Bourbaki

§ 3. Poincaré groupoid

1. Poincaré groupoid

Definition 1 01YE

Let X be a topological space, let c0c_0 and c1c_1 be paths in X and σ:I×I→X\sigma :\mathbf{I}\times \mathbf{I}\rightarrow X a homotopy joining c0c_0 to c1c_1. One says that σ\sigma is a strict homotopy if the mappings s↦σ(0,s)s\mapsto \sigma (0, s) and s↦σ(1,s)s\mapsto \sigma (1, s) are constant.

One says that paths c0c_0 and c1c_1 in X are strictly homotopic if there exists a strict homotopy joining c0c_0 to c1c_1.

Two strictly homotopic paths have the same origin and the same end.

Example 01YF

Let cc be a path in X and let φ:I→I\varphi :\mathbf{I}\rightarrow \mathbf{I} be a continuous mapping such that φ(0)=0\varphi (0) = 0 and φ(1)=1\varphi (1) = 1. The paths cc and c∘φc\circ \varphi are strictly homotopic. In fact, the mapping σ:I×I→X\sigma :\mathbf{I}\times \mathbf{I}\rightarrow X defined by σ(t,s)=c((1−s)t+sφ(t))\sigma (t, s) =c((1-s)t+s\varphi (t)) is a strict homotopy joining cc to c∘φc\circ \varphi.

Let X be a topological space. Recall (cf. III, p. 257) that Λ(X) denotes the topological space Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) of paths in X and that for xx, y∈Xy\in X, Λx,y(X)\Lambda_{x,y}(X) is the subspace of Λ(X) formed by the paths with origin xx and end yy. The family of sets Λx,y(X)\Lambda_{x,y}(X), for x,y∈Xx,y\in X, is a partition of the space of paths in X. By the canonical bijection (III, p. 257, remark 2) of C(I×I;X)\mathscr{C}(\mathbf{I}\times \mathbf{I}; X) onto C(I;Λ(X))\mathscr{C}(\mathbf{I}; \Lambda (X)), strict homotopies correspond to paths c:I→Λ(X)c:\mathbf{I}\rightarrow \Lambda (X) whose image is contained in a subspace of the form Λx,y(X)\Lambda_{x,y}(X). The relation “the paths c0c_0 and c1c_1 are strictly homotopic” is therefore an equivalence relation in Λ(X) (III, p. 259, prop. 5) and the equivalence classes for this relation are the arcwise connected components of the subspaces Λx,y(X)\Lambda_{x,y}(X) of the space of paths in X. One denotes by ϖx,y(X)\varpi_{x,y}(X) the set π0(Λx,y(X))\pi_0(\Lambda_{x,y}(X)) and calls every element of ϖx,y(X)\varpi_{x,y}(X) a class of paths joining xx to yy.

Definition 2 01YG

One calls the space of loops of X, and denotes by Ω(X), the subspace of Λ(X) consisting of loops (III, p. 256, def. 1) in X.

One denotes by Ωx(X)\Omega_x(X) the set Λx,x(X)\Lambda_{x,x}(X). The elements of Ωx(X)\Omega_x(X) are called loops in X at xx and the elements of ϖx,x(X)\varpi_{x,x}(X) are called classes of loops in X at xx. The constant mapping ex:I→Xe_x:\mathbf{I}\rightarrow X with image xx is a loop, called the constant loop at xx; its strict homotopy class is denoted by εx\varepsilon_x. The mapping x↦exx\mapsto e_x of X into Λ(X) is continuous (III, p. 257, prop. 1).

Let X be a topological space and let x,y,zx,y,z be points of X. By passing to arcwise connected components, one deduces from the continuous mapping c↦c‾c\mapsto \overline{c} of Λx,y(X)\Lambda_{x,y}(X) into Λy,x(X)\Lambda_{y,x}(X) (III, p. 258, corollary) a mapping of ϖx,y(X)\varpi_{x,y}(X) into ϖy,x(X)\varpi_{y,x}(X) which is denoted by γ↦γ‾\gamma \mapsto \overline{\gamma}. If γ∈\gamma \in ϖx,y(X),γ‾\varpi_{x,y}(X),\overline{\gamma} is called the inverse of the path class γ\gamma.

Analogously, if one identifies the sets π0(Λx,y(X))×π0(Λy,z(X))\pi_0(\Lambda_{x,y}(X))\times \pi_0(\Lambda_{y,z}(X)) and π0(Λx,y(X)×Λy,z(X))\pi_0(\Lambda_{x,y}(X)\times \Lambda_{y,z}(X)) (III, p. 260, Prop. 6), one deduces from the continuous mapping (c,d)↦c∗d(c, d)\mapsto c*d of Λx,y(X)×Λy,z(X)\Lambda_{x,y}(X)\times \Lambda_{y,z}(X) into Λx,z(X)\Lambda_{x,z}(X) (III, p. 258, corollary), by passing to arcwise connected components, a mapping Cx,y,z:ϖx,y(X)×ϖy,z(X)→ϖx,z(X)C_{x,y,z}:\varpi_{x,y}(X)\times \varpi_{y,z}(X)\rightarrow \varpi_{x,z}(X). For γ∈ϖx,y(X)\gamma \in \varpi_{x,y}(X) and δ∈ϖy,z(X)\delta \in \varpi_{y,z}(X), the strict homotopy class Cx,y,z(γ,δ)C_{x,y,z}(\gamma , \delta ) is denoted by γδ\gamma \delta. It is called the composite of the composable path classes γ\gamma and δ\delta.

