Bourbaki

§ 2. Homotopy and paths

1. Paths

Definition 1 01WN

Let X be a topological space. A path in X is called any continuous mapping cc of I\mathbf{I} into X. The point c(0)c(0) is called the origin, the point c(1)c(1) the term of the path cc. A loop in X is called a path in X whose origin is equal to the term.

Let xx and yy be points of X. One says that a path cc in X joins xx to yy if xx is its origin and yy is its term.

Definition 2 01WO

Let X be a topological space. One says that paths cc and dd in X are juxtaposed if one has c(1)=d(0)c(1) =d(0). One then calls juxtaposed path of cc and dd the path c∗dc*d defined by the formula:

c(2t)c(2t) for 0⩽t⩽1/20\leqslant t\leqslant 1/2,

(c∗d)(t)=(1) (c*d)(t) = \tag{1}

d(2t−1)d(2t-1) for 1/2⩽t⩽11/2\leqslant t\leqslant 1.

Its origin is that of cc, its term that of dd.

Let cc be a path in X; one calls the path opposite to cc and denotes by c‾\overline{c} the path defined by c‾(t)=c(1−t)\overline{c}(t) =c(1-t) for t∈It\in \mathbf{I}.

one hasIf ccccand∗dd=*dd=are two juxtaposed paths ind‾∗c‾\overline{d}*\overline{c}. For every path cc in X, one hasX,c‾‾d‾=,\overline{\overline{\overline{c}}d}=andcc.‾c\overline{c.} are, and

Remark 1 01WP

Let X be a topological space and let P be a topological space reduced to a point. Identify P×IP\times \mathbf{I} with I\mathbf{I} by the projection pr2_2. The set C(P×I;X)\mathscr{C}(P\times \mathbf{I}; X) of homotopies between (necessarily continuous) mappings from P into X is then identified with the set C(I;X)\mathscr{C}(\mathbf{I}; X) of paths in X. To a homotopy joining the constant mapping of image xx to the constant mapping of image yy there corresponds a path with origin xx and term yy. The preceding identifications are compatible with the notions of juxtaposition and passing to the opposite (cf. III, p. 230).

Remark 2 01WQ

Let X and Y be topological spaces. The canonical mapping

C(X×I;Y)→C(I;Cc(X;Y)) \mathscr{C}(X\times \mathbf{I}; Y)\rightarrow \mathscr{C}(\mathbf{I};\mathscr{C}_c(X; Y))

associates to every homotopy σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y joining two mappings ff and gg from X into Y a path with origin ff, term gg, in the space Cc(X;Y)\mathscr{C}_c(X; Y). If X is a locally compact space, this canonical mapping is bijective (TG, X, p. 28, theorem 3).

Let X be a topological space. One calls the space of paths of X, and denotes by Λ(X), the topological space Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) whose elements are the paths in X and whose topology is that of compact convergence (cf. TG, X, p. 27). If xx is a point of X, one denotes by Λx(X)\Lambda_x(X) the subspace of Λ(X) formed by paths with origin xx. If yy is a second point of X, one denotes by Λx,y(X)\Lambda_{x,y}(X) the subspace of Λ(X) formed by paths with origin xx and term yy.

Proposition 1 01WR

Let X and Z be topological spaces. For a mapping φ\varphi of Z into the space of paths Cc(I;X)\mathscr{C}_c(\mathbf{I}; X) to be continuous, it is necessary and sufficient that the mapping (z,t)↦φ(z)(t)(z, t)\mapsto \varphi (z)(t) be a continuous mapping of Z×IZ\times \mathbf{I} into X. In particular, 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 is continuous.

The topological space I\mathbf{I} being locally compact, the proposition follows from TG, X, p. 28, th. 3.

We denote oo and call origin mapping the mapping c↦c(0)c\mapsto c(0) of Λ(X) into X; we denote ee and call term mapping the mapping c↦c(1)c\mapsto c(1) of Λ(X) into X. The mappings oo and ee are continuous (TG, X, p. 27, remark 1). The pairs of juxtaposed paths in X are the elements of the fibered product of the X-spaces (Λ(X),e)(\Lambda (X), e) and (Λ(X),o)(\Lambda (X), o).

Proposition 2 01WS

Let X be a topological space. The map which associates with a path cc the opposite path c‾\overline{c} is a homeomorphism of Λ(X) onto itself. The juxtaposition of paths (c,d)↦c∗d(c, d)\mapsto c*d is a homeomorphism of the space (Λ(X),e)×X(Λ(X),o)(\Lambda (X), e)\times_X(\Lambda (X), o) onto Λ(X).

The mapping t↦1−tt\mapsto 1-t is a homeomorphism of the space I\mathbf{I} onto itself; the first assertion follows.

Let C denote the fibered product (Λ(X),e)×X(Λ(X),o)(\Lambda (X), e)\times_X(\Lambda (X), o). Let γ:C→\gamma : C\rightarrow Λ(X) be the mapping (c,d)↦c∗d(c, d)\mapsto c*d and δ:C×I→\delta : C\times \mathbf{I}\rightarrow X the mapping ((c,d),t)↦(c∗d)(t)((c, d), t)\mapsto (c*d)(t). The restrictions of the mapping δ\delta to C×[0,12]C\times [0,\frac{1}{2}] and C×[12,1]C\times [\frac{1}{2},1] are continuous by virtue of formula (1) and of prop. 1. The mapping δ\delta is therefore continuous (TG, I, p. 19, prop. 4), as is the mapping γ\gamma (prop. 1). We have c(t)=(c∗d)(t2)c(t) = (c*d)(\frac{t}{2}) and d(t)=(c∗d)(1+t2)d(t) = (c*d)(\frac{1+t}{2}), hence γ\gamma is injective.

Let gg be a path in X; for t∈It\in \mathbf{I}, put

cg(t)=g(t2)c_g(t) =g(\frac{t}{2}) and dg(t)=g(1+t2)d_g(t) =g(\frac{1 + t}{2}).

The mappings cgc_g and dgd_g thus defined are juxtaposed paths in X and we have cg∗dg=gc_g*d_g=g. Moreover, the mappings g↦cgg\mapsto c_g and g↦dgg\mapsto d_g are continuous mappings of the space Λ(X) into itself (prop. 1). It follows that the mapping γ\gamma is a homeomorphism.

Corollary 01WT

Let X be a topological space and let x,y,zx,y,z be points of X. The mappings c↦c‾c\mapsto \overline{c} of Λx,y(X)\Lambda_{x,y}(X) into Λy,x(X)\Lambda_{y,x}(X) and (c,d)↦(c, d)\mapsto c∗dc*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) are continuous.

2. Arcwise connected spaces

Definition 3 01WU

A topological space X is said to be arcwise connected if for every pair (x,y)(x, y) of points of X, there exists a path of origin xx and of term yy.

Example 1 01WV

Every interval of R\mathbf{R}, every convex subset of a numerical space or, more generally, of a topological vector space over R\mathbf{R} is arcwise connected.

Proposition 3 01WW

The image by a continuous mapping of an arcwise connected space is arcwise connected.

Let X be a path-connected space, f:X→Yf: X\rightarrow Y a continuous mapping. Let xx and yy be two points of f(X)f(X); let x′x' and y′y' be points of X such that f(x′)=xf(x') =x and f(y′)=yf(y') =y. There exists a path cc in X whose origin is x′x' and whose term is y′y'. Then, f∘cf\circ c is a path in f(X)f(X) with origin xx and term yy.

Proposition 4 01WX

A path-connected topological space is connected.

Since every interval of R\mathbf{R} is connected, the image of a path is connected (TG, I, p. 82, prop. 4). The empty space is connected. Since a nonempty path-connected space is the union of the images of the paths issuing from one of its points, it is connected (TG, I, p. 81, prop. 2).

Taking into account the identification (III, p. 256, remark 1) of the paths in X with the homotopies between mappings of a topological space P reduced to a point into X, Proposition 1 of III, p. 230 gives:

Proposition 5 01WY

In a topological space, the relation “there exists a path with origin xx and term yy” is an equivalence relation.

The path-connected components of a topological space X are called the equivalence classes for the above relation. A path-connected component is a path-connected space. Let xx be a point of X. The path-connected component of xx is the union of the path-connected subspaces of X containing xx. It is also the union of the images of the paths with origin xx in X.

Example 2 01WZ

The union of a family of path-connected sets, whose intersection is nonempty, is path-connected (cf. TG, I, p. 81, prop. 2).

Example 3 01X0

A subset of a numerical space (or, more generally, a topological vector space over R\mathbf{R}) which is star-shaped at one of its points is path-connected.

Let X be a topological space. We denote by π0(X)\pi_0(X) the set of path-connected components of X. Let P be a topological space reduced to a point. The mapping from X into [P; X] which, to every point xx of X, associates the homotopy class φx\varphi_x of the mapping fx:P→Xf_x: P\rightarrow X having image xx, defines by passing to the quotient a bijection, called canonical, of π0(X)\pi_0(X) onto [P; X]. We have π0(∅)=∅\pi_0(\emptyset ) =\emptyset. For a nonempty topological space to be path-connected, it is necessary and sufficient that π0(X)\pi_0(X) be a set with one element.

Let X and Y be topological spaces and f:X→Yf: X\rightarrow Y a continuous mapping. The image f(C)f(C) of every path-connected component C of X is path-connected (III, p. 258, prop. 3), hence contained in a path-connected component of Y. We denote by π0(f):π0(X)→π0(Y)\pi_0(f):\pi_0(X)\rightarrow \pi_0(Y) the mapping which, to a path-connected component C of X, associates the unique path-connected component C′C' of Y such that f(C)⊂C′f(C)\subset C'. If we identify π0(X)\pi_0(X) and π0(Y)\pi_0(Y) with [P; X] and [P; Y] respectively, the mapping π0(f)\pi_0(f) identifies with the mapping χ↦[f]∘χ\chi \mapsto [f]\circ \chi from [P; X] into [P; Y]. In particular, if ff and gg are homotopic mappings from X into Y, we have π0(f)=π0(g)\pi_0(f) =\pi_0(g) (III, p. 230, prop. 2).

Let Z be a topological space and let g:Y→g: Y\rightarrow Z be a continuous mapping; we have π0(g∘f)=π0(g)∘π0(f)\pi_0(g\circ f) =\pi_0(g)\circ \pi_0(f). In particular, if Z = X and if ff and gg are homotopy inverses of one another up to homotopy, we have π0(g)∘π0(f)=π0\pi_0(g)\circ \pi_0(f) =\pi_0(IdX_X) == Idπ0(X)_{\pi_0(X)} and π0(f)∘π0(g)=π0\pi_0(f)\circ \pi_0(g) =\pi_0(IdY_Y) == Idπ0(Y)_{\pi_0(Y)}, which proves that π0(f)\pi_0(f) and π0(g)\pi_0(g) are inverse bijections of one another. A space homotopy equivalent to a path-connected space is therefore itself path-connected. In particular, a space homotopy equivalent to a point is path-connected.

Proposition 6 01X1

Let (Yj)j∈J(Y_j)_{j\in J} be a family of topological spaces. The mapping

(π0(\pi_0(prj_j)):π0(∏j∈JYj)→∏j∈Jπ0(Yj):\pi_0(\prod_{j\in J}Y_j)\rightarrow \prod_{j\in J}\pi_0(Y_j)

is bijective. In particular, the product space of a family of path-connected spaces is path-connected.

This follows from Proposition 3 of III, p. 231 where one takes for X a space reduced to a point.

3. Locally path-connected spaces

Definition 4 01X2

A topological space is said to be locally path-connected if each of its points possesses a fundamental system of path-connected neighbourhoods.

Proposition 7 01X3

A locally path-connected topological space is locally connected. Its connected components coincide with its path-connected components. In particular, if a locally path-connected topological space is connected, it is path-connected.

The first assertion follows from prop. 4. Let us prove the second assertion. Let X be a locally path-connected topological space and let C be a path-connected component of X. Every point of C possesses a path-connected neighbourhood, hence contained in C; consequently, C is an open subset of X. The path-connected components forming a partition of X, such a component is also closed. Since it is connected (III, p. 258, prop. 4), it is a connected component (TG, I, p. 83). The third assertion follows from the second.

Thus we see that there is no ambiguity in saying that a topological space is connected and locally path-connected.

Corollary 1 01X4

Every connected open subset of a locally path-connected topological space is path-connected.

Corollary 2 01X5

Let B be a locally path-connected topological space and let E be a B-space étalé. The space E is locally path-connected. If it is connected, it is path-connected.

The first assertion follows immediately from def. 4 and the definition of an étale mapping (I, p. 28, def. 2). The second follows from it by prop. 7.

In order that a topological space X be locally path-connected, it is necessary and sufficient that every path-connected component of an open set of X be an open set of X (cf. I, p. 85, proof of prop. 11). If the space X is locally path-connected, every point of X therefore possesses a fundamental system of open path-connected neighbourhoods.

Proposition 8 01X6

Every quotient space of a locally path-connected space is locally path-connected.

Let X be a locally path-connected space, let R be an equivalence relation in X and let φ:X→X/R\varphi : X\rightarrow X/R be the canonical mapping. It suffices to prove that the path-connected components of an open subset of X/RX/R are open. Let thus A be an open subset of X/RX/R and let C be a path-connected component of A. Let x∈φ−1(C)x\in \overset{-1}{\varphi}(C), and let K be the path-connected component of xx in φ−1(A)\overset{-1}{\varphi}(A). The set φ(K)\varphi (K) is path-connected (III, p. 258, prop. 3), contained in A and contains φ(x)\varphi (x); hence φ(K)⊂C\varphi (K)\subset C, whence K⊂φ−1(C)K\subset \overset{-1}{\varphi}(C). This proves that φ−1(C)\overset{-1}{\varphi}(C) is the union of path-connected components of φ−1(A)\overset{-1}{\varphi}(A). Since X is locally path-connected and φ−1(A)\overset{-1}{\varphi}(A) is open in X,φ−1(C),\overset{-1}{\varphi}(C) is open in X. Consequently, C is open in X/RX/R. This proves the proposition.

