. 3) compatible with the operations of G on E×GE\times G and on E×BEE\times_BE given by ((x,g),g′)↦(x,gg′)((x, g), g')\mapsto (x, gg') and ((x,y),g′)↦(x,yg′)((x, y), g')\mapsto (x, yg') respectively. By passing to quotients, θ\theta therefore induces a homeomorphism θ′\theta ' of (E×G)×GF(E\times G)\times^GF onto (E×B(E\times_B E)×GFE)\times^GF. The mapping μ\mu is the composite of the homeomorphism E×F→E\times F\rightarrow (E×G)×GF(E\times G)\times^GF (example 2), of θ′\theta ', and of the canonical homeomorphism (E×BE)×GF→E×B(E×GF)(E\times_BE)\times^GF\rightarrow E\times_B(E\times^GF) (I, p. 105, lemma), whence b).

Under the hypotheses of proposition 7, the locally trivial fibred B-space (E×GF,pFE\times^GF, p^F) is called the locally trivial fibred space of fibre type F associated with the principal fibred space (E,p)(E, p). All the fibres of the B-space E×GFE\times^GF are homeomorphic to the space F. In particular, if the space F is discrete, the mapping pFp^F is a covering.

Example 4 01RB

Let B be a topological space, G a topological group, and let (E,p)(E, p) be a principal fibred B-space with group G. Let H be a subgroup of G.

Let us denote by φ:E→E×(G/H)\varphi : E\rightarrow E\times (G/H) and ψ:E×(G/H)→E/H\psi : E\times (G/H)\rightarrow E/H the mappings defined by φ(x)=(x,H)\varphi (x) = (x,H) and ψ(x,gH)=(x⋅g)H\psi (x, gH) = (x\cdot g)H. They are compatible with the projections onto B and with the operations of G and define, by passing to quotients, morphisms of B-spaces φ‾:E/H→E×G(G/H)\overline{\varphi}: E/H\rightarrow E\times^G(G/H) and ψ‾:E×G(G/H)→E/H\overline{\psi}: E\times^G(G/H)\rightarrow E/H, inverse to one another. One says that φ‾\overline{\varphi} is the canonical homeomorphism of E/HE/H onto E×G(G/H)E\times^G(G/H). In particular, the topological space E/HE/H endowed with the continuous mapping pH:E/H→Bp_H: E/H\rightarrow B is a locally trivial fibred B-space of fibre type G/HG/H.

If, moreover, H is a normal subgroup of G, the action of G endows the B-space E/HE/H, by passing to quotients, with a structure of principal fibred space with group G/HG/H.

In particular, if E is a principal covering of B with group G, E/HE/H is a covering of B; if H is normal in G, it is a principal covering with group G/HG/H.

Example 5 01RC

Let B be a topological space, let G be a topological group, and let E be a principal fibred B-space with group G. Let F be a topological homogeneous space relative to G (TG, III, p. 12). Let yy be a point of F and let GyG_y be its stabiliser. The mapping φy:x↦(x,y)\varphi_y:x\mapsto (x, y) of E into E×FE\times F defines, by passing to quotients, a homeomorphism φ‾y\overline{\varphi}_y of E/GyE/G_y onto E×GFE\times^GF. When the group G is abelian, the subgroup GyG_y does not depend on the point yy, but the homeomorphism φ‾y\overline{\varphi}_y, in general, does depend on it.

Example 6 01RD

Let B be a topological space, let G be a topological group, let (E,p)(E, p) be a principal fibred B-space with group G. Let H be a topological group and let f:G→Hf: G\rightarrow H be a morphism of topological groups; let us endow H with the left operation of G given by g⋅h=f(g)hg\cdot h=f(g)h.

Let q:E×H→E×GHq: E\times H\rightarrow E\times^GH be the canonical mapping; it is open, hence universally strict (I, p. 20, corollary 11). The mapping m:(x,h,h′)↦q(x,hh′)m: (x, h, h')\mapsto q(x, hh') of E×H×HE\times H\times H into E×GHE\times^GH is continuous. Let (x,h)∈E×H,h′∈H(x, h)\in E\times H,h'\in H and g∈Gg\in G; we have m(xg,f(g)−1h,h′)=m(xg, f(g)^{-1}h, h') = q(xg,f(g)−1hh′)=q(x,hh′)=m(x,h,h′)q(xg, f(g)^{-1}hh') =q(x, hh') =m(x, h, h'). Consequently, there exists a unique mapping

m′:(E×GH)×H→E×GH m': (E\times^GH)\times H\rightarrow E\times^GH

such that m′(q(x,h),h′)=q(x,h,h′)m'(q(x, h), h') =q(x, h, h') for all x∈Ex\in E and all h,h′∈Hh, h'\in H. Since qq is universally strict, the mapping m′m' is continuous. This is a right operation of the topological group H on the B-space E×GHE\times^GH.

Let U be an open subset of B such that EUE_U is isomorphic to the principal fibre bundle over U, U×GU\times G. Endowed with the operation of H, the U-space (E×GH)U(E\times^GH)_U is identified with the principal fibre bundle U×HU\times H. This proves that E×GHE\times^GH is a principal fibre bundle over B with group H.

Definition 4 01RE

Let B be a topological space and let G be a topological group. A locally trivial fibre space X over B is said to be associated with a principal fibre bundle E over B with group G if there exists a topological space F on which the group G operates continuously on the left and a B-isomorphism of E×GFE\times^GF onto X.

Let E be a principal fibre bundle over B with group G and let X be a locally trivial fibre space over B associated with E. If the principal fibre bundle E is trivializable, X is trivializable (Prop. 7). If B′B' is a topological space and h:B′→Bh: B'\rightarrow B a continuous mapping, the locally trivial fibre space over B′B', B′×BXB'\times_BX, deduced from X by base change is associated with the principal fibre bundle over B′B', B′×BEB'\times_BE (I, p. 105, lemma).

