---
title: Three-Page Link Presentations
url: https://www.emergentmind.com/topics/three-page-presentation
type: topic
---

# Three-Page Link Presentations

Searching arXiv for the provided papers and closely related work on “three-page presentation” to ground the article and resolve the topic’s usage across domains.
arXiv search query: "2501.03936 OR 2601.12846 OR 2507.13633 OR \"three-page presentation\" OR \"three-page presentations\""
A three-page presentation is, in knot theory, an embedding of a link into exactly three pages of an open book, each page containing a finite collection of pairwise disjoint properly embedded arcs whose union is the link. The associated three-page index $\alpha_3(L)$ is the minimum number of arcs, equivalently the minimum number of binding points on the axis, among all such presentations of a link $L$ [2507.13633]. A complementary diagrammatic formulation replaces the binding axis by a binding circle in a planar diagram and encodes the three pages by outside arcs, inside-over arcs, and inside-under arcs [2601.12846]. The term also appears in an unrelated computational setting, where a three-slide presentation is generated from a document by an edit-based workflow; that usage concerns document-to-slides synthesis rather than link embeddings [2501.03936].

## 1. Definition and relation to arc presentations

In the open-book formulation, $\mathbb{R}^3$ is viewed as a book whose binding axis is the $z$-axis and whose pages are half-planes $P_\theta=\{(r,\phi,z)\in\mathbb{R}^3:\phi=\theta,\; r\ge 0\}$. An arc presentation of a link embeds $L$ into a union of distinct pages $P_1,\dots,P_n$ with exactly one properly embedded arc on each page; the smallest such $n$ is the arc index $\alpha(L)$. A three-page presentation modifies this constraint by fixing the number of pages to be exactly three while allowing several disjoint arcs on the same page [2507.13633].

This distinction is fundamental. In the classical arc-presentation setting, different arcs must lie on different pages. In the three-page setting, several disjoint arcs may occupy a single page, but the ambient decomposition has only three pages. The invariant $\alpha_3(L)$ therefore measures a different notion of complexity from $\alpha(L)$: it records how efficiently a link can be organized when page count is fixed and arc count becomes the optimization target [2507.13633].

A common misconception is to identify the three-page index with the ordinary arc index. The available results do not support that identification. Instead, the basic relation stated for all links is $\alpha(L)\le \alpha_3(L)$, since any three-page presentation with $k$ arcs can be spread into an arc presentation with $k$ pages [2507.13633].

## 2. Planar and circular formulations

Three-page presentations admit a planar encoding that is often more convenient for construction and counting. One draws a planar link diagram $D$ on $S^2\cong\mathbb{R}^2\cup\{\infty\}$ together with a simple closed curve $\gamma\subset S^2$, called the binding circle, such that all crossings of $D$ lie in the interior of $\gamma$, $\gamma$ meets $D$ transversely in finitely many points, each arc cut by $\gamma$ either lies outside $\gamma$ or lies inside and is exclusively over-crossing or exclusively under-crossing, and no two arcs of the same type are adjacent along $\gamma$ [2601.12846].

The three pages are then read off directly from the position and crossing type of the cut arcs. Page 1 consists of arcs outside $\gamma$; Page 2 consists of inside-arcs that go over at every crossing; Page 3 consists of inside-arcs that go under at every crossing. The number of intersection points $|\gamma\cap D|$ equals the total number of arcs in the corresponding three-page presentation [2601.12846].

This circular model is not merely a visualization device. It converts the spatial embedding problem into a diagrammatic optimization problem: choose $\gamma$ so that the induced arc count is as small as possible while respecting the crossing-type alternation conditions. That reformulation underlies the strongest general upper bounds presently stated in the supplied sources.

## 3. Existence and basic inequalities

Dynnikov’s fundamental result is that every tame link admits a three-page presentation. An alternative proof for every non-split link proceeds from maximal overpasses in a planar diagram. If a diagram has no overpasses then the link is the trivial knot, which admits a three-page presentation of three arcs. Otherwise, one slides all maximal overpasses onto a single straight line in the plane, nests those that overlap, and then pushes the overpasses into a perpendicular half-plane. The resulting embedding occupies exactly three pages: two pages carry the under-arcs on the two sides of the line, and the third page carries the overpasses [2507.13633].

This construction yields immediate inequalities. If $br(L)$ denotes the bridge number, then in any three-page presentation each page contains at least $br(L)$ arcs, so $\alpha_3(L)\ge 3\,br(L)$. Together with $\alpha(L)\le\alpha_3(L)$, this places the three-page index between two classical invariants that reflect different decompositional constraints [2507.13633].

These inequalities clarify what the invariant does and does not measure. It is not simply a reparameterization of bridge number or arc index. Rather, it interpolates between diagrammatic overpass complexity and open-book complexity under the fixed-page constraint. This suggests a structural role for $\alpha_3(L)$ in comparing link presentations that privilege different ambient decompositions.

## 4. Diagrammatic bounds and the Hopf-link equality case

A reduced, non-split link diagram $D$ with $n=c(D)$ crossings determines a cellular decomposition $X(D)$ of $S^2$ whose $0$-cells are the crossings, $1$-cells are the arcs between crossings, and $2$-cells are the complementary regions. The key construction is to choose a connected contractible subcomplex $Y\subset X(D)$ and take the binding circle to be the boundary of a regular neighborhood $\partial N(Y)$. Counting intersections of $\partial N(Y)$ with $D$ then produces upper bounds for $\alpha_3(L)$ [2601.12846].

