Exercises
1.4 Maps between CW complexes
(Xn× {0})∪((Xn−1∪A)×I) is a strong deformation retract of (Xn∪A)×I.
It follows immediately thatZn−1 is a strong deformation retract of Zn. By 1.3.14,Z−1 is a strong deformation ofZ, as claimed. The last part is proved
as in the proof of 1.3.12.
Examination of the proof of 1.3.15 shows a little more which can be useful:
Addendum 1.3.16.Let(X, A)be a CW pair and letF : (X×{0})∪(A×I)→ Z be a map. The mapF extends to a mapF˜ :X×I→Z such that, for every cell eαof X−A,F˜1(eα) =F0(eα)∪F˜(•eα×I).
The definition of the homotopy extension property has nothing to do with CW complexes. Therefore the last part of 1.3.15 is more powerful than it might appear; it says that the existence of a CW pair structure on the pair of spaces (X, A) ensures that (X, A) has the homotopy extension property with respect to every space.
Corollary 1.3.17.If(X, A)is a CW pair andZ is a contractible space, every
map A→Z extends to a map X→Z.
Exercises
1. Prove 1.3.4–1.3.7.
For the next two exercises (X, A) is a CW pair.
2. Prove that if A is a weak deformation retract of X then A is a deformation retract ofX.
3. Prove that ifA is a deformation retract of X thenA is a strong deformation retract ofX.
Hint for 2. and 3.: Apply 1.3.15 appropriately.
1.4 Maps between CW complexes 29 Proposition 1.4.1.Let m > n. Let U be an open subset of Rn such that the spaceclU is compact. Letf : clU →Bmbe such that f(frU)⊂Sm−1. Then f is homotopic, rel fr U, to a map g such that some point of Bm−Sm−1 is not in the image ofg.
Proof. There are many proofs in the literature. The most common involve
“simplicial approximation” off (see, for example, [146, 7.6.15], ). We give a proof here which assumes instead knowledge of multivariable calculus.
Let V =f−1(Bm−Sm−1). Define : V →(0,∞) by (x) = distance14 fromxto frV. Using partitions of unity (see [85, pp. 41–45] ), there exists a C1map ˆg:V →Bm−Sm−1such that|f(x)−ˆg(x)|< (x) for allx∈V. Since
|f(x)−g(x)ˆ | →0 asxapproaches frRmV, ˆg can be extended continuously to g : clU →Bm by definingg = ˆg onV andg =f on (clU)−V. Thenf is homotopic tog by the homotopy
Ht(x) = (1−t)f(x) +tg(x).
We haveg(clU)∩(Bm−Sm−1) =g(V). We will show thatg(V) is a proper subset ofBm−Sm−1.
IdentifyRn withRn× {0} ⊂Rm. ThenV ⊂Rn× {0} ⊂Rm andg(V) = gp(V) wherep:Rm→Rn is projection.
Anm-cube CinRmof side λ >0 is a productC=I1×. . .×Imof closed intervals each of lengthλ. The m-dimensional measure ofC isλm. A subset X ⊂Rm hasmeasure 0 if for every >0,X can be covered by a countable set ofm-cubes the sum of whose measures is less15than.
Clearly V ⊂ Rm has m-dimensional measure 0. Since ˆg is C1, so also is ˆgp |: p−1(V) → Bm −Sm−1. By the Mean Value Theorem, given any compact subsetK⊂p−1(V), there existsµ(K)>0 such that|ˆgp(x)−gp(y)ˆ | ≤ µ(K)|x−y|for all x, y ∈K; we may takeµ(K) = sup{||D(ˆgp)x|| | x∈K}. It follows that ifC⊂Kis anm-cube of sideλ, thengp(C) lies in anm-cube C of side√
m.µ(K).λ. ThusC has measure (√
m.µ(K))m.(measure ofC).
It is a theorem of elementary topology that every open cover of an open subset of Rn has a countable subcover; see pages 174 and 65 of [51] for a proof. Hence we may write V =
∞ i=1
Vi where each set Vi lies in the interior of a closedm-ballBi⊂p−1(V). Let >0 be given. Each Vi can be covered by countably manym-cubesCi1, Ci2, . . .such that eachCij⊂intBi, and
j
measure (Cij)< . Hence, by the above discussion,g(Vi) can be covered by countably many cubesCij such that
j
measure (Cij )< (√
m.µ(Bi))m. So
14All distances in euclidean spaces refer to the usual euclidean metric |x−y| = (P
(x2i −yi2))1/2.
15Of course, this ism-dimensional Lebesgue measure, but we need almost no mea- sure theory!
gp(Vi) has measure 0. But, clearly, the countable union of sets of measure 0 has measure 0, so gp(V)(= g(V)) has measure 0. Clearly, Bm−Sm−1 does
not have measure 0. Sog(V) is a proper subset.
Proposition 1.4.2.Letenαbe ann-cell of the CW complexX, and letz∈e◦nα. Then Xn−◦enα is a strong deformation retract of Xn− {z}.
