Cellular homology and cellular approximation

Intuition [Beginner]

A CW complex comes with a canonical bookkeeping system: dimension by dimension, count the cells. The miracle of cellular homology is that this bookkeeping is enough to compute the homology of the space. You do not need to triangulate, you do not need to choose a singular chain, you do not need to evaluate any integrals. You list the cells, you record how each cell's boundary wraps around the cells one dimension below, and you read off the homology from the resulting count.

For a sphere $S^2$ presented with one 0-cell and one 2-cell, the count is two cells in dimensions zero and two, nothing in dimension one. The homology comes out to $\mathbb{Z}$ in dimension zero, $\mathbb{Z}$ in dimension two, $0$ in dimension one — and that matches the answer the singular theory works much harder to produce.

The reason this matters: most spaces studied in algebraic topology come naturally as CW complexes (spheres, projective spaces, Lie groups, classifying spaces), and the cellular method turns the homology computation into a finite combinatorial problem instead of an infinite-dimensional one. A bonus: every continuous map between CW complexes can be deformed by a homotopy until it respects the cell structure on both sides — the cellular approximation theorem — so cellular methods cover all of homotopy theory, not just the cellular maps.

Visual [Beginner]

A schematic of the cellular chain complex: a row of free abelian groups, one per dimension, generated by the cells of that dimension; arrows between consecutive groups labelled by the cellular boundary. For ${\mathbb{CP}}^2$, the diagram has $\mathbb{Z}$ in dimensions zero, two, four, and zeros in odd dimensions; every arrow is the zero map.

The picture captures the entire computation. Counting cells per dimension gives the chain groups; tracking how each cell wraps around the cells below gives the boundary; reading off kernels modulo images gives the homology.

Worked example [Beginner]

Compute the homology of complex projective space ${\mathbb{CP}}^n$ using the cellular method.

Step 1. Recall the minimal CW structure on ${\mathbb{CP}}^n$: one cell in each even dimension $0,2,4,\dots ,2n$, and no cells in odd dimensions.

Step 2. Write down the cellular chain group in each dimension. In even dimension $2k$ with $0\le k\le n$, there is one cell, so the cellular chain group is $\mathbb{Z}$. In odd dimensions and in dimensions above $2n$, there are no cells, so the group is $0$.

Step 3. Determine the cellular boundary in each dimension. The boundary goes from the chain group in dimension $m$ to the chain group in dimension $m-1$. For ${\mathbb{CP}}^n$, every chain group sits in even dimension and every neighbouring group (one dimension below) sits in odd dimension and is therefore zero. So every cellular boundary is the zero map.

Step 4. Read off the homology. With every boundary zero, kernels equal the whole chain group and images equal $0$. So $H_k({\mathbb{CP}}^n)=\mathbb{Z}$ for $k=0,2,4,\dots ,2n$, and $H_k({\mathbb{CP}}^n)=0$ otherwise.

What this tells us: the homology of ${\mathbb{CP}}^n$ is $\mathbb{Z}$ in every even dimension up to $2n$ and zero elsewhere. The entire computation took four steps and never required a single chain integral. Two cells per even dimension, no cells per odd dimension, all boundaries zero — the ring structure of ${\mathbb{CP}}^n$'s homology pops out of pure cell-counting.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a CW complex with skeleton filtration $X^{(-1)}=\varnothing \subseteq X^{(0)}\subseteq X^{(1)}\subseteq \cdots$ and let $J_n$ index the $n$-cells of $X$. The cellular chain complex $(C_{\ast }^{\mathrm{cell}}(X),d_{\ast })$ is defined by

$$C_n^{\mathrm{cell}}(X)=H_n(X^{(n)},X^{(n-1)})$$

where the right-hand side is relative singular homology of consecutive skeleta. As a free abelian group, $C_n^{\mathrm{cell}}(X)$ has one generator $e_{\alpha }^n$ per $n$-cell $\alpha \in J_n$; this is the Hatcher §2.2 lemma identifying $H_n(X^{(n)},X^{(n-1)})$ with the free abelian group on $n$-cells via the characteristic-map fundamental classes $({\Phi }_{\alpha }{)}_{\ast }[D^n,S^{n-1}]$.

