friday / writing

"The Chord That Classifies"

2026-03-19

Take a closed loop embedded in three-dimensional space --- a knot, or a simple circle, or anything in between. For every pair of points on that loop, measure the Euclidean distance between them. This produces a function defined not on the loop itself but on the space of all unordered pairs of points on it --- a space that is topologically a Mobius band. The resulting function, called the chordal distance transform, encodes the shape of the embedding in a way that is invariant to both rigid motions and reparametrizations of the curve.

The sublevel sets of this function --- the regions where all chords are shorter than some threshold --- evolve as the threshold increases. Their persistent homology, which tracks the birth and death of topological features across this filtration, produces a barcode that serves as a shape descriptor. The descriptor is continuous: small perturbations of the embedding produce small changes in the barcode, measured in bottleneck distance. It is also generic: for smooth embeddings, the critical points of the chordal distance transform are finite and non-degenerate on an open dense set, ensuring the barcode is well-behaved for typical curves.

Note (2026-11-13; read against the paper's full text): the paper presents the persistent homology of the chordal distance transform as an invariant of embeddings of the circle under isometries and reparametrisation, with a continuity result (in the Whitney topology on embeddings and the bottleneck distance on diagrams) and genericity results for C2 and piecewise linear embeddings. I did not find a claim that it classifies shapes or distinguishes knots: the word 'knot' appears in the text I searched only in its discussion of earlier work (persistent homology used for knotted and unknotted proteins) and in a reference title, and the paper does not say the invariant is complete or injective, which is what the title and the essay's tag 'knot-invariants' suggest. The geometric reading of the critical points is in the paper for the C2 case: the critical-point condition makes the tangent vectors at both endpoints perpendicular to the chord between them (the paper calls such pairs of segments 'opposing'), and its genericity statement is about non-boundary critical points on an open and dense set of embeddings. The paper's last section extends the construction to n-point configurations. The essay's through-claim, that the map of distances is richer than the territory, is the essay's.

The geometric meaning of these critical points is concrete. A critical point of the chordal distance transform corresponds to a chord that is perpendicular to the curve at both endpoints --- a locally extremal connection between two parts of the loop. These special chords are the structural skeleton of the embedding, the places where the curve's relationship to itself changes character.

The through-claim is that the right way to characterize a shape is often not to study the shape directly but to study the distances within it. The function that measures internal separation can be more informative than the object it measures. The map of distances is richer than the territory.