Simplicial and $\Delta$-complex homology

Intuition [Beginner]

A space built by gluing triangles, tetrahedra, and their higher-dimensional cousins is called a $\Delta$-complex. The pieces are standard simplices: a 0-simplex is a point, a 1-simplex is a line segment with two labelled endpoints, a 2-simplex is a triangle with three labelled vertices, a 3-simplex is a solid tetrahedron. You glue them together by identifying entire faces, and you are allowed to identify two faces of the same simplex (so a single 2-simplex can become a torus by gluing each pair of opposite edges).

Once a space is presented this way, you can count things. Each simplex contributes a generator to a chain group; the boundary of an $n$-simplex is the alternating sum of its $(n-1)$-dimensional faces. Simplicial homology counts the cycles modulo boundaries: how many independent loops, how many independent surfaces, how many independent solid regions, and so on.

The point is that this counting is finite and combinatorial. A torus, a Klein bottle, a genus-$g$ surface, a real projective space all become matrices of integers. You compute by row-reducing.

Visual [Beginner]

A schematic showing the standard simplices ${\Delta }^0,{\Delta }^1,{\Delta }^2,{\Delta }^3$ side by side, then a 2-simplex with its three oriented edges and the formula reminding the reader that the boundary of the triangle is "edge 1 minus edge 2 plus edge 3", and finally a $\Delta$-complex picture of the torus as a single square (two triangles) with opposite edges identified.

The two triangles share a diagonal. Opposite edges of the square get identified to form the torus. The combinatorial data is: one vertex, three edges, two triangles.

Worked example [Beginner]

Compute the simplicial homology of the circle $S^1$ presented as one vertex $v$ and one edge $e$ with both endpoints attached to $v$.

Step 1. List the simplices. There is one 0-simplex $v$ and one 1-simplex $e$. So $C_0=\mathbb{Z}⟨v⟩$ and $C_1=\mathbb{Z}⟨e⟩$. All higher chain groups are zero.

Step 2. Compute the boundary of $e$. The two endpoints of $e$ both go to $v$. The boundary is "endpoint 2 minus endpoint 1", which is $v-v=0$.

Step 3. Read off homology. Since the boundary of $e$ is zero, every chain in $C_1$ is a cycle and there are no boundaries (no 2-simplices to take boundaries of), so $H_1=\mathbb{Z}⟨e⟩=\mathbb{Z}$. In dimension 0, the cycles are all of $C_0=\mathbb{Z}⟨v⟩$ and again there are no boundaries from above, so $H_0=\mathbb{Z}$. In all other dimensions the homology is 0.

What this tells us: the circle has one connected component (recorded by $H_0=\mathbb{Z}$) and one independent loop (recorded by $H_1=\mathbb{Z}$). Two cells, one matrix calculation, the entire homology of the circle.

Check your understanding [Beginner]

Formal definition [Intermediate+]

A $\Delta$-complex structure on a topological space $X$ is a collection of continuous maps ${\sigma }_{\alpha }:{\Delta }^{n_{\alpha }}\to X$, indexed by $\alpha$ in some set $J$, where ${\Delta }^n=\{(t_0,\dots ,t_n)\in {\mathbb{R}}^{n+1}:t_i\ge 0,\sum t_i=1\}$ is the standard $n$-simplex, such that:

1. The restriction ${\sigma }_{\alpha }{|}_{{\mathring{\Delta }}^{n_{\alpha }}}$ to the open simplex is injective, and every point of $X$ lies in the image of exactly one such open simplex.
2. Each restriction of ${\sigma }_{\alpha }$ to a codimension-1 face ${\Delta }^{n_{\alpha }-1}\hookrightarrow {\Delta }^{n_{\alpha }}$ equals some ${\sigma }_{\beta }$ (where $\beta$ ranges over the indices of $(n_{\alpha }-1)$-simplices). The face inclusions are the standard ones $[v_0,\dots ,\widehat{v_i},\dots ,v_n]\hookrightarrow [v_0,\dots ,v_n]$.
3. A subset $A\subseteq X$ is open iff ${\sigma }_{\alpha }^{-1}(A)$ is open in ${\Delta }^{n_{\alpha }}$ for every $\alpha$.

The maps ${\sigma }_{\alpha }$ are the characteristic maps of the $\Delta$-structure. A classical simplicial complex is the special case in which each characteristic map is also injective on the closed simplex and any two simplices intersect in a common face.

