CW complexes

Soren Hansen
This note is meant to give a short introduction to CW complexes. 1. Notation and conventions In the following a space is a topological space and a map f : X Y between topological spaces X and Y is a function which is continuous. If X is a space, then a subspace of X is a subset A X with the relative (or induced) topology (induced by the topology in X ). The n-dimensional disk (just called the n-disk in the following) is the following subspace of Rn : Dn = { x Rn : |x| 1 }, where | | : Rn [0, [ is the standard norm on Rn . Thus the n-disk is the closed n-disk and is a closed subset of Rn . The open n-disk, denoted int(Dn ), is the interior of Dn in Rn . Thus int(Dn ) = { x Rn : |x| < 1 }. The boundary of Dn in Rn is the standard (n 1)-sphere S n1 = { x Rn : |x| = 1 }. We note that the 0-disk D0 is equal to R0 = {0} by denition. We have int(D0 ) = D0 = {0}. A topological space X is called quasi-compact if every open cover of X has a nite subcover, i.e. whenever {Ui }iI is a family of open subsets of X s.t. X = iI Ui then there exist i1 , . . . , in I s.t. X = n j =1 Uij . A topological space X is called compact if it is Hausdor and quasi-compact. As is standard we will write i for if and only if. 2. Cell decompositions and CW-complexes Denition 2.1. An n-cell is a space homeomorphic to the open n-disk int(Dn ). A cell is a space which is an n-cell for some n 0. Note that int(Dm ) and int(Dn ) are homeomorphic if and only if m = n. This e.g. follows by noting that int(Dn ) is homeomorphic to Rn (via the map x tan( |x|/2)x), and by the fact that Rm is homeomorphic to Rn i m = n, cf. [Ha, Theorem 2.26 p. 126]. Thus we can talk about the dimension of a cell. An n-cell will be said to have dimension n. Denition 2.2. A cell-decomposition of a space X is a family E = {e | I } of subspaces of X such that each e is a cell and X = I e (disjoint union of sets). The n-skeleton of X is the subspace Xn =
I :dim(e )n e .

Note that if E is a cell-decomposition of a space X , then the cells of E can have many dierent dimensions. E.g. one cell-decomposition of S 1 is given by E = {ea , eb }, where ea is an arbitrary point p S 1 and eb = S 1 \ {p}. Here ea is a 0-cell and eb is a 1-cell. There are no restrictions on the number of cells in a cell-decomposition. Thus we can have uncountable many cells in such a decomposition. E.g. any space X has a cell-decomposition where each point of X is a 0-cell. A nite cell-decomposition is a cell decomposition consisting of nitely many cells. Denition 2.3. A pair (X, E ) consisting of a Hausdor space X and a cell-decomposition E of X is called a CW-complex if the following 3 axioms are satised:
1

Axiom 1: (Characteristic Maps) For each n-cell e E there is a map e : Dn X restricting to a homeomorphism e |int(Dn ) : int(Dn ) e and taking S n1 into X n1 . Axiom 2: (Closure Finiteness) For any cell e E the closure e intersects only a nite number of other cells in E . Axiom 3: (Weak Topology) A subset A X is closed i A e is closed in X for each e E. Here e of course is the closure of e in X . Note that the Axioms 2 and 3 are only needed in case E is innite (i.e. they are automatically satised if E is nite). It is not dicult to give examples of pairs (X, E ) with X a Hausdor space and E an innite cell-decomposition of X such that Axiom 1 is satised and either Axiom 2 or Axiom 3 is satised, see e.g. [J, p. 97]. Thus Axiom 2 and 3 are independent of each other. Note that the characteristic map for a 0-cell e X is simply the map mapping 0 to the one-point space e. Lemma 2.4. Let (X, E ) be a Hausdor space X together with a cell-decomposition E . If (X, E ) satises Axiom 1 in Denition 2.3 then we have e = e (Dn ) for any cell e E . In particular e is a compact subspace of X and the cell boundary e \ e = e (S n1 ) lies in X n1 . Proof. For any map f : Y Z between topological spaces Y and Z and any subset B Y we ) f (B ), see e.g. [D, Theorem III.8.3 pp. 79-80] or [A, Theorem 2.9 p. 33]. Thus have f (B e = e (int(Dn )) e (Dn ) e. But e (Dn ) is compact hence closed in X since X is Hausdor. Thus e (Dn ) = e . By Axiom 1 n n 1 n 1 we have e (int(D )) = e and e (S ) e = so e (S )=e \ e. Note, if X is not Hausdor we still have e (Dn ) e but we dont necessarily have equality. We have no garantee that e (Dn ) is closed in X . 3. Subcomplexes Let (X, E ) be a CW-complex, E E a set of cells in it and X = eE e. Lemma 3.1. The following 3 conditions are equivalent: (a) The pair (X , E ) is a CW-complex. (b) The subset X is closed in X . (c) The closure e X for each e E , where e is the closure of e in X . For a proof, see [J, p. 98]. Denition 3.2. Let (X, E ) be a CW-complex and let (X , E ) be as above. Then (X , E ) is called a subcomplex (of (X, E )) if the 3 equivalent conditions (a), (b) and (c) in the above lemma are satised. We have some immediate consequences: Corollary 3.3. Let (X, E ) be a CW-complex. Then (1) Let {Ai | i I } be any family of subcomplexes of (X, E ). Then iI Ai and iI Ai are subcomplex of (X, E ). (2) The n-skeleton X n is a subcomplex of (X, E ) for each n 0.