Proof. Leth : (Bn, Sn−1) →(enα,e•nα) be a characteristic map. For any y ∈ Bn−Sn−1, the formula x→x+ (1− |x|)y defines a homeomorphism ofBn fixing each point ofSn−1and sending 0 toy. Hence we may assumeh(0) =z.
LetH : (Bn−{0})×I→Bn−{0}be the homotopy (x, t)→(1−t)x+tx/|x|. Note thatH is a strong deformation retraction ofBn−{0}ontoSn−1. It is an easy exercise to check that a functionHexists making the following diagram commute:
(Bn− {0})×I −−−−→H Bn− {0}
⏐⏐ h|×id
⏐⏐ h| (enα− {z})×I −−−−→H enα− {z}
i.e., that if (h(x1), t1) = (h(x2), t2) thenh(H(x1, t1)) =h(H(x2, t2)). Sinceh is a quotient map, so ish|, and also (h|)×id by 1.3.11. ThusHis continuous and is therefore the required strong deformation retraction.
If X is a CW complex and S ⊂ X, the carrier of S is the intersection of all subcomplexes ofX which contain S. It is a subcomplex ofX, denoted C(S).
Theorem 1.4.3. (Cellular Approximation Theorem) Let f : X → Y be a map between CW complexes and let A be a subcomplex of X such that f |A:A→Y is cellular. Then f is homotopic, relA, to a cellular map.
Proof. We will constructH :X×I→Y such thatH0=f,Ht|A=f |Afor allt, andg:=H1is cellular. ThisH is defined by induction on (Xn∪A)×I.
For each vertexe0 ofX0−A, there is a unique celldsuch thatf(e0)∈d.◦ Define H | {e0} ×I to be any path inC(d) fromf(e0) to a vertex of C(d):
call that vertexg(e0). (There is such a path becauseC(d) is path connected.) AssumeH already defined on (Xn−1∪A)×Iwith the desired properties.
Extend H to agree with f on Xn× {0}. By 1.3.15, H extends to a map H¯ : (Xn∪A)×I → Y. Let ¯g = ¯H1. Then ¯g | Xn−1 = g | Xn−1 and is cellular, but ¯g might not map n-cells of X −A into Yn. We claim that ¯g is homotopic rel Xn−1∪A to a cellular map. For eachn-cell en of X, the carrier of ¯g(en),C(¯g(en)), is a finite subcomplex ofY, by 1.2.13. Letdmbe a top-dimensional cell ofC(¯g(en)). Writeq :en →C(¯g(en)) for the restriction of ¯g. If m ≤ n, the next step can be omitted, so suppose m > n. Since q(•en)⊂Yn−1, q−1(d−d)• ⊂en−•en. We want a homotopy of q, rele•n, to
1.5 Neighborhoods and complements 31 a map which misses a point of d◦m, for then, by 1.4.2, there is a homotopy of q, rel•en, to a map ofen into C(¯g(en))−d, and after finitely many such◦ homotopies, rel•en, we will obtain a map of en intoYn.
If q(en)∩d◦ = ∅, there is nothing to do. Otherwise, let y ∈ q(en)∩d.◦ q−1(y) is a compact set lying in◦en. LetDbe a neighborhood ofy ind◦which is homeomorphic to a closed ball, and let U = q−1(intD). Then cl U is a compact subset ofe◦n and q|: clU →D. By 1.4.1, the desired homotopy of q, relen−U (hence rel•en), exists.
Doing this on each n-cell of X, we get the desired homotopy of ¯g, and, hence, in the obvious way, the desiredH : (Xn∪A)×I→Y. Addendum 1.4.4.The homotopy H : X×I →Y obtained in the proof of 1.4.3 has the property that for each cell eof X,H(e×I)⊂C(f(e)).
Proof. Assume the induction hypothesisH(em×I)⊂C(f(em)) for allm < n.
It is clearly true when n = 1. Let e•n ⊂ ◦emα1
1 ∪. . .∪e◦mαr
r. Then ¯g(en) ⊂ H¯(en ×I) = H(en×0)∪H(e•n×I) ⊂ f(en)∪
r i=1
C(f(emαi
i)) ⊂ C(f(en)).
(The last inclusion holds because, if ◦em∩•en = ∅, then C(em) ⊂ C(en).) The other half of the homotopy ofen takes place in C(¯g(en))⊂C(f(en)). So
H(en×I)⊂C(f(en)).
Corollary 1.4.5.LetX be a CW complex;X is path connected iffX1 is path connected.
Proof. As in the proof of 1.4.3, there is a path from any point ofX to a point ofX1, soX is path connected ifX1 is path connected. For the converse, let ω:I→X be a path withω(0), ω(1)∈X1. Without loss of generality assume ω(0), ω(1) ∈ X0. Apply 1.4.3 to produce a cellular path, necessarily in X1,
fromω(0) toω(1).