Bourbaki

§ 1. Homotopies, homeotopies

In this paragraph and those that follow, I\mathbf{I} denotes the interval [0,1][0,1] of R\mathbf{R}.

1. Homotopic continuous mappings

Definition 1 01V4

Let X and Y be topological spaces and let ff and gg be continuous mappings of X into Y. A homotopy relating ff to gg is by definition a continuous mapping σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y such that, for every x∈Xx\in X, one has σ(x,0)=f(x)\sigma (x,0) =f(x) and σ(x,1)=g(x)\sigma (x,1) =g(x). One says that ff is homotopic to gg if there exists a homotopy relating ff to gg.

One says that ff is the origin and gg the term of the homotopy σ(\sigma (cf. below, III, p. 257, remark 2).

Let A be a subset of X. One says that the homotopy σ\sigma is fixed on A if, for every a∈Aa\in A, the mapping t↦σ(a,t)t\mapsto \sigma (a, t) of I\mathbf{I} into Y is constant. In this case, the origin and the term of σ\sigma coincide at every point of A.

Let X and Y be topological spaces. One says that homotopies σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y and τ:X×I→Y\tau : X\times \mathbf{I}\rightarrow Y are juxtaposed if the term of σ\sigma is

N. Bourbaki and Springer-Verlag Berlin Heidelberg 2 016

N.© Bourbaki, Algebraic Topology, DOI 10.1007/978-3-662-49361-8_3 229 the origin of τ\tau, in other words if one has σ(x,1)=τ(x,0)\sigma (x,1) =\tau (x,0) for every x∈Xx\in X. In this case, the mapping σ∗τ\sigma *\tau of X×IX\times \mathbf{I} into Y defined by

σ(x,2t)\sigma (x,2t) for 0⩽t⩽1/20\leqslant t\leqslant 1/2

(σ∗τ)(x,t)=(1) (\sigma *\tau )(x, t) = \tag{1}

τ(x,2t−1)\tau (x,2t-1) for 1/2⩽t⩽11/2\leqslant t\leqslant 1

is continuous (TG, I, p. 19, prop. 4) and is a homotopy relating the origin of σ\sigma to the term of τ\tau. It is called the juxtaposed homotopy of the homotopies σ\sigma and τ\tau.

If σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y is a homotopy, the mapping σ‾:X×I→Y\overline{\sigma}: X\times \mathbf{I}\rightarrow Y defined bythe origin of(σx,t(\sigma x, t. One has)↦σσ(=x,1σ−)\mapsto \frac{\sigma}{\sigma}(=x,1\sigma -. Ift)σt)\sigmais a homotopy relating the term ofand τ\tau are juxtaposed homotopiesσ\sigma to

of X×IX\times \mathbf{I} into Y, the homotopies τ‾\overline{\tau} and σ‾\overline{\sigma} are juxtaposed and one has σ∗τ‾=τ‾∗σ‾\overline{\sigma*\tau}=\overline{\tau}*\overline{\sigma}.

Proposition 1 01V5

Let X and Y be topological spaces. The relation “ ff is homotopic to gg ” is an equivalence relation in the set C(X;Y)\mathscr{C}(X; Y) of continuous mappings of X into Y.

Let ff be an element of C(X;Y)\mathscr{C}(X; Y). The mapping f∘f\circ pr1:X×I→Y_1: X\times \mathbf{I}\rightarrow Y is a homotopy relating ff to ff; this relation is therefore reflexive.

Let ff and gg be elements of C(X;Y)\mathscr{C}(X; Y) and σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y a homotopy joining ff to gg. The mapping σ‾:X×I→Y\overline{\sigma}: X\times \mathbf{I}\rightarrow Y is then a homotopy joining gg to ff; the relation considered is therefore symmetric.

Finally, let us prove that it is transitive. If f,gf,g and hh are elements of C(X;Y),σ\mathscr{C}(X; Y),\sigma a homotopy joining ff to gg and τ\tau a homotopy joining gg to hh, then σ\sigma and τ\tau are juxtaposed and σ∗τ\sigma *\tau is a homotopy joining ff to hh.

Let X and Y be topological spaces. The equivalence relation “ff is homotopic to gg” in C(X;Y)\mathscr{C}(X; Y) (Prop. 1) is called the homotopy relation. The quotient set of C(X;Y)\mathscr{C}(X; Y) by this relation is denoted by [X; Y]. Its elements are called homotopy classes of continuous mappings of X into Y. The homotopy class of a continuous mapping f:X→Yf: X\rightarrow Y will often be denoted by [f][f].

Proposition 2 01V6

Let X, Y and Z be topological spaces, ff and f′f' continuous mappings of X into Y,g,g and g′g' continuous mappings of Y into Z. If ff is homotopic to f′f' and if gg is homotopic to g′g', then g∘fg\circ f is homotopic to g′∘f′g'\circ f'.

Let σ\sigma be a homotopy joining ff to f′f' and τ\tau a homotopy joining gg to g′g'. Then the mapping θ:X×I→Z\theta : X\times \mathbf{I}\rightarrow Z defined by θ(x,t)=τ(σ(x,t),t)\theta (x, t) =\tau (\sigma (x, t), t) is a homotopy joining g∘fg\circ f to g′∘f′g'\circ f'.

Let X, Y and Z be topological spaces. Given homotopy classes φ∈[X;Y]\varphi \in [X; Y] and ψ∈[Y;Z]\psi \in [Y; Z], the mappings g∘f:X→Zg\circ f: X\rightarrow Z, where f∈φ,g∈ψf\in \varphi ,g\in \psi, all belong to one and the same homotopy class (Prop. 2), which is denoted by ψ∘φ\psi \circ \varphi and is called the composite homotopy class of the classes ψ\psi and φ\varphi. The mapping [X;Y]×[Y;Z]→[X;Z][X; Y]\times [Y; Z]\rightarrow [X; Z] which associates ψ∘φ\psi \circ \varphi to (φ,ψ)(\varphi , \psi ) is called the composition mapping.

Let φ∈[X;Y]\varphi \in [X; Y] ; we have φ=φ∘\varphi =\varphi \circ [IdX]=_X] = [IdY]∘φ_Y]\circ \varphi.

Let X, Y, Z and T be topological spaces, let φ∈[X;Y]\varphi \in [X; Y], ψ∈[Y;Z]\psi \in [Y; Z] and χ∈[Z;T]\chi \in [Z; T]. The homotopy class χ∘(ψ∘φ)\chi \circ (\psi \circ \varphi ) is equal to (χ∘ψ)∘φ(\chi \circ \psi )\circ \varphi; it is denoted by χ∘ψ∘φ\chi \circ \psi \circ \varphi.

Proposition 3 01V7

Let X be a topological space and let (Yj)j∈J(Y_j)_{j\in J} be a family of topological spaces. The mapping of [X;∏j∈JYj][X;\prod_{j\in J}Y_j] into the product set ∏j∈J[X;Yj]\prod_{j\in J}[X; Y_j] defined by φ↦\varphi \mapsto ([prj]∘φ_j]\circ \varphi )j∈J_{j\in J} is bijective.

Surjectivity results immediately from I, p. 25, prop. 1. Let us prove injectivity. Let ff and gg be continuous mappings of X into ∏j∈JYj\prod_{j\in J}Y_j. For each j∈Jj\in J, put fj=f_j= prj∘f_j\circ f and gj=g_j= prj∘g_j\circ g; suppose fjf_j homotopic to gjg_j and let σj\sigma_j be a homotopy joining fjf_j to gjg_j. The mapping σ=(σj)\sigma = (\sigma_j) of X×IX\times \mathbf{I} into ∏j∈JYj\prod_{j\in J}Y_j is continuous (loc. cit.) ; it is a homotopy joining ff to gg, whence the proposition.

Corollary 01V8

Let (Xj)j∈J(X_j)_{j\in J} and (Yj)j∈J(Y_j)_{j\in J} be families of topological spaces having the same index set. For each j∈Jj\in J, let fjf_j and gjg_j be continuous mappings of XjX_j into YjY_j. If, for each j∈Jj\in J, the mappings fjf_j and gjg_j are homotopic, the same is true of the product mappings f:(xj)↦(fj(xj))f: (x_j)\mapsto (f_j(x_j)) and g:(xj)↦(gj(xj))g: (x_j)\mapsto (g_j(x_j)) of ∏j∈JXj\prod_{j\in J}X_j into ∏j∈JYj\prod_{j\in J}Y_j.

Suppose that we have [fj]=[gj][f_j] = [g_j] for each j∈Jj\in J. We have [prj]∘[f]=_j]\circ [f] = [fj∘[f_j\circ prj]_j] and [prj]∘[g]=[gj∘_j]\circ [g] = [g_j\circ prj]_j], hence [prj]∘[f]=_j]\circ [f] = [prj]∘[g]_j]\circ [g] by prop. 2, whence [f]=[g][f] = [g] by prop. 3.

2. Pointed homotopies

Let X and Y be topological spaces and let xx be a point of X. A homotopy σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y is said to be pointed at xx if it is fixed on {x}\{x\}, that is to say if the mapping t↦σ(x,t)t\mapsto \sigma (x, t) from I\mathbf{I} into Y is constant. The juxtaposition of two homotopies pointed at xx is pointed at xx. The origin and the end of a homotopy pointed at xx take the same value yy at xx; they are therefore pointed continuous mappings from (X,x)(X, x) into (Y,y)(Y, y).

Let (X,x)(X, x) and (Y,y)(Y, y) be pointed topological spaces (I, p. 120, definition 1). The relation « ff is related to gg by a homotopy pointed at xx » is an equivalence relation in the set C((X,x);(Y,y))\mathscr{C}((X, x); (Y, y)), called the pointed homotopy relation. The quotient set of C((X,x);(Y,y))\mathscr{C}((X, x); (Y, y)) by this equivalence relation is denoted by [(X,x);(Y,y)][(X, x); (Y, y)]. Its elements are called pointed homotopy classes of pointed continuous mappings from (X,x)(X, x) into (Y,y)(Y, y).

The set C((X,x);(Y,y))\mathscr{C}((X, x); (Y, y)) is a subset of C(X;Y)\mathscr{C}(X; Y). It should be noted that pointed homotopy is a finer equivalence relation than the relation induced by homotopy in C(X;Y)\mathscr{C}(X; Y). It is in general a strictly finer relation (see III, p. 321, exerc. 1 and III, p. 234, example 3).

Let also (Z,z)(Z, z) be a pointed topological space. Let ff and f′f' be pointed continuous mappings from (X,x)(X, x) into (Y,y),g(Y, y),g and g′g' pointed continuous mappings from (Y,y)(Y, y) into (Z,z)(Z, z). If ff and f′f' are related by a homotopy σ\sigma pointed at xx, and if gg and g′g' are related by a homotopy τ\tau pointed at yy, then g∘fg\circ f and g′∘f′g'\circ f' are related by the homotopy θ:X×I→Z\theta : X\times \mathbf{I}\rightarrow Z, (u,t)↦τ(σ(u,t),t)(u, t)\mapsto \tau (\sigma (u, t), t), which is pointed at xx. As above, this makes it possible to define the composition mapping of [(X,x);(Y,y)]×[(Y,y);(Z,z)][(X, x); (Y, y)]\times [(Y, y); (Z, z)] into [(X,x);(Z,z)][(X, x); (Z, z)]. We leave to the reader the task of formulating and proving for pointed mappings and homotopies the statements analogous to prop. 3 and its corollary.

3. Homotopic spaces

Definition 2 01V9

Let X and Y be topological spaces. A homotopy class φ∈\varphi \in [X; Y] is said to be invertible if there exists a homotopy class ψ∈[Y;X]\psi \in [Y; X] such that ψ∘φ=\psi \circ \varphi = [IdX]_X] and φ∘ψ=\varphi \circ \psi = [IdY]_Y]. A continuous mapping is said to be a homotopy equivalence if its homotopy class is invertible.

