Is a persistent Betti number always an integer?

In the finite-dimensional field-valued setting described here, it is a vector-space dimension and therefore a nonnegative integer.

Can it exceed the Betti number at the later stage?

No. It is the dimension of an image inside the later homology space, so it cannot exceed that space's dimension. It also cannot exceed the earlier dimension.

Does a value of zero mean there are no features at either endpoint?

No. Both stages can have features even when no earlier class survives nontrivially to the later one.

AI Glossary / Data science and analytics / Advanced

Persistent Betti number

Definition

A persistent Betti number counts the independent homology classes present at an earlier filtration parameter that remain nonzero and independent at a later parameter. With field coefficients, it is the rank of the homology map between those two stages in a chosen dimension.

What it means

An ordinary Betti number describes topology at one stage: for example, the number of connected components or independent loops. A persistent Betti number asks a two-stage question: how much of the earlier topology survives into the later stage?

The answer is a nonnegative integer, but its interpretation requires a homological dimension and two parameter values. For zero-dimensional homology it concerns component information; for one-dimensional homology it concerns loop classes.

It is not simply the smaller of the two ordinary Betti numbers. Features present later may have been born after the earlier stage, and earlier classes may merge or become trivial. The connecting map determines survival, so counting the features at the two endpoints alone is insufficient.

How it works

Let K(s) be the earlier complex and K(t) the later one, with s no greater than t. Their inclusion induces a map from H(k, K(s)) to H(k, K(t)), where k specifies the homology dimension. The persistent Betti number is the dimension of that map's image over the chosen field.

In a finite single-parameter barcode using intervals [birth, death), the same value is obtained by counting bars with birth no greater than s and death strictly greater than t. These are the bars spanning the entire queried range, including both queried stages.

At equal parameters, the map is the identity, so the persistent Betti number reduces to the ordinary Betti number at that stage. With the earlier parameter fixed, extending the later parameter cannot increase the count of surviving earlier classes.

Software conventions, coefficient fields and the filtration's units should be recorded. A parameter might be radius, edge distance or image threshold, each giving a different meaning to the queried interval.

Three practical examples

A researcher counts how many component classes present at a fine scale remain distinct at a coarser scale.
An image-analysis team measures the number of loop classes surviving across a selected intensity-threshold range.
A machine-learning experiment uses values of a persistent Betti function as candidate features, while checking whether the representation improves held-out performance.

Think of it this way

Think of counting which items from an earlier checklist are still represented after a later consolidation. Newly added items do not count as surviving earlier items, and merged items may no longer contribute independently.

This is only an analogy. Homology classes are algebraic objects, and their linear dependence matters. A persistent Betti number is not a count of uniquely named physical objects or a literal audit trail of individual records.

Worked example

Suppose an invented one-dimensional barcode contains [0, 4), [1, 7), [3, 6) and [5, infinity). We ask for the persistent Betti number from s = 2 to t = 5.

The first bar is present at 2 but dies before 5, so it does not count. The second is present at 2 and remains alive at 5, so it counts. The third and fourth are born after 2, so neither contributes to this earlier-to-later survival count. The result is 1.

At parameter 2 there are two live bars, and at parameter 5 there are three. Taking the smaller endpoint count would give 2, which is wrong: only one earlier class survives independently across the whole interval.

If t were 7, the second bar would also be excluded under the right-open convention. Its death at 7 means it is not alive at that endpoint.

Illustrative case study

Illustrative fictional case: Summit Imaging produces a report comparing shape summaries across inspection images. An analyst calculates the minimum of the Betti counts at two thresholds and labels it “persistent features.”

A topology specialist explains that this ignores whether the later features are continuations of the earlier ones. The team changes the calculation to use the rank of the connecting map, or the equivalent count of spanning barcode intervals, and documents its endpoint convention.

It then checks whether the resulting measure relates to the inspection question using labeled examples. This fictional case demonstrates a definition error and its correction; it does not claim that persistent Betti numbers alone diagnose defects or establish a particular classification accuracy.

Common misconceptions and mistakes

Do not subtract or take the minimum of endpoint Betti numbers and assume the result measures persistence. Do not include features born after the earlier parameter simply because they are alive at the later one.

Avoid ignoring open and closed interval endpoints when reproducing counts. A persistent Betti number is not a significance level or a probability that a feature is real. Its value depends on filtration, preprocessing and coefficient field; comparisons require matching definitions.

Business use

Persistent Betti numbers can provide compact summaries for specialized shape analysis. They suit questions about features surviving across scales, rather than topology at one threshold.

Specify the range before interpreting results and assess sensitivity to nearby choices. If an AI model uses these counts, evaluate it against simpler baselines and retain the transformation details needed to reproduce them. For a non-specialist audience, explain the selected dimension and parameter units, and separate the mathematical count from any proposed business interpretation.

Check your understanding

A loop is born at parameter 4 and survives until 10. Does it contribute to the persistent Betti number from 2 to 6?

Answer: No. It was not present at the earlier parameter 2. Surviving at the later parameter is necessary but not sufficient for this two-stage count.

Related terms

Persistent homology group
Persistence barcode
Betti number
rank invariant
image of a linear map
filtration
homology class

Learn next

Read persistence barcode and image of a linear map. Draw four intervals with different birth and death values, choose two parameters and count only those spanning the query. Compare the result with ordinary Betti counts at each endpoint.

Sources

Robert Ghrist, barcode intervals and ranks of persistent homology groups: https://graphics.stanford.edu/courses/cs164-09-spring/Handouts/paper_barcodes.pdf

Chazal and Michel, persistent Betti numbers and modules: https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2021.667963/full

GUDHI, persistent Betti number and interval interfaces: https://gudhi.inria.fr/doc/latest/class_gudhi_1_1persistent__cohomology_1_1_persistent__cohomology.html

The barcode calculation and inspection case are original fictional examples.

Previous: Persistence module Next: Persistent contrastive divergence