---
title: Ribbonlength Crossing Number Problem
url: https://www.emergentmind.com/topics/ribbonlength-crossing-number-problem
type: topic
---

# Ribbonlength Crossing Number Problem

Searching arXiv for the cited ribbonlength papers and related work.
The ribbonlength crossing number problem asks how the scale-invariant amount of material needed to realize a knot or link as a folded ribbon in the plane grows as the combinatorial complexity of that knot or link, measured by crossing number, increases. In Kauffman’s folded ribbon model, one starts with a polygonal knot or link diagram in the plane, thickens it to a strip of fixed width, inserts fold lines at vertices, and studies the ratio of centerline length to ribbon width. The resulting invariant, usually written as folded ribbonlength or ribbonlength, is defined as an infimum over all folded ribbon realizations of a given knot or link type. The central question is whether universal inequalities of the form
$$
c_1\,\operatorname{Cr}(L)^{\alpha}\le \operatorname{Rib}([L])\le c_2\,\operatorname{Cr}(L)^{\beta}
$$
hold for all knots or links, and what the optimal exponents $\alpha$ and $\beta$ should be. Recent work has resolved the upper-bound side with a universal linear estimate and has shown that the universal lower-bound exponent must be $\alpha=0$, contrary to earlier conjectures of positive lower growth [2409.13572] [2512.12830].

## 1. Definitions and geometric model

A polygonal knot diagram is the image of a piecewise-linear immersion $K:S^1\to \mathbb{R}^2$ with crossing information. At each vertex $v_i$, consecutive edges $e_{i-1}$ and $e_i$ determine a fold angle $\theta_i\in[0,\pi]$. In Kauffman’s model, an oriented folded ribbon knot of width $w$, denoted $K_w$, is obtained by placing at each vertex with $\theta_i<\pi$ a fold line of length $w/\cos(\theta_i/2)$ perpendicular to the angle bisector, centered at the vertex, then connecting fold-line endpoints by boundary segments parallel to the edges so that each boundary lies at distance $w/2$ from the diagram. The ribbon must be immersed except at disjoint fold lines, and the crossing information must be consistent with both the diagram and the chosen overfold or underfold data [2512.12830].

The folded ribbonlength of a realization is the length-to-width ratio. In the notation used across the literature,
$$
\operatorname{Rib}(L_w)=\frac{\operatorname{length}(L)}{w},
$$
and the infimal folded ribbonlength of the knot or link type is
$$
\operatorname{Rib}([L])=\inf_{L_w\in [L]}\operatorname{Rib}(L_w).
$$
Most constructive papers normalize to $w=1$, so ribbonlength becomes the Euclidean length of the planar centerline. The crossing number $\operatorname{Cr}(L)$ or $c(L)$ is the minimal number of crossings among all regular projections of the knot or link type [2512.12830] [2409.13572].

Several papers emphasize that admissibility is nontrivial. Allowed folded ribbons require disjoint fold lines and consistency of crossing data; for sufficiently small widths, allowed realizations exist for a given polygonal diagram. Other formulations enlarge the admissible space. In the disk-diagram framework, ribbon loops are embedded into a larger space of $C^1$ immersed planar loops satisfying a weak separation condition and a unit-disk-in-each-region condition, which is technically useful because minimizers in disk space exist and are finite concatenations of arcs of unit circles and straight segments, but such minimizers need not satisfy the stricter ribbon conditions [2010.04188] [2005.13168].

## 2. Formulation of the ribbonlength–crossing number problem

The modern statement of the problem seeks universal constants $c_1,c_2$ and exponents $\alpha,\beta$ such that, for every knot or link,
$$
c_1\cdot \operatorname{Cr}(L)^{\alpha}\le \operatorname{Rib}([L])\le c_2\cdot \operatorname{Cr}(L)^{\beta}.
$$
Historically, Diao and Kusner conjectured $\alpha=\tfrac12$ and $\beta=1$. The upper exponent $\beta$ measures how efficiently ribbonlength can be realized as crossing number grows; the lower exponent $\alpha$ measures how much growth is forced across all knot and link types. A value $\alpha=0$ means that no positive power of crossing number can serve as a universal lower bound [2512.12830] [2409.13572].