Given a $\Delta$-complex structure, the simplicial chain complex is

$$C_n^{\Delta }(X)=\underset{\alpha :n_{\alpha }=n}{⨁}\mathbb{Z}⟨{\sigma }_{\alpha }⟩={\mathbb{Z}}^{|J_n|},$$

the free abelian group on the $n$-simplices of the structure. The boundary ${\partial }_n:C_n^{\Delta }(X)\to C_{n-1}^{\Delta }(X)$ is defined on a generator ${\sigma }_{\alpha }$ (with $n_{\alpha }=n$) as the alternating sum of face restrictions:

$${\partial }_n{\sigma }_{\alpha }=\sum _{i=0}^n(-1{)}^i{\sigma }_{\alpha }{|}_{[v_0,\dots ,\widehat{v_i},\dots ,v_n]}.$$

By condition (2), each face restriction is itself one of the indexed $(n-1)$-simplices, so the right-hand side lives in $C_{n-1}^{\Delta }(X)$. A direct computation, identical to the singular case, gives ${\partial }_{n-1}\circ {\partial }_n=0$. The simplicial homology of $X$ is

$$H_n^{\Delta }(X)=\ker ({\partial }_n)/\mathrm{im}({\partial }_{n+1}).$$

The singular chain complex $C_n(X)$ is the free abelian group on all continuous maps $\sigma :{\Delta }^n\to X$, with the same alternating-face boundary. Each simplicial generator ${\sigma }_{\alpha }$ is in particular a singular simplex, so there is an inclusion of chain complexes

$${\iota }_{\ast }:C_{\ast }^{\Delta }(X)\hookrightarrow C_{\ast }(X).$$

This ${\iota }_{\ast }$ commutes with boundaries because the simplicial boundary is the singular boundary applied to a simplicial generator, and by definition each face is again a simplicial generator. Hence ${\iota }_{\ast }$ induces a homomorphism on homology,

$${\iota }_{\ast }:H_n^{\Delta }(X)\to H_n(X).$$

Counterexamples to common slips

• The boundary formula $\partial \sigma =\sum (-1{)}^i\sigma {|}_{i\text{-th face}}$ depends on the vertex order. A $\Delta$-structure is the data needed to make this choice canonical: each ${\sigma }_{\alpha }$ has its vertices labelled by $0,1,\dots ,n$, and reversing the order changes signs.
• A subset of $X$ that is a union of open simplices is not automatically a $\Delta$-subcomplex: it must be closed under taking faces. Closure under faces is what makes the chain complex of the subspace sit inside the chain complex of the whole space.
• The simplicial chain group $C_n^{\Delta }(X)$ is countable (or finite, for finite $\Delta$-complexes); the singular chain group $C_n(X)$ is uncountable. The comparison map ${\iota }_{\ast }$ injects a countable subgroup into an uncountable one.

Key theorem with proof [Intermediate+]

Theorem (simplicial homology equals singular homology). Let $X$ be a $\Delta$-complex and let $A\subseteq X$ be a $\Delta$-subcomplex. Then the inclusion $\iota_ : C_*^\Delta(X, A) \to C_*(X, A)$ of chain complexes induces an isomorphism on homology in every degree:*

$${\iota }_{\ast }:H_n^{\Delta }(X,A)\stackrel{\cong }{\to }{H}_n(X,A).$$

(Hatcher Theorem 2.27.)

Proof. Reduce to the absolute case $A=\varnothing$ by a five-lemma argument applied to the long exact sequences of the pairs (the simplicial and singular long exact sequences are natural in the inclusion of chain complexes), so it suffices to prove the absolute statement.

Filter $X$ by its skeleta $X^k={\bigcup }_{n_{\alpha }\le k}\mathrm{im}({\sigma }_{\alpha })$, and prove the result by induction on $k$ for the relative pair $(X^k,X^{k-1})$, then pass to the colimit over $k$.