Proposition 9 01X7

The product of a family of locally path-connected spaces, connected with the exception of a finite number of them, is locally path-connected.

Let (Xj)j∈J(X_j)_{j\in J} be a family of locally path-connected spaces which are connected, with the exception of a finite number of them. Let x=x= (xj)j∈J(x_j)_{j\in J} be a point of the product space X = ∏j∈JXj\prod_{j\in J}X_j. By assumption, a fundamental system of neighbourhoods of xx in X consists of sets of the form V=∏j∈JVjV =\prod_{j\in J}V_j where, for every j∈Jj\in J, VjV_j is a path-connected neighbourhood of xjx_j in XjX_j and where Vj=XjV_j= X_j except for a finite set of indices j∈Jj\in J (TG, I, p. 24). These sets being path-connected (III, p. 260, prop. 6), X is locally path-connected.

Corollary 01X8

Every open subset of a numerical space is locally path-connected.

Every point of R\mathbf{R} possesses a basis of neighbourhoods formed by intervals; consequently, R\mathbf{R} is locally path-connected. According to Proposition 9, the numerical space Rn\mathbf{R}^n is locally path-connected, for every integer n⩾1n\geqslant 1. Moreover, an open subset of a locally path-connected topological space is again path-connected, whence the corollary.

Example 01X9

Let G be the subgroup of GL(n,R)\mathbf{G}\mathbf{L}(n,\mathbf{R}) formed by the square matrices of order nn whose determinant is strictly positive. The group G is connected and locally path-connected.

According to A, III, p. 104, prop. 17, the group SL(n,R)\mathbf{S}\mathbf{L}(n,\mathbf{R}) is generated by the elements Bij(λ)B_{ij}(\lambda ) (for 1⩽i,j⩽n1\leqslant i, j\leqslant n such that i=j̸i=\not j and λ∈R\lambda \in \mathbf{R}). The mappings λ↦Bij(λ)\lambda \mapsto B_{ij}(\lambda ) from R\mathbf{R} into GL(n,R)\mathbf{G}\mathbf{L}(n,\mathbf{R}) are continuous, their images are connected subsets of SL(n,R)\mathbf{S}\mathbf{L}(n,\mathbf{R}); since they all contain the identity matrix InI_n, their union is connected (TG, I, p. 81, prop. 2). Consequently, the group SL(n,R)\mathbf{S}\mathbf{L}(n,\mathbf{R}) is connected (TG, III, p. 8, prop. 7). Let A be the subgroup of G formed by the matrices of the form diag(1,...,1,λ, . . . ,1, \lambda ), with λ∈R+∗\lambda \in \mathbf{R}_+^*; it is connected and one has GL(n,R)=A⋅SL(n,R)\mathbf{G}\mathbf{L}(n,\mathbf{R}) = A\cdot \mathbf{S}\mathbf{L}(n,\mathbf{R}). It follows that the group G is connected. Since it is the inverse image of R+∗\mathbf{R}_+^* by the determinant mapping of Mn(R)\mathbf{M}_n(\mathbf{R}) into R\mathbf{R}, it is an open subset of Mn(R)\mathbf{M}_n(\mathbf{R}); it is therefore locally arcwise connected (corollary above), as well as arcwise connected (III, p. 261, corollary 1).

Proposition 10 01XA

Let X be a topological space. The mapping (o,e)(o, e) of Λ(X)=Cc(I;X)\Lambda (X) =\mathscr{C}_c(\mathbf{I}; X) into X×XX\times X, which associates with a path cc in X the pair (c(0),c(1))(c(0), c(1)), is continuous. If the space X is locally arcwise connected, this mapping is open.

We already know that the origin mappings oo and term ee of Λ(X) into X are continuous (III, p. 257), whence the first assertion.

Lemma 01XB

Let c:I→Xc:\mathbf{I}\rightarrow X be a path and let W be a neighbourhood of cc in the space Cc(I;X)\mathscr{C}_c(\mathbf{I}; X). There exists a real number ε∈]0,1/2]\varepsilon \in ]0,1/2], a neighbourhood V0V_0 of c(0)c(0) and a neighbourhood V1V_1 of c(1)c(1) in X having the following properties: one has c([0,ε])⊂V0,c([1−ε,1])⊂V1c([0, \varepsilon ])\subset V_0,c([1-\varepsilon ,1])\subset V_1, and every path c′:I→Xc':\mathbf{I}\rightarrow X such that

c′(t)∈V0c'(t)\in V_0 for 0⩽t⩽ε0\leqslant t\leqslant \varepsilon,

(2) c′(t)=c(t)c'(t) =c(t) for ε⩽t⩽1−ε\varepsilon \leqslant t\leqslant 1-\varepsilon,

c′(t)∈V1c'(t)\in V_1 for 1−ε⩽t⩽11-\varepsilon \leqslant t\leqslant 1,

belongs to W.

By definition of the topology of compact convergence (TG, X, p. 26, déf. 1), there exists a finite set J, a family (Uj)j∈J(U_j)_{j\in J} of open sets in X and a family (Kj)j∈J(K_j)_{j\in J} of compact subsets of I\mathbf{I} such that the set W′W' of paths c′c' satisfying c′(Kj)⊂Ujc'(K_j)\subset U_j for every index jj is a neighbourhood of cc contained in W. Let us then denote by A0A_0 (resp. A1A_1) the set of indices jj such that 0∈Kj0\in K_j (resp. 1∈Kj1\in K_j); put V0=⋂j∈A0UjV_0=\bigcap_{j\in A_0}U_j and V1=⋂j∈A1UjV_1=\bigcap_{j\in A_1}U_j.

Since the mapping cc is continuous, there exists a real number ε∈\varepsilon \in ]0,1/2]]0,1/2] such that c([0,ε])⊂V0,c([1−ε,1])⊂V1,[0,ε]∩Kj=∅c([0, \varepsilon ])\subset V_0,c([1-\varepsilon ,1])\subset V_1,[0, \varepsilon ]\cap K_j=\emptyset for all j∉A0j\notin A_0 and [1−ε,1]∩Kj=∅[1-\varepsilon ,1]\cap K_j=\emptyset for all j∉A1j\notin A_1. Let then c′c' be a path satisfying the conditions (2). Let us prove that c′∈W′c'\in W'. Let j∈Jj\in J and let t∈Kjt\in K_j. If ε⩽t⩽1−ε,c′(t)=c(t)\varepsilon \leqslant t\leqslant 1-\varepsilon ,c'(t) =c(t) belongs to UjU_j. If 0⩽t⩽ε,c′(t)∈V00\leqslant t\leqslant \varepsilon ,c'(t)\in V_0; by the choice of ε\varepsilon, we have j∈A0j\in A_0, hence c′(t)∈Ujc'(t)\in U_j. Analogously, if 1−ε⩽t⩽11-\varepsilon \leqslant t\leqslant 1, we have j∈A1j\in A_1 and c′(t)∈V1⊂Ujc'(t)\in V_1\subset U_j. Thus, c′(Kj)⊂Ujc'(K_j)\subset U_j and c′c' belongs to W′W', hence to W.

Let us now prove the second assertion of Proposition 10. Suppose that the space X is locally connected by arcs. Let cc be a path in X and let W be a neighbourhood of cc in Cc(I;X)\mathscr{C}_c(\mathbf{I}; X). Let ε,V0\varepsilon ,V_0 and V1V_1 be as in the lemma. Let T0T_0 and T1T_1 be connected by arcs neighbourhoods of c(0)c(0) and c(1)c(1) contained in V0V_0 and V1V_1 respectively. There exists a real number θ\theta such that 0<θ<ε0< \theta < \varepsilon and such that c([0,θ])⊂T0,c([1−θ,1])⊂T1c([0, \theta ])\subset T_0,c([1-\theta ,1])\subset T_1. Let x0∈T0x_0\in T_0 and x1∈T1x_1\in T_1; let c0c_0 be a path of origin x0x_0 and of term c(θ)c(\theta ) in T0T_0 and let c1c_1 be a path of origin x1x_1 and of term c(1−θ)c(1-\theta ) in T1T_1. Put

c0(t/θ)c_0(t/\theta ) for 0⩽t⩽θ0\leqslant t\leqslant \theta,

c′(t)=c(t)c'(t) =c(t) for θ⩽t⩽1−θ\theta \leqslant t\leqslant 1-\theta,

c1((1−t)/θ)c_1((1-t)/\theta ) for 1−θ⩽t⩽11-\theta \leqslant t\leqslant 1.

One thus defines a path c′c' joining x0x_0 to x1x_1 and satisfying the conditions (2). This proves that the image of W in X×XX\times X by the mapping (o,e)(o, e) contains the neighbourhood T0×T1T_0\times T_1 of (c(0),c(1))(c(0), c(1)), whence the proposition.

Corollary 01XC

Soit X a topological space locally connected by arcs and let xx be a point of X. The mapping c↦c(1)c\mapsto c(1) of Λx(X)\Lambda_x(X) into X is open.

According to the proposition, the mapping φ:Λ(X)→X×X\varphi : \Lambda (X)\rightarrow X\times X defined by φ(c)=(c(0),c(1))\varphi (c) = (c(0), c(1)) is open and one has Λx(X)=φ−1({x}×X)\Lambda_x(X) =\overset{-1}{\varphi}(\{x\} \times X). Consequently, the mapping Λx(X)→{x}×X\Lambda_x(X)\rightarrow \{x\} \times X deduced from φ\varphi is open (TG, I, p. 30, prop. 2), as is its composite with the second projection pr2_2.

4. Links between connectedness and connectedness by arcs

A space connected by arcs is connected (III, p. 258, prop. 4). There exist connected spaces, even locally connected spaces, which are not connected by arcs (cf. III, p. 331, exerc. 2 and 4). However:

Proposition 11 01XD

A connected and locally connected topological space, whose topology can be defined by a distance for which it is complete, is connected by arcs.

Let X be a topological space. Call a train in X any nonempty finite sequence T=(Wi)1⩽i⩽nT = (W_i)_{1\leqslant i\leqslant n} of connected open subsets of X such that Wi∩Wi+1=∅̸W_i\cap W_{i+1}=\not\emptyset for 1⩽i⩽n−11\leqslant i\leqslant n-1. We say that nn is the length of the train T and that the WiW_i are its wagons. If X is endowed with a distance compatible with its topology, we call the width of the train T the maximum of the diameters of its wagons. We say that the train joins a point aa to a point bb if aa belongs to the first and bb to the last wagon. We call a refinement of T any pair (T′,f)(T', f) formed by a train T′=(Wj′)1⩽j⩽mT'= (W'_j)_{1\leqslant j\leqslant m} and a strictly increasing mapping f:{0,1,...,n}→{0,1,...,m}f:\{0,1, . . . , n\} \rightarrow \{0,1, . . . , m\} such that f(0)=0,f(n)=mf(0) = 0,f(n) =m and Wj′⊂WiW'_j\subset W_i for 1⩽i⩽n1\leqslant i\leqslant n and f(i−1)<j⩽f(i)f(i-1)< j\leqslant f(i).

Lemma 1 01XE

Let X be a connected and locally connected metric space, let aa and bb be points of X and let ε\varepsilon be a real number >0>0. There exists in X a train of width ⩽ε\leqslant \varepsilon joining aa to bb.

More precisely, every train T joining aa to bb possesses a refinement (T′,f)(T', f), where T′T' is a train of width ⩽ε\leqslant \varepsilon joining aa to bb.

The relation “there exists a train of width ⩽ε\leqslant \varepsilon joining xx to yy” is an equivalence relation between xx and yy in X. The equivalence class of a point xx contains every connected open neighbourhood of xx of diameter ⩽ε\leqslant \varepsilon, and xx possesses such a neighbourhood since X is locally connected. The equivalence classes for this relation are therefore open, and consequently also closed. There is at most one of them, since X is connected. This proves the first assertion.

Let T=(Wi)1⩽i⩽nT = (W_i)_{1\leqslant i\leqslant n} be a train in X joining aa to bb. Put x0=ax_0=a, xn=bx_n=b and choose for 1⩽i⩽n−11\leqslant i\leqslant n-1 a point xix_i in the nonempty set Wi∩Wi+1W_i\cap W_{i+1}. For 1⩽i⩽n1\leqslant i\leqslant n, the open set WiW_i is connected and locally connected and xi−1,xix_{i-1},x_i are two of its points; there exists by the preceding paragraph a train (Wi,k)1⩽k⩽mi(W_{i,k})_{1\leqslant k\leqslant m_i} in WiW_i, of width ⩽ε\leqslant \varepsilon, joining xi−1x_{i-1} to xix_i.

Put m=m1+⋯+mnm=m_1+\cdots +m_n. For 1⩽j⩽m1\leqslant j\leqslant m, put Wj′=Wi,kW'_j= W_{i,k}, where (i,k)(i, k) is the unique pair of integers such that 1⩽i⩽n,1⩽k⩽mi1\leqslant i\leqslant n,1\leqslant k\leqslant m_i and j=m1+⋯+mi−1+kj=m_1+\cdots +m_{i-1}+k. For 0⩽i⩽n0\leqslant i\leqslant n, put f(i)=m1+⋯+mif(i) =m_1+\cdots +m_i. Then T′=(Wj′)1⩽j⩽mT'= (W'_j)_{1\leqslant j\leqslant m} is a train of width ⩽ε\leqslant \varepsilon in X joining aa to bb, and (T′,f)(T', f) is a refinement of T.