Let φ∈\varphi \in [X; Y] be an invertible homotopy class. There exists a unique homotopy class ψ∈[Y;X]\psi \in [Y; X] having the properties of definition 2. Let indeed ψ,ψ′\psi ,\psi ' be classes having these properties; one has

ψ=ψ∘\psi =\psi \circ [IdY]=ψ∘φ∘ψ′=_Y] =\psi \circ \varphi \circ \psi '= [IdX]∘ψ′=ψ′_X]\circ \psi '=\psi '.

This unique class is called the inverse of the homotopy class φ\varphi and is denoted φ−1\varphi^{-1}.

Let Z be a topological space. We have χ∘φ∘φ−1=χ\chi \circ \varphi \circ \varphi^{-1}=\chi for all χ∈\chi \in [Y; Z] and θ∘φ−1∘φ=θ\theta \circ \varphi^{-1}\circ \varphi =\theta for all θ∈[X;Z]\theta \in [X; Z]. It follows that the mappings χ↦χ∘φ\chi \mapsto \chi \circ \varphi of [Y; Z] into [X; Z] and θ↦θ∘φ−1\theta \mapsto \theta \circ \varphi^{-1} of [X; Z] into [Y; Z] are bijections inverse to one another. Analogously, the mapping χ↦φ∘χ\chi \mapsto \varphi \circ \chi of [Z; X] into [Z; Y] is bijective and its inverse bijection is the mapping θ↦φ−1∘θ\theta \mapsto \varphi^{-1}\circ \theta of [Z; Y] into [Z; X].

Let X, Y, Z be topological spaces, φ∈[X;Y]\varphi \in [X; Y] and ψ∈[Y;Z]\psi \in [Y; Z] invertible homotopy classes. Then the class ψ∘φ\psi \circ \varphi is invertible, with inverse φ−1∘ψ−1\varphi^{-1}\circ \psi^{-1}. Indeed, we have

(ψ∘φ)∘(φ−1∘ψ−1)=ψ∘(φ∘φ−1)∘ψ−1 (\psi \circ \varphi )\circ (\varphi^{-1}\circ \psi^{-1}) =\psi \circ (\varphi \circ \varphi^{-1})\circ \psi^{-1}

=ψ∘=\psi \circ [IdY]∘ψ−1_Y]\circ \psi^{-1}

=ψ∘ψ−1==\psi \circ \psi^{-1}= [IdZ]_Z]

and, analogously, (φ−1∘ψ−1)∘(ψ∘φ)=(\varphi^{-1}\circ \psi^{-1})\circ (\psi \circ \varphi ) = [IdX]_X].

Let X, Y be topological spaces and let f:X→Yf: X\rightarrow Y be a homotopy equivalence. Any continuous mapping g:Y→Xg: Y\rightarrow X whose class is the inverse of that of ff is called a reciprocal (or inverse) of ff up to homotopy. Such a mapping gg is a homotopy equivalence.

A homeomorphism ff is a homotopy equivalence, and [f]−1=[f−1][f]^{-1}= [f^{-1}]. The composite mapping of two homotopy equivalences is a homotopy equivalence. The relation "X and Y are topological spaces and there exists a homotopy equivalence of X into Y" is an equivalence relation.

Definition 3 01VA

One says that topological spaces X and Y are homotopy equivalent if there exists a homotopy equivalence of X into Y.

Example 1 01VB

The empty topological space is homotopy equivalent only to itself.

Example 2 01VC

Let X be a nonempty topological space. In order that X be homotopy equivalent to a space reduced to one point, it is necessary and sufficient that there exist a constant mapping p:X→p: X\rightarrow X which is homotopic to the identity mapping of X. (This is then so for every constant mapping q:X→Xq: X\rightarrow X, since [p]=[p]∘[q]=[p] = [p]\circ [q] = [IdX]∘[q]=[q]_X]\circ [q] = [q].) Indeed, let P be a space reduced to one point, f:P→Xf: P\rightarrow X a mapping and g:X→Pg: X\rightarrow P the unique mapping of X into P. We have g∘f=g\circ f= IdP_P; in order that ff be a homotopy equivalence, it is necessary and sufficient that f∘gf\circ g be homotopic to IdX_X. Now, f∘gf\circ g is constant, of image f(P)f(P).

Example 3 01VD

One says that a pointed topological space (X,x)(X, x) is contractible, or that the topological space X is contractible at xx, if there exists a homotopy pointed at xx connecting IdX_X to the constant mapping of X into X of image {x}\{x\}. Such a space is homotopy equivalent to a point. There nevertheless exist spaces homotopy equivalent to a point which are not contractible at any of their points, and spaces contractible at one point, but not at every point (III, p. 321, exerc. 1).

Example 4 01VE

Let E be the numerical space of nn dimensions (or, more generally, a topological vector space over R\mathbf{R}) and let X be a subset of E. One says that the set X is star-shaped at a point xx of X if, for all y∈Xy\in X and all t∈It\in \mathbf{I}, the point tx+(1−t)ytx+ (1-t)y belongs to X. A convex subset (I, p. 122) of E is star-shaped at each of its points.

A topological subspace X of E which is star-shaped at one of its points xx is contractible at this point. Indeed, the mapping σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X defined by σ(y,t)=tx+(1−t)y\sigma (y, t) =tx+ (1-t)y is a pointed homotopy at xx joining IdX_X to the constant mapping with image {x}\{x\}.

In particular, 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 contractible at each of its points.

Example 5 01VF

We shall prove later (TA, V) that the Euclidean spheres Sn,n⩾1\mathbf{S}_n,n\geqslant 1, are not spaces homotopic to a point. The unit sphere of a Hilbert space of countable type and infinite dimension is contractible at each of its points (EVT, V, p. 71, exerc. 13).

4. Relative homotopies

Let X, Y be topological spaces, let A be a subspace of X and let B be a subspace of Y.

Let f:X→Yf: X\rightarrow Y be a continuous mapping such that f(A)⊂Bf(A)\subset B. We say that ff is a homotopy of the pair (X,A)(X,A) onto the pair (Y,B)(Y,B) if there exists a continuous mapping g:Y→Xg: Y\rightarrow X such that g(B)⊂Ag(B)\subset A, a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X, fixed on A, joining IdX_X to g∘fg\circ f, and a homotopy τ:Y×I→Y\tau : Y\times \mathbf{I}\rightarrow Y, fixed on B, joining IdY_Y to f∘gf\circ g.

Suppose these conditions are satisfied. Then:

a) The mapping ff is a homotopy of X into Y, the mapping gg is a homotopy of Y into X, and these homotopies are inverses of one another up to homotopy.

b) The mapping gg is then a homotopy of the pair (Y,B)(Y,B) onto the pair (X,A)(X,A), called the inverse of ff up to homotopy.

c) The mappings ff and gg define, by passing to the subspaces, homeomorphisms of A onto B which are inverse to one another.

We say that the pairs (X,A)(X,A) and (Y,B)(Y,B) are homotopic if there exists a homotopy of the pair (X,A)(X,A) onto the pair (Y,B)(Y,B). The relation “X, Y are topological spaces, A is a subspace of X, B is a subspace of Y and the pairs (X,A)(X,A) and (Y,B)(Y,B) are homotopic” is an equivalence relation.

5. Retractions and contractions

Definition 4 01VG

Let X be a topological space and let A be a subset of X.

A retraction of X onto A is called a continuous mapping from X into A which is a retraction of the canonical injection of A into X. If there exists a retraction of X onto A, we say that X can be retracted onto A, or also that A is a retract of X.

Let X be a topological space and let A be a subspace of X.

Suppose that there exists a retraction rr of X onto A. The subspace A is identified with the set of points x∈Xx\in X such that x=r(x)x=r(x). If X is a separated space, A is then closed in X.

According to the following lemma, for a subspace A to be a retract of X, it is necessary and sufficient that every continuous mapping from A into a topological space Y extend to a continuous mapping from X into Y.

Lemma 1 01VH

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. The following conditions are equivalent:

(i) For every topological space Z and every continuous mapping g:X→Zg: X\rightarrow Z, there exists a continuous mapping g′:Y→Zg': Y\rightarrow Z such that g=g= g′∘fg'\circ f;

(ii) The mapping ff is injective and possesses a continuous retraction ;

(iii) The mapping ff defines a homeomorphism of X onto its image f(X)f(X), which is a retract of Y.

Suppose that ff is injective and that r:Y→r: Y\rightarrow X is a continuous retraction of the mapping ff. For every continuous mapping g:X→Zg: X\rightarrow Z, the mapping g′=g∘r:Y→Zg'=g\circ r: Y\rightarrow Z satisfies g′∘f=g∘r∘f=gg'\circ f=g\circ r\circ f=g. This proves that (ii)⇒\Rightarrow(i).

Suppose that condition (i) is satisfied. Let g:X→Xg: X\rightarrow X be the identity mapping. By hypothesis, there exists a continuous mapping g′:Y→Xg': Y\rightarrow X such that g′∘f=gg'\circ f=g. This implies that the mapping ff is injective and that g′g' is a continuous retraction of ff.

The equivalence of the properties (ii) and (iii) is immediate.

Definition 5 01VI

Let X be a topological space and let A be a subset of X. A contraction of X onto A is a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X fixed on A whose origin is the identity mapping of X and whose term is a retraction of X onto A. If there exists a contraction of X onto A, one says that X can be contracted onto A, or again that A is a retractile subset of X.

Remark 01VJ

In other words, a contraction of X onto A is a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X having the following properties:

(i) σ(x,0)=x\sigma (x,0) =x for every x∈Xx\in X;

(ii) σ(x,1)∈A\sigma (x,1)\in A for every x∈Xx\in X;

(iii) σ(x,t)=x\sigma (x, t) =x for every x∈Ax\in A and every t∈It\in \mathbf{I}.

For there to exist a contraction of X onto A, it is necessary and sufficient that the canonical injection of A into X be a homotopy of the pair (A,A)(A,A) onto the pair (X,A)(X,A).

Let aa be a point of X. A contraction of X onto the subspace {a}\{a\} is nothing other than a homotopy pointed at aa joining the mapping IdX_X to the constant mapping with image {a}\{a\}. Consequently, X contracts onto {a}\{a\} if and only if X is contractible at aa (Example 3 of III, p. 234).

Let X be a topological space and let A be a subset of X. Let σ\sigma be a contraction of X onto A. The mapping rr of X into A defined by r(x)=σ(x,1)r(x) =\sigma (x,1) is a retraction of X onto A and σ\sigma is a homotopy joining IdX_X to rr. The relations r∘i=r\circ i= IdA_A and i∘r=ri\circ r=r then imply that the mappings ii and rr are homotopies, inverse to one another up to homotopy.

Definition 6 01VK

With the preceding notations, one says that σ\sigma is a strong contraction if, moreover, one has r(σ(x,t))=r(x)r(\sigma (x, t)) =r(x) for every x∈Xx\in X and every t∈It\in \mathbf{I},

Example 01VL

Let X be the complement of the origin in Bn\mathbf{B}_n. The mapping of X×IX\times \mathbf{I} into X given by (x,t)↦((1−t)+t1∥x∥)x(x, t)\mapsto ((1-t) +t\frac{1}{\|x\|})x is a strong contraction of X onto Sn−1\mathbf{S}_{n-1}. The retraction of X onto Sn−1\mathbf{S}_{n-1} associated with it is the mapping given by x↦x/∥x∥x\mapsto x/\|x\|.

Lemma 2 01VM

Let X be a topological space, let U be an open subset of X and let σ\sigma be a strong contraction of X onto X-U. Then σ(U×I)⊂U‾\sigma (U\times \mathbf{I})\subset \overline{U}. In particular, σ(U×{1})\sigma (U\times \{1\}) is contained in the boundary of U.

Let x∈Ux\in U. The set of real numbers t∈It\in \mathbf{I} such that σ(x,t)∈U\sigma (x, t)\in U is open in I\mathbf{I} and contains 0; let ss be its least upper bound. We have σ(x,s)∈U‾\sigma (x, s)\in \overline{U}. If s<1,σ(x,s)∉Us <1,\sigma (x, s)\notin U, by definition of ss; the same is true if s=1s= 1 since σ(x,1)∈X−U\sigma (x,1)\in X-U. Consequently, σ(x,s)∈\sigma (x, s)\in Fr(U). By definition of a strong contraction, we then have σ(x,s)=σ(σ(x,s),1)=σ(x,1)\sigma (x, s) =\sigma (\sigma (x, s),1) =\sigma (x,1); in particular, σ(x,1)∈U‾\sigma (x,1)\in \overline{U}. Consequently, σ(x,1)\sigma (x,1) belongs to the boundary U‾∩(X−U)\overline{U}\cap (X-U) of U.

