Cellular Homology
2.5 Orientation and incidence number
2. In the definition of “suspension” prove that Σg is continuous when gis; and thatΣg1 andΣg2 are homotopic wheng1 andg2 are.
3. Prove that homeomorphic CW complexes have the same dimension.
4. Describe a CW complex structure forRn.
5. (for those who know category theory) Show that “wedge” is the category theoret- ical coproduct in the category Pointed Spaces. What is the category theoretical coproduct in the category Spaces?
2.5 Orientation and incidence number 53 Proof. First, we observe thatr|:Sn−1→Sn−1is not homotopic to id. Forn= 1, this is obvious. Forn >1, consider the homeomorphismt12:Sn−1→Sn−1 (x1, x2, x3, . . . , xn)→(x2, x1, x3, . . . , xn). By 2.4.3,r|=t−121◦fn−1,−1◦t12. If r|were homotopic to id, thenfn−1,−1would be homotopic to id (conjugating the homotopy byt12), which is false by 2.4.4.
It follows that r : (Bn, Sn−1) → (Bn, Sn−1) is not equivalent to (i.e., pairwise homotopic to) the identity map. Clearly (h◦r):Bn/Sn−1→enα/•enα is the compositionh◦r. Sokn◦(h)−1◦(h◦r)◦kn−1=kn◦r◦kn−1, and this map does not have degree 1. Thushandh◦rdetermine different orientations ofenα. By 2.5.1, every orientation is given by one of these.
The proof of 2.5.1 depends on some lemmas.
Lemma 2.5.3.Let n≥1. Leta= (0, . . . ,0,1)∈Sn andb= (0, . . . ,0,−1)∈ Sn. Let en+ and en− be the closed upper and lower hemispheres, respectively.
Letf, g:Sn→Sn be maps such thata∈f(en−)∪g(en−)andb∈f(en+)∪g(en+).
If f andg have the same degree, thenf| g|:Sn−1→Sn− {a, b}.
Proof. We considerf andgas maps (Sn, Sn−1)→(Sn, Sn−{a, b}). As in the first part of the proof of 2.4.18,f andg are (pairwise) homotopic to mapsf andg which preserve hemispheres, hence, as in the proof of 2.4.5, to suspen- sionsΣf andΣg, where f, g:Sn−1→Sn−1. These homotopies restrict to homotopiesSn−1×I→Sn− {a, b}. Sof| inclusion◦fandg| inclu- sion◦ g (as mapsSn−1 →Sn− {a, b}). Now degree(f) = degree(Σf) = degree(f) = degree(g) = degree(Σg) = degree(g). So f g, by 2.4.5, hencef| g|. Note that, while 2.4.18 is false forn= 1, the part of the proof
used here works forn= 1.
Corollary 2.5.4.Let f, g : (Bn, Sn−1) → (enα,•enα) be characteristic maps inducing f, g : Bn/Sn−1 → enα/•enα. For t ∈ (0,1), let At be the annulus {x ∈ Bn | t < |x| < 1} and let St = {x ∈ Bn | |x| = t}. Let z ∈ e◦nα. If there exists t such that f(At)∪g(At) ⊂ e◦nα− {z}, and if f g, then f | g|:St→enα− {z}.
Proof. SinceBn/Sn−1andenα/•enαare homeomorphic toSn, the result follows from 2.5.3. In fact,f | g|as maps fromSttoe◦nα− {z}. Lemma 2.5.5.Letz∈e◦nα, letf andgbe two maps(Sn−1×I, Sn−1×{1})→ (enα− {z},e•nα), and let H :Sn−1× {0} ×I →enα− {z} be a homotopy from f |Sn−1× {0} tog|Sn−1× {0}. Then there is a homotopy K: (Sn−1×I× I, Sn−1× {1} ×I)→(enα− {z},e•nα)from f tog extendingH.
Proof. The map (Sn−1× {0} ×I)∪(Sn−1×I× {0,1})→enα− {z}, which agrees with H onSn−1× {0} ×I, withf on Sn−1×I× {0} and withg on Sn−1×I× {1}, extends, by 1.3.15, to a mapL:Sn−1×I×I →enα− {z}.