6. Associated Coverings

Let B be a topological space, let G be a discrete topological group and let E be a principal covering of B with group G. Let F be a G-set; if F is endowed with the discrete topology, the group G acts continuously on F. The B-space E×GFE\times^GF is then a covering. It is called the covering of B with fibre type F associated with the principal covering E.

Definition 5 01RF

Let B be a topological space, let G be a discrete topological group and let E be a principal covering of B with group G. A covering X of B is said to be associated with the principal covering E if there exists a G-set F and a B-isomorphism of E×GFE\times^GF onto X.

Let B be a topological space, let G be a discrete topological group and let E be a covering of B, principal with group G.

For every covering X of B, the group G operates on the left on CB(E;X)\mathscr{C}_B(E; X) by the law defined by (g⋅h)(x)=h(x⋅g)(g\cdot h)(x) =h(x\cdot g) for x∈Ex\in E, g∈G,h∈CB(E;X)g\in G,h\in \mathscr{C}_B(E; X).

For every G-set F endowed with the discrete topology, let us define a mapping αF:F→CB(E;E×GF)\alpha_F: F\rightarrow \mathscr{C}_B(E; E\times^GF) by:

(2) αF(y)(x)=π(x,y)\alpha_F(y)(x) =\pi (x, y) for y∈Fy\in F and x∈Ex\in E,

where π:E×F→E×GF\pi : E\times F\rightarrow E\times^GF is the canonical surjection. The mapping αF\alpha_F is compatible with the operations of G on F and on CB(E;E×GF)\mathscr{C}_B(E; E\times^GF).

For every covering X of B, there exists a unique mapping βX\beta_X from E×GCB(E;X)E\times^G\mathscr{C}_B(E; X) into X such that:

(3) βX(π(x,h))=h(x)\beta_X(\pi (x, h)) =h(x) for h∈CB(E;X)h\in \mathscr{C}_B(E; X) and x∈Ex\in E.

In fact, one has (g−1⋅h)(x⋅g)=h(x)(g^{-1}\cdot h)(x\cdot g) =h(x) for x∈E,g∈Gx\in E,g\in G and h∈CB(E;X)h\in \mathscr{C}_B(E; X), by definition of the operation of G on CB(E;X)\mathscr{C}_B(E; X). When the set CB(E;X)\mathscr{C}_B(E; X) is endowed with the discrete topology, the mapping βX\beta_X is a B-morphism of coverings.

When X=E×GFX = E\times^GF, one has

(4) βX(π(x,αF(y)))=π(x,y)\beta_X(\pi (x, \alpha_F(y))) =\pi (x, y) for x∈Xx\in X and y∈Fy\in F.

Proposition 8 01RG

Let B be a connected nonempty topological space. Let G be a discrete group and let (E,p)(E, p) be a principal covering of B with group G. Suppose that E is connected. With the above notation, one has:

a) For every G-set F endowed with the discrete topology, the mapping αF\alpha_F is an isomorphism of the G-set F onto the G-set CB(E;E×GF)\mathscr{C}_B(E; E\times^GF).

b) Let X be a covering of B. The B-morphism βX\beta_X is an isomorphism of E×GCB(E;X)E\times^G\mathscr{C}_B(E; X) onto X if and only if the covering (E×BX(E\times_BX, pr1)_1) of E is trivializable.

a) Let yy and y′y' be points of F such that αF(y)=αF(y′)\alpha_F(y) =\alpha_F(y'). The space E is not empty; let us choose a point xx in it. We have π(x,y)=π(x,y′)\pi (x, y) =\pi (x, y') in E×GFE\times^GF. Hence there exists g∈Gg\in G such that x⋅g=xx\cdot g=x and g−1⋅y=y′g^{-1}\cdot y=y'. The first equality implies that gg is the identity element ee of G, hence y=y′y=y'. Thus the mapping αF\alpha_F is injective. On the other hand, let h∈h\in CB(E;E×GF)\mathscr{C}_B(E; E\times^GF) and let xx be a point of E. Let x′∈E,y′∈Fx'\in E,y'\in F be such that h(x)=π(x′,y′)h(x) =\pi (x', y'). In particular, p(x)=p(x′)p(x) =p(x'); there then exists an element gg of G such that x′=x⋅gx'=x\cdot g, and we also have h(x)=π(x,y)h(x) =\pi (x, y), where we have set y=g⋅y′y=g\cdot y'. The B-morphisms hh and αF(y)\alpha_F(y) coincide at the point xx of E; they are therefore equal since the space E is connected (I, p. 34, cor. 1 of prop. 11), and this proves that the mapping αF\alpha_F is surjective.

b) By proposition 7, b) of I, p. 105 applied to F=CB(E;X)F =\mathscr{C}_B(E; X), the covering p∗(E×GCB(E;X))p^*(E\times^G\mathscr{C}_B(E; X)) of E is isomorphic to the trivial covering E×CB(E;X)E\times \mathscr{C}_B(E; X). If βX\beta_X is an isomorphism, the covering p∗(X)p^*(X) of E is therefore trivializable. Conversely, suppose that the covering p∗(X)p^*(X) is trivializable and let us prove that the B-morphism βX\beta_X is bijective; it will follow that βX\beta_X is a B-isomorphism (I, p. 30, cor. 2 of prop. 6).