The cellular boundary $d_n:C_n^{\mathrm{cell}}(X)\to C_{n-1}^{\mathrm{cell}}(X)$ is defined as the composite

$$H_n(X^{(n)},X^{(n-1)})\stackrel{\partial }{\to }{H}_{n-1}\left(X^{(n-1)}\right)\stackrel{j_{\ast }}{\to }{H}_{n-1}(X^{(n-1)},X^{(n-2)})$$

where $\partial$ is the connecting homomorphism in the long exact sequence of the pair $(X^{(n)},X^{(n-1)})$ and $j_{\ast }$ is induced by the inclusion of pairs $(X^{(n-1)},\varnothing )\hookrightarrow (X^{(n-1)},X^{(n-2)})$. That $d_n\circ d_{n+1}=0$ follows because $j_{\ast }$ factors through $H_{n-1}(X^{(n-1)})$ and the composition of two consecutive connecting maps in adjacent long exact sequences vanishes.

Degree formula. With respect to the cell-basis $\{e_{\alpha }^n\}$, the cellular boundary is given explicitly by

$$d_n(e_{\alpha }^n)=\sum _{\beta \in J_{n-1}}\mathrm{deg}({\varphi }_{\alpha \beta })\cdot e_{\beta }^{n-1}$$

where ${\varphi }_{\alpha \beta }$ is the composition

$$S^{n-1}\stackrel{{\varphi }_{\alpha }}{\to }{X}^{(n-1)}\stackrel{q}{\to }{X}^{(n-1)}/X^{(n-2)}=\underset{\beta \in J_{n-1}}{\bigvee }{S}_{\beta }^{n-1}\stackrel{p_{\beta }}{\to }{S}_{\beta }^{n-1}$$

with ${\varphi }_{\alpha }$ the attaching map of the cell $e_{\alpha }^n$, $q$ the quotient by the $(n-2)$-skeleton (which collapses each $(n-1)$-cell to a sphere), and $p_{\beta }$ the projection to the $\beta$-th wedge factor. The integer $\mathrm{deg}({\varphi }_{\alpha \beta })$ is the topological degree of this map $S^{n-1}\to S^{n-1}$.

The cellular homology of $X$ is

$$H_n^{\mathrm{cell}}(X)=\ker (d_n)/\mathrm{image}(d_{n+1}).$$

Sign and orientation conventions. Each cell carries a chosen characteristic map ${\Phi }_{\alpha }:D^n\to X^{(n)}$, which fixes an orientation on the cell. The degree $\mathrm{deg}({\varphi }_{\alpha \beta })$ is computed with respect to the orientations on source and target spheres induced by these characteristic-map orientations.

Counterexamples to common slips

• The cellular chain complex is not a sub-complex of the singular chain complex; the relation between the two is the agreement theorem below, not a direct inclusion.
• The boundary $d_n$ is not the singular boundary $\partial$ restricted to cellular chains. It is the composition of $\partial$ with the projection $j_{\ast }$ to the next relative homology group.
• The degree formula requires an oriented choice of characteristic map for each cell. Reversing the orientation flips the sign of every degree involving that cell.
• An infinite-dimensional CW complex has a cellular chain complex of unbounded length; its cellular homology is computed via the same formula and still equals singular homology, but the agreement theorem requires the weak (colimit) topology to handle the inductive limit properly.

Key theorem with proof [Intermediate+]

Theorem (cellular = singular). Let $X$ be a CW complex. Then there is a natural isomorphism $H_n^{\mathrm{cell}}(X)\cong H_n(X)$ between cellular and singular homology, for every $n\ge 0$.

Proof. Fix $n\ge 0$. The strategy is to compute $H_n(X)$ by climbing the skeleton filtration, using the long exact sequences of the pairs $(X^{(k)},X^{(k-1)})$ for $k$ near $n$, and to identify the resulting expression with $\ker (d_n)/\mathrm{image}(d_{n+1})$.

Step 1 (singular vanishing on skeleta). For $k>n$, the relative homology $H_n(X^{(k)},X^{(k-1)})=0$, because attaching $k$-cells changes only $H_k$ and $H_{k-1}$. Specifically, $(X^{(k)},X^{(k-1)})$ is a good pair, so $H_{\ast }(X^{(k)},X^{(k-1)})\cong {\tilde{H}}_{\ast }(X^{(k)}/X^{(k-1)})$, and $X^{(k)}/X^{(k-1)}={\bigvee }_{\alpha }{S}^k$ has reduced homology ${\mathbb{Z}}^{|J_k|}$ concentrated in degree $k$. So $H_n(X^{(k)},X^{(k-1)})=0$ for $k\ne n$.

