Singular Homology

Singular homology assigns to every topological space

X

a sequence of abelian groups

H_{n} (X)

that measure

n

-dimensional "holes" using continuous maps of standard simplices. Key result: it is a homotopy-invariant, excisive, functorial sequence of groups satisfying the Eilenberg–Steenrod axioms, computable via the long exact sequence and Mayer–Vietoris. Where it applies: all topological spaces (no manifold or CW structure required). Who/when: crystallized by Eilenberg (1944), building on Poincaré (1895), Vietoris, Čech, and Lefschetz.

Singular homology is a functor $H_{n} (-) : Top \to Ab$ that detects topological holes of each dimension. Its definition uses singular simplices — arbitrary continuous maps $σ : Δ^{n} \to X$ from the standard $n$ -simplex — assembled into chain complexes whose homology groups are $H_{n} (X)$ . It matters because it is defined for every space, is functorial, and is invariant under homotopy equivalence, making it a powerful tool to distinguish spaces and prove fixed-point and division theorems. A surprising fact: although the singular chain groups $C_{n} (X)$ are typically uncountably generated (huge free abelian groups), the resulting homology groups are often finitely generated and computable. The theory is uniquely characterized — up to natural isomorphism on nice spaces — by the Eilenberg–Steenrod axioms together with a normalization on the point.

The Core Insight

A hole is a cycle that is not a boundary. Probe $X$ with continuous images of solid simplices; the boundary operator $\partial$ sends an $n$ -simplex to the alternating sum of its faces. Then $\partial^{2} = 0$ , so boundaries sit inside cycles, and the quotient "cycles modulo boundaries" exactly captures $n$ -dimensional voids that fail to be filled in by an $(n + 1)$ -dimensional region. Replacing rigid combinatorial simplices with arbitrary continuous maps frees the theory from any chosen triangulation, buying full topological invariance at the cost of enormous chain groups.

Formal Statement

Standard simplex and faces

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 convex span of the standard basis vectors

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

; each point is a unique barycentric combination.

with vertices $e_{0}, \dots, e_{n}$ . The $i$ -th face map $d^{i} : Δ^{n - 1} \to Δ^{n}$ is the affine embedding onto the face opposite $e_{i}$ :

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

⟳insert a zero in the

i

-th coordinate slot, skipping vertex

e_{i}

, preserving barycentric coordinates.

Singular chains and boundary

A singular $n$ -simplex in a space $X$ is a continuous map $σ : Δ^{n} \to X$ . The group of singular $n$ -chains is the free abelian group on these:

C_{n} (X) = Z ⟨ σ : Δ^{n} \to X continuous ⟩, C_{n} (X) = 0 for n < 0.

⟳take formal

Z

-linear combinations

\sum_{σ} n_{σ} σ

with finitely many nonzero coefficients.

The boundary homomorphism $\partial_{n} : C_{n} (X) \to C_{n - 1} (X)$ is defined on generators by the alternating sum of face restrictions:

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

⟳each

σ \circ d^{i} : Δ^{n - 1} \to X

is the

i

-th face of

σ

; the signs make

\partial^{2} = 0

via cancellation of double faces.

The singular homology groups are

H_{n} (X) = \frac{ker ( \partial _{n} : C _{n} \to C _{n - 1} )}{im ( \partial _{n + 1} : C _{n + 1} \to C _{n} )} = \frac{Z _{n} ( X )}{B _{n} ( X )},

⟳

Z_{n} = ker \partial_{n}

(cycles) contains

B_{n} = im \partial_{n + 1}

(boundaries) precisely because

\partial^{2} = 0

; the quotient is well-defined.

Fundamental properties (Eilenberg–Steenrod)

Singular homology $H_{*}$ with coefficients in $Z$ satisfies:

Functoriality: a map $f : X \to Y$ induces $f_{*} : H_{n} (X) \to H_{n} (Y)$ with $(g \circ f)_{*} = g_{*} f_{*}$ and $(id)_{*} = id$ .
Homotopy invariance: $f ≃ g \Rightarrow f_{*} = g_{*}$ .
Long exact sequence of a pair $(X, A)$ :

\dots \to H_{n} (A) i_{*} H_{n} (X) j_{*} H_{n} (X, A) \partial H_{n - 1} (A) \to \dots

⟳the short exact sequence of chain complexes

0 \to C_{*} (A) \to C_{*} (X) \to C_{*} (X, A) \to 0

yields this via the snake lemma.