The mapping βX\beta_X is deduced by passing to the quotient from the mapping γ:E×CB(E;X)→X\gamma : E\times \mathscr{C}_B(E; X)\rightarrow X defined by γ(x,h)=h(x)\gamma (x, h) =h(x). Let us prove that the mapping γ\gamma is surjective. Let xx be a point of X. The projection of the B-space E is surjective, for it is a principal covering; there therefore exists a point yy of E such that (y,x)∈E×BX(y, x)\in E\times_BX. There then exists a continuous section ss of the trivializable covering pr1:E×BX→E_1: E\times_BX\rightarrow E such that s(y)=(y,x)s(y) = (y, x). The mapping h=h= pr2∘s:E→X_2\circ s: E\rightarrow X is a B-morphism, and we have h(y)=xh(y) =x, whence the surjectivity of the mapping γ\gamma and, consequently, that of the mapping βX\beta_X.

Let us prove finally that βX\beta_X is injective. Let (x,h)(x, h) and (x′,h′)(x', h') be elements of E×CB(E;X)E\times \mathscr{C}_B(E; X) such that h(x)=h′(x′)h(x) =h'(x'). Observe that xx and x′x' have the same projection in B; there therefore exists an element gg of G such that x′=x⋅gx'=x\cdot g. We then have h(x)=h′(x⋅g)=(g⋅h′)(x)h(x) =h'(x\cdot g) = (g\cdot h')(x). Since the space E is connected, we have h=g⋅h′(h=g\cdot h'(I, p. 34, cor. 1 of prop. 11). Thus, (x,h)(x, h) and (x′,h′)(x', h') have the same class in E×GCB(E;X)E\times^G\mathscr{C}_B(E; X), which proves that the mapping βX\beta_X is injective, and completes the proof.

Corollary 1 01RH

Let (E,p)(E, p) be a principal covering of B of group G; suppose that E is connected and nonempty. A covering X of B is associable to E if and only if the covering p∗(X)p^*(X) is trivializable.

If the covering p∗(X)p^*(X) is trivializable, it follows from proposition 8, b), that the covering X is associable to E. In this case, the mapping βX\beta_X identifies the covering X with the covering of fibre-type CB(E;X)\mathscr{C}_B(E; X) associated to E. Conversely, let F be a G-set endowed with the discrete topology and suppose that one has X=E×GFX = E\times^GF. Then αF\alpha_F is an isomorphism (loc. cit., a)) and formula (4) implies that βX\beta_X is an isomorphism. Consequently, the covering p∗(X)p^*(X) is trivializable (prop. 8, b)), whence the corollary.

Corollary 2 01RI

Let B be a connected and locally connected topological space, let (E,p)(E, p) be a principal covering of B of group G. Let E0E_0 be a connected component of E and let G0G_0 be the subgroup of G stabilizer of E0E_0. The B-space (E0,p∣E0)(E_0, p|E_0) is a principal covering of group G0(G_0(I, p. 103, prop. 6).

Every covering X of B which is associable to the principal covering E is associable to the principal covering E0E_0. In particular, the covering E is associable to the principal covering E0E_0.

Indeed, if the covering X is associable to E, the covering p∗(X)p^*(X) is trivializable and the covering p0∗(X)p^*_0(X) induced by p∗(X)p^*(X) above E0E_0 is therefore trivializable.

More precisely, note that the mapping (x,g)↦x⋅g(x, g)\mapsto x\cdot g of E0×GE_0\times G into E induces, by passing to the quotient, a B-isomorphism of principal coverings of E0×G0GE_0\times^{G_0}G onto E.

Proposition 9 01RJ

Let B be a topological space, let E be a principal covering of B of group G, connected and nonempty. Let F be a nonempty G-set endowed with the discrete topology. In order that the space E×GFE\times^GF be connected, it is necessary and sufficient that G operate transitively in F.

Let U be a subset open and closed of E×GFE\times^GF. If π:E×F→\pi : E\times F\rightarrow E×GFE\times^GF denotes the canonical surjection, π−1(U)\overset{-1}{\pi}(U) is a subset open and closed of E×FE\times F which is stable under G. Since E is connected, there exists a subset F′⊂FF'\subset F, stable under G, such that π−1(U)=E×F′\overset{-1}{\pi}(U) = E\times F', whence U=π(E×F′)U =\pi (E\times F'). Let F′F' and F′′F'' be subsets of F stable under G such that π(E×F′)=π(E×F′′)\pi (E\times F') =\pi (E\times F''); since E is not empty, one has F′=F′′F'= F''. The mapping F′↦π(E×F′)F'\mapsto \pi (E\times F') is a bijection of the set of subsets of F stable under G onto the set of subsets open and closed of E×GFE\times^GF. The proposition follows.

Proposition 10 01RK

Let B be a topological space, let (E,p)(E, p) be a principal covering of B of group G and let X be a covering of B. Suppose that E and X are connected and nonempty. The following properties are equivalent:

(i) The covering X is associable to the principal covering E;

(ii) There exists a subgroup H of G such that X is B-isomorphic to E/HE/H;

(iii) There exists a surjective B-morphism h:E→Xh: E\rightarrow X;

(iv) For every point (y,x)(y, x) of E×BXE\times_BX, there exists a B-morphism r:E→Xr: E\rightarrow X such that r(y)=xr(y) =x.

Suppose these conditions satisfied and let H be a subgroup of G such that X is B-isomorphic to E/HE/H. The covering X is Galois if and only if the subgroup H is normal in G.

(i)⇒\Rightarrow(ii) : Let F be a discrete G-set such that the covering E×GFE\times^GF is B-isomorphic to X. The set F is nonempty and the space E×GFE\times^GF is connected, hence the group G operates transitively in F (proposition 9). Then the space E×GFE\times^GF is B-isomorphic to E/HE/H, where H is the subgroup of G fixing a point of F (I, p. 106, example 4).

(ii)⇒\Rightarrow(iii) : In fact, the canonical surjection of E onto E/HE/H is a surjective B-morphism.

