Bourbaki

§ 6. Classifying spaces

1. Extension of homotopies

Proposition 1 023V

Let X′X' be a normal topological space, let X be a subspace of X′X', let A be a closed subspace of X′X' contained in X, and let U be a neighbourhood of A in X.

Let us denote by iAi_A the canonical injection of A into X,iU,i_U the canonical injection of U into X; let jAj_A and jUj_U denote the corresponding injections of Cyl(iAi_A) and Cyl(iUi_U) into X×IX\times \mathbf{I}.

There exists a continuous mapping r:X×I→r: X\times \mathbf{I}\rightarrow Cyl(iUi_U) such that jU∘j_U\circ r(x)=xr(x) =x for every point x∈jAx\in j_A(Cyl(iAi_A)).

Let U′U' be an open set of X′X' such that A⊂X∩U′⊂UA\subset X\cap U'\subset U. By definition of a normal space (TG, IX, p. 41), there exists an open neighbourhood V′V' of A in X′X' such that A⊂V′⊂V′‾⊂U′A\subset V'\subset \overline{V'}\subset U' and a continuous function φ′:X′→I\varphi ': X'\rightarrow \mathbf{I} which is equal to 1 at every point of A and to 0 at every point of ∁V′\complement V'. Put φ=φ′∣X\varphi =\varphi '|X and V=X∩V′V = X\cap V'. Let us also denote by α:U×I→\alpha : U\times \mathbf{I}\rightarrow Cyl(iUi_U) and β:X→\beta : X\rightarrow Cyl(iUi_U) the canonical mappings (III, p. 238).

Let rr be the mapping of X×IX\times \mathbf{I} into Cyl(iUi_U) defined by

α(x,1−(1−t)φ(x))\alpha (x,1-(1-t)\varphi (x)) for (x,t)∈U×I(x, t)\in U\times \mathbf{I},

r(x,t)= r(x, t) =

β(x)\beta (x) otherwise.

The mapping rr is continuous on U×IU\times \mathbf{I}. For every point (x,t)∈U×I(x, t)\in U\times \mathbf{I} such that x∉Vx\notin V, we have φ(x)=0\varphi (x) = 0, whence r(x,t)=α(x,1)=β(x)r(x, t) =\alpha (x,1) =\beta (x). It follows that the mappings rr and β∘\beta \circ pr1_1 coincide on (X−V)×I(X-V)\times \mathbf{I}, which implies that rr is continuous on this subspace. Since U and the interior of X-V cover X, the mapping rr is continuous.

Let yy be a point of Cyl(iAi_A) ; put (x,t)=jA(y)(x, t) =j_A(y). If x∈x\in A, φ(x)=1\varphi (x) = 1 and r(x,t)=α(x,t)r(x, t) =\alpha (x, t), whence jU(r(x,t))=(x,t)j_U(r(x, t)) = (x, t). Otherwise, t=1t= 1 and one has r(x,t)=α(x,1)r(x, t) =\alpha (x,1) if x∈Ux\in U and r(x,t)=β(x)r(x, t) =\beta (x) otherwise ; hence jU(r(x,t))=(x,t)j_U(r(x, t)) = (x, t) in this case. This entails that jU∘rj_U\circ r maps every point of jAj_A(Cyl(iAi_A)) to itself, whence the proposition.

Corollary 1 023W

Let X be a normal topological space, let ff be a continuous mapping of X into a topological space Z. Let A be a closed subspace of X, let U be a neighbourhood of A in X and let σ:U×\sigma : U\times I→Z\mathbf{I}\rightarrow Z be a homotopy whose term is the mapping f∣Uf|U. There exists a homotopy σ~:X×I→Z\widetilde{\sigma}: X\times \mathbf{I}\rightarrow Z whose term is the mapping ff and which coincides with σ\sigma in A×IA\times \mathbf{I}.

Put X′=XX'= X and let r:X×I→r: X\times \mathbf{I}\rightarrow Cyl(iUi_U) be the continuous mapping given by Prop. 1. With the notations introduced in the proof of this proposition, there exists a unique continuous mapping F: Cyl(iUi_U)→Z\rightarrow Z such that F(α(x,t))=σ(x,t)F(\alpha (x, t)) =\sigma (x, t) for (x,t)∈U×I(x, t)\in U\times \mathbf{I} and F(β(x))=f(x)F(\beta (x)) =f(x) for x∈x\in X (III, p. 238, Prop. 4). The mapping F∘r:X×I→F\circ r: X\times \mathbf{I}\rightarrow Z is a homotopy ; its term maps a point x∈x\in X to F(r(x,1))=F(β(x))=f(x)F(r(x,1)) = F(\beta (x)) =f(x). If (x,t)∈A×I(x, t)\in A\times \mathbf{I}, F(r(x,t))=F(α(x,t))=σ(x,t)F(r(x, t)) = F(\alpha (x, t)) =\sigma (x, t), therefore this homotopy coincides with σ\sigma on A×IA\times \mathbf{I}.

Corollary 2 023X

Let X be a normal topological space, let A be a closed subspace of X and let U be a neighbourhood of A in X. Let ff and f′f' be continuous homotopic mappings of U into a topological space Z. If ff possesses a continuous extension to X, the same is true of the mapping f′∣Af'|A.

Let F be a continuous mapping of X into Z such that F∣U=fF|U =f. Let us choose a homotopy σ:A×I→Z\sigma : A\times \mathbf{I}\rightarrow Z joining f′f' to ff. By Cor. 1, there exists a homotopy σ~:X×I→Z\widetilde{\sigma}: X\times \mathbf{I}\rightarrow Z with term F which extends σ\sigma. The origin of σ~\widetilde{\sigma} is then a continuous mapping of X into Z which coincides with f′f' on A.

Corollary 3 023Y

Let X be a normal topological space, let A be a closed subspace of X and let U be a neighbourhood of A in X. Let Z be a topological space homotopic to a point and let g:U→g: U\rightarrow Z be a continuous mapping. There exists a continuous mapping g~:X→Z\widetilde{g}: X\rightarrow Z which coincides with gg on A.

Since Z is homotopic to a point, the mapping gg is homotopic to a constant mapping ff. The mapping ff extends continuously to X ; the assertion therefore follows from Cor. 2.

Corollary 4 023Z

Let X be a paracompact topological space, let A be a closed subspace of X. Let nn be an integer ⩾0\geqslant 0 and let V be an open subset of Rn\mathbf{R}^n. Let f:X→Vf: X\rightarrow V be a continuous mapping and let σ:A×I→V\sigma : A\times \mathbf{I}\rightarrow V be a homotopy with terminal map f∣Af|A. There exists a homotopy σ~:X×I→V\widetilde{\sigma}: X\times \mathbf{I}\rightarrow V with terminal map ff which extends σ\sigma.

Let us denote by iAi_A the canonical injection of A into X, by αA:A×I→\alpha_A: A\times \mathbf{I}\rightarrow Cyl(iAi_A) and βA:X→\beta_A: X\rightarrow Cyl(iAi_A) the canonical mappings, and also by jA:j_A: Cyl(iAi_A)→\rightarrow X×IX\times \mathbf{I} the canonical injection. Since A is closed in X, jAj_A defines a homeomorphism of Cyl(iAi_A) onto the closed subspace (A×I)∪(X×{1})(A\times \mathbf{I})\cup (X\times \{1\}) of X×IX\times \mathbf{I}. Let σ~0\widetilde{\sigma}_0 be the unique mapping of Cyl(iAi_A) into V such that σ~0∘αA=σ\widetilde{\sigma}_0\circ \alpha_A=\sigma and σ~0∘βA=f\widetilde{\sigma}_0\circ \beta_A=f; it is continuous (III, p. 238, prop. 4).