Excision: if $\overline{Z} \subset int (A)$ , then $H_{n} (X ∖ Z, A ∖ Z) ≅ H_{n} (X, A)$ .
Dimension axiom: $H_{n} (pt) = Z$ for $n = 0$ and $0$ otherwise.
Additivity: $H_{n} (∐_{α} X_{α}) ≅ ⨁_{α} H_{n} (X_{α})$ .

Why It Holds / Derivation

The chain complex condition $\partial^2 = 0$

We verify the cornerstone identity that makes homology well-defined.

Step 1 — Face identity. For

i < j

, composing face maps satisfies the simplicial identity

d^{j} \circ d^{i} = d^{i} \circ d^{j - 1} (i < j),

⟳inserting zeros in coordinate slots

i

then

j

equals inserting in slot

j - 1

then

i

, because inserting at the earlier slot shifts later indices.

Step 2 — Expand $\partial^{2}$ . 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)^{i + j} σ \circ d^{j} \circ d^{i} .

⟳apply [4] twice; the inner

d^{i}

ranges over faces of the

(n - 1)

-chain

σ \circ d^{j}

Step 3 — Split and cancel. Separate the double sum into terms with

i < j

and

i \geq j

. Using [7] on the

i < j

block and reindexing the

i \geq j

block (

i^{'} = i

j^{'} = j

with

i^{'} \geq j^{'}

) shows the two blocks are negatives of one another. Hence every term cancels:

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

⟳the substitution

(i, j) \mapsto (j - 1, i)

i < j

matches the

i \geq j

terms with opposite sign

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

flip.

Thus $B_{n} = im \partial_{n + 1} \subseteq ker \partial_{n} = Z_{n}$ , and the quotient $H_{n} = Z_{n} / B_{n}$ in [5] is well-defined. $■$

Homotopy invariance via the prism operator

Let $H : X \times [0, 1] \to Y$ be a homotopy from $f = H (-, 0)$ to $g = H (-, 1)$ . Define a chain homotopy $P : C_{n} (X) \to C_{n + 1} (Y)$ by subdividing the prism $Δ^{n} \times [0, 1]$ into $(n + 1)$ -simplices with vertices $(e_{0}, 0), \dots, (e_{i}, 0), (e_{i}, 1), \dots, (e_{n}, 1)$ :

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

⟳triangulate the prism by the standard "staircase" decomposition, with

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

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

A direct boundary computation on the prism's top, bottom, and side faces gives the identity

\partial P + P \partial = g_{#} - f_{#},

⟳the top face yields

g_{#}

, the bottom

- f_{#}

, and the interior side faces cancel in

\partial P + P \partial

Chain homotopic chain maps induce equal maps on homology, so $f_{*} = g_{*}$ . $■$

Samuel Eilenberg | 1913–1998

Polish-American topologist who, in his 1944 paper "Singular Homology Theory," gave the modern definition using all continuous maps of simplices, eliminating the dependence on triangulations that plagued earlier combinatorial approaches. With Norman Steenrod he axiomatized homology (1945, book 1952), reducing the subject to a few characterizing properties. Eilenberg co-founded category theory with Mac Lane — partly to make functoriality of homology precise — and later co-authored the Eilenberg–Zilber and Eilenberg–Moore machinery. His insistence on naturality reshaped algebraic topology into its current functorial form.

Building Intuition

Think of $H_{n} (X)$ as cataloguing " $n$ -dimensional surfaces in $X$ that enclose nothing yet bound nothing." A 1-cycle is a loop; if it cannot be filled by a disk in $X$ , it survives in $H_{1}$ . The torus has two independent loops (the meridian and longitude), so $H_{1} (T^{2}) = Z^{2}$ . The 2-sphere has no nontrivial loops ( $H_{1} = 0$ ) but a genuine 2-dimensional void enclosed by its surface, so $H_{2} (S^{2}) = Z$ . The "singular" adjective signals that we allow degenerate, crumpled, even constant maps of simplices — the simplices are probes, not pieces of a rigid mesh. Most probes are boundaries (uninteresting); homology extracts only the equivalence classes that resist being filled.

