Applied Topology 6: Homology and Holes

This section covers computational methods for homology. Every finitely generated abelian group decomposes as Z^r ⊕ T, where T is a direct sum of cyclic groups with orders d₁ | d₂ | ... | d_k. The Smith normal form provides an algorithm to compute homology: boundary maps are represented as matrices with entries 0, 1, -1, and the Smith normal form determines Betti numbers β_q = rank(H_q) and torsion coefficients. For the torus with 22 vertices, 27 edges, 18 faces, χ = 22 - 27 + 18 = 3. Adequate subcomplexes allow computing homology by restricting to smaller subcomplexes. The Möbius strip has H₀ = Z, H₁ = Z, H₂ = 0, reflecting its non-orientable nature.

Smith Normal Form is a matrix reduction technique that transforms any integer matrix into a diagonal form where non-zero entries satisfy divisibility conditions, enabling efficient computation of homology groups for triangulated topological spaces; by reducing the boundary matrix to Smith Normal Form, one can determine the Betti numbers (the ranks of homology groups) by counting the number of zero columns (which gives the number of independent cycles) and the number of non-zero pivots (which gives the number of boundaries), thereby revealing the topological structure of the space.

Boundary matrices represent simplicial complexes as square matrices where entries indicate simplex inclusion in boundaries. For a triangle with vertices A,B,C and edges AB, BC, AC: Z₀ = span{A,B,C} (rank 3), B₀ = span{A+B, B+C, A+C} (rank 2), so H₀ = Z₀/B₀ has rank 1 (one connected component). Similarly, Z₁ = span{A+B+C} (rank 1), B₁ = 0, so H₁ = Z₁/B₁ has rank 1 (one 1-dimensional hole). This demonstrates how matrix operations reveal topological structure.

Homology computation uses matrix representations of boundary operators. The Smith Normal Form diagonalizes these matrices, revealing the structure of homology groups. The kernel dimension gives the rank, while diagonal entries reveal torsion. This provides a complete algebraic description. The algorithm runs in polynomial time for fixed dimension, making homology computationally tractable for practical applications.

This section establishes the computational framework for computing homology groups. It covers setting up boundary matrices Delta(n+1) and Delta(n), performing simultaneous row and column operations to reduce matrices while preserving homological structure, and identifying cycles (Z_n) and boundaries (B_n). The rank-nullity theorem connects simplex counts to homology dimensions. After Smith normal form reduction, matrix structure reveals homology group generators, with non-zero diagonal entries indicating free generators and zero entries signaling torsion. This computational pipeline transforms abstract topological concepts into concrete matrix operations.

The cup product is a bilinear operation that takes two cohomology classes from H^j(X) and H^k(X) and produces a cohomology class in H^(j+k)(X), defined by evaluating the first cochain on the first j vertices of a (j+k)-simplex and the second cochain on the remaining k vertices, then multiplying the results; this operation is well-defined because the boundary operator satisfies the product rule ∂(φ ∪ ψ) = ∂φ ∪ ψ + (-1)^j φ ∪ ∂ψ, ensuring that coboundaries produce coboundaries and cycles produce cycles.

Cohomology theory is built on singular chain complexes with face differentials, whose duals form cochain complexes with natural cup products. The Eilenberg-Steenrod axioms characterize ordinary cohomology: homotopy invariance, exactness, excision, additivity, and the dimension axiom. Removing the dimension axiom yields extraordinary cohomology. All ordinary cohomology theories with the same coefficient ring are equivalent. Cohomology naturally carries a graded ring structure through the cup product, defined via the diagonal map. For the torus T², H*(T²; Z) is an exterior algebra on two generators. Poincaré duality provides pairings Hk(M) × Hn-k(M) → R for orientable manifolds. Real projective spaces RP^n have cellular structures yielding cohomology rings H*(RP^n; Z/2) ≅ Z/2[x]/(x^{n+1}). The Hopf invariant one problem asks for which n there exists a non-trivial map f: S^{2n-1} → S^n. Using cohomology, the adjunction space X = S^n ∪f D^{2n} has Hn(X) generated by α and H2n(X) by β, with α² = kβ for some integer k (the Hopf invariant). Adams proved Hopf invariant one exists only for n = 1, 2, 4, 8, corresponding to real, complex, quaternion, and octonion division algebras.

The cup product is a bilinear operation on cohomology groups that multiplies two cochains (functions on chains) to produce a new cochain of higher dimension, defined by c ∪ η(σ) = c(σ<) × η(σ>) for a simplex σ, where σ< denotes the first k vertices and σ> denotes the last l vertices; this operation descends to well-defined products on cohomology classes because if c and η are cocycles, then c ∪ η is also a cocycle, and perturbing either cochain by a coboundary only changes the product by a coboundary, thereby giving cohomology the structure of a graded algebra over a field, which distinguishes spaces with identical homology but different cohomology algebras such as the torus versus the wedge of a sphere and two circles.