The problem has several distinct interpretations in the literature. In constructive work, it is often treated as a search for explicit families of realizations whose length can be counted exactly or nearly exactly. In variational work, it becomes a question about properties of ribbonlength-minimizing diagrams and how their planar graph structure constrains crossing number. The latter viewpoint produces bounds of the form “a minimizer of length $\ell$ can have at most so many crossings,” rather than directly producing realizations with prescribed crossing number [2005.13168].

A persistent source of confusion is the difference between upper and lower growth statements. Linear or sublinear upper bounds for specific families do not imply a universal lower bound of the same order. Conversely, a universal lower bound would have to survive against families whose crossing number diverges while ribbonlength remains bounded. The current state of the subject separates these two sides sharply: the universal upper side is linear, while the universal lower side is constant-order in the exponent sense [2512.12830] [2509.18370].

## 3. Universal and family-specific upper bounds

The upper-bound side developed through a sequence of progressively sharper estimates. Early general bounds were quadratic, then $O(c^{3/2})$, and eventually linear. Alongside these, many specific knot families admitted better constants than the general theory.

| Result | Bound | Scope |
|---|---:|---|
| Tian | $\operatorname{Rib}(K)\le 2\,\operatorname{Cr}(K)^2+6\,\operatorname{Cr}(K)+4$ | all knots/links |
| Denne | $\operatorname{Rib}(K)\le 72\,\operatorname{Cr}(K)^{3/2}+32\,\operatorname{Cr}(K)+12\sqrt{\operatorname{Cr}(K)}+6$ | all knots/links |
| Kim–No–Yoo | $\operatorname{Rib}(K)\le 2.5\,c(K)+1$ | all knots/links |
| 2-bridge knots/links | $\operatorname{Rib}(K)\le 2c(K)+2$ | all 2-bridge knots/links |
| Alternating links with bipartite dual graph | $\operatorname{Rib}(L)\le \sqrt{3}\,c(L)$ | alternating links admitting such a diagram |

The universal linear bound
$$
\operatorname{Rib}(K)\le 2.5\,c(K)+1
$$
was proved using binary grid diagrams and bisected vertex leveling. The construction converts a minimal-crossing diagram into a binary grid diagram, reduces the block types, rearranges the diagram, and realizes the relevant blocks by “paper planes,” each contributing normalized length $2$. The combinatorial block count then yields the coefficient $2.5$ and additive constant $1$ [2409.13572].

Before that result, two main general methods were dominant. One used arc-presentations and spoked form to obtain explicit quadratic bounds with small leading constants, such as the odd-case bound
$$
\operatorname{Rib}(K)\le 0.64\,\operatorname{Cr}(K)^2+2.55\,\operatorname{Cr}(K)+2.03,
$$
while another used Hamiltonian lattice projections to prove
$$
\operatorname{Rib}(K)\le 72\,\operatorname{Cr}(K)^{3/2}+32\,\operatorname{Cr}(K)+12\sqrt{\operatorname{Cr}(K)}+6
$$
in general, and
$$
\operatorname{Rib}(K)\le 9\,\operatorname{Cr}(K)^{3/2}+8\,\operatorname{Cr}(K)+6\sqrt{\operatorname{Cr}(K)}+6
$$
for minimally Hamiltonian diagrams [2010.03611].