(3) Let {ei |i I } be an arbitrary family of n-cells in E . Then X n1 (iI ei ) is a subcomplex. Proof. For (1) note that iI Ai is a subcomplex by characterization (c) in Lemma 3.1 and iI Ai is a subcomplex by (b) in that lemma. Both (2) and (3) follow by using characterization (c) in Lemma 3.1 together with Lemma 2.4. Note that (2) is a special case of (3). By the above we have that the n-skeleton X n of a CW-complex (X, E ) is a closed subset of X . 4. Identification Topology and Quotient Spaces In the next section we need a general proceedure for constructing new spaces from old spaces by gluing spaces together via maps. Let us in this section describe the general concepts from point set topology needed. 4.1. Identication Topology. Let X be a topological space and let Y be an arbitrary set and let p : X Y be a surjection. Then we can dene a topology in Y by: a subset U Y is open i p1 (U ) is open in X . This topology is the largest topology in Y for which p : X Y is continuous. We call it the identication topology in Y determined by p, and p : X Y is called an identication map. If X and Y are two spaces and p : X Y a surjective map, then p is called an identication map if the topology in Y is the identication topology determined by p. Lemma 4.1. Let X be a compact space and Y a Hausdor space and let p : X Y be a surjective map. Then p is an identication map. Proof. It is enough to prove that C Y is closed i p1 (C ) is closed. Since p is continuous we only have to prove that C is closed in Y if p1 (C ) is closed in X . But if p1 (C ) is closed in X then it is compact, since X is compact. Thus C = p(p1 (C )) is compact in the Hausdor space Y , hence C is closed in Y . Lemma 4.2. Let p : X Y be an identication map and let Z be a space. Then f : Y Z is continuous i f p : X Z is continuous. Proof left to the reader. 4.2. Quotient spaces. Let X be a set and let be an equivalence relation on X . Let X/ be the set of equivalence classes and let : X X/ be the canonical projection, i.e. the function mapping x to the equivalence class containing x. Recall here that the equivalence classes are mutually disjoint subsets of X and that X is the disjoint union of these equivalence classes. Oppositely, if we have given a disjoint family {Ai }iI of subsets of X covering X , i.e. X = iI Ai , then we can dene an equivalence relation on X by x y if and only if i I s.t. x, y Ai . The equivalence classes for that equivalence relation are nothing but our subsets Ai . Thus an equivalence relation in X is nothing but a partition of X into subsets. Now, given a space X and an equivalence relation we equip X/ with the identication topology determined by the canonical projection : X X/ . This topology is normally called the quotient topology and X/ is called a quotient space of X (the quotient of X by ). If p : X Y is an identication map, then we can identify Y with a quotient space. Namely, the subsets p1 (y ), y Y , gives a partition of X . Thus the equivalence relation in X induced by this partition is x x i p(x) = p(x ). We thus have a bijection q : X/ Y given by

q ( (x)) = p(x), where : X X/ is the canonical projection. By Lemma 4.2 both q and q 1 are continuous since p and are both identication maps and q = p and q 1 p = are continuous. Thus q : X/ Y is a homeomorphism so from a topological point of view we consider Y and X/ to be the same space. Thus quotient spaces and identication spaces are one and the same thing. There are many standard constructions in algebraic topology using the above ideas. In particular we mention: Collapsing a subspace to a point. Let X be a topological space and let A be some non-empty subspace. Then we let X/A = X/ , where the equivalence classes w.r.t. are A and the singletons {x}, x X \ A. The Wedge of Spaces. Given two spaces X and Y and chosen points x0 X and y0 Y we let X Y = (X Y )/{x0 , y0 }. That is, we collapse the subset {x0 , y0 } to a point. Here X Y is the topological disjoint union of X and Y (sometimes called the topological sum of X and Y ), see below. We note that the construction depends on the points x0 and y0 . However, dierent choices can lead to homeomorphic results. Thus e.g. S 1 S 1 does not depend on the choice of the points x0 and y0 . More generally we can talk about the wedge of a family of spaces {Xi }iI , namely if xi Xi we let iI Xi = iI Xi /A, where A = {xi }iI , and where iI Xi is the topological disjoint union of the spaces Xi . Here and in the following we use the following construction. Let {Xi }iI be a family of topological spaces. Then the topological disjoint union X = Xi is the following topological space. As a set it is equal to iI {i} Xi . We identify {i} Xi with Xi . The topology on X is given by: U is open in X i U Xi is open in Xi for all i I . Note that if fi : Xi Y , i I is a family of maps then we get a unique map f = iI fi : X Y s.t. f |Xi = fi for each i I . A function f : X Y is continuous i f |Xi : Xi Y is continuous for each i I . Attaching a space to another via map. Let X and Y be two spaces and let A be a subspace of X and f : A Y a map. Then X
f