(iii)⇒\Rightarrow(iv) : Let (y,x)(y, x) be a point of E×BXE\times_BX. Since the mapping hh is surjective, there exists a point y′y' of E such that h(y′)=xh(y') =x. The points yy and y′y' have the same projection in B. Since the covering E is principal with group G, there exists g∈Gg\in G such that y⋅g=y′y\cdot g=y'. The mapping r:E→Xr: E\rightarrow X defined by z↦h(z⋅g)z\mapsto h(z\cdot g) is a B-morphism and one has r(y)=xr(y) =x.

(iv)⇒\Rightarrow(i) : Since X is not empty and the projection of E onto B is surjective, the space E×BXE\times_BX is not empty. Hence there exists a B-morphism rr of E into X. The mapping (IdE,r_E, r):E→E×BX: E\rightarrow E\times_BX is a section of the covering p∗(X)p^*(X) of E. Since the space E is connected, it follows from Corollary 2 of proposition 1 of I, p. 69 that the covering p∗(X)p^*(X) is trivializable, which proves that the covering X is associable to the Galois covering pp (proposition 8).

Suppose now that these conditions are satisfied. We denote by X the B-space E/HE/H, by qq its projection, and by h:E→E/Hh: E\rightarrow E/H the canonical mapping.

If the group H is normal in G, then X is a principal covering with group G/HG/H (I, p. 106, example 4). Since X and B are connected and nonempty, X is a Galois covering of B (I, p. 102, theorem 2). Conversely, suppose that E/HE/H is a Galois covering of B and let K = AutB(E/H)∘_B(E/H)^{\circ}. One defines a mapping α:E×G→K\alpha : E\times G\rightarrow K by associating to (t,g)∈E×G(t, g)\in E\times G the unique element kk of K such that h(t⋅g)=h(t)⋅kh(t\cdot g) =h(t)\cdot k. It is continuous because it is obtained by composing the continuous mapping (t,g)↦(h(t),h(t⋅g))(t, g)\mapsto (h(t), h(t\cdot g)) of E×GE\times G into X×BXX\times_BX with the canonical trivialization (I, p. 95, remark 1) X×BX→X×KX\times_BX\rightarrow X\times K of q∗(X)q^*(X). Since E is connected and K is a discrete group, the continuous mapping t↦α(t,g)t\mapsto \alpha (t, g) is constant, for every g∈Gg\in G; we denote by α(g)\alpha (g) its value. If t∈E,g∈Gt\in E,g\in G and g′∈Hg'\in H, one then has

h(t)=h(t⋅g−1g)=h(t⋅g−1)⋅α(g)=h(t⋅g−1⋅g′)⋅α(g)=h(t⋅(g−1g′g)) h(t) =h(t\cdot g^{-1}g) =h(t\cdot g^{-1})\cdot \alpha (g) =h(t\cdot g^{-1}\cdot g')\cdot \alpha (g) =h(t\cdot (g^{-1}g'g))

It follows that g−1g′gg^{-1}g'g belong to H and hence that H is normal in G.

Corollary 01RL

Let E be a Galois covering of B and let X be a covering of B, connected and nonempty. Suppose that X is a finite covering or else that the space B is locally connected. If there exists a B-morphism h:E→Xh: E\rightarrow X, the covering X is associable to the principal covering E.

In both cases, the B-morphism hh is a covering (I, p. 77, Cor. 3 of theorem 1, and I, p. 78, proposition 5), and a nonempty covering of a connected space is surjective (I, p. 68).

Theorem 3 01RM

Let B be a nonempty, connected and locally connected topological space. Every covering of B is associable to a Galois covering; every finite covering of B is associable to a finite Galois covering.

Let X be a covering of B, and let us denote by qq its projection. Since the space B is connected and nonempty, the covering X has a degree; let us denote this degree by F and endow the set F with the discrete topology. Since the space B is locally connected, the sheaf I\mathscr{I} = I\mathscr{I}somB(B×F;X)_B(B\times F; X) is locally constant and at every point bb of B its stalk Ib\mathscr{I}_b is canonically isomorphic to the set B(F;Xb)\mathscr{B}(F; X_b) of bijections of F onto the fibre Xb(X_b(I, p. 89, prop. 12). Let E=EIE = E_{\mathscr{I}} be the covering associated with the locally constant sheaf I(\mathscr{I}(I, p. 86, corollary).

Let G be the group of permutations of F. For every g∈Gg\in G, one defines as follows a morphism γ′(g)\gamma '(g) of the sheaf I\mathscr{I} into itself: for every open subset U of B, the mapping γ′(g)U\gamma '(g)_U associates with a U-isomorphism φ:U×F→q−1(U)\varphi : U\times F\rightarrow \overset{-1}{q}(U) the U-isomorphism defined by (b,f)↦φ(b,g(f))(b, f)\mapsto \varphi (b, g(f)) for b∈Ub\in U and f∈Ff\in F. One has γ′\gamma '(IdF_F) == IdI_{\mathscr{I}} and γ′(g∘g′)=γ′(g′)∘γ′(g)\gamma '(g\circ g') =\gamma '(g')\circ \gamma '(g), for g,g′∈Gg,g'\in G, so that for every g∈G,γ′(g)g\in G,\gamma '(g) is an automorphism of the sheaf I\mathscr{I}. The B-morphism γ(g):E→E\gamma (g): E\rightarrow E associated with γ′(g)\gamma '(g) is an automorphism (I, p. 50). If x=[U,φ,b]x= [U, \varphi , b] is an element of E, where U is an open subset of B,b,b a point of U and φ\varphi a U-isomorphism of U×FU\times F onto q−1(U)\overset{-1}{q}(U), one has γ(g)(x)=[U,ψ,b]\gamma (g)(x) = [U, \psi , b], where ψ\psi is defined by ψ(a,f)=φ(a,g(f))\psi (a, f) =\varphi (a, g(f)), for a∈a\in U and f∈f\in F. If gg and g′g' are elements of G, one has γ(g∘g′)=γ(g′)∘γ(g)\gamma (g\circ g') =\gamma (g')\circ \gamma (g). One has γ\gamma(IdF_F) == IdE_E. One thus defines a continuous right operation law of G in E by setting x⋅g=γ(g)(x)x\cdot g=\gamma (g)(x), for x∈Ex\in E and g∈Gg\in G. The group G operates simply transitively on every fibre of E, so that the covering E endowed with this operation is a principal covering with group G (I, p. 97, prop. 4). Let hh be the mapping from E×FE\times F into X defined by h([U,φ,b],f)=φ(b,f)h([U, \varphi , b], f) =\varphi (b, f) for every open subset U of B, every point bb of U and every U-isomorphism φ\varphi of U×FU\times F onto q−1(U)\overset{-1}{q}(U). By definition of the topology of E, it is continuous; it is a B-morphism. For every element gg of G and every point (x,f)(x, f) of E×FE\times F, one has h(x,g(f))=h(x⋅g,f)h(x, g(f)) =h(x\cdot g, f). The mapping hh therefore defines, by passing to the quotient, a B-morphism h′:E×GF→Xh': E\times^GF\rightarrow X. For every point bb of B, the mapping hb:Eb×F→Xbh_b: E_b\times F\rightarrow X_b identifies with the mapping (φ,f)↦φ(f)(\varphi , f)\mapsto \varphi (f) from B(F;Xb)×F\mathscr{B}(F; X_b)\times F into XbX_b, so that the mapping hb′h'_b is bijective. Consequently, h′h' is a B-isomorphism of E×GFE\times^GF onto X (I, p. 30, cor. 2 of prop. 6).