Step 1 (relative simplicial homology of the pair $(X^k,X^{k-1})$). The simplicial chain complex $C_{\ast }^{\Delta }(X^k,X^{k-1})$ has $C_k^{\Delta }={\mathbb{Z}}^{|J_k|}$ in degree $k$ (one generator per $k$-simplex) and is zero in all other degrees, because every $\Delta$-simplex of dimension less than $k$ already lies in $X^{k-1}$ and every $\Delta$-simplex of dimension greater than $k$ is absent at this stage. So

$$H_n^{\Delta }(X^k,X^{k-1})=\{\begin{array}{ll}{\mathbb{Z}}^{|J_k|}& n=k,\\ 0& n\ne k.\end{array}$$

Step 2 (relative singular homology of the same pair). The pair $(X^k,X^{k-1})$ is good in the sense that $X^{k-1}$ is a deformation retract of a neighbourhood in $X^k$ (slightly thicken each closed simplex to a collar of its boundary). Excision combined with the long exact sequence of a good pair gives

$$H_n(X^k,X^{k-1})\cong {\tilde{H}}_n(X^k/X^{k-1}),$$

and $X^k/X^{k-1}$ is homeomorphic to a wedge ${\bigvee }_{\alpha }{S}^k$, one $k$-sphere for each $k$-simplex (each closed $k$-simplex with its boundary collapsed is an $S^k$). The reduced homology of a wedge of $k$-spheres is ${\mathbb{Z}}^{|J_k|}$ in degree $k$ and $0$ elsewhere. So

$$H_n(X^k,X^{k-1})=\{\begin{array}{ll}{\mathbb{Z}}^{|J_k|}& n=k,\\ 0& n\ne k.\end{array}$$

The map ${\iota }_{\ast }$ on the only non-zero degree sends each simplicial generator $[{\sigma }_{\alpha }]$ to the class of the same singular simplex, which under the identification with ${\tilde{H}}_k(\bigvee S^k)$ is the class of the corresponding wedge factor. Both sides are free abelian on the same indexing set $J_k$, so ${\iota }_{\ast }$ is an isomorphism on $(X^k,X^{k-1})$.

Step 3 (inductive step on absolute homology). Compare the long exact sequences of the pair $(X^k,X^{k-1})$ in simplicial and singular homology:

$$\begin{array}{cccccccccc}\cdots & \to & H_n^{\Delta }(X^{k-1})& \to & H_n^{\Delta }(X^k)& \to & H_n^{\Delta }(X^k,X^{k-1})& \to & H_{n-1}^{\Delta }(X^{k-1})& \to \cdots \\ & & \downarrow {\iota }_{\ast }& & \downarrow {\iota }_{\ast }& & \downarrow {\iota }_{\ast }& & \downarrow {\iota }_{\ast }& \\ \cdots & \to & H_n(X^{k-1})& \to & H_n(X^k)& \to & H_n(X^k,X^{k-1})& \to & H_{n-1}(X^{k-1})& \to \cdots \end{array}$$

The relative comparison maps are isomorphisms by Step 2. By the inductive hypothesis applied to $X^{k-1}$, the absolute comparison maps for $X^{k-1}$ are isomorphisms in all degrees. Apply the five lemma to conclude that ${\iota }_{\ast }:H_n^{\Delta }(X^k)\to H_n(X^k)$ is an isomorphism. The base case $X^0$ is a discrete set of points, where both theories return ${\mathbb{Z}}^{|J_0|}$ in degree 0 and zero elsewhere, and ${\iota }_{\ast }$ is the identity.

Step 4 (pass to the colimit). If $X$ is finite-dimensional, $X=X^k$ for some $k$ and the result follows from Step 3. For infinite-dimensional $X$, both $C_{\ast }^{\Delta }(X)={\mathrm{colim}}_k{C}_{\ast }^{\Delta }(X^k)$ and $C_{\ast }(X)={\mathrm{colim}}_k{C}_{\ast }(X^k)$ are filtered colimits of chain complexes (every chain has compact support, hence lives in some finite skeleton). Filtered colimits of abelian groups commute with homology, so

$$H_n^{\Delta }(X)={\mathrm{colim}}_k{H}_n^{\Delta }(X^k)\stackrel{{\iota }_{\ast }}{\to }{\mathrm{colim}}_k{H}_n(X^k)=H_n(X),$$

