Singular Homology and Chain Complexes

Singular homology assigns to each topological space

X

a sequence of abelian groups

H_{n} (X)

measuring "

n

-dimensional holes." It is built by mapping standard simplices into

X

, forming free abelian groups of these maps (chains), taking formal boundaries, and computing

ker \partial / im \partial

. It is a homotopy-invariant functor from topological spaces to graded abelian groups — the foundational tool of algebraic topology, developed by Poincaré (1895), Veblen, Alexander, and made rigorous by Eilenberg (1944).

Singular homology is a functor $H_{∙} : Top \to grAb$ that converts the geometric problem of distinguishing spaces into the algebraic problem of distinguishing groups. The construction proceeds in two stages: first one builds the singular chain complex $C_{∙} (X)$ , a sequence of free abelian groups connected by boundary maps $\partial_{n}$ satisfying $\partial_{n - 1} \circ \partial_{n} = 0$ ; then homology extracts the failure of exactness, $H_{n} = ker \partial_{n} / im \partial_{n + 1}$ . Its power lies in being homotopy invariant, functorial, and computable via the long exact sequence and the Mayer–Vietoris sequence. The surprising fact is that this combinatorial machinery, defined using arbitrary continuous maps of simplices (uncountably many in general), nevertheless yields finitely generated, often finite, groups that agree exactly with the more rigid simplicial homology on triangulable spaces — and it does so without requiring any triangulation at all.

The Core Insight

A "hole" is a cycle that is not a boundary. To detect $n$ -dimensional holes in a space $X$ , probe $X$ with maps of the standard $n$ -simplex, package these probes into a chain complex whose algebra encodes "boundary of a boundary is zero," and measure exactly how far cycles fail to be boundaries. The identity $\partial^{2} = 0$ is the single algebraic axiom that makes everything work.

Formal Statement

Standard simplex and singular simplices

The standard $n$ -simplex is the convex hull

Δ^{n} = {(t_{0}, \dots, t_{n}) \in R^{n + 1} : t_{i} \geq 0, \sum_{i = 0}^{n} t_{i} = 1} .

⟳it is the set of barycentric coordinates; the vertices

e_{0}, \dots, e_{n}

are the standard basis vectors, and any point is a convex combination

\sum t_{i} e_{i}

A singular $n$ -simplex in a topological space $X$ is a continuous map $σ : Δ^{n} \to X$ . There is no requirement of injectivity, smoothness, or embedding — hence "singular." The set of all singular $n$ -simplices is denoted $Sing_{n} (X)$ .

Singular chain groups and face maps

The group of singular $n$ -chains is the free abelian group on singular $n$ -simplices:

C_{n} (X) = Z [Sing_{n} (X)] = {\sum_{i} n_{i} σ_{i} : n_{i} \in Z, finite sum}, C_{n} (X) = 0 for n < 0.

⟳take the free abelian group functor

Z [-]

applied to the set

Sing_{n} (X)

; elements are finite formal

Z

-linear combinations of maps

σ

For each $0 \leq i \leq n$ the $i$ -th face map $δ_{i} : Δ^{n - 1} \to Δ^{n}$ inserts a zero in the $i$ -th coordinate:

δ_{i} (t_{0}, \dots, t_{n - 1}) = (t_{0}, \dots, t_{i - 1}, 0, t_{i}, \dots, t_{n - 1}) .

⟳

δ_{i}

is the affine embedding onto the face opposite vertex

e_{i}

, sending

e_{j} \mapsto e_{j}

for

j < i

and

e_{j} \mapsto e_{j + 1}

for

j \geq i

Boundary operator

The boundary map $\partial_{n} : C_{n} (X) \to C_{n - 1} (X)$ is defined on a generator $σ$ by the alternating sum of its faces $σ^{(i)} := σ \circ δ_{i}$ :

\partial_{n} σ = i = 0 \sum n (- 1)^{i} σ \circ δ_{i},

extended

Z

-linearly to all of

C_{n} (X)

⟳the sign

(- 1)^{i}

encodes the induced orientation of the

i

-th face; the alternation is exactly what is needed for

\partial^{2} = 0

below.

Fundamental identity $\partial^2=0$

For every $n \geq 1$ ,

\partial_{n - 1} \circ \partial_{n} = 0.

⟳expand the double sum and use the simplicial identity

δ_{j} δ_{i} = δ_{i} δ_{j - 1}

for

i < j

; the terms cancel in pairs with opposite signs.