The sete◦nα−(image ofL) =:U is a neighborhood ofzin ◦enα. By 2.4.14, there is a homotopy D : X ×I → X such that D0 = id, Dt = id on X −◦enα for all t, and D1(enα−U) ⊂ e•n. Define M : Sn−1×I×I → enα− {z} by M(x, s, t) = L(x, s,2t) if 0≤ t ≤ 12 and by M(x, s, t) = D2t−1(L(x, s,1)) if
1
2 ≤t≤1. It is a matter of planar geometry to construct a homeomorphism h : I × I → I × Iagreeing with id onI× {0}, taking (0, t) to (0,2t) and (1, t) to (1,2t) when 0≤t≤ 12, and taking the rest of frR2(I×I) intoI× {1}.
The requiredK isM◦h−1.
Proof (of 2.5.1). “Only if” is obvious. We prove “if”. By the hypothesis and 2.4.5h1 h2 :Bn/Sn−1→ enα/•enα. Letz ∈e◦nα. There exists t∈(0,1) such that At is mapped into e◦nα− {z} by h1 and by h2. By 2.5.4, there exists H :St×I→enα− {z}such thatH0=h1 |St andH1=h2 |St. Therefore,
by 2.5.5 and 1.3.17,h1h2.
An oriented CW complex is a CW complexX together with a choice of orientation on each cell. If A is a subcomplex of X, it is understood to be given the inherited orientation (each cell oriented as in X): we call (X, A) anoriented CW pair. A quotient complexX/∼, as in Sect. 1.2, is given the quotient orientation: the vertices {Aα} are oriented by +1; all other cells receive their orientation via the quotient mapX→X/∼.
LetX be an oriented CW complex,enα ann-cell andenβ−1 an (n−1)-cell.
First, assumen≥2. Choose characteristic mapshα: (Bn, Sn−1)→(enα,e•nα) andhβ: (Bn−1, Sn−2)→(enβ−1,•enβ−1) representing the orientations. Consider the commutative diagram:
Xn−1
oo ? _en−1β
oo hβ Bn−1
•
enα
inclusion
77o
oo oo oo oo oo oo oo
sαβ //Xn−1/(Xn−1−◦enβ−1)oo
rβ enβ−1/•enβ−1oo
hβ Bn−1/Sn−2
kn−1
Sn−1
hα|
OO
Sn−1
Here rβ is induced by inclusion and is clearly a homeomorphism. Unmarked arrows are quotient maps. sαβ is the indicated composition, and kn−1 is as before. Sincehβ is a homeomorphism, we obtain a map
kn−1◦(hβ)−1◦r−β1◦sαβ◦hα|:Sn−1−→Sn−1
whose degree, denoted by [enα:enβ−1], is theincidence number of the oriented cells enαand enβ−1. One should think of it as the (algebraic) number of times enαis attached to enβ−1. By 2.5.1 we have:
2.5 Orientation and incidence number 55 Proposition 2.5.6.For n ≥2, the definition of [enα : enβ−1] depends on the orientations but not on the specific mapshα andhβ. Note that the definition of [enα :enβ−1] also depends on kn−1. Indeed, the homeomorphisms{kn|n≥1}have been chosen once and for all. We will say more about how we want them to have been chosen later (see 2.5.16).
We now extend the definition of incidence number to the case n = 1.
Given an oriented 1-cell e1α and a 0-cell e0β oriented by ∈ {±1}, choose a characteristic map hα : (B1, S0) → (e1α,•e1α) representing the orientation.
Define theincidence number [e1α:e0β] by:
[e1α:e0β] =
⎧⎪
⎪⎪
⎨
⎪⎪
⎪⎩
ifhα(1) =e0β and hα(−1)=e0β
− ifhα(1)=e0β and hα(−1) =e0β 0 ifhα(1) =hα(−1) =e0β
0 ifhα(1)=e0β and hα(−1)=e0β. Clearly we have:
Proposition 2.5.7.The definition of [e1α :e0β] depends only on the orienta-
tions.
Proposition 2.5.8.Letenα andenβ−1be (oriented) cells,n≥1. Ifenβ−1 is not a subset of enα, then[enα:enβ−1] = 0.
Proof. Whenn≥2, the mapSn−1→Sn−1 whose degree defines [enα :enβ−1] is not surjective, so, by 1.3.4 and 1.3.7, it is homotopic to a constant map. By 2.4.3 its degree is 0. Whenn= 1, the Proposition is obvious.