a colimit of isomorphisms, hence an isomorphism. $\square$

Bridge. This builds toward the cellular homology theory of the next unit 03.12.13 and the full Eilenberg-Steenrod axiomatic framework: the comparison theorem is the prototype of every uniqueness statement for ordinary homology, and the same skeleton-by-skeleton five-lemma argument appears again in 03.12.13 to identify cellular homology with singular homology. The foundational reason simplicial and singular homology agree is exactly the wedge-of-spheres computation in Step 2: the relative homology of one skeleton modulo the next is concentrated in a single degree, so the long exact sequence collapses to a chain complex whose $n$-th group is precisely the simplicial chain group. This pattern recurs: the same trick shows that any cell-filtration of a space produces a chain complex computing its singular homology. Putting these together, the bridge is the recognition that simplicial homology is not a separate theory but a finite presentation of singular homology, generalises directly to the cellular case, and is dual to the simplicial cohomology / cup-product computations introduced later in 03.04.13.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not yet ship a $\Delta$-complex API tied to a topological space. The closest existing infrastructure is the simplicial-object machinery in Mathlib.AlgebraicTopology.SimplicialObject and the alternating-face-map chain complex in Mathlib.AlgebraicTopology.AlternatingFaceMapComplex, plus the singular complex in Mathlib.AlgebraicTopology.SingularSet. The intended definition of the $\Delta$-complex chain complex would read schematically:

[object Promise]

The Hatcher-2.27 statement is the formalisation target. The current Mathlib treatment of the Dold-Kan correspondence supplies the abstract algebraic skeleton, but the geometric input — that the realisation map is compatible with the topology on $X$ and that the comparison induces an isomorphism on singular homology — is not yet formalised. See lean_mathlib_gap in the frontmatter for the concrete Mathlib roadmap.

Advanced results [Master]

Theorem (relative simplicial homology of skeleta). For a $\Delta$-complex $X$ with $k$-skeleton $X^k$, the relative simplicial chain complex $C_^\Delta(X^k, X^{k-1})$isconcentratedindegree$k$,whereitequalsthefreeabeliangroupontheset$J_k$of$k$-simplices.Consequently$H_n^\Delta(X^k, X^{k-1}) = \mathbb{Z}^{|J_k|}$for$n = k$and$0$ otherwise.*

This is the input to the Hatcher 2.27 comparison theorem and to the cellular-homology theory of 03.12.13. The same statement, applied with the skeleton filtration of a CW complex, is what defines the cellular chain complex.

Theorem (Eilenberg-Steenrod uniqueness on $\Delta$-complexes). Let $h_$beanyhomologytheoryonthecategoryof$\Delta$-complexesand$\Delta$-mapssatisfyingtheEilenberg-Steenrodaxioms(homotopy,exactness,excision,dimension,additivity).Then$h_*$ is naturally isomorphic to simplicial homology.*

The proof is the standard induction on skeleta combined with the comparison theorem. The dimension axiom $h_{\ast }(\mathrm{pt})=\mathbb{Z}$ in degree 0 fixes the homology of a point; the wedge-of-spheres computation in Step 2 of the comparison proof transports through any such $h_{\ast }$; and the long-exact-sequence comparison via the five lemma propagates the isomorphism through all skeleta. This is the sense in which "ordinary homology with integer coefficients" is unique on $\Delta$-complexes.

Theorem (additivity and naturality). The simplicial homology functor $H_^\Delta$isfunctorialunder$\Delta$-maps(continuousmapsthatsendeachsimplextoasimplexbyanaffinemappreservingvertexorder).Ondisjointunions$X = \bigsqcup X_i$of$\Delta$-complexes,$H_n^\Delta(X) = \bigoplus_i H_n^\Delta(X_i)$.Thecomparisonmap$\iota_*$ to singular homology is a natural transformation between these functors.*

Naturality and additivity are immediate from the chain-level definitions; the substantive content is that arbitrary continuous maps between $\Delta$-complexes need not be $\Delta$-maps, but they are homotopic to $\Delta$-maps under simplicial approximation (see Munkres §16). Combined with homotopy invariance of singular homology, this establishes that simplicial homology is a homotopy invariant of the underlying topological space, independent of the chosen $\Delta$-structure.

Theorem (compactness and finite generation). If $X$ is a finite $\Delta$-complex (finitely many simplices in finitely many dimensions), then each $H_n^{\Delta }(X)$ is a finitely generated abelian group, and only finitely many are nonzero. The Smith normal form of the boundary matrices computes the homology by row-and-column reduction over $\mathbb{Z}$.

The computational content is that simplicial homology is algorithmic: input a finite $\Delta$-complex, output the homology groups. This is why simplicial homology is the model of ordinary homology used in practice for explicit calculations and for computer verification.