One has γ‾‾=γ\overline{\overline{\gamma}}=\gamma and γδ‾=δγ\overline{\gamma \delta}=\delta \gamma.

Proposition 1 01YH

Let X be a topological space, let x,y,z,ux,y,z,u be points of X, and let γ1∈ϖx,y(X),γ2∈ϖy,z(X),γ3∈ϖz,u(X)\gamma_1\in \varpi_{x,y}(X),\gamma_2\in \varpi_{y,z}(X),\gamma_3\in \varpi_{z,u}(X) be path classes. One has

εxγ1=γ1εy=γ1(1) \varepsilon_x\gamma_1=\gamma_1\varepsilon_y=\gamma_1 \tag{1}
γ1γ‾1=εx,γ‾1γ1=εy(2) \gamma_1\overline{\gamma}_1=\varepsilon_x,\overline{\gamma}_1\gamma_1=\varepsilon_y \tag{2}
(γ1γ2)γ3=γ1(γ2γ3)(3) (\gamma_1\gamma_2)\gamma_3=\gamma_1(\gamma_2\gamma_3) \tag{3}

Let c1∈Λx,y(X),c2∈Λy,z(X),c3∈Λz,u(X)c_1\in \Lambda_{x,y}(X),c_2\in \Lambda_{y,z}(X),c_3\in \Lambda_{z,u}(X) be representatives of γ1,γ2\gamma_1,\gamma_2 and γ3\gamma_3 respectively. Let φ:I→I\varphi :\mathbf{I}\rightarrow \mathbf{I} be the function defined by

t/2t/2 for 0⩽t⩽1/20\leqslant t\leqslant 1/2,

(4) φ(t)=t−1/4\varphi (t) =t-1/4 for 1/2⩽t⩽3/41/2\leqslant t\leqslant 3/4,

2t−12t-1 for 3/4⩽t⩽13/4\leqslant t\leqslant 1.

The function φ\varphi is affine on each of the three intervals [0,1/2][0,1/2], [1/2,3/4][1/2,3/4] and [3/4,1][3/4,1]; it is therefore continuous. One has φ(0)=0\varphi (0) = 0 and φ(1)=1\varphi (1) = 1. It follows from formula (1) of III, p. 256 defining the juxtaposition of paths that

c1∗(c2∗c3)=((c1∗c2)∗c3)∘φ c_1*(c_2*c_3) = ((c_1*c_2)*c_3)\circ \varphi

By the example of III, p. 289, the paths c1∗(c2∗c3)c_1*(c_2*c_3) and (c1∗c2)∗c3(c_1*c_2)*c_3 are strictly homotopic, whence equality (3).

Analogously, the function ψ:I→I\psi :\mathbf{I}\rightarrow \mathbf{I} defined by

2t2t for 0⩽t⩽1/20\leqslant t\leqslant 1/2,

ψ(t)=(5) \psi (t) = \tag{5}

1 for 1/2⩽t⩽11/2\leqslant t\leqslant 1

is continuous and satisfies ψ(0)=0,ψ(1)=1\psi (0) = 0,\psi (1) = 1. The equality γ1εy=γ1\gamma_1\varepsilon_y=\gamma_1 then follows from the fact that

c1∗ey=c1∘ψ c_1*e_y=c_1\circ \psi

and the equality εxγ1=γ1\varepsilon_x\gamma_1=\gamma_1 is proved analogously, whence (1).

The mapping σ:I×I→X\sigma :\mathbf{I}\times \mathbf{I}\rightarrow X defined by

c1(2ts)c_1(2ts) for 0⩽t⩽1/20\leqslant t\leqslant 1/2,

σ(t,s)=(6) \sigma (t, s) = \tag{6}

c1(2(1−t)s)c_1(2(1-t)s) for 1/2⩽t⩽11/2\leqslant t\leqslant 1

is continuous; it is a strict homotopy joining the path exe_x to the path c1∗c1‾c_1*\overline{c_1}, whence the first equality in (2). The second results from the first and from the fact that, for every path cc, one has c=c‾‾c=\overline{\overline{c}}.

Remark 1 01YI

Let X be a topological space. Let nn be an integer ⩾1\geqslant 1 and let (c1,...,cn)(c_1, . . . , c_n) be a sequence of paths in X such that cic_i and ci+1c_{i+1} are juxtaposable for 1⩽i⩽n−11\leqslant i\leqslant n-1 (such a sequence is called a sequence of juxtaposable paths). Let us denote by cc the path

c1∗(c2∗(⋯∗(cn−1∗cn)...)) c_1*(c_2*(\cdots *(c_{n-1}*c_n). . .))

and by c′c' the path defined by c′(t)=ci(nt−i+1)c'(t) =c_i(nt-i+ 1) for 1⩽i⩽n1\leqslant i\leqslant n and t∈[i−1n,in]t\in [\frac{i-1}{n},\frac{i}{n}]. The paths cc and c′c' have the same image and are strictly homotopic: one is the composite of the other with a homeomorphism of I\mathbf{I} leaving 0 and 1 fixed (cf. III, p. 289, example). We shall sometimes denote by c1∗c2∗⋯∗cnc_1*c_2* \cdots *c_n the path c′c'.