Special families long supplied the strongest evidence for $\beta=1$. For 2-bridge knots and links, an explicit rational-tangle construction gives
$$
\operatorname{Rib}(K)\le 2c(K)+2,
$$
with the additive $+2$ coming from the denominator closure [2208.03669]. Earlier family-specific work established
$$
\operatorname{Ribbonlength}(w)\le 2\,\operatorname{Cr}(T(2,q))
$$
for $(2,q)$ torus knots, and
$$
\operatorname{Ribbonlength}(w)\le 2\,\operatorname{Cr}(T_n)+2
$$
for twist knots [2010.04188]. More recent small-crossing constructions improved these constants to
$$
\operatorname{Rib}(T(2,q))=q+3,\qquad \operatorname{Rib}(T_n)=n+6,
$$
and, for 3-strand pretzel links,
$$
\operatorname{Rib}(P(p,q,r))=|p|+|q|+|r|+6,
$$
which in alternating same-sign cases becomes $\operatorname{Rib}\le \operatorname{cr}+6$ [2510.16190].

For alternating links admitting an alternating diagram whose checkerboard dual graph is bipartite, a different construction based on rotated circular three-page presentations yields
$$
\operatorname{Rib}(L)\le \sqrt{3}\,c(L).
$$
The constant $\sqrt{3}$ arises because the construction uses exactly $3n$ arcs for $n=c(L)$ crossings and assigns one equilateral triangle of unit-width contribution $1/\sqrt{3}$ to each binding point. This bound is sharp for the Hopf link, for which the exact ribbonlength is
$$
\operatorname{Rib}(\text{Hopf})=2\sqrt{3}
$$
[2601.10278].

## 4. Lower bounds, the exponent $\alpha$, and the collapse of the universal lower-growth conjecture

The lower-bound side of the problem was long guided by the conjecture that $\alpha=\tfrac12$. That conjecture became untenable once infinite families were found whose crossing number grows without forcing ribbonlength to grow. The decisive logic is simple: if a universal lower bound
$$
c\cdot \operatorname{Cr}(L)^\alpha\le \operatorname{Rib}([L])
$$
held with $\alpha>0$, then any family with unbounded crossing number and uniformly bounded ribbonlength would violate it [2512.12830].

For knots, Denne–Patterson established this phenomenon using twist knots and $(2,q)$ torus knots. They showed that any $(2,q)$-torus knot can be constructed with folded ribbonlength at most $8\sqrt{3}+\epsilon$, and that twist knots admit uniform bounds at most $9\sqrt{3}+2+\epsilon$ in the odd case and $8\sqrt{3}+2+\epsilon$ in the even case, all independent of the number of half-twists. These constructions use an “escape accordion” and repeated half-wraps at fold angle $\pi/3$, with the small terms absorbed by taking the spacing parameter $d$ sufficiently small [2509.18370].

For links, the same phenomenon was extended to pretzel families. The central result is that any 3-strand pretzel link $P(p,q,r)$ satisfies
$$
\operatorname{Rib}([P(p,q,r)])\le \frac{55}{\sqrt{3}}\approx 31.755,
$$
with the sharper parity-refined bound
$$
\operatorname{Rib}([P(p,q,r)])\le \frac{49}{\sqrt{3}}\approx 28.291
$$
when one of $p,q,r$ has parity opposite to the other two. More generally, any $n$-strand pretzel link satisfies
$$
\operatorname{Rib}([P(p_1,\dots,p_n)])\le \frac{18n+1}{\sqrt{3}}.
$$
These bounds depend only on the number of strands, not on the twist parameters. Since pretzel links contain infinite subfamilies with arbitrarily large crossing number, such as alternating same-sign examples with
$$
\operatorname{Cr}(P(p,q,r))=|p|+|q|+|r|,
$$
the only possible universal lower exponent for links is $\alpha=0$ [2512.12830].

This settles the universal lower-bound exponent differently from the universal upper exponent. The current picture is therefore asymmetric: $\beta=1$ is valid universally on the upper side, while $\alpha=0$ is forced universally on the lower side. This does not preclude stronger lower bounds for restricted subclasses. The pretzel-link paper explicitly notes that refined lower bounds under alternation, positivity, or additional geometric constraints remain of interest [2512.12830].