Computational examples.

• Spheres. $S^n$ admits a $\Delta$-structure with two $k$-simplices in each dimension $0\le k\le n$ (the upper and lower hemispheres of each sub-sphere). The boundary alternates: in even degrees the difference of the two simplices is a cycle, and in odd degrees one half maps to the other. The result is $H_k(S^n)=\mathbb{Z}$ for $k=0,n$ and $0$ otherwise.

• Real projective space. ${\mathbb{RP}}^n$ has a $\Delta$-structure with one cell in each dimension. The boundary of the $k$-cell is $(1+(-1{)}^k)$ times the $(k-1)$-cell, by direct computation of the antipodal-quotient face map. So odd-degree boundaries vanish and even-degree boundaries are multiplication by $2$. The resulting homology is the famous table: $H_0=\mathbb{Z}$, $H_k=\mathbb{Z}/2$ for $k$ odd with $0<k<n$, $H_k=0$ for $k$ even with $0<k\le n$ (except $H_n=\mathbb{Z}$ if $n$ is odd). The $\mathbb{Z}/2$ in odd degrees is the prototype of torsion in homology.

• Complex projective space. ${\mathbb{CP}}^n$ has a $\Delta$-structure with one cell in each even dimension (the cells of the standard CW structure subdivided). All boundary maps are zero because the cells are in non-adjacent degrees. So $H_{2k}({\mathbb{CP}}^n)=\mathbb{Z}$ for $0\le k\le n$ and $H_j=0$ otherwise.

• Genus-$g$ surface. ${\Sigma }_g$ has a one-vertex, $2g$-edge, one-2-cell $\Delta$-structure with boundary word ${\prod }_{i=1}^g[a_i,b_i]$. All boundary maps vanish, giving $H_0=\mathbb{Z}$, $H_1={\mathbb{Z}}^{2g}$, $H_2=\mathbb{Z}$.

• Lens spaces. $L(p,q)=S^3/(\mathbb{Z}/p)$ has a $\Delta$-structure with $p$ cells in each dimension $0,1,2,3$; the resulting homology is $H_0=\mathbb{Z}$, $H_1=\mathbb{Z}/p$, $H_2=0$, $H_3=\mathbb{Z}$.

Synthesis. Simplicial homology is the original combinatorial incarnation of homology, dating to Poincaré 1895, and the comparison theorem identifies it with the topologically-canonical singular theory introduced by Eilenberg in 1944. The bridge is the wedge-of-spheres calculation: every relative pair $(X^k,X^{k-1})$ has its homology concentrated in a single degree, and that degree is exactly the simplicial chain group. This is the foundational reason any cell-filtration argument works in algebraic topology, and the same pattern appears again in the cellular homology of CW complexes 03.12.13, in the spectral-sequence calculation of homology from a filtration, and in the Eilenberg-MacLane construction 03.12.05 where the cell-by-cell killing of homotopy groups uses exactly this skeleton arithmetic. Putting these together, the bridge is the recognition that simplicial homology, cellular homology, and singular homology are three reformulations of one functor, distinguished only by which combinatorial input is used to compute it: a $\Delta$-structure, a CW structure, or no structure at all. Each generalises the previous in flexibility and specialises it in computability. The comparison theorem identifies them as natural-isomorphism-equivalent functors on the appropriate categories, and this identification underlies every classical computation in homology and recurs in the dual cohomology theory developed in 03.04.13. The Eilenberg-Steenrod uniqueness theorem makes the identification axiomatic: any homology theory satisfying the standard axioms agrees with simplicial homology on $\Delta$-complexes.

Full proof set [Master]

Theorem (Hatcher 2.27, simplicial = singular), full proof. Given in the Intermediate-tier Key theorem section above. The four steps — reduction to the absolute case, skeleton-by-skeleton five-lemma argument, wedge-of-spheres computation of the relative singular homology, and colimit over skeleta — together prove that ${\iota }_{\ast }$ is an isomorphism on every $\Delta$-complex pair.