As the space B is connected, locally connected and nonempty, the covering X, associated with the principal covering E, is associated with a Galois covering (I, p. 110, corollary 2). If the covering X is finite, the same is true of the covering E, hence X is associated with a finite Galois covering (loc. cit.).

7. Principal Fibre Spaces Defined by Cocycles

Definition 6 01RN

Let B be a topological space, let G be a topological group, and let U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} be an open covering of B. A continuous 1-cocycle on B with values in G, subordinate to U\mathscr{U}, is by definition the data, for each pair (i,j)(i, j) of elements of I, of a continuous mapping gi,jg_{i,j} of Ui∩UjU_i\cap U_j into G, such that for every triplet (i,j,k)∈I×I×I(i, j, k)\in I\times I\times I and every point bb of Ui∩Uj∩UkU_i\cap U_j\cap U_k, one has

gi,k(b)=gi,j(b)gj,k(b) g_{i,k}(b) =g_{i,j}(b)g_{j,k}(b)

We denote by Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G) the set of continuous 1-cocycles on B with values in G, subordinate to the covering U\mathscr{U}.

In this section, we shall simply say cocycle instead of continuous 1-cocycle.

Let (E,p)(E, p) be a principal fibre space over B of group G, and let U\mathscr{U} = (Ui)i∈I(U_i)_{i\in I} be a covering of B by open sets UiU_i. One says that E is trivializable over U\mathscr{U} if, for every i∈Ii\in I, E is trivializable over UiU_i. One then calls a U\mathscr{U} -trivialization of E a family (fi)i∈I(f_i)_{i\in I}, where fi:p−1(Ui)→Ui×Gf_i:\overset{-1}{p}(U_i)\rightarrow U_i\times G is a trivialization of E over UiU_i.

Let U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} be a covering of B by open sets, and let (E,p)(E, p) be a principal fibre space over B of group G endowed with a U\mathscr{U} -trivialization (fi)i∈I(f_i)_{i\in I}. For every pair (i,j)∈I×I(i, j)\in I\times I, let us denote by gijg_{ij} the mapping of Ui∩UjU_i\cap U_j into G defined by

(b,gij(b))=fi∘fj−1(b,e) (b, g_{ij}(b)) =f_i\circ f_j^{-1}(b, e)

where ee denotes the identity element of G. It is continuous.

Since fif_i and fjf_j are compatible with the operations of G, one has

(fi∘fj−1)(b,g)=(b,gij(b)g) (f_i\circ f_j^{-1})(b, g) = (b, g_{ij}(b)g)

for b∈Ui∩Ujb\in U_i\cap U_j and g∈Gg\in G. If bb is a point of Ui∩Uj∩UkU_i\cap U_j\cap U_k (where i,ji,j, k∈Ik\in I), one has

(fi∘fk−1)(b,e)=(b,gik(b)) (f_i\circ f_k^{-1})(b, e) = (b, g_{ik}(b))

and

(fi∘fk−1)(b,e)=(fi∘fj−1)∘(fj∘fk−1)(b,e) (f_i\circ f_k^{-1})(b, e) = (f_i\circ f_j^{-1})\circ (f_j\circ f_k^{-1})(b, e)
=(fi∘fj−1)(b,gjk(b))=(b,gij(b)gjk(b)) = (f_i\circ f_j^{-1})(b, g_{jk}(b)) = (b, g_{ij}(b)g_{jk}(b))

It follows that

gik(b)=gij(b)gjk(b) g_{ik}(b) =g_{ij}(b)g_{jk}(b)

so that the family (gij)(g_{ij}) is a cocycle on B with values in G, subordinate to the covering U\mathscr{U}. It is called the cocycle defined by the family of trivializations (fi)i∈I(f_i)_{i\in I}.