There exists a unique oriented graph ϖ(X)\varpi (X) whose set of vertices is X and whose set of arrows joining a point xx to a point yy is ϖx,y(X)\varpi_{x,y}(X), and in which the mappings Cx,y,zC_{x,y,z} define a law of composition. By proposition 1,ϖ(X)1,\varpi (X) is a groupoid (II, p. 162, def. 4). For every x∈Xx\in X, the identity element at the vertex xx of this groupoid is the class of the constant loop with image xx. The inverse of an arrow γ\gamma is the arrow γ‾\overline{\gamma}, which we shall also denote by γ−1\gamma^{-1}. In particular, the law of composition Cx,x,xC_{x,x,x} endows, for every x∈Xx\in X, the set ϖx,x(X)\varpi_{x,x}(X) with a group structure; this group is denoted by π1(X,x)\pi_1(X, x).

Definition 3 01YJ

Let X be a topological space. The groupoid ϖ(X)\varpi (X) is called the Poincaré groupoid, or fundamental groupoid, of the space X. Let xx be a point of X; the group π1(X,x)\pi_1(X, x) of classes of loops at xx is called the Poincaré group, or fundamental group, of the space X at the point xx.

Let U\mathscr{U} be a set of subsets of X whose interiors cover X. The classes of paths in X whose image is contained in one of the subsets belonging to U\mathscr{U} generate the groupoid ϖ(X)\varpi (X) (lemma 4 of III, p. 272).

The orbits of the groupoid ϖ(X)\varpi (X) (II, p. 162) coincide with the arcwise connected components of the space X. In particular (loc. cit.), one has:

Proposition 2 01YK

Let X be a topological space.

a) If the space X is arcwise connected, the groups π1(X,x)\pi_1(X, x) and π1(X,y)\pi_1(X, y) are isomorphic for all points xx and yy of X.

b) Let xx be a point of X; the following conditions are equivalent:

(i) The group π1(X,x)\pi_1(X, x) is trivial;

(ii) Two paths with origin xx in X which have the same end

are strictly homotopic;

(iii) Every loop with origin xx in X is strictly homotopic

to the constant loop with image xx.

More precisely, let X be an arcwise connected topological space and let xx and yy be points of X. For every element δ\delta of ϖx,y(X)\varpi_{x,y}(X), the mapping uδ:γ↦δγδ−1u_{\delta}:\gamma \mapsto \delta \gamma \delta^{-1} of π1(X,y)\pi_1(X, y) into π1(X,x)\pi_1(X, x) is a group isomorphism whose inverse isomorphism is uδ−1u_{\delta^{-1}}. For δ∈ϖx,y(X)\delta \in \varpi_{x,y}(X) and γ∈π1(X,y)\gamma \in \pi_1(X, y), we set δγ=uδ(γ)^{\delta}\gamma =u_{\delta}(\gamma ). When x=yx=y, one has δγ=^{\delta}\gamma = Int(δ\delta )(γ)(\gamma ), i.e. uδ=u_{\delta}= Int(δ\delta ).

Let x,y,zx,y,z be points of X. For δ∈ϖx,y(X)\delta \in \varpi_{x,y}(X) and η∈ϖy,z(X)\eta \in \varpi_{y,z}(X), one has uδη=uδ∘uηu_{\delta \eta}=u_{\delta}\circ u_{\eta}, which may also be written δηγ=δ(ηγ)^{\delta \eta}\gamma =^{\delta}(^{\eta}\gamma ) for γ∈π1(X,z)\gamma \in \pi_1(X, z).

Let x,yx,y be points of X and let δ,δ′\delta ,\delta ' be elements of ϖx,y(X)\varpi_{x,y}(X). One has

(7) uδ′=uδ∘u_{\delta'}=u_{\delta}\circ Int(δ−1δ′\delta^{-1}\delta ') == Int(δ′δ−1\delta '\delta^{-1})∘uδ\circ u_{\delta}.

Remark 2 01YL

Let X be a topological space and let C be an arcwise connected component of X. If xx and yy are points of C, the topological space Λx,y(C)\Lambda_{x,y}(C) is identified with the topological space Λx,y(X)\Lambda_{x,y}(X), so that the set ϖx,y(C)\varpi_{x,y}(C) is identified with the set ϖx,y(X)\varpi_{x,y}(X). Thus, the fundamental groupoid ϖ(C)\varpi (C) is identified with the full subgroupoid of ϖ(X)\varpi (X) having C as set of points. In particular, for every point xx of C, the group π1(C,x)\pi_1(C, x) is identified with the group π1(X,x)\pi_1(X, x).