Homology with coefficients in an abelian group G is defined by tensoring the singular chain complex with G, yielding Cₗ*(X; G) = Cₗ*(X) ⊗ G. Cohomology uses the Hom functor: Cᵤ*(X; G) = Hom(Cₗ*(X), G), with differentials going in the opposite direction. These constructions satisfy properties analogous to integral homology: long exact sequences (with reversed arrows), excision, homotopy invariance, and disjoint union properties. For a point, H₀(point; G) = G. Cohomology with ring coefficients naturally forms a graded ring under cup product, providing additional multiplicative structure not available in homology. This framework generalizes to cohomology theories satisfying Eilenberg-Steenrod axioms.

This comprehensive segment covers the foundational concepts of cohomology in algebraic topology: (1) Key corrections from previous lectures: cohomology indices run opposite to homology (requiring N-1 instead of N), formulas only work for principal ideal domains (PIDs) not all rings like Z6, and dual chain complexes require careful sign conventions with (-1)^(dim(F)+1) for Koszul compatibility; (2) The cup product is a fundamental operation inducing a ring structure on cohomology groups, with commutativity established through the Alexander-Whitney map and transposition operation T(a⊗b) = (-1)^(|a|·|b|) b⊗a; (3) The cap product takes a cohomology class and homology class to produce a homology class, defined using the diagonal map and pairing between cohomology and homology; (4) The cohomology of a space is a graded commutative ring with multiplication given by the cup product, commutative up to sign: a∪b = (-1)^(|a|·|b|) b∪a; (5) The free graded commutative ring on generators is constructed by tensoring polynomial rings on even-degree generators with exterior algebras on odd-degree generators.

Topological Data Analysis (TDA) provides dimensionality reduction techniques for big data by encoding the shape of point cloud data into topological signatures called barcodes or persistence diagrams, which can then be translated into machine learning-compatible formats such as Betti curves, persistence landscapes, or persistence images to enable classification and other machine learning tasks.

Topological data analysis has proven successful across diverse applications: in medicine, identifying breast cancer subgroups with unique mutation profiles and excellent survival outcomes; in network analysis, understanding contagion spread and how long-distance connections affect disease transmission; in material science, discovering nanoporous materials with optimal cavity structures; and in sensor networks, detecting coverage gaps and evasion paths. These applications demonstrate how abstract topological concepts translate into practical solutions, revealing that data shape matters significantly across scientific disciplines.

Topological data analysis (TDA) provides a framework for understanding data structure using topology. Simplices are fundamental building blocks: zero simplexes are points, one simplexes connect two points, two simplexes connect three points with convex hulls. Simplicial complexes glue these together along faces, providing finite combinatorial representations of continuous spaces. The nerve theorem states that constructing a simplicial complex from an open cover captures all relevant topology. Category theory provides the formal framework: functors enable rigorous translation between domains, limits solve constraint systems, and colimits glue complex systems. Manifold learning assumes data lies on manifolds that locally resemble Euclidean space, with each patch having a different local metric. The topological dimension reduction algorithm transforms high-dimensional data into lower dimensions while preserving topological structure. For word embeddings, the distributional hypothesis represents words as multinomial distributions of co-occurring words. Topological clustering defines clusters as connected components of level sets of the underlying probability density function. Practical applications include dimension reduction on MNIST and Fashion-MNIST, metric learning with Siamese and triplet networks, and topological word embeddings.

Topological Data Analysis (TDA) applies algebraic topology concepts to data analysis, revealing hidden structures in complex datasets. Three key applications demonstrate its power: (1) Chemistry - analyzing molecular conformation spaces by mapping molecules to high-dimensional points and discovering geometric structures like spheres and cylinders; (2) Biology - studying grid cell activity patterns in brains to reveal toroidal structures underlying spatial perception; (3) Medicine - classifying breast cancer subtypes by transforming patient genomic data into high-dimensional point clouds and identifying branching tree structures that revealed previously unknown survival-positive patient groups. TDA bridges pure mathematics with practical data science by focusing on the underlying topology of data rather than raw numerical values.

Topological data analysis (TDA) provides a framework for extracting global structural properties from data using persistent homology. This approach captures topological features (connected components, loops, voids) and their lifetimes across different scales, vectorizing them for machine learning applications. For zero-dimensional homology, computation reduces to finding minimum spanning trees of complete graphs with pairwise distances as edge weights. The birth and death times of features form barcodes, which are stable under small perturbations for points in general position. TDA can be applied as feature extraction for molecular property prediction, as regularization in neural networks, and for optimizing point clouds to match target shapes. However, persistent homology diagrams lose substantial geometric information, meaning many different shapes can produce identical diagrams, causing optimization to converge to shapes with correct topology but incorrect geometry.