Homology Groups

Why homology matters

• Algebraic topology backbone. Homology assigns computable abelian groups to spaces, turning topological problems into linear algebra over ℤ. Two homeomorphic spaces have isomorphic homology, so a difference in Hₙ certifies that two spaces are not homeomorphic. The sphere S² and the torus T² differ in H₁ (zero vs ℤ²), proving they are different topologically.
• Persistent homology in data analysis. Topological data analysis (TDA) builds a filtration of simplicial complexes from a point cloud at varying scales and tracks which holes persist across scales. Long-lived bars in a barcode signal real topological features in the data; short-lived bars are noise. Used in cancer subtyping, neuroscience (place cells encode environment topology), and material science.
• Sensor coverage. Given sensors with known communication graph but unknown positions, the Vietoris-Rips complex on the graph has H₁ = 0 if and only if the sensors cover a contractible region. Computing H₁ certifies coverage without coordinates — used in distributed robotics and wireless networks.
• Gauge theory and physics. De Rham cohomology, the differential-form analogue of homology, classifies magnetic monopoles by H²(X) and Aharonov-Bohm phases by H¹(X). Yang-Mills instantons are classified by topological charges in H⁴ of the gauge bundle.
• Brouwer fixed-point and degree. The degree of a map S^n → S^n is captured by H_n; this degree is what makes the no-retraction proof of Brouwer work. Lefschetz fixed-point theorem generalizes — fixed points of f: X → X are counted by an alternating trace on Hₙ.
• Mayer-Vietoris and computability. The Mayer-Vietoris sequence computes homology of X = U ∪ V from homology of U, V, U ∩ V. Combined with cellular homology (using a CW-decomposition), it makes Betti numbers of most familiar spaces calculable by hand.

The chain complex construction

Begin with a triangulation or CW-decomposition of X. Let C_n = free abelian group on n-simplices, with each simplex equipped with an orientation. Define the boundary ∂_n: C_n → C_{n-1} by sending an oriented n-simplex [v₀, v₁, …, vₙ] to the alternating sum Σᵢ (−1)ⁱ [v₀, …, v̂ᵢ, …, vₙ] of its (n-1)-dimensional faces with the i-th vertex omitted.

The crucial identity ∂_{n-1} ∘ ∂_n = 0 follows because each (n-2)-face appears twice in the double boundary with cancelling signs. So we get a sequence:

… → C_{n+1} → C_n → C_{n-1} → … → C_1 → C_0 → 0,

and the n-th homology is Hₙ(X) = ker(∂_n) / im(∂_{n+1}) = (n-cycles) / (n-boundaries). An n-cycle that is not a boundary is a non-trivial class — an n-dimensional hole.

Computed examples

1. Point. H₀ = ℤ (one component), Hₙ = 0 for n ≥ 1. Reflects "no holes anywhere."
2. Circle S¹. H₀ = ℤ (one component), H₁ = ℤ (one loop), Hₙ = 0 for n ≥ 2.
3. Sphere S². H₀ = ℤ, H₁ = 0 (every loop on S² contracts), H₂ = ℤ (one cavity).
4. Torus T². H₀ = ℤ, H₁ = ℤ² (meridian + longitude), H₂ = ℤ (the surface itself is a cycle).
5. Klein bottle K. H₀ = ℤ, H₁ = ℤ ⊕ ℤ/2 (one free + one torsion), H₂ = 0 (non-orientable, no fundamental class over ℤ).
6. Real projective plane ℝℙ². H₀ = ℤ, H₁ = ℤ/2, H₂ = 0. Pure torsion in H₁.
7. Genus-g surface Σ_g. H₀ = ℤ, H₁ = ℤ^{2g}, H₂ = ℤ. Each handle adds two generators to H₁.

Common misconceptions

• Homology equals fundamental group. H₁(X) is the abelianization of π₁(X), not π₁ itself. The figure-8 has π₁ = F₂ (free group on two letters, non-abelian) but H₁ = ℤ². Homology forgets non-commutative information.
• Homology is easy to compute. For nice spaces with explicit cell structure, yes — by Mayer-Vietoris or the cellular chain complex. For arbitrary spaces, computation can be intractable; the homology of a moduli space or a high-dimensional manifold may require deep machinery (spectral sequences, equivariant homology).
• Only for nice spaces. Singular homology is defined for any topological space, even pathological ones like the topologist's sine curve or the Hawaiian earring. The values may be exotic — the Hawaiian earring has uncountable H₁ — but they are well-defined.
• Betti numbers determine the space. They do not. The torus T² and the disjoint union S² ⊔ S¹ ⊔ S¹ have the same Betti numbers (β₀ = 1 vs 3 differs, so this example fails — but spaces with the same Betti numbers but different homology with torsion exist: ℝℙ² ⊔ S² and the Klein bottle, for instance, share Betti numbers but differ in torsion).
• Coefficients do not matter. Hₙ(X; ℤ) carries torsion; Hₙ(X; ℚ) does not. Universal coefficient theorem relates them. ℝℙ² has H₁(−; ℤ) = ℤ/2 but H₁(−; ℚ) = 0. Choice of coefficients affects what is visible.
• Homology distinguishes all spaces. No. The Whitehead manifold has the homology of a point but is not contractible. There are simply connected manifolds with the same homology as the sphere that are not the sphere (Mazur manifolds, exotic spheres in higher dimensions). Homology is a coarse invariant.

Eilenberg-Steenrod axioms

In 1952 Eilenberg and Steenrod showed that any homology theory satisfying five axioms — homotopy, excision, exact sequence of a pair, dimension, and additivity — is uniquely determined on CW-complexes, with the singular construction as the canonical example. Dropping the dimension axiom yields generalized homology theories: K-theory, cobordism, stable homotopy. The axiomatic perspective made homology a self-contained theory rather than a sequence of ad hoc constructions.