Since X is paracompact and I\mathbf{I} is compact, the space X×IX\times \mathbf{I} is paracompact (TG, I, p. 70, prop. 17), hence normal (TG, IX, p. 49, prop. 4). There therefore exists a continuous mapping k:X×I→Rnk: X\times \mathbf{I}\rightarrow \mathbf{R}^n such that k∘jA=σ~0k\circ j_A=\widetilde{\sigma}_0 (TG, IX, p. 45, corollary).

The set U of points x∈x\in X such that k(x,t)∈k(x, t)\in V for all t∈It\in \mathbf{I} is open in X (IV, p. 439, lemma 1). Hence it is a neighbourhood of A. By corollary 1, applied to the mapping ff and to the homotopy k∣U×Ik|U\times \mathbf{I}, there exists a homotopy σ~:X×I→V\widetilde{\sigma}: X\times \mathbf{I}\rightarrow V with initial map ff which coincides with kk, hence with σ\sigma, on A×IA\times \mathbf{I}.

Lemma 1 0240

Let Z be a topological space, K a compact topological space and W an open subset of Z×KZ\times K. The set U of points z∈Zz\in Z such that {z}×K\{z\} \times K is contained in W is open in Z.

The first projection pr1:Z×K→Z_1: Z\times K\rightarrow Z is proper (TG, I, p. 77, cor. 5), hence closed. Consequently, ∁U=\complement U = pr1(∁W)_1(\complement W) is closed in Z and U is open.

Corollary 5 0241

Let X′X' be a normal topological space, let X be a subspace of X′X', let A be a closed subspace of X′X' contained in X, and let U be a neighbourhood of A in X.

Let Y and Z be topological spaces; let f:X×{1}×Y→f: X\times \{1\} \times Y\rightarrow X×{1}×ZX\times \{1\} \times Z be an X×{1}X\times \{1\}-morphism and let g:U×I×Y→U×I×Zg: U\times \mathbf{I}\times Y\rightarrow U\times \mathbf{I}\times Z be a U×IU\times \mathbf{I}-morphism which coincides with ff on U×{1}×YU\times \{1\} \times Y.

There then exists an X×IX\times \mathbf{I}-morphism h:X×I×Y→X×I×Zh: X\times \mathbf{I}\times Y\rightarrow X\times \mathbf{I}\times Z which coincides with ff on X×{1}×YX\times \{1\} \times Y and with gg on A×I×YA\times \mathbf{I}\times Y.

Moreover, if ff and gg are homeomorphisms, one can choose a homeomorphism hh having the required properties.

Let U′U' be an open neighbourhood of A in X′X' such that U′∩X⊂UU'\cap X\subset U. Since the space X′X' is normal, there exists an open neighbourhood V′V' of A in X′X' such that A⊂V′⊂V′‾⊂U′A\subset V'\subset \overline{V'}\subset U' (TG, IX, p. 41). Replacing U by V′‾∩X\overline{V'}\cap X if necessary, we may thus suppose that U is closed in X. Let us then take up again the notation of Prop. 1 and of its proof; let r:X×I→r: X\times \mathbf{I}\rightarrow Cyl(iUi_U) be a continuous mapping such that jU∘r(x)=xj_U\circ r(x) =x for every point x∈jAx\in j_A(Cyl(iAi_A)).

Put also f′=f'= pr3∘f_3\circ f and g′=g'= pr3∘g_3\circ g. Since f′f' and g′g' coincide on U×{1}×YU\times \{1\} \times Y, there exists a unique mapping

φ:\varphi : Cyl(iUi_U)×Y→\times Y\rightarrow Cyl(iUi_U)×Z\times Z

such that φ(α(x,t),y)=(α(x,t),g′(x,t,y))\varphi (\alpha (x, t), y) = (\alpha (x, t), g'(x, t, y)) for (x,t,y)∈U×I×Y(x, t, y)\in U\times \mathbf{I}\times Y and φ(β(x),y)=(β(x),f′(x,1,y))\varphi (\beta (x), y) = (\beta (x), f'(x,1, y)) for (x,y)∈X×Y(x, y)\in X\times Y. Since U is closed in X, the canonical surjection π\pi of (U×I)∪X(U\times \mathbf{I})\cup X onto Cyl(iUi_U) is universally strict (III, p. 239, remark 2). Since the mapping φ∘(π×\varphi \circ (\pi \times IdY)_Y) is continuous, the mapping φ\varphi is continuous. It is therefore a morphism of Cyl(iUi_U)-spaces, and even an isomorphism if ff and gg are homeomorphisms.

Let then h:X×I×Y→X×I×Zh: X\times \mathbf{I}\times Y\rightarrow X\times \mathbf{I}\times Z be the mapping r∗(φ)r^*(\varphi ) deduced from φ\varphi by the base change rr. It is given by h(x,t,y)=h(x, t, y) = (x,t(x, t, pr2(φ(r(x,t),y)))_2(\varphi (r(x, t), y))) for (x,t,y)∈X×I×Y(x, t, y)\in X\times \mathbf{I}\times Y. It is a morphism of X×IX\times \mathbf{I}-spaces, and even an isomorphism if φ\varphi is one. For (x,t,y)∈(x, t, y)\in A×I×YA\times \mathbf{I}\times Y, one has r(x,t)=(x,t)r(x, t) = (x, t), whence

h(x,t,y)=(x,th(x, t, y) = (x, t, pr2(φ(x,t,y)))=(x,t,g′(x,t,y))=g(x,t,y)_2(\varphi (x, t, y))) = (x, t, g'(x, t, y)) =g(x, t, y),

which proves that hh coincides with gg on A×I×YA\times \mathbf{I}\times Y. Analogously, for x∈Xx\in X and y∈Yy\in Y, we have r(x,1)=(x,1)r(x,1) = (x,1), hence

h(x,1,y)=(x,1h(x,1, y) = (x,1, pr2(φ(x,1,y)))=(x,1,f′(x,1,y))=f(x,1,y)_2(\varphi (x,1, y))) = (x,1, f'(x,1, y)) =f(x,1, y)

so that hh coincides with ff on X×{1}×YX\times \{1\} \times Y. The corollary is thus proved.

2. Locally trivial fibre spaces with base B×IB\times \mathbf{I}

Proposition 2 0242

Let B be a paracompact topological space and let (E,p)(E, p) be a locally trivial fibre space over B×IB\times \mathbf{I}. Put E1=p−1(B×{1})E_1=\overset{-1}{p}(B\times \{1\}) and denote by p1:E1→Bp_1: E_1\rightarrow B the mapping pr1∘p∣E1_1\circ p|E_1. Then the B×IB\times \mathbf{I}-spaces (E,p)(E, p) and (E1×I,p1×(E_1\times \mathbf{I}, p_1\times IdI)_{\mathbf{I}}) are isomorphic.

Let us first prove two lemmas.

Lemma 2 0243

Let α,β,γ\alpha ,\beta ,\gamma be real numbers such that α⩽β⩽γ\alpha \leqslant \beta \leqslant \gamma, let B be a topological space and let p:E→B×[α,γ]p: E\rightarrow B\times [\alpha , \gamma ] be a continuous mapping. Put B0=B×[α,β]B_0= B\times [\alpha , \beta ], B1=B×[β,γ]B_1= B\times [\beta , \gamma ], E0=p−1(B0)E_0=\overset{-1}{p}(B_0), E1=E_1= p−1(B1)\overset{-1}{p}(B_1) and denote by p0:E0→B0,p1:E1→B1p_0: E_0\rightarrow B_0,p_1: E_1\rightarrow B_1 the mappings deduced from pp. If (E0,p0)(E_0, p_0) and (E1,p1)(E_1, p_1) are trivializable fibre spaces, the same is true of (E,p)(E, p).

