Is the persistent homology group always the same as ordinary homology?

At equal stages, the identity map gives the ordinary homology group. Between different stages, the image can be smaller than either endpoint space.

Can an earlier cycle disappear without its edges being removed?

Yes. Adding higher-dimensional pieces can make the cycle a boundary, so its homology class becomes zero.

Why do many implementations use a field?

Field coefficients allow homology and its maps to be handled with vector-space methods. Other coefficient systems involve additional algebraic considerations.

AI Glossary / Data science and analytics / Advanced

Persistent homology group

Definition

A persistent homology group is the image of the homology map from an earlier stage of a filtration to a later stage. It represents the earlier homology classes that remain visible in the later stage, identifying classes that have become equivalent or trivial there.

What it means

Ordinary homology describes a space at one scale. Persistent homology examines how that description changes across nested spaces. The persistent homology group for two stages focuses on what passes from the first to the second.

The word “group” refers to an algebraic structure, not a customer segment or a cluster of data records. With coefficients in a field, the homology groups are vector spaces, and the persistent group is a vector subspace of the later homology space.

Its dimension is the corresponding persistent Betti number. The group is the algebraic object; the Betti number is a numerical summary of its size in independent directions. This distinction separates a feature count from the definition of persistence.

How it works

Let K(s) be contained in K(t), where s is no greater than t. Inclusion induces a map H(k, K(s)) → H(k, K(t)). The persistent homology group is the image of this map: all later classes obtained from earlier classes.

For a reader familiar with chains, the same information can be expressed as earlier k-cycles modulo those earlier cycles that are boundaries by the later stage. A cycle may still be drawable in the larger complex while no longer representing a nonzero homology class because it has been filled in.

Classes newly born after s do not belong to the image unless represented by information already coming from the earlier stage. The construction therefore differs from simply taking all homology at t.

In computational applications, field coefficients make linear-algebra methods available. The coefficient choice, filtration and dimension remain part of the definition and can affect the resulting answer.

Three practical examples

A researcher studies which loops in an image-derived complex remain nontrivial after a threshold is increased.
A shape-analysis developer distinguishes the later homology space from the subspace inherited from an earlier resolution.
An AI research team uses persistent homology summaries as candidate inputs and investigates what information is lost when the groups are reduced to counts or other numerical features.

Think of it this way

Think of asking which themes from an early report are still represented in a later revision, excluding entirely new themes. Some earlier distinctions may disappear when the later account combines or resolves them.

The analogy has limits: homology classes obey exact algebraic rules, not subjective theme matching. A class can be represented by different cycles, and becoming zero means becoming a boundary within the mathematical construction, not being forgotten by a reader.

Worked example

Consider an invented filtration with two independent loop classes, a and b, at an early stage. Over a chosen field, the first homology space is two-dimensional, spanned by a and b.

At a later stage, the loop represented by a has been filled, while b remains unfilled. A new independent loop c has also appeared. The later homology space is again two-dimensional, now spanned by b and c.

The induced map sends a to zero and b to b. Its image is the one-dimensional space spanned by b. Therefore, the persistent homology group between these stages is isomorphic to the coefficient field as a one-dimensional vector space, and its persistent Betti number is 1.

Both ordinary Betti numbers are 2, but they describe different stage-specific information. The new loop c contributes to the later homology and not to this persistent group. The kernel records the earlier direction a that has become trivial.

Illustrative case study

Illustrative fictional case: Riverton Materials builds a prototype that examines pore-like structures in simplified images. A team member says that every visible outline retained in a later complex must represent a persistent loop.

A specialist demonstrates that adding filled regions can make a previously nontrivial cycle into a boundary even though its outline remains part of the complex. The team distinguishes drawn cycles from homology classes and verifies the maps in its summaries.

It also keeps the image-processing assumptions separate from physical claims about material properties. This fictional case illustrates an algebraic distinction. It does not establish a validated inspection method or claim that every topological loop corresponds to a physical pore.

Common misconceptions and mistakes

Do not identify the persistent homology group with the entire later homology group. Do not assume that a nested inclusion of spaces produces an injective map on homology.

Avoid treating every drawn cycle as a different class: cycles differing by a boundary represent the same homology class. Do not interpret “group” as a set of original observations. Finally, a nonzero persistent group establishes a feature of the chosen construction, not its practical cause, statistical significance or usefulness for prediction.

Business use

Specialist teams use this concept to design topological analyses and interpret software results. It identifies earlier features surviving into a later scale.

For an applied project, ask for the filtration, parameter units, coefficient field and homology dimension, then request an explanation of the summary delivered to decision-makers. If a model consumes only persistent Betti numbers, recognize that the numerical representation has a specific scope. Evaluate the complete pipeline against the business question and inspect representative data rather than relying on mathematical terminology as evidence of value.

Check your understanding

The later complex has three independent loop classes, but only one is in the image of the earlier homology map. What is the persistent Betti number for these stages?

Answer: It is 1. The persistent group contains the inherited image, not every loop class present later.

Related terms

Cycle
Image
Kernel
Persistent Betti number
Persistence module
Homology group
boundary
induced map
filtration

Learn next

Read cycle, boundary and image of a linear map. Recreate the worked example by drawing two loop outlines, filling one and adding another loop. Explain why the later loop count and the rank of the earlier-to-later map answer different questions.

Sources

Robert Ghrist, persistent homology as the image of an induced map: https://graphics.stanford.edu/courses/cs164-09-spring/Handouts/paper_barcodes.pdf

Edelsbrunner and Morozov, homology, cycles, boundaries and persistence: https://dav.lbl.gov/PDFs/PH-handbook.pdf

Chazal and Michel, persistence modules and their interpretation: https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2021.667963/full

The loop construction and materials-team case are original fictional illustrations.

Previous: Persistent failure model Next: Personal access token