Remark 3 01YM

Let X be a topological space and let x,y,zx,y,z be points of X. The mapping (c,d)↦c∗d(c, d)\mapsto c*d of Λx,y(X)×Λy,z(X)\Lambda_{x,y}(X)\times \Lambda_{y,z}(X) into Λx,z(X)\Lambda_{x,z}(X) is continuous (III, p. 258, corollary of Prop. 2). It should be noted that the composition mapping ϖx,y(X)×ϖy,z(X)→ϖx,z(X)\varpi_{x,y}(X)\times \varpi_{y,z}(X)\rightarrow \varpi_{x,z}(X) deduced from it is not necessarily continuous when the sets ϖx,y(X),ϖy,z(X)\varpi_{x,y}(X),\varpi_{y,z}(X) and ϖx,z(X)\varpi_{x,z}(X) are endowed with the quotient topologies (cf. TG, I, p. 35 and III, p. 259). However, for every γ0∈ϖx,y(X)\gamma_0\in \varpi_{x,y}(X) and every δ0∈ϖy,z(X)\delta_0\in \varpi_{y,z}(X), the partial mappings γ↦γδ0\gamma \mapsto \gamma \delta_0 of ϖx,y(X)\varpi_{x,y}(X) into ϖx,z(X)\varpi_{x,z}(X) and δ↦γ0δ\delta \mapsto \gamma_0\delta of ϖy,z(X)\varpi_{y,z}(X) into ϖx,z(X)\varpi_{x,z}(X) are homeomorphisms. In fact, let c0∈Λx,y(X)c_0\in \Lambda_{x,y}(X) be a path of class γ0\gamma_0. The mapping d↦c0∗dd\mapsto c_0*d is a continuous mapping of Λy,z(X)\Lambda_{y,z}(X) into Λx,z(X)\Lambda_{x,z}(X). The mapping δ↦γ0δ\delta \mapsto \gamma_0\delta is deduced from it by passing to quotients, and is therefore continuous. The same holds for the mapping δ′↦γ0−1δ′\delta '\mapsto \gamma_0^{-1}\delta ' of ϖx,z(X)\varpi_{x,z}(X) into ϖy,z(X)\varpi_{y,z}(X). Since these two mappings are inverse to one another, they are homeomorphisms. One reasons analogously for the mapping γ↦γδ0\gamma \mapsto \gamma \delta_0. See also IV, p. 374.

2. Functoriality of the Poincaré groupoid

Let X, Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. The mapping c↦f∘cc\mapsto f\circ c is a continuous mapping, denoted Λ(f)\Lambda (f), of Λ(X) = Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) into Λ(Y) = Cc(I;Y)\mathscr{C}_c(\mathbf{I}; Y) (I, p. 132, lemma). By passing to subsets it defines continuous mappings

Λx(f):Λx(X)→Λf(x)(Y) \Lambda_x(f): \Lambda_x(X)\rightarrow \Lambda_{f(x)}(Y)

for x∈Xx\in X,

Λx,y(f):Λx,y(X)→Λf(x),f(y)(Y) \Lambda_{x,y}(f): \Lambda_{x,y}(X)\rightarrow \Lambda_{f(x),f(y)}(Y)

for x,y∈Xx,y\in X and

Ω(f):Ω(X)→Ω(Y) \Omega (f): \Omega (X)\rightarrow \Omega (Y)

For x∈Xx\in X, the mapping Λx,x(f)\Lambda_{x,x}(f) is also denoted Ωx(f)\Omega_x(f). By passing to path-connected components (III, p. 290), one deduces from the mapping Λx,y(f)\Lambda_{x,y}(f) a mapping

ϖx,y(f):ϖx,y(X)→ϖf(x),f(y)(Y) \varpi_{x,y}(f):\varpi_{x,y}(X)\rightarrow \varpi_{f(x),f(y)}(Y)

Let x,y,zx,y,z be points of X and let c∈Λx,y(X),d∈Λy,z(X)c\in \Lambda_{x,y}(X),d\in \Lambda_{y,z}(X) be paths; by definition of the juxtaposition of paths, one has

f∘(c∗d)=(f∘c)∗(f∘d) f\circ (c*d) = (f\circ c)*(f\circ d)

By passing to strict homotopy classes, one obtains the relation

ϖx,z(f)(γδ)=(ϖx,y(f)(γ))(ϖy,z(f)(δ)) \varpi_{x,z}(f)(\gamma \delta ) = (\varpi_{x,y}(f)(\gamma ))(\varpi_{y,z}(f)(\delta ))

for every γ∈ϖx,y(X)\gamma \in \varpi_{x,y}(X) and every δ∈ϖy,z(X)\delta \in \varpi_{y,z}(X). Thus, the continuous mapping ff and the mappings ϖx,y(f)\varpi_{x,y}(f), for xx and y∈y\in X, define a morphism of the groupoid ϖ(X)\varpi (X) into the groupoid ϖ(Y)\varpi (Y) (II, p. 161, def. 3). It is denoted by ϖ(f)\varpi (f) and is called the morphism of Poincaré groupoids deduced from the continuous mapping ff. In particular, if xx is a point of X, the mapping ϖx,x(f)\varpi_{x,x}(f) is a homomorphism of the group π1(X,x)\pi_1(X, x) into the group π1(X,f(x))\pi_1(X, f(x)); this homomorphism is also denoted by π1(f,x)\pi_1(f, x).

Remark 1 01YN

The homomorphism ϖx,y(f)\varpi_{x,y}(f) is continuous if one endows the sets ϖx,y(X)\varpi_{x,y}(X) and ϖf(x),f(y)(Y)\varpi_{f(x),f(y)}(Y) with the quotient topology of the topology of compact convergence on Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) and Cc(I;Y)\mathscr{C}_c(\mathbf{I}; Y).

