cospan cotrace


One can naturally think of a cospan as the abstraction of a cobordism. For instance an interval object cospan models the standard topological interval [0,1][0,1] regarded as a cobordism from pt to pt. The cospan cotrace on the interval glues the two ends of the interval together to produce a circle regarded as a cospan from \emptyset to itself.

The concrete dual of a cospan, obtained by mapping it into some target object, is a span, which in the context of groupoidification and geometric function theory can be interpreted as a generalized linear map. On such a generalized linear map, there is a notion of trace, the span trace.

The cospan cotrace is the concept dual to that: the image of the cotrace of a cospan under mapping it into a target object is the span trace of the result of mapping the original cospan to that target object.



T in out Σ Σ \array{ && T \\ & {}^{in}\nearrow && \nwarrow^{out} \\ \Sigma &&&& \Sigma }

a cospan with identical left and right index object Σ\Sigma, its cospan cotrace cotr(T)cotr(T) is the composite of the result

T inout ΣΣ \array{ && T \\ & {}^{in \sqcup out}\nearrow && \nwarrow \\ \Sigma \sqcup \Sigma &&&& \emptyset }

of dualizing one leg of the cospan with the cospan

Σ IdId ΣΣ \array{ && \Sigma \\ & {}^{}\nearrow && \nwarrow^{Id \sqcup Id} \\ \emptyset &&&& \Sigma \sqcup \Sigma }

i.e. the pushout

cotrT Σ T IdId inout ΣΣ \array{ &&&& \mathrm{cotr}T \\ &&& \nearrow && \nwarrow \\ && \Sigma &&&& T \\ & {}^{}\nearrow && \nwarrow^{Id \sqcup Id} && {}^{in \sqcup out}\nearrow && \nwarrow \\ \emptyset &&&& \Sigma \sqcup \Sigma &&&& \emptyset }

regarded as a cospan from the initial object \emptyset to \emptyset

cotr(T) . \array{ && cotr(T) \\ & {}^{}\nearrow && \nwarrow \\ \emptyset &&&& \emptyset } \,.

Definition for multi-cospans

More generally, the trace of a multi-cospan over nn identical of its index objects Σ\Sigma is the composite with the multi-cospan

Σ Id Id Σ Σ Σ \array{ & \Sigma \\ & {}^{Id}\nearrow \uparrow^{Id} & \cdots \\ \Sigma & \Sigma & \cdots & \Sigma & \cdots }


Cotracing topological interval to circle

Let the ambient category be Top, let I=[0,1]I = [0,1] be the standard topological interval and let e:=[0,ϵ]e := [0,\epsilon] be a small interval, for some 0<ϵ<1/20 \lt \epsilon \lt 1/2 – to be thought here as a collar of the point ptpt.


I 1ϵ+() e e \array{ && I \\ & {}\nearrow && \nwarrow^{1-\epsilon+(-)} \\ e &&&& e }

be the interval regarded as a collared cobordisms from the point to the point. Its cotrace, the pushout

cotr(I) I inout e IdId ee \array{ cotr(I) &\leftarrow& I \\ \uparrow && \uparrow^{in \sqcup out} \\ e &\stackrel{Id \sqcup Id}{\leftarrow}& e \sqcup e }

is the result of gluing the ends of the interval to each other, i.e. the circle

cotr(I)=S 1. cotr(I) = S^1 \,.

Urs: This may require a bit more care with the topology involved. I still need to check the reference below for more details.

See also

  • Marco Grandis, Collared cospans, cohomotopy and TQFT (Cospans in Algebraic Topology II) (pdf)

Cotracing category interval object to the natural numbers

Let the ambient category be Cat, let I={ab}I = \{a \to b\} be the standard interval object in Cat and let pt={}pt = \{\bullet\} be the terminal category.


I pta ptb pt pt \array{ && I \\ & {}^{pt \mapsto a}\nearrow && \nwarrow^{pt \mapsto b} \\ pt &&&& pt }

be the standard interval object in Cat regarded in the standard way as a cospan from the point to the point.

Dualizing it to

I inout ptpt \array{ && I \\ & {}^{in \sqcup out}\nearrow && \nwarrow^{} \\ pt \sqcup pt &&&& \emptyset }

corresponds to thinking of it as a “bent interval”

pt pt. \array{ pt \\ & \searrow \\ && \downarrow \\ & \swarrow \\ pt } \,.

Accordingly, the co-span

pt IdId ptpt \array{ && pt \\ & {}^{}\nearrow && \nwarrow^{Id \sqcup Id} \\ \emptyset &&&& pt \sqcup pt }

can be thought of as

pt pt. \array{ & pt \\ \nearrow \\ \nwarrow \\ & pt } \,.

Gluing these two arcs together yields the cotrace, the pushout

cotr(I) I inout pt IdId ptpt, \array{ cotr(I) &\leftarrow& I \\ \uparrow && \uparrow^{in \sqcup out} \\ pt &\stackrel{Id \sqcup Id}{\leftarrow}& pt \sqcup pt } \,,

which is the result of gluing the ends of the interval object to each other, which here is the skeleton of the fundamental category of the directed circle

, \array{ && \rightarrow \\ & \nearrow && \searrow \\ \uparrow &&&& \downarrow \\ & \nwarrow && \swarrow \\ && \leftarrow } \,,

namely the monoid of natural numbers, regarded as a one-object category:

cotr(I)=B={n|n}. cotr(I) = \mathbf{B} \mathbb{N} = \{\bullet \stackrel{n}{\to} \bullet | n \in \mathbb{N}\} \,.

If instead we start with the standard interval object in groupoids, I inv={ab}I_{inv} = \{a \stackrel{\simeq}{\to} b\} with the nontrivial morphism from aa to bb being an isomorphism, then the co-trace in question is the skeleton of the fundamental groupoid of the ordinary topological circle

cotr(I inv)=B={n|n}. cotr(I_{inv}) = \mathbf{B} \mathbb{Z} = \{\bullet \stackrel{n}{\to} \bullet | n \in \mathbb{Z}\} \,.



While the concept is obvious, it is apparently (?) not discussed yet in the (young) literature on the subject. On the blog the concept was mentioned in

Revised on August 5, 2017 01:58:41 by David Corfield (