By https://arxiv.org/abs/1406.4419 (The fundamental groupoid as a terminal costack, Ilia Pirashvili), we know that for a topological space $X$, the $2$-functor $$\text{Top}(X)\rightarrow \text{Gpd}, \quad (U\rightarrow X)\mapsto \Pi_1(U)$$ is a $2$-cosheaf, in fact the terminal one. In particular, it follows that $$\text{Top}(X)^{op}\rightarrow \text{Gpd}, \quad (U\rightarrow X)\mapsto \text{Hom}(\Pi_1(U),\mathcal{G})$$ is a $2$-sheaf, where $\mathcal{G}$ is a groupoid. However, if we assume $X$ is a manifold (probably locally simply connected does the job, but I want to be cautious), then for any point $x\in X$, open simply connected sets $U\ni x$ are final in all open neighborhoods of $x$. Hence the stalk of this $2$-sheaf is $$\text{colim}_{U\ni x}\text{Hom}(\Pi_1(U),\mathcal{G})\cong \text{colim}_{U\ni x,\text{ simp. conn}}\text{Hom}(\Pi_1(U),\mathcal{G})\cong \text{colim}_{U\ni x,\text{ simp. conn}}\text{Hom}(\{x\},\mathcal{G})$$ where I denote by $\{x\}$ the groupoid whose objects are $\{x\}$ and has no automorphism (I think this is called the trivial groupoid). If we assume that $\mathcal{G}$ is a group $G$, then we get that the stalk consits only of the trival morphism.

However, having trivial stalks means that $\text{Hom}(\Pi_1(-),G)$ is trivial $2$-sheaf. But surely this can not be true, as that would mean $$\text{Hom}(\Pi_1(X),\text{Gl}_n(\mathbb{C})$$ is trivial, which is not true. Hence there is a gap/flaw in my reasoning, but I don't see it.

Basically, as $\Pi_1(-)$ is the terminal $2$-cosheaf, $\text{Hom}(\Pi_1(-),\mathcal{G})$ is the initial $2$-sheaf, but that would mean that there are almost no representations of the fundamental group, which seems wrong to me.

  • $\begingroup$ Do you mean $\hom$ with values in $Gpd$ or with values in $Set$ ? (note, by the way, that you never use, in your "proof", that $\Pi_1(-)$ is the terminal $2$-cosheaf) $\endgroup$ Sep 23 '20 at 19:40
  • $\begingroup$ I mean $\text{hom}$ in $Gpd$ to get the sheaf property. $\endgroup$ Sep 23 '20 at 19:45
  • $\begingroup$ But then $\hom(\{x\},\mathcal G)\cong \mathcal G$. So you probably want $\hom$ to take values in $Set$, so the thing you're after is the fact that $\Pi_1(-)$ is a $1$-cosheaf (which is also claimed in the paper) $\endgroup$ Sep 23 '20 at 19:52
  • $\begingroup$ Shouldn't $\{x\}$ be the initial groupoid? So in partiuclar $\text{hom}(\{x\},\mathcal{G})$ would be a singleton as a set. $\endgroup$ Sep 23 '20 at 20:01
  • $\begingroup$ So you're looking at $\hom$ valued in $Set$, which is what I first asked (it's not the initial groupoid, it's the terminal one -in particular $\hom(\{x\}, \mathcal G) \cong Ob(\mathcal G)$ in general, although that doesn't change your question when $\mathcal G= BG$ has only one object) $\endgroup$ Sep 23 '20 at 20:03

First of all, note that you haven't used the fact that $\Pi_1(-)$ was the terminal $2$-cosheaf, just that it was a ($2$-)cosheaf.

Then, as I pointed out in the comments, there's a question of whether you're considering the usual $\hom$ functor into $Set$, or the internal $\hom$ with values in $Gpd$.

  • If you're considering the usual, external $\hom$, then the question is more about whether $\Pi_1(-)$ is a cosheaf than a $2$-cosheaf. But note that if you're considering $\hom$ into $Set$, then $\hom$ is not equivalence invariant. In particular, $\hom(\Pi_1(U),\mathcal G) \not\cong \hom(\{x\},\mathcal G)$ even when $U$ is simply-connected. For instance if you take for $\mathcal G$ an indiscrete groupoid on a set $S$ of objects, the former is $\hom_{Set}(U,S)$, while the latter is $\hom_{Set}(\{x\},S)$. So your stalk computation doesn't work anymore.

(to perhaps get a better grasp on this, compare $\hom(2^{in},BG)$ and $\hom(\{x\},BG)$ where $2^{in}$ is the indiscrete groupoid on $2$ objects, and $BG$ the one-object groupoid associated to the group $G$)

  • If you're considering the internal $\hom$, and looking at stacks/$2$-sheaves (so that you do have invariance under equivalence), then $\hom(\{x\},\mathcal G)\cong \mathcal G$, so it's definitely nontrivial.

(moreover, note that in this context of stacks, or $2$-sheaves, you need to be more careful with colimits: this would probably be a $2$-colimit, so you'd need to check that the subcategory of simply-connected neighbourhoods of $x$ is final in the $2$-categorical sense in the category of neighbourhoods. This is a rather stronger condition than just finality in the $1$-categorical sense, and here I'm not sure whether it would be satisfied in general - I'm definitely not an expert on stacks and $2$-categories so I can't say much more)

  • 2
    $\begingroup$ Directed sets have the wonderful property that cofinal = homotopy cofinal, so it isn't a problem. $\endgroup$
    – Zhen Lin
    Sep 23 '20 at 22:17
  • $\begingroup$ I'm still confused by your statement that $hom_{gpd}(\{x\},\mathcal{G})\cong \mathcal{G}$. To define a functor $\{x\}\rightarrow \mathcal{G}$ corresponds to a choice of $\mathcal{G}$, so therefore shouldn't $hom_{gpd}(\{x\},\mathcal{G})\cong \text{Ob}(\mathcal{G})$? $\endgroup$ Sep 23 '20 at 23:59
  • $\begingroup$ You should read more carefully : there are two flavours of $\hom$, one which is set-valued (for this one, you are correct, and I explain in the first bullet point why the argument doesn't work for this one), and one which is groupoid-valued : this is a different functor $\endgroup$ Sep 24 '20 at 7:09
  • $\begingroup$ @ZhenLin : the poset of simply-connected neighbourhoods if $x$ isn't directed in general, is it ? Ah maybe up to a cofinal subset, in manifolds ? $\endgroup$ Sep 24 '20 at 7:36
  • $\begingroup$ The poset of open neighbourhoods, ordered by reverse inclusion, is directed. The simply connected open neighbourhoods are cofinal in this poset, for a manifold or more generally a locally simply connected space. $\endgroup$
    – Zhen Lin
    Sep 24 '20 at 7:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.