Let us now prove Proposition 11. Equip the connected and locally connected space X with a distance dd, compatible with its topology, for which it is complete. Let aa and bb be points of X. Lemma 1 permits us to construct by induction sequences (Ts)s⩾0(T_s)_{s\geqslant 0} and (fs)s⩾1(f_s)_{s\geqslant 1} such that, for every s⩾0s\geqslant 0, Ts=(Ws,i)1⩽i⩽nsT_s= (W_{s,i})_{1\leqslant i\leqslant n_s} is a train of width ⩽2−s\leqslant 2^{-s} joining aa to bb and (Ts+1,fs+1)(T_{s+1}, f_{s+1}) is a refinement of TsT_s.

We can choose by induction, for s⩾0s\geqslant 0, a strictly increasing mapping gs:{0,1,...,ns}→Ig_s:\{0,1, . . . , n_s\} \rightarrow \mathbf{I} such that gs(0)=0g_s(0) = 0 and gs(ns)=1g_s(n_s) = 1, in such a way that gs+1∘fs=gsg_{s+1}\circ f_s=g_s. Define, for s⩾0s\geqslant 0, a subset AsA_s of I×X\mathbf{I}\times X by putting

As=⋃1⩽i⩽ns([gs(i−1),gs(i)]×Ws,i) A_s=\bigcup_{1\leqslant i\leqslant n_s}([g_s(i-1), g_s(i)]\times W_{s,i})

The sequence (As)s⩾0(A_s)_{s\geqslant 0} is decreasing: indeed, for every integer j∈{1,...,ns+1}j\in \{1, . . . , n_{s+1}\}, there exists a unique integer i∈{1,...,ns}i\in \{1, . . . , n_s\} such that fs(i−1)<j⩽fs(i)f_s(i-1)< j\leqslant f_s(i), and one has

[gs+1(j−1),gs+1(j)]⊂[gs(i−1),gs(i)],Ws+1,j⊂Ws,i [g_{s+1}(j-1), g_{s+1}(j)]\subset [g_s(i-1), g_s(i)],W_{s+1,j}\subset W_{s,i}

Let t∈It\in \mathbf{I}. For every s⩾0s\geqslant 0, denote by As(t)A_s(t) the set of the x∈Xx\in X such that (t,x)∈As(t, x)\in A_s. The set As(t)A_s(t) is either one of the cars, or the union of two consecutive cars of the train TsT_s; it is therefore a nonempty subset of X, of diameter ⩽21−s\leqslant 2^{1-s}. The sequence (As(t))s⩾0(A_s(t))_{s\geqslant 0} is decreasing. The set of its terms is a Cauchy filter basis. This one converges to a point c(t)c(t) since the metric space X is complete.

Since aa belongs to each of the sets As(0)=Ws,1A_s(0) = W_{s,1}, we have c(0)=ac(0) =a; analogously c(1)=bc(1) =b. Let t∈It\in \mathbf{I} and let ss be an integer ⩾0\geqslant 0. The point tt possesses in I\mathbf{I} a neighbourhood V of one of the following forms: [gs(0),gs(1)[[g_s(0), g_s(1)[, ]gs(i−1),gs(i+1)[]g_s(i-1), g_s(i+ 1)[ for an integer ii such that 1⩽i⩽1\leqslant i\leqslant ns−1n_s-1, or ]gs(ns−1),gs(ns)]]g_s(n_s-1), g_s(n_s)]. According to the case, the set c(V)c(V) is contained in the closure of the first car, of the union of two consecutive cars, or of the last car of the train TsT_s. It is therefore of diameter ⩽21−s\leqslant 2^{1-s}, and we have d(c(t),c(t′))⩽21−sd(c(t), c(t'))\leqslant 2^{1-s} for t′∈Vt'\in V. This proves that the mapping c:I→c:\mathbf{I}\rightarrow X is continuous. Thus cc is a path in X joining aa to bb, and X is arcwise connected.

Corollary 1 01XF

A locally connected topological space, whose topology can be defined by a distance for which it is complete, is locally arcwise connected.

Let X be such a space and let U be an open and connected subset of X. According to Lemma 2 below, there exists a distance on U compatible with its topology for which U is complete. Since U is locally connected, it follows from Proposition 11 that U is arcwise connected.

Lemma 2 01XG

Let X be a complete metric space and let U be an open subset of X. There exists a distance on U compatible with its topology for which U is complete.

We shall take up again the arguments of the proof of prop. 2 of TG, IX, p. 57. We may suppose U distinct from X. Let V be the subset of the product R×X\mathbf{R}\times X formed by the points (t,x)(t, x) such that td(x,X−U)=1t d(x,X-U) = 1 ; the subspace V of R×X\mathbf{R}\times X is closed and the mapping (t,x)↦x(t, x)\mapsto x from V into U is a homeomorphism (TG, IX, p. 13, prop. 3). There exists on R×X\mathbf{R}\times X a distance d′d' compatible with its topology for which R×X\mathbf{R}\times X is complete (TG, IX, p. 15, cor. 2 and TG, II, p. 17, prop. 10). The space V is complete for the distance induced by d′d' (TG, II, p. 16, prop. 8), whence the lemma.

Corollary 2 01XH

A locally compact, locally connected and metrizable topological space is locally arcwise connected. If it is connected, it is arcwise connected.

It suffices to prove the first assertion (III, p. 260, prop. 7). Let X be a locally compact and locally connected metric space. The open, connected and relatively compact subsets of X constitute a basis of the topology of X. Let U be such a set; since U is an open subset of its closure, which is a compact metric space, hence complete, there exists according to Lemma 2 a distance on U compatible with its topology for which U is complete. It follows from prop. 11 of III, p. 264 that U is arcwise connected, whence the corollary.

5. Arcwise continuous mappings

Definition 5 01XI

Let X and Y be topological spaces. A mapping f:X→Yf: X\rightarrow Y is said to be arcwise continuous if, for every path cc in X, the mapping f∘c:I→Yf\circ c:\mathbf{I}\rightarrow Y is continuous.

Remark 01XJ

Suppose that the space X is arcwise connected. Let xx be a point of X. For ff to be arcwise continuous, it suffices that, for every path cc with origin xx, the mapping f∘cf\circ c be continuous. Indeed, let dd be an arbitrary path in X and let cc be a path in X with origin xx and with term d(0)d(0). If the mapping f∘(c∗d)f\circ (c*d) is continuous, the same is true of the mapping f∘df\circ d because we have f∘d(t)=f∘(c∗d)((t+1)/2)f\circ d(t) =f\circ (c*d)((t+ 1)/2).

A continuous mapping is arcwise continuous. The converse is not always true; the propositions 12 and 13 below provide criteria making it possible to assert that an arcwise continuous mapping is continuous.

Proposition 12 01XK

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a mapping. Suppose that the space X is locally arcwise connected and that every point of X possesses a countable fundamental system of neighbourhoods. If the mapping ff is arcwise continuous, it is continuous.

According to (TG, IX, p. 18), it suffices to prove that, for every point xx of X and every sequence (xn)n⩾1(x_n)_{n\geqslant 1} of points of X which tends to xx, the sequence (f(xn))n⩾1(f(x_n))_{n\geqslant 1} tends to f(x)f(x). By deleting, if necessary, the first terms of the sequence (xn)n⩾1(x_n)_{n\geqslant 1}, we reduce to the case where all the terms of the sequence belong to the same arc-connected neighbourhood of xx in X. According to the lemma below, there then exists a path c:I→Xc:\mathbf{I}\rightarrow X such that c(0)=xc(0) =x and c(1/n)=xnc(1/n) =x_n for n⩾1n\geqslant 1. If the mapping ff is continuous by arcs, the mapping f∘cf\circ c is continuous and the element f(x)=f(c(0))f(x) =f(c(0)) is the limit of the sequence (f(c(1/n)))n⩾1(f(c(1/n)))_{n\geqslant 1}, that is to say of the sequence (f(xn))n⩾1(f(x_n))_{n\geqslant 1}, whence the corollary.

Lemma 01XL

Let X be a topological space connected and locally connected by arcs and let xx be a point of X. Suppose that the point xx has a countable fundamental system of neighbourhoods. Then, for every sequence (xn)n⩾1(x_n)_{n\geqslant 1} of points of X tending to xx, there exists a path cc in X such that c(0)=xc(0) =x and c(1/n)=xnc(1/n) =x_n for n⩾1n\geqslant 1.

Let (Wm)m⩾1(W_m)_{m\geqslant 1} be a fundamental system of neighbourhoods of xx. Put V0=XV_0= X and for every m⩾1m\geqslant 1, let VmV_m be an arc-connected neighbourhood of xx contained in Vm−1∩WmV_{m-1}\cap W_m.

For every integer n⩾1n\geqslant 1, denote by mnm_n the largest integer m⩽nm\leqslant n such that xk∈Vmx_k\in V_m for every k⩾nk\geqslant n. The sequence (mn)n⩾1(m_n)_{n\geqslant 1} is increasing by definition; it tends to infinity, because the sequence (xn)n⩾1(x_n)_{n\geqslant 1} tends to xx. For every integer n⩾1n\geqslant 1, let cn:I→Vmnc_n:\mathbf{I}\rightarrow V_{m_n} be a path with origin xn+1x_{n+1} and endpoint xnx_n in VmnV_{m_n}. Define a mapping c:I→Xc:\mathbf{I}\rightarrow X by putting c(0)=xc(0) =x and c(t)=cn(n(n+1)t−n)c(t) =c_n(n(n+ 1)t-n) if 1/(n+1)<t⩽1/(n+ 1)< t\leqslant 1/n1/n. We have c(1/n)=xnc(1/n) =x_n for every n⩾1n\geqslant 1. The mapping cc is therefore continuous on every interval of the form [1/(n+1),1/n][1/(n+ 1),1/n] with n⩾1n\geqslant 1, hence on the interval ]0,1]]0,1]. If t⩽1/nt\leqslant 1/n, the point c(t)c(t) belongs to VmnV_{m_n}; the mapping cc is therefore continuous at 0.

Proposition 13 01XM

Let p:E→Bp: E\rightarrow B be an étale and separated mapping and let s:B→Es: B\rightarrow E be a section of pp. If the space B is locally path-connected and if the section ss is path-continuous, it is continuous.

Let bb be a point of B; let us prove that ss is continuous at the point bb. Since pp is an étale mapping, there exist a neighbourhood V of bb and a continuous local section s′s' of pp defined in V such that s′(b)=s(b)s'(b) =s(b) (I, p. 33, prop. 9). Since B is locally path-connected, we may suppose that V is path-connected. For every path cc in V, with origin bb, the mappings s∘cs\circ c and s′∘cs'\circ c are two continuous liftings to B of the mapping c:I→Xc:\mathbf{I}\rightarrow X and we have s∘c(0)=s′∘c(0)=s(b)s\circ c(0) =s'\circ c(0) =s(b). By Corollary 1 of I, p. 34, we have s∘c=s′∘cs\circ c=s'\circ c and in particular s∘c(1)=s′∘c(1)s\circ c(1) =s'\circ c(1). Since every point of V is the end point of a path in V with origin bb, the mappings ss and s′s' coincide in V. The mapping ss is therefore continuous in V.

Corollary 01XN

Let B be a topological space, let (E,p)(E, p) and (E′,p′)(E', p') be two B-spaces. Suppose that the mapping pp is étale and separated and that the space E′E' is locally path-connected. Then every path-continuous mapping f:E′→Ef: E'\rightarrow E such that p∘f=p′p\circ f=p' is continuous.

The mapping pr1:E′×BE→E′_1: E'\times_BE\rightarrow E' is étale and separated (I, p. 31, prop. 8 and I, p. 27, prop. 4) and the mapping x↦(x,f(x))x\mapsto (x, f(x)) is a path-continuous section of it. By Proposition 13, it is continuous, hence ff is continuous.

6. Complements on compact metrizable topological spaces

Let us endow the set with two elements {0,1}\{0,1\} with the discrete topology and the set {0,1}N\{0,1\}^{\mathbf{N}} with the product topology. The topological space {0,1}N\{0,1\}^{\mathbf{N}} is compact (TG, I, p. 63, th. 3), metrizable (TG, IX, p. 15, cor. 2), nonempty, totally disconnected (TG, I, p. 84, prop. 10) and has no isolated point.

Proposition 14 01XO

Every compact, metrizable, nonempty, totally disconnected topological space without isolated point is homeomorphic to {0,1}N\{0,1\}^{\mathbf{N}}.

Let X be such a topological space. Let us endow it with a distance compatible with its topology. Since the space X is compact, it is complete for this distance (TG, II, p. 27, th. 1).

Lemma 3 01XP

Let ε\varepsilon be a real number >0>0. There exists an integer m⩾1m\geqslant 1 such that, for every integer n⩾mn\geqslant m, X admits a partition formed of nn nonempty open and closed sets of diameter ⩽ε\leqslant \varepsilon.

Every point of X admits an open and closed neighbourhood of diameter ⩽ε\leqslant \varepsilon (TG, II, p. 32, corollary of prop. 6). Since the space X is compact, it has a finite covering by such sets. Choose one, (Ui)1⩽i⩽m(U_i)_{1\leqslant i\leqslant m}, for which mm is minimal. We have m⩾1m\geqslant 1 since X is nonempty. For 1⩽i⩽m1\leqslant i\leqslant m, denote by ViV_i the intersection of UiU_i and the X−UkX-U_k for k<ik < i. Then (Vi)1⩽i⩽m(V_i)_{1\leqslant i\leqslant m} is a partition of X, formed of mm nonempty open and closed sets of diameter ⩽ε\leqslant \varepsilon.