Soit t∈It\in \mathbf{I}; si σ(x,t)∉U\sigma (x, t)\notin U, il vient encore σ(x,t)=σ(σ(x,t),1)=\sigma (x, t) =\sigma (\sigma (x, t),1) = σ(x,1)\sigma (x,1), donc σ(x,t)∈U‾\sigma (x, t)\in \overline{U}, d’où le lemme.

6. Cylinder of a mapping

Let X and Y be topological spaces and let ff be a continuous mapping of X into Y. Denote by U the topological space sum of the topological space U1=X×IU_1= X\times \mathbf{I} and of the space U2U_2 = Y and identify U1U_1 and U2U_2 with subspaces of U by the canonical injections. Let R be the equivalence relation on U the finest for which the points (x,1)(x,1) of U1U_1 and f(x)f(x) of U2U_2 are equivalent, for every x∈Xx\in X.

Definition 7 01VN

The cylinder of the mapping ff is called and is denoted by Cyl(ff), the topological space quotient U/RU/R.

Denote by αf:X×I→\alpha_f: X\times \mathbf{I}\rightarrow Cyl(ff) and βf:Y→\beta_f: Y\rightarrow Cyl(ff) the restrictions to U1U_1 and U2U_2 of the canonical surjection of U onto U/RU/R. The mapping αf\alpha_f is a homotopy and its term is βf∘f\beta_f\circ f.

Proposition 4 (Universal Property of Cylinders)

Let Z be a topological space, let β:Y→Z\beta : Y\rightarrow Z be a continuous mapping and let α:X×I→Z\alpha : X\times \mathbf{I}\rightarrow Z be a homotopy whose term is β∘f\beta \circ f. There exists a unique continuous mapping φ\varphi from Cyl(ff) into Z such that α=φ∘αf\alpha =\varphi \circ \alpha_f and β=φ∘βf\beta =\varphi \circ \beta_f.

The mapping ψ\psi from U into Z which coincides with α\alpha in X×IX\times \mathbf{I} and with β\beta in Y is continuous. We have α(x,1)=β(f(x))\alpha (x,1) =\beta (f(x)) for every x∈Xx\in X, hence ψ\psi defines by passing to the quotient a continuous mapping φ\varphi from Cyl(ff) into Z such that α=φ∘αf\alpha =\varphi \circ \alpha_f and β=φ∘βf\beta =\varphi \circ \beta_f. Since the images of αf\alpha_f and βf\beta_f cover Cyl(ff),φ,\varphi is the only mapping from Cyl(ff) into Z satisfying these relations.

Example 1 01VO

Let X′,Y′X',Y' be topological spaces and let f′:X′→Y′f': X'\rightarrow Y' be a continuous mapping. Let u:X→X′u: X\rightarrow X' and v:Y→Y′v: Y\rightarrow Y' be continuous mappings such that f′∘u=v∘ff'\circ u=v\circ f. There exists a unique continuous mapping w:w: Cyl(ff)→\rightarrow Cyl(f′f') such that w∘αf=αf′∘(u×w\circ \alpha_f=\alpha_{f'}\circ (u\times IdI)_{\mathbf{I}}) and w∘βf=βf′∘vw\circ \beta_f=\beta_{f'}\circ v.

According to prop. 4, applied to Z = Y and β=\beta = IdY_Y, there exists a unique continuous mapping γf:\gamma_f: Cyl(ff)→Y\rightarrow Y such that γf(αf(x,s))=f(x)\gamma_f(\alpha_f(x, s)) =f(x) and γf(βf(y))=y\gamma_f(\beta_f(y)) =y for x∈X,s∈Ix\in X,s\in \mathbf{I} and y∈Yy\in Y.

Proposition 5 01VP

The mapping αf\alpha_f induces a homeomorphism of X×[0,1[X\times [0,1[ onto an open subset of Cyl(ff). The mapping βf\beta_f defines a homeomorphism of Y onto the complement of this open subset. The mapping βf∘γf\beta_f\circ \gamma_f is a continuous retraction of Cyl(ff) onto βf(Y)\beta_f(Y).

The set X×[0,1[X\times [0,1[ is an open subset of U, saturated for the relation R, and the equivalence relation induced by R in X×[0,1[X\times [0,1[ is the equality relation. The first assertion follows from this (TG, I, p. 23, cor. 10).

The complement of αf(X×[0,1[)\alpha_f(X\times [0,1[) is βf(Y)\beta_f(Y). Since one has γf∘βf=\gamma_f\circ \beta_f= IdY,βf_Y,\beta_f defines a homeomorphism of Y onto βf(Y)\beta_f(Y) and βf∘γf\beta_f\circ \gamma_f is a continuous retraction of the canonical injection of βf(Y)\beta_f(Y) into Cyl(ff).

The closed subspace βf(Y)\beta_f(Y) of Cyl(ff) is called the basis of the cylinder of ff. The mapping βf∘γf\beta_f\circ \gamma_f is called the canonical retraction of Cyl(ff) onto its basis.

Consider the mappings σ1:U1×I→\sigma_1: U_1\times \mathbf{I}\rightarrow Cyl(ff) and σ2:U2×I→\sigma_2: U_2\times \mathbf{I}\rightarrow Cyl(ff) defined by

σ1((x,s),t)=αf(x,(1−t)s+t)\sigma_1((x, s), t) =\alpha_f(x,(1-t)s+t) for (x,s)∈X×I(x, s)\in X\times \mathbf{I} and t∈It\in \mathbf{I}

σ2(y,t)=βf(y)\sigma_2(y, t) =\beta_f(y) for y∈Yy\in Y and t∈It\in \mathbf{I}.

They are continuous. For x∈Xx\in X and t∈It\in \mathbf{I}, one has

σ1((x,1),t)=αf(x,1)=βf(f(x))=σ2(f(x),t) \sigma_1((x,1), t) =\alpha_f(x,1) =\beta_f(f(x)) =\sigma_2(f(x), t)

There therefore exists a unique mapping σf\sigma_f of Cyl(ff)×I\times \mathbf{I} into Cyl(ff) such that σf∘(αf×\sigma_f\circ (\alpha_f\times IdI)=σ1_{\mathbf{I}}) =\sigma_1 and σf∘(βf×\sigma_f\circ (\beta_f\times IdI)=σ2_{\mathbf{I}}) =\sigma_2. The mapping σf\sigma_f is continuous (I, p. 19, prop. 10).

Proposition 6 01VQ

The mapping σf:\sigma_f: Cyl(ff)×I→\times \mathbf{I}\rightarrow Cyl(ff) is a strong contraction of Cyl(ff) onto its base. Its term is the canonical retraction of Cyl(ff) onto its base.

The mapping σf\sigma_f is a homotopy. The relations σf(αf(x,s),0)=\sigma_f(\alpha_f(x, s),0) = αf(x,s)\alpha_f(x, s) and σf(βf(y),0)=βf(y)\sigma_f(\beta_f(y),0) =\beta_f(y) imply that the origin of σf\sigma_f is the identity mapping of Cyl(ff). Let rfr_f denote the term of σf\sigma_f. The relations

σf(αf(x,s),1)=αf(x,1)=βf(f(x))=(βf∘γf)(αf(x,s)) \sigma_f(\alpha_f(x, s),1) =\alpha_f(x,1) =\beta_f(f(x)) = (\beta_f\circ \gamma_f)(\alpha_f(x, s))

and σf(βf(y),1)=βf(y)=(βf∘γf)(βf(y))\sigma_f(\beta_f(y),1) =\beta_f(y) = (\beta_f\circ \gamma_f)(\beta_f(y)) imply that rfr_f is the canonical retraction βf∘γf\beta_f\circ \gamma_f of Cyl(ff) onto its base.

For (x,s)∈X×I(x, s)\in X\times \mathbf{I} and t∈It\in \mathbf{I}, we have

rf(σf(αf(x,s),t))=rf(αf(x,s(1−t)+t))=βf(f(x))=rf(αf(x,s)) r_f(\sigma_f(\alpha_f(x, s), t)) =r_f(\alpha_f(x, s(1-t) +t)) =\beta_f(f(x)) =r_f(\alpha_f(x, s))

For y∈Yy\in Y and t∈It\in \mathbf{I}, we have rf(σf(βf(y),t))=rf(βf(y))r_f(\sigma_f(\beta_f(y), t)) =r_f(\beta_f(y)). Consequently, σf\sigma_f is a strong contraction of Cyl(ff) onto its base.

The mapping σf\sigma_f is called the canonical contraction of the cylinder of ff onto its base.

Remark 1 01VR

Let A be a subset of X×IX\times \mathbf{I} and let A1A_1 be the set of points x∈Xx\in X such that (x,1)∈A(x,1)\in A. We have αf−1(αf(A))=A∪f−1(f(A1))×{1}\alpha^{-1}_f(\alpha_f(A)) = A\cup \overset{-1}{f}(f(A_1))\times \{1\} and βf−1(αf(A))=f(A1)\beta^{-1}_f(\alpha_f(A)) =f(A_1). It follows that the mapping αf\alpha_f is closed (resp. open) if ff is.

Remark 2 01VS

Suppose that the mapping ff is proper. Let P be a point of Cyl(ff). If P = αf(x,t)\alpha_f(x, t), with 0 ⩽t<\leqslant t < 1 and x∈x\in X, we have αf−1(P)=(x,t)\alpha^{-1}_f(P) = (x, t). In the contrary case, there exists y∈Yy\in Y such that P=βf(y)P =\beta_f(y) and αf−1(P)=f−1(y)×{1}\alpha^{-1}_f(P) =\overset{-1}{f}(y)\times \{1\}. This proves that the fibres of the mapping αf\alpha_f are quasi-compact. The mapping αf\alpha_f is then proper (TG, I, p. 75, th. 1), since it is closed.

By prop. 5 and TG, I, p. 72, prop. 2, the mapping βf\beta_f is itself proper. It follows that, if ff is proper, the canonical surjection of U onto Cyl(ff) is proper and hence, in particular, universally strict (I, p. 20, corollary).

Remark 3 01VT

If the spaces X and Y are separated, the same is true of the mapping cylinder of the mapping ff. Indeed, let zz and z′z' be two distinct points of Cyl(ff) and let us prove that they possess disjoint neighbourhoods. Distinguish three cases.

– There exist (x,t)(x, t) and (x′,t′)∈X×[0,1[(x', t')\in X\times [0,1[ such that z=αf(x,t)z=\alpha_f(x, t) and z′=αf(x′,t′)z'=\alpha_f(x', t').

In this case, the assertion follows from the fact that the space X×[0,1[X\times [0,1[ is separated (TG, I, p. 54, prop. 7) and that the mapping αf\alpha_f induces a homeomorphism of this space onto an open subspace of Cyl(ff).

– There exists (x,t)∈X×[0,1[(x, t)\in X\times [0,1[ such that z=αf(x,t)z=\alpha_f(x, t) and y′∈Yy'\in Y such that z′=βf(y′)z'=\beta_f(y').

Then, αf(X×[0,t+12[)\alpha_f(X\times [0,\frac{t+1}{2}[) and Cyl(ff)−αf(X×[0,t+12])-\alpha_f(X\times [0,\frac{t+1}{2}]) are disjoint open neighbourhoods of zz and z′z' in Cyl(ff).

– There exist yy and y′∈Yy'\in Y such that z=βf(y),z′=βf(y′)z=\beta_f(y),z'=\beta_f(y').

In this case, y≠y′y=\not y'; since Y is separated, there exists an open neighbourhood V of yy in Y and an open neighbourhood V′V' of y′y' in Y such that V∩V′=∅V\cap V'=\emptyset. Then, (βf∘γf)−1(V)(\beta_f\circ \gamma_f)^{-1}(V) and (βf∘γf)−1(V′)(\beta_f\circ \gamma_f)^{-1}(V') are disjoint open subsets of Cyl(ff) containing respectively yy and y′y'.