Let g0:E0→B0×F0g_0: E_0\rightarrow B_0\times F_0 and g1:E1→B1×F1g_1: E_1\rightarrow B_1\times F_1 be trivializations of E0E_0 and E1E_1 respectively. Let g0′g'_0 and g1′g'_1 denote the trivializations of the B×{β}B\times \{\beta \}-fibered space p−1(B×{β})\overset{-1}{p}(B\times \{\beta \}) deduced from g0g_0 and g1g_1 by restriction. The mapping h=g0′∘(g1′)−1h=g'_0\circ (g'_1)^{-1} is a B×{β}B\times \{\beta \}-isomorphism of B×{β}×F1B\times \{\beta \} \times F_1 onto B×{β}×F0B\times \{\beta \} \times F_0. Define a continuous mapping h′h' of B×F1B\times F_1 into F0F_0 by putting h′(a,y)=h'(a, y) = pr3∘h(a,β,y)_3\circ h(a, \beta , y) for (a,y)∈B×F1(a, y)\in B\times F_1. For (a,t,y)∈B×[β,γ]×F1(a, t, y)\in B\times [\beta , \gamma ]\times F_1, put H(a,t,y)=(a,t,h′(a,y))H(a, t, y) = (a, t, h'(a, y)). The mapping H resulting is a (B×[β,γ])(B\times [\beta , \gamma ])-isomorphism of B×[β,γ]×F1B\times [\beta , \gamma ]\times F_1 onto B×[β,γ]×F0B\times [\beta , \gamma ]\times F_0, and one has g0∣p−1(B×{β})=H∘g1∣p−1(B×{β})g_0|\overset{-1}{p}(B\times \{\beta \}) = H\circ g_1|\overset{-1}{p}(B\times \{\beta \}). There therefore exists a continuous mapping g:E→B×[α,γ]×F0g: E\rightarrow B\times [\alpha , \gamma ]\times F_0 such that g∣E0=g0g|E_0=g_0 and g∣E1=H∘g1g|E_1= H\circ g_1. The mapping gg is an isomorphism of B×[α,γ]B\times [\alpha , \gamma ]-spaces, hence E is trivializable.

Lemma 3 0244

Let B be a topological space and let (E,p)(E, p) be a locally trivial B×IB\times \mathbf{I}-fibered space. Every point aa of B possesses a neighbourhood V such that the V×IV\times \mathbf{I}-space EV×IE_{V\times\mathbf{I}} is trivializable.

Let aa be a point of B; for every point tt of I\mathbf{I}, there exists an open neighbourhood WtW_t of tt in I\mathbf{I} and a neighbourhood VtV_t of aa in B such that E is trivializable over Vt×WtV_t\times W_t. There then exist an integer n>0n >0 and, for every integer ii such that 1 ⩽i⩽n\leqslant i\leqslant n, a point tit_i of I\mathbf{I} such that the interval [i−1n,in][\frac{i-1}{n},\frac{i}{n}] is contained in Wti(W_{t_i}(III, p. 272, lemma 4). Put V=∩1⩽i⩽nVtiV =\cap_{1\leqslant i\leqslant n}V_{t_i}. The fibered space E is trivializable over V×V\times [i−1n,in][\frac{i-1}{n},\frac{i}{n}] for every integer ii such that 1⩽i⩽n1\leqslant i\leqslant n. Lemma 3 then follows from Lemma 2 by induction on nn.

Let us now prove the proposition. By lemma 3, there exists an open covering (Uj)j∈J(U_j)_{j\in J} of B such that, for every j∈Jj\in J, E is trivializable over Uj×IU_j\times \mathbf{I}. Since the space B is paracompact, we may suppose the covering (Uj)j∈J(U_j)_{j\in J} locally finite (TG, IX, p. 49) and choose a covering (Aj)j∈J(A_j)_{j\in J} of B where, for every j∈Jj\in J, the set AjA_j is closed in B and contained in UjU_j (TG, IX, p. 49, prop. 4 and p. 48, cor. 1).

For every open subset U of B, let us denote by F(U)\mathscr{F}(U) the set of U×IU\times \mathbf{I}-isomorphisms of p−1(U×I)\overset{-1}{p}(U\times \mathbf{I}) onto p1−1(U)×I\overset{-1}{p_{1}}(U)\times \mathbf{I} which induce the identical mapping of p−1(U×{1})\overset{-1}{p}(U\times \{1\}) onto p1−1(U)×{1}\overset{-1}{p_{1}}(U)\times \{1\}. For every pair (V,U)(V,U) of open sets of B such that U⊂VU\subset V, let us denote by rUV:F(V)→F(U)r_{UV}:\mathscr{F}(V)\rightarrow \mathscr{F}(U) the mapping which, to a (V×I)(V\times \mathbf{I})-isomorphism g:p−1(V×I)→p0−1(V)×Ig:\overset{-1}{p}(V\times \mathbf{I})\rightarrow \overset{-1}{p_{0}}(V)\times \mathbf{I}, associates the (U×I)(U\times \mathbf{I})-isomorphism deduced from gg by passing to the subspaces. The pair F=((F(U)),(rUV))\mathscr{F}= ((\mathscr{F}(U)),(r_{UV})) is a sheaf on B (I, p. 45, example 4). To prove the proposition, it is enough to prove that the sheaf F\mathscr{F} is soft (I, p. 64).

Let j∈Jj\in J, let A be a closed subset of AjA_j, let V be an open set in B such that A⊂V⊂UjA\subset V\subset U_j, and let gg be an element of F(V)\mathscr{F}(V). There exists an open neighbourhood W of A such that W‾⊂V\overline{W}\subset V, for a paracompact space is normal (TG, IX, p. 49, prop. 4). We shall prove that there exists an element g′g' of F(Uj)\mathscr{F}(U_j) such that g′∣W=g∣Wg'|W =g|W. Corollary 2 (I, p. 65) of prop. 6 then implies that the sheaf F\mathscr{F} is soft.

Since the B×IB\times \mathbf{I}-space E is trivializable over Uj×IU_j\times \mathbf{I}, we may suppose that p−1(Uj×I)=Uj×I×F\overset{-1}{p}(U_j\times \mathbf{I}) = U_j\times \mathbf{I}\times F, where F is a topological space. The element gg of F(V)\mathscr{F}(V) is then a V×IV\times \mathbf{I}-isomorphism of V×I×FV\times \mathbf{I}\times F onto itself which induces the identical mapping of V×V\times {1}×F\{1\}\times F. Apply cor. 5 of IV, p. 439, to the spaces X′=XX'= X, X=UjX = U_j, A = W, U = V, and Y = Z = F, to the mapping g:V×I×F→V×I×Fg: V\times \mathbf{I}\times F\rightarrow V\times \mathbf{I}\times F and to the identical mapping of Uj×{1}×FU_j\times \{1\} \times F. There therefore exists a Uj×IU_j\times \mathbf{I}-isomorphism g′g' of Uj×I×FU_j\times \mathbf{I}\times F onto itself which induces the identical mapping of Uj×{1}×FU_j\times \{1\} \times F and which coincides with gg on W‾×I×F\overline{W}\times \mathbf{I}\times F, and a fortiori on W×I×FW\times \mathbf{I}\times F. Hence the proposition.

