Is this knot-forming process a categorical construction?

58 Views Asked by At

Every embedded planar graph gives rise to a link (collection of knots) according to a particular procedure (I've heard it called the Mercat procedure). I am trying to determine whether this construction is universal. So far, I have constructed a family of alternative procedures for turning planar graphs into links; I am trying to show that the Mercat procedure is optimal among these alternatives. One hypothesis is that the Mercat procedure produces a link with the fewest number of components, but I am having trouble seeing whether this is true or gaining intuition about it. Details follow.


  • The "dart space" of an embedded undirected graph $G=(V,E)$ is the set consisting of two points for every edge $e\in E$; intuitively, it consists of the set of edges where one or the other of its endpoints has been marked as special. Categorically, we can identify the dart space of $G$ with the set of graph morphisms $\hom(\mathbf{2}, G)$, where $\mathbf{2}$ is the graph with two vertices and one edge.

  • The Mercat procedure uses the fact that every graph can be characterized by two permutations on its darts. Let $\alpha:D\rightarrow D$ be the involution which, for each dart, flips which endpoint is marked as special. Because the graph is embedded, let $r:D\rightarrow D$ be the permutation which sends each dart $\langle e, v\rangle$ to its immediate clockwise neighbor $\langle e^\prime, v\rangle$, turning around $v$. These data comprise a combinatorial map, and completely characterize the embedded graph.

  • We can describe the Mercat procedure as a permutation on the space $D \times \{\mathsf{over},\mathsf{under}\}$, which equips each dart with a pair of crossing threads, one over and one under. Equivalently, we can describe it as a permutation on $D+D$.

    Let $\beta$ be the involution on $D+D$ which exchanges $(d,\mathsf{over})\leftrightarrow (d,\mathsf{under})$. Then the Mercat procedure is defined as $$\mathcal{M}:D+D\xrightarrow{r + r^{-1}}D+D \xrightarrow{\alpha + \alpha} D+D \xrightarrow{\beta} D+D$$ and the components of the associated link are the orbits of this procedure.

  • We can generalize this procedure, and hopefully show that the Mercat procedure is an optimal choice among this newly-created family of procedures, essentially by replacing the embedded rotation $r$ with any permutation that behaves in a pivot-like way, i.e. by mapping each dart to a dart with the same designated endpoint.

    Categorically, we can formalize this condition by fixing an arrow $\mathsf{pivot}:\mathbf{1}\rightarrow \mathbf{2}$ which identifies which of the endpoints of a dart is the "pivot" endpoint. Then a pivot-like endomorphism $\rho$ of $D$ is just one where every dart has the same special endpoint as its image under $\rho$.

$$\hom(\mathbf{2},D) \xrightarrow{\rho \circ -} \hom(\mathbf{2},D) \xrightarrow{-\circ \mathsf{pivot}} \hom(\mathbf{1}, D) \quad=\quad \hom(\mathbf{2},D) \xrightarrow{-\circ \mathsf{pivot}} \hom(\mathbf{1}, D) $$

  • Such an endomorphism $\rho$ no longer needs to be spatially realizable or coherent, so we can abandon the requirement that $G$ be concretely embedded in a surface. For each pivot-like endomorphism $\rho:D\rightarrow D$, we have a corresponding Mercat-like procedure

    $$D+D\xrightarrow{\rho + \rho^{-1}}D+D \xrightarrow{\alpha + \alpha} D+D \xrightarrow{\beta} D+D$$ and the components of the associated link are the orbits of this procedure.

  • The generality of $\rho$ allows for some bizarre and degenerate knotwork constructions. For example when $\rho = \mathsf{id}_D$, there is a pair of intersecting loops wrapped around every edge, forming $|E|$ separate components.

  • Within this context, I am wondering whether I can show that there is something optimal/special about the Mercat procedure, e.g. whether all choices of $\rho$ produce links with fewer components. Certainly the pivot $r$ seems like the least-degenerate sort of pivot permutation on darts, as it has no fixed points except for when obligatory—in cases like loops and vertices of degree 1. I wonder if I can show that all such minimally-degenerate pivot permutations produce links with the same number of components (e.g. using a deletion-contraction recurrence), even if some of these pivot permutations are not physically realizable as coherent rotations.