Since X has no isolated point, every nonempty open and closed subset V of X contains at least two points. Since moreover X is compact and totally disconnected, V is the union of two disjoint nonempty open and closed subsets (loc. cit.). It follows, by induction, that for every integer n⩾mn\geqslant m, X admits a partition formed of nn nonempty open and closed sets of diameter ⩽ε\leqslant \varepsilon.

Let us now finish the proof of the prop. 14. Every nonempty open and closed subspace of X is a compact, totally disconnected metric space and has no isolated point. Lemma 3 therefore allows us to construct by induction a sequence (Jn)n⩾0(J_n)_{n\geqslant 0} of finite sets and, for every n⩾0n\geqslant 0, a mapping φn\varphi_n from the set Cn=J0×⋯×JnC_n= J_0\times \cdots \times J_n into the set of nonempty open and closed subsets of X of diameter ⩽2−n\leqslant 2^{-n}, in such a way that :

(i) For every n⩾0n\geqslant 0, there exists an integer mn⩾1m_n\geqslant 1 such that Card(Jn_n) == 2mn2^{m_n};

(ii) The family (φ0(c))c∈C0(\varphi_0(c))_{c\in C_0} is a partition of X ;

(iii) For every n⩾0n\geqslant 0 and every c∈Cnc\in C_n, the family (φn+1(c,j))j∈Jn+1(\varphi_{n+1}(c, j))_{j\in J_{n+1}} is a partition of φn(c)\varphi_n(c).

Let us denote by pnp_n the canonical projection of Cn+1C_{n+1} onto CnC_n. The sequence C = (Cn,pn)n⩾0(C_n, p_n)_{n\geqslant 0} is a sieve (TG, IX, p. 63, def. 8). The topological space associated with this sieve (TG, IX, p. 63) is identified with the topological space J, product of the discrete topological spaces JnJ_n. The sieve C and the sequence of mappings (φn)n⩾0(\varphi_n)_{n\geqslant 0} define a strict sieving of the metric space X (TG, IX, p. 63 and p. 64). The mapping f:J→Xf: J\rightarrow X deduced from this sieving is continuous and bijective (TG, IX, p. 65). Since the topological space J is compact (TG, I, p. 63, th. 3) and X is separated, ff is a homeomorphism (TG, I, p. 63, cor. 2). Since JnJ_n is homeomorphic to {0,1}mn\{0,1\}^{m_n} for every n⩾0n\geqslant 0, J is homeomorphic to {0,1}N\{0,1\}^{\mathbf{N}} (TG, I, p. 25, prop. 2).

Example 01XQ

Let K be the triadic Cantor set (TG, IV, p. 9, example). For every n⩾0n\geqslant 0, put Jn={0,1}J_n=\{0,1\} and define a mapping KnK_n from the set Cn=J0×⋯×JnC_n= J_0\times \cdots \times J_n into the set of closed intervals of [0,1][0,1] in the following manner: put K0(0)=[0,13]K_0(0) = [0,\frac{1}{3}] and K0(1)=[23,1]K_0(1) = [\frac{2}{3},1]; for every n⩾0n\geqslant 0 and every c∈Cn,Kn+1(c,0)c\in C_n,K_{n+1}(c,0) and Kn+1(c,1)K_{n+1}(c,1) are respectively the “left third” and the “right third” of Kn(c)K_n(c). If c=(j0,j1,...,jn)∈Cn,Kn(c)c= (j_0, j_1, . . . , j_n)\in C_n,K_n(c) is the interval denoted by Kn,pK_{n,p} in loc. cit., with p=2nj0+2n−1j1+⋯+jn+1p= 2^nj_0+2^{n-1}j_1+\cdots +j_n+1, it is also the interval [a,a+13n+1][a, a+\frac{1}{3^{n+1}}], where a=2(j03+j132+⋯+jn3n+1)a= 2(\frac{j_0}{3}+\frac{j_1}{3^2}+\cdots +\frac{j_n}{3^{n+1}}). For n⩾0n\geqslant 0 and c∈Cnc\in C_n, put φn(c)=Kn(c)∩K\varphi_n(c) = K_n(c)\cap K. The family (φn(c))c∈Cn(\varphi_n(c))_{c\in C_n} is a partition of K formed by closed sets. These sets are therefore also open in K; they are nonempty and have diameter 13n+1\frac{1}{3^{n+1}}, because the endpoints of the intervals Kn(c)K_n(c) belong to K. The sequence C=(Cn,pn)n⩾0C = (C_n, p_n)_{n\geqslant 0}, where pn:Cn+1→Cnp_n: C_{n+1}\rightarrow C_n is the canonical projection (c,j)↦c(c, j)\mapsto c, is a sieve, and the topological space associated with this sieve is identified with {0,1}N\{0,1\}^{\mathbf{N}}. The sieve C and the sequence of mappings (φn)n⩾0(\varphi_n)_{n\geqslant 0} define a strict sieving of the metric space K. The mapping f:{0,1}N→Kf:\{0,1\}^{\mathbf{N}}\rightarrow K deduced from this sieving is a homeomorphism, given by the formula

f((jn)n⩾0)=2∑n=0∞jn3n+1 f((j_n)_{n\geqslant 0}) = 2\sum_{n=0}^{\infty}\frac{j_n}{3^{n+1}}

Corollary 01XR

Let X be a metrizable, compact and nonempty topological space. There exists a continuous and surjective mapping from {0,1}N\{0,1\}^{\mathbf{N}} into X.

Every compact and metrizable topological space is homeomorphic to a necessarily closed subspace of the topological space IN\mathbf{I}^{\mathbf{N}} (TG, IX, p. 18, prop. 12 and p. 21, prop. 16). Let A be a closed subspace of IN\mathbf{I}^{\mathbf{N}} and hh a homeomorphism of A onto X.

Let K = {0,1}N\{0,1\}^{\mathbf{N}} and, for α=(an)∈\alpha = (a_n)\in K, put f(α)=f(\alpha ) = ∑n=0∞an2−n−1\sum^{\infty}_{n=0}a_n2^{-n-1}; we have f(α)∈If(\alpha )\in \mathbf{I}. The mapping f:K→If: K\rightarrow \mathbf{I} thus resulting is surjective (TG, IV, p. 42) and continuous. Indeed, if two elements α\alpha and β\beta of K have the same coordinates of index <n< n, we have ∣f(β)−f(α)∣⩽2−n|f(\beta )-f(\alpha )|\leqslant 2^{-n}. Let gg denote the mapping (αn)↦(f(αn))(\alpha_n)\mapsto (f(\alpha_n)) of KNK^{\mathbf{N}} into IN\mathbf{I}^{\mathbf{N}}; it is continuous and surjective. The topological space KNK^{\mathbf{N}} is compact (TG, I, p. 63, th. 3), metrizable (TG, IX, p. 15, cor. 2) and totally disconnected (TG, I, p. 84, prop. 10) and the same is true of its closed subspace g−1(A)\overset{-1}{g}(A); the latter is nonempty since the mappings gg and hh are surjective.

Then, the space g−1(A)×K\overset{-1}{g}(A)\times K is homeomorphic to {0,1}N\{0,1\}^{\mathbf{N}} (prop. 14), since it is a compact, metrizable, totally disconnected topological space without isolated point and the mapping (x,y)↦h(g(x))(x, y)\mapsto h(g(x)) of g−1(A)×K\overset{-1}{g}(A)\times K into X is continuous surjective.

7. Topological properties of the image of a path

Proposition 15 01XS

The image of a path in a separated topological space is a compact, metrizable, connected and locally connected by arcs topological space.

Let X be a separated topological space and c:I→Xc:\mathbf{I}\rightarrow X a continuous mapping. Let R denote the equivalence relation c(s)=c(t)c(s) =c(t) in I\mathbf{I}. The topological space c(I)c(\mathbf{I}) is separated (TG, I, p. 63, cor. 1), the space I/R\mathbf{I}/R is quasi-compact (TG, I, p. 62, th. 2), hence the bijection I/R→c(I)\mathbf{I}/R\rightarrow c(\mathbf{I}) deduced from cc is a homeomorphism (TG, I, p. 63, cor. 2). Consequently, the space c(I)c(\mathbf{I}) is compact, metrizable (TG, IX, p. 22, prop. 17), connected (TG, I, p. 82, prop. 6) and locally connected by arcs (III, p. 261, prop. 8).

Theorem 1 (Hahn and Mazurkiewicz) 01XT

Every metrizable, compact, nonempty, connected and locally connected topological space is homeomorphic to a quotient space of the segment [0,1][0,1].

Lemma 4 01XU

Let K be a compact topological space and let R\mathscr{R} be a set of open subsets of K covering K. There exists an entourage V of the uniform structure of K (TG, II, p. 27, th. 1) such that, for every x∈Kx\in K, V(x)V(x) is contained in one of the sets belonging to R\mathscr{R}.

For every point xx of K, there exists an entourage WxW_x of the uniform structure of K such that Wx(x)W_x(x) is contained in one of the sets belonging to R\mathscr{R}. Let VxV_x be an entourage of the uniform structure of K such that Vx2\overset{2}{V_{x}} is contained in WxW_x. The interiors of the Vx(x)V_x(x) cover K; since the space K is compact, there exists a finite subset F of K such that the family (Vy(y))y∈F(V_y(y))_{y\in F} is a covering of K (TG, I, p. 59). Let V denote the intersection of the family (Vy)y∈F(V_y)_{y\in F}; the set V is an entourage of the uniform structure of K. For every point x∈Kx\in K, there exists a point y∈Fy\in F such that xx belongs to Vy(y)V_y(y). Consequently, the set

V(x)V(x) is contained in Vy2(y)\overset{2}{V_{y}}(y), hence in one of the sets belonging to R\mathscr{R}.

Lemma 5 01XV

Let X and Y be uniform spaces and ff a mapping of X into Y. Let F\mathscr{F} be a set of closed subsets of X covering X and possessing the following properties:

(i) There exists a set F0∈FF_0\in \mathscr{F} which meets all the sets F∈FF\in \mathscr{F};

(ii) For every entourage U of the uniform structure of X, there are only a finite number of sets F∈FF\in \mathscr{F} which are not small of order U;

(iii) For every entourage V of the uniform structure of Y, there are only a finite number of sets F∈FF\in \mathscr{F} such that f(F)f(F) is not small of order V.

Thus, if the restriction of ff to each of the sets F∈FF\in \mathscr{F} is continuous, ff is continuous.

Let xx be a point of X. Let us prove that ff is continuous at xx. There exists a set F1∈FF_1\in \mathscr{F} such that x∈F1x\in F_1. The restriction of ff to F0∪F1F_0\cup F_1 is continuous (TG, I, p. 19, prop. 4). By replacing F0F_0 by F0∪F1F_0\cup F_1, we reduce to the case where x∈F0x\in F_0.

Let V be an entourage of the uniform structure of Y. Choose

an entourage V′V' of this same uniform structure such that V′2⊂V\overset{2}{V'}\subset V. Since the restriction of ff to F0F_0 is continuous, there exists an entourage U of the uniform structure of X such that f(z)∈V′(f(x))f(z)\in V'(f(x)) for every z∈F0∩U(x)z\in F_0\cap U(x). Let U′U' be an entourage of the uniform structure of X

such that U′2⊂U\overset{2}{U'}\subset U. Let A denote the union of F0F_0, of the sets F∈FF\in \mathscr{F} which are not small of order U′U' and of those such that f(F)f(F) is not small of order V′V'. By assumption, A is the union of a finite number of sets belonging to F\mathscr{F}, and the restriction of ff to A is continuous (loc. cit.). There therefore exists a neighbourhood W of xx in X, contained in U′(x)U'(x), such that f(y)∈V(f(x))f(y)\in V(f(x)) for y∈A∩Wy\in A\cap W. To conclude, it will suffice to prove that one also has f(y)∈V(f(x))f(y)\in V(f(x)) for every point y∈(X−A)∩Wy\in (X-A)\cap W. Let yy be such a point. Let F be an element of F\mathscr{F} such that y∈Fy\in F. By definition of A, F is small of order U′U' and f(F)f(F) is small of order V′V'. By assumption, F meets F0F_0. Let z∈F∩F0z\in F\cap F_0. We have z∈U′(y)z\in U'(y) since F is small of order U′U' and y∈U′(x)y\in U'(x) since W

is contained in U′(x)U'(x), whence z∈U′2(x)z\in \overset{2}{U'}(x) and fortiori z∈U(x)z\in U(x). But then, since zz belongs to F0F_0, we have f(z)∈V′(f(x))f(z)\in V'(f(x)). Moreover f(F)f(F) is small of order V′V', whence f(y)∈V′(f(z))f(y)\in V'(f(z)). It follows that we have

f(y)∈V′2(f(x))f(y)\in \overset{2}{V'}(f(x)) and finally f(y)∈V(f(x))f(y)\in V(f(x)). This concludes the proof of lemma 5.

Let us now prove theorem 1. Let X be a compact nonempty, connected and locally connected metric space. Such a space is connected by arcs and locally connected by arcs (III, p. 267, corollary 2). According to the corollary and the example of III, p. 270, there exists a continuous and surjective mapping ff from the triadic Cantor set K (TG, IV, p. 9, example) into X. We shall construct a continuous extension gg of ff to [0,1][0,1], which will prove theorem 1.

Let ε\varepsilon be a real number >0>0. The open and arcwise connected subsets of X of diameter ⩽ε\leqslant \varepsilon cover X. Let R\mathscr{R} denote the set of their inverse images by ff: it is a set of open sets of K covering K. According to Lemma 4 of III, p. 272, there exists a real number α>0\alpha >0 such that every closed ball of K of radius α\alpha is contained in an element of R\mathscr{R}. In particular, if tt and t′t' are points of K such that ∣t−t′∣⩽α|t-t'|\leqslant \alpha, there exists a path in X joining f(t)f(t) to f(t′)f(t') whose image has diameter ⩽ε\leqslant \varepsilon.

