Bourbaki

§ 5. Homotopy and coverings (case of locally arcwise connected spaces)

1. Homotopic condition for lifting continuous mappings

Proposition 1 01ZL

Let B be a topological space and let (E,p)(E, p) be a covering of B. Let Y be a topological space and let f:Y→Bf: Y\rightarrow B be a continuous mapping. Let y∈y\in Y,x∈,x\in E,b∈,b\in B be points such that f(y)=p(x)=bf(y) =p(x) =b. Suppose the space Y connected and locally arcwise connected. In order that there exist a continuous lifting g:Y→Eg: Y\rightarrow E of ff such that g(y)=xg(y) =x, it is necessary and sufficient that the image of the homomorphism π1(f,y):π1(Y,y)→π1(B,b)\pi_1(f, y):\pi_1(Y, y)\rightarrow \pi_1(B, b) be contained in the image of the homomorphism π1(p,x):π1(E,x)→π1(B,b)\pi_1(p, x):\pi_1(E, x)\rightarrow \pi_1(B, b).

The condition is necessary without any assumption on the space Y. In fact, if such a lifting gg exists, we have π1(f,y)=π1(p,x)∘π1(g,y)\pi_1(f, y) =\pi_1(p, x)\circ \pi_1(g, y).

Let us prove that it is sufficient. Let s:Λb(B)→Λx(E)s: \Lambda_b(B)\rightarrow \Lambda_x(E) denote the homeomorphism inverse to the homeomorphism c↦p∘c(c\mapsto p\circ c(III, p. 302, cor. 2 of prop. 3) and let φ:Λy(Y)→Λx(E)\varphi : \Lambda_y(Y)\rightarrow \Lambda_x(E) be the mapping d↦d\mapsto s(f∘d)s(f\circ d). The mapping φ\varphi is continuous (I, p. 132, lemma).

Let dd and d′∈Λy(Y)d'\in \Lambda_y(Y) be paths with origin yy having the same term; let us prove that the paths φ(d)\varphi (d) and φ(d′)\varphi (d') have the same term. Put c=f∘d,c′=f∘d′c=f\circ d,c'=f\circ d'. Since the path d∗d′‾d*\overline{d'} is a loop in Y at yy, the path c∗c′‾c*\overline{c'} is a loop in B at bb and its class belongs to the image of the homomorphism π1(f,y)\pi_1(f, y), hence to the image of the homomorphism π1(p,x)\pi_1(p, x) by assumption. By cor. 2 of prop. 4 (III, p. 303), the path s(c∗c′‾)s(c*\overline{c'}) is a loop in E at xx. The same is true of the path s(c′∗c‾)s(c'*\overline{c}) which, by uniqueness of the lifting of paths, is equal to s(c∗c′‾)s(c*\overline{c'}). Therefore s(c′∗c‾)(12)=s(c′)(1)=s(c)(1)s(c'*\overline{c})(\frac{1}{2}) =s(c')(1) =s(c)(1), as was to be proved.

Let us denote by eE:Λx(E)→Ee_E: \Lambda_x(E)\rightarrow E and eY:Λy(Y)→Ye_Y: \Lambda_y(Y)\rightarrow Y respectively the term mappings. Since the space Y is supposed connected and locally arcwise connected, the mapping eYe_Y is surjective and open (III, p. 262, prop. 10). By the preceding paragraph, there exists a unique mapping g:Y→Eg: Y\rightarrow E such that eE∘φ=g∘eYe_E\circ \varphi =g\circ e_Y. It is continuous, for the mapping eYe_Y is strict ( I, p. 18, example 2).

Let us verify finally that the mapping gg lifts the mapping ff and that g(y)=xg(y) =x. Every point zz of Y is the term of a path cc with origin yy. The path φ(c)\varphi (c) is a lifting of f∘cf\circ c with origin xx and with term the point g(z)g(z). Therefore p(g(z))=f(z)p(g(z)) =f(z). For z=yz=y and c=eyc=e_y, we have φ(ey)=ex\varphi (e_y) =e_x, whence g(y)=xg(y) =x.

Corollary 1 01ZM

Let B be a connected topological space locally connected by paths, and let bb be a point of B. In order that a covering (E,p)(E, p) of B be trivializable, it is necessary and sufficient that, for every point xx of the fibre EbE_b, the homomorphism π1(p,x)\pi_1(p, x) be bijective.

Recall that the homomorphism π1(p,x)\pi_1(p, x) is injective (III, p. 303, cor. 1 of prop. 4). The statement therefore follows from prop. 1 and I, p. 81, cor. 4 of prop. 6.

Remark 01ZN

Let B be a topological space locally connected by paths, let bb be a point of B, and let V be an open connected neighbourhood of bb such that the image of the homomorphism from π1(V,b)\pi_1(V, b) into π1(B,b)\pi_1(B, b) is the subgroup reduced to the identity element. By cor. 1, every covering of B is trivializable over V, and fortiori over every subset of B contained in V.

Corollary 2 01ZO

Let B be a connected topological space locally connected by paths. If, for a point bb of B, the group π1(B,b)\pi_1(B, b) is reduced to the identity element, the space B is simply connected.

In fact, it follows from corollary 1 that, under these hypotheses, every covering of B is trivializable.

