Persistent Homology

The primary visual and quantitative outputs of persistent homology. A barcode is a set of horizontal lines, each representing a topological feature (e.g., a connected component or loop). The line's start and end points correspond to the birth and death scales (filtration parameter ε) at which the feature appears and disappears. A persistence diagram plots these (birth, death) pairs as points in a 2D plane. Points far from the diagonal represent persistent features (long-lived, structurally significant), while points near the diagonal are considered topological noise.

The core computational process that builds a nested sequence of topological spaces from the data. Common filtrations include:

• Vietoris-Rips Filtration: For a point cloud, builds a simplicial complex where a k-simplex is formed if all pairwise distances between its vertices are less than a scale parameter ε. As ε increases, the complex grows.
• Čech Filtration: Similar but uses intersections of balls of radius ε/2 centered on points; it is more computationally expensive but has stronger theoretical guarantees.
• Alpha Filtration: For point clouds, uses the Delaunay triangulation to create a more geometrically accurate complex.
• Sublevel Set Filtration: For functions or grayscale images, the space is the set of points where the function value is below a threshold ε.

Persistent homology computes homology groups at each scale in the filtration. These algebraic structures classify topological features by their dimension:

• H₀: Counts connected components.
• H₁: Counts 1-dimensional loops or cycles.
• H₂: Counts 2-dimensional voids or cavities. The Betti numbers (β₀, β₁, β₂, ...) are the ranks of these homology groups, providing a count of features in each dimension. Persistent homology tracks how these Betti numbers change with the scale ε.

Vectorized representations derived from persistence diagrams, enabling the use of standard machine learning algorithms.

• Persistence Landscape: Transforms a persistence diagram into a sequence of piecewise-linear functions that are easier to integrate into statistical frameworks. It provides a functional summary of topological activity.
• Persistence Image: Creates a 2D histogram by placing a kernel (e.g., Gaussian) over each point in the persistence diagram and summing them. This yields a fixed-size vector that is stable to small perturbations in the data. These are crucial for tasks like topological feature extraction for classification or regression models.

Metrics used to quantify the similarity or difference between two persistence diagrams, essential for statistical analysis and hypothesis testing.

• Bottleneck Distance: The maximum distance between matched points in a bijection between two diagrams, where points can also be matched to the diagonal. It measures the worst-case difference.
• Wasserstein Distance (p-th): The cost of the optimal matching between points, raised to the p-th power and summed. The 1-Wasserstein and 2-Wasserstein distances are common. They provide a more holistic measure of distributional difference. These distances satisfy stability theorems, guaranteeing that small changes in input data lead to bounded changes in the diagrams.

In Synthetic Data Fidelity Assessment, persistent homology provides a powerful, geometry-aware tool for comparison. It can detect structural mismatches that simple statistical tests miss.

• Process: Compute persistence diagrams for both the real dataset and the synthetic dataset.
• Comparison: Calculate the Wasserstein distance between the diagrams. A small distance indicates the synthetic data has successfully captured the multiscale topological 'shape' of the real data.
• Insight: It can reveal if synthetic data fails to replicate specific holes (H₁) or clusters (H₀) present in the real data's underlying manifold, indicating a synthetic-to-real gap in data geometry.

Topological Data Analysis is a field that applies principles from algebraic topology to study the 'shape' of data. It focuses on features that are invariant under continuous deformation, such as connected components, loops, and voids. Unlike traditional statistical methods, TDA provides a multiscale, coordinate-free view of data structure, making it robust to noise and useful for high-dimensional datasets where geometric intuition fails.

A simplicial complex is a combinatorial object used to build a topological space from discrete data points. It is constructed from simplices:

• A 0-simplex is a point.
• A 1-simplex is a line segment connecting two points.
• A 2-simplex is a filled triangle.
• A 3-simplex is a solid tetrahedron. Complexes are built by gluing these simplices together along their faces. In TDA, a common construction is the Vietoris-Rips complex, which connects points that are within a specific distance ε, forming the basis for computing persistent homology.

A filtration is a nested sequence of simplicial complexes, parameterized by a scale (e.g., a distance threshold ε). As ε increases from 0 to infinity, simplices are added: points appear first, then edges form as points connect, followed by triangles, and so on. This growing sequence captures how the topological features (components, loops, cavities) of the data appear (are 'born') and later merge or fill in ('die') at different scales. The persistent homology is computed directly from this filtration.

These are the two primary visual outputs of a persistent homology calculation.

• A barcode is a graphical representation where each topological feature (e.g., a connected component) is shown as a horizontal bar spanning from its birth scale to its death scale. Long bars represent persistent, significant features; short bars often indicate noise.
• A persistence diagram plots each feature as a point in a 2D plane, with birth on the x-axis and death on the y-axis. Points far from the diagonal (where birth=death) represent persistent features. This diagram is a stable summary under small perturbations of the data.

Betti numbers are integer topological invariants that count the number of independent k-dimensional holes in a topological space. For a given scale ε in a filtration:

• β₀ counts the number of connected components.
• β₁ counts the number of one-dimensional loops or cycles.
• β₂ counts the number of two-dimensional voids or cavities. In persistent homology, we track how these Betti numbers change over the filtration scale, providing a multiscale signature of the data's shape. A persistent feature is indicated by a range of scales where a specific Betti number is stable.

The Wasserstein distance (or Earth Mover's Distance) applied to persistence diagrams is a fundamental metric for comparing the topological signatures of two datasets. It measures the minimum cost of matching points between two diagrams, where the cost of matching a point to the diagonal (birth=death) is its persistence. This distance is stable, meaning small changes in the input data lead to small changes in the diagram distance. It is crucial for tasks like clustering datasets by shape or quantifying the topological difference between real and synthetic data distributions.