Corollary 1 0245

Let B be a paracompact topological space and let (E,p)(E, p) be a locally trivial fibre space over B×IB\times \mathbf{I} (I, p. 71, Corollary 2). For t∈It\in \mathbf{I}, let us denote by (Et,pt)(E_t, p_t) the locally trivial fibre space over B it∗Ei^*_tE, where it:B→B×{t}i_t: B\rightarrow B\times \{t\} is the mapping b↦(b,t)b\mapsto (b, t). The locally trivial fibre spaces over B E0E_0 and EtE_t are isomorphic for every t∈It\in \mathbf{I}.

Corollary 2 0246

Let B and B′B' be topological spaces, and let E be a locally trivial fibre space over B. Let f0f_0 and f1f_1 be continuous mappings of B′B' into B. Suppose that the space B′B' is paracompact. If the mappings f0f_0 and f1f_1 are homotopic, the locally trivial fibre spaces over B′B' f0∗Ef_0^*E and f1∗Ef_1^*E are isomorphic.

Let σ:B′×I→B\sigma : B'\times \mathbf{I}\rightarrow B be a homotopy joining f0f_0 to f1f_1 and let us denote by E′E' the space over B′×IB'\times \mathbf{I} σ∗E\sigma^*E; this is a locally trivial fibre space. Let us denote by i0i_0 and i1i_1 the mappings of B′B' into B′×IB'\times \mathbf{I} given by x↦(x,0)x\mapsto (x,0) and x↦(x,1)x\mapsto (x,1). By Cor. 1, the fibre spaces over B′B' i0∗E′i^*_0E' and i1∗E′i^*_1E' are isomorphic. Since σ∘i0=f0\sigma \circ i_0=f_0, the space over B′B' i0∗E′i^*_0E' is identified with f0∗Ef_0^*E; analogously, the space over B′B' i1∗E′i^*_1E' is identified with f1∗Ef_1^*E. Consequently, the fibre spaces over B′B' f0∗Ef_0^*E and f1∗Ef_1^*E are isomorphic, as was to be proved.

Corollary 3 0247

Let B be a paracompact topological space. If B is homotopic to a point, every locally trivial fibre space with base B is trivializable.

Corollary 4 0248

Let B and B′B' be topological spaces, let (E,p)(E, p) be a locally trivial fibre space over B, let ff be a continuous mapping of B′B' into B, let σ:B′×I→B\sigma : B'\times \mathbf{I}\rightarrow B be a homotopy starting from ff, and let f~\widetilde{f} be a continuous lifting of ff to E. If the space B′B' is paracompact, there exists a homotopy σ~:B′×I→E\widetilde{\sigma}: B'\times \mathbf{I}\rightarrow E starting from f~\widetilde{f} which is a lifting of σ\sigma to E.

Let (E′,p′)(E', p') be the B′×IB'\times \mathbf{I}-space deduced from (E,p)(E, p) by base change by the mapping σ:B′×I→B\sigma : B'\times \mathbf{I}\rightarrow B. It is a locally trivial fibre space (I, p. 71, Corollary 2), and the mapping s:B′×{0}→E′s: B'\times \{0\} \rightarrow E' defined by s((a,0))=((a,0),f~(a))s((a,0)) = ((a,0),\widetilde{f}(a)) for a∈B′a\in B' is a continuous section of p′p' over B′×{0}B'\times \{0\}. By Proposition 2, there exists a continuous section s~\widetilde{s} of p′p' extending ss. The mapping σ~=\widetilde{\sigma}= pr2∘s~_2\circ \widetilde{s} has the required property.

3. Principal fibre spaces with base B×IB\times \mathbf{I}

Let G be a topological group. We shall see that Proposition 2 and its corollaries remain valid when, in each statement, one replaces “locally trivial fibre space” by “principal fibre space with group G”.

Lemma 4 0249

Let B be a topological space, let G be a topological group and let E, E′E' be principal fibre spaces with group G and base B. Let F denote the topological space G endowed with the left operation of G×GG\times G given by (g,g′)⋅f=g′fg−1(g, g')\cdot f=g'f g^{-1}, and let M=(E×BE′)×G×GFM = (E\times_BE')\times^{G\times G}F denote the locally trivial fibre space with fibre type F and base B associated with it.

The sheaf S\mathscr{S} on B of sections of M is isomorphic to the sheaf on B of isomorphisms of principal fibre spaces from E onto E′E'.

Let us denote by p,p′p,p' and qq the projections of the B-spaces E, E′E' and M. For (x,x′)∈E×BE′(x, x')\in E\times_BE' and f∈Ff\in F, let [x,x′,f][x, x', f] denote the class in M of the element ((x,x′),f)∈(E×BE′)×F((x, x'), f)\in (E\times_BE')\times F. Let M\mathscr{M} denote the sheaf on B of isomorphisms of principal fibre spaces from E onto E′E'; its sections over an open set U of B are the isomorphisms of principal fibre spaces from EUE_U onto EU′E'_U.

Let ee denote the identity element of G. Let U be an open set of B and let φ:EU→EU′\varphi : E_U\rightarrow E'_U be an isomorphism of principal fibre spaces with group G and base U. The mapping from EUE_U into (E×BE′)×G×GF(E\times_BE')\times^{G\times G}F defined by x↦[x,φ(x),e]x\mapsto [x, \varphi (x), e] is continuous. For g∈Gg\in G and x∈EUx\in E_U, it sends x⋅gx\cdot g to [x⋅g,φ(x⋅g),e]=[x,φ(x),geg−1]=[x,φ(x),e][x\cdot g, \varphi (x\cdot g), e] = [x, \varphi (x), geg^{-1}] = [x, \varphi (x), e]. There therefore exists a unique continuous mapping αU(φ):U→M\alpha_U(\varphi ): U\rightarrow M such that αU(φ)(p(x))=[x,φ(x),e]\alpha_U(\varphi )(p(x)) = [x, \varphi (x), e] for every x∈EUx\in E_U; we have αU(φ)∈S(U)\alpha_U(\varphi )\in \mathscr{S}(U).

It is immediate that the mappings αU\alpha_U define a morphism of sheaves α\alpha from M\mathscr{M} into S\mathscr{S}.

Lemma 5 024A

The morphism of sheaves α\alpha is an isomorphism.

Suppose first that the principal fibre spaces E and E′E' are both trivializable; choose sections i:B→Ei: B\rightarrow E and i′:B→E′i': B\rightarrow E' of them. Then there exists a unique continuous mapping θ\theta from M into B×GB\times G such that one has

θ([i(b)⋅g,i′(b)⋅g′,f])=(b,g′fg−1) \theta ([i(b)\cdot g, i'(b)\cdot g', f]) = (b, g'f g^{-1})

for b∈B,g∈G,g′∈Gb\in B,g\in G,g'\in G and f∈f\in F; it is an isomorphism of B-spaces. Let φ\varphi be an isomorphism of principal fibre spaces from E onto E′E'; there exists a unique continuous mapping γ:B→G\gamma : B\rightarrow G such that φ(i(b))=i′(b)⋅γ(b)\varphi (i(b)) =i'(b)\cdot \gamma (b) for b∈Bb\in B. The image of φ\varphi under the mapping αB\alpha_B is the mapping b↦θ−1(b,γ(b))b\mapsto \theta^{-1}(b, \gamma (b)) from B into M. This implies that αB\alpha_B is a bijection.

Consequently, αU\alpha_U is a bijection for every open set U of B such that the principal fibre spaces EUE_U and EU′E'_U are trivializable. By corollary 2 of I, p. 55, this implies that α\alpha is an isomorphism of sheaves.