To simplify the notation, if x,y∈Xx,y\in X and γ∈ϖx,y(X)\gamma \in \varpi_{x,y}(X), we shall sometimes write f∗(γ)f_*(\gamma ) for the element ϖx,y(f)(γ)\varpi_{x,y}(f)(\gamma ) of ϖf(x),f(y)(Y)\varpi_{f(x),f(y)}(Y).

Let X, Y, Z be topological spaces, let f:X→Y,g:Y→Zf: X\rightarrow Y,g: Y\rightarrow Z be continuous mappings. For every path cc in X, one has (g∘f)∘c=(g\circ f)\circ c= g∘(f∘c)g\circ (f\circ c). It follows that one has

ϖ(g∘f)=ϖ(g)∘ϖ(f) \varpi (g\circ f) =\varpi (g)\circ \varpi (f)

Let X and Y be topological spaces, let f0f_0 and f1f_1 be continuous mappings of X into Y, and let σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y be a homotopy joining f0f_0 to f1f_1. Since the mapping (c,t)↦c(t)(c, t)\mapsto c(t) of Cc(I;X)×I\mathscr{C}_c(\mathbf{I}; X)\times \mathbf{I} into X (III, p. 257, prop. 1) is continuous, the same is true of the mapping of Cc(I;X)×I×I\mathscr{C}_c(\mathbf{I}; X)\times \mathbf{I}\times \mathbf{I} into Y given by (c,t,s)↦σ(c(t),s)(c, t, s)\mapsto \sigma (c(t), s). Consequently, the mapping Σ:(c,s)↦σ(c(⋅),s)\Sigma : (c, s)\mapsto \sigma (c(\cdot ), s) is a continuous mapping of Cc(I;X)×I\mathscr{C}_c(\mathbf{I}; X)\times \mathbf{I} into Cc(I;Y)\mathscr{C}_c(\mathbf{I}; Y) (loc. cit.). The mapping Σ is a homotopy joining the mapping Λ(f0)\Lambda (f_0) to the mapping Λ(f1)\Lambda (f_1). By restriction to the loop spaces, the mapping Σ induces a homotopy Ω(X)×I→Ω(Y)\Omega (X)\times \mathbf{I}\rightarrow \Omega (Y) joining Ω(f0)\Omega (f_0) to Ω(f1)\Omega (f_1). Let xx be a point of X; suppose that the homotopy σ\sigma is a homotopy pointed at xx, and put y=f0(x)=f1(x)y=f_0(x) =f_1(x). The mapping Σ then induces a continuous mapping of Ωx(X)×I\Omega_x(X)\times \mathbf{I} into Ωy(Y)\Omega_y(Y) which is a homotopy pointed at exe_x, joining the mapping Ωx(f0)\Omega_x(f_0) to the mapping Ωx(f1)\Omega_x(f_1).

Proposition 3 01YO

Let X and Y be topological spaces, f0f_0 and f1f_1 continuous mappings of X into Y, and let σ:X×I→\sigma : X\times \mathbf{I}\rightarrow Y be a homotopy joining f0f_0 to f1f_1. Let xx be a point of X; put y0=f0(x)y_0=f_0(x), y1=f1(x)y_1=f_1(x), and denote by δ∈ϖy0,y1(Y)\delta \in \varpi_{y_0,y_1}(Y) the class of the path dd defined by d(t)=σ(x,t)d(t) =\sigma (x, t) for t∈It\in \mathbf{I}. For every γ∈π1(X,x)\gamma \in \pi_1(X, x), one has (f1)∗(γ)=(f_1)_*(\gamma ) = δ−1((f0)∗(γ))δ\delta^{-1}((f_0)_*(\gamma ))\delta.

Let cc be a loop in X at xx and let γ\gamma be its strict homotopy class. Put γ0=(f0)∗(γ)\gamma_0= (f_0)_*(\gamma ) and γ1=(f1)∗(γ)\gamma_1= (f_1)_*(\gamma ); these are the strict homotopy classes of f0∘cf_0\circ c and f1∘cf_1\circ c. For (t,s)∈I×I(t, s)\in \mathbf{I}\times \mathbf{I}, put φ(t,s)=σ(c(t),s)\varphi (t, s) =\sigma (c(t), s). For every t∈It\in \mathbf{I}, one has φ(t,0)=(f0∘c)(t),φ(t,1)=(f1∘c)(t)\varphi (t,0) = (f_0\circ c)(t),\varphi (t,1) = (f_1\circ c)(t) and φ(0,t)=φ(1,t)=d(t)\varphi (0, t) =\varphi (1, t) =d(t). The relation γ0δ=δγ1\gamma_0\delta =\delta \gamma_1 therefore follows from the following lemma.

Lemma 1 01YP

Let Y be a topological space and let φ:I×I→Y\varphi :\mathbf{I}\times \mathbf{I}\rightarrow Y be a continuous mapping. For t∈It\in \mathbf{I}, put c0(t)=φ(t,0),c1(t)=φ(t,1)c_0(t) =\varphi (t,0),c_1(t) =\varphi (t,1), d0(t)=φ(0,t)d_0(t) =\varphi (0, t) and d1(t)=φ(1,t)d_1(t) =\varphi (1, t). The paths c0∗d1c_0*d_1 and d0∗c1d_0*c_1 are strictly homotopic.