The preceding paragraph makes it possible to construct by induction a strictly increasing sequence (nk)k⩾0(n_k)_{k\geqslant 0} of integers ⩾\geqslant 0 possessing the following property: for every integer k⩾0k\geqslant 0 and every pair (t,t′)(t, t') of points of K such that ∣t−t′∣⩽3−nk|t-t'|\leqslant 3^{-n_k}, there exists a path in X joining f(t)f(t) to f(t′)f(t') whose image has diameter ⩽2−k\leqslant 2^{-k}. The complement of K in [0,1][0,1] is the union of a family (In,p)(I_{n,p}) of pairwise disjoint open intervals, where nn runs through the set of integers ⩾0\geqslant 0 and pp the set of integers between 1 and 2n2^n (TG, IV, p. 9, example). Consider one of these intervals In,pI_{n,p} and write it ]a,b[]a, b[. The points aa and bb belong to K and we have b−a=3−n−1b-a= 3^{-n-1}. The function gg is defined on the interval In,pI_{n,p} in the following way: one chooses a path cc in X joining f(a)f(a) to f(b)f(b), whose image has diameter ⩽2−k\leqslant 2^{-k} if nk⩽n<nk+1n_k\leqslant n < n_{k+1}, and one sets g(t)=c(t−ab−a)g(t) =c(\frac{t-a}{b-a}) for t∈In,pt\in I_{n,p}. The function g:[0,1]→Xg: [0,1]\rightarrow X thus defined extends ff. It is continuous on K as well as on each of the closed intervals In,p‾\overline{I_{n,p}}. The latter meet K. Moreover, for every real number ε>0\varepsilon >0, there are only finitely many intervals In,p‾\overline{I_{n,p}} of length >ε> \varepsilon, and there are only finitely many intervals In,p‾\overline{I_{n,p}} whose images by gg have diameter >ε> \varepsilon. According to Lemma 5, the mapping gg is continuous.

8. Characterizations of the interval

Lemma 6 01XW

Let D be a countable totally ordered set, not reduced to one element, possessing a smallest and a greatest element. Suppose that D has no gap (E, III, p. 73, exerc. 19), that is to say that every open interval ]x,y[]x, y[, where xx and yy are elements of D such that x<yx < y, is nonempty. There then exists an isomorphism of ordered sets from I∩Q\mathbf{I}\cap \mathbf{Q} onto D.

Let aa be the smallest element and bb the greatest element of D. By assumption, we have b=a̸b=\not a and ]a,x[=∅̸]a, x[=\not\emptyset for every x∈D−{a}x\in D-\{a\}. The set D−{a}-\{a\} is totally ordered, is not empty and has no smallest element; it is therefore infinite (E, III, p. 34, cor. 1 of prop. 3). The sets I∩Q\mathbf{I}\cap \mathbf{Q} and D, infinite and countable, are equipotent to N\mathbf{N}.

Choose bijections n↦ann\mapsto a_n and n↦bnn\mapsto b_n of N\mathbf{N} onto I∩Q\mathbf{I}\cap \mathbf{Q} and D respectively, such that a0=0,a1=1,b0=a,b1=ba_0= 0,a_1= 1,b_0=a,b_1=b. There exists a unique strictly increasing mapping f:I∩Q→Df:\mathbf{I}\cap \mathbf{Q}\rightarrow D possessing the following properties: we have f(0)=af(0) =a and f(1)=bf(1) =b; for n⩾2n\geqslant 2, we have f(an)=bmf(a_n) =b_m, where mm is the smallest natural number for which the mapping of {a0,...,an}\{a_0, . . . , a_n\} into D which coincides with ff in {a0,...,an−1}\{a_0, . . . , a_{n-1}\} and maps ana_n onto bmb_m is strictly increasing. These properties indeed define f(an)f(a_n) by induction on nn, the existence of the integer mm being ensured by the fact that D is without gaps.

Since the mapping ff is strictly increasing and I∩Q\mathbf{I}\cap \mathbf{Q} is totally ordered, ff defines an isomorphism of ordered sets from I∩Q\mathbf{I}\cap \mathbf{Q} onto its image (E, III, p. 14, prop. 11). It remains for us to prove that ff is surjective. For this, let us prove by induction that bmb_m belongs to the image of ff for every m∈Nm\in \mathbf{N}.

One has b0=f(0)b_0=f(0) and b1=f(1)b_1=f(1). Suppose that we have m⩾2m\geqslant 2 and that, for 0⩽k⩽m−10\leqslant k\leqslant m-1, there exists ck∈I∩Qc_k\in \mathbf{I}\cap \mathbf{Q} such that f(ck)=bkf(c_k) =b_k. We have c0=0c_0= 0 and c1=1c_1= 1, since ff is injective. Consider the smallest integer n∈Nn\in \mathbf{N} for which ana_n does not belong to {c0,...,cm−1}\{c_0, . . . , c_{m-1}\} and the mapping gg of {c0,...,cm−1,an}\{c_0, . . . , c_{m-1}, a_n\} into D which coincides with ff in {c0,...,cm−1}\{c_0, . . . , c_{m-1}\} and maps ana_n onto bmb_m is strictly increasing; such an integer exists since I∩Q\mathbf{I}\cap \mathbf{Q} has no gap. Let f′f' be the mapping of {a0,...,an}\{a_0, . . . , a_n\} into D which coincides with ff in {a0,...,an−1}\{a_0, . . . , a_{n-1}\} and which maps ana_n onto bmb_m. Let us prove that it is strictly increasing. Let j∈{0,...,n−1}j\in \{0, . . . , n-1\}; by definition of the integer nn, there exists k∈{0,...,m−1}k\in \{0, . . . , m-1\} such that aj=cka_j=c_k, or aj<cka_j< c_k and bm⩾f(ck)b_m\geqslant f(c_k), or aj>cka_j> c_k and bm⩽f(ck)b_m\leqslant f(c_k). Suppose bk<bmb_k< b_m; we then have g(ck)=f(ck)=bk<bm=g(an)g(c_k) =f(c_k) =b_k< b_m=g(a_n), whence ck<anc_k< a_n since gg is strictly increasing, then aj⩽ck<ana_j\leqslant c_k< a_n; moreover, f′(aj)=f'(a_j) = f(aj)⩽f(ck)=bk<bm=f′(an)f(a_j)\leqslant f(c_k) =b_k< b_m=f'(a_n). Analogously, if bk>bmb_k> b_m, we obtain an<ck⩽aja_n< c_k\leqslant a_j and f′(aj)=f(aj)⩾f(ck)=bk>bm=f′(an)f'(a_j) =f(a_j)\geqslant f(c_k) =b_k> b_m=f'(a_n). Since ff itself is strictly increasing, it follows that the mapping f′f' is strictly increasing.

If m′m' is the integer such that f(an)=bm′f(a_n) =b_{m'}, we have m′⩽mm'\leqslant m, by definition of ff. If we had m′<mm'< m, we would have f(an)=bm′=f(cm′)f(a_n) =b_{m'}=f(c_{m'}), whence an=cm′a_n=c_{m'}, since ff is injective, which contradicts the definition of ana_n. Thus, m′=mm'=m and f(an)=bmf(a_n) =b_m, which proves that bmb_m belongs to the image of ff and completes by induction the proof of the surjectivity of ff.

Proposition 16 01XX

Let E be a totally ordered set not reduced to one element. Suppose that every subset of E has a supremum and that there exists a countable subset of E which meets every open interval ]x,y[]x, y[, where xx and yy are elements of E such that x<yx < y. There then exists an isomorphism of ordered sets of I\mathbf{I} onto E.

Since ∅\emptyset and E each have a supremum in E, E has a smallest element aa and a greatest element bb. These are distinct since E is not reduced to one element. Let D′D' be a countable subset of E which meets every open interval of E of the form ]x,y[]x, y[, with x<yx < y. Put D=D′∪{a,b}D = D'\cup \{a, b\}. The set D is totally ordered and without gaps. By lemma 6, there exists an isomorphism of ordered sets ff of I∩Q\mathbf{I}\cap \mathbf{Q} onto D.

For every t∈It\in \mathbf{I}, let g(t)g(t) be the supremum of f([0,t]∩Q)f([0, t]\cap \mathbf{Q}) in E. For every x∈Ex\in E, let h(x)h(x) be the supremum of f−1([a,x]∩D)f^{-1}([a, x]\cap D) in I\mathbf{I}. The mappings g:I→Eg:\mathbf{I}\rightarrow E and h:E→Ih: E\rightarrow \mathbf{I} thus defined are increasing, gg coincides with ff in I∩Q\mathbf{I}\cap \mathbf{Q} and hh coincides with f−1f^{-1} in D.

We therefore have g(h(y))=yg(h(y)) =y for every y∈Dy\in D. Let x∈Ex\in E. If one had g(h(x))>xg(h(x))> x, the interval ]x,g(h(x))[]x, g(h(x))[ would contain a point yy of D and the relations g(h(y))=y<g(h(x))g(h(y)) =y < g(h(x)) would contradict the fact that g∘hg\circ h is increasing. Analogously, one does not have g(h(x))<xg(h(x))< x. We therefore have g(h(x))=xg(h(x)) =x, which proves that g∘hg\circ h is the identity mapping of E. One proves analogously that h∘gh\circ g is the identity mapping of I\mathbf{I}. Thus, g:I→Eg:\mathbf{I}\rightarrow E and h:E→Ih: E\rightarrow \mathbf{I} are reciprocal isomorphisms of ordered sets.

Remark 01XY

Let E be a totally ordered set. The set of open intervals of E (bounded or not) is stable under finite intersection. It is a basis of a topology T0(E)\mathscr{T}_0(E) on E (TG, I, p. 91, exerc. 5). The topology T0(I)\mathscr{T}_0(\mathbf{I}) is identical with the topology induced on I\mathbf{I} by that of R\mathbf{R}. It follows that, in prop. 16, every isomorphism of ordered sets of I\mathbf{I} onto E is a homeomorphism of I\mathbf{I} onto the topological space obtained by endowing E with the topology T0(E)\mathscr{T}_0(E).

Corollary 01XZ

Let R be an equivalence relation in I\mathbf{I}. The following conditions are equivalent:

(i) Every equivalence class according to R is a closed interval of I\mathbf{I}, distinct from I\mathbf{I};

(ii) There exists an increasing surjective mapping u:I→Iu:\mathbf{I}\rightarrow \mathbf{I} such that R is the equivalence relation associated with uu.

Such a mapping uu, when it exists, is continuous and defines by passing to the quotient a homeomorphism of I/R\mathbf{I}/R onto I\mathbf{I}.

We shall denote by p:I→I/Rp:\mathbf{I}\rightarrow \mathbf{I}/R the canonical surjection.

Suppose condition (i) is satisfied. For A and B equivalence classes following R, write A<BA<B if one has a<ba < b for every a∈Aa\in A and every b∈Bb\in B. In I/R\mathbf{I}/R, the relation « A = B or A<BA<B » is an order relation. Indeed, it is reflexive; it is antisymmetric since one cannot have simultaneously A<BA<B and B<AB<A; it is transitive since the relations A<BA<B and B<CB<C imply A<CA<C. If A and B are distinct elements of I/R\mathbf{I}/R, they are disjoint closed intervals of I\mathbf{I}, and one then has either A<BA<B, or B<AB<A. Endowed with the order relation thus defined, I/R\mathbf{I}/R is therefore totally ordered. The mapping p:I→I/Rp:\mathbf{I}\rightarrow \mathbf{I}/R is increasing.

The set I/R\mathbf{I}/R is not reduced to one element, by virtue of (i).

Let F be a subset of I/R\mathbf{I}/R. Let us prove that F has a least upper bound in I/R\mathbf{I}/R. Put F′=p−1(F)F'=\overset{-1}{p}(F); denote by aa the least upper bound of F′F' in I\mathbf{I} and by A the equivalence class of aa following R. Since aa is an upper bound of F′F' in I\mathbf{I}, A is an upper bound of F in I/R\mathbf{I}/R. Conversely, let B∈I/RB\in \mathbf{I}/R be an upper bound of F; put b=b= sup(B). Every element of F′F' is then bounded above by bb. We therefore have a⩽ba\leqslant b. Since the equivalence classes following R are closed intervals, aa belongs to A and bb to B. Consequently we have A=p(a)⩽p(b)=BA =p(a)\leqslant p(b) = B. This proves that A is the least upper bound of F.

Let A and B be elements of I/R\mathbf{I}/R such that A<BA<B. Let aa be the greatest lower bound of A and bb the least upper bound of B. Since a∈Aa\in A and b∈Bb\in B, we have a<ba < b. The equivalence class following R of any element of ]a,b[]a, b[ is an element of the interval ]A,B[]A,B[ of I/R\mathbf{I}/R. Since I∩Q\mathbf{I}\cap \mathbf{Q} meets ]a,b[]a, b[, its image by pp meets ]A,B[]A,B[.

We have thus proved that the totally ordered set I/R\mathbf{I}/R satisfies the hypotheses of prop. 16. There therefore exists an isomorphism of ordered sets f:I/R→If:\mathbf{I}/R\rightarrow \mathbf{I}. The mapping u=f∘pu=f\circ p is a surjective and increasing mapping of I\mathbf{I} onto I\mathbf{I} and the equivalence relation associated with uu is the relation R; this proves that condition (ii) is satisfied.

Conversely, suppose that condition (ii) is satisfied. Let u:I→Iu:\mathbf{I}\rightarrow \mathbf{I} be an increasing and surjective mapping such that R is the equivalence relation associated with uu.