Corollary 3 01ZP

Let

E′f′{E'}^{f'} E

p′pp'p

B′f{B'}^f B

be a cartesian square. Suppose that E is a covering of B and that the space B′B' is connected and locally connected by paths. Let b′b' be a point of B′B' and b=f(b′)b=f(b'). In order that (E′,p′)(E', p') be a trivializable covering of B′B', it is necessary and sufficient that, for every point xx of the fibre EbE_b, the image of π1(p,x)\pi_1(p, x) contain the image of π1(f,b)\pi_1(f, b).

This follows from prop. 1 and I, p. 81, cor. 5 of prop. 6.

2. Operations of the Poincaré group and morphisms of coverings

Let B be a connected topological space locally connected by paths and let bb be a point of B.

Let E be a covering of B. Since the space B is assumed connected, the fibre EbE_b is not empty if E is not empty (I, p. 74, prop. 4). By assertion a) of theorem 1 of III, p. 305, the covering E is connected and nonempty if and only if the group π1(B,b)\pi_1(B, b) operates transitively on EbE_b.

Proposition 2 01ZQ

Let E and E′E' be coverings of B.

a) Let f:E→E′f: E\rightarrow E' be a B-morphism. The mapping fb:Eb→Eb′f_b: E_b\rightarrow E'_b derived from ff is compatible with the operations of π1(B,b)\pi_1(B, b) on EbE_b and Eb′E'_b respectively. It is bijective if and only if ff is an isomorphism.

b) The mapping f↦fbf\mapsto f_b is a bijection of the set CB(E;E′)\mathscr{C}_B(E; E') of B-morphisms from E into E′E' onto the set Fπ1(B,b)(Eb;Eb′)\mathscr{F}_{\pi_1(B,b)}(E_b; E'_b) of π1(B,b)\pi_1(B, b)-morphisms from EbE_b into Eb′E'_b.

If ff is a B-morphism from E into E′E', the mapping fb:Eb→Eb′f_b: E_b\rightarrow E'_b is a π1(B,b)\pi_1(B, b)-morphism (cf. III, p. 305). Moreover, two B-morphisms from E into E′E' which coincide on the fibre EbE_b are equal (I, p. 80, cor. 3 of prop. 6).

Let φ:Eb→Eb′\varphi : E_b\rightarrow E'_b be a morphism of π1(B,b)\pi_1(B, b)-sets; let us prove that there exists a B-morphism ff of E into E′E' such that fb=φf_b=\varphi. We may suppose that the spaces E and E′E' are connected and non-empty, so that EbE_b and Eb′E'_b are homogeneous π1(B,b)\pi_1(B, b)-sets. Let xx be a point of EbE_b; its stabilizer G is the image of the mapping π1(p,x)\pi_1(p, x) (III, p. 305, Theorem 1). Since the mapping φ\varphi is a π1(B,b)\pi_1(B, b)-morphism, the group G fixes the point x′=φ(x)x'=\varphi (x) of Eb′E'_b, and is therefore contained in the image of the mapping π1(p′,x′)\pi_1(p', x'). By Prop. 1 of III, p. 308, there exists a B-morphism ff of E into E′E' such that f(x)=x′f(x) =x'. The mappings fbf_b and φ\varphi are π1(B,b)\pi_1(B, b)-morphisms of the homogeneous π1(B,b)\pi_1(B, b)-set EbE_b into Eb′E'_b which agree at the point xx; they are therefore equal.

If f:E→E′f: E\rightarrow E' is a B-isomorphism, the mapping fb:Eb→Eb′f_b: E_b\rightarrow E'_b is bijective. Conversely, suppose that the mapping fbf_b is bijective, and let g:E′→g: E'\rightarrow E be a B-morphism such that gb=(fb)−1g_b= (f_b)^{-1}. The B-morphism g∘f:E→Eg\circ f: E\rightarrow E induces on EbE_b the identical mapping; it therefore follows from the proposition that g∘f=g\circ f= IdE_E. Analogously, f∘g=f\circ g= IdE′_{E'}, which proves that ff is a B-isomorphism.

Corollary 1 01ZR

The coverings E and E′E' are isomorphic if and only if the π1(B,b)\pi_1(B, b)-sets EbE_b and Eb′E'_b are isomorphic.

Corollary 2 01ZS

Let (E,p)(E, p) and (E′,p′)(E', p') be connected coverings of B. Let xx be a point of EbE_b and x′x' a point of Eb′E'_b. In order that there exist a B-morphism g:E→E′g: E\rightarrow E' such that g(x)=x′g(x) =x', it is necessary and sufficient that one have p∗(π1(E,x))⊂p∗′(π1(E′,x′))p_*(\pi_1(E, x))\subset p'_*(\pi_1(E', x')). Such a morphism is then unique, and is an isomorphism if and only if the subgroups p∗(π1(E,x))p_*(\pi_1(E, x)) and p∗′(π1(E′,x′))p'_*(\pi_1(E', x')) of π1(B,b)\pi_1(B, b) are equal.