Proposition 3 024B

Let G be a topological group, let B be a paracompact topological space and let (E,p)(E, p) be a principal fibre space of group G and base B×IB\times \mathbf{I}. Put E1=p−1(B×{1})E_1=\overset{-1}{p}(B\times \{1\}) and let p1:E0→Bp_1: E_0\rightarrow B be the mapping pr1∘p∣E1_1\circ p|E_1. Then, (E,p)(E, p) and (E1×I,p1×(E_1\times \mathbf{I}, p_1\times IdI)_{\mathbf{I}}) are isomorphic principal fibre spaces of group G and base B×IB\times \mathbf{I}.

Let F be the topological space G endowed with the left operation of the group G×GG\times G given by (g,g′)⋅f=g′fg−1(g, g')\cdot f=g'f g^{-1}. Let (M,q)(M, q) be the locally trivial fibre space of base B×IB\times \mathbf{I} and fibre type F associated with the principal fibre space E×B×I(E1×I)E\times_{B\times\mathbf{I}}(E_1\times \mathbf{I}) of group G×GG\times G. Put M1=q−1(B×{1})M_1=\overset{-1}{q}(B\times \{1\}) and q1=q_1= pr1∘p∣M1_1\circ p|M_1; the B-space (M1,q1)(M_1, q_1) is identified with the locally trivial fibre space of fibre type F associated with E1×BE1E_1\times_BE_1. By lemma 4, where one takes for principal fibre spaces E and E′E' both equal to E1E_1, the B-space M1M_1 has a section. Since the B×IB\times \mathbf{I}-spaces (M,q)(M, q) and (M1×I,q1×(M_1\times \mathbf{I}, q_1\timesIdI)_{\mathbf{I}}) are isomorphic (IV, p. 440, prop. 2), the B×IB\times \mathbf{I}-space (M,q)(M, q) has a section, which implies that the principal fibre spaces of group G, E and E1×IE_1\times \mathbf{I}, are isomorphic.

Corollary 1 024C

Let B be a paracompact topological space, let G be a topological group and let (E,p)(E, p) be a principal fibre B×IB\times \mathbf{I}-space of group G. For t∈It\in \mathbf{I}, let (Et,pt)(E_t, p_t) denote the principal fibre B-space it∗Ei^*_tE, where it:B→B×{t}i_t: B\rightarrow B\times \{t\} is the mapping b↦(b,t)b\mapsto (b, t). The principal fibre B-spaces E0E_0 and EtE_t are isomorphic for every t∈It\in \mathbf{I}.

Corollary 2 024D

Let B and B′B' be topological spaces, let G be a topological group, and let E be a B-principal fibre space with group G. Let f0f_0 and f1f_1 be continuous mappings of B′B' into B. Suppose that the space B′B' is paracompact. If the mappings f0f_0 and f1f_1 are homotopic, the B′B'-principal fibre spaces f0∗Ef_0^*E and f1∗Ef_1^*E are isomorphic.

Corollary 3 024E

Let B be a paracompact topological space and let G be a topological group. If B is homotopic to a point, every principal fibre space with group G is trivialisable.

Remark 024F

An alternative proof of these results would consist in verifying that the isomorphisms of fibre spaces constructed in prop. 2 and its corollaries are isomorphisms of principal fibre spaces.

4. Universal Fibre Spaces

Let G be a topological group, let B and B′B' be topological spaces, and let (E,p)(E, p) and (E′,p′)(E', p') be principal fibre spaces with group G and with bases B and B′B' respectively.

Let U be an open subset of B and let f′:p−1(U)→E′f':\overset{-1}{p}(U)\rightarrow E' be a continuous mapping which is compatible with the operations of G in p−1(U)\overset{-1}{p}(U) and E′E' respectively. Then there exists a unique continuous mapping f:U→B′f: U\rightarrow B' such that f∘pU=p′∘f′f\circ p_U=p'\circ f' and the commutative square

EUf′E′ E_U^{f'}E'
pUp′p_Up'

U fB′^fB'

is then cartesian (I, p. 94, example (FP)).

For every open subset U of B, let us then denote by F(U)\mathscr{F}(U) the set of continuous mappings g:EU→E′g: E_U\rightarrow E' which are compatible with the operations of G in p−1(U)\overset{-1}{p}(U) and E′E' respectively. For every pair (U,V)(U,V) of open subsets of B such that U⊂VU\subset V, let rUV:F(V)→F(U)r_{UV}:\mathscr{F}(V)\rightarrow \mathscr{F}(U) denote the mapping defined by rUV(g)=g∣EUr_{UV}(g) =g|E_U. It is immediately verified that in this way a sheaf F=((F(U)),(rUV))\mathscr{F}= ((\mathscr{F}(U)),(r_{UV})) on B has been defined. We shall call this sheaf the sheaf on B of morphisms of principal fibre spaces with group G from E into E′E'.

Proposition 4 024G

If the space B is paracompact and if the space E′E' is homotopic to a point, the sheaf on B of morphisms of principal fibre spaces with group G from E into E′E' is a soft sheaf.

There exists an open covering (Uj)j∈J(U_j)_{j\in J} of B such that, for every j∈Jj\in J, the fibre space EUjE_{U_j} is trivialisable. Since the space B is paracompact, we may suppose that the covering (Uj)j∈J(U_j)_{j\in J} is locally finite (TG, IX, p. 49) and choose a covering (Aj)j∈J(A_j)_{j\in J} of B where, for every j∈Jj\in J, the set AjA_j is closed in B and contained in UjU_j (TG, IX, p. 49, prop. 4 and p. 48, cor. 1).

By I, p. 65, cor. 2 of prop. 6, it is enough, in order to prove the proposition, to establish the following assertion: let U be an open subset of B such that the principal fibre space (EU,pU)(E_U, p_U) is trivialisable, let A be a closed subset of B contained in U, let V be an open neighbourhood of A contained in U, and let ff be an element of F(V)\mathscr{F}(V); then there exists an open neighbourhood W of A in V and an element f′f' of F(U)\mathscr{F}(U) such that rWU(f′)=rWV(f)r_{WU}(f') =r_{WV}(f). Let us prove this assertion.

Let W be an open subset of B such that A⊂W⊂W‾⊂VA\subset W\subset \overline{W}\subset V. Let s:U→EUs: U\rightarrow E_U be a section of (EU,pU)(E_U, p_U). Apply Corollary 3 (IV, p. 438) to the space B, the closed subset W‾\overline{W} and the neighbourhood V of W‾\overline{W}, and to the mapping g=f∘(s∣V)g=f\circ (s|_V) of EVE_V into E′E'. There then exists a continuous mapping g~:B→E′\widetilde{g}: B\rightarrow E' which coincides with gg on W‾\overline{W}. Put h=g~∣Uh=\widetilde{g}|_U. We have h∣W=g∣W=f∘(s∣W)h|_W=g|_W=f\circ (s|_W).

The mapping H:U×G→E′H: U\times G\rightarrow E' defined by H(x,g)=h(x)⋅gH(x, g) =h(x)\cdot g for (x,g)∈U×G(x, g)\in U\times G is continuous and compatible with the operations of G in U×GU\times G and E′E'. Put f′=H∘s−1f'= H\circ s^{-1}; this is an element of F(U)\mathscr{F}(U). For every x∈Wx\in W, the mappings ff and f′f' coincide at the point s(x)s(x), hence at every point of p−1(x)\overset{-1}{p}(x), since they are morphisms of principal fibre spaces. The proposition follows.

Theorem 1 024H

Let G be a topological group, let BuB_u be a topological space, and let (Eu,pu)(E_u, p_u) be a principal fibre space with group G and base BuB_u. Suppose that the space EuE_u is homotopic to a point.