Step 2 (singular homology stabilises beyond dimension $n$). From the long exact sequence of $(X^{(k)},X^{(k-1)})$ in degrees near $n$,

$$H_{n+1}(X^{(k)},X^{(k-1)})\to H_n(X^{(k-1)})\to H_n(X^{(k)})\to H_n(X^{(k)},X^{(k-1)}),$$

both end terms vanish for $k\ne n,n+1$ by Step 1. So for $k\ge n+2$, the inclusion-induced map $H_n(X^{(k-1)})\to H_n(X^{(k)})$ is an isomorphism, and the colimit $H_n(X)={\mathrm{colim}}_k{H}_n(X^{(k)})$ stabilises: $H_n(X)=H_n(X^{(n+1)})=H_n(X^{(k)})$ for any $k\ge n+1$.

Step 3 (extract the cellular kernel). From the long exact sequence of $(X^{(n)},X^{(n-1)})$,

$$H_n(X^{(n-1)})\to H_n(X^{(n)})\stackrel{j_{\ast }}{\to }{H}_n(X^{(n)},X^{(n-1)})\stackrel{\partial }{\to }{H}_{n-1}(X^{(n-1)}).$$

The first term $H_n(X^{(n-1)})=0$ because $X^{(n-1)}$ has cells only in dimensions $\le n-1$ (apply Step 1 inductively to the filtration $X^{(0)}\subseteq \cdots \subseteq X^{(n-1)}$ and use that $H_n$ is preserved by attaching cells of dimension $>n$). So $j_{\ast }$ is injective, giving $H_n(X^{(n)})\cong \ker (\partial :H_n(X^{(n)},X^{(n-1)})\to H_{n-1}(X^{(n-1)}))$.

Step 4 (extract the cellular image). From the long exact sequence of $(X^{(n+1)},X^{(n)})$,

$$H_{n+1}(X^{(n+1)},X^{(n)})\stackrel{\partial }{\to }{H}_n(X^{(n)})\to H_n(X^{(n+1)})\to H_n(X^{(n+1)},X^{(n)}).$$

The last term is $0$ by Step 1. So $H_n(X^{(n+1)})\cong H_n(X^{(n)})/\mathrm{image}(\partial :H_{n+1}(X^{(n+1)},X^{(n)})\to H_n(X^{(n)}))$.

Step 5 (assemble). Combining Steps 2–4,

$$H_n(X)=H_n(X^{(n+1)})=\frac{\ker (\partial :H_n(X^{(n)},X^{(n-1)})\to H_{n-1}(X^{(n-1)}))}{\mathrm{image}(\partial :H_{n+1}(X^{(n+1)},X^{(n)})\to H_n(X^{(n)}))}.$$

The bottom $\mathrm{image}$ has its target inside $\ker$ (the image lands in $\ker ({\partial }^{\prime })$ where ${\partial }^{\prime }$ is the connecting map from $H_n(X^{(n)})$ down — cellular two-step boundary).

Now identify the right-hand side with cellular homology. The numerator's denominator $H_{n-1}(X^{(n-1)})$ is replaced by its quotient $H_{n-1}(X^{(n-1)},X^{(n-2)})=C_{n-1}^{\mathrm{cell}}(X)$ via $j_{\ast }$, which is injective on the relevant subgroups (using Step 3 applied to dimension $n-1$). So the kernel of $\partial$ in the singular long exact sequence becomes the kernel of the cellular boundary $d_n=j_{\ast }\partial$. The image of $\partial$ in dimension $n+1$ becomes the image of the cellular boundary $d_{n+1}$ via the same identification. Therefore

$$H_n(X)=\ker (d_n)/\mathrm{image}(d_{n+1})=H_n^{\mathrm{cell}}(X).$$

Naturality in $X$ follows because every step uses functorial constructions (the long exact sequence is functorial in the pair, the colimit is functorial in the filtered diagram). $\square$