Let aa be a point of I\mathbf{I}. The set A=u−1(a)A =\overset{-1}{u}(a) is an interval of I\mathbf{I}, since the mapping uu is increasing. Since uu is surjective, A is neither empty nor equal to I\mathbf{I}. Let bb be the greatest lower bound of A in I\mathbf{I}. We have u(b)⩾au(b)\geqslant a. If u(b)>au(b)> a, there exist c∈]a,u(b)[c\in ]a, u(b)[ and d∈Id\in \mathbf{I} such that u(d)=cu(d) =c since uu is surjective. Since uu is increasing and a<u(d)<u(b),da < u(d)< u(b),d majorizes every element of A and we have d<bd < b, which contradicts the assumption that bb is the greatest lower bound of A. We therefore have u(b)=au(b) =a, that is to say b∈Ab\in A. Analogously one proves that A contains its greatest lower bound. The set A is therefore a closed interval of I\mathbf{I}, distinct from I\mathbf{I}. This proves that condition (i) is satisfied.

Let us now prove that the mapping uu is continuous. It suffices for this to prove that, for every a∈Ia\in \mathbf{I}, the sets u−1(]a,→[)\overset{-1}{u}(]a,\rightarrow [) and u−1(]←,a[)\overset{-1}{u}(]\leftarrow , a[) are open. Let bb be the greatest lower bound of u−1(a)\overset{-1}{u}(a); we have u(b)=au(b) =a. If x∈Ix\in \mathbf{I} satisfies u(x)>au(x)> a, we necessarily have x>bx > b; conversely, if x>bx > b, we have u(x)⩾au(x)\geqslant a and u(x)=a̸u(x)=\not a since bb is the greatest lower bound of u−1(a)\overset{-1}{u}(a). Consequently u−1(]a,→[)=]b,→[\overset{-1}{u}(]a,\rightarrow [) = ]b,\rightarrow [, which proves that the inverse image by uu of the interval ]a,→[]a,\rightarrow [ is open. Analogously one proves that that of ]←,a[]\leftarrow , a[ is open. It follows that the mapping uu is continuous.

The mapping v:I/R→Iv:\mathbf{I}/R\rightarrow \mathbf{I} deduced from uu by passing to the quotient is then continuous and bijective. Since I\mathbf{I} is compact, I/R\mathbf{I}/R is quasi-compact (TG, I, p. 62, th. 2) and vv is a homeomorphism (TG, I, p. 63, cor. 2).

Proposition 17 01Y0

Let X be a connected and compact topological space. Let aa be a point of X, let U be a nonempty open and closed subset of X −{a}-\{a\}.

(a) The closure U‾\overline{U} of U in X is equal to U∪{a}U\cup \{a\} and is connected.

b) Let a′a' be a point of X distinct from aa, and let U′U' be a nonempty open and closed subset of X −{a′}-\{a'\}. If a∉U′a\notin U' and if U‾∩U′‾=∅̸\overline{U}\cap \overline{U'}=\not\emptyset, then U′‾⊂U\overline{U'}\subset U. Conversely, if a∈U′a\in U' and X =U∪̸U′= U\not\cup U', then U‾⊂U′\overline{U}\subset U'.

c) There exists a point bb of U such that X −{b}-\{b\} is connected.

Let us prove assertion a). Denote by V the complement of U in X −{a}-\{a\}. Since U is closed in X −{a}-\{a\}, V is open in X−{a}-\{a\} and fortiori in X. We therefore have U⊂U‾⊂X−V=U∪{a}U\subset \overline{U}\subset X-V = U\cup \{a\}. Analogously, U is an open subset of X. Since X is connected, U is not closed in X, whence the equality U‾=U∪{a}\overline{U}= U\cup \{a\}. We likewise have V‾=V∪{a}\overline{V}= V\cup \{a\}.

Let F and G be disjoint closed subsets of U‾\overline{U} such that U=F∪GU = F\cup G. Suppose that a∈Ga\in G and let us prove that F is empty; one would reason in the same way if a∈Fa\in F. The set G∪V=G∪V‾G\cup V = G\cup \overline{V} is a closed subset of X, disjoint from F, and we have X=U‾∪V=F∪(G∪V)X =\overline{U}\cup V = F\cup (G\cup V). Since X is connected and G is not empty, F is empty. This proves that U‾\overline{U} is connected.

Let us prove b). Suppose that a∉U′a\notin U' and that U‾∩U′‾=∅̸\overline{U}\cap \overline{U'}=\not\emptyset. By assertion a), we have U=U∪{a}U = U\cup \{a\} and U′‾=U′∪{a′}\overline{U'}= U'\cup \{a'\}. Since a∉U′a\notin U' and a≠a′a=\not a', the subsets U and U′‾\overline{U'} have a common point. Still by a), U′‾\overline{U'} is a connected subset of X; since it is contained in X−{a}-\{a\} and it meets the open and closed subset U of the latter, we have U′‾⊂U\overline{U'}\subset U, which was to be proved. The second assertion follows from the first by considering the complements in X of U‾\overline{U} and U′‾\overline{U'} respectively.

Let us finally prove assertion c). Suppose, by contradiction, that, for every x∈Ux\in U, the set X −{x}-\{x\} is not connected and choose subsets UxU_x and VxV_x, open and closed in X −{x}-\{x\}, disjoint and nonempty, such that X −{x}=Ux∪Vx-\{x\}= U_x\cup V_x and a∈Vxa\in V_x. By assertion b), applied to the open and closed subsets U, UxU_x of X−{a}-\{a\} and X−{x}-\{x\} respectively, we have Ux‾⊂U\overline{U_x}\subset U for every x∈Ux\in U. Let xx and yy be points of U such that x∈Uyx\in U_y; we then have x=y̸x=\not y. Again by assertion b), applied to the open and closed subsets UxU_x and UyU_y of X −{x}-\{x\} and X −{y}-\{y\}, the relation x∈Uyx\in U_y implies the relation Ux‾⊂Uy‾\overline{U_x}\subset \overline{U_y}. Consequently, the relations x∈Uyx\in U_y and Ux‾⊂Uy‾\overline{U_x}\subset \overline{U_y} are equivalent.

The set of subsets S of U such that ⋂x∈SUx‾=∅̸\bigcap_{x\in S}\overline{U_x}=\not\emptyset is of finite character (E, III, p. 34), since the space X is compact. By E, III, p. 35, th. 1, there exists a maximal subset S of U such that C=⋂x∈SUx‾C =\bigcap_{x\in S}\overline{U_x} is not empty. Let cc be a point of C. For every element xx of S, we have c∈Ux‾c\in \overline{U_x}, whence c∈Uc\in U and then Uc‾⊂Ux‾\overline{U_c}\subset \overline{U_x}. Consequently, we have Uc‾⊂C\overline{U_c}\subset C and then, by maximality of S, C=UcC = U_c. For every x∈Cx\in C such that x=c̸x=\not c, we also have Ux‾⊂Uc‾\overline{U_x}\subset \overline{U_c} and Ux‾=U̸c\overline{U_x}= U\not_c. By maximality of S, C={c}C =\{c\}, hence Uc‾={c}\overline{U_c}=\{c\}, which contradicts the hypothesis that UcU_c is a nonempty open and closed subset of X −{c}-\{c\}.

Corollary 01Y1

Let X be a compact connected topological space, let N be the set of points of X whose complement is connected. The set X is the unique connected and compact subset of X which contains N.

Let S be a connected and compact subset of X such that N⊂SN\subset S. Suppose that S =X̸= X\not and let x∈X−Sx\in X-S. By hypothesis, X −{x}-\{x\} is not connected; there therefore exist open and closed subsets U and V of X −{x}-\{x\}, disjoint and nonempty, such that X −{x}=U∪V-\{x\}= U\cup V. We may suppose that S⊂VS\subset V. We have U‾=U∪{x}\overline{U}= U\cup \{x\} and V=V∪{x}V = V\cup \{x\} and these spaces are connected (III, p. 278, prop. 17, a)). By assertion c) of this proposition, there exists y∈Uy\in U such that X −{y}-\{y\} is connected, which contradicts the inclusions N⊂S⊂VN\subset S\subset V.

Lemma 7 01Y2

Soit T un espace topologique localement connexe. Soit U\mathscr{U} un ensemble filtrant croissant de parties ouvertes de T, de réunion T. Pour t∈Tt\in T et U∈UU\in \mathscr{U}, notons UtU_t la composante connexe de tt dans U si t∈Ut\in U et posons Ut=∅U_t=\emptyset si t∉Ut\notin U. Pour tout t∈T,⋃U∈UUtt\in T,\bigcup_{U\in\mathscr{U}}U_t est la composante connexe de tt dans T.

For t∈Tt\in T, put Ct=⋃U∈UUtC_t=\bigcup_{U\in\mathscr{U}}U_t. We have t∈Ctt\in C_t. The sets UtU_t are open and connected and the same is true of CtC_t since we have t∈Utt\in U_t if Ut=∅̸U_t=\not\emptyset (TG, I, p. 81, prop. 2). Let u,vu,v be points of T such that Cu∩Cv=∅̸C_u\cap C_v=\not\emptyset. Let tt be a point of Cu∩CvC_u\cap C_v; let U∈UU\in \mathscr{U} be such that t∈Uut\in U_u and let V∈UV\in \mathscr{U} be such that v∈Vvv\in V_v. Since U\mathscr{U} is increasing filtrant, there exists W∈UW\in \mathscr{U} which contains U∪VU\cup V. Then, Wu∩Wv=∅̸W_u\cap W_v=\not\emptyset, hence Wu=WvW_u= W_v. More generally, one has Wu′=Wv′W'_u= W'_v for every W′∈UW'\in \mathscr{U} such that W⊂W′W\subset W', hence Cu=CvC_u= C_v. This proves that the sets of the form CtC_t form a partition of T into open connected subsets. Consequently, for every t∈Tt\in T, CtC_t is the connected component of tt in T.

Theorem 2 01Y3

Let X be a compact connected topological space possessing a countable everywhere dense subset. Let a,ba,b be distinct points of X. The following conditions are equivalent:

(i) There exists a homeomorphism f:I→Xf:\mathbf{I}\rightarrow X such that f(0)=af(0) =a and f(1)=bf(1) =b;

(ii) Every connected subset of X which contains {a,b}\{a, b\} is equal to X ;

(iii) The space X is locally connected and every connected compact subset of X which contains {a,b}\{a, b\} is equal to X ;

(iv) For every x∈X−{a,b}x\in X-\{a, b\}, the space X −{x}-\{x\} is not connected.

Assertion (i) implies all the others.

Let xx be a point of X −{a,b}-\{a, b\}. Since X −{x}-\{x\} contains {a,b}\{a, b\}, assertion (ii) implies that it is not a connected subset of X, so that (ii) implies (iv).

Let us show that (iii) implies (iv). Suppose the assumptions of (iii) are satisfied. Let x∈X−{a,b}x\in X-\{a, b\} and suppose that X−{x}-\{x\} is connected. Let T be the space X −{x}-\{x\} and let U\mathscr{U} be the set of subsets of T of the form X-V, where V is a compact neighbourhood of xx. By Lemma 7, there exists a compact neighbourhood V of xx such that aa and bb belong to the same connected component of X-V. They belong in particular to the same connected component of X−V˚X-\mathring{V}; this is a compact connected set of X, distinct from X, which contradicts the assumption (iii).

It remains to prove that assertion (iv) implies condition (i).

Let xx be a point of X −{a,b}-\{a, b\} and let U, V be open and closed, disjoint and nonempty subsets of X−{x}-\{x\}, such that X−{x}=U∪V-\{x\}= U\cup V. According to III, p. 278, prop. 17, c), there exists a point of U (resp. of V) whose complement in X is connected. Since X admits only two such points, aa and bb, one of them, say aa, is contained in U and the other in V. According to loc. cit., a nonempty open and closed subset, U′U' of U, contains a point of {a,b}\{a, b\}, hence contains aa. Applying this to U−U′U-U', it follows that U=U′U = U'. Consequently, U is connected; it is the connected component of aa in X−{x}-\{x\}. Analogously, V is the connected component of bb in X −{x}-\{x\}.

For every x∈X−{a,b}x\in X-\{a, b\}, denote thus by UxU_x and VxV_x the connected components of aa and bb in X −{x}-\{x\}. From what precedes, they are open and closed in X −{x}-\{x\}, disjoint, and their union is equal to X −{x}-\{x\}.

Denote by ≼\preccurlyeq the relation in X defined in the following way: on the one hand, a≼xa\preccurlyeq x and x≼bx\preccurlyeq b for every x∈Xx\in X, on the other hand, if xx and yy belong to X −{a,b}-\{a, b\}, then x≼yx\preccurlyeq y if x∈Uy‾x\in \overline{U_y}. For xx and yy in X −{a,b}-\{a, b\}, the subsets VxV_x and VyV_y have the point bb in common, the relation x≼yx\preccurlyeq y is in fact equivalent to the assertion Ux‾⊂Uy‾(\overline{U_x}\subset \overline{U_y}(III, p. 278, prop. 17, b)). It follows that the relation ≼\preccurlyeq is an order relation in X.