According to III, p. 305, Theorem 1, the mapping of π1(B,b)\pi_1(B, b) onto EbE_b defined by γ↦x⋅γ\gamma \mapsto x\cdot \gamma induces by passing to the quotient an isomorphism of the π1(B,b)\pi_1(B, b)-set p∗(π1(E,x))\π1(B,b)p_*(\pi_1(E, x))\backslash \pi_1(B, b) onto EbE_b. Analogously, there exists a unique isomorphism of π1(B,b)\pi_1(B, b)-sets of p∗(π1(E′,x′))\π1(B,b)p_*(\pi_1(E', x'))\backslash \pi_1(B, b) onto Eb′E'_b. According to Proposition 2, there exists a B-morphism g:E→E′g: E\rightarrow E' such that g(x)=x′g(x) =x' if and only if there exists a morphism of π1(B,b)\pi_1(B, b)-sets of p∗(π1(E,x))\π1(B,b)p_*(\pi_1(E, x))\backslash \pi_1(B, b) onto p∗′(π1(E′,x′))\π1(B,b)p'_*(\pi_1(E', x'))\backslash \pi_1(B, b) which sends the class p∗(π1(E,x))p_*(\pi_1(E, x)) onto the class p∗′(π1(E′,x′))p'_*(\pi_1(E', x')). Such a morphism exists if and only if one has p∗(π1(E,x))⊂p∗′(π1(E′,x′))p_*(\pi_1(E, x))\subset p'_*(\pi_1(E', x')). It is then unique, since the space E is assumed connected (I, p. 34, cor. 2 of prop. 11). It is an isomorphism if and only if the subgroups p∗(π1(E,x))p_*(\pi_1(E, x)) and p∗′(π1(E′,x′))p'_*(\pi_1(E', x')) of π1(B,b)\pi_1(B, b) are equal.

Corollary 3 01ZT

Let (E,p)(E, p) be a connected covering of B and let xx be a point of the fibre EbE_b. Let N denote the normalizer of p∗(π1(E,x))p_*(\pi_1(E, x)) in π1(B,b)\pi_1(B, b). For every element γ\gamma of N, there exists a unique B-automorphism gg of E such that g(x)=x⋅γg(x) =x\cdot \gamma, and the mapping γ↦g\gamma \mapsto g defines by passing to the quotient a group isomorphism of N/p∗(π1(E,x))N/p_*(\pi_1(E, x)) onto AutB(E)_B(E).

Let γ∈π1(B,b)\gamma \in \pi_1(B, b); one has p∗(π1(E,x))=p_*(\pi_1(E, x)) = Int(γ\gamma )(p∗(π1(E,x⋅γ)))(p_*(\pi_1(E, x\cdot \gamma ))) (III, p. 305, Theorem 1, c)). According to Corollary 2, in order that there should exist a B-automorphism gg of E such that g(x)=x⋅γg(x) =x\cdot \gamma, it is necessary and sufficient that γ\gamma belong to N. Such an isomorphism is then unique. Let α:N→\alpha : N\rightarrow AutB(E)_B(E) denote the mapping γ↦g\gamma \mapsto g thus defined.

Let γ\gamma and γ′\gamma ' be two elements of N. Put g=α(γ),g′=α(γ′)g=\alpha (\gamma ),g'=\alpha (\gamma '); one has g(g′(x))=g(x⋅γ′)=g(x)⋅γ′=(x⋅γ)⋅γ′=x⋅(γγ′)g(g'(x)) =g(x\cdot \gamma ') =g(x)\cdot \gamma '= (x\cdot \gamma )\cdot \gamma '=x\cdot (\gamma \gamma ') according to relations (4), III, p. 305 and (1), p. 304. Consequently, α\alpha is a group homomorphism. In order that α(γ)=\alpha (\gamma ) = IdE_E, it is necessary and sufficient that one have x⋅γ=xx\cdot \gamma =x, by virtue of the uniqueness of α(γ)\alpha (\gamma ), that is to say γ∈p∗(π1(E,x))\gamma \in p_*(\pi_1(E, x)) (III, p. 305, Theorem 1, b)). Finally, if gg is a B-automorphism of E, there exists a path cc joining xx to g(x)g(x) in E (which is arcwise connected), and one then has g(x)=x⋅γg(x) =x\cdot \gamma, where γ\gamma is the class of the path p∘cp\circ c. This proves that the homomorphism α\alpha is surjective.

The group AutB(E)_B(E) operates on the fibre EbE_b and is identified with the automorphism group of the homogeneous (right) π1(B,b)\pi_1(B, b)-set Eb(E_b(cf. A, I, p. 56, prop. 5 and 6).

Corollary 4 01ZU

Let (E,p)(E, p) be a connected covering of B and let xx be a point of the fibre EbE_b. In order that E be a Galois covering of B, it is necessary and sufficient that p∗(π1(E,x))p_*(\pi_1(E, x)) be a distinguished subgroup of π1(B,b)\pi_1(B, b). The group AutB(E)_B(E) is then isomorphic to the quotient group π1(B,b)/p∗(π1(E,x))\pi_1(B, b)/p_*(\pi_1(E, x)).

If p∗(π1(E,x))p_*(\pi_1(E, x)) is a distinguished subgroup of π1(B,b)\pi_1(B, b), the group AutB(E)_B(E) operates transitively on the fibre EbE_b by Corollary 3, so that E is a Galois covering of B (I, p. 102, th. 2). The converse is assertion d) of Theorem 1 (III, p. 305). The last assertion follows from Corollary 3.

