Four-panel gallery: trefoil 3₁, figure-eight 4₁, 5₂ knot, and a Hopf-linked pair rendered from the actual hex lattice
Four examples extracted from the model: trefoil 3₁ (9 arcs, 3 crossings), figure-eight 4₁ (23 arcs, 6 crossings), 5₂ knot (26 arcs, 6 crossings), and a Hopf-linked loop pair. Over-strands are drawn continuous; under-strands show the canonical diagram gap.

Every loop is a genuine knot diagram

The tile's printed paths cross inside the tile at a single point. Because the printing is physical, one arc must pass over the other — and the model fixes this globally: the 2–4 arc is always over, the 3–5 arc is always under. This assignment is rotation-independent: rotating a tile changes which physical sides each arc connects, but does not change which arc is on top.

Consequence: every closed loop that visits a crossing tile simultaneously inherits an over/under label at that crossing. The loop, together with its crossing information, is precisely a knot diagram — a projection of a closed curve in 3-space onto the plane with explicit crossing data. The topology is not just an analogy; the invariants (determinant, Jones polynomial, genus) apply directly.

The 2–4 over 3–5 rule means the model is a knot-diagram generator. Every closed loop carries a knot type, automatically.

The smallest knotted loop is the trefoil 3₁, which requires a loop of at least 9 arcs spanning 6 tiles. The smallest loops (3 and 4 arcs, the triangle and rhombus) are always unknotted — they have no self-crossings. The first crossing that can contribute to a non-trivial knot appears at arc-length 6 (size-6 loops exist with 0 or 1 crossing), but the full trefoil requires 3 crossings simultaneously, which needs at least 9 arcs.

How common are knots?

2.5% of all loops are knotted
25.6% among loops with ≥3 crossings
9 arcs in the smallest knot (trefoil)
177 arcs: FWD half-knottedness length N₀

Most loops are short (the arc-length distribution is a power law s−2.11), so only a fraction are long enough to be knotted. But the knotting probability rises sharply with length: by s ≈ 177 arcs, half of all loops are knotted; by s ≈ 400 arcs, nine in ten are.

This is the Frisch-Wasserman-Delbruck (FWD) law:

P(unknot | s) ≈ e−s / N₀,  N₀ ≈ 177 arcs

The exponential form is proven for random polygons in 3-space; the measured decay constant N₀ ≈ 177 arcs is specific to this model. The complementary probability P(knotted | s) = 1 − e−s/177 is shown in the interactive widget below.

Three-panel figure: (a) P(knotted) sigmoid vs arc-length, (b) Seifert-genus histogram, (c) knot-type bar chart
Knot statistics from the model. (a) P(knotted | s): the sigmoid climb from ~0 at short lengths to near 1 for long loops, fitted to e−s/177. (b) Seifert-genus histogram among knotted loops — genus 1 (trefoil, fig-8) dominates. (c) Knot-type bar chart: 3₁ is overwhelmingly the most common, followed by 4₁, 5₂, 5₁, 6₁, 6₂, 6₃, with a rare exotic tail including composite knots with determinants up to 10146.
FRISCH-WASSERMAN-DELBRUCK EXPLORER

Drag the length slider to watch P(knotted | s) climb. The N₀ slider lets you explore how the decay constant changes the curve. The yellow dashed line marks N₀ — the length at which exactly half of loops are knotted.

50
177
P(knotted | s) = P(unknot) =

The small-knot census

Among knotted loops, the trefoil 3₁ is overwhelmingly dominant. The full zoo in order of frequency:

3₁ Trefoil
Most common
9 arcs min · 3 crossings
det = 3 · genus 1
Jones: −t⁻⁴+t⁻³+t⁻¹
4₁ Figure-eight
2nd most common
23 arcs min · 6 crossings
det = 5 · genus 1 · amphichiral
Jones: t⁻²−t⁻¹+1−t+t²
5₂
3rd
26 arcs min · 6 crossings
det = 7 · genus 2
Jones: −t⁻⁶+t⁻⁵−t⁻⁴+2t⁻³−t⁻²+t⁻¹
5₁ Torus
4th
≥30 arcs · 5 crossings
det = 5 · genus 2
torus knot T(2,5)
6₂
5th
6 crossings · det = 11
genus 2
6₁, 6₃
6th–7th
6 crossings
det = 9, 13 respectively
Exotic tail
Composite knots: granny, square, torus knots; determinants up to 10146; rare but present for very long loops.

The trefoil's dominance is not surprising: at any fixed crossing number, the trefoil has the smallest arc-length, so it is encountered at the leading edge of the knotting transition. The fig-8 is amphichiral (mirror of itself), so it is found with both handedness conventions equally — in this model only left-handed trefoils appear, since the 2–4-over-3–5 rule fixes the chirality of all crossings.

Build a knot arc by arc

The widget below draws real loops extracted from the model as proper knot diagrams. Under-strands show the canonical gap at each crossing. Step through arcs one at a time to see how the knot structure assembles — and how the crossing handedness forces a non-trivial topology.

The trefoil in 9 arcs across 6 tiles is the global minimum: no smaller loop in this model can be non-trivially knotted. The figure-eight is the smallest amphichiral knot and requires 23 arcs — 2.5× as long. The 5₂ knot also needs 6 crossings but a different arrangement of 26 arcs.

Slipknots: a knot hiding inside an unknot

Strictly, an open arc has trivial topology: it can always be isotoped to a straight line segment. "Is this open path knotted?" is ill-posed without a closure.

Our open paths, however, have both ends on the single outer boundary of the tiling. This gives a canonical exterior closure: join the two endpoints through the unbounded outer face without adding any new crossings. The closure is isotopy-unique and basepoint-invariant — it defines a genuine knot type for each open path.

Under the exterior closure:

Five-panel figure: FWD curve for open paths, slipknot illustration, ACN and writhe scaling plots
Open-path knot statistics. (a) FWD sigmoid for open paths under exterior closure, N₀ ≈ 186. (b) Slipknot illustration: the full strand closes to the unknot, but a 15-arc sub-window closes to the trefoil. (c–d) Average crossing number ACN ∝ arc-length (linear); |writhe| ∝ √(arc-length) — the two distinct entanglement scales.

Slipknots: ~50% of long unknotted strands

A slipknot is an open path that is globally unknotted under exterior closure, yet contains a knotted sub-arc. The knotted region "slips out" past the free endpoints of the full strand — like a real slipknot knot that unties when you pull on both ends.

Scanning windows along long strands in the model: 89% of long strands contain a localized knotted sub-arc (median knot size ≈ 18 arcs), and ~50% of globally-unknotted long strands are slipknots. The paradox is genuine: the full strand is unknotted, but a sub-arc of it closes to a trefoil.

ACN ∝ length · |writhe| ∝ √length

Even without closure, open strands accumulate entanglement. Two distinct measures capture different aspects:

These two scalings — linear ACN vs. √-writhe — reflect two different facets of the geometry: ACN counts the total entanglement (which grows with the number of crossing pairs = quadratic in length, but averaged over projections gives linear scaling), while writhe is the net chirality, which random-walks.

Long strands are deeply entangled even without closure. The knotting, slipknot, ACN, and writhe results all confirm that topological structure accumulates proportionally to length — there is no characteristic scale beyond which strands are "safe" from entanglement.