Let B, G and U\mathscr{U} be as in definition 6, and let (gij)∈(g_{ij})\in Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G) be a cocycle. For every pair (i,j)∈I×I(i, j)\in I\times I, the mapping γij:(Ui∩Uj)×G→(Ui∩Uj)×G\gamma_{ij}: (U_i\cap U_j)\times G\rightarrow (U_i\cap U_j)\times G defined by

(5) γij(b,g)=(b,gij(b)g)\gamma_{ij}(b, g) = (b, g_{ij}(b)g) for b∈Ui∩Ujb\in U_i\cap U_j and g∈Gg\in G

is an isomorphism of principal fibre spaces with base Ui∩UjU_i\cap U_j. For every triplet (i,j,k)∈I×I×I(i, j, k)\in I\times I\times I and every pair (b,g)∈(Ui∩Uj∩Uk)×G(b, g)\in (U_i\cap U_j\cap U_k)\times G, one has :

γik(b,g)=γij∘γjk(b,g) \gamma_{ik}(b, g) =\gamma_{ij}\circ \gamma_{jk}(b, g)

Let F be the topological space obtained by gluing the spaces Ui×GU_i\times G along the (Ui∩Uj)×G(U_i\cap U_j)\times G by means of the bijections γij\gamma_{ij} (TG, I, p. 16). For each i∈Ii\in I, the image of Ui×GU_i\times G in F is an open set of F (TG, I, p. 17, Prop. 9). The canonical projections pi:Ui×G→Uip_i: U_i\times G\rightarrow U_i glue together to give a continuous mapping pp of F into B. Since the mappings γij\gamma_{ij} are compatible with the right operations of G in the spaces Ui×GU_i\times G, one deduces a continuous law of operation of G on the right in F which makes F into a principal fibre space with base B and group G, endowed with a trivialization over each UiU_i. One says that F is the principal fibre space defined by the cocycle (gij)(g_{ij}).

Definition 7 01RO

Let B be a topological space, G a topological group and U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} an open covering of B. Two cocycles (gij)(g_{ij}) and (gij′)(g'_{ij}) of Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G) are said to be cohomologous if there exists a family (hi)i∈I(h_i)_{i\in I} of continuous mappings hi:Ui→Gh_i: U_i\rightarrow G such that one has

gij′(b)=hi(b)gij(b)hj(b)−1(6) g'_{ij}(b) =h_i(b)g_{ij}(b)h_j(b)^{-1} \tag{6}

for every pair (i,j)∈I×I(i, j)\in I\times I and every b∈Ui∩Ujb\in U_i\cap U_j.

The relation “(gij)(g_{ij}) is cohomologous to (gij′)(g'_{ij})” is an equivalence relation in the set Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G). We denote by Hcont1(B,U,G)H^1_{cont}(B,\mathscr{U},G) the quotient set of Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G) for this equivalence relation.

Proposition 11 01RP

Let B be a topological space, G a topological group and U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} an open covering of B.

a) Every principal fibre space over B with group G which is trivializable over U\mathscr{U} is isomorphic to a principal fibre space defined by a cocycle of Zcont1(B,U,G)Z^1_{cont}(B,\mathscr{U},G).

b) Let (E,p)(E, p) and (E′,p′)(E', p') be principal fibre spaces over B which are trivializable over U\mathscr{U}. Let (fi)i∈I((f_i)_{i\in I}( resp. (fi′)i∈I)(f'_i)_{i\in I}) be a trivialization of (E,p)(E, p) ( resp. of (E′,p′)(E', p')) adapted to U\mathscr{U}, and let us denote by (gij)(i,j)∈I×I(g_{ij})_{(i,j)\in I\times I} ( resp. (gi,j′)(i,j)∈I×I(g'_{i,j})_{(i,j)\in I\times I}) the cocycle defined by this trivialization. Then the principal fibre spaces (E,p)(E, p) and (E′,p′)(E', p') are isomorphic if and only if these cocycles are cohomologous.

Let us prove b). Let φ:E→E′\varphi : E\rightarrow E' be an isomorphism of principal fibre spaces with base B and group G. For each i∈Ii\in I, let hih_i be the continuous mapping of UiU_i into G defined by

(b,hi(b))=fi′∘φ∘fi−1(b,e) (b, h_i(b)) =f'_i\circ \varphi \circ f_i^{-1}(b, e)

Since, for each i∈I,fii\in I,f_i and fi′f'_i are compatible with the operations of G, one has for every b∈Uib\in U_i and every g∈Gg\in G,

(fi′∘φ∘fi−1)(b,g)=(b,hi(b)g) (f'_i\circ \varphi \circ f_i^{-1})(b, g) = (b, h_i(b)g)

and

(fi∘φ−1∘(fi′)−1)(b,g)=(b,hi(b)−1g) (f_i\circ \varphi^{-1}\circ (f'_i)^{-1})(b, g) = (b, h_i(b)^{-1}g)

Hence, for every pair (i,j)∈I×I(i, j)\in I\times I and every point bb of Ui∩Uj:U_i\cap U_j: fi′∘(fj′)−1(b,e)=(fi′∘φ∘fi−1)∘(fi∘fj−1)∘(fj∘φ−1∘(fj′)−1)(b,e)f'_i\circ (f'_j)^{-1}(b, e) = (f'_i\circ \varphi \circ f_i^{-1})\circ (f_i\circ f_j^{-1})\circ (f_j\circ \varphi^{-1}\circ (f'_j)^{-1})(b, e)

=(b,hi(b)gij(b)hj(b)−1) = (b, h_i(b)g_{ij}(b)h_j(b)^{-1})

so that one has

gij′(b)=hi(b)gij(b)hj(b)−1 g'_{ij}(b) =h_i(b)g_{ij}(b)h_j(b)^{-1}

This proves that the cocycles (gij)(g_{ij}) and (gij′)(g'_{ij}) are cohomologous.