3. Operations without local monodromy of the Poincaré groupoid

Lemma 1 01ZV

Let B be a topological space locally connected by paths, let p:E→Bp: E\rightarrow B and p′:E′→Bp': E'\rightarrow B be étale and separated mappings satisfying the path-lifting property (III, p. 302). Let g:E→E′g: E\rightarrow E' be a mapping such that p′∘g=pp'\circ g=p. Suppose that gg is compatible with the canonical operations of the groupoid ϖ(B)\varpi (B) on E and E′E'. Then gg is continuous.

Let cc be a path in E; put c′=g∘cc'=g\circ c and prove that the mapping c′c' is continuous. Let dd denote the path p∘cp\circ c in B and, for every t∈It\in \mathbf{I}, let dtd_t denote the path s↦d(st)s\mapsto d(st). For every t∈It\in \mathbf{I}, we have c(t)=c(0)⋅dt(c(t) =c(0)\cdot d_t(III, p. 304, remark 3), whence c′(t)=c′(0)⋅dtc'(t) =c'(0)\cdot d_t by the hypothesis made on gg, which proves that c′c' is a path lifting the path dd, by the same remark. This proves that the mapping gg is pathwise continuous. Since the space E is locally connected by paths (III, p. 261, cor. 2), the mapping gg is continuous (III, p. 269, corollary to prop. 13).

Let B be a topological space. Consider an operation φ\varphi = (φa,b)(a,b)∈B×B(\varphi_{a,b})_{(a,b)\in B\times B} of the groupoid ϖ(B)\varpi (B) on a set E, relative to a mapping p:E→Bp: E\rightarrow B. One says that ϖ(B)\varpi (B) operates without monodromy on E (cf. II, p. 168) if for every b∈b\in B and every class of loops γ∈π1(B,b)\gamma \in \pi_1(B, b), the action of γ\gamma on the fibre EbE_b is trivial. If B is connected by paths, it is enough that this be so for one point of B (loc. cit.). We shall say that the operation φ\varphi of the groupoid ϖ(B)\varpi (B) is without local monodromy if every point of B possesses a neighbourhood V such that ϖ(V)\varpi (V) operates without monodromy on the set EVE_V = p−1(V)\overset{-1}{p}(V) relative to the mapping pV=p∣p−1(V)p_V=p|\overset{-1}{p}(V).

Remark 01ZW

Let B be a topological space locally connected by paths and suppose that every point bb of B possesses a neighbourhood V such that the image of π1(V,b)\pi_1(V, b) in π1(B,b)\pi_1(B, b) reduces to the identity element (cf. IV, p. 340, def. 2). Then every operation of the groupoid ϖ(B)\varpi (B) is without local monodromy.

In fact, let us consider a set E, a mapping p:E→Bp: E\rightarrow B and an operation φ\varphi of the groupoid ϖ(B)\varpi (B) on E relative to pp. Let aa be a point of B and let V be a neighbourhood of aa such that the image of π1(V,a)\pi_1(V, a) in π1(B,a)\pi_1(B, a) is reduced to the identity element. Let U be a path-connected neighbourhood of aa contained in V. Let bb be a point of U and let γ∈π1(U,b)\gamma \in \pi_1(U, b). Let δ\delta be the class of a path joining aa to bb in U. Then δγδ−1\delta \gamma \delta^{-1} is the class of a loop at aa in U; its image in π1(B,a)\pi_1(B, a) is therefore trivial. Thus φa,a(δγδ−1)=\varphi_{a,a}(\delta \gamma \delta^{-1}) = IdEa_{E_a}, whence φb,b(γ)=\varphi_{b,b}(\gamma ) = IdEb_{E_b}. This proves that the operation of ϖ(U)\varpi (U) on the U-space EUE_U is without monodromy. Consequently, the operation φ\varphi is without local monodromy.

Proposition 3 01ZX

Let B be a topological space locally path-connected and let E be a set endowed with an operation without local monodromy φ\varphi of the groupoid ϖ(B)\varpi (B), relative to a mapping p:E→Bp: E\rightarrow B. Then there exists on E a unique topology for which the following conditions are satisfied:

(i) The set E endowed with this topology and the mapping pp is a covering of B;

(ii) The canonical operation of ϖ(B)\varpi (B) on this covering is identical with the operation φ\varphi.

The uniqueness of such a topology follows from lemma 1 of III, p. 312, where one takes for gg the identical mapping of E. To prove its existence, one may moreover suppose that B is connected and nonempty.

Let us suppose first of all that the operation φ=(φa,b)(a,b)∈B×B\varphi = (\varphi_{a,b})_{(a,b)\in B\times B} of the groupoid ϖ(B)\varpi (B) on the set E, relative to pp, is without monodromy.