Theorem (relative simplicial homology of skeleta), proof. The simplicial chain complex of the pair $(X^k,X^{k-1})$ has chain group $C_n^{\Delta }(X^k)/C_n^{\Delta }(X^{k-1})$. By construction, every $\Delta$-simplex of dimension $<k$ lies in $X^{k-1}$, every $\Delta$-simplex of dimension $>k$ is absent in $X^k$, and the $k$-simplices of $X^k$ that are not in $X^{k-1}$ are precisely the $k$-simplices indexed by $J_k$. Hence the relative chain complex has $C_k^{\Delta }={\mathbb{Z}}^{|J_k|}$ in degree $k$ and zero elsewhere. The boundary ${\partial }_k$ vanishes (its image lies in $C_{k-1}^{\Delta }(X^k)/C_{k-1}^{\Delta }(X^{k-1})=0$), so the homology equals the chain group itself in degree $k$. $\square$

Theorem (Eilenberg-Steenrod uniqueness), proof sketch. Let $h_{\ast }$ be a homology theory on $\Delta$-complexes satisfying the axioms. The dimension axiom and additivity give $h_n(\mathrm{pt})=\mathbb{Z}$ for $n=0$ and zero otherwise, hence $h_n(⨆\mathrm{pt})=⨁\mathbb{Z}$ for the discrete 0-skeleton. By induction on $k$: assume $h_n(X^{k-1})\cong H_n^{\Delta }(X^{k-1})$ naturally. The pair $(X^k,X^{k-1})$ is excisive by axiom; the relative homology of a wedge of $k$-spheres is ${\mathbb{Z}}^{|J_k|}$ in degree $k$ by the suspension-isomorphism consequence of the axioms (suspension shifts degree by one and reduces to the $S^0$ case via dimension). Comparing the long exact sequences of the pair in $h_{\ast }$ and in $H_{\ast }^{\Delta }$ via the natural transformation ${\iota }_{\ast }$ extended formally, the five lemma propagates the isomorphism to $h_n(X^k)$. Pass to colimits as in the Hatcher 2.27 proof.

Theorem (additivity and naturality), proof. Naturality: a $\Delta$-map $f:X\to Y$ sends each generator ${\sigma }_{\alpha }$ to a singular simplex $f\circ {\sigma }_{\alpha }$ which is a $\Delta$-simplex of $Y$ by hypothesis, hence to a generator of $C_{\ast }^{\Delta }(Y)$; the induced chain map commutes with $\partial$ because face restrictions are preserved by $f$; and applied homology gives $f_{\ast }:H_{\ast }^{\Delta }(X)\to H_{\ast }^{\Delta }(Y)$. The comparison map ${\iota }_{\ast }$ is natural because a $\Delta$-map is in particular a continuous map. Additivity: the chain complex of a disjoint union splits as a direct sum of the chain complexes of the summands, so homology splits accordingly.

Theorem (compactness and finite generation), proof. A finite $\Delta$-complex has finitely many simplices in each degree and zero in all but finitely many degrees, so each $C_n^{\Delta }$ is a finitely generated free $\mathbb{Z}$-module and only finitely many are nonzero. The boundary $\partial$ is a $\mathbb{Z}$-linear map between finitely generated free $\mathbb{Z}$-modules. Smith normal form (the structure theorem for finitely generated modules over a PID applied to the boundary matrix) gives $\partial =UDV$ with $U,V$ unimodular and $D$ diagonal with positive integer diagonal entries. Reading off $\ker$ and $\mathrm{im}$ from $D$ produces $H_n$ as a finitely generated abelian group, of the form ${\mathbb{Z}}^{r_n}\oplus {⨁}_i\mathbb{Z}/d_{n,i}$ where the $d_{n,i}$ are the elementary divisors. $\square$

Stated without proof — see Hatcher §2.1 ^{[Hatcher §2.1]} and Munkres §13 ^{[Munkres §13]}. The simplicial approximation theorem (any continuous map $f:K\to L$ between simplicial complexes is homotopic to a simplicial map after sufficient barycentric subdivision of $K$), used implicitly in the homotopy invariance of simplicial homology; the Hauptvermutung in dimensions $\le 3$ (any two triangulations of a topological 3-manifold have a common subdivision); the failure of the Hauptvermutung in dimension $\ge 5$ (Milnor 1961 counterexample); the existence of non-triangulable topological 4-manifolds (Casson 1985; Freedman's $E_8$-manifold).

Connections [Master]

• CW complex 03.12.10. Every CW complex with characteristic maps that respect a cell-by-cell ordering carries a canonical $\Delta$-structure on a subdivision; conversely, every $\Delta$-complex is a CW complex with the obvious cell decomposition. The skeletal filtration argument used in the comparison theorem is the prototype for the cellular-homology argument that identifies cellular homology of a CW complex with its singular homology.

