Mathematics
Topological Data Analysis with Persistent Homology
Quick fact
Persistent homology can detect the presence of a circle in a point cloud even when the data is contaminated with noise, by counting how long a loop survives across different connection scales.
Why this is interesting
If you looked at a cloud of points, could you tell if they form a circle or just a random blob? Turns out, noise can fool our eyes—but mathematics has a way to see through it.
Read the full explanation
Understanding Topological Data Analysis with Persistent Homology
Imagine you have a scatterplot of points that might come from a circle, but there is noise. If you connect every pair of points that are closer than some distance d, you get a network. For a small d, you have many isolated points; as d grows, points link into edges, then triangles, and eventually the whole structure fills in. At some stage, a loop appears (like a circle), and later it gets filled in. Topological data analysis (TDA) looks at this whole process across all possible d values. The key idea is that features that persist over a wide range of d are likely real structure, while those that appear and disappear quickly are noise. Persistent homology is the tool that tracks these features—connected components, loops, voids—and records their 'birth' and 'death' times. This gives a multi-scale summary of the data's shape.
A deeper explanation
Persistent homology formalizes the intuition of the growing network. Starting from a point cloud, we build a sequence of simplices (points, edges, triangles, and higher-dimensional analogs) by increasing a scale parameter, such as the distance threshold. This creates a filtration: a nested family of simplicial complexes K0 ⊆ K1 ⊆ … ⊆ Km. For each complex, we compute its homology groups, which count the number of connected components (H0), loops (H1), voids (H2), etc. As the scale increases, these features appear (birth) and disappear (death). The collection of intervals (birth, death) is the persistence barcode. A feature with a long interval is considered persistent, hence likely to be a true topological feature of the underlying shape. This works because topological invariants are robust to continuous deformations, and noise typically creates short-lived features. Persistent homology is a key tool in TDA, used in material science, biology, and network analysis, to extract shape-like structure from data.