Y := X

Y / ,

where X Y is given the disjoint union topology, see above, and the equivalence classes w.r.t. are the singletons {p}, p X \ A and p Y \ f (A), and the subsets {y } f 1 (y ), y f (A). We will only need this in situations where A is a closed subset of X . Lemma 4.3. Let the situation be as above with A a closed subset of X , let : X Y X f Y be the canonical projection and let i : Y X Y and j : X \ A X Y be the inclusion maps. Then i : Y X f Y is an embedding with image a closed subset of X f Y and j : X \ A X f Y is an embedding with image an open subset of X f Y . Exercise 4.4. Prove the above lemma. 5. CW-complexes considered as spaces obtained by attaching cells to each other We start by a small Lemma 5.1. Let (X, E ) be a CW-complex and let Y be any space and let f : X Y be a function. Then the following are equivalent. (1) f : X Y is continuous. (2) The restriction f |e Y is continuous for all e E . :e

(3) The restriction f |X n : X n Y is continuous for all n 0. Exercise 5.2. Prove the above lemma. Proposition 5.3. Let (X, E ) be a CW-complex. Then X n is obtained from X n1 by attaching of the n-cells in X . Proof. Let En be the n-cells in X and let for each e En , e : Dn X be the characteristic map of e. Since e (S n1 ) X n1 we can consider e = e |S n1 : S n1 X n1 . For each n = {e} D n and identify this with D n . (Here e is just some index.) We let e En , let De n n and = n1 . Thus | n = e . Z = eEn De , A = eEn De (Here eEn e : A X De n n 1 n 1 De = {e} S S .) We let = Y = Z X n1 and let n : Z X n1 Y be the canonical projection. We have a surjective map f =(
eEn e )

j:Z

X n1 X n ,

where j : X n1 X n is the inclusion. The proof is nalized by the following exercise. Exercise 5.4. Let the situation be as in the above proof. Prove the following facts: (1) There is a unique bijection : Y X n such that n = f . (2) The bijection is a homeomorphism. (Hint: To prove that 1 is continuous use Lemma 5.1 n1 and 1 | = | n : D n Y for each e E . and prove that 1 |e = n |e if e X e e n De n e n is an identication map.) Prove and use that e : De e (3) The composition n |X n1 : X n1 X n is the inclusion. We can now formulate an alterantive denition of CW-complexes. Proposition/Denition 5.5. A CW-complex is a space together with a ltration of subspaces = X 1 X 0 X 1 . . . X n . . . X such that n X n1 , (1) X n is obtained by attaching of n-cells to X n1 . That is, we have maps : D n n In , n 0, and homeomorphisms n : Y X such that n n |X n1 : X n1 X n is the inclusion, where Y n = Z n n X n1 ,
n and = n n1 , and where : Z n where Z n = In D n n In : In D X n 1 n X Y is the canonical projection. (2) X = n0 X n . (3) X carries the weak topology w.r.t. the family of spaces {X n }n0 . That is, a subset A X is closed i A X n is closed in X n for all n 0.

Thus by (1) we have (inductively) a unique topology on X n for each n 0 starting by noting 0 ), and by (3) the total space that X 0 necessarily have the discrete topology (since X 0 = I0 D X then has the weak topology w.r.t. these topological spaces. Note that we have not included in the denition above that X is Hausdor. That follows automatically as the proof will reveal. Proof. Exercise. See later version.

References
[A] [D] [Ha] [J] M. A. Armstrong, Basic Topology, Undergraduate Texts in Mathematics, SpringerVerlag (1983). J. Dugundji, Topology, Allyn and Bacon, Inc. (1966). A. Hatcher, Algebraic Topology, Cambridge University Press (2002). K. J anich, Topology, Undergraduate Texts in Mathematics, SpringerVerlag (1984).