A 1-cycle on the torus that is not a boundary — generator of $H_1(T^2)$

<svg viewBox="0 0 460 280" xmlns="http://www.w3.org/2000/svg">
<rect width="460" height="280" fill="#0a0e17"/>
<ellipse cx="230" cy="140" rx="160" ry="90" fill="none" stroke="#94a3b8" stroke-width="2"/>
<path d="M120 130 Q230 200 340 130" fill="none" stroke="#94a3b8" stroke-width="2"/>
<path d="M140 142 Q230 100 320 142" fill="none" stroke="#94a3b8" stroke-width="2"/>
<ellipse cx="230" cy="140" rx="40" ry="80" fill="none" stroke="#d4a853" stroke-width="3" stroke-dasharray="6 4"/>
<text x="250" y="60" fill="#e2e8f0" font-size="14" font-family="serif">torus T²</text>
<text x="270" y="150" fill="#d4a853" font-size="14" font-family="serif">cycle z, [z]≠0 in H₁</text>
<text x="180" y="260" fill="#e2e8f0" font-size="13" font-family="serif">z bounds no 2-chain in T²</text>
</svg>

Worked Examples

Reduced homology of the $n$-sphere

Given: Compute

H_{k} (S^{n})

for all

k, n \geq 1

using Mayer–Vietoris. Step 1. Cover

S^{n}

by two open hemispheres

U, V

, each contractible, with

U \cap V ≃ S^{n - 1}

(an equatorial band). Step 2. Reduced Mayer–Vietoris gives

H_{k} (U) \oplus H_{k} (V) \to H_{k} (S^{n}) \to H_{k - 1} (S^{n - 1}) \to H_{k - 1} (U) \oplus H_{k - 1} (V) .

⟳the Mayer–Vietoris sequence of

{U, V}

with

U, V

contractible kills the outer terms.

Step 3. Since

U, V

contractible,

H_{*} (U) = H_{*} (V) = 0

, so the connecting map is an isomorphism:

H_{k} (S^{n}) ≅ H_{k - 1} (S^{n - 1}) .

⟳a map flanked by zeros is an isomorphism by exactness.

Step 4. Base case

S^{0}

= two points:

H_{0} (S^{0}) = Z

, else

0

. Induct. Boxed answer:

H_{k} (S^{n}) = Z for k = 0 and k = n (n \geq 1); H_{k} (S^{n}) = 0 otherwise.

(For

n = 0

H_{0} (S^{0}) = Z^{2}

$H_*$ of the real projective plane $\mathbb{RP}^2$ via cellular chains

Given:

RP^{2}

has a CW structure with one cell each in dimensions

0, 1, 2

. The cellular chain complex is

Z d_{2} Z d_{1} Z

. Step 1. The 1-cell is a loop, so

d_{1} = 0

. Step 2. The 2-cell attaches along the loop traversed twice (degree 2 map

S^{1} \to S^{1}

), so the cellular boundary is

d_{2} = \times 2 : Z \to Z .

⟳the attaching map of the top cell of

RP^{2}

wraps the 1-skeleton circle twice, giving degree

2

Step 3. Compute homology of

0 \to Z 2 Z 0 Z \to 0

$H_{2} = ker (\times 2) = 0$ ,
$H_{1} = Z /2 Z$ (image of $\times 2$ is $2 Z$ ),
$H_{0} = Z$ (connected).

By the equivalence of cellular and singular homology, these are the singular groups.

Boxed answer:

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

The torsion class

Z /2

is invisible to rational homology but real in

Z

— a hallmark of nonorientability.

Common Misconceptions

Singular chains are not "loops" or "surfaces." A singular simplex can be degenerate (e.g., a constant map), wildly self-intersecting, or measure-zero. Geometric intuition ("a 2-cycle is an embedded surface") is heuristic only; cycles are algebraic combinations. $H_{0}$ counts path-components, not points. For a space with

k

path-components,

H_{0} (X) ≅ Z^{k}

(free). Reduced homology

H_{0}

has rank

k - 1

. Homology does not determine the space. Non-homotopy-equivalent spaces can share all homology groups (e.g., a space with the homology of a sphere need not be a sphere). Homology is invariant but not complete. Coefficients matter.

H_{n} (X; Q)

destroys all torsion. Always specify the coefficient group; the universal coefficient theorem relates them.

Homology is not a complete invariant

The Poincaré homology sphere $Σ$ (quotient $S^{3} / I^{*}$ by the binary icosahedral group) has exactly the integral homology of $S^{3}$ :

H_{0} (Σ) = Z, H_{1} (Σ) = 0, H_{2} (Σ) = 0, H_{3} (Σ) = Z .

Yet

Σ

is not homeomorphic (nor homotopy equivalent) to

S^{3}

: its fundamental group

π_{1} (Σ)

is the binary icosahedral group of order

120

, while

π_{1} (S^{3}) = 1

. Homology, being abelian, cannot see this nonabelian fundamental group. This counterexample motivated Poincaré to invent

π_{1}

and refine his conjecture.