We sayeβ is aface ofeαifeβ⊂eα.
Now we turn to numbers associated with maps. LetX andY be oriented CW complexes, and letf :X →Y be a map such that, for somen,f(Xn)⊂ Yn andf(Xn−1)⊂Yn−1; of course cellular maps have this property for all n. For eachenα in X and ˜enβ in Y we wish to define an integer [enα : ˜enβ : f] measuring the (algebraic) number of timesf mapsenαonto ˜enβ. First, letn≥1 and consider the commutative diagram:
Bn hα //
enα f| //
Yn oo ? _
˜
enβ oo ˜hβ
Bn
Bn/Sn−1
hα
//
kn
enα/•enα
fαβ
//Yn/(Yn−◦˜enβ)oo
rβ ˜enβ/e•˜nβoo
˜hβ Bn/Sn−1
kn
Sn Sn.
Here ˜enβ is a cell ofY; ˜hβis a characteristic map defining the orientation;fαβ is induced byf, and is well defined sincef(Xn, Xn−1)⊂(Yn, Yn−1);rβ is a homeomorphism as before. Let [enα: ˜enβ :f] be the degree of the map
kn◦(˜hβ)−1◦r−β1◦fαβ ◦hα◦kn−1:Sn→Sn.
For the case n = 0, let e0α ⊂ X have orientation and let ˜e0β ⊂ Y have orientation ˜. Define [e0α : ˜e0β : f] = ˜ iff(e0α) = ˜e0β, and [e0α : ˜e0β : f] = 0 otherwise. From 2.5.1 we conclude:
Proposition 2.5.9.The definition of[enα: ˜enβ:f]depends only on the orien-
tations.
As with 2.5.8 we have:
Proposition 2.5.10.Let enα and e˜nβ be oriented n-cells of X and Y respec- tively. If˜enβ is not a subset off(enα)then[enα: ˜enβ:f] = 0.
The number [enα: ˜enβ :f] is called themapping degree off with respect to enαand ˜enβ.
Next, we discuss product orientations. LetX andY be oriented CW com- plexes. We define the product orientation on the CW complex X ×Y by specifying an orientation for eachemα ×e˜nβ whereemα, ˜enβare cells of X andY respectively: (i) if m >0 and n > 0, and if hα : (Im,I•m) → (emα,e•mα) and
˜hβ : (In,•In)→(˜enβ,˜e•nβ) are characteristic maps defining the separate orien- tations,emα טenβ is to be oriented byhαטhβ. (ii) Ifm >0 andn= 0, so that
˜
e0β={yβ}, ifhαorientsemα, and if 1 orients ˜e0β, thenemα טe0βis to be oriented by the mapIm → emα טe0β, x→ (hα(x), yβ); if−1 orients ˜e0β, the opposite orientation is to be used. (iii) If m= 0 andn >0, the rule of orientation is similar to that given in (ii). (iv) Ifm= 0 =n, withe0α oriented by, and ˜e0β oriented by ˜, then e0α×e˜0β is to be oriented by˜.
We can use this to define a preferred orientation on the CW complexIn as follows: whenn= 0, thecanonical orientation is 1; whenn= 1, thecanonical orientation onI1(with two vertices and one 1-cell) is: 1 on each vertex, and (the equivalence class of) id :I1→I1on the 1-cell. Thecanonical orientation onIn is then defined inductively to be the product orientation onIn−1×I1. We leave the proofs of the next two statements to the reader:
Proposition 2.5.11.If In is factorized as Ir×Is, and each factor carries the canonical orientation, then the product orientation on Ir×Is coincides
with the canonical orientation onIn.
Proposition 2.5.12.Let r ≥ 1. Let e be the r-cell of In obtained by con- straining the ijth coordinate to be j ∈ {1,−1}, for 1 ≤ j ≤ n−r. Let the remaining r coordinates be indexed by p1 < p2 < . . . < pr. The canonical
2.5 Orientation and incidence number 57 orientation on e is given by the characteristic map (Ir,•Ir) → (e,e)• taking (x1, . . . , xr)to(y1, . . . , yn)whereyij =j for 1≤j ≤n−randypj =xj for
1≤j≤r.
Proposition 2.5.13.Letri:I•n→I•nbe the homeomorphism(x1, . . . , xn)→ (x1, . . . , xi−1,−xi, xi+1, . . . , xn). Then ri is orientation reversing.