Bridge. This builds toward the cellular approximation theorem stated in the next section: the agreement of cellular and singular homology is what makes the cellular boundary formula a genuine computational tool, and the same skeleton-by-skeleton argument appears again in 03.12.05 (Eilenberg-MacLane spaces), where the inductive cell-attachment construction relies on cellular homology to verify the homotopy groups at each stage. The foundational reason cellular homology works is exactly the long-exact-sequence pattern of the pair $(X^{(n)},X^{(n-1)})$: the relative homology of consecutive skeleta isolates a single dimension, which generalises to an instance of a more general organising idea — that filtrations on a space induce spectral sequences computing the homology of the colimit. Putting these together, the bridge is the recognition that cellular homology is the spectral sequence of the skeleton filtration, collapsed at $E^2$ because the relative groups are concentrated in a single row; the cellular chain complex is precisely the $E^1$ page, and the cellular boundary is the $d^1$ differential. This is exactly the same as the Atiyah-Hirzebruch spectral sequence specialised to ordinary homology and CW filtrations.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not yet ship a dedicated cellular-homology API. The closest existing infrastructure is the chain-complex framework in Mathlib.Algebra.Homology.HomologicalComplex and the singular-homology functor as an unstable construction. A schematic statement of the cellular chain complex would read:

[object Promise]

Compiling this requires the CWComplex structure (gap from 03.12.10) and a definition of singular homology as a Mathlib-citizen functor. The lean_mathlib_gap field in the frontmatter records the concrete formalisation roadmap.

Advanced results [Master]

Theorem (cellular approximation; Hatcher 4.8). Every continuous map $f:X\to Y$ between CW complexes is homotopic to a cellular map — a map sending $X^{(n)}$ into $Y^{(n)}$ for every $n$. Moreover, if $f$ is already cellular on a subcomplex $A\subseteq X$, the homotopy can be chosen to fix $f{|}_A$.

The proof proceeds by induction on the skeleta of $X$, using the homotopy extension property (HEP) of CW pairs to extend each successive deformation. The technical core is a transversality argument: any continuous map of a compact $n$-dimensional space into a CW complex meets only finitely many cells, and any meeting with a cell of dimension strictly greater than $n$ can be deformed off that cell because the image is too low-dimensional to fill it. Iterating dimension by dimension, with HEP gluing the per-cell deformations into a global homotopy, produces the cellular map.

Corollary (homotopy groups of spheres in low dimensions). For $i<n$, ${\pi }_i(S^n)=0$. Apply cellular approximation to a representative map $f:S^i\to S^n$. Give $S^i$ the minimal CW structure (one 0-cell, one $i$-cell) and $S^n$ the minimal CW structure (one 0-cell, one $n$-cell). The cellular approximation sends $S^i$'s $i$-skeleton (which is all of $S^i$) into $S^n$'s $i$-skeleton, which is just the 0-cell because $i<n$. So $f$ is homotopic to a constant map.

Theorem (homotopy invariance of cellular homology). Cellular homology is a homotopy-invariant functor on CW complexes. Combining cellular = singular with the homotopy invariance of singular homology gives the result; alternatively, use cellular approximation to show that homotopic maps induce chain-homotopic cellular chain maps.

Computational consequences.

• Spheres. $H_{\ast }(S^n)=\mathbb{Z}$ in dimensions $0$ and $n$, $0$ elsewhere.
• Real projective space. $H_{\ast }({\mathbb{RP}}^n;\mathbb{Z})$ has $\mathbb{Z}/2$ summands in odd dimensions $<n$ and a $\mathbb{Z}$ in dimension $n$ when $n$ is odd. With $\mathbb{Z}/2$ coefficients, $H_{\ast }({\mathbb{RP}}^n;\mathbb{Z}/2)=\mathbb{Z}/2$ in every dimension $0,1,\dots ,n$ (all boundaries become zero mod 2).
• Complex projective space. $H_{\ast }({\mathbb{CP}}^n)=\mathbb{Z}$ in even dimensions $0,2,\dots ,2n$, $0$ elsewhere; the cup-product structure is $H^{\ast }({\mathbb{CP}}^n)=\mathbb{Z}[x]/(x^{n+1})$ with $|x|=2$.
• Surfaces. Closed orientable surfaces of genus $g$: $H_{\ast }({\Sigma }_g)=(\mathbb{Z},{\mathbb{Z}}^{2g},\mathbb{Z})$. Closed non-orientable surfaces $N_k$ of non-orientable genus $k$: $H_{\ast }(N_k)=(\mathbb{Z},{\mathbb{Z}}^{k-1}\oplus \mathbb{Z}/2,0)$.
• Lens spaces. $L(p,q)=S^3/{\mathbb{Z}}_p$ has cellular structure with one cell in each dimension $0,1,2,3$ and homology $(\mathbb{Z},\mathbb{Z}/p,0,\mathbb{Z})$.
• Lie groups. Compact Lie groups $G$ admit CW structures (via Bott's Morse theory or Schubert decomposition of $G/T$). The cellular complex computes $H_{\ast }(G)$ and exhibits the well-known result that $H_{\ast }(U(n))$ is an exterior algebra on odd-degree generators.