7. The homotopy extension property

Definition 8 01VU

Let X be a topological space and let A be a subset of X. We say that the pair (X,A)(X,A) possesses the homotopy extension property if, for every topological space Y, every continuous mapping f:X→Yf: X\rightarrow Y and every homotopy σ:A×I→Y\sigma : A\times \mathbf{I}\rightarrow Y whose term is the mapping f∣Af|A, there exists a homotopy τ:X×I→Y\tau : X\times \mathbf{I}\rightarrow Y extending σ\sigma and whose term is the mapping ff.

Remark 1 01VV

Let X be a topological space and let A be a subset of X such that the pair (X,A)(X,A) possesses the homotopy extension property. Let Y be a topological space, let f:X→Yf: X\rightarrow Y be a continuous mapping and let σ:A×I→Y\sigma : A\times \mathbf{I}\rightarrow Y be a homotopy whose origin is the mapping f∣Af|A. The mapping σ‾:A×I→Y\overline{\sigma}: A\times \mathbf{I}\rightarrow Y defined by (a,t)↦σ(a,1−t)(a, t)\mapsto \sigma (a,1-t) is a homotopy with term f∣Af|A; let τ:X×I→Y\tau : X\times \mathbf{I}\rightarrow Y be a homotopy with term ff which extends σ‾\overline{\sigma}. The mapping τ‾:X×I→Y\overline{\tau}: X\times \mathbf{I}\rightarrow Y given by (x,t)↦τ(x,1−t)(x, t)\mapsto \tau (x,1-t) is then a homotopy which extends σ\sigma and whose origin is the mapping ff.

Let X be a topological space, let A be a subspace of X and let i:A→Xi: A\rightarrow X be the canonical injection. Let αi:A×I→\alpha_i: A\times \mathbf{I}\rightarrow Cyl(ii) and βi:X→\beta_i: X\rightarrow Cyl(ii) denote the canonical mappings. Let j:j: Cyl(ii)→X×I\rightarrow X\times \mathbf{I} be the unique continuous mapping such that j(αi(a,s))=(i(a),s)j(\alpha_i(a, s)) = (i(a), s) and j(βi(x))=(x,1)j(\beta_i(x)) = (x,1) for a∈A,s∈Ia\in A,s\in \mathbf{I} and x∈Xx\in X. It is injective; its image is the subspace (A×I)∪(X×{1})(A\times \mathbf{I})\cup (X\times \{1\}) of X×IX\times \mathbf{I}. The mapping jj is closed if A is closed in X. It is not always strict (III, p. 325, exerc. 17).

Proposition 7 01VW

With the above notations, the following assertions are equivalent:

(i) The pair (X,A)(X,A) has the homotopy extension property;

(ii) For every topological space Y and every continuous mapping g:g: Cyl(ii)→Y\rightarrow Y, there exists a continuous mapping g′:X×I→Yg': X\times \mathbf{I}\rightarrow Y such that g=g′∘jg=g'\circ j;

(iii) The injection jj has a continuous retraction;

(iv) The mapping jj is strict and there exists a contraction of X×IX\times \mathbf{I} onto the image of jj.

Suppose that the pair (X,A)(X,A) has the homotopy extension property. Let g:g: Cyl(ii)→Y\rightarrow Y be a continuous mapping; put σ=g∘αi\sigma =g\circ \alpha_i and f=g∘βif=g\circ \beta_i. The mapping f:X→f: X\rightarrow Y is continuous and σ:A×I→Y\sigma : A\times \mathbf{I}\rightarrow Y is a homotopy whose end is the mapping f∣Af|A. There therefore exists a homotopy g′:X×I→Yg': X\times \mathbf{I}\rightarrow Y with end ff extending σ\sigma. We have g′∘j(αi(a,s))=g′(a,s)=σ(a,s)=g(αi(a,s))g'\circ j(\alpha_i(a, s)) =g'(a, s) =\sigma (a, s) =g(\alpha_i(a, s)) and g′∘j(βi(x))=g'\circ j(\beta_i(x)) = g′(x,1)=f(x)=g(βi(x))g'(x,1) =f(x) =g(\beta_i(x)) for a∈A,s∈Ia\in A,s\in \mathbf{I} and x∈Xx\in X. Consequently, g′∘j=gg'\circ j=g, whence (ii).

Conversely, suppose that assertion (ii) is satisfied, and let us prove that the pair (X,A)(X,A) has the homotopy extension property. Let Y be a topological space, let f:X→Yf: X\rightarrow Y be a continuous mapping and let σ:A×I→Y\sigma : A\times \mathbf{I}\rightarrow Y be a homotopy whose end is equal to f∣Af|A. There exists a unique continuous mapping g:g: Cyl(ii)→Y\rightarrow Y such that g∘αi=σg\circ \alpha_i=\sigma and g∘βi=fg\circ \beta_i=f (III, p. 238, prop. 4). Any mapping g′:X×I→Yg': X\times \mathbf{I}\rightarrow Y such that g=g′∘jg=g'\circ j is then a homotopy with end ff extending σ\sigma.

The equivalence of assertions (ii) and (iii) is a particular case of lemma 1 of III, p. 235.

Suppose that the injection j:j: Cyl(ii)→X×I\rightarrow X\times \mathbf{I} admits a continuous retraction rr. The mapping jj therefore defines a homeomorphism of Cyl(ii) onto the subspace T=(A×I)∪(X×{1})T = (A\times \mathbf{I})\cup (X\times \{1\}) of X×I(X\times \mathbf{I}(loc. cit.). It is therefore strict.

Put ρ=\rho = pr1∘j∘r_1\circ j\circ r and θ=\theta = pr2∘j∘r_2\circ j\circ r. If x∈Ax\in A or if s=1s= 1, we have ρ(x,s)=x\rho (x, s) =x and θ(x,s)=s\theta (x, s) =s For (x,s)∈X×I(x, s)\in X\times \mathbf{I} and t∈It\in \mathbf{I}, put

σ((x,s),t)=(ρ(x,(1−t)+st),(1−t)s+tθ(x,s)) \sigma ((x, s), t) = (\rho (x,(1-t) +st),(1-t)s+t\theta (x, s))

The mapping σ:(X×I)×I→X×I\sigma : (X\times \mathbf{I})\times \mathbf{I}\rightarrow X\times \mathbf{I} is continuous. For (x,s)∈X×I(x, s)\in X\times \mathbf{I}, we have

σ((x,s),0)=(ρ(x,1),s)=(x,s) \sigma ((x, s),0) = (\rho (x,1), s) = (x, s)

and

σ((x,s),1)=(ρ(x,s),θ(x,s))=j∘r(x,s) \sigma ((x, s),1) = (\rho (x, s), \theta (x, s)) =j\circ r(x, s)

for x∈Xx\in X and t∈It\in \mathbf{I}, we have

σ((x,1),t)=(ρ(x,1),(1−t)+tθ(x,1))=(x,1) \sigma ((x,1), t) = (\rho (x,1),(1-t) +t\theta (x,1)) = (x,1)

whereas, for (x,s)∈A×I(x, s)\in A\times \mathbf{I} and t∈It\in \mathbf{I}, we have

σ((x,s),t)=(x,(1−t)s+ts)=(x,s) \sigma ((x, s), t) = (x,(1-t)s+ts) = (x, s)

Hence σ\sigma is a contraction of X×IX\times \mathbf{I} onto T. This shows that assertion (iii) implies assertion (iv).

The implication (iv)⇒\Rightarrow(iii) is obvious.

Remark 2 01VX

Let X be a separated topological space and let A be a subset of X such that the pair (X,A)(X,A) has the homotopy extension property. Let us denote by i:A→i: A\rightarrow X the canonical injection. Let j:j: Cyl(ii)→X×I\rightarrow X\times \mathbf{I} be the canonical injection and let rr be a continuous retraction of the mapping jj. The subspace jj(Cyl(ii)) is equal to the set of pairs (x,t)∈X×I(x, t)\in X\times \mathbf{I} such that j(r(x,t))=(x,t)j(r(x, t)) = (x, t). Since the space X×IX\times \mathbf{I} is separated, the subset jj(Cyl(ii)) is closed in X×IX\times \mathbf{I}. The set A, equal to the set of x∈Xx\in X such that (x,0)∈j(x,0)\in j(Cyl(ii)), is then a closed subset of X.

Lemma 3 01VY

Let X and Y be topological spaces, let p:X→Yp: X\rightarrow Y be a continuous proper and open mapping, and let f:X→Rf: X\rightarrow \mathbf{R} be a continuous mapping. The mapping g:Y→R‾g: Y\rightarrow \overline{\mathbf{R}} given by y↦y\mapsto supx∈p−1(y)f(x)_{x\in\overset{-1}{p}(y)}f(x) is continuous.

Let b∈Yb\in Y. Since pp is proper, its fibre p−1(b)\overset{-1}{p}(b) is a quasi-compact space; hence we have g(b)∈R∪{−∞}g(b)\in \mathbf{R}\cup \{-\infty \}.

Let m∈Rm\in \mathbf{R} such that g(b)<mg(b)< m. For every a∈p−1(b)a\in \overset{-1}{p}(b), one has f(a)⩽f(a)\leqslant g(b)<mg(b)< m; let VaV_a be a neighbourhood of aa in X such that f(x)<mf(x)< m for every x∈Vax\in V_a. The union V of the sets VaV_a is a neighbourhood of p−1(b)\overset{-1}{p}(b) in X. By Lemma 5 (I, p. 75), there exists a neighbourhood W of bb in Y such that p−1(W)⊂V\overset{-1}{p}(W)\subset V. For every y∈Wy\in W, one has g(y)⩽mg(y)\leqslant m. This proves that gg is upper semi-continuous at bb.

Let us now prove that gg is lower semi-continuous at bb. We may suppose that g(b)∈Rg(b)\in \mathbf{R}. Let m∈Rm\in \mathbf{R} such that m<g(b)m < g(b). Let a∈p−1(b)a\in \overset{-1}{p}(b) such that m<f(b)m < f(b); let then V be a neighbourhood of aa in X such that f(x)>mf(x)> m for every x∈Vx\in V. It follows that g(y)>mg(y)> m for every y∈p(V)y\in p(V). Since pp is open, p(V)p(V) is a neighbourhood of bb, so that gg is lower semi-continuous.

The lemma is thus proved.

Theorem 1 01VZ

Let X be a topological space and let A be a closed subspace of X; denote by i:A→Xi: A\rightarrow X the canonical injection. The following assertions are equivalent:

(i) The pair (X,A)(X,A) has the homotopy extension property ;

(ii) There exists a continuous mapping φ:X→I\varphi : X\rightarrow \mathbf{I} such that A=φ−1(0)A =\overset{-1}{\varphi}(0) and a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X fixed on A whose term is the identity mapping of X and such that σ(x,0)∈A\sigma (x,0)\in A for every point x∈Xx\in X such that φ(x)=1̸\varphi (x)= 1\not .

(iii) There exists a continuous mapping φ:X→R+\varphi : X\rightarrow \mathbf{R}_+ such that A=φ−1(0)A =\overset{-1}{\varphi}(0) and a homotopy σ:φ−1(I)×I→X\sigma :\overset{-1}{\varphi}(\mathbf{I})\times \mathbf{I}\rightarrow X, fixed on A, such that σ(x,1)=x\sigma (x,1) =x and σ(x,0)∈A\sigma (x,0)\in A for every x∈φ−1(I)x\in \overset{-1}{\varphi}(\mathbf{I}).

(iv) There exists a continuous mapping φ:X→R+\varphi : X\rightarrow \mathbf{R}_+ such that A⊂φ−1(0)A\subset \overset{-1}{\varphi}(0) and a continuous mapping

σ:{(x,t)∈X×I∣t+φ(x)⩾1}→X \sigma :\{(x, t)\in X\times \mathbf{I}|t+\varphi (x)\geqslant 1\} \rightarrow X

such that σ(x,1)=x\sigma (x,1) =x for every x∈Xx\in X and σ(x,1−φ(x))∈A\sigma (x,1-\varphi (x))\in A for every x∈Xx\in X such that φ(x)⩽1\varphi (x)\leqslant 1.