• Cellular homology 03.12.13 (planned, parallel batch). The next unit upgrades the simplicial-homology theory to the more flexible cellular-homology theory: cells replace simplices, and the cellular boundary is computed via degrees of attaching maps rather than face restrictions. The same skeleton-by-skeleton five-lemma argument identifies cellular homology with singular homology, and the Hatcher 2.27 comparison is the prototype of that identification.

• Singular cohomology 03.04.13. Dualising the simplicial chain complex of a $\Delta$-complex produces the simplicial cochain complex; its cohomology is simplicial cohomology, and the same comparison theorem gives an isomorphism with singular cohomology. The cup-product on simplicial cohomology is the Alexander-Whitney coproduct dualised, and the cohomology ring of a finite $\Delta$-complex is computable from a finite combinatorial recipe.

• Eilenberg-MacLane spaces 03.12.05. The construction of $K(\pi ,n)$ proceeds by attaching cells to kill homotopy groups; the homology of each stage is computed via the cellular / simplicial chain complex of the $\Delta$-or-CW structure built so far. Hatcher 2.27 ensures that this combinatorial computation matches the topologically-canonical singular homology used to detect the killed classes.

• Mayer-Vietoris and excision (planned units). The excision axiom for singular homology is what justifies the wedge-of-spheres computation in Step 2 of the comparison proof; conversely, simplicial homology gives an explicit model in which excision can be verified by direct chain-level inspection, and the agreement of the two theories transports excision back to the singular side.

• Riemann-Roch and intersection theory (downstream applications). On a triangulated Riemann surface, the simplicial homology of the underlying real surface ($H_1={\mathbb{Z}}^{2g}$) is the home of the symplectic intersection pairing; on a triangulated complex algebraic variety, the simplicial homology of the underlying real variety carries the cycle classes of subvarieties. Both are central to the Riemann-Roch theorem.

Historical & philosophical context [Master]

Henri Poincaré introduced the simplicial chain complex in Analysis Situs (Journal de l'École Polytechnique, 1895) and its five subsequent supplements (1899-1904) ^{[Poincaré Analysis Situs]}. His original definition triangulated each manifold into a finite collection of simplices, took the formal $\mathbb{Z}$-linear combinations as chains, and defined the boundary as the alternating sum of faces. The resulting "Betti numbers" — the ranks of the homology groups — were shown to be topological invariants in his second supplement, and the existence of torsion in $H_1({\mathbb{RP}}^n)$ for $n$ odd was identified in the fifth supplement. The dependence of the construction on the choice of triangulation was a known difficulty that Poincaré addressed via the Hauptvermutung (the conjecture that any two triangulations admit a common subdivision); the proof of independence from triangulation was given for low dimensions during the early twentieth century.

The modern $\Delta$-complex framing — semi-simplicial sets allowing identification of faces — is the framework of Eilenberg-Zilber 1950 ("Semi-simplicial complexes and singular homology", Annals of Mathematics 51, 499-513) and Eilenberg-Steenrod 1952 (Foundations of Algebraic Topology, Princeton) ^{[Eilenberg-Steenrod 1952]}. Eilenberg-Steenrod axiomatised the homology functor (homotopy, exactness, excision, dimension, additivity) and proved that any theory satisfying the axioms agrees with simplicial homology on the category of $\Delta$-complex pairs. Eilenberg had earlier (1944) introduced the singular chain complex, defined for arbitrary topological spaces, which removed the dependence on triangulation entirely; the comparison theorem identifying simplicial with singular homology on $\Delta$-complexes is what justified the use of the older combinatorial theory for computation while granting the newer theory its topological canonicity.

The category of $\Delta$-complexes was later subsumed into the more general category of simplicial sets (Kan 1957, Quillen 1967) where the same comparison theorem is the geometric-realisation / total-singular-complex adjunction's statement that geometric realisation preserves homology. The Dold-Kan correspondence (Dold 1958, Kan 1958) gave the algebraic incarnation, identifying simplicial abelian groups with non-negatively-graded chain complexes via the alternating-face-map functor, and this is the modern home of the simplicial chain complex in homological algebra.

Bibliography [Master]

[object Promise]