Connections

Singular homology sits at the hub of algebraic topology. It is isomorphic to Simplicial Homology on triangulable spaces and to Cellular Homology on CW complexes, and to Čech Cohomology's dual on nice spaces. Its abelianized shadow of Fundamental Group is captured by the Hurewicz theorem: $H_{1} (X) ≅ π_{1} (X)^{ab}$ for path-connected $X$ . Dualizing the chain complex with $Hom (-, G)$ produces Singular Cohomology, linked to homology by the Universal Coefficient Theorem and endowed with the cup-product ring structure. The interaction of homology with maps and fiber bundles is organized by Spectral Sequences, while Poincaré Duality relates $H_{k}$ and $H_{n - k}$ on closed oriented $n$ -manifolds. Through the Eilenberg–Steenrod Axioms, singular homology is one instance of a generalized homology theory, a family that includes K-Theory and Cobordism.

Real-World Applications

Brouwer fixed-point theorem. Using $H_{n} (S^{n - 1})$ , one proves every continuous self-map of $D^{n}$ has a fixed point — foundational in economics (Nash equilibria via Kakutani) and PDE existence theory.
Topological data analysis (persistent homology). Computes $H_{*}$ of a filtered simplicial complex built from point clouds; used in genomics, neuroscience (analyzing neural firing patterns), and materials science to detect voids in porous media.
Sensor network coverage. Čech/Vietoris–Rips homology certifies that a region is fully covered by sensors without coordinate data — De Silva–Ghrist's "coverage holes."
Robotics & configuration spaces. $H_{*}$ of configuration spaces detects motion-planning obstructions and counts homotopy classes of paths (topological complexity).
Condensed matter physics. Homology/cohomology classifies topological insulators and quantum Hall plateaus (Chern numbers are evaluations of cohomology classes).
Computer graphics / mesh repair. Homology generators locate and fix handles and tunnels in 3D scanned meshes.
Electromagnetism on complex domains. $H_{1}$ of a conductor's complement governs the number of independent loop currents (cohomological formulation of Maxwell's equations).

Historical Development

1895 — Henri Poincaré, Analysis Situs: introduces homology via "Betti numbers" and torsion coefficients of manifolds; intuitive, manifold-bound.
1900s — Poincaré discovers the homology sphere, realizing homology fails to characterize $S^{3}$ ; invents the fundamental group.
1920s — Veblen, Alexander, and Lefschetz develop combinatorial (simplicial) homology rigorously; the boundary operator formalized.
1926–1932 — Heinz Hopf and Emmy Noether reframe homology in terms of abelian groups rather than mere Betti numbers and torsion coefficients — the decisive algebraic turn.
1933 — Čech and (independently) Vietoris develop homology theories for general spaces; Vietoris–Rips complexes appear.
1935 — Lefschetz coins "singular" and explores singular chains; foundations still shaky (issues with degenerate simplices).
1944 — Samuel Eilenberg, Singular Homology Theory: gives the clean, fully functorial modern definition; resolves earlier difficulties.
1945–1952 — Eilenberg & Steenrod axiomatize homology; Foundations of Algebraic Topology (1952) proves uniqueness on triangulable spaces.
1950s onward — Cup products, spectral sequences (Serre, Leray), and generalized theories (K-theory, cobordism) extend the framework; persistent homology emerges around 2000 (Edelsbrunner–Letscher–Zomorodian).

Subtleties & Limits

Where singular homology strains.

Pathological spaces. For non-CW spaces (e.g., the Warsaw circle, the Hawaiian earring) singular and Čech homology disagree; singular homology can miss "wild" structure. The Hawaiian earring has uncountable, non-free $H_{1}$ behavior — its $π_{1}$ is famously non-free and not finitely generated.
Computability. Although homology is "computable in principle," for an arbitrary space presented abstractly there is no algorithm; computation requires combinatorial models (CW/simplicial), and even then the word problem in resulting group presentations can be undecidable when extracting finer invariants.
Failure to detect homotopy type. Beyond the homology sphere, the broader question of which algebraic invariants suffice to classify spaces up to homotopy drives stable and chromatic homotopy theory; the full structure (Steenrod operations, $k$ -invariants, Postnikov towers) is needed, and much remains open in unstable contexts.
Generalized theories. Whether a given exotic homology theory satisfies a desired form of excision or duality on broad classes of spaces remains an active research interface with $\infty$ -categories and motivic homotopy theory.