Conversely, suppose that these cocycles are cohomologous, and let (hi)i∈I(h_i)_{i\in I} be a family of continuous mappings hi:Ui→Gh_i: U_i\rightarrow G such that gij′(b)=hi(b)gij(b)hj(b)−1g'_{ij}(b) =h_i(b)g_{ij}(b)h_j(b)^{-1} for i,j∈Ii,j\in I and b∈Ui∩Ujb\in U_i\cap U_j. For

i∈Ii\in I, let φi:p−1(Ui)→(−p′1)(Ui)\varphi_i:\overset{-1}{p}(U_i)\rightarrow (^-{p'}^1)(U_i) be the mapping defined by fi′∘φi∘f'_i\circ \varphi_i\circ fi−1(b,g)=(b,hi(b)g)f_i^{-1}(b, g) = (b, h_i(b)g), for b∈Uib\in U_i and g∈Gg\in G. It is an isomorphism of principal fibre spaces with basis UiU_i and group G. For (i,j)∈(i, j)\in I×II\times I, let us denote by γij\gamma_{ij} and γij′\gamma '_{ij} the gluing homeomorphisms associated as above with the cocycles (gij)(g_{ij}) and (gij′)(g'_{ij}) respectively (I, p. 115, formula (5)), so that fi(b,g)=γij∘fj(b,g)f_i(b, g) =\gamma_{ij}\circ f_j(b, g) and fi′(b,g)=γij′∘fj′(b,g)f'_i(b, g) =\gamma '_{ij}\circ f'_j(b, g) for every (b,g)∈(Ui∩Uj)×G(b, g)\in (U_i\cap U_j)\times G. Consequently, for (i,j)∈I×I(i, j)\in I\times I and (b,g)∈(Ui∩Uj)×G(b, g)\in (U_i\cap U_j)\times G, we have the relations

fi′∘φj∘fi−1(b,g)=γij′∘(fj′∘φj∘fj−1)(b,gij(b)−1g) f'_i\circ \varphi_j\circ f_i^{-1}(b, g) =\gamma '_{ij}\circ (f'_j\circ \varphi_j\circ f_j^{-1})(b, g_{ij}(b)^{-1}g)
=γij′(b,hj(b)gij(b)−1g) =\gamma '_{ij}(b, h_j(b)g_{ij}(b)^{-1}g)
=(b,gij′(b)hj(b)gij(b)−1g) = (b, g'_{ij}(b)h_j(b)g_{ij}(b)^{-1}g)
=(b,hi(b)g)=fi′∘φi∘fi−1(b,g) = (b, h_i(b)g) =f'_i\circ \varphi_i\circ f_i^{-1}(b, g)

This proves that φi\varphi_i and φj\varphi_j coincide on p−1(Ui∩Uj)\overset{-1}{p}(U_i\cap U_j). The morphisms φi\varphi_i therefore glue together to a B-morphism of principal fibre spaces from E to E′E'. Assertion b) follows, since every morphism of principal fibre spaces with basis B and group G is an isomorphism (I, p. 93, Prop. 1).

Let us now prove a). Let (E,p)(E, p) be a principal fibre space with group G endowed, for each i∈Ii\in I, with a trivialization fi:p−1(Ui)→f_i:\overset{-1}{p}(U_i)\rightarrow Ui×GU_i\times G over UiU_i. Let (gij)(g_{ij}) be the cocycle defined by this family, and let F then be the principal fibre space defined by the cocycle (gij)(g_{ij}). By construction, the principal fibre space F is endowed with a trivialization over U\mathscr{U}; this trivialization defines the cocycle (gij)(g_{ij}). By assertion b), the principal fibre space (E,p)(E, p) is isomorphic to F.

Let B be a topological space and G a topological group. Let (E,p)(E, p) be a principal fibre space with basis B and group G. Let ss be a section of the surjective mapping pp (E, II, p. 19, Prop. 8). The mapping f:B×G→Ef: B\times G\rightarrow E defined by f(b,g)=s(b)⋅gf(b, g) =s(b)\cdot g is a bijection compatible with the action of G. Endow B×GB\times G with the topology obtained by transport of structure; (B×G(B\times G, pr1)_1) is then a principal fibre space with basis B and group G isomorphic to E. There therefore exists a set T of principal fibre spaces with basis B and group G such that every principal fibre space with basis B and group G is isomorphic to an element of T. Let us denote by P(B,G)P(B,G) the set of isomorphism classes of principal fibre spaces with basis B and group G (E, II, p. 47).

Let U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} be an open covering of B. Let P(B,U,G)P(B,\mathscr{U},G) denote the subset of P(B,G)P(B,G) consisting of the isomorphism classes of principal bundles which are trivializable over U\mathscr{U}. By proposition 11, there exists a mapping rUr_{\mathscr{U}} of Hcont1(B,U,G)H^1_{cont}(B,\mathscr{U},G) into P(B,U,G)P(B,\mathscr{U},G) which associates with the class of a cocycle (gij)∈Zcont1(B,U,G)(g_{ij})\in Z^1_{cont}(B,\mathscr{U},G) the isomorphism class of the principal fibre space defined by this cocycle; it is a bijection (loc. cit.). The inverse bijection sUs_{\mathscr{U}} associates with the isomorphism class in Hcont1(B,U,G)H^1_{cont}(B,\mathscr{U},G) of a principal fibre space E, trivializable over U\mathscr{U}, the class of the cocycle defined by an arbitrary family of trivializations of E over U\mathscr{U}. Under this correspondence, the isomorphism class of the trivial principal fibre space B×GB\times G corresponds to the cohomology class of the constant cocycle gij=eg_{ij}=e, also called the trivial cocycle.

By virtue of the definition of a principal fibre space, the set P(B,G)P(B,G) is the union of the sets of the form P(B,U,G)P(B,\mathscr{U},G), where U\mathscr{U} is an open covering of B.

