Alec Kriebel

A four-vertex counterexample to Knudson's persistence-gradient conjecture

· Zenodo record and DOI

Persistence pairs come from a boundary-matrix reduction. Gradient paths come from a matching on the simplices. This note gives a small filtration where a nonincident persistence pair has no gradient path in the specific matching obtained by retaining the incident persistence pairs.

Read the paper

Open the original PDF · Download the PDF · BibTeX

The PDF prints version 1.0.1 and 23 September 2026. Zenodo declares that publication date, was created on 24 September 2026 UTC, and supplies no version field. The version shown here identifies the printed manuscript.

The filtration and the counterexample

Take four vertices a, b, c, d and three edges cd, bd, ac, with no higher simplices. The space is the interval a–c–d–b. Add its nonempty simplices one at a time in this order:

a, b, c, d, cd, bd, ac.

Every vertex appears before its incident edges. Ordinary persistence reduction gives the finite pairs

P = {(d, cd), (c, bd), (b, ac)}.

Only d is a face of its paired edge. Retaining the incident pairs therefore gives the fixed matching

VP = {(d, cd)}.

The pair (b, ac) is nonincident, and both b and ac are critical in this matching. There are zero gradient paths from ac to b in VP. This refutes Conjecture 2 in Knudson’s contribution to Oberwolfach Report 29/2008, which predicts exactly one such path for every nonincident persistence pair.

Why the path stops

With vertex rows ordered a < b < c < d, the three reduced edge columns over the two-element field have these pivots:

Boundary reduction in the filtration order
EdgeReduced boundaryPivotPersistence pair
cdc + dd(d, cd)
bdb + cc(c, bd)
aca + bb(b, ac)

In the directed incidence graph, unmatched incidences point down from an edge to a vertex; matched incidences point up. Here the complete arrows are

ac → a, ac → c, bd → b, bd → d, d → cd, cd → c.

A path from ac can start only by descending to a or c. Both are unmatched vertices, so it stops there. The full reachable set is {ac, a, c}, which excludes b. The graph is acyclic, so the incident matching is a discrete gradient.

The same example works over every field. Orienting each edge from its alphabetically earlier vertex to its later one gives reduced columns d − c, c − b and b − a, all with pivot coefficient 1. The persistence pairs, matching and absence of a path remain the same. No minimality claim is made.

Cancellation changes the field

The earlier nonincident pair (c, bd) does have the unique path

bd → d → cd → c.

Reversing that path gives a different matching, V′ = {(d, bd), (c, cd)}. In this changed field the path

ac → c → cd → d → bd → b

now exists. The conjecture specifies the original fixed VP. A path appearing after cancellation does not supply a path in that original field. The note makes no counterexample claim against the ordered cancellation and surface simplification results of Bauer, Lange and Wardetzky.

Scope, credit and priority limits

The target is the incident submatching of the persistence pairing for a simplex-wise filtration. The counterexample does not require a lower-star filtration, a surface or a special placement of the interval. It concerns the fixed matching, rather than an updated sequence of cancellations.

Knudson’s conjecture supplies the question; the standard persistence-reduction method is due to Edelsbrunner, Letscher and Zomorodian. The paper distinguishes its fixed-field statement from the simplification results of Bauer, Lange and Wardetzky.

A bounded literature search does not establish first priority or current worldwide openness. The manuscript records inaccessible sources and related work listed as in preparation; those historical findings leave limits on any priority claim.

Original manuscript abstract

Knudson conjectured that every nonincident persistence pair is joined by exactly one gradient path in the discrete vector field obtained by retaining the incident persistence pairs. We give a simplex-wise filtration of a triangulated interval with four vertices and three edges for which one such pair has no gradient path. This disproves Conjecture 2 in his contribution to Oberwolfach Report 29/2008. The proof consists of three boundary reductions and an inspection of the matching. The same example works over any field.

Original files and citation

The original files are provided in the Zenodo record’s order:

  1. SHA256SUMS (170 bytes) · Zenodo original checksums
  2. source-and-verification.zip (105,727 bytes) · Zenodo original ZIP
  3. paper.pdf (49,210 bytes) · Zenodo original PDF

Alec Kriebel. A four-vertex counterexample to Knudson's persistence-gradient conjecture. Printed manuscript version 1.0.1, 23 September 2026. doi:10.5281/zenodo.22929556. Download BibTeX.

The manuscript and explanatory text use CC BY 4.0. Python and website code use the MIT License. Cited third-party works retain their own rights.

Status and collaboration

The candidate construction came from the author’s prior AI-assisted work. OpenAI Codex and separate AI agents assisted preparation and checking. This preprint is unrefereed; internal AI-assisted checks are not independent human peer review, journal acceptance or formal proof-assistant certification. The proof is written in the note and does not depend on running the supporting programs.

Independent verification, substantive mathematical extensions and development toward a journal submission would be useful contributions. Collaboration and any authorship would reflect actual scholarly contributions.

Related literature

  • K. P. Knudson, Persistent homology and discrete Morse theory, in Computational Algebraic Topology, Oberwolfach Report 29/2008, 1628–1630. Conjecture 2, p. 1629. Publisher DOI.
  • H. Edelsbrunner, D. Letscher and A. Zomorodian, Topological persistence and simplification, Discrete & Computational Geometry 28 (2002), 511–533. Publisher DOI.
  • U. Bauer, C. Lange and M. Wardetzky, Optimal topological simplification of discrete functions on surfaces, Discrete & Computational Geometry 47 (2012), 347–377. Publisher DOI.