Computational advantage over singular. For a finite CW complex with $c_n$ cells in dimension $n$, the cellular chain complex has $C_n^{\mathrm{cell}}={\mathbb{Z}}^{c_n}$ — finitely generated, often very small. The singular chain complex by contrast has uncountable rank in every dimension. Cellular homology turns the homological algebra of $X$ into a finite computation; singular homology turns it into a colimit of infinite-dimensional vector spaces. The agreement theorem says these compute the same answer, and cellular is by orders of magnitude the better road for explicit calculation.

Synthesis. Cellular homology identifies the homology of a CW complex with the homology of a finite, explicit chain complex generated by cells. The foundational reason this works is the long-exact-sequence pattern of consecutive skeleta: the relative groups $H_n(X^{(n)},X^{(n-1)})$ are concentrated in a single row, so the spectral sequence of the skeleton filtration collapses at $E^2$ and the cellular chain complex is exactly the $E^1$ page. The cellular boundary is the $d^1$ differential, computed by the degree formula. This is exactly the same as a special case of the Atiyah-Hirzebruch spectral sequence — for ordinary homology and CW filtrations, the spectral-sequence machinery degenerates to cellular homology — and generalises to a generalised cohomology theory where the $E^1$ page is the cellular chain complex with coefficients in the coefficient ring of the theory.

Putting these together, the bridge is the recognition that cellular methods are not a separate computational gadget but the simplest case of a general filtration-spectral-sequence-converges-to-the-colimit pattern. The same colimit-cell pattern appears again in the cellular cochain complex (which computes singular cohomology by the same agreement theorem), in cellular cohomology with local coefficients (which couples the cell structure to a representation of ${\pi }_1$), and in the spectral sequence of a fibration (where the $E^1$ page reads off cells of the base with coefficients in the cohomology of the fibre). The cellular approximation theorem is what guarantees that the cellular methods cover all of homotopy theory: every map between CW complexes is homotopic to one that respects the cell structure, so every homological invariant of every continuous map can be computed cellularly. This is the same identification used in 03.12.05 (Eilenberg-MacLane spaces) to verify the homotopy groups of cellular constructions, and recurs in the postnikov-tower construction of every CW complex as an inverse limit of cellularly-attached fibrations.

Full proof set [Master]

Theorem (cellular approximation), full proof. Let $f:X\to Y$ be continuous between CW complexes. Inductively assume $f$ has been homotoped so that $f{|}_{X^{(n-1)}}$ lands in $Y^{(n-1)}$. Consider an $n$-cell $e_{\alpha }^n$ of $X$ with characteristic map ${\Phi }_{\alpha }:D^n\to X^{(n)}$. The image $f({\Phi }_{\alpha }(D^n))$ is compact in $Y$; by the standard CW result that compact subsets of CW complexes meet only finitely many open cells, this image meets only finitely many cells of $Y$.

If any of those cells has dimension $>n$, deform $f$ off the cell as follows. Pick a point $y_0$ in the open cell that is not in $f({\Phi }_{\alpha }(\partial D^n))$ — such a point exists because the boundary image is contained in $Y^{(n-1)}$ by inductive hypothesis, while the open cell is disjoint from $Y^{(n-1)}$. The image of the small ball $B\subseteq D^n$ around the preimage of $y_0$ has dimension $\le n<\dim (\text{cell})$, so a generic-position argument (smooth approximation if $f$ is smooth, simplicial approximation if it is piecewise-linear, or the polyhedral neighborhood argument in the general continuous case — see Hatcher Lemma 4.10) deforms $f$ to miss $y_0$ and then radially deforms within the cell to push the entire image off the open cell. This homotopy is supported on a small ball in $D^n$ that is mapped to a small neighbourhood of $y_0$ in the cell, so it leaves $f$ unchanged outside.