Let U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} be an open covering of B and let V=(Vk)k∈K\mathscr{V}= (V_k)_{k\in K} be an open covering of B finer than U\mathscr{U}. We have P(B,U,G)⊂P(B,\mathscr{U},G)\subset P(B,V,G)P(B,\mathscr{V},G); let iVUi_{\mathscr{V} \mathscr{U}} denote the canonical injection defined by this inclusion. Choose a mapping φ:K→I\varphi : K\rightarrow I such that Vk⊂Uφ(k)V_k\subset U_{\varphi(k)} for every k∈Kk\in K. Given a cocycle (gij)∈Zcont1(B,U,G)(g_{ij})\in Z^1_{cont}(B,\mathscr{U},G), set, for every pair (k,ℓ)∈K×K,g‾kℓ=gφ(k)φ(ℓ)∣Vk∩Vℓ(k, \ell )\in K\times K,\overline{g}_{k\ell}=g_{\varphi(k)\varphi(\ell)}|V_k\cap V_{\ell}.

The family (g‾kℓ)(\overline{g}_{k\ell}) is a cocycle on B, with values in G, subordinate to V\mathscr{V}. If (gij′)∈Zcont1(B,U,G)(g'_{ij})\in Z^1_{cont}(B,\mathscr{U},G) is a cocycle cohomologous to the cocycle (gij)(g_{ij}), the cocycle (g‾kℓ′)(\overline{g}'_{k\ell}) deduced from (gkℓ′)(g'_{k\ell}) is cohomologous to the cocycle (g‾kℓ)(\overline{g}_{k\ell}). It follows that there is a mapping

c(φ):Hcont1(B,U,G)→Hcont1(B,V,G) c(\varphi ): H^1_{cont}(B,\mathscr{U},G)\rightarrow H^1_{cont}(B,\mathscr{V},G)

which associates with the class of (gij)(g_{ij}) the class of (g‾kℓ)(\overline{g}_{k\ell}).

Let E be a principal fibre space with base B and group G and, for every i∈i\in I, let fi:EUi→Ui×Gf_i: E_{U_i}\rightarrow U_i\times G be a trivialization of the principal UiU_i-bundle EUiE_{U_i}. For every k∈k\in K, let fk′:EVk→Vk×Gf'_k: E_{V_k}\rightarrow V_k\times G be the trivialization deduced from fφ(k)f_{\varphi(k)} by passing to subsets. Let (gij)∈Zcont1(B,U,G)(g_{ij})\in Z^1_{cont}(B,\mathscr{U},G) be the cocycle defined by the family (fi)(f_i); the cocycle defined by the family (fk′)(f'_k) is precisely the cocycle (g‾kℓ)(\overline{g}_{k\ell}) defined above. Thus, the following diagram is commutative:

Hcont1(B,U,G)c(φ)Hcont1(B,V,G) H^1_{cont}(B,\mathscr{U},G)^{c(\varphi)}H^1_{cont}(B,\mathscr{V},G)
rUrVr_{\mathscr{U}}r_{\mathscr{V}}
P(B,U,G)iVUP(B,V,G) P(B,\mathscr{U},G)^{i_{\mathscr{V} \mathscr{U}}}P(B,\mathscr{V},G)

Since the mappings rUr_{\mathscr{U}} and rVr_{\mathscr{V}} are bijective, the mapping c(φ)c(\varphi ), analogously to iVUi_{\mathscr{V} \mathscr{U}}, is an injection and does not depend on the choice of φ\varphi. Henceforth we shall write cVUc_{\mathscr{V} \mathscr{U}} instead of c(φ)c(\varphi ).

Let R\mathscr{R} be the set of elements of P(P(B))\mathfrak{P}(\mathfrak{P}(B)) which are open coverings of B. For every open covering U\mathscr{U} of B, there exists an open covering V\mathscr{V} belonging to R\mathscr{R} such that U\mathscr{U} is both finer and coarser than U\mathscr{U}. The set R\mathscr{R} is ordered and filtering for the relation ⩽\leqslant defined by U⩽V\mathscr{U}\leqslant \mathscr{V} if V\mathscr{V} is a covering finer than U\mathscr{U}. It follows from the foregoing that an inductive system (Hcont1(B,U,G),cVU)(H^1_{cont}(B,\mathscr{U},G), c_{\mathscr{V} \mathscr{U}}) relative to the filtered ordered set R\mathscr{R} has been defined and that the family (rU)(r_{\mathscr{U}}) is an inductive system of bijective mappings of (Hcont1(B,U,G),cVU)(H^1_{cont}(B,\mathscr{U},G), c_{\mathscr{V} \mathscr{U}}) into (P(B,U,G),iVU)(P(B,\mathscr{U},G), i_{\mathscr{V} \mathscr{U}}). If we denote by Hcont1(B,G)H^1_{cont}(B,G) the inductive limit of the system (Hcont1(B,U,G),cVU)(H^1_{cont}(B,\mathscr{U},G), c_{\mathscr{V} \mathscr{U}}) and by r:Hcont1(B,G)→P(B,G)r: H^1_{cont}(B,G)\rightarrow P(B,G) the inductive limit of the family (rU)(r_{\mathscr{U}}), we therefore have:

Theorem 4 01RQ

The mapping r:Hcont1(B,G)→P(B,G)r: H^1_{cont}(B,G)\rightarrow P(B,G) is bijective.

Let U=(Ui)i∈I\mathscr{U}= (U_i)_{i\in I} be an open covering of B; let us denote by cUc_{\mathscr{U}} the canonical mapping Hcont1(B,U,G)→Hcont1(B,G)H^1_{cont}(B,\mathscr{U},G)\rightarrow H^1_{cont}(B,G). If (gij)(g_{ij}) is a cocycle on B, with values in G, subordinate to the open covering U\mathscr{U}, the element cU((gij))c_{\mathscr{U}}((g_{ij})) of Hcont1(B,G)H^1_{cont}(B,G) is called the cohomology class of the cocycle (gij)(g_{ij}).

Exercises

See the exercises for § 5.

Machine translation of the English, by gpt-5.4, against glossary version 34. Not checked by a person.