Let cc denote the path in I×I\mathbf{I}\times \mathbf{I} obtained by juxtaposing the paths t↦(t,0)t\mapsto (t,0) and t↦(1,t)t\mapsto (1, t); let dd denote the path in I×I\mathbf{I}\times \mathbf{I} obtained by juxtaposing the paths t↦(0,t)t\mapsto (0, t) and t↦(t,1)t\mapsto (t,1). The paths cc and dd have the same origin (0,0)(0,0) and the same end (1,1)(1,1). The mapping (t,s)↦(t, s)\mapsto (1−s)c(t)+sd(t)(1-s)c(t)+sd(t) of I×I\mathbf{I}\times \mathbf{I} into I×I\mathbf{I}\times \mathbf{I} is a strict homotopy joining cc to dd. One has c0∗d1=φ∘cc_0*d_1=\varphi \circ c and d0∗c1=φ∘dd_0*c_1=\varphi \circ d; these two paths are therefore strictly homotopic.

Corollary 1 01YQ

Let X and Y be topological spaces and let xx be a point of X. Let f0f_0 and f1f_1 be continuous mappings of X into Y. If there exists a pointed homotopy at xx joining f0f_0 to f1f_1, one has π1(f0,x)=\pi_1(f_0, x) = π1(f1,x)\pi_1(f_1, x).

Let σ\sigma be a pointed homotopy at xx joining f0f_0 to f1f_1. With the notation of Prop. 3,δ3,\delta is the class of a constant path with image f0(x)=f_0(x) = f1(x)f_1(x). The assertion follows.

Corollary 2 01YR

Let X and Y be topological spaces and let f:X→f: X\rightarrow Y be a homeomorphism. For every point xx of X, the homomorphism

π1(f,x):π1(X,x)→π1(Y,f(x)) \pi_1(f, x):\pi_1(X, x)\rightarrow \pi_1(Y, f(x))

is an isomorphism.

Let gg be a continuous mapping of Y into X, inverse to ff up to homotopy. Let xx be a point of X. It follows from Prop. 3 applied to the homotopic mappings IdX_X and g∘f:X→Xg\circ f: X\rightarrow X that the mapping π1(g∘f,x)\pi_1(g\circ f, x) is an isomorphism of the group π1(X,x)\pi_1(X, x) onto the group π1(X,g∘f(x))\pi_1(X, g\circ f(x)). Since

π1(g∘f,x)=π1(g,f(x))∘π1(f,x) \pi_1(g\circ f, x) =\pi_1(g, f(x))\circ \pi_1(f, x)

the homomorphism π1(f,x)\pi_1(f, x) is injective and the homomorphism π1(g,f(x))\pi_1(g, f(x)) is surjective. Since the mapping gg is also a homotopy equivalence, the homomorphism π1(g,f(x))\pi_1(g, f(x)) is injective; it is therefore an isomorphism. Hence π1(f,x)\pi_1(f, x) is an isomorphism.

Example 01YS

Let G be a topological group, let ee be its identity element. For every point gg of G, the left and right translations, x↦gxx\mapsto gx and x↦xgx\mapsto xg, are homeomorphisms of G onto itself (TG, III, p. 2) which map ee onto gg. By Corollary 2, they induce isomorphisms of π1(G,e)\pi_1(G, e) onto π1(G,g)\pi_1(G, g). These isomorphisms are not necessarily equal (IV, p. 459, exerc. 1).

Corollary 3 01YT

Let X be a topological space homotopic to a point. For every point xx of X, the group π1(X,x)\pi_1(X, x) is reduced to the identity element.

Cor. 3 applies in particular when X is the numerical space of nn dimensions Rn\mathbf{R}^n and more generally when X is a subset of the space Rn\mathbf{R}^n which is star-shaped (III, p. 234) with respect to one of its points.

Proposition 4 01YU

Let X be the product space of a family (Xj)j∈J(X_j)_{j\in J} of topological spaces. The morphism of the groupoid ϖ(X)\varpi (X) into the product of the groupoids ϖ(Xj)\varpi (X_j), for j∈Jj\in J, defined by the family of morphisms (ϖ(\varpi(prj_j))j∈J_{j\in J} is an isomorphism.

Let φ\varphi denote this morphism of groupoids. The mapping deduced from it by passing to the vertices is the identity mapping X→∏jXjX\rightarrow \prod_jX_j. Let x=(xj)x= (x_j) and y=(yj)y= (y_j) be two points of X. Let φx,y\varphi_{x,y} denote the mapping of ϖx,y(X)\varpi_{x,y}(X) into ∏jϖxj,yj(Xj)\prod_j\varpi_{x_j,y_j}(X_j) deduced from φ\varphi. If for every j∈J,cj:I→j\in J,c_j:\mathbf{I}\rightarrow XjX_j is a path joining xjx_j to yjy_j, the mapping t↦(cj(t))t\mapsto (c_j(t)) is a path in X joining xx to yy (TG, I, p. 25, prop. 1). This proves that φx,y\varphi_{x,y} is surjective. Let cc and dd be two paths in X joining xx to yy. Suppose that for every j∈Jj\in J there exists a strict homotopy σj:I×I→Xj\sigma_j:\mathbf{I}\times \mathbf{I}\rightarrow X_j joining prj∘c_j\circ c to prj∘d_j\circ d. Then the mapping (t,s)↦(σj(t,s))(t, s)\mapsto (\sigma_j(t, s)) of I×I\mathbf{I}\times \mathbf{I} into X is a strict homotopy joining cc to d(d(loc. cit.). This proves that φx,y\varphi_{x,y} is injective.