We shall denote by ii the canonical injection of A into X,αi:A×I→,\alpha_i: A\times \mathbf{I}\rightarrow Cyl(ii) and βi:X→\beta_i: X\rightarrow Cyl(ii) the canonical mappings, and by jj the mapping of Cyl(ii) into X×IX\times \mathbf{I} such that j(αi(x,s))=(x,s)j(\alpha_i(x, s)) = (x, s) for (x,s)∈A×I(x, s)\in A\times \mathbf{I} and j(βi(x))=(x,1)j(\beta_i(x)) = (x,1) for x∈Xx\in X.

Suppose that the pair (X,A)(X,A) has the homotopy extension property and let r:X×I→r: X\times \mathbf{I}\rightarrow Cyl(ii) be a continuous retraction of the mapping j(j(III, p. 241, Proposition 7). Let us denote by σ\sigma the continuous mapping pr1∘j∘r_1\circ j\circ r of X×IX\times \mathbf{I} into X. For every (x,t)∈X×I(x, t)\in X\times \mathbf{I}, one has ∣|pr2(j(r(x,t)))−t∣⩽1_2(j(r(x, t)))-t|\leqslant 1. Since I\mathbf{I} is compact, the first projection pr1:X×I→X_1: X\times \mathbf{I}\rightarrow X is a proper mapping; it is also open. By Lemma 3, the mapping φ:X→I\varphi : X\rightarrow \mathbf{I} given by

x↦x\mapsto supt∈I(∣_{t\in\mathbf{I}}(|pr2(j(r(x,t)))−t∣)_2(j(r(x, t)))-t|)

is therefore continuous.

For x∈Xx\in X, one has σ(x,1)=x\sigma (x,1) =x. Let x∈Xx\in X be such that σ(x,0)∉A\sigma (x,0)\notin A; then pr2(j(r(x,0)))=1_2(j(r(x,0))) = 1 and φ(x)⩾1\varphi (x)\geqslant 1. For x∈Ax\in A and t∈It\in \mathbf{I}, one has j(r(x,t))=(x,t)j(r(x, t)) = (x, t); consequently, σ(x,t)=x\sigma (x, t) =x and φ(x)=0\varphi (x) = 0. Conversely, let x∈Xx\in X be such that φ(x)=0\varphi (x) = 0. One has therefore pr2(j(r(x,t))⩽t_2(j(r(x, t))\leqslant t for every t∈It\in \mathbf{I}; if t<1t <1, this implies j(r(x,t))∈A×Ij(r(x, t))\in A\times \mathbf{I}; since A is closed in X, one has therefore j(r(x,1))∈A×Ij(r(x,1))\in A\times \mathbf{I}, hence x∈Ax\in A.

This proves that (i) implies (ii).

Let φ\varphi and σ\sigma be mappings satisfying the properties of assertion (ii). Put φ1=2φ\varphi_1= 2\varphi and let σ1\sigma_1 be the restriction of σ\sigma to φ−11(I)×I\overset{-1}{\varphi}_{1}(\mathbf{I})\times \mathbf{I}. We have φ−11(0)=A\overset{-1}{\varphi}_{1}(0) = A; for a∈A,σ1(a,t)=aa\in A,\sigma_1(a, t) =a for all a∈Aa\in A. Let x∈Xx\in X be such that φ1(x)⩽1\varphi_1(x)\leqslant 1; we have σ1(x,1)=x\sigma_1(x,1) =x; moreover, φ(x)=φ1(x)/2<1\varphi (x) =\varphi_1(x)/2<1, hence σ1(x,0)∈A\sigma_1(x,0)\in A. Thus, (ii) implies (iii).

Let us prove that (iii) implies (iv). Let φ:X→R\varphi : X\rightarrow \mathbf{R} and σ:φ−1(I)×\sigma :\overset{-1}{\varphi}(\mathbf{I})\times I→X\mathbf{I}\rightarrow X be as in the statement; put B=φ−1(I)B =\overset{-1}{\varphi}(\mathbf{I}) and C=φ−1([1,+∞[)C =\overset{-1}{\varphi}([1,+\infty [).

Let u1u_1 be the mapping from B×IB\times \mathbf{I} into X such that u1(x,t)=σ(x,1−u_1(x, t) =\sigma (x,1- (1−t)/2φ(x))(1-t)/2\varphi (x)) if t+2φ(x)⩾1t+ 2\varphi (x)\geqslant 1 and φ(x)>0\varphi (x)>0 and u1(x,t)=σ(x,0)u_1(x, t) =\sigma (x,0) otherwise. By the following lemma 4, it is continuous.

Let u2u_2 be the mapping from B×IB\times \mathbf{I} into X such that u2(x,t)=u_2(x, t) = σ(x\sigma (x, sup(0,1−2(1−t)(1−φ(x)),1-2(1-t)(1-\varphi (x)))). It is continuous. For x∈Bx\in B such that φ(x)=1/2\varphi (x) = 1/2 and t∈It\in \mathbf{I}, we have u1(x,t)=σ(x,t)=u2(x,t)u_1(x, t) =\sigma (x, t) =u_2(x, t). Let u:B×I→Xu: B\times \mathbf{I}\rightarrow X denote the mapping such that u(x,t)=u1(x,t)u(x, t) =u_1(x, t) if φ(x)⩽1/2\varphi (x)\leqslant 1/2 and u(x,t)=u2(x,t)u(x, t) =u_2(x, t) if 1/2<x⩽11/2< x\leqslant 1 ; it is continuous since its restrictions to the closed subspaces φ−1([0,1/2])×I\overset{-1}{\varphi}([0,1/2])\times \mathbf{I} and φ−1([1/2,1])×I\overset{-1}{\varphi}([1/2,1])\times \mathbf{I} of B×IB\times \mathbf{I} are continuous (TG, I, p. 19, prop. 4).

For x∈Xx\in X such that φ(x)=1\varphi (x) = 1 and t∈It\in \mathbf{I}, we have u(x,t)=u2(x,t)=u(x, t) =u_2(x, t) = σ(x,1)=x\sigma (x,1) =x. There therefore exists a unique mapping τ:X×I→X\tau : X\times \mathbf{I}\rightarrow X which coincides with uu in B×IB\times \mathbf{I} and with the mapping pr1_1 in C×IC\times \mathbf{I}; it is continuous (loc. cit.).

Let x∈Xx\in X; it is verified that one has τ(x,1)=1\tau (x,1) = 1. If, moreover, x∈Ax\in A, then one has φ(x)=0\varphi (x) = 0, hence τ(x,t)=u1(x,t)=σ(x,0)=x\tau (x, t) =u_1(x, t) =\sigma (x,0) =x. Finally, if 2φ(x)⩽12\varphi (x)\leqslant 1, then φ(x)⩽1/2\varphi (x)\leqslant 1/2, hence τ(x,1−2φ(x))=σ(x,0)∈A\tau (x,1-2\varphi (x)) =\sigma (x,0)\in A.

This proves that assertion (iv) is satisfied.

Suppose finally that assertion (iv) is satisfied, and let us prove that the pair (X,A)(X,A) has the homotopy extension property.

Let C1C_1 (resp. C2C_2) be the set of pairs (x,t)∈X×I(x, t)\in X\times \mathbf{I} such that t+φ(x)⩽1t+\varphi (x)\leqslant 1 (resp. t+φ(x)⩾t+\varphi (x)\geqslant 1). These are closed sets. For (x,t)∈C1(x, t)\in C_1, one has σ(x,1−φ(x))∈A\sigma (x,1-\varphi (x))\in A; let then ρ1:C1→\rho_1: C_1\rightarrow Cyl(ii) be the mapping defined by (x,t)↦αi(σ(x,1−φ(x)),t+φ(x))(x, t)\mapsto \alpha_i(\sigma (x,1-\varphi (x)), t+\varphi (x)); it is continuous. Let also ρ2:C2→\rho_2: C_2\rightarrow Cyl(ii) be the continuous mapping defined by (x,t)↦βi(σ(x,t))(x, t)\mapsto \beta_i(\sigma (x, t)). For (x,t)∈C1∩C2(x, t)\in C_1\cap C_2, one has t+φ(x)=1t+\varphi (x) = 1, hence

ρ1(x,t)=αi(σ(x,1−φ(x)),1)=βi(σ(x,1−φ(x))=ρ2(x,t) \rho_1(x, t) =\alpha_i(\sigma (x,1-\varphi (x)),1) =\beta_i(\sigma (x,1-\varphi (x)) =\rho_2(x, t)

There therefore exists a unique mapping ρ:X×I→\rho : X\times \mathbf{I}\rightarrow Cyl(ii) which coincides with ρ1\rho_1 in C1C_1 and with ρ2\rho_2 in C2C_2; it is continuous (TG, I, p. 19, prop. 4).

For x∈Ax\in A and t∈It\in \mathbf{I}, one has φ(x)=0\varphi (x) = 0, hence t+φ(x)⩽1t+\varphi (x)\leqslant 1

ρ(j(αi(x,t)))=ρ(x,t)=αi(σ(x,1),t)=αi(x,t) \rho (j(\alpha_i(x, t))) =\rho (x, t) =\alpha_i(\sigma (x,1), t) =\alpha_i(x, t)

For x∈Xx\in X, one also has

ρ(j(βi(x)))=ρ(x,1)=βi(σ(x,1))=βi(x) \rho (j(\beta_i(x))) =\rho (x,1) =\beta_i(\sigma (x,1)) =\beta_i(x)

Since the images of the mappings αi\alpha_i and βi\beta_i cover Cyl(ii), it follows that the mapping ρ\rho is a continuous retraction of the mapping jj. Consequently, the pair (X,A)(X,A) has the homotopy extension property (III, p. 241, prop. 7), which concludes the proof of the theorem.

Lemma 4 01W0

Let X and Y be topological spaces, let φ:X→R+\varphi : X\rightarrow \mathbf{R}_+ be a continuous mapping, and put A=φ−1(0)A =\overset{-1}{\varphi}(0). Let σ:X×I→Y\sigma : X\times \mathbf{I}\rightarrow Y be a homotopy which is fixed on A. The mapping σ′:X×I→Y\sigma ': X\times \mathbf{I}\rightarrow Y which maps (x,s)(x, s) to σ(x,s/φ(x))\sigma (x, s/\varphi (x)) if s<φ(x)s < \varphi (x) and to σ(x,1)\sigma (x,1) if s⩾φ(x)s\geqslant \varphi (x) is continuous.

The mapping σ′\sigma ' is continuous at every point of the closed subset φ−1([1,+∞[)×I\overset{-1}{\varphi}([1,+\infty [)\times \mathbf{I}; it is therefore enough to prove that its restriction to φ−1(I)×I\overset{-1}{\varphi}(\mathbf{I})\times \mathbf{I} is continuous. We may therefore suppose that φ(X)⊂I\varphi (X)\subset \mathbf{I}. Let C and C′C' be the subspaces of X×IX\times \mathbf{I} consisting of pairs (x,s)(x, s) such that s⩽φ(x)s\leqslant \varphi (x) and s⩾φ(x)s\geqslant \varphi (x) respectively. They are closed and cover X×IX\times \mathbf{I}. The mapping σ′\sigma ' is continuous on C′C'; let us prove that it is continuous on C.

Let α:X×I→X×I\alpha : X\times \mathbf{I}\rightarrow X\times \mathbf{I} be the continuous mapping defined by α(x,t)=\alpha (x, t) = (x,tφ(x))(x, t\varphi (x)). Its image is equal to C and σ′∘α\sigma '\circ \alpha is the continuous mapping σ\sigma. Let us prove that α\alpha is a proper map. Consider in fact an ultrafilter U\mathfrak{U} on X×IX\times \mathbf{I} and a point (x,t)∈X×I(x, t)\in X\times \mathbf{I} adherent to the ultrafilter basis α(U)\alpha (\mathfrak{U}). Since pr1∘α=_1\circ \alpha = pr1_1, the ultrafilter basis pr1(U)_1(\mathfrak{U}) on X converges to xx. Since I\mathbf{I} is compact, there exists a point s∈Is\in \mathbf{I} such that the ultrafilter basis pr2(U)_2(\mathfrak{U}) converges to ss. Then U\mathfrak{U} converges to (x,s)(x, s). Since α\alpha is continuous, the ultrafilter basis α(U)\alpha (\mathfrak{U}) converges to (x,sφ(x))(x, s\varphi (x)). Since I\mathbf{I} is separated, we have sφ(x)=ts\varphi (x) =t, whence α(x,s)=(x,t)\alpha (x, s) = (x, t), so that α\alpha is proper (TG, I, p. 75, Theorem 1). It then follows from I, p. 18, example 2 and Proposition 9 that σ′∣C\sigma '|C is continuous.