Proof. If n = 1, this is obvious. Next, let n = 2 and i = 1, let Mt be the matrix
cos(tπ) sin(tπ) sin(tπ)−cos(tπ)
, and letan:In →Bn be the homeomorphism given in Sect. 1.2. Then a−21M a2 | I•2 gives a homotopy from a−21f1,−1a2 to r1. For all other cases, embed this matrix in the appropriaten×nmatrix. (For another proof, adapt the first paragraph of the proof of 2.5.2.) Now we discuss the incidence numbers ofIn with its (n−1)-dimensional faces. For =±1, let Fi, be {(x1, . . . , xn) ∈ In | xi = }. Each Fi, is an (n−1)-cell ofIn. All cells are canonically oriented.
Proposition 2.5.14.[In:Fi,1] =−[In:Fi,−1].
Proof. This is clear whenn= 1. Assumen≥2. Recall our permanent choice of homeomorphism,an, for identifyingInwithBn; see Sect. 1.2. Consider the following commutative diagram:
Sn−1 a
−1n | //I•n
bn //
ri
I•n/(•In−F◦i,1)oo
∼= Fi,1/F•i,1
Fi,1
aaCCCCCCCC
In−1
hi,1
ffMMMMM MMMMMMM
hi,−1
xxqqqqqqqqqqqq
//In−1/•In−1
kn−1
∼= //
∼= hi,−1
∼= hi,1
OO
Sn−1
Fi,−1
}}||||||||
Sn−1 a
−1n | //I•n
cn //I•n/(•In−F◦
i,−1)oo ∼
= Fi,−1/F•i,−1
Here,hi,:In−1→Fi, is the canonical characteristic map
(x1, . . . , xn−1)→(x1, . . . , xi−1, , xi+1, . . . , xn−1) ofFi,, given by 2.5.12, and ri is as in 2.5.13. All other arrows are the obvious quotients or inclusions or are induced by such. By 2.5.13,ri is orientation reversing.
The required incidence numbers are the degrees off andgwheref [resp.
g] is the composition of maps leading from the top left [resp. bottom left] copy ofSn−1to the right copy ofSn−1. Note thatf =g◦(an◦ri◦a−n1). By 2.5.13, ri has degree−1. So, by 2.4.19, deg(f) =−deg(g).
Proposition 2.5.15.The mapsbnandcnin the above diagram are homotopy equivalences. Hence the incidence numbers in 2.5.14 are±1.
Proof. Leten(t) ={x∈ Sn | t ≤xn+1 ≤1}. The spaceen(0) is the upper closed hemisphere, while en(1) = {north pole}. The reader can easily con- struct a homotopyH :Sn×I →Sn such thatH0 = id, Ht(en(0)) =en(t), andHtmapsSn−en(0) bijectively ontoSn−en(t), for allt. The mapH1 is a quotient map (see Sect. 1.1), and sinceH1 id,H1 is a homotopy equiv- alence. Clearly this implies that bn and cn are homotopy equivalences. The
second part then follows from 2.4.20.
Proposition 2.5.14 makes a relative statement about incidence numbers.
But even though we know, by 2.5.15, that these incidence numbers are±1, we cannot say which is which (whenn≥2) until we specify the homeomorphism kn−1 which “was chosen once and for all” earlier in this section. In fact we are only making our choice now:
Convention 2.5.16.For n ≥ 2, the canonical homeomorphism kn−1:Bn−1/Sn−2→Sn−1 is to be chosen to make [In:Fn,1] = (−1)n+1. Proposition 2.5.17.[In:Fi,1] = (−1)i+1. Hence7 [In:Fi,−1] = (−1)i. Proof. This is clear when n = 1. Assumen ≥2. The orientation onFn,1 is given byIn−1→Fn,1, (x1,· · · , xn−1)→(x1,· · ·, xn−1,1); and the orientation on Fn−1,1 is In−1 → Fn−1,1, (x1,· · ·, xn−1) → (x1,· · ·, xn−2,1, xn−1). Let h : In → In be the restriction of the linear automorphism of Rn which is the identity in all but the last two coordinates and is given by the matrix [0 11 0] in those coordinates. Thenhis orientation reversing onI•n(compare the proof of 2.5.13) but matches up the above orientations on the (n−1)-faces.