Let aa be a point of B. For every point b∈Bb\in B, there exists a path cc in B joining aa to bb. If γ∈ϖa,b(B)\gamma \in \varpi_{a,b}(B) denotes the class of the path cc, the bijection φa,b(γ):Ea→Eb\varphi_{a,b}(\gamma ): E_a\rightarrow E_b is independent of the path cc, since the operation is without monodromy; let fa,bf_{a,b} denote this bijection. The mapping Φa:B×Ea→\Phi_a: B\times E_a\rightarrow E defined by (b,x)↦fa,b(x)(b, x)\mapsto f_{a,b}(x) is a bijection; the inverse bijection associates with x∈Ex\in E the pair (p(x),fp(x),a(x))(p(x), f_{p(x),a}(x)). Endow the set EaE_a with the discrete topology, so that the B-space B×EaB\times E_a is a trivial covering of B. By transport of structure, the bijection Φa\Phi_a endows E with a topology which makes it into a covering of B.

Let us prove that the canonical operation of ϖ(B)\varpi (B) on E is identical with the operation φ\varphi. Let xx be a point of EaE_a and bb a point of B. Let cc be a path in B joining the point aa to the point bb; the mapping t↦Φa(c(t),x)t\mapsto \Phi_a(c(t), x) is then a path in E joining the point xx to the point Φa(b,x)=fa,b(x)\Phi_a(b, x) =f_{a,b}(x). If γ∈ϖa,b(B)\gamma \in \varpi_{a,b}(B) denotes the class of the path cc, we thus have x⋅γ=fa,b(x)x\cdot \gamma =f_{a,b}(x). This proves that the canonical operation of ϖ(B)\varpi (B) on the covering E and the operation φ\varphi coincide on the classes of paths with origin aa. Since these classes generate the groupoid ϖ(B)\varpi (B), the two operations are equal.

Let us now treat the general case. Let B\mathscr{B} be the set of open subsets V of E such that ϖ(V)\varpi (V) operates without monodromy on p−1(V)\overset{-1}{p}(V). By hypothesis, the elements of B\mathscr{B} cover B. From the preceding, there exists for every V∈BV\in \mathscr{B} a unique topology on the set p−1(V)\overset{-1}{p}(V) such that (p−1(V),pV)(\overset{-1}{p}(V), p_V) is a covering of V and that the canonical operation of ϖ(V)\varpi (V) on this covering coincides with the operation induced by φ\varphi.

Let V and V′∈BV'\in \mathscr{B}. The topology on p−1(V∩V′)\overset{-1}{p}(V\cap V') induced by the topology of p−1(V)\overset{-1}{p}(V) (resp. of p−1(V′)\overset{-1}{p}(V')) defined above makes it into a covering of V∩V′V\cap V' on which the canonical operation of ϖ(V∩V′)\varpi (V\cap V') is induced by the operation φ\varphi. These topologies therefore coincide with that of p−1(V∩V′)\overset{-1}{p}(V\cap V'). There then exists a unique topology on E inducing on each p−1(V)\overset{-1}{p}(V) the topology previously defined (cf. I, § 2, p. 16).

When E is endowed with this topology, the mapping pp is continuous and the B-space E is a covering. The canonical operation of ϖ(B)\varpi (B) on this covering coincides with the operation φ\varphi on the classes of paths whose image is contained in one of the open sets of B\mathscr{B}. By lemma 4 of III, p. 272, these classes generate the groupoid ϖ(B)\varpi (B). It follows that these two operations are equal (II, p. 167).

Corollary 01ZY

Let B be a connected topological space, locally connected by arcs, let (E,p)(E, p) be a B-space whose projection pp is etale, separated, and has the path-lifting property. If the canonical operation of ϖ(B)\varpi (B) on the set E, relative to pp, is without local monodromy, then E is a covering of B.

Let us endow E with the canonical operation of ϖ(B)\varpi (B) defined by path lifting (III, p. 303, no3^o3). There exists a topology on E for which conditions (i) and (ii) of prop. 3 are satisfied. This topology coincides with the given topology on E by lemma 1 of III, p. 312.

4. Admissible topology of Poincaré groups

Let B be a topological space and let aa be a point of B. A subgroup H of π1(B,a)\pi_1(B, a) is said to be admissible if every point bb of B has a neighbourhood V such that one has γi∗(δ)γ−1∈H\gamma i_*(\delta )\gamma^{-1}\in H for every γ∈ϖa,b(B)\gamma \in \varpi_{a,b}(B) and every δ∈π1(V,b)\delta \in \pi_1(V, b), where i:V→i: V\rightarrow B is the canonical injection. If H is a normal subgroup of π1(B,a)\pi_1(B, a), it is enough, in order that H be admissible, that this condition be satisfied for one class of paths γ∈ϖa,b(B)\gamma \in \varpi_{a,b}(B).

Proposition 4 01ZZ

There exists a unique topology on π1(B,a)\pi_1(B, a), compatible with its group structure, for which the admissible normal subgroups of π1(B,a)\pi_1(B, a) form a fundamental system of neighbourhoods of the identity element. For this topology, the open subgroups are exactly the admissible subgroups.

The following facts result from the definition of an admissible subgroup:

a) The group π1(B,a)\pi_1(B, a) is admissible;

b) A subgroup containing an admissible subgroup is admissible;

c) The intersection of a finite family of admissible subgroups is admissible;

d) For every admissible subgroup H of π1(B,a)\pi_1(B, a) the intersection of the subgroups γHγ−1\gamma H\gamma^{-1}, when γ\gamma runs through π1(B,a)\pi_1(B, a), is admissible. In particular, the set of admissible normal subgroups is a filter basis formed of subgroups of π1(B,a)\pi_1(B, a) satisfying axiom (GVIII′'_{III}) of III, p. 4, whence the first part of the proposition after TG, III, p. 5, example. The second part is then immediate (cf. TG, III, p. 7, corollary of prop. 4).