Let B be a paracompact topological space.

a) Every principal fibre space with group G and base B is isomorphic to a principal fibre space of the form f∗Euf^*E_u, where f:B→Buf: B\rightarrow B_u is a continuous mapping.

b) Let f0f_0 and f1f_1 be continuous mappings of B into BuB_u. In order that f0∗Euf_0^*E_u and f1∗Euf_1^*E_u should be isomorphic principal fibre spaces with group G and base B, it is necessary and sufficient that the mappings f0f_0 and f1f_1 should be homotopic.

In other words, there exists a mapping of [B,Bu][B,B_u] into P(B; G) which associates with the homotopy class of a continuous mapping ff of B into BuB_u the isomorphism class of the principal fibre space f∗Euf^*E_u. This mapping is bijective.

Let E be a principal fibre space with group G and base B. By Prop. 4, the sheaf F\mathscr{F} on B of morphisms of principal fibre spaces of E into EuE_u is soft. Hence F(B)\mathscr{F}(B) is not empty, whence a).

Let f0f_0 and f1f_1 be continuous mappings of B into BuB_u. If the mappings f0f_0 and f1f_1 are homotopic, the principal fibre spaces f0∗Euf_0^*E_u and f1∗Euf_1^*E_u are isomorphic (IV, p. 445, corollary 2). Let us prove the converse. For α∈{0,1}\alpha \in \{0,1\}, let EαE_{\alpha} denote the principal fibre BuB_u-space fα∗Euf_{\alpha}^*E_u, gα:Eα→Eug_{\alpha}: E_{\alpha}\rightarrow E_u the first projection and pα:Eα→Bp_{\alpha}: E_{\alpha}\rightarrow B the second projection. Let i:E0→E1i: E_0\rightarrow E_1 be an isomorphism of principal fibre spaces. Let pp be the mapping p0×p_0\times IdI:E0×I→B×I_{\mathbf{I}}: E_0\times \mathbf{I}\rightarrow B\times \mathbf{I}.

Since the space B×IB\times \mathbf{I} is paracompact (TG, IX, p. 70, proposition 17), the sheaf G\mathscr{G} on B×IB\times \mathbf{I} of morphisms of principal fibre spaces from E0×IE_0\times \mathbf{I} into EuE_u is soft (IV, p. 446, proposition 4). Put A=B×{0,1}A = B\times \{0,1\}, U=B×([0,12[∪]12,1])U = B\times ([0,\frac{1}{2}[\cup ]\frac{1}{2},1]), and define an element gg of G(U)\mathscr{G}(U) by setting

g0(x)g_0(x) for (x,t)∈E0×[0,1[(x, t)\in E_0\times [0,^1[,

g(x,t)=g(x, t) = 2

g1∘i(x)g_1\circ i(x) for (x,t)∈E0×]12,1](x, t)\in E_0\times ]\frac{1}{2},1].

Since the sheaf G\mathscr{G} is soft, there exist an element h∈G(B×I)h\in \mathscr{G}(B\times \mathbf{I}) and an open neighbourhood V of A in U such that h∣V=g∣Vh|V =g|V; such an element is a continuous mapping H:E0×I→EuH: E_0\times \mathbf{I}\rightarrow E_u, compatible with the operations of G and such that H(x,0)=g0(x)H(x,0) =g_0(x), H(x,1)=g1(i(x))H(x,1) =g_1(i(x)) for every x∈E0x\in E_0. There then exists a mapping h′:B×I→Buh': B\times \mathbf{I}\rightarrow B_u such that h′(p0(x),t)=pu(H(x,t))h'(p_0(x), t) =p_u(H(x, t)) for x∈E0x\in E_0 and t∈It\in \mathbf{I}; this mapping is continuous, and it is a homotopy joining f0f_0 to f1f_1.

Corollary 024I

Let G be a topological group, let B and B′B' be paracompact topological spaces. Let (E,p)(E, p) and (E′,p′)(E', p') be principal fibre spaces with group G and with base B and B′B' respectively. Suppose that the spaces E and E′E' are both homotopic to a point. The spaces B and B′B' are homotopic.

For there exists a continuous mapping f:B→B′f: B\rightarrow B' such that the principal fibre space E is isomorphic to the principal fibre space f∗E′f^*E' and a continuous mapping g:B′→Bg: B'\rightarrow B such that the principal fibre space E′E' is isomorphic to the principal fibre space g∗Eg^*E (Theorem 1, a)). The principal fibre spaces (g∘f)∗E(g\circ f)^*E and E are then isomorphic, therefore the mapping g∘fg\circ f is homotopic to the mapping IdB_B (Theorem 1, b)). Analogously, the mapping f∘gf\circ g is homotopic to the mapping IdB′_{B'}.

Let G be a topological group, let B be a topological space and let E be a principal fibre space with group G and with base B. Suppose that the space E is homotopic to a point. One says that the principal fibre space E is universal for principal fibre spaces with group G and paracompact base, and one says that the space B is a classifying space for G. If two classifying spaces for G are paracompact, they are homotopic. When there exists a classifying space, the study of isomorphism classes of principal fibre spaces with group G and paracompact base may be regarded as a homotopy problem.

Example 024J

Endowed with the mapping p:R→S1,t↦e2πitp:\mathbf{R}\rightarrow \mathbf{S}^1,t\mapsto e^{2\pi it}, and with the operation of Z\mathbf{Z} by translation, the space R\mathbf{R} is a principal covering space with group Z\mathbf{Z}. The space S1\mathbf{S}^1 is thus a classifying space for the group Z\mathbf{Z}.

For every discrete group G, we shall construct in the following number a metrisable space which is a classifying space for G.

5. Classifying space for a discrete group

Let G be a topological group; let us denote by ee its identity element. Let G∗G^* be the set of mappings h:[0,1[→Gh: [0,1[\rightarrow G for which there exists a finite sequence (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} with 0=t0<t1<⋯<tn=10 =t_0< t_1<\cdots < t_n= 1 such that hh is constant on the intervals [ti−1,ti[[t_{i-1}, t_i[ for 1⩽i⩽n1\leqslant i\leqslant n. Such a sequence will be said to be adapted to hh. For every finite subset of G∗G^*, there exists a sequence adapted to each of its elements.

The set G∗G^* is a subgroup of G[0,1[G^{[0,1[}. We denote by e∗e^* its identity element; the inverse of an element h∈G∗h\in G^* is the mapping t↦t\mapsto h(t)−1h(t)^{-1}, denoted by h−1h^{-1}.

Let V be a neighbourhood of ee in G. Let h∈G∗h\in G^* and let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to hh. The set of elements t∈[0,1[t\in [0,1[ such that h(t)∉Vh(t)\notin V is the union of certain of the intervals [ti−1,ti[[t_{i-1}, t_i[; the sum of the lengths ti−ti−1t_i-t_{i-1} of these intervals does not depend on the sequence (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} chosen; let us denote it by pV(h)p_V(h). For every real number ε\varepsilon such that ε>0\varepsilon >0, let us then denote by Vε∗V^*_{\varepsilon} the set of h∈G∗h\in G^* such that pV(h)<εp_V(h)< \varepsilon.

Proposition 5 024K

There exists a unique topology on G∗G^* compatible with its group structure for which the sets Vε∗V^*_{\varepsilon} form a basis of neighbourhoods of the identity element.

Let us verify that the sets Vε∗V_{\varepsilon}^* satisfy axioms (GVI′'_I), (GVII′'_{II}) and (GVIII′'_{III}) of TG, III, p. 4.