Consequently $(C_{∙} (X), \partial_{∙})$ is a chain complex, and one defines:

$n$ -cycles $Z_{n} (X) = ker \partial_{n}$ ;
$n$ -boundaries $B_{n} (X) = im \partial_{n + 1}$ ;
since $\partial^{2} = 0$ , $B_{n} (X) \subseteq Z_{n} (X)$ , so the $n$ -th singular homology group is well-defined:

H_{n} (X) = Z_{n} (X) / B_{n} (X) = ker \partial_{n} / im \partial_{n + 1} .

⟳

B_{n} \subseteq Z_{n}

is exactly the content of

\partial_{n} \partial_{n + 1} = 0

; the quotient of abelian groups is well-defined and itself abelian.

Chain complex, abstractly

A chain complex of abelian groups is a sequence $(C_{∙}, \partial_{∙})$

\dots \partial_{n + 1} C_{n} \partial_{n} C_{n - 1} \partial_{n - 1} \dots \partial_{1} C_{0} \partial_{0} 0, \partial_{n} \circ \partial_{n + 1} = 0.

⟳the defining condition is that consecutive maps compose to zero, equivalently

im \partial_{n + 1} \subseteq ker \partial_{n}

at every degree.

A chain map $f_{∙} : (C_{∙}, \partial) \to (D_{∙}, \partial^{'})$ is a family $f_{n} : C_{n} \to D_{n}$ with $f_{n - 1} \partial_{n} = \partial_{n}^{'} f_{n}$ . The homology $H_{n} (C_{∙}) = ker \partial_{n} / im \partial_{n + 1}$ is a functor on the category of chain complexes.

Functoriality and Eilenberg–Steenrod axioms

Singular homology is a functor $Top \to grAb$ : a continuous map $f : X \to Y$ induces $f_{#} : C_{∙} (X) \to C_{∙} (Y)$ , $σ \mapsto f \circ σ$ , which is a chain map and descends to $f_{*} : H_{n} (X) \to H_{n} (Y)$ with $(g \circ f)_{*} = g_{*} \circ f_{*}$ and $(id)_{*} = id$ . It satisfies the Eilenberg–Steenrod axioms: homotopy invariance, the long exact sequence of a pair, excision, additivity over disjoint unions, and the dimension axiom $H_{n} (pt) = Z$ if $n = 0$ , else $0$ .

Why It Holds / Derivation

Proof that $\partial^2 = 0$

Step 1 — The key simplicial identity. For

0 \leq i < j \leq n

, the face maps satisfy

δ_{j} \circ δ_{i} = δ_{i} \circ δ_{j - 1} (as maps Δ^{n - 2} \to Δ^{n}, i < j) .

⟳both sides insert zeros in coordinate slots

i

and

j

; reading the order of insertions shows the index shift

j \to j - 1

on the right because inserting at

i

first pushes the later slot down by one.

Step 2 — Expand the composite. For a singular simplex

σ : Δ^{n} \to X

\partial_{n - 1} \partial_{n} σ = j = 0 \sum n i = 0 \sum n - 1 (- 1)^{j} (- 1)^{i} σ \circ δ_{j} \circ δ_{i} .

⟳apply

\partial_{n - 1}

to each face

σ \circ δ_{j}

\partial_{n} σ

; each contributes faces

σ \circ δ_{j} \circ δ_{i}

with sign

(- 1)^{j} (- 1)^{i}

Step 3 — Split into $i < j$ and $i \geq j$ . Write the double sum as

S_{<} + S_{\geq}

where

S_{<}

collects terms with

i < j

and

S_{\geq}

those with

i \geq j

. In

S_{\geq}

relabel: terms have

δ_{j} \circ δ_{i}

with

j \leq i

. Step 4 — Match using Step 1. In

S_{<}

, use

δ_{j} δ_{i} = δ_{i} δ_{j - 1}

. Setting

i^{'} = j - 1

j^{'} = i

in this rewritten term gives a term with

j^{'} \leq i^{'}

and sign

(- 1)^{j} (- 1)^{i} = (- 1)^{(j^{'} + 1)} (- 1)^{i^{'}} = - (- 1)^{j^{'}} (- 1)^{i^{'}}

. Thus every term of (rewritten)

S_{<}

exactly cancels a term of

S_{\geq}

. Step 5 — Conclusion. All terms cancel in pairs, so

\partial_{n - 1} \partial_{n} σ = 0

. By linearity

\partial_{n - 1} \partial_{n} = 0

on all chains.