The topology on the group π1(B,a)\pi_1(B, a) characterized in proposition 4 is called the admissible topology.

Remark 1 0200

If B0B_0 denotes the path-connected component of aa in B, the canonical isomorphism π1(B0,a)→π1(B,a)\pi_1(B_0, a)\rightarrow \pi_1(B, a) is an isomorphism of topological groups when these groups are endowed with the admissible topology.

Remark 2 0201

Let B be a topological space. Let bb and b′b' be points of B which belong to the same path-connected component and let γ\gamma be an element of ϖb,b′(B)\varpi_{b,b'}(B). It follows from the definition of an admissible subgroup that, for every admissible subgroup H of π1(B,b′)\pi_1(B, b'), the subgroup γHγ−1\gamma H\gamma^{-1} of π1(B,b)\pi_1(B, b) is admissible. Consequently, the isomorphism uγ:δ↦γδγ−1u_{\gamma}:\delta \mapsto \gamma \delta \gamma^{-1} of π1(B,b′)\pi_1(B, b') onto π1(B,b)\pi_1(B, b) (cf. III, p. 292) is a homeomorphism when these groups are endowed with the admissible topologies.

Remark 3 0202

Let A and B be topological spaces, let aa be a point of A, and let f:A→Bf: A\rightarrow B be a continuous mapping. Put b=f(a)b=f(a). If H is an admissible subgroup of π1(B,b)\pi_1(B, b), its inverse image under the homomorphism π1(f,a)\pi_1(f, a) is an admissible subgroup of π1(A,a)\pi_1(A, a). Consequently, the group homomorphism π1(f,a):π1(A,a)→π1(B,b)\pi_1(f, a):\pi_1(A, a)\rightarrow \pi_1(B, b) is continuous, when these groups are endowed with the admissible topology.

Proposition 5 0203

Let B be a topological space locally arcwise connected, and let aa be a point of B. In order that a subgroup H of π1(B,a)\pi_1(B, a) be admissible, it is necessary and sufficient that there exist a covering (E,p)(E, p) of B and a point x∈Eax\in E_a such that H=p∗(π1(E,x))H =p_*(\pi_1(E, x)).

Let (E,p)(E, p) be a covering of B and let xx be a point of EaE_a; put H=p∗(π1(E,x))H =p_*(\pi_1(E, x)) and let us prove that this is an admissible subgroup of π1(B,a)\pi_1(B, a). Let bb be a point of B and let V be a neighbourhood of bb such that EV=E_V= (p−1(V),pV)(\overset{-1}{p}(V), p_V) is a trivializable covering of V. Let γ∈ϖa,b(B)\gamma \in \varpi_{a,b}(B) ; we shall prove that for every element δ∈π1(V,b)\delta \in \pi_1(V, b), the class of paths γδγ−1\gamma \delta \gamma^{-1} belongs to H.

By III, p. 301, prop. 3, there exists a unique strict homotopy class γ′\gamma ' of path with origin xx in E such that p∗(γ′)=γp_*(\gamma ') =\gamma Let yy be the end of γ′\gamma '; we have p(y)=b(p(y) =b(III, p. 302, prop. 4). Let then δ′\delta ' be the unique class of path with origin yy in E such that p∗(δ′)=δp_*(\delta ') =\delta. Since EVE_V is trivializable, δ′\delta ' is the class of a loop at yy. Then γ′δ′(γ′)−1\gamma '\delta '(\gamma ')^{-1} is the class of a loop at aa in E whose image under p∗p_* is the class γδγ−1\gamma \delta \gamma^{-1}, which was to be proved.

Conversely, let H be an admissible subgroup of π1(B,a)\pi_1(B, a). Let λa(B)\lambda_a(B) be the quotient of the space Λa(B)\Lambda_a(B) of paths with origin aa by the relation of strict homotopy; since two strictly homotopic paths have the same end-point, the end-point mapping e:Λa(B)→Be: \Lambda_a(B)\rightarrow B defines, by passing to the quotient, a mapping ε:λa(B)→B\varepsilon :\lambda_a(B)\rightarrow B. The composition of classes of paths endows the set λa(B)\lambda_a(B) with a left action of the group π1(B,a)\pi_1(B, a). Endow it with the action of the group H deduced by restriction, and denote by H\λa(B)H\backslash \lambda_a(B) the set of its orbits.

The mapping ε:λa(B)→B\varepsilon :\lambda_a(B)\rightarrow B induces, by passing to the quotient, a mapping q:H\λa(B)→Bq: H\backslash \lambda_a(B)\rightarrow B. The composition of classes of paths endows the set H\λa(B)H\backslash \lambda_a(B) with a right operation of the groupoid ϖ(B)\varpi (B) relative to the mapping qq.