Let xx and yy be points of X such that one does not have x≼yx\preccurlyeq y. Necessarily, x=a̸x=\not a and y=b̸y=\not b, and x∈Vyx\in V_y. If x=bx=b or y=ay=a, one has y≼xy\preccurlyeq x. Suppose then that xx and yy are distinct from aa and bb. Since the parts UxU_x and UyU_y have the point aa in common, one has the inclusion Vx‾⊂Vy‾(\overline{V_x}\subset \overline{V_y}(loc. cit.), whence, taking the closures of the complements, Uy‾⊂Ux‾\overline{U_y}\subset \overline{U_x} and, fortiori, y≼xy\preccurlyeq x. The relation ≼\preccurlyeq in the space X is therefore a total order relation. For every x∈X−{a,b}x\in X-\{a, b\}, one moreover has Ux=]←,x[U_x= ]\leftarrow , x[ and Vx=]x,→[V_x= ]x,\rightarrow [; for x,y∈X−{a,b}x, y\in X-\{a, b\}, one has ]x,y[=Uy∩Vx]x, y[ = U_y\cap V_x. When x,yx, y run through the points of X −{a,b}-\{a, b\}, the sets Uy∩VxU_y\cap V_x, the sets UyU_y and the sets VxV_x form a basis of a topology on X. Let X~\widetilde{X} be the topological space corresponding to it and let i:X→X~i: X\rightarrow \widetilde{X} be the identical mapping of X. Since, for every x∈Xx\in X, the sets UxU_x and VxV_x are open in X, the mapping ii is continuous.

The space X~\widetilde{X} is separated. Indeed, let xx and yy be distinct points of X~\widetilde{X} such that x≼yx\preccurlyeq y. The parts ]←,y[]\leftarrow , y[ and ]x,→[]x,\rightarrow [ are open neighbourhoods of xx and yy; if they have a common point zz, ]←,z[]\leftarrow , z[ and ]z,→[]z,\rightarrow [ are then disjoint open neighbourhoods of xx and yy. Since X is compact, the mapping ii is therefore a homeomorphism (TG, I, p. 63, cor. 2).

Consequently, the image by ii of a countable everywhere dense subset of X meets every nonempty open interval of the ordered set (X,≼)(X,\preccurlyeq ). It follows then from Prop. 16 of III, p. 276 that there exists an isomorphism cc of the ordered set I\mathbf{I} onto (X,≺)(X,\prec ). According to the remark following this proposition, this isomorphism is a homeomorphism of I\mathbf{I} onto the topological space X~\widetilde{X}. The mapping f=i−1∘cf=i^{-1}\circ c is then a homeomorphism of I\mathbf{I} onto X which maps 0 onto aa and 1 onto bb.

9. Injective paths

Proposition 18 01Y4

Let X be a separated topological space. Let aa and bb be distinct points of X which belong to the same arcwise connected component of X. There exists an injective path joining aa to bb in X.

Let f:I→Xf:\mathbf{I}\rightarrow X be a path joining aa to bb in X. Let U\mathscr{U} be the set of open subsets U of ]0,1[]0,1[ such that f(x)=f(y)f(x) =f(y) for every connected component ]x,y[]x, y[ of U.

Lemma 8 01Y5

The set U\mathscr{U}, ordered by inclusion, is inductive.

Let V\mathscr{V} be a totally ordered subset of U\mathscr{U}. Let us prove that the union V of the sets belonging to V\mathscr{V} is an element of U\mathscr{U}, that is to say that, for every connected component ]u,v[]u, v[ of V, one has f(u)=f(v)f(u) =f(v).

Let xx be a point of ]u,v[]u, v[. Let Vx\mathscr{V}_x be the set of the U∈VU\in \mathscr{V} such that x∈Ux\in U; for such a U, denote by ]uU,vU[]u_U, v_U[ the connected component of xx in U. One has uU′⩽uU<vU⩽vU′u_{U'}\leqslant u_U< v_U\leqslant v_{U'} if U and U′U' are elements of Vx\mathscr{V}_x such that U⊂U′U\subset U'. Since the union for U∈UxU\in \mathscr{U}_x of the ]uU,vU[]u_U, v_U[ is equal to ]u,v[]u, v[ according to Lemma 7 of III, p. 280, one has u=u= lim uUu_U and v=v= lim vUv_U, where the limits are taken following the filtered set Vx\mathscr{V}_x. Since the mapping ff is continuous and the topological space X is separated, the equalities f(uU)=f(vU)f(u_U) =f(v_U) for every U∈VxU\in \mathscr{V}_x imply that f(u)=f(v)f(u) =f(v), which was to be proved.

By virtue of E, III, p. 20, Th. 2, there exists an open subset U belonging to U\mathscr{U} which is maximal for the relation of inclusion.

Let g:I→Xg:\mathbf{I}\rightarrow X be the mapping defined as follows. If t∉Ut\notin U, we put g(t)=f(t)g(t) =f(t); otherwise, denoting by ]u,v[]u, v[ the connected component of tt in U, we put g(t)=f(u)=f(v)g(t) =f(u) =f(v), so that the mapping g∣[u,v]g|[u, v] is constant with image f(u)f(u).

Prop. 18 follows from the following lemma.

Lemma 9 01Y6

There exists a continuous, increasing and surjective mapping u:I→Iu:\mathbf{I}\rightarrow \mathbf{I} and an injective path c:I→Xc:\mathbf{I}\rightarrow X joining aa to bb such that g=c∘ug=c\circ u.

Let us first prove that the mapping gg is continuous. Let tt be a point of I\mathbf{I}. If t∈U,gt\in U,g is constant in a neighbourhood of tt, hence continuous at tt. Suppose that t∉Ut\notin U and let W be an open neighbourhood of g(t)g(t) in X. Let V be an open interval in I\mathbf{I} containing tt such that f(V)⊂Wf(V)\subset W; such an interval exists since ff is continuous at tt. Let us prove that g(V)⊂Wg(V)\subset W. Let x∈Vx\in V; if x∉Ux\notin U, we have g(x)=f(x)g(x) =f(x), hence g(x)∈Wg(x)\in W. Suppose then that x∈Ux\in U and let ]u,v[]u, v[ be the connected component of xx in U. Observe that ]u,v[]u, v[ does not contain tt. Consequently, if x<tx < t, we have x<v⩽tx < v\leqslant t whence v∈Vv\in V and g(x)=g(v)=f(v)∈f(V)⊂Wg(x) =g(v) =f(v)\in f(V)\subset W. The same argument shows that g(x)∈Wg(x)\in W if x>tx > t. Thus, gg is continuous at tt.

Let uu and vv be points of I\mathbf{I} such that g(u)=g(v)g(u) =g(v) and u<vu < v. Let us prove that ]u,v[⊂U]u, v[\subset U. Put U′=U∪]u,v[U'= U\cup ]u, v[; it is an open subset of ]0,1[]0,1[.

If uu belongs to U, let u′u' denote the greatest lower bound in I\mathbf{I} of the connected component of uu in U; put u′=uu'=u if uu does not belong to U. Analogously, if vv belongs to U, let v′v' denote the least upper bound in I\mathbf{I} of the connected component of vv in U; put v′=vv'=v if vv does not belong to U. Then, ]u′,v′[]u', v'[ is a connected component of U′U' and one has f(u′)=g(u)=g(v)=f(v′)f(u') =g(u) =g(v) =f(v'). Since the connected components of U′U' distinct from ]u′,v′[]u', v'[ are connected components of U, the open set U′U' is an element of U\mathscr{U}. Since U is a maximal element of U\mathscr{U} for the relation of inclusion, one has U′=UU'= U and ]u,v[]u, v[ is contained in U. This proves that gg is constant on the interval [u,v][u, v]. The fibres of gg are therefore intervals of I\mathbf{I}, and these intervals are closed because gg is continuous.

Let R be the equivalence relation associated with gg and let pp be the canonical surjection of I\mathbf{I} onto I/R\mathbf{I}/R. There exists a unique continuous mapping g′g' of I/R\mathbf{I}/R into X such that g=g′∘pg=g'\circ p; this mapping is injective. By the corollary, III, p. 276, of prop. 16, there exists a mapping uu, increasing, continuous and surjective, such that R is the equivalence relation associated with uu. Since the space I\mathbf{I} is compact and the space X is separated, the mapping uu is closed, hence strict (I, p. 18, example 2). It defines by passing to the quotient a homeomorphism u′u' of I/R\mathbf{I}/R onto I\mathbf{I}. Then, the mapping g′∘(u′)−1g'\circ (u')^{-1} of I\mathbf{I} into X is an injective path of origin aa and of term bb.

10. Lifting of paths

Theorem 3 01Y7

Let I be an interval of R\mathbf{R} and let X be a nonempty and separated topological space. Let p:X→Ip: X\rightarrow I be a continuous, open and proper mapping whose fibres are totally disconnected (TG, I, p. 83). The mapping pp is surjective. For every point xx of X, it possesses a continuous section ss such that s(p(x))=xs(p(x)) =x.

The set p(X)p(X) is an open, closed (TG, I, p. 72, prop. 1), nonempty subset of I. Since I is connected, one has p(X)=Ip(X) = I; the mapping pp is therefore surjective.

For every pair (a,b)∈I×I(a, b)\in I\times I such that a⩽ba\leqslant b, let Fa,bF_{a,b} denote the set of pairs (y,z)∈p−1(a)×p−1(b)(y, z)\in \overset{-1}{p}(a)\times \overset{-1}{p}(b) such that yy and zz belong to one and the same connected component of p−1([a,b])\overset{-1}{p}([a, b]).

Lemma 10 01Y8

Let a,ba,b be points of I such that a⩽ba\leqslant b.

a) The set Fa,bF_{a,b} is closed in p−1(a)×p−1(b)\overset{-1}{p}(a)\times \overset{-1}{p}(b).

b) One has pr1(Fa,b)=p−1(a)_1(F_{a,b}) =\overset{-1}{p}(a) and pr2(Fa,b)=p−1(b)_2(F_{a,b}) =\overset{-1}{p}(b).

c) Let c∈Ic\in I such that b⩽cb\leqslant c. If (y,z)(y, z) belongs to Fa,bF_{a,b} and (z,t)(z, t) belongs to Fb,cF_{b,c}, then (y,t)(y, t) belongs to Fa,cF_{a,c}.

The set p−1([a,b])\overset{-1}{p}([a, b]) is compact (TG, I, p. 77, Prop. 7). Hence, for a pair (y,z)∈p−1(a)×p−1(b)(y, z)\in \overset{-1}{p}(a)\times \overset{-1}{p}(b) to belong to Fa,bF_{a,b}, it is necessary and sufficient that every open and closed subset of p−1([a,b])\overset{-1}{p}([a, b]) which contains yy should contain zz (TG, II, p. 32, Prop. 6).

Put Y=p−1([a,b])Y =\overset{-1}{p}([a, b]). For every open and closed subset U of Y, the set ((U×U)∪((Y−U)×(Y−U)))∩(p−1(a)×p−1(b))((U\times U)\cup ((Y-U)\times (Y-U)))\cap (\overset{-1}{p}(a)\times \overset{-1}{p}(b)) is closed in p−1(a)×p−1(b)\overset{-1}{p}(a)\times \overset{-1}{p}(b). The intersection of these sets is equal to Fa,bF_{a,b}, whence a).

Let y∈p−1(a)y\in \overset{-1}{p}(a). Let U\mathscr{U} denote the set of open and closed neighbourhoods of yy in Y. The mapping of Y into [a,b][a, b] deduced from pp by passing to subspaces is open and proper (I, p. 17, Prop. 8), hence also closed. It follows that, for every set U belonging to U,p(U)\mathscr{U},p(U) is a nonempty open and closed subset of [a,b][a, b]; since the interval [a,b][a, b] is connected, one has p(U)=[a,b]p(U) = [a, b] and in particular U∩p−1(b)=∅̸U\cap \overset{-1}{p}(b)=\not\emptyset. Hence, (U∩p−1(b))U∈U(U\cap \overset{-1}{p}(b))_{U\in\mathscr{U}} is a decreasing filtering family of nonempty closed subsets of the compact space p−1(b)\overset{-1}{p}(b). The intersection of this family is not empty (TG, I, p. 59); let zz be one of its elements. One has (y,z)∈Fa,b(y, z)\in F_{a,b}. We have proved the relation pr1(Fa,b)=p−1(a)_1(F_{a,b}) =\overset{-1}{p}(a). The relation pr2(Fa,b)=p−1(b)_2(F_{a,b}) =\overset{-1}{p}(b) is deduced from it by replacing pp, I,a,b,a,b by −p-p, −I,−b,−a-I,-b,-a.

Under the hypotheses of c), the pair (y,t)(y, t) belongs to p−1(a)×p−1(c)\overset{-1}{p}(a)\times \overset{-1}{p}(c), the set {y,z}\{y, z\} is contained in a connected part C of p−1([a,b])\overset{-1}{p}([a, b]) and the set {z,t}\{z, t\} is contained in a connected part C′C' of p−1([b,c])\overset{-1}{p}([b, c]). Then, C∪C′C\cup C' is a connected part of p−1([a,c])\overset{-1}{p}([a, c]) (TG, I, p. 81, prop. 2) and contains {y,t}\{y, t\}, whence the relation (y,t)∈Fa,c(y, t)\in F_{a,c}.

Let us return to the proof of theorem 3. Each fibre of the mapping pp is compact (TG, I, p. 77, prop. 7). According to Tychonoff's theorem (TG, I, p. 63, th. 3), the product space K=∏a∈Ip−1(a)K =\prod_{a\in I}\overset{-1}{p}(a) is compact. Let K′K' be the set of elements (ya)a∈I(y_a)_{a\in I} of K such that yp(x)y_{p(x)} is equal to xx and such that one has (ya,yb)∈Fa,b(y_a, y_b)\in F_{a,b} for every pair (a,b)(a, b) of elements of I such that a<ba < b. Theorem 3 follows from the following lemma.

Lemma 11 01Y9

a) The set K′K' is not empty.

b) Let (sa)a∈I(s_a)_{a\in I} be an element of K′K'. The mapping s:a↦sas:a\mapsto s_a from I into X is a continuous section of pp such that s(p(x))=xs(p(x)) =x.