Let V be a neighbourhood of ee in G and let ε\varepsilon be a strictly positive real number. Let W be a neighbourhood of ee in G such that W⋅W⊂VW\cdot W\subset V. Let hh and h′h' be elements of G∗G^*. If t∈[0,1[t\in [0,1[ is such that h(t)h′(t)∉Vh(t)h'(t)\notin V, one has h(t)∉Wh(t)\notin W or h′(t)∉Wh'(t)\notin W. It follows that pV(hh′)⩽pW(h)+pW(h′)p_V(hh')\leqslant p_W(h)+p_W(h'). Consequently, Wε/2∗⋅Wε/2∗⊂Vε∗W^*_{\varepsilon /2}\cdot W^*_{\varepsilon /2}\subset V^*_{\varepsilon}, which proves axiom (GVI′'_I).

Let W be a neighbourhood of ee in G such that W−1⊂VW^{-1}\subset V. Then (Wε∗)−1=(W^*_{\varepsilon})^{-1}= (W−1)ε∗⊂Vε∗(W^{-1})^*_{\varepsilon}\subset V^*_{\varepsilon}, whence axiom (GVII′'_{II}).

Let kk be an element of G∗G^*; since the function kk takes only a finite number of values, there exists a neighbourhood W of ee in G such that k(t)Wk(t)−1⊂Vk(t)Wk(t)^{-1}\subset V for all t∈[0,1[t\in [0,1[. Let hh then be an element of G∗G^*. For t∈[0,1[t\in [0,1[, if k(t)h(t)k(t)−1∉k(t)h(t)k(t)^{-1}\notin V, then h(t)∉Wh(t)\notin W. Consequently, pV(khk−1)⩽pW(h)p_V(khk^{-1})\leqslant p_W(h). This proves that kWε∗k−1⊂Vε∗kW^*_{\varepsilon}k^{-1}\subset V^*_{\varepsilon}, whence axiom (GVIII′'_{III}).

Proposition 6 024L

The space G∗G^* is contractible and locally contractible at each of its points.

For h∈G∗h\in G^* and t∈It\in \mathbf{I}, let us denote by σ(h,t)\sigma (h, t) the mapping of [0,1[[0,1[ into G defined by σ(h,t)(x)=h(x)\sigma (h, t)(x) =h(x) if 0⩽x<t0\leqslant x < t and σ(h,t)=e\sigma (h, t) =e otherwise.

Let us show that the mapping σ:G∗×I→G∗\sigma : G^*\times \mathbf{I}\rightarrow G^* resulting is continuous. Let in fact k∈G∗,u∈Ik\in G^*,u\in \mathbf{I}, let V be a neighbourhood of ee in G and let ε\varepsilon be a strictly positive real number. The element σ(h,t)σ(k,u)−1\sigma (h, t)\sigma (k, u)^{-1} of G∗G^* is the mapping ff of [0,1[[0,1[ into G defined by

h(x)k(x)−1h(x)k(x)^{-1} if 0⩽x<0\leqslant x < min(t,ut, u) ;

h(x)h(x) if u⩽x<tu\leqslant x < t;

f(x)= f(x) =

k(x)−1k(x)^{-1} if t⩽x<ut\leqslant x < u;

ee otherwise.

Consequently,

pV(σ(h,t)σ(k,u)−1)⩽pV(hk−1)+∣t−u∣ p_V(\sigma (h, t)\sigma (k, u)^{-1})\leqslant p_V(hk^{-1}) +|t-u|

in other words, for σ(h,t)∈Vε∗σ(k,u)\sigma (h, t)\in V^*_{\varepsilon}\sigma (k, u), it is enough that one have ∣t−u∣⩽ε2|t-u|\leqslant \frac{\varepsilon}{2} and h∈Vε/2∗kh\in V^*_{\varepsilon /2}k, which proves the continuity of σ\sigma at (k,u)(k, u).

For every h∈G∗,σ(h,0)h\in G^*,\sigma (h,0) is the constant mapping with image {e}\{e\}, whereas σ(h,1)=h\sigma (h,1) =h. Moreover, σ(e,t)=e\sigma (e, t) =e for every t∈It\in \mathbf{I}. Consequently, σ\sigma is a pointed homotopy at e∈G∗e\in G^* joining the constant mapping with image {e}\{e\} to the identity mapping of G∗G^*. This proves that G∗G^* is contractible at e∗e^*.

Moreover, for every neighbourhood V of ee in G and every real number ε>0\varepsilon >0, one has σ(Vε∗×I)⊂Vε∗\sigma (V^*_{\varepsilon}\times \mathbf{I})\subset V^*_{\varepsilon}. Hence Vε∗V^*_{\varepsilon} is also contractible at e∗∈e^*\in G∗G^*, so that G∗G^* is locally contractible at e∗e^*.

Since G∗G^* is a topological group, it is contractible and locally contractible at each of its points.

Let ι\iota be the mapping of G into G∗G^* which associates to g∈Gg\in G the constant mapping with image {g}\{g\} of [0,1[[0,1[ into G. The mapping ι\iota is an injective homomorphism of groups. Let V be a neighbourhood of ee in G and let ε\varepsilon be a strictly positive real number. One has ι−1(Vε∗)=V\overset{-1}{\iota}(V^*_{\varepsilon}) = V if ε⩽1\varepsilon \leqslant 1 and ι−1(Vε∗)=G\overset{-1}{\iota}(V^*_{\varepsilon}) = G otherwise. The inverse image of a neighbourhood of the identity element of G∗G^* is a neighbourhood of the identity element of G, whence the continuity of ι\iota. Moreover, ι(V)=V1∗∩ι(G)\iota (V) = V^*_1\cap \iota (G) for every neighbourhood V of ee in G. Hence ι\iota defines an isomorphism of topological groups of G onto its image.

Remark 1 024M

If G is a separated topological group, ι(G)\iota (G) is closed in G∗G^*.

Let indeed h∈G∗h\in G^* be such that h∉ι(G)h\notin \iota (G), let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to hh, put ε=\varepsilon = min1⩽i⩽n(ti−ti−1)_{1\leqslant i\leqslant n}(t_i-t_{i-1}). Let V be a neighbourhood of ee in G such that h(ti)−1h(tj)∉Vh(t_i)^{-1}h(t_j)\notin V, for every pair (i,j)(i, j) of integers such that 0⩽i,j⩽n−10\leqslant i, j\leqslant n-1 and h(ti)=h̸(tj)h(t_i)=\not h(t_j); there exists one since G is separated. Let W be a neighbourhood of ee in G such that W⋅W−1⊂VW\cdot W^{-1}\subset V. Let us prove that then hWε∗hW^*_{\varepsilon} does not meet ι(G)\iota (G).

Let us reason by contradiction. Let ff be an element of Wε∗W^*_{\varepsilon} and gg an element of G such that hfhf = ι(g)\iota (g). We therefore have f(t)=h(t)−1gf(t) =h(t)^{-1}g for all t∈[0,1[t\in [0,1[, so that the sequence (ti)(t_i) is also adapted to the function ff. If the value taken by ff on the interval [ti−1,ti[[t_{i-1}, t_i[ does not belong to W, then ti−ti−1<εt_i-t_{i-1}< \varepsilon, since f∈Wε∗f\in W^*_{\varepsilon}. This inequality being false, by definition of ε\varepsilon, we have f(t)∈Wf(t)\in W for all t∈[0,1[t\in [0,1[. Let ii and jj then be elements of {0,...,n−1}\{0, . . . , n-1\} such that h(ti)=h̸(tj)h(t_i)=\not h(t_j); we have

h(ti)−1h(tj)=f(ti)g−1gf(tj)−1=f(ti)f(tj)−1∈W⋅W−1 h(t_i)^{-1}h(t_j) =f(t_i)g^{-1}gf(t_j)^{-1}=f(t_i)f(t_j)^{-1}\in W\cdot W^{-1}

which contradicts the choice of W. The subgroup ι(G)\iota (G) of G∗G^* is therefore closed.