This operation is without local monodromy. In fact, let bb be a point of B and let V be a neighbourhood of bb such that γi∗(π1(V,b))γ−1⊂H\gamma i_*(\pi_1(V, b))\gamma^{-1}\subset H for every class of paths γ\gamma joining aa to bb in B, where ii denotes the inclusion of V in B. Since B is locally arcwise connected, one may moreover suppose that V is arcwise connected. Let c∈Vc\in V, let δ\delta be the class in π1(B,c)\pi_1(B, c) of a loop at cc contained in V, and let δ′\delta ' be an element of λa(B)\lambda_a(B) such that ε(δ′)=c\varepsilon (\delta ') =c. Let δ′′\delta '' be an element of ϖc,b(V)\varpi_{c,b}(V) and put γ=δ′δ′′\gamma =\delta '\delta ''. By definition of V, the element δ′δ(δ′)−1=γ((δ′′)−1δδ′′)γ−1\delta '\delta (\delta ')^{-1}=\gamma ((\delta '')^{-1}\delta \delta '')\gamma^{-1} of π1(B,a)\pi_1(B, a) belongs to H. Then Hδ′⋅δ=Hδ′H\delta '\cdot \delta = H\delta '; this proves that π1(V,c)\pi_1(V, c) acts trivially on the set q−1(c)\overset{-1}{q}(c).

By III, p. 313, Proposition 3, there exists a unique topology on H\λa(B)H\backslash \lambda_a(B) for which qq is continuous and the B-space (H\λa(B),q)(H\backslash \lambda_a(B), q) is a covering such that the canonical operation of ϖ(B)\varpi (B) on this covering is the operation defined above. By assertion b) of Theorem 1 of III, p. 305, the group q∗(π1(H\λa(B),H))q_*(\pi_1(H\backslash \lambda_a(B),H)) is equal to H.

We shall say that an operation of the group π1(B,b)\pi_1(B, b) on a set X is admissible if the kernel of the canonical mapping π1(B,b)→SX\pi_1(B, b)\rightarrow \mathfrak{S}_X is an open subgroup of π1(B,b)\pi_1(B, b). This amounts to saying that the homomorphism π1(B,b)→SX\pi_1(B, b)\rightarrow \mathfrak{S}_X is continuous if the group SX\mathfrak{S}_X is endowed with the discrete topology. The mapping π1(B,b)×X→X\pi_1(B, b)\times X\rightarrow X is then continuous, when X is endowed with the discrete topology. Conversely, consider a continuous operation of π1(B,b)\pi_1(B, b) on a discrete space X. Let xx be a point of X; its stabiliser is an open subgroup H of π1(B,b)\pi_1(B, b) by hypothesis. For γ∈π1(B,b)\gamma \in \pi_1(B, b), the stabiliser of γ⋅x\gamma \cdot x is the subgroup γHγ−1\gamma H\gamma^{-1}, so that the subgroup of π1(B,b)\pi_1(B, b) fixing each element of the orbit of xx is the intersection of the subgroups γHγ−1\gamma H\gamma^{-1}, for γ\gamma ranging over π1(B,b)\pi_1(B, b). It is an open subgroup because the open normal subgroups of π1(B,b)\pi_1(B, b) form a basis of its topology (III, p. 315, Proposition 4). It follows that a continuous operation of π1(B,b)\pi_1(B, b) on a discrete space X is admissible if it is transitive or, more generally, if the set of orbits of the elements of X is finite.

For every discrete group G, every principal covering E of B with group G, and every point x∈Ebx\in E_b, let us recall that the mapping h(E,x)h_{(E,x)} from π1(B,b)\pi_1(B, b) into G which, to γ∈π1(B,b)\gamma \in \pi_1(B, b), associates the unique element g∈Gg\in G such that x⋅g=x⋅γ−1x\cdot g=x\cdot \gamma^{-1} is a group homomorphism (III, p. 306, prop. 5).

Proposition 6 0204

Let B be a topological space and let bb be a point of B. Endow the group π1(B,b)\pi_1(B, b) with the admissible topology.

a) For every discrete group G, every principal covering E of B with group G, and every x∈Ebx\in E_b, the homomorphism h(E,x)h_{(E,x)} is a continuous homomorphism of topological groups.

b) For every covering E of B, the canonical action of π1(B,b)\pi_1(B, b) on the fibre EbE_b is admissible.

It is enough to prove the second assertion. Let K be the set of elements k∈π1(B,b)k\in \pi_1(B, b) such that x⋅k=xx\cdot k=x for every x∈Ebx\in E_b; let us prove that K is an admissible subgroup of π1(B,b)\pi_1(B, b).

Let b′b' be a point of B and let V′V' be a neighbourhood of b′b' in B over which the covering E is trivializable. Let γ\gamma be the class of a path cc joining bb to b′b' in B and let c′c' be the unique path in E with origin xx which lifts cc. For every δ∈π1(V′,b′)\delta \in \pi_1(V', b') and every x′∈Eb′x'\in E_{b'}, one has x′⋅δ=x′x'\cdot \delta =x'. Consequently, for x∈Ebx\in E_b and δ∈π1(V′,b′)\delta \in \pi_1(V', b'), one has

x⋅γδγ−1=((x⋅γ)⋅δ)⋅γ−1=(x⋅γ)⋅γ−1=x x\cdot \gamma \delta \gamma^{-1}= ((x\cdot \gamma )\cdot \delta )\cdot \gamma^{-1}= (x\cdot \gamma )\cdot \gamma^{-1}=x

so that γδγ−1\gamma \delta \gamma^{-1} belongs to K. In other words, K is an admissible subgroup, whence the proposition.