Corollary 01YV

Let x=(xj)j∈Jx= (x_j)_{j\in J} be a point of X. The mapping

π1(X,x)→∏j∈Jπ1(Xj,xj) \pi_1(X, x)\rightarrow \prod_{j\in J}\pi_1(X_j, x_j)

deduced from the mappings π1\pi_1(prj,xj_j, x_j) is a group isomorphism.

This isomorphism is said to be canonical. In the sequel, we shall often identify π1(X,x)\pi_1(X, x) with ∏j∈Jπ1(Xj,xj)\prod_{j\in J}\pi_1(X_j, x_j) by means of this isomorphism.

Remark 2 01YW

Let (Xj)j∈J(X_j)_{j\in J} be a family of topological spaces. Let X denote the product topological space ∏j∈JXj\prod_{j\in J}X_j and let x=(xj)x= (x_j) be a point of X.

For every i∈Ji\in J, let ui:Xi→Xu_i: X_i\rightarrow X be the mapping such that, for z∈Xiz\in X_i, pri∘ui(z)=z_i\circ u_i(z) =z and, for every j∈Jj\in J distinct from ii, prj∘ui(z)=_j\circ u_i(z) = xjx_j. The mapping uiu_i is continuous and the mapping π1(ui,xi)\pi_1(u_i, x_i) is identified with the canonical injection of the factor π1(Xi,xi)\pi_1(X_i, x_i) into the product group of the family (π1(Xj,xj))j∈J((\pi_1(X_j, x_j))_{j\in J}(cf. A, I, p. 45).

Suppose that the set J is finite and, for every j∈Jj\in J, let γj\gamma_j be an element of π1(Xj,xj)\pi_1(X_j, x_j). The element (γj)(\gamma_j) of π1(X,x)\pi_1(X, x) is the composite of the classes of loops (uj)∗(γj),j∈J(u_j)_*(\gamma_j),j\in J, these classes being pairwise permutable.

Remark 3 01YX

Let (Xj)j∈J(X_j)_{j\in J} be a family of topological spaces. Let X denote the product topological space ∏j∈JXj\prod_{j\in J}X_j and let x=(xj)x= (x_j) be a point of X.

Endow the sets π1(X,x)\pi_1(X, x) and π1(Xj,xj)\pi_1(X_j, x_j) with the quotient topology of the topology of compact convergence on the spaces Λx(X)\Lambda_x(X) and Λxj(Xj)\Lambda_{x_j}(X_j). The isomorphism π1(X,x)→∏j∈Jπ1(Xj,xj)\pi_1(X, x)\rightarrow \prod_{j\in J}\pi_1(X_j, x_j) is then a homeomorphism. It is continuous (III, p. 294, remark 1). The topology of compact convergence on Λ(X) is generated by the subsets of the form T(K,U)\mathbf{T}(K,U), where K is a compact subset of I\mathbf{I} and U an open subset of X. For j∈Jj\in J, let UjU_j be an open subset of XjX_j, such that ∏j∈JUj⊂U\prod_{j\in J}U_j\subset U. Then (prj_j)∗(T(K,U))_*(\mathbf{T}(K,U)) contains T(K,Uj)\mathbf{T}(K,U_j). This shows that the mappings (prj_j)∗:Λx(X)→Λxj(Xj)_*: \Lambda_x(X)\rightarrow \Lambda_{x_j}(X_j) are open, and the mappings π1\pi_1(prj,xj_j, x_j) are also open. Since they are surjective, the mapping π1(X,x)→∏j∈Jπ1(Xj,xj)\pi_1(X, x)\rightarrow \prod_{j\in J}\pi_1(X_j, x_j) is open (TG, I, p. 34, Prop. 8). Being continuous and bijective, it is a homeomorphism (TG, I, p. 30, example 2).

Proposition 5 01YY

Let X be a topological space and let (Ai)i∈I(A_i)_{i\in I} be an increasing family of subsets of X, indexed by a filtered ordered set I, such that every quasi-compact subset of X is contained in one of the AiA_i. The canonical groupoid morphism

ρ:\rho : lim⟶i∈Iϖ(Ai)→ϖ(X)\longrightarrow_{i\in I}\varpi (A_i)\rightarrow \varpi (X),

induced by the canonical injections of AiA_i into X, is an isomorphism.

If i⩽ji\leqslant j, let us denote by ρj,i\rho_{j,i} the groupoid morphism ϖ(Ai)→ϖ(Aj)\varpi (A_i)\rightarrow \varpi (A_j) induced by the injection of AiA_i into AjA_j. Since the mapping induced by ρj,i\rho_{j,i} by passing to the vertices is the injection Ai→AjA_i\rightarrow A_j and since the AiA_i cover X, the mapping induced by ρ\rho by passing to the vertices is bijective.