Corollary 1 01W1

Let X be a normal topological space and let A be a closed subspace of X. Suppose that there exist a neighbourhood V of A in X and a contraction of V onto A, together with a continuous mapping f:X→Rf: X\rightarrow \mathbf{R} such that f−1(0)=A\overset{-1}{f}(0) = A. Then the pair (X,A)(X,A) has the homotopy extension property.

Let ρ:V×I→V\rho : V\times \mathbf{I}\rightarrow V be a contraction of V onto A. By definition of a normal space (TG, IX, p. 41, Definition 1), there exists a continuous mapping g:X→Ig: X\rightarrow \mathbf{I} which is 0 on A and 1 at every point of X - V. Let φ:X→R\varphi : X\rightarrow \mathbf{R} be the mapping defined by φ(x)=∣f(x)∣+g(x)\varphi (x) =|f(x)|+g(x) for x∈Xx\in X; it is continuous. We have φ−1(0)=A\overset{-1}{\varphi}(0) = A and φ−1(I)⊂V\overset{-1}{\varphi}(\mathbf{I})\subset V. Let σ\sigma be the mapping from φ−1(I)×I\overset{-1}{\varphi}(\mathbf{I})\times \mathbf{I} into X defined by σ(x,t)=ρ(x,1−t)\sigma (x, t) =\rho (x,1-t) for x∈φ−1(I)x\in \overset{-1}{\varphi}(\mathbf{I}) and t∈It\in \mathbf{I}. For x∈φ−1(I)x\in \overset{-1}{\varphi}(\mathbf{I}), we have σ(x,1)=ρ(x,0)=x\sigma (x,1) =\rho (x,0) =x and σ(x,0)=ρ(x,1)∈A\sigma (x,0) =\rho (x,1)\in A.

The mappings φ\varphi and σ\sigma satisfy the conditions of assertion (iii) of theorem 1 of III, p. 243; hence the pair (X,A)(X,A) possesses the homotopy extension property.

Example 1 01W2

Take for space X the ball Bn\mathbf{B}_n and for subspace A the sphere Sn−1\mathbf{S}_{n-1}. If V = X−{0}-\{0\}, there exists a strong contraction of V onto Sn−1(\mathbf{S}_{n-1}(III, p. 237, example). Hence the pair (Bn,Sn−1)(\mathbf{B}_n,\mathbf{S}_{n-1}) possesses the homotopy extension property.

Corollary 2 01W3

Let X and Y be topological spaces, let A be a closed subspace of X, let B be a closed subspace of Y. If the pairs (X,A)(X,A) and (Y,B)(Y,B) possess the homotopy extension property, the same is true of the pair (X×Y,(X×B)∪(A×Y))(X\times Y,(X\times B)\cup (A\times Y)).

Let φ:X→R+\varphi : X\rightarrow \mathbf{R}_+ and σ:φ−1(I)×I→\sigma :\overset{-1}{\varphi}(\mathbf{I})\times \mathbf{I}\rightarrow X, resp. φ′:Y→R+\varphi ': Y\rightarrow \mathbf{R}_+

and σ′:ψ−1(I)×I→Y\sigma ':\overset{-1}{\psi}(\mathbf{I})\times \mathbf{I}\rightarrow Y, satisfy the conditions of assertion (iv) of theorem 1 for the pair (X,A)(X,A), resp. for the pair (Y,B)(Y,B). Let ψ:X×Y→R+\psi : X\times Y\rightarrow \mathbf{R}_+ be the mapping given by (x,y)↦(x, y)\mapsto inf(φ(x),φ′(x)\varphi (x), \varphi '(x)) ; it is continuous; one has also ψ(x,y)=0\psi (x, y) = 0 for every (x,y)∈X×Y(x, y)\in X\times Y such that x∈Ax\in A or y∈By\in B. For (x,y,t)∈X×Y×I(x, y, t)\in X\times Y\times \mathbf{I} such that t+ψ(x)⩾1t+\psi (x)\geqslant 1, one has t+φ(x)⩾1t+\varphi (x)\geqslant 1 and t+φ′(x)⩾1t+\varphi '(x)\geqslant 1. Thus one defines a continuous mapping

τ:{(x,y,t)∈X×Y×I∣t+ψ(x)⩾1}→X×Y \tau :\{(x, y, t)\in X\times Y\times \mathbf{I}|t+\psi (x)\geqslant 1\} \rightarrow X\times Y

by putting τ(x,y,t)=(σ(x,t),σ′(y,t))\tau (x, y, t) = (\sigma (x, t), \sigma '(y, t)). For every (x,y)∈X×Y(x, y)\in X\times Y, one has τ(x,y,1)=(σ(x,1),σ′(y,1))=(x,y)\tau (x, y,1) = (\sigma (x,1), \sigma '(y,1)) = (x, y). If, moreover, ψ(x,y)⩽1\psi (x, y)\leqslant 1, then φ(x)⩽1\varphi (x)\leqslant 1 or φ′(y)⩽1\varphi '(y)\leqslant 1, so that σ(x,1)∈A\sigma (x,1)\in A or σ′(y,1)∈B\sigma '(y,1)\in B; it follows that τ(x,y,1−ψ(x))\tau (x, y,1-\psi (x)) belong to (A×Y)∪(X×B)(A\times Y)\cup (X\times B). This verifies assertion (iv) of theorem 1, whence the corollary.

Example 2 01W4

Here are other cases in which the pair (X,A)(X,A) possesses the homotopy extension property.

(i) The space X is a paracompact differentiable manifold of class C1C^1 and A a closed submanifold of X. In view of corollary1, this follows from the fact that X is perfectly normal (IX, p. 103, exerc. 11) and that A possesses a tubular neighbourhood in X;

(ii) The space X is a cellular space and A a full subspace with respect to a given cellular decomposition of X.

8. Attachment of a topological space

Let X,BX,B be topological spaces, let A be a subspace of B, and let f:A→Xf: A\rightarrow X be a continuous mapping. Let Y denote the topological sum of the spaces Y1=XY_1= X and Y2=BY_2= B, and identify Y1Y_1 and Y2Y_2 with subspaces of Y by the canonical injections. Let R be the finest equivalence relation in Y for which the elements aa of Y2Y_2 and f(a)f(a) of Y1Y_1 are equivalent, for every a∈Aa\in A. The relation R is the equality relation in Y1Y_1. Let x∈Y1x\in Y_1 and let b∈Y2b\in Y_2; one has xRbxRb if and only if b∈Ab\in A and f(b)=xf(b) =x. Let b,b′∈Y2b, b'\in Y_2; in order that one have bRb′bRb', it is necessary and sufficient that b=b′b=b', or that b,b′∈Ab, b'\in A and f(b)=f(b′)f(b) =f(b').

Definition 9 01W5

The quotient topological space Y/RY/R is called the space obtained by attaching the space B to the space X along the mapping ff. It is denoted by X∪fBX\cup_fB.

Let αf:X→X∪fB\alpha_f: X\rightarrow X\cup_fB and βf:B→X∪fB\beta_f: B\rightarrow X\cup_fB denote the restrictions to Y1Y_1 and Y2Y_2 of the canonical surjection of Y onto Y/RY/R. One has αf∘f=βf∣A\alpha_f\circ f=\beta_f|A.

Remark 1 01W6

The mapping αf\alpha_f is injective. For every subset U

of X, one has αf−1(αf(U))=U\alpha^{-1}_f(\alpha_f(U)) = U and βf−1(αf(U))=f−1(U)\beta^{-1}_f(\alpha_f(U)) =\overset{-1}{f}(U). If A is an open (resp. closed) subset of B, these relations imply that αf\alpha_f is an open (resp. closed) mapping of X into X∪fBX\cup_fB.

Let U be an open subset of X and let V be an open subset of B such that f−1(U)=V∩A\overset{-1}{f}(U) = V\cap A. The union of the subsets U of Y1Y_1 and V of Y2Y_2 is a saturated open subset of Y; its image in X∪fBX\cup_fB is therefore open, and its trace on αf(X)\alpha_f(X) is αf(U)\alpha_f(U). It follows that the mapping αf\alpha_f defines a homeomorphism of X onto its image in X∪fBX\cup_fB.

Remark 2 01W7

Let V be a subset of B. One has αf−1(βf(V))=f(V∩A)\alpha^{-1}_f(\beta_f(V)) =f(V\cap A) and

βf−1(βf(V))=V∪f−1(f(V∩A))\beta^{-1}_f(\beta_f(V)) = V\cup \overset{-1}{f}(f(V\cap A)). If A is closed in B and if the mapping ff is closed, the mapping βf\beta_f is therefore closed. Analogously, if A is open in B and if the mapping ff is open, the mapping βf\beta_f is open.

In these two cases, the mapping βf\beta_f then induces a homeomorphism of B-A onto its image.

Remark 3 01W8

Suppose that A is closed and that the mapping ff is proper. We have just seen that the mapping βf\beta_f is closed. For every x∈Xx\in X,

βf−1(αf(x))=f−1(x)\beta^{-1}_f(\alpha_f(x)) =\overset{-1}{f}(x). It is therefore a quasi-compact subset of A (TG,

I, p. 75, Theorem 1), hence also of B. For every b∈B,βf−1(βf(b))b\in B,\beta^{-1}_f(\beta_f(b))

equal to {b}\{b\} if b∈B−Ab\in B-A, and to f−1(f(b))\overset{-1}{f}(f(b)) if b∈Ab\in A; in both cases, it is a quasi-compact subset of B. The fibres of the mapping βf\beta_f are therefore quasi-compact; consequently (loc. cit.), the mapping βf\beta_f is proper.

The mapping αf\alpha_f is likewise proper (TG, I, p. 72, prop. 2). It follows that the canonical mapping of Y onto X∪fBX\cup_fB is proper.

Example 1 01W9

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. The cylinder Cyl(ff) is none other than the space obtained by attaching the space Y to the space X×IX\times \mathbf{I} along the mapping f∘f\circ pr1_1 of X×{1}X\times \{1\} into Y.

Example 2 01WA

Let X be a topological space, let B be a topological space and let A be a subspace of B, and let f:A→Xf: A\rightarrow X be a continuous mapping inducing a homeomorphism of A onto its image.

Put A′=f(A)A'=f(A); let f′f' be the mapping of A′A' into B which associates to each x∈A′x\in A' the unique inverse image of xx by ff. The mapping f′∘ff'\circ f is continuous; since ff is strict, the mapping f′f' is continuous. There exists a unique continuous mapping uu of the space X∪fBX\cup_fB into the space B∪f′XB\cup_{f'}X which maps αf(x)\alpha_f(x) to βf′(x)\beta_{f'}(x) for x∈Xx\in X and βf(b)\beta_f(b) to αf′(b)\alpha_{f'}(b) for b∈Bb\in B; it is a homeomorphism mapping the subspace αf(X)\alpha_f(X) onto the subspace βf′(X)\beta_{f'}(X).