■

Homotopy invariance via the prism operator

Claim. If

f, g : X \to Y

are homotopic, then

f_{*} = g_{*}

on homology. Step 1 — Setup. Let

H : X \times [0, 1] \to Y

be a homotopy with

H_{0} = f

H_{1} = g

. Define

i_{0}, i_{1} : X \to X \times [0, 1]

x \mapsto (x, 0)

and

(x, 1)

; then

H \circ i_{0} = f

H \circ i_{1} = g

. Step 2 — Prism decomposition. Triangulate the prism

Δ^{n} \times [0, 1]

into

(n + 1)

-simplices using vertices

[v_{0}, \dots, v_{i}, w_{i}, \dots, w_{n}]

(where

v_{j} = (e_{j}, 0)

w_{j} = (e_{j}, 1)

). Define the prism operator

P : C_{n} (X) \to C_{n + 1} (X \times I)

P (σ) = i = 0 \sum n (- 1)^{i} (σ \times id)_{[v_{0}, \dots, v_{i}, w_{i}, \dots, w_{n}]} .

⟳

P

"sweeps" each

n

-simplex through the unit interval, splitting the resulting prism into ordered

(n + 1)

-simplices with alternating signs matching the boundary bookkeeping.

Step 3 — Chain homotopy identity. A direct boundary computation yields

\partial P + P \partial = (i_{1})_{#} - (i_{0})_{#} .

⟳in

\partial P σ

the top and bottom faces give

i_{1}

and

i_{0}

images, and the remaining lateral faces cancel against

P \partial σ

by the same sign-matching as in

\partial^{2} = 0

Step 4 — Transport along $H$ . Compose with

H_{#}

and set

D = H_{#} \circ P

. Then

\partial D + D \partial = g_{#} - f_{#}

: the maps

f_{#}, g_{#}

are chain homotopic. Step 5 — Pass to homology. If

z

is a cycle (

\partial z = 0

), then

g_{#} z - f_{#} z = \partial D z \in B_{∙}

, so

[g_{#} z] = [f_{#} z]

. Hence

f_{*} = g_{*}

■

The boundary of a 2-simplex: an oriented triangle whose boundary is a 1-cycle

Samuel Eilenberg | 1913–1998

Polish-American mathematician who in 1944 gave the definitive modern formulation of singular homology, replacing the cumbersome earlier "singular simplices up to subdivision" approaches with the clean free-abelian-group construction used today. With Norman Steenrod he axiomatized homology theory (1945–1952), proving that any functor satisfying the Eilenberg–Steenrod axioms agrees with singular homology on CW complexes — turning a zoo of competing constructions into a single characterized theory. With Saunders Mac Lane he founded category theory in 1945, the language in which functoriality of $H_{∙}$ is naturally expressed. His insistence on naturality and functoriality reshaped all of homological algebra.

Building Intuition

Think of $H_{n} (X)$ as counting $n$ -dimensional "voids." A cycle is something with no boundary — a closed loop ( $n = 1$ ), a closed surface ( $n = 2$ ). A boundary is something that bounds — a loop that is the edge of a disk, a sphere that is the surface of a ball. Homology declares two cycles equivalent if their difference bounds. A loop around a hole in an annulus is a cycle that does not bound (you cannot fill it within the annulus), so it represents a nontrivial class; a loop that can be contracted bounds a disk and is trivial. The whole apparatus of chains, signs, and $\partial^{2} = 0$ is bookkeeping to make "boundary of a boundary is empty" precise: the boundary of a filled triangle is its three edges, and the boundary of that loop (each edge's two endpoints) cancels because each vertex is the end of one edge and the start of another. Singular homology refuses to demand any triangulation: it throws every possible continuous simplex at $X$ at once, then quotients away the redundancy with algebra.