Proposition 7 0205

Let B be a connected topological space, locally connected by arcs, and let bb be a point of B. Endow the group π1(B,b)\pi_1(B, b) with the admissible topology.

a) Let G be a discrete topological group. For every continuous homomorphism f:π1(B,b)→f:\pi_1(B, b)\rightarrow G, there exist a principal covering E of B with group G and a point xx of EbE_b such that h(E,x)=fh_{(E,x)}=f.

b) For every discrete topological space F endowed with an admissible right action of the group π1(B,b)\pi_1(B, b), there exists a covering E of B such that the π1(B,b)\pi_1(B, b)-sets F and EbE_b are isomorphic.

Let us prove a). Let ff be a continuous homomorphism from π1(B,b)\pi_1(B, b) into a discrete group G. Its kernel is an open invariant subgroup K of π1(B,b)\pi_1(B, b); let H denote the group π1(B,b)/K\pi_1(B, b)/K and f‾:H→G\overline{f}: H\rightarrow G the group homomorphism deduced from ff by passing to the quotient. By III, p. 316, prop. 5, there exists a connected covering E′E' of B such that the subgroup K is the stabilizer of every point of the fibre Eb′E'_b; the covering E′E' is principal with group H=π1(B,b)/KH =\pi_1(B, b)/K. Let x′x' be a point of Eb′E'_b; the homomorphism h(E′,x′):π1(B,b)→Hh_{(E',x')}:\pi_1(B, b)\rightarrow H is surjective with kernel K (III, p. 306, prop. 5). There is therefore a unique group homomorphism φ:H→G\varphi : H\rightarrow G such that φ∘h(E′,x′)=f\varphi \circ h_{(E',x')}=f. The associated covering E=E′×HGE = E'\times^HG of B is principal with group G and one has h(E,x)=φ∘h(E′,x′)=fh_{(E,x)}=\varphi \circ h_{(E',x')}=f (III, p. 307, example 2). This proves assertion a).

Let us prove b). Let F be a set endowed with an admissible right action of the group π1(B,b)\pi_1(B, b). Let f:π1(B,b)→SFf:\pi_1(B, b)\rightarrow \mathfrak{S}_F denote this action and endow the group SF\mathfrak{S}_F with the discrete topology. By a), there exist a covering E of B, principal with group SF\mathfrak{S}_F, and a point x∈Ex\in E such that the homomorphism h(E,x)h_{(E,x)} is equal to ff. The canonical action of the group π1(B,b)\pi_1(B, b) on the fiber over bb of the associated covering E×SFFE\times^{\mathfrak{S}_F}F of B is identified with the action of π1(B,b)\pi_1(B, b) on F (III, p. 306, Example 1).

Remark 4 0206

The admissible topology on π1(B,b)\pi_1(B, b) is the least fine topology for which the action of π1(B,b)\pi_1(B, b) on EbE_b is continuous for every connected covering E of B (Prop. 6 of III, p. 318 and Prop. 7 of III, p. 318). If E is a connected covering of B and xx a point of the fiber EbE_b, the mapping c′↦c′(1)c'\mapsto c'(1) of Λx(E)\Lambda_x(E) into E is continuous. Consequently, the mapping c↦x⋅cc\mapsto x\cdot c of Λb(B)\Lambda_b(B) into E is continuous (III, p. 302, Cor. 2 of Prop. 3). By restriction to Ωb(B)\Omega_b(B) and passing to the quotient, it follows that the mapping γ↦x⋅γ\gamma \mapsto x\cdot \gamma is continuous when the group π1(B,b)\pi_1(B, b) is endowed with the quotient-space topology of Ωb(B)\Omega_b(B). The admissible topology on π1(B,b)\pi_1(B, b) is therefore less fine than the quotient topology of the topology of compact convergence. It may be strictly less fine (III, p. 337, Exercise 7).

Remark 5 0207

Let B be a connected and locally arcwise connected topological space, let aa be a point of B, and let H be a subgroup of π1(B,a)\pi_1(B, a). By the preceding remark, if H is admissible, it is also open for the quotient topology of the topology of compact convergence. Conversely, suppose that H is a normal subgroup of π1(B,a)\pi_1(B, a) which is open for the quotient topology of the topology of compact convergence. Let xx be a point of B and let γ∈ϖa,x(B)\gamma \in \varpi_{a,x}(B). The subgroup γ−1Hγ\gamma^{-1}H\gamma of π1(B,x)\pi_1(B, x) is again open for the quotient topology of the topology of compact convergence (III, p. 293, Remark 3). There therefore exists a neighbourhood V of xx in B such that γ−1Hγ\gamma^{-1}H\gamma contains the class of every loop at xx whose image is contained in V. Thus γπ1(V,x)γ−1⊂H\gamma \pi_1(V, x)\gamma^{-1}\subset H; since H is normal, this implies that H is an admissible subgroup of π1(B,a)\pi_1(B, a).

Exercises

See the exercises for § 5.

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