## 5. Principal constructions and the geometry behind the constants

Several distinct construction paradigms dominate the subject. Each controls ribbonlength by introducing a local gadget whose contribution can be counted exactly.

One major paradigm is the escape-accordion method. Here one fixes width $w=1$ and chooses fold angle $\theta=\pi/3$, because prior work shows that $\pi/3$ minimizes local ribbonlength at folds and crossings. An accordion consists of alternating left-right overfolds with constant spacing $d$, producing a narrow zig-zag stack. A half-wrap then winds one ribbon around another thin accordion while preserving the same angle and spacing. In the pretzel-link work, a key theorem states that, apart from end effects,
$$
(k\text{ half-twists})=\frac{12}{\sqrt{3}}+\epsilon
$$
for any $\epsilon>0$ and any number $k$ of half-twists, by choosing $d$ sufficiently small. This is the geometric reason twist regions can have essentially constant cost independent of crossing number [2512.12830].

The pretzel constants are obtained by adding the fixed costs of twist regions and joining patterns. For three strands, the three twist bands contribute approximately $3\cdot (12/\sqrt{3})=36/\sqrt{3}$. Two joins contribute $2\cdot (6/\sqrt{3})$, and the last contributes $7/\sqrt{3}$, totaling
$$
\frac{55}{\sqrt{3}}.
$$
For $n$ strands, the formula
$$
\frac{12n+6(n-1)+7}{\sqrt{3}}=\frac{18n+1}{\sqrt{3}}
$$
follows from $n$ twist regions, $(n-1)$ interior joins, and one closing join. The paper is explicit that these are upper bounds and does not claim optimality [2512.12830].

A second paradigm is the wrap method for small-crossing families. This method concentrates all half-twists into a single square “twist box,” uses only fold angles $0$ and $\pi/2$, and reuses square units of area to minimize added length. Its guiding lemma is that a twist box with $n$ half-twists uses at least $|n|+2$ units of ribbonlength and produces $|n|+3$ sticks in the centerline. From this, one obtains the exact formulas
$$
\operatorname{Rib}(T(2,q))=q+3,\qquad \operatorname{Rib}(T_n)=n+6,\qquad \operatorname{Rib}(P(p,q,r))=|p|+|q|+|r|+6.
$$
These formulas give the best known upper bounds to date for small-crossing knots in those families, but they are not designed to be uniform in the parameters; by contrast, the escape-accordion constructions are tailored precisely to eliminate dependence on crossing number at large parameter values [2510.16190].

A third paradigm is graph- and diagram-based. Binary grid diagrams, bisected vertex leveling, arc-presentations, lattice embeddings, and rotated three-page presentations all translate knot diagrams into combinatorial layouts where every local piece has a controlled ribbonlength cost. This suggests that much of the problem is a question of finding a diagrammatic normal form that minimizes the cost per crossing or per twist region. A plausible implication is that improvements in the universal constant on the upper side may depend less on new inequalities and more on more efficient local gadgets for standard diagrammatic pieces [2409.13572] [2601.10278].

## 6. Variational approaches, minimizers, and diagrammatic subtleties

Not all progress comes from explicit folded constructions. A variational approach embeds ribbon diagrams into a larger disk-diagram space. In this setting, each complementary region must contain an open disk of radius $1$, minimizers exist, and every minimizer is a finite $C^1$ concatenation of unit-circle arcs and straight segments. If such a disk-space minimizer is also ribbon, then it solves the ribbonlength problem in ribbon space [2005.13168].

