Does a persistence module have to come from homology?

No. The algebraic definition is broader, although homology of a filtration is a central application.

Are the maps always one-to-one?

No. A class can map to zero, and different earlier classes can become dependent at a later stage.

Does every persistence module have a barcode?

Not in unrestricted generality. Interval decompositions require appropriate conditions, and multi-parameter settings differ substantially from the standard finite single-parameter case.

AI Glossary / Data science and analytics / Advanced

Persistence module

Definition

A persistence module is an indexed family of algebraic objects together with maps that describe how information passes from one index to later indices. In the standard field-valued setting, the objects are vector spaces and the connecting maps are linear maps satisfying consistency rules.

What it means

In topological data analysis, each index often represents a scale or threshold. At one scale, data may form several connected components; at a larger scale, some components merge. A persistence module records both the information at each stage and how it is related across stages.

This is more than a list of feature counts. Two analyses can have the same number of features at every stage while differing in which features survive from one stage to the next. The connecting maps capture that difference.

Persistence modules provide the algebraic language behind many barcode and persistence-diagram constructions. They can also be studied abstractly. “Module” here is a mathematical term, not a software plug-in or a package of application code.

How it works

Write V(t) for the vector space at index t. For s no greater than t, there is a map from V(s) to V(t). The map from an index to itself must be the identity. Moving from s through t to u must give the same result as moving directly from s to u.

A filtration supplies a common example. Nested spaces induce maps on their homology, producing a persistence module in each homological dimension. The maps can send a class to zero or make previously independent classes dependent; they need not be invertible.

Under suitable single-parameter conditions, including the usual finite-dimensional finite-filtration setting, the module decomposes into interval modules. Each interval piece represents one algebraic feature present over an interval, yielding a barcode.

Such statements need their hypotheses. More general modules, especially those indexed by several independent parameters, cannot simply be assumed to have the same complete interval description as the familiar one-parameter case.

Three practical examples

A topological analyst uses a zero-dimensional module to track how components merge as a distance threshold increases.
A researcher studies one-dimensional homology maps to distinguish a loop that persists across scales from separate loops appearing at successive scales.
A developer uses the module framework to reason about what information a barcode represents and which assumptions its computation relies on.

Think of it this way

Think of a sequence of project handovers that records not only how many work items each team holds but also how earlier items are carried forward. Equal counts alone cannot show whether the same work survived.

The analogy has limits. Module maps are linear transformations between algebraic objects, not literal assignments of named tasks. Features can become dependent or map to zero, so their behavior is richer than moving cards between boards.

Worked example

Consider two invented modules at indices 0, 1 and 2. Both have a one-dimensional vector space at 0, a one-dimensional vector space at 1 and the zero space at 2. Thus their dimension lists are both 1, 1, 0.

In module A, the map from 0 to 1 is the identity. Its one feature survives from the first stage to the second and disappears at 2. Its interval description is [0, 2).

In module B, the map from 0 to 1 is zero. The feature at 0 does not survive into stage 1; the feature present at stage 1 is new. Its interval description is [0, 1) together with [1, 2).

All maps into stage 2 are zero in both examples. These assignments satisfy the composition rule. Despite identical dimensions at individual stages, the rank of the map from 0 to 1 is 1 in A and 0 in B. That is information a list of counts cannot recover.

Illustrative case study

Illustrative fictional case: Northbank Research compares two shape-analysis pipelines by checking that they produce the same Betti number at each sampled threshold. The team initially concludes that the persistence information is identical.

A reviewer asks whether the features are connected across thresholds in the same way. The researchers inspect the maps and interval summaries, recognizing that matching stage-by-stage counts does not establish matching persistence modules.

They update the comparison procedure to examine the claimed invariant directly and document the coefficient field and parameter conventions. This fictional case illustrates a mathematical verification issue; it does not assert that every pair of differing pipelines must produce different barcodes or identify a real software defect.

Common misconceptions and mistakes

Do not confuse a persistence module with its dimension at each index. Do not assume all connecting maps are inclusions merely because the underlying spaces are nested: the induced maps on homology can have kernels.

Avoid claiming interval decomposition without the relevant conditions. Do not infer that an algebraic feature has a unique physical location or explanation. Finally, a module captures selected information about the construction; it is not a complete description of every geometric property of the original data.

Business use

The module viewpoint helps researchers and specialist teams building or evaluating topological methods. It clarifies what a summary retains and prevents comparisons based only on visually similar counts.

For an applied project, ask the specialist to identify the parameter, the spaces being analyzed, the connecting maps and the resulting summary. Record the assumptions under which the summary is valid. If a vendor describes a multi-parameter method with a simple barcode, request a precise explanation of what has been reduced or summarized before interpreting it as a complete representation.

Check your understanding

Two modules have dimensions 2, 2 and 1 at the same three indices. Does this prove that they encode the same persistence information?

Answer: No. Their connecting maps can differ. One must examine the relationships between stages, not just the number of dimensions at each stage.

Related terms

Persistence barcode
Linear map
filtration
vector space
persistent homology
interval module
rank invariant
multiparameter persistence

Learn next

Read vector space and linear map before interval decomposition. Recreate the two small modules in the worked example and calculate the ranks of their maps. Explain why the stage-by-stage counts agree even though their barcodes differ.

Linear map

Sources

Edelsbrunner and Morozov, persistence-module definitions and homology maps: https://dav.lbl.gov/PDFs/PH-handbook.pdf

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

Robert Ghrist, Barcodes: The Persistent Topology of Data: https://graphics.stanford.edu/courses/cs164-09-spring/Handouts/paper_barcodes.pdf

The two modules and research-team case are original teaching examples.

Previous: Persistence barcode Next: Persistent Betti number