By considering the diagram defining incidence numbers, one sees that this implies [In:Fn−1,1] =−[In, Fn,1]. The conclusion follows, by induction, from
2.5.14, 2.5.15 and 2.5.16.
We now interpret the incidence number [enα:enβ−1] and the mapping degree [enα: ˜enγ :f] homologically, wheref :X→Y is a cellular map of oriented CW complexes. It is to be understood that singular homology is taken with Z- coefficients. We saw in Sect. 2.3 that the coefficient of ¯hβ∗(λn−1) in∂¯hα∗(λn) is the image of 1 under a certain homomorphismZ →Z. We wish to claim
7 The choice in 2.5.16 which leads to the incidence numbers stated in 2.5.17 is not a universal convention. For example, choices used in [42] lead to [In : Fi,1] = (−1)n−i.
2.5 Orientation and incidence number 59 that this integer is precisely the incidence number [enα : enβ−1]. However, for this to be true some further once-and-for-all choices must be made correctly.
By an easy exercise in singular homology using excision as outlined in Sect.
2.2, there are canonical isomorphisms
Hn∆(Bn, Sn−1) ∼= //Hn∆(Bn/Sn−1, v)oo ∼= Hn∆(Bn/Sn−1) kn∗ //Hn∆(Sn) when n ≥ 1. Let κn ∈ Hn∆(Sn) be the image of the generator λn ∈ Hn∆(Bn, Sn−1) under this composition of isomorphisms. We also have a ho- momorphism∂∗:Hn∆(Bn, Sn−1)→Hn∆−1(Sn−1) which is an isomorphism for n≥2 and a monomorphism whenn= 1; see 2.2.1. Indeed, the image of λ1
under∂∗ :H1∆(B1, S0)→H0∆(S0)∼=Z⊕Z is the homology class of the sin- gular 0-cycle±({1} − {−1}), with the + or−depending on which generator ofH1∆(B1, S0) has been chosen asλ1.
Now we make our choices. Let κ0 ∈ H0∆(S0) be the homology class of {1} − {−1}. Define the generatorsλn∈Hn∆(Bn, Sn−1) andκn∈Hn∆(Sn) by H0∆(S0)oo oo H1∆(B1, S0) ∼= //H1∆(S1)oo∼= H2∆(B2, S1) ∼= //H2∆(S2)oo · · ·
κ0oo λ1 //κ1oo λ2 //κ2 oo · · · This specifiesλn,n≥1, for the first time. We also defineλ0∈H0∆(B0) to be the homology class of the singular 0-cycle{0}. With these choices we get:
Proposition 2.5.18.The coefficient of ¯hβ∗(λn−1) in ∂¯hα∗(λn) is the inci- dence number[enα : enβ−1].
Proof. We leave the casen = 1 as an exercise. For n ≥2 the claim can be read off from the following commutative diagram of singular homology groups (∆’s omitted):
Hn(Bn, Sn−1) //
∂∗
Hn(enα,•enα) //
∂∗
Hn(Xn, Xn−1)
∂∗
Hn−1(Sn−1) //Hn−1(•enα) //Hn−1(Xn−1, Xn−2)
tt oo ∼=
γ Hn−1(en−1γ ,•en−1γ ) projection
uu
Hn−1(Xn−1, Xn−1−◦en−1β )
∼=
oo Hn−1(en−1β ,•en−1β )oo
∼=
Hn−1(Bn−1, Sn−2)
//Hn−1(Sn−1)
Hn−1(Xn−1/Xn−1−◦en−1β )oo Hn−1(en−1β /•en−1β )oo Hn−1(Bn−1/Sn−2).
55k
kk kk kk kk k
The point here is that, by our choices,λn→κn−1on the left andλn−1→
κn−1 on the right.
Similarly, one proves (exercise):
Proposition 2.5.19.The coefficient of ¯hβ∗(λn)in f#(¯hα∗(λn))is the map-
ping degree[enα : ˜enβ : f].
Exercises
1. LetX andY be oriented CW complexes. GiveX×Y the product orientation.
Prove
(a) [enαטemγ :enβ−1×e˜mγ] = [enα:enβ−1] (b) [enαטemγ :enα×e˜mδ−1] = (−1)n[˜emγ : ˜emδ−1] 2. Prove 2.5.19.
3. Prove then= 1 case of 2.5.18.