This framework yields a universal crossing bound for a ribbon- or disk-length minimizing planar projection of total core length $\ell$:
$$
c\le \frac{\ell}{\pi}-\frac12\left(\sqrt{1+\frac{4\ell}{\pi}}-1\right).
$$
In ribbonlength variables, using the width-$2$ normalization of that paper,
$$
c \le \frac{2\,\mathrm{Rib}}{\pi}-\frac12\left(\sqrt{1+\frac{8\,\mathrm{Rib}}{\pi}}-1\right).
$$
The proof is isoperimetric: a diagram with $c$ transverse 4-valent crossings has $c+1$ bounded complementary regions, each containing a unit disk and therefore contributing area at least $\pi$ and boundary length at least $2\pi$ [2005.13168].

This approach also clarifies an important subtlety: disk-space minimizers can fail to be ribbon minimizers because they violate the separation or non-overlap conditions. The figure-eight knot and Whitehead link provide explicit cases where a disk-space length minimizer exists but is not ribbon. Thus, “minimizer in a relaxed space” and “true folded ribbon minimizer” are not interchangeable notions [2005.13168].

Another subtlety concerns equivalence. A survey of folded ribbon knots distinguishes knot diagram equivalence, topological ribbon equivalence, and ribbon link equivalence. Fold choices can change ribbon linking number and can alter achievable ribbonlength. The survey also records that, even for 3-stick unknots, the fold pattern matters: if all three folds are of the same type, the minimal ribbonlength equals $3\sqrt{3}$, while if exactly one fold differs from the other two, the minimal ribbonlength is at most $\sqrt{3}$. This shows that ribbonlength is sensitive not only to the diagram but also to the ribbon structure imposed on it [1807.00691].

## 7. Small-crossing data, exact values, and open directions

For knots and links with small crossing number, the strongest upper bounds are typically family-specific rather than universal. Specialized formulas such as $\operatorname{Rib}(T(2,q))=q+3$, $\operatorname{Rib}(T_n)=n+6$, $\operatorname{Rib}(P(p,q,r))=|p|+|q|+|r|+6$, and $\operatorname{Rib}(K)\le 2c(K)+2$ for 2-bridge knots often beat the general bound $2.5c(K)+1$ in low-crossing regimes [2510.16190] [2208.03669].

A few exact or conjecturally exact values are especially prominent. The Hopf link satisfies
$$
\operatorname{Rib}(\text{Hopf})=2\sqrt{3},
$$
proved by combining a six-equilateral-triangle construction with a lower bound from the aspect ratio of embedded paper Möbius bands [2601.10278]. The trefoil admits constructions of ribbonlength $6$ in several frameworks, including $(2,q)$ torus constructions [2010.04188] [2510.16190]. For the figure-eight knot, the wrap method gives
$$
\operatorname{Rib}(T_2)=8,
$$
and that value is conjectured to be the infimum; earlier constructions gave upper bound $10$ [2510.16190] [2010.04188].

The main unresolved questions now concern sharp constants and restricted lower bounds rather than existence of power-law upper or lower exponents. On the upper side, it remains open whether the universal linear constant $2.5$ can be significantly improved, and whether a $\sqrt{3}$-type coefficient can be extended from the bipartite-dual alternating class to all alternating links or beyond [2601.10278] [2409.13572]. On the lower side, the universal exponent is settled at $\alpha=0$, but family-specific lower bounds under alternation, positivity, adequacy, homogeneity, or other geometric constraints remain open [2512.12830]. The disk-space framework further suggests a program of refining the isoperimetric crossing bound by incorporating more structure of cs minimizers or alternating diagrams [2005.13168].

Taken together, the literature shows that the ribbonlength crossing number problem is no longer a single asymptotic conjecture but a stratified subject. Universally, ribbonlength is at most linear in crossing number and cannot be bounded below by any positive power of crossing number. Within specific families and diagram classes, however, the constant factors, the exact infima, and the mechanisms of efficiency remain highly nontrivial, with constructions ranging from grid and lattice models to escape accordions, wrap layouts, and three-page presentations [2512.12830] [2409.13572].

Source: https://www.emergentmind.com/topics/ribbonlength-crossing-number-problem