If $Y$ is any spanning tree of the $1$-skeleton, then a counting argument gives $|\partial N(Y)\cap D|\le 3n+1$, hence $\alpha_3(L)\le 3\,c(L)+1$. If $Y$ is enlarged to an extended spanning tree, meaning that it is connected and contractible, contains all $0$-cells, and no two $2$-cells in $Y$ share a common $1$-cell, then each included face removes exactly one intersection point. If $Y$ contains $m$ faces, the bound becomes $\alpha_3(L)\le (3\,c(L)+1)-m$ [2601.12846].

| Bound or characterization | Statement | Scope |
|---|---|---|
| Spanning-tree bound | $\alpha_3(L)\le 3\,c(L)+1$ | General |
| Extended spanning-tree bound | $\alpha_3(L)\le (3\,c(L)+1)-m$ | With $m$ added faces |
| Improved bound | $\alpha_3(L)\le 3\,c(L)-1$ | Non-split, nontrivial $L\neq$ Hopf link |
| Equality case | $\alpha_3(L)=3\,c(L)$ exactly for split unions of Hopf links | Complete characterization |

The improved theorem states that if $L$ is a non-split, non-trivial link other than the Hopf link, then
$$
\alpha_3(L)\le 3\,c(L)-1.
$$
The proof outline given in the source enlarges a spanning tree by two non-adjacent faces, so that the extended spanning tree has $m\ge 2$ faces and the refined count yields the stated inequality. The equality case is also sharp and explicit: for the Hopf link one has $c=2$ and $\alpha_3=6=3\times 2$, and more generally $\alpha_3(L)=3\,c(L)$ occurs exactly when $L$ is a split union of Hopf links [2601.12846].

## 5. Torus links and explicit examples

For torus links, the supplied results go beyond general bounds and determine exact values in several families. Assuming $2\le p\le q$, the torus link $T(p,q)$ is the closure of the braid $(\sigma_1\sigma_2\cdots \sigma_{|q|-1})^{|p|}$ on $\max\{|p|,|q|\}$ strands and has $\gcd(p,q)$ components. The exact formula
$$
\alpha_3\bigl(T(n,n)\bigr)=4n-2
$$
holds for every integer $n\ge 2$ [2507.13633].

More generally, the stated upper bound is
$$
\alpha_3\bigl(T(p,q)\bigr)\le 2p+2q-2.
$$
When $2\le p<2p\le q$, the improved bound becomes
$$
\alpha_3\bigl(T(p,q)\bigr)\le 2p+2q-3.
$$
These are obtained by rewriting the braid word into forms that cluster crossings, introducing kinked arcs, choosing a horizontal binding line, and separating the resulting diagram into three pages according to whether arcs lie above the line, below it, or pass over it [2507.13633].

The small examples recorded in the sources illustrate how the general theory specializes. For the trefoil knot, the theorem $\alpha_3(L)\le 3\,c(L)-1$ gives $\alpha_3\le 8$, and an explicit three-page presentation with $8$ arcs is described; in the torus-link treatment this is sharpened to the exact identity $\alpha_3(T(2,3))=8$ [2601.12846]. For the figure-eight knot, the bound gives $\alpha_3\le 11$, and an explicit drawing with $11$ arcs has been produced; conjecturally no better presentation exists [2601.12846].

These cases show that the general linear estimates are often close to exact, and in some structured families exact. A plausible implication is that torus-link braid structure is particularly well matched to three-page decompositions, because the constructions exploit repeated crossing patterns and controlled closure operations.

## 6. Distinct computational usage in document-to-slides generation

In an unrelated literature on automatic presentation generation, a three-page presentation refers literally to a three-slide artifact rather than a link embedded in a three-page book. The system “PPTAgent: Generating and Evaluating Presentations Beyond Text-to-Slides” uses a two-stage, edit-based approach. In Stage I, reference presentations are analyzed to extract slide-level functional types and content schemas; in Stage II, an outline is drafted and editing actions are iteratively generated based on selected reference slides to create new slides [2501.03936].

The supplied three-slide outline consists of “PPTAgent Framework,” “Two-Stage Edit-Based Workflow,” and “Results & Impact.” The formalization contrasts a conventional formulation,
$$
S=\sum_{i=1}^{n} e_i=f(C),
$$
with an agent formulation,
$$
A=\sum_{i=1}^{m} a_i=f\bigl(C\mid R_j\bigr).
$$
The editing actions include operations such as `clone_paragraph`, `replace_span`, and `del_image`, and the workflow uses a code-in-HTML REPL for self-correction. The associated evaluation framework, PPTEval, assesses presentations across Content, Design, and Coherence. The reported metrics in the supplied material are success rate $\ge 95\%$ across models and average Eval score $3.67/5$ with Content $3.28$, Design $3.27$, and Coherence $4.48$; the reported ablations include “w/o Outline” decreasing coherence by $0.36$ and “w/o CodeRender” decreasing SR by $20.4$ percentage points [2501.03936].

This usage is terminologically distinct from the link-theoretic notion. In the former, “three-page” denotes the length and structure of a slide deck synthesized from a document; in the latter, it denotes an embedding of a link into exactly three pages of an open book and the corresponding minimization problem for arc count. The coexistence of these usages is a matter of terminology rather than shared mathematical or algorithmic content.

Source: https://www.emergentmind.com/topics/three-page-presentation