For every finite subset S of I containing the point p(x)p(x), denote by KSK_S the set of elements (ya)a∈I(y_a)_{a\in I} of K satisfying the relation yp(x)=xy_{p(x)}=x and the relations (ya,yb)∈Fa,b(y_a, y_b)\in F_{a,b} for every pair (a,b)(a, b) of elements of S such that a<ba < b. The sets KSK_S are closed in K (lemma 10, a)) and form a decreasing filtering family of subsets of K, whose intersection is K′K'. To prove that K′K' is not empty, it is enough to prove that, for every finite subset S of I containing the point p(x)p(x), the set KSK_S is not empty (TG, I, p. 59).

Let S be such a subset; order its elements in a strictly increasing sequence (a1,...,an)(a_1, . . . , a_n) and denote by ii the integer such that p(x)=aip(x) =a_i. Put yai=xy_{a_i}=x. Lemma 10, b), permits us to construct by induction elements yaj∈p−1(aj)y_{a_j}\in \overset{-1}{p}(a_j), for i<j⩽ni < j\leqslant n, and by descending induction elements yaj∈p−1(aj)y_{a_j}\in \overset{-1}{p}(a_j) for 1⩽j<i1\leqslant j < i, in such a way that one has (yaj,yaj+1)∈Faj,aj+1(y_{a_j}, y_{a_{j+1}})\in F_{a_j,a_{j+1}} for every integer jj such that 1⩽j<n1\leqslant j < n. According to lemma 10, c), one has (ya,yb)∈Fa,b(y_a, y_b)\in F_{a,b} for every pair (a,b)(a, b) of elements of S such that a<ba < b. Since the mapping pp is surjective, we can choose for every a∈I−Sa\in I-S an element ya∈p−1(a)y_a\in \overset{-1}{p}(a). The family (ya)a∈I(y_a)_{a\in I} thus constructed belongs to KSK_S, hence KSK_S is not empty.

Let us prove b). By definition of K′,sK',s is a section of pp such that s(p(x))=xs(p(x)) =x. Let a∈Ia\in I; let us prove the continuity of ss at the point aa. Let U be an open neighbourhood of sas_a in X. Since p−1(a)\overset{-1}{p}(a) is a compact space (TG, I, p. 77, prop. 7) and totally disconnected, sas_a possesses in p−1(a)\overset{-1}{p}(a) an open and closed neighbourhood C, contained in U (TG, II, p. 32, corollary). The sets C and p−1(a)−C\overset{-1}{p}(a)-C are closed in p−1(a)\overset{-1}{p}(a), hence compact, and they are disjoint. Since X is separated, they possess in X open and disjoint neighbourhoods V and V′V' (TG, I, p. 61, prop. 3). The set (V∩U)∪V′(V\cap U)\cup V' is an open neighbourhood of the fibre p−1(a)\overset{-1}{p}(a) in X; since the mapping pp is closed (TG, I, p. 72, prop. 1), (V∩U)∪V′(V\cap U)\cup V' contains a set of the form p−1(J)\overset{-1}{p}(J), where J⊂IJ\subset I is an open interval containing a(a(I, p. 75, lemma). Put W=V∩U∩p−1(J)W = V\cap U\cap \overset{-1}{p}(J). The set W is open in p−1(J)\overset{-1}{p}(J); it is also closed in p−1(J)\overset{-1}{p}(J) since one has p−1(J)−W=V′∩p−1(J)\overset{-1}{p}(J)-W = V'\cap \overset{-1}{p}(J). Let b∈Jb\in J. The closed interval of I with endpoints aa and bb is contained in J. By assumption, (sa,sb)(s_a, s_b) belongs to Fa,bF_{a,b} if a⩽ba\leqslant b and (sb,sa)(s_b, s_a) belongs to Fb,aF_{b,a} if b⩽ab\leqslant a. There therefore exists a connected subset of p−1(J)\overset{-1}{p}(J) containing {sa,sb}\{s_a, s_b\}. Consequently, the point sbs_b belongs to every open and closed subset of p−1(J)\overset{-1}{p}(J) which contains sas_a, hence in particular to W; fortiori, sbs_b belongs to U. We therefore have s(J)⊂Us(J)\subset U, which proves the continuity of ss at the point aa.

Corollary 01YA

Let X and B be topological spaces and let p:X→Bp: X\rightarrow B be a continuous, open, proper and separated mapping whose fibres are totally disconnected. Let I be an interval of R\mathbf{R}, let f:I→Bf: I\rightarrow B be a continuous mapping, let aa be a point of I and let xx be a point of X such that f(a)=p(x)f(a) =p(x). There exists a continuous mapping g:I→Xg: I\rightarrow X such that p∘g=fp\circ g=f and g(a)=xg(a) =x.

Posons X′=I×BXX'= I\times_BX and denote by p′:X′→p': X'\rightarrow I and f′:X′→f': X'\rightarrow X the canonical projections. The mapping p′p' is continuous, open, proper and separated (I, p. 17, prop. 8 and p. 27, prop. 4). Since the space I is separated, the space X′X' is separated (I, p. 26, remark 3). The fibres of p′p' are totally disconnected ( I, p. 10, corollary, a)). According to Theorem 3, there exists a continuous section s′s' of p′p' which takes in aa the value (a,x)(a, x). The mapping g=f′∘s′g=f'\circ s' of I into X is continuous, we have p∘g=p∘f′∘s′=f∘p′∘s′=fp\circ g=p\circ f'\circ s'=f\circ p'\circ s'=f and g(a)=xg(a) =x, whence the corollary.

Theorem 4 01YB

Let X be a topological space, let G be a discrete group operating properly in X and let p:X→X/Gp: X\rightarrow X/G be the canonical mapping. Let I be an interval of R\mathbf{R}, let f:I→X/Gf: I\rightarrow X/G be a continuous mapping, let aa be a point of I and let xx be a point of X such that f(a)=p(x)f(a) =p(x). There exists a continuous mapping φ:I→X\varphi : I\rightarrow X such that p∘φ=fp\circ \varphi =f and φ(a)=x\varphi (a) =x.

We first treat the case where I is a closed bounded interval of R\mathbf{R}.

According to TG, III, p. 29, prop. 3, the space X is separated.

Let yy be a point of X/GX/G and let x∈Xx\in X such that y=p(x)y=p(x). The stabilizer KxK_x of xx is a finite subgroup of G; moreover, there exist neighbourhoods UxU_x of xx in X and VyV_y of yy in X/GX/G such that UxU_x is stable by Kx,gUx∩Ux=∅K_x,gU_x\cap U_x=\emptyset if g∉Kxg\notin K_x and pp induces a homeomorphism of Ux/KxU_x/K_x onto VyV_y (TG, III, p. 32, proposition 8). Moreover, for every g∈G,pg\in G,p induces a homeomorphism of gUxgU_x onto VyV_y. Since I is compact, there exist integers mm and nn such that m⩽0⩽nm\leqslant 0\leqslant n and a finite sequence (ai)m⩽i⩽n(a_i)_{m\leqslant i\leqslant n} of elements of I such that a0=a,I=[am,an]a_0=a,I = [a_m, a_n], and such that, for every i∈{m,...,n−1},f([ai,ai+1])i\in \{m, . . . , n-1\},f([a_i, a_{i+1}]) is contained in an open set VyiV_{y_i} of X/GX/G constructed as above.

Soit x0x_0 the unique element of p−1(y0)\overset{-1}{p}(y_0) such that x∈Ux0x\in U_{x_0}. Let q0q_0 be the canonical mapping of Ux0U_{x_0} onto Ux0/Kx0U_{x_0}/K_{x_0}; by passing to the quotient, the mapping pp induces a homeomorphism i0i_0 of Ux0/Kx0U_{x_0}/K_{x_0} onto Vy0V_{y_0} such that i0∘q0=p∣Ux0i_0\circ q_0=p|U_{x_0}. The mapping q0q_0 is proper (TG, III, p. 29, prop. 3), open (TG, I, p. 31, example 1) and separated, since X is separated. Its fibres are totally disconnected, since they are finite. According to the corollary, III, p. 286, there exists a continuous mapping φ0:[a0,a1]→Ux0\varphi_0: [a_0, a_1]\rightarrow U_{x_0} such that p∘φ0=f∣[a0,a1]p\circ \varphi_0=f|[a_0, a_1].

In the same way, by induction on the integer i∈{0,...,n−1}i\in \{0, . . . , n-1\}, one constructs a point xi∈p−1(yi)x_i\in \overset{-1}{p}(y_i), a continuous mapping φi:[ai,ai+1]→X\varphi_i: [a_i, a_{i+1}]\rightarrow X whose image is contained in UxiU_{x_i} such that p∘φi=f∣[ai,ai+1]p\circ \varphi_i=f|[a_i, a_{i+1}] and such that φi(ai+1)=φi+1(ai+1)\varphi_i(a_{i+1}) =\varphi_{i+1}(a_{i+1}) if 0⩽i<n−10\leqslant i < n-1.

Analogously, by decreasing induction on the integer i∈{m,...,−1}i\in \{m, . . . ,-1\}, one constructs a point xi∈p−1(yi)x_i\in \overset{-1}{p}(y_i), a continuous mapping φi:[ai,ai+1]→X\varphi_i: [a_i, a_{i+1}]\rightarrow X whose image is contained in UxiU_{x_i} such that p∘φi=p\circ \varphi_i= f∣[ai,ai+1]f|[a_i, a_{i+1}] and such that φi(ai+1)=φi+1(ai+1)\varphi_i(a_{i+1}) =\varphi_{i+1}(a_{i+1}) if m⩽i<0m\leqslant i <0.

There exists a unique mapping φ:I→\varphi : I\rightarrow X which coincides with φi\varphi_i in [ai,ai+1][a_i, a_{i+1}] for m⩽i<nm\leqslant i < n. It is continuous (TG, I, p. 19, prop. 4). It is a continuous lifting of ff to X such that φ(a)=x\varphi (a) =x.

This proves the theorem when I is compact. In the general case, there exist sequences (an)n∈N(a_n)_{n\in\mathbf{N}} and (bn)n∈N(b_n)_{n\in\mathbf{N}} such that (an)(a_n) is stationary with limit inf(I), (bn)(b_n) is stationary with limit sup(I),a=a0=b0,a=a_0=b_0, and such that (an)(a_n) (resp. (bn)(b_n)) is constant if I has a smallest (resp. greatest) element. From what precedes, there exists for every integer n∈Nn\in \mathbf{N} a continuous lifting φn\varphi_n of f∣[an,an+1]f|[a_n, a_{n+1}] to X, a continuous lifting φn′\varphi '_n of f∣[bn+1],bn]f|[b_{n+1}], b_n] to X such that φ0(a0)=φ0′(b0)=x\varphi_0(a_0) =\varphi '_0(b_0) =x φn+1(an+1)=φn(an+1),φn+1′(bn+1)=φn′(bn+1)\varphi_{n+1}(a_{n+1}) =\varphi_n(a_{n+1}),\varphi '_{n+1}(b_{n+1}) =\varphi '_n(b_{n+1}). There exists a unique mapping φ:I→X\varphi : I\rightarrow X which coincides with φn\varphi_n in [an,an+1][a_n, a_{n+1}] and with φn′\varphi '_n in [bn+1,bn][b_{n+1}, b_n], for every n∈Nn\in \mathbf{N}. It is continuous (loc. cit.) and it is a continuous lifting of ff to X such that phi(a)=xphi(a) =x. The theorem is thus proved.

Example 1 01YC

Let X be a separated topological space and let G be a finite group, endowed with the discrete topology, which operates continuously in X. The operation is then proper (TG, III, p. 28, prop. 2). The assertion of theorem 4 follows directly from the corollary of theorem 3.

Example 2 01YD

Soit nn an integer ⩾0\geqslant 0. Let PnP_n denote the set of monic polynomials P∈P\in C[X]\mathbf{C}[X] of degree nn, endowed with the topology for which the mapping (c0,...,cn−1)↦Xn+cn−1Xn−1+⋯+c0(c_0, . . . , c_{n-1})\mapsto X^n+c_{n-1}X^{n-1}+\cdots +c_0 is a homeomorphism of Cn\mathbf{C}^n onto PnP_n. The mapping pp of Cn\mathbf{C}^n into PnP_n defined by p(z1,...,zn)=(X−z1)...(X−zn)p(z_1, . . . , z_n) = (X-z_1). . .(X-z_n) is continuous. The symmetric group Sn\mathfrak{S}_n operates on Cn\mathbf{C}^n by permutation of the factors and pp defines by passing to the quotient a homeomorphism of Cn/Sn\mathbf{C}^n/\mathfrak{S}_n onto PnP_n (TG, VIII, p. 22, prop. 1, I, p. 23, cor. 1 and TG, VIII, p. 20). We therefore deduce the following statement:

Let I be an interval of R\mathbf{R}, let (c0,...,cn−1)(c_0, . . . , c_{n-1}) be a sequence of continuous mappings from I into C\mathbf{C}, let aa be a point of I and let (z1,...,zn)(z_1, . . . , z_n) be a sequence of complex numbers such that one has (X−z1)...(X−zn)=Xn+(X-z_1). . .(X-z_n) = X^n+ cn−1(a)Xn−1+⋯+c0(a)c_{n-1}(a)X^{n-1}+\cdots +c_0(a). There exists a sequence (λ1,...,λn)(\lambda_1, . . . , \lambda_n) of continuous mappings from I into C\mathbf{C} such that one has λi(a)=zi\lambda_i(a) =z_i for 1⩽i⩽n1\leqslant i\leqslant n and (X−λ1(t))...(X−λn(t))=Xn+cn−1(t)Xn−1+⋯+c0(t)(X-\lambda_1(t)). . .(X-\lambda_n(t)) = X^n+c_{n-1}(t)X^{n-1}+\cdots +c_0(t) for every t∈It\in I.

Exercises

See the exercises for § 2.

Machine translation of the English, by gpt-5-6-mini, gpt-5-mini, gpt-5-6, gpt-5.4, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.