Worked Examples

Homology of the circle $S^1$

Given. Compute

H_{n} (S^{1})

for all

n

. Method (Mayer–Vietoris). Cover

S^{1} = U \cup V

by two open arcs, each contractible, with

U \cap V

two disjoint arcs (so

U \cap V ≃

two points). The reduced Mayer–Vietoris sequence gives

\dots \to \tilde{H}_{n} (U) \oplus \tilde{H}_{n} (V) \to \tilde{H}_{n} (S^{1}) \to \tilde{H}_{n - 1} (U \cap V) \to \tilde{H}_{n - 1} (U) \oplus \tilde{H}_{n - 1} (V) \to \dots

⟳this is the long exact sequence associated to the short exact sequence of chain complexes

0 \to C_{∙} (U \cap V) \to C_{∙} (U) \oplus C_{∙} (V) \to C_{∙} (U + V) \to 0

, with excision identifying the last term's homology with that of

S^{1}

Steps.

$U, V$ contractible $\Rightarrow \tilde{H}_{*} (U) = \tilde{H}_{*} (V) = 0$ .
$U \cap V ≃ {p, q}$ $\Rightarrow \tilde{H}_{0} (U \cap V) = Z$ , all higher $= 0$ .
For $n = 1$ : $0 \to \tilde{H}_{1} (S^{1}) \to \tilde{H}_{0} (U \cap V) \to 0$ , so $\tilde{H}_{1} (S^{1}) ≅ Z$ .
For $n \geq 2$ : sandwiched between zeros, so $\tilde{H}_{n} (S^{1}) = 0$ .

Answer.

H_{0} (S^{1}) = Z, H_{1} (S^{1}) = Z, H_{n} (S^{1}) = 0 (n \geq 2) .

The generator of

H_{1}

is the fundamental loop, "winding once."

Homology of the $n$-sphere $S^n$ by induction

Given. Compute

H_{k} (S^{n})

for

n \geq 1

. Method. Write

S^{n} = U \cup V

with

U, V

open hemispheres (each

≃

pt, contractible) and

U \cap V ≃ S^{n - 1}

(an equatorial band). Mayer–Vietoris gives, for

k \geq 1

\tilde{H}_{k} (S^{n}) ≅ \tilde{H}_{k - 1} (S^{n - 1}) .

⟳contractibility of

U, V

kills their reduced homology, so the connecting map in MV becomes an isomorphism

\tilde{H}_{k} (S^{n}) \sim \tilde{H}_{k - 1} (U \cap V) = \tilde{H}_{k - 1} (S^{n - 1})

Steps.

Base case $S^{0}$ = two points: $\tilde{H}_{0} (S^{0}) = Z$ , all else $0$ .
Iterate (13): $\tilde{H}_{k} (S^{n}) ≅ \tilde{H}_{k - n} (S^{0})$ , which is $Z$ iff $k = n$ , else $0$ .
Restore unreduced $H_{0}$ : every nonempty space has $H_{0}$ containing a $Z$ from path-components; $S^{n}$ is path-connected for $n \geq 1$ .

Answer.

H_{k} (S^{n}) = {Z 0 k = 0 or k = n, otherwise, (n \geq 1) .

Common Misconceptions

Singular homology is not computed directly from the definition. The chain groups

C_{n} (X)

are typically of uncountable rank —

Sing_{n} (X)

is enormous. One never computes

H_{n}

by literally taking kernels and images of these maps. Computation always goes through tools: homotopy invariance, long exact sequences, Mayer–Vietoris, the equivalence with cellular/simplicial homology on CW/triangulable spaces, and the comparison theorems. The free-abelian construction exists to guarantee functoriality and the axioms, not to enable hand computation. $H_{n}$ is a quotient, so a homology class is an equivalence class of cycles, not a single cycle. Writing "

[z]

" matters:

z

and

z + \partial w

represent the same class. Confusing a cycle with its class leads to false claims (e.g., asserting a specific loop "is" the generator when only its class is canonical). Also, homology with

Z

coefficients can have torsion (e.g.,

H_{1} (RP^{2}) = Z /2

), unlike de Rham cohomology over

R

Homology does not detect homotopy type fully

Singular homology is a homotopy invariant, but it is not a complete invariant — spaces with the same homology need not be homotopy equivalent. The Poincaré homology sphere $Σ$ (the quotient $S^{3} / I^{*}$ by the binary icosahedral group) has the homology of $S^{3}$ : $H_{0} = H_{3} = Z$ , $H_{1} = H_{2} = 0$ . Yet $Σ$ is not homotopy equivalent to $S^{3}$ , because its fundamental group $π_{1} (Σ) = I^{*}$ is a nontrivial group of order $120$ . Homology abelianizes (indeed kills) this information: $H_{1}$ is the abelianization of $π_{1}$ , and $I^{*}$ is perfect, so $H_{1} = 0$ despite $π_{1} \neq = 0$ .