Iterate over the finitely many higher-dimensional cells meeting $f({\Phi }_{\alpha }(D^n))$ to deform $f\circ {\Phi }_{\alpha }$ entirely into $Y^{(n)}$. Use the homotopy extension property of the CW pair $(X^{(n)},X^{(n-1)}\cup {\Phi }_{\alpha }(\partial D^n))$ to extend the per-cell deformation back to a homotopy of $f$ on $X^{(n)}$ that fixes $f{|}_{X^{(n-1)}}$. Iterate over $\alpha$, then increment $n$. The colimit structure of $X$ ensures the inductive homotopies assemble into a global homotopy on all of $X$. The result is a cellular map homotopic to $f$. $\square$

Theorem (cellular = singular), proof of naturality and chain-map property. The construction in the Intermediate-tier proof gives an isomorphism on homology objects. To verify it comes from a chain map $C_{\ast }^{\mathrm{cell}}(X)\to C_{\ast }^{\mathrm{sing}}(X)/\partial$, observe that the diagram

$$\begin{array}{ccc}{H}_n(X^{(n)},X^{(n-1)})& \stackrel{\partial }{\to }& H_{n-1}(X^{(n-1)})\\ \downarrow j_{\ast }& & \downarrow j_{\ast }\\ H_n(X^{(n)},X^{(n-1)})& \stackrel{d_n}{\to }& H_{n-1}(X^{(n-1)},X^{(n-2)})\end{array}$$

commutes by definition of $d_n=j_{\ast }\partial$. Functoriality in $X$ follows because every long exact sequence is functorial in pairs and every cell index $J_n$ is functorial under cellular maps.

Theorem (degree formula for the cellular boundary), proof. Let $e_{\alpha }^n$ be an $n$-cell with attaching map ${\varphi }_{\alpha }:S^{n-1}\to X^{(n-1)}$. Under the identification $C_n^{\mathrm{cell}}(X)=H_n(X^{(n)},X^{(n-1)})$, the cell corresponds to $({\Phi }_{\alpha }{)}_{\ast }[D^n,S^{n-1}]$, the pushforward of the relative fundamental class.

The connecting map $\partial$ sends $({\Phi }_{\alpha }{)}_{\ast }[D^n,S^{n-1}]$ to $({\varphi }_{\alpha }{)}_{\ast }[S^{n-1}]\in H_{n-1}(X^{(n-1)})$. Then $j_{\ast }$ projects this to the relative group $H_{n-1}(X^{(n-1)},X^{(n-2)})={⨁}_{\beta }\mathbb{Z}⟨e_{\beta }^{n-1}⟩$, which under good-pair isomorphism is ${\tilde{H}}_{n-1}(X^{(n-1)}/X^{(n-2)})={\tilde{H}}_{n-1}({\bigvee }_{\beta }{S}^{n-1})$.

Composing ${\varphi }_{\alpha }$ with the quotient $q:X^{(n-1)}\to X^{(n-1)}/X^{(n-2)}={\bigvee }_{\beta }{S}_{\beta }^{n-1}$ gives a map $S^{n-1}\to {\bigvee }_{\beta }{S}_{\beta }^{n-1}$. Apply ${\tilde{H}}_{n-1}$ and use the wedge decomposition to write ${\tilde{H}}_{n-1}(\bigvee S_{\beta }^{n-1})={⨁}_{\beta }\mathbb{Z}⟨[S_{\beta }^{n-1}]⟩$. The image of the fundamental class $[S^{n-1}]$ under this composition decomposes coordinate-wise as ${\sum }_{\beta }\mathrm{deg}({\varphi }_{\alpha \beta })\cdot [S_{\beta }^{n-1}]$ where ${\varphi }_{\alpha \beta }=p_{\beta }\circ q\circ {\varphi }_{\alpha }$ is the projection to the $\beta$-th wedge factor and $\mathrm{deg}$ is the topological degree. So $d_n(e_{\alpha }^n)={\sum }_{\beta }\mathrm{deg}({\varphi }_{\alpha \beta })\cdot e_{\beta }^{n-1}$. $\square$

Stated without proof — see Hatcher §2.2 and §4.1 ^{[Hatcher §2.2 and §4.1]} and Massey Ch. IX ^{[Massey Ch. IX]}. The cellular cochain complex (dual construction computing singular cohomology); cellular homology with local coefficients (twisted by a ${\pi }_1$-representation); the relationship between cellular homology and the spectral sequence of a CW filtration (collapses at $E^2$); the cellular Mayer-Vietoris sequence for a CW complex written as a union of two subcomplexes.