Suppose that G is a metrisable topological group and let dd be a distance on G which defines its topology. Let hh and h′∈G∗h'\in G^* and let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to hh and h′h'. The real number

∑i=1n(ti−ti−1)d(h(ti−1),h′(ti−1)) \sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), h'(t_{i-1}))

does not depend on the sequence (ti)(t_i) chosen; let us denote it by d∗(h,h′)d^*(h, h').

Let us prove that d∗d^* is a distance on G∗G^*. We have d∗(h,h′)=d^*(h, h') = d∗(h′,h)d^*(h', h) for h,h′∈G∗h,h'\in G^*, and d∗(h,h)=0d^*(h, h) = 0 for all h∈G∗h\in G^*. Conversely, let hh and h′h' be elements of G∗G^* such that d∗(h,h′)=0d^*(h, h') = 0. Let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to hh and h′h'. Since

0=d∗(h,h′)=∑i=1n(ti−ti−1)d(h(ti−1),h′(ti−1)) 0 =d^*(h, h') =\sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), h'(t_{i-1}))

and all the terms of this sum are positive or zero, we have d(h(ti−1),h′(ti−1))=0d(h(t_{i-1}), h'(t_{i-1})) = 0 for all i∈{1,...,n}i\in \{1, . . . , n\}, whence h=h′h=h'. Finally, let h,h′,h′′h,h',h'' be elements of G∗G^* and let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to each of them. Then,

d∗(h,h′′)=∑i=1n(ti−ti−1)d(h(ti−1),h′′(ti−1)) d^*(h, h'') =\sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), h''(t_{i-1}))
⩽∑i=1n(ti−ti−1)(d(h(ti−1),h′(ti−1))+d(h(ti−1),h′′(ti−1))) \leqslant \sum_{i=1}^n(t_i-t_{i-1})(d(h(t_{i-1}), h'(t_{i-1}))+d(h(t_{i-1}), h''(t_{i-1})))
=d∗(h,h′)+d∗(h,h′′) =d^*(h, h') +d^*(h, h'')

therefore d∗d^* satisfies the triangle inequality.

This distance d∗d^* is invariant under right (resp. left) translations if dd is.

Proposition 7 024N

Suppose that G is a metrizable topological group. Then the topological group G∗G^* is metrizable. More precisely, if dd is a bounded distance on G which defines its topology, the topology of G∗G^* is defined by the distance d∗d^*.

Let dd be a distance on G which defines its topology. Then, the mapping d′d' given by d′(h,h′)=d'(h, h') = inf(d(h,h′),1d(h, h'),1) is a bounded distance on G which also defines its topology (TG, IX, p. 3). It is therefore enough to prove that d∗d^* defines the topology of G∗G^* under the assumption that dd is bounded.

Let V be a neighbourhood of ee in G and let ε\varepsilon be a strictly positive real number. Let δ\delta be a strictly positive real number such that V contains the ball B(e,δ)B(e, \delta ). Let h∈G∗h\in G^* and let (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} be a sequence adapted to hh. Then,

pV(h)=∑h(tii−=11)∉Vn(ti−ti−1) p_V(h) =\sum_{h(t_ii_-=1_1)\notin V}^n(t_i-t_{i-1})
⩽∑n(ti−ti−1)d(h(ti−1),e)δ \leqslant \sum^n(t_i-t_{i-1})\frac{d(h(t_{i-1}), e)}{\delta}
i=1 i=1
d(h(ti−1),e)⩾δd(h(t_{i-1}),e)\geqslant \delta
⩽d∗(h,e∗)δ \leqslant \frac{d^*(h, e^*)}{\delta}

Consequently, the ball B(e∗,εδ)B(e^*, \varepsilon \delta ) in G∗G^* is contained in Vε∗V_{\varepsilon}^*.

Conversely, let δ\delta be a strictly positive real number and let Δ be a strictly positive upper bound for the diameter of G. Let V be a neighbourhood of ee in G contained in the ball B(e,δ/2)B(e, \delta /2). For a function h∈G∗h\in G^* and a sequence (ti)0⩽i⩽n(t_i)_{0\leqslant i\leqslant n} adapted to hh, one has

d∗(h,e∗)=∑i=1n(ti−ti−1)d(h(ti−1),e) d^*(h, e^*) =\sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), e)

= ∑i=1n(ti−ti−1)d(h(ti−1),e)\sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), e)

d(h(ti−1),e)⩽δ/2d(h(t_{i-1}),e)\leqslant \delta /2

+ ∑i=1n(ti−ti−1)d(h(ti−1),e)\sum_{i=1}^n(t_i-t_{i-1})d(h(t_{i-1}), e)

d(h(ti−1),e)>δ/2d(h(t_{i-1}),e)>\delta /2 ⩽δ2+Δ∑i=1n(ti−ti−1)\leqslant \frac{\delta}{2}+ \Delta \sum_{i=1}^n(t_i-t_{i-1}) h(ti−1)∉Vh(t_{i-1})\notin V
⩽δ2+ΔpV(h) \leqslant \frac{\delta}{2}+ \Delta p_V(h)

The preceding inequality implies that, for every element hh of Vδ/2Δ∗V_{\delta /2\Delta}^*, one has d∗(h,e∗)⩽δd^*(h, e^*)\leqslant \delta. Consequently, every ball of G∗G^* for the distance d∗d^* contains a neighbourhood of the identity element.

Remark 2 024O

Let dd be a bounded distance which defines the topology of G. When the topological group G∗G^* is endowed with the distance d∗d^*, the homomorphism ι:G→G∗\iota : G\rightarrow G^* is an isometry.

Remark 3 024P

Suppose that G is a discrete topological group. The subgroup ι(G)\iota (G) is then a discrete subgroup of G∗G^*. In fact, the topology of G is defined by the distance dd on G given by d(g,g′)=1d(g, g') = 1 if g≠g′g=\not g' and d(g,g′)=0d(g, g') = 0 otherwise. The assertion then follows from the preceding remark.

Theorem 2 024Q

Let G be a discrete topological group. Let G operate on the right in G∗G^* by h⋅g=hι(g)h\cdot g=h\iota (g) and let B denote the quotient topological space G∗/GG^*/G.

The space G∗G^* is a principal covering of B with group G; it is simply connected by paths.

The space B is a metrizable topological space, connected by paths, locally contractible, and its Poincaré group at every point is isomorphic to G.

Thus, the space B is a classifying space for the group G.

The group G∗G^* is contractible to its identity element (IV, p. 450, prop. 6). In particular it is connected by paths (III, p. 260) and simply connected by paths (IV, p. 340).

The group ι(G)\iota (G) is closed in G∗G^* (TG, III, p. 7, prop. 5), hence the space G∗/GG^*/G is metrisable (TG, III, p. 13, prop. 18 and TG, IX, p. 25, prop. 4). It is also connected by arcs (III, p. 258, prop. 3). Since the group ι(G)\iota (G) is discrete, it follows from corollary 2 of I, p. 100, that G∗G^* is a principal G-covering of B. The pointed covering (G∗,e∗)(G^*, e^*) is then a universal covering of the space B pointed at the image of e∗e^*; the Poincaré group of B (at each of its points) is isomorphic to G.

Exercises

See the exercises for § 6.

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