Let aa and bb be points of X and let cc be a path joining aa to bb in X. The image of cc is a quasi-compact subset of X (TG, I, p. 62, Theorem 2), since I\mathbf{I} is compact. There therefore exists an element i∈Ii\in I such that the image of cc is contained in AiA_i. Consequently, the mapping induced by ρ\rho by passing to the sets of arrows is surjective.

Let i∈Ii\in I, let aa and bb be points of AiA_i, let c,c′c,c' be paths joining aa to bb in AiA_i; let hh be a strict homotopy joining cc to c′c' in X. Since I×I\mathbf{I}\times \mathbf{I} is compact, h(I×I)h(\mathbf{I}\times \mathbf{I}) is a quasi-compact subset of X (loc. cit.) and there exists an element i∈Ii\in I such that the image of hh is contained in AiA_i. The paths cc and c′c' are strictly homotopic in AiA_i; fortiori, the path classes [c][c] and [c′][c'] have the same image in lim⟶ϖ(Ai)\longrightarrow \varpi (A_i). Consequently, ρ\rho is injective.

Corollary 01YZ

Let aa be a point of X and let J be the set of i∈Ii\in I such that a∈Aia\in A_i. The canonical homomorphism

lim⟶i∈Jπ1(Ai,a)→π1(X,a)\longrightarrow_{i\in J}\pi_1(A_i, a)\rightarrow \pi_1(X, a)

is bijective.

Remark 4 01Z0

The proposition and its corollary apply in particular when (Ai)i∈I(A_i)_{i\in I} is an increasing family of subsets of X indexed by a directed ordered set I such that the interiors of the AiA_i cover X.

3. Freely Homotopic Loops

Definition 4 01Z1

Let X be a topological space and let cc and c′c' be two loops in X. A free homotopy joining cc to c′c' means a homotopy σ\sigma joining cc to c′c' such that σ(0,s)=σ(1,s)\sigma (0, s) =\sigma (1, s) for every s∈Is\in \mathbf{I}. One says that cc is freely homotopic to c′c' if there exists a free homotopy joining cc to c′c'.

The free homotopies joining cc to c′c' correspond to the paths joining cc to c′c' in the space Ω(X) of loops in X. Consequently, the relation “cc is freely homotopic to c′c'” is an equivalence relation in Ω(X) whose equivalence classes are the arcwise connected components of Ω(X).

Remark 01Z2

Let φ\varphi be the canonical mapping of R\mathbf{R} onto T=R/Z\mathbf{T}=\mathbf{R}/\mathbf{Z} (TG, V, p. 2). The mapping f↦f∘φ∣If\mapsto f\circ \varphi |\mathbf{I} is a homeomorphism of Cc(T;X)\mathscr{C}_c(\mathbf{T}; X) onto Ω(X), whence, by passing to arcwise connected components, a bijection of the set [T;X][\mathbf{T}; X] (III, p. 230) onto the set of free homotopy classes of loops in X.

Proposition 6 01Z3

Let X be an arcwise connected topological space and let xx be a point of X.

a) Every loop in X is freely homotopic to a loop at xx. More precisely, if cc is a loop at yy and dd a path with origin yy and end x,cx,c is freely homotopic to the loop (d‾∗c)∗d(\overline{d}*c)*d at xx.

b) Two loops in X at xx are freely homotopic if and only if their strict homotopy classes are conjugate in the group π1(X,x)\pi_1(X, x).

Let us prove a). For every s∈[0,1]s\in [0,1], let dsd_s denote the path in X defined by ds(t)=d(st)d_s(t) =d(st) for t∈It\in \mathbf{I}; its origin is yy. Since the mapping (s,t)↦d(st)(s, t)\mapsto d(st) is continuous, the mapping s↦dss\mapsto d_s of I\mathbf{I} into Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) is continuous (III, p. 257, prop. 1). The mapping s↦(ds‾∗c)∗dss\mapsto (\overline{d_s}*c)*d_s is then a path in Ω(X) (III, p. 257, prop. 2) joining (ey∗c)∗ey(e_y*c)*e_y to (d‾∗c)∗d(\overline{d}*c)*d, whence a).

Let cc and c′c' be two loops in X at xx. If their strict homotopy classes are conjugate in π1(X,x)\pi_1(X, x), there exists a loop dd at xx such that c′c' is strictly homotopic to the loop (d‾∗c)∗d(\overline{d}*c)*d. It follows from a) that cc and c′c' are freely homotopic. Conversely, suppose that there exists a free homotopy φ\varphi joining cc to c′c'. Put d(t)=φ(0,t)d(t) =\varphi (0, t); one has also d(t)=φ(1,t)d(t) =\varphi (1, t) and dd is a loop at xx. By lemma 1 of III, p. 295, the loops c∗dc*d and d∗c′d*c' are strictly homotopic. The strict homotopy classes of cc and c′c' are therefore conjugate in π1(X,x)\pi_1(X, x).

Scholium 01Z4

Let X be an arcwise connected topological space and let xx be a point of X. Proposition 6 makes it possible to define a canonical bijection of the set of free homotopy classes of loops in X onto the set of conjugacy classes in π1(X,x)\pi_1(X, x).

Exercises

See the exercises for § 3.

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