Proposition 8 01WB

Let Z be a topological space.

a) Let u:X→Zu: X\rightarrow Z and v:B→Zv: B\rightarrow Z be continuous mappings such that v(a)=u(f(a))v(a) =u(f(a)) for every a∈Aa\in A. Then there exists a unique continuous mapping w:X∪fB→Zw: X\cup_fB\rightarrow Z such that w∘αf=uw\circ \alpha_f=u and w∘βf=vw\circ \beta_f=v.

b) Let σ:X×I→Z\sigma : X\times \mathbf{I}\rightarrow Z and τ:B×I→Z\tau : B\times \mathbf{I}\rightarrow Z be homotopies. Suppose that τ\tau is fixed on A and that σ(f(a),t)=τ(a,t)\sigma (f(a), t) =\tau (a, t) for every a∈Aa\in A and every t∈It\in \mathbf{I}. Then there exists a unique homotopy η:(X∪fB)×I→Z\eta : (X\cup_fB)\times \mathbf{I}\rightarrow Z such that η(αf(x),t)=σ(x,t)\eta (\alpha_f(x), t) =\sigma (x, t) and η(βf(b),t)=τ(b,t)\eta (\beta_f(b), t) =\tau (b, t) for x∈X,b∈Bx\in X,b\in B and t∈It\in \mathbf{I}. This homotopy is fixed on βf(A)\beta_f(A).

The proposition results immediately from the definition of a quotient space and prop. 10 (I, p. 19).

Corollary 1 01WC

Let B′B' be a topological space, let A′A' be a subspace of B′B', and let f′:A′→f': A'\rightarrow X be a continuous mapping. Let v:B→B′v: B\rightarrow B' be a continuous mapping such that v(A)⊂A′v(A)\subset A' and ff = f′∘(v∣A)f'\circ (v|A). Let w:X∪fB→X∪f′B′w: X\cup_fB\rightarrow X\cup_{f'}B' be the unique continuous mapping such that w∘αf=αf′w\circ \alpha_f=\alpha_{f'} and w∘βf=βf′∘vw\circ \beta_f=\beta_{f'}\circ v. If vv defines a homotopy equivalence of the pair (B,A)(B,A) onto the pair (B′,A′)(B',A'), then ww defines a homotopy equivalence of the pair (X∪fB,αf(X))(X\cup_fB, \alpha_f(X)) onto the pair (X∪f′B′,αf′(X))(X\cup_{f'}B', \alpha_{f'}(X)).

Let v′:B′→Bv': B'\rightarrow B be a continuous mapping, let τ:B×I→B\tau : B\times \mathbf{I}\rightarrow B be a homotopy fixed on A joining IdB_B to v′∘vv'\circ v, and let τ′:B′×I→B′\tau ': B'\times \mathbf{I}\rightarrow B' be a homotopy fixed on A′A' joining IdB′_{B'} to v∘v′v\circ v'. For a′∈A′a'\in A' and a=v′(a′)a=v'(a'), we have the relations a′=v(a)a'=v(a) and βf(v′(a′))=βf(a)=αf(f(a))=αf(f′(a′))\beta_f(v'(a')) =\beta_f(a) =\alpha_f(f(a)) =\alpha_f(f'(a')). By Prop. 8, a), there exists a unique mapping w′:X∪f′B′→w': X\cup_{f'}B'\rightarrow X∪fBX\cup_fB such that w′∘αf′=αf′w'\circ \alpha_{f'}=\alpha_{f'} and w′∘βf′=βf∘v′w'\circ \beta_{f'}=\beta_f\circ v'. By Prop. 8, b), there exists a unique homotopy η:(X∪fB)×I→X∪f′B′\eta : (X\cup_fB)\times \mathbf{I}\rightarrow X\cup_{f'}B' such that η(αf(x),t)=αf′(x)\eta (\alpha_f(x), t) =\alpha_{f'}(x) and η(βf(b),t)=βf′(τ(b,t))\eta (\beta_f(b), t) =\beta_{f'}(\tau (b, t)) for x∈X,b∈Bx\in X,b\in B and t∈It\in \mathbf{I}. Analogously there exists a unique homotopy η′:(X∪f′B′)×I→\eta ': (X\cup_{f'}B')\times \mathbf{I}\rightarrow X∪fBX\cup_fB such that η′(αf′(x),t)=αf(x)\eta '(\alpha_{f'}(x), t) =\alpha_f(x) and η′(βf′(b),t)=βf(τ′(b,t))\eta '(\beta_{f'}(b), t) =\beta_f(\tau '(b, t)) for x∈X,b∈B′x\in X,b\in B' and t∈It\in \mathbf{I}. The mapping from (X∪fB)×I(X\cup_fB)\times \mathbf{I} into X∪fBX\cup_fB given by (x,t)↦η′(η(x,t),t)(x, t)\mapsto \eta '(\eta (x, t), t) is then a homotopy fixed on αf(X)\alpha_f(X) joining the identity mapping of X∪fBX\cup_fB to the mapping w′∘ww'\circ w. Analogously, the mapping from (X∪f′B′)×I(X\cup_{f'}B')\times \mathbf{I} into X∪f′B′X\cup_{f'}B' given by (x,t)↦η(η′(x,t),t)(x, t)\mapsto \eta (\eta '(x, t), t) is a homotopy fixed on αf′(X)\alpha_{f'}(X) joining the identity mapping of X∪f′B′X\cup_{f'}B' to the mapping w∘w′w\circ w'. The corollary follows.

Example 3 01WD

Let ii be the canonical injection of A into B and let us take for the space B′B' the cylinder of the mapping ii; let αi:A×I→\alpha_i: A\times \mathbf{I}\rightarrow Cyl(ii) and βi:B→\beta_i: B\rightarrow Cyl(ii) be the canonical mappings and ri:r_i: Cyl(ii)→B\rightarrow B the canonical retraction of the cylinder Cyl(ii) onto its base. Let A0A_0 be the subspace αi(A×{0})\alpha_i(A\times \{0\}) of Cyl(ii) and let f0:A0→Xf_0: A_0\rightarrow X be the mapping f∘rif\circ r_i. For a∈Aa\in A, we have

βf∘ri(αi(a,0))=βf(a)=αf(f0(αi(a,0)))=αf(f(a)) \beta_f\circ r_i(\alpha_i(a,0)) =\beta_f(a) =\alpha_f(f_0(\alpha_i(a,0))) =\alpha_f(f(a))

Let η:X∪f0\eta : X\cup_{f_0} Cyl(ii)→X∪fB\rightarrow X\cup_fB be the unique continuous mapping such that η∘αf0=αf\eta \circ \alpha_{f_0}=\alpha_f and η∘βf0=βf∘ri\eta \circ \beta_{f_0}=\beta_f\circ r_i. Suppose that the pair (B,A)(B,A) has the homotopy extension property. Then the mapping rir_i defines a homotopy equivalence of the pair (Cyl(ii),A0,A_0) with the pair (B,A)(B,A) (III, p. 243, th. 1). By Corollary 1, the mapping η\eta then defines a homotopy equivalence of the pair (X∪f′B′,αf′(X))(X\cup_{f'}B', \alpha_{f'}(X)) with the pair (X∪fB,αf(X))(X\cup_fB, \alpha_f(X)). In particular, η\eta is a homotopy equivalence.

Corollary 2 01WE

Let X′X' be a topological space, let u:X→X′u: X\rightarrow X' be a continuous mapping, and put f′=u∘ff'=u\circ f. Let w:X∪fB→X′∪f′Bw: X\cup_fB\rightarrow X'\cup_{f'}B be the unique continuous mapping such that w∘αf=αf′∘uw\circ \alpha_f=\alpha_{f'}\circ u and w∘βf=βf′w\circ \beta_f=\beta_{f'}. If uu defines a homotopy equivalence of the pair (X,f(A))(X, f(A)) with the pair (X′,f′(A))(X', f'(A)), then ww defines a homotopy equivalence of the pair (X∪fB,βf(B))(X\cup_fB, \beta_f(B)) with the pair (X′∪f′B,βf′(B))(X'\cup_{f'}B, \beta_{f'}(B)).

The proof is analogous to that of Corollary 1.

Proposition 9 01WF

Let X and B be topological spaces, let A be a subspace of B, and let f:A→Xf: A\rightarrow X be a continuous mapping.

a) If the pair (B,A)(B,A) has the homotopy extension property, the same is true of the pair (X∪fB,αf(X))(X\cup_fB, \alpha_f(X)).

b) Suppose that the mapping ff is injective and strict. If the pair (X,f(A))(X, f(A)) has the homotopy extension property, the same is true of the pair (X∪fB,βf(B))(X\cup_fB, \beta_f(B)).

a) Suppose that the pair (B,A)(B,A) has the homotopy extension property. Let Z be a topological space, let u:X∪fB→Zu: X\cup_fB\rightarrow Z be a continuous mapping, and let σ:αf(X)×I→Z\sigma :\alpha_f(X)\times \mathbf{I}\rightarrow Z be a homotopy whose final term is the restriction of uu to the subspace αf(X)\alpha_f(X).

Let us put v1=u∘αfv_1=u\circ \alpha_f and denote by τ1:X×I→Z\tau_1: X\times \mathbf{I}\rightarrow Z the mapping given by (x,t)↦σ(αf(x),t)(x, t)\mapsto \sigma (\alpha_f(x), t). Put v2=u∘βfv_2=u\circ \beta_f. Since βf∣A=αf∘f\beta_f|A =\alpha_f\circ f, the mapping from A×IA\times \mathbf{I} into Z which maps (a,t)(a, t) to σ(αf(f(a)),t)\sigma (\alpha_f(f(a)), t) for a∈Aa\in A and t∈It\in \mathbf{I} is a homotopy whose final term is equal to the mapping v2∣Av_2|A. Since the pair (B,A)(B,A) has the homotopy extension property, there exists a homotopy τ2:B×I→Z\tau_2: B\times \mathbf{I}\rightarrow Z whose final term is the mapping v2v_2 and such that τ2(a,t)=σ(αf(f(a)),t)\tau_2(a, t) =\sigma (\alpha_f(f(a)), t) for (a,t)∈A×I(a, t)\in A\times \mathbf{I}.

For a∈Aa\in A and t∈It\in \mathbf{I}, we have τ2(a,t)=τ1(f(a),t)\tau_2(a, t) =\tau_1(f(a), t). By Proposition 8, there exists a unique continuous mapping σ:(X∪fB)×I→Z\sigma : (X\cup_fB)\times \mathbf{I}\rightarrow Z such that σ(αf(x),t)=σ1(x,t)\sigma (\alpha_f(x), t) =\sigma_1(x, t) and σ(βf(b),t)=σ2(b,t)\sigma (\beta_f(b), t) =\sigma_2(b, t), for x∈Xx\in X, b∈Bb\in B and t∈It\in \mathbf{I}. It is a homotopy with final term uu which extends τ\tau, hence assertion a).

b) In view of Example 2 (III, p. 249), assertion b) follows from assertion a).

9. Space obtained by contraction of a subspace

Let X be a topological space and let A be a subset of X. Consider the finest equivalence relation R in X for which all the elements of A are equivalent: two elements of X are equivalent under this relation if they are equal or if they both belong to A.

Definition 10 01WG

The quotient space of X by R is denoted by X/AX/A and is called the space obtained from X by contraction of the subset A.

Let us denote by ρ:X→X/A\rho : X\rightarrow X/A the canonical surjection. Let Y be a topological space and let f:X→Yf: X\rightarrow Y be a continuous mapping. In order that there should exist a continuous mapping g:X/A→Yg: X/A\rightarrow Y such that g∘ρ=fg\circ \rho =f, it is necessary and sufficient that ff should be constant on A.

If the set A is closed (resp. open) in X, the mapping ρ\rho is closed (resp. open) and induces a homeomorphism of X-A onto its image. If A is a closed and quasi-compact subset of X, the mapping ρ\rho is proper (TG, I, p. 75, th. 1).

If the set A is empty, ρ\rho is a homeomorphism. If it is not empty, A is a point of X/AX/A called the base point of X/AX/A and denoted by sX/As_{X/A}. The space X/AX/A is then identified with the space obtained by attaching the space X to the space {sX/A}\{s_{X/A}\} by means of the unique mapping from A into {sX/A}\{s_{X/A}\}.

Proposition 10 01WH

Let X be a topological space and let A be a subset of xx. Suppose that there exists a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X with origin IdX_X, constant on A×{1}A\times \{1\} and such that σ(A×I)⊂A\sigma (A\times \mathbf{I})\subset A. Then the canonical mapping ρ\rho of X into X/AX/A is a homotopy equivalence.

