Homology: An Overview

How is the shape of data understood without coordinates? How are holes in a molecule detected algorithmically? How are independent cycles counted in a network? Homology is the algebraic answer: it turns topological holes into numbers (Betti numbers) that can be computed and compared.

• **TDA:** Betti numbers β₀, β₁, β₂ characterize the shape of data; persistent homology is the main tool for point-cloud analysis
• **Bioinformatics:** shape of protein molecules, holes in RNA graph structures, cavities in crystal lattices
• **Computer graphics:** Betti numbers of a surface affect UV unwrapping and rendering; χ = V−E+F is used to validate mesh topology

Предварительные знания

• Covering Spaces

Simplicial Complexes

A **simplex** is the convex hull of affinely independent points: a 0-simplex is a point, a 1-simplex is an edge, a 2-simplex is a triangle, a 3-simplex is a tetrahedron. A **simplicial complex** K is a collection of simplices closed under taking faces: if σ ∈ K and τ is a face of σ, then τ ∈ K.

**Geometric realization:** a simplicial complex K defines a topological space |K|, the union of its simplices. Every 'reasonable' topological space is homeomorphic to |K| for some K (triangulation theorem). Simplices are the cellular building blocks of topology.

In ML and TDA, data is often turned into a simplicial complex: a point cloud becomes a Vietoris-Rips complex (connect points within distance ε). Varying ε and tracking how topological features appear and disappear gives **persistent homology**.

Chain Complexes and the Boundary Operator

To compute homology, build a **chain complex**: the chain group Cₙ is the free abelian group generated by n-simplices (with integer coefficients). The **boundary operator** ∂ₙ: Cₙ → Cₙ₋₁ maps an oriented n-simplex to an alternating sum of its (n−1)-dimensional faces.

**Boundary formula:** ∂ₙ([v₀, v₁, ..., vₙ]) = Σᵢ (−1)ⁱ [v₀, ..., v̂ᵢ, ..., vₙ], where v̂ᵢ means vertex vᵢ is omitted. **Fundamental property:** ∂ₙ ∘ ∂ₙ₊₁ = 0 - the boundary of a boundary is zero.

Homology Groups

Since ∂ ∘ ∂ = 0, we have im(∂ₙ₊₁) ⊆ ker(∂ₙ). The group of **n-cycles** Zₙ = ker(∂ₙ) consists of chains with zero boundary. The group of **n-boundaries** Bₙ = im(∂ₙ₊₁) consists of boundaries of (n+1)-chains. The **homology group** is Hₙ = Zₙ / Bₙ = ker(∂ₙ) / im(∂ₙ₊₁).

**Interpretation:** Hₙ counts 'n-dimensional holes'. H₀ = connected components (one component → H₀ = ℤ). H₁ = loops/tunnels (abelianization of π₁). H₂ = enclosed voids. Hₙ = {0} when there are no n-dimensional holes.

Betti Numbers and Euler Characteristic

**Betti numbers** βₙ = rank(Hₙ) count independent n-dimensional holes. β₀ = connected components, β₁ = loops/tunnels, β₂ = enclosed voids. The **Euler characteristic** is χ = Σₙ (−1)ⁿ βₙ = β₀ − β₁ + β₂ − ···.

**Betti numbers in TDA:** For a Vietoris-Rips complex built from data, β₀ counts clusters, β₁ counts loops, β₂ counts voids. Persistent Betti numbers (varying with scale ε) give 'topological barcodes' that are stable signatures of the data's shape.

Key Ideas

• **Simplicial complexes:** built from simplices; closed under faces; every space can be triangulated
• **Boundary operator ∂:** maps a simplex to an oriented alternating sum of faces; ∂∘∂ = 0
• **Homology groups:** Hₙ = ker(∂ₙ)/im(∂ₙ₊₁); H₀ = connectivity, H₁ = loops, H₂ = voids
• **Betti numbers:** βₙ = rank(Hₙ); χ = Σ(−1)ⁿβₙ is a topological invariant

Related Topics

Homology is the primary tool of computational topology:

• Fundamental Group — H₁(X) is the abelianization of π₁(X); homology is the commutative version of π₁
• Topological Data Analysis — Persistent homology = homology varying with scale; Betti numbers of data

Вопросы для размышления

• Compute the homology groups of the sphere Sⁿ for arbitrary n. State the pattern and explain it geometrically.
• Using the Mayer-Vietoris sequence, compute H*(S²) by writing S² = D² ∪_{S¹} D² (two disks glued along their boundary).
• In TDA, β₁ = 1 for data sampled from a circle and β₁ = 0 for data from a disk. How is this used to distinguish the shape of point clouds in practice?

Связанные уроки

• aa-20-homological