Connections [Master]

• Eilenberg-MacLane spaces 03.12.05. The construction of $K(\pi ,n)$ proceeds by attaching cells dimension by dimension to kill all homotopy groups except ${\pi }_n=\pi$; cellular homology is exactly the tool used to verify that the resulting space has the desired homology and homotopy. The cellular chain complex of $K(\mathbb{Z},n)$ in low dimensions reads off the integer Eilenberg-MacLane invariants.

• Whitehead tower 03.12.07. The successive killing of low homotopy groups in the Whitehead tower is performed by attaching cells, and cellular homology is what verifies that the homology of each tower stage tracks the prescribed homotopy filtration. Combined with the Hurewicz theorem, cellular homology turns the abstract tower construction into a finite computation per stage.

• Singular homology 03.12.11. The agreement theorem identifies cellular and singular homology; this is the source of cellular homology's power. The proof of the theorem turns the singular long exact sequence of consecutive skeleta into the cellular boundary by a connecting-map argument.

• CW complex 03.12.10. The cellular chain complex is the canonical algebraic invariant of the cellular structure; the existence of a CW structure on a space is what makes cellular methods applicable. Whitehead's theorem (a weak homotopy equivalence between CW complexes is a homotopy equivalence) closes the loop: cellular invariants detect homotopy type for CW complexes.

• Spectral sequences (planned future unit). Cellular homology is the simplest case of the spectral sequence of a filtration: the $E^1$ page is the cellular chain complex, the $d^1$ differential is the cellular boundary, the spectral sequence collapses at $E^2$ for ordinary homology, and the abutment is singular homology. For generalised cohomology theories, the same filtration produces the Atiyah-Hirzebruch spectral sequence, which does not collapse in general.

• Postnikov towers (within 03.12.05). Every CW complex is the inverse limit of its Postnikov sections, each obtained by attaching cells to kill higher homotopy groups; cellular homology of the Postnikov stages is the bookkeeping tool that controls obstruction theory and the $k$-invariant classification.

Historical & philosophical context [Master]

J. H. C. Whitehead developed cellular methods in his 1949 paper Combinatorial homotopy II (Bull. AMS 55, 453-496) ^{[Whitehead 1949 II]} as the algebraic companion to the CW-complex theory introduced in Combinatorial homotopy I. The cellular chain complex appears there in essentially modern form, with the boundary computed via attaching-map degrees. The motivation was computational: simplicial methods required at least $n+2$ vertices to triangulate $S^n$, and the resulting chain complex for any reasonable space was unmanageably large. Whitehead's cellular reformulation gave $S^n$ a chain complex with two non-zero entries, ${\mathbb{CP}}^n$ a chain complex with $n+1$ entries, and Lie groups chain complexes whose entries matched the Schubert decomposition of the corresponding flag varieties.

The agreement of cellular and singular homology was made precise in the Eilenberg-Steenrod axiomatic framework: Eilenberg and Steenrod 1952 Foundations of Algebraic Topology ^{[Eilenberg-Steenrod]} proved the uniqueness theorem stating that any homology theory satisfying their axioms (homotopy, exactness, excision, dimension) is determined on CW pairs by its values on a point. Cellular homology satisfies the axioms; singular homology satisfies the axioms; they agree on a point; therefore they agree on all CW pairs. The uniqueness theorem retroactively justified Whitehead's cellular methods as a computational presentation of the same invariant produced abstractly by the axioms.

The cellular approximation theorem was proved in essentially the same period; Hatcher attributes the modern proof to Whitehead, with the technical refinements (PL transversality, smooth approximation) developed by Whitney and others through the 1950s. The combination — cellular homology to compute, cellular approximation to handle morphisms, agreement to identify cellular with the abstract axiomatic theory — became the standard computational toolkit of algebraic topology in the era after Eilenberg-Steenrod, and remains the standard toolkit today.

Bibliography [Master]

[object Promise]