Let ff be the terminal mapping of σ\sigma. It is a continuous mapping of X into X, homotopic to IdX_X. It is constant on A; there therefore exists a unique continuous mapping g:X/A→Xg: X/A\rightarrow X such that g∘ρ=fg\circ \rho =f. On the other hand, since σ(A×I)⊂A\sigma (A\times \mathbf{I})\subset A, there exists a mapping σ′\sigma ' of X/A×IX/A\times \mathbf{I} into X/AX/A such that σ′(ρ(x),t)=ρ(σ(x,t))\sigma '(\rho (x), t) =\rho (\sigma (x, t)) for every x∈Xx\in X and every t∈It\in \mathbf{I}. By I, p. 19, prop. 10, the mapping σ′\sigma ' is continuous. It is a homotopy joining the identity mapping of X/AX/A to the mapping ρ∘g\rho \circ g. Consequently, the mappings gg and ρ\rho are homotopy inverses of one another.

Remark 01WI

Let X be a topological space and let A be a subset of X.

1) If the canonical mapping of X onto X/AX/A is a homotopy equivalence, there exists a homotopy σ:X×I→X\sigma : X\times \mathbf{I}\rightarrow X whose initial mapping is IdX_X and which is constant on A×{1}A\times \{1\}. But it may happen that there exists no such homotopy satisfying moreover the condition σ(A×I)⊂A\sigma (A\times \mathbf{I})\subset A (III, p. 322, exerc. 4).

2) There may exist a homotopy joining the identity mapping of X to a mapping of X into X which is constant on A, without the spaces X and X/AX/A being homotopy equivalent (III, p. 325, exerc. 14).

3) Suppose that A is contractible. If the pair (X,A)(X,A) has the homotopy extension property, the canonical mapping of X onto X/AX/A is a homotopy equivalence. Let in fact σ:A×I→A\sigma : A\times \mathbf{I}\rightarrow A be a homotopy whose initial mapping is the identity mapping of A and whose terminal mapping is a constant mapping. There then exists a homotopy σ′\sigma ' with initial mapping IdX_X which extends σ\sigma. The assertion then follows from proposition 10.

4) Let X and Y be topological spaces, let A be a nonempty subspace of X and let B be a subspace of Y. Let f:X→Yf: X\rightarrow Y be a continuous mapping such that f(A)⊂Bf(A)\subset B. Let φ:X/A→Y/B\varphi : X/A\rightarrow Y/B denote the continuous mapping deduced from ff by passing to quotients. If ff defines a homotopy equivalence of the pair (X,A)(X,A) onto the pair (Y,B)(Y,B), the mapping φ\varphi is a homotopy equivalence of the pair (X/A,sX/A)(X/A, s_{X/A}) onto the pair (Y/B,sY/B)(Y/B, s_{Y/B}). Let P be a topological space reduced to a point, and let α\alpha and β\beta be the constant mappings of A and B into P. Observe that the mapping φ\varphi is identified with the mapping of P∪αXP\cup_{\alpha}X into P∪βYP\cup_{\beta}Y deduced from ff and the identity mapping of P. The assertion then follows from corollary 1 of III, p. 249.

10. Cone of a mapping

Let X and Y be topological spaces and let ff be a continuous mapping of X into Y. Let Cyl(ff) be the cylinder of the mapping ff. Let αf\alpha_f denote the canonical mapping of X×IX\times \mathbf{I} into Cyl(ff) and f0f_0 the mapping x↦αf(x,0)x\mapsto \alpha_f(x,0) of X into Cyl(ff); these mappings induce homeomorphisms of X×IX\times \mathbf{I} and X respectively onto their images in Cyl(ff).

Definition 11 01WJ

The cone of the mapping ff is the topological space deduced from Cyl(ff) by contracting f0(X)f_0(X); it is denoted by Côn(f)(f).

Let βf′:Y→\beta '_f: Y\rightarrow Côn(f)(f) denote the composition of the canonical mapping βf:Y→\beta_f: Y\rightarrow Cyl(ff) and the canonical surjection of Cyl(ff) onto Côn(f)(f). The mapping βf′\beta '_f is continuous and defines a homeomorphism of Y onto a closed subset of Côn(f)(f), called the base of the cone, and which we shall thus identify with Y.

Let αf′:X×I→\alpha '_f: X\times \mathbf{I}\rightarrow Côn(f)(f) denote the composition of the mapping αf\alpha_f and the canonical surjection of Cyl(ff) onto Côn(f)(f). The mapping αf′\alpha '_f is a homotopy whose origin is a constant mapping and whose term is the mapping βf′∘f\beta '_f\circ f.

If X is empty, the mapping βf′\beta '_f is a homeomorphism of Y onto Côn(f)(f).

Suppose that X is not empty. Let ss then denote the base point of the space Cyl(ff)/f0(X)/f_0(X); it is said to be the vertex of the cone Côn(f)(f). Since f0(X)f_0(X) is closed in Cyl(ff), the canonical mapping π:\pi : Cyl(ff)→\rightarrow Côn(f)(f) induces a homeomorphism of Cyl(ff)−f0(X)-f_0(X) onto Côn(f)−{s}((f)-\{s\}(III, p. 252). The vertex of the cone Côn(f)(f) does not belong to its base; the canonical injection of Y into Côn(f)−{s}(f)-\{s\} is a homotopy equivalence (III, p. 239, prop. 6).

Let σf:\sigma_f: Cyl(ff)×I→\times \mathbf{I}\rightarrow Cyl(ff) be the canonical contraction of the cylinder of ff onto its base. For c∈c\in Cyl(ff)−f0(X)-f_0(X) and t∈It\in \mathbf{I}, we have σf(c,t)∉f0(X)\sigma_f(c, t)\notin f_0(X). Hence there exists a unique mapping σf′:(\sigma '_f: (Côn(f)−{s})×I→(f)-\{s\})\times \mathbf{I}\rightarrow Côn(f)−{s}(f)-\{s\} such that σf′(π(c),t)=π(σf(c,t))\sigma '_f(\pi (c), t) =\pi (\sigma_f(c, t)) for c∈c\in Cyl(ff)−f0(X)-f_0(X) and t∈It\in \mathbf{I}. It is continuous and is a strong contraction of Côn(f)−{s}(f)-\{s\} onto Y. It is said to be the canonical contraction of Côn(f)−{s}(f)-\{s\} onto its base. Its term is a retraction of Côn(f)−{s}(f)-\{s\} onto Y, called the canonical retraction of the cone deprived of its vertex onto its base.

Proposition 11 (Universal Property of Cones)

Let Z be a topological space, let β:Y→Z\beta : Y\rightarrow Z be a continuous mapping and let α:X×I→Z\alpha : X\times \mathbf{I}\rightarrow Z be a homotopy whose origin is a constant mapping and whose term is equal to β∘f\beta \circ f. There exists a unique continuous mapping φ\varphi of Côn(f)(f) into Z such that α=φ∘αf′\alpha =\varphi \circ \alpha '_f and β=φ∘βf′\beta =\varphi \circ \beta '_f.

From the universal property of cylinders (III, p. 238, prop. 4), there exists a unique continuous mapping h:h: Cyl(ff)→\rightarrow Z such that α=h∘αf\alpha =h\circ \alpha_f and β=h∘βf\beta =h\circ \beta_f. Since the source of α\alpha is a constant mapping, the restriction of hh to the subspace αf(X×{0})\alpha_f(X\times \{0\}) is constant. There therefore exists a unique continuous mapping φ:\varphi : Cone(f)→Z(f)\rightarrow Z such that h=φ∘πh=\varphi \circ \pi, where π\pi denotes the canonical surjection of Cyl(ff) onto Cone(f)(f). The mapping φ\varphi satisfies α=φ∘αf′\alpha =\varphi \circ \alpha '_f and β=φ∘βf′\beta =\varphi \circ \beta '_f and it is the only one having these properties, since the images of αf′\alpha '_f and βf′\beta '_f cover Cone(f)(f).

Example 1 01WK

Let X be a topological space. One denotes by C(X), and calls the cone of the space X, the cone of the mapping IdX_X; it is the space obtained from X×IX\times \mathbf{I} by contracting X×{0}X\times \{0\}. Let π\pi be the canonical mapping of X×IX\times \mathbf{I} into C(X); the image C′(X)C'(X) of X×[0,1[X\times [0,1[ under π\pi is called the open cone of the space X.

Assume that X is not empty. Then the cone C(X) is not empty and its base point is again called the vertex of the cone.

The mapping ((x,t),u)↦(x,t(1−u))((x, t), u)\mapsto (x, t(1-u)) of (X×I)×I(X\times \mathbf{I})\times \mathbf{I} into X×IX\times \mathbf{I} is a homotopy joining the identity mapping of X×IX\times \mathbf{I} to the mapping (x,t)↦(x,0)(x, t)\mapsto (x,0). One deduces from this, by passing to quotient sets, a mapping σ:C(X)×I→C(X)\sigma : C(X)\times \mathbf{I}\rightarrow C(X) such that σ(π(x,t),u)=π(x,t(1−u))\sigma (\pi (x, t), u) =\pi (x, t(1-u)) for x∈X,t,u∈Ix\in X,t, u\in \mathbf{I}. The mapping σ\sigma is continuous by I, p. 19, prop. 10. The mapping σ\sigma is then a pointed homotopy at ss joining the identity mapping of C(X) to the constant mapping with image {s}\{s\}. Consequently, the cone C(X) is contractible to its vertex ss. Since σ(C′(X)×I)\sigma (C'(X)\times \mathbf{I}) is contained in C′(X)C'(X), the mapping σ\sigma defines, by passing to subspaces, a pointed homotopy at ss joining the identity mapping of C′(X)C'(X) to the constant mapping with image {s}\{s\}; the open cone C′(X)C'(X) is therefore contractible to ss.

Remark 1 01WL

Let X be a topological space and let A be a subspace of X; let ii denote the canonical injection of A into X. The canonical retraction rir_i of the cylinder Cyl(ii) onto its base maps the subspace αi(A×{0})\alpha_i(A\times \{0\}) into A. Let ρ:\rho : Cone(i)→X/A(i)\rightarrow X/A be the continuous mapping deduced from it by passing to the quotient spaces. Suppose that the pair (X,A)(X,A) has the homotopy extension property. By III, p. 243, Theorem 1, the mapping rir_i defines a homotopy equivalence of the pair (Cyl(ii),αi(A×{0}), \alpha_i(A\times \{0\})) with the pair (X,A)(X,A). By Remark 4 of III, p. 253, the mapping ρ\rho is then a homotopy equivalence of the pair (Cone(i),s(i), s) with the pair (X,sX/A)(X, s_{X/A}). In particular it is a homotopy equivalence.

Remark 2 01WM

Let X and Y be topological spaces and let f:X→Yf: X\rightarrow Y be a continuous mapping. The canonical mapping αf′:X×I→\alpha '_f: X\times \mathbf{I}\rightarrow Cone(f)(f) is constant on X×{0}X\times \{0\} and therefore defines a continuous mapping γf:C(X)→\gamma_f: C(X)\rightarrow Cone(f)(f). The restriction of the mapping γf\gamma_f to C′(X)C'(X) is injective and strict, and defines by passing to the subspaces a homeomorphism of the open cone C′(X)C'(X) onto the complement of Y in Cone(f)(f).

Let us identify the base of the cone C(X) with the space X and let uu denote the continuous mapping of Y∪fC(X)Y\cup_fC(X) into Cone(f)(f) deduced from the mappings βf′:Y→\beta '_f: Y\rightarrow Cone(f)(f) and γf:C(X)→\gamma_f: C(X)\rightarrow Cone(f)(f). Conversely, let v:v: Cone(f)→Y∪fC(X)(f)\rightarrow Y\cup_fC(X) be the unique continuous mapping such that v∘αf′v\circ \alpha '_f is the canonical mapping of X×IX\times \mathbf{I} onto C(X) and v∘βf′v\circ \beta '_f is the canonical mapping of Y onto Y∪fC(X)Y\cup_fC(X). The mappings uu and vv are inverse homeomorphisms of one another.

Exercises

See the exercises for § 1.

Machine translation of the English, by gpt-5.4, gpt-5-6-mini, gpt-5-mini, 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.