Connections

Singular homology sits at a hub of algebraic topology and homological algebra.

Simplicial Homology and Cellular Homology agree with singular homology on triangulable/CW spaces, providing the actual computational engines; the equivalence is itself proved using singular theory.
Singular Cohomology dualizes the chain complex via $Hom (C_{∙}, Z)$ ; the Universal Coefficient Theorem relates $H^{n}$ to $H_{n}$ and its torsion through $Ext$ .
Homotopy Groups $π_{n}$ refine homology; the Hurewicz Theorem says the first nonvanishing $π_{n}$ equals the corresponding $H_{n}$ .
de Rham Cohomology on smooth manifolds is isomorphic to singular cohomology with $R$ coefficients (de Rham theorem), linking differential forms to topology.
Homological Algebra abstracts the chain complex, derived functors $Tor, Ext$ , and the snake lemma from which the long exact sequence arises.

Real-World Applications

Sensor network coverage (TDA): the Čech/Vietoris–Rips complexes of sensor positions use $H_{1}$ to detect coverage holes — a nonzero $H_{1}$ class signals an uncovered region (de Silva–Ghrist).
Persistent homology / topological data analysis: filtered chain complexes track how homology classes are born and die across scales, used in genomics, neuroscience (brain connectome cycles), and materials science (pore structure of glasses).
Robot motion planning: configuration spaces' homology obstructs the existence of continuous, globally defined motion planners (Farber's topological complexity).
Condensed matter / topological phases: $K$ -theory and homology classify gapped quantum phases and topological insulators; Chern numbers are homological invariants.
Electrical engineering / Kirchhoff theory: $H_{1}$ of a circuit graph counts independent current loops (cycle space); $H_{0}$ counts connected components for voltage references.
Computer graphics & mesh repair: computing $H_{1}$ , $H_{2}$ of triangulated surfaces detects handles and voids, guiding hole-filling and genus reduction.
Cosmology and image analysis: Betti numbers of density fields characterize the cosmic web's loops and voids (Minkowski functionals, persistent topology).

Historical Development

1895 — Henri Poincaré's Analysis Situs introduces homology informally via formal sums of submanifolds and the notion of "bounding"; defines Betti numbers and torsion coefficients.
1899–1904 — Poincaré's Compléments repairs gaps, introduces simplicial/triangulation methods and the dual cell decomposition (toward Poincaré duality).
1912 — L. E. J. Brouwer proves invariance of dimension and the fixed-point theorem using simplicial approximation and degree.
1915–1922 — James Alexander and Oswald Veblen develop combinatorial topology; Alexander duality emerges.
1925–1927 — Emmy Noether reframes homology in terms of abelian groups and their homomorphisms rather than Betti numbers and torsion coefficients, founding the group-theoretic viewpoint.
1933–1935 — Čech, Vietoris, and Alexandrov develop homology for general (non-triangulable) spaces; nerve and Čech constructions appear.
1944 — Samuel Eilenberg publishes the modern singular homology definition via free abelian groups on continuous simplices, cleanly homotopy-invariant.
1945 — Eilenberg–Mac Lane found category theory; Eilenberg–Steenrod begin the axiomatic program.
1952 — Eilenberg–Steenrod's Foundations of Algebraic Topology gives the axioms and uniqueness theorem, unifying all homology theories on CW complexes.

Subtleties & Limits

Singular homology is exquisitely well-behaved on CW complexes but exhibits pathologies on wild spaces. The Warsaw circle and Hawaiian earring have singular homology that disagrees with Čech homology and resists intuition — for the Hawaiian earring, $H_{1}$ is an uncountable group, not the naive free product of countably many copies of $Z$ . More broadly, singular homology fails continuity (it does not commute with inverse limits), which is why Čech homology/cohomology exists as an alternative; reconciling them on non-nice spaces is the domain of shape theory (Borsuk, Mardešić). At the categorical frontier, the question of which generalized homology theories arise and how to compute them feeds into stable homotopy theory and the still-open structure of the stable homotopy groups of spheres; chromatic homotopy theory organizes these via formal group laws, but a complete computation remains far out of reach. The recent rise of persistent homology raises its own open issues: stability of persistence diagrams under noise is well-understood, but multi-parameter persistence has no complete discrete invariant, and classifying multidimensional persistence modules is provably wild.