# Gerbes and Stacks

The definition of a gerbe on a smooth manifold that I know is that - after fixing an open cover $U_i$, a gerbe consists of the data of line bundles $L_{ij}$ on two-fold-intersections $U_{ij}$, isomorphisms $\alpha_{ijk}: L_{ij} \otimes L_{jk} \longrightarrow L_{ik}$ on three-fold intersections that satisfy a co-cycle condition on four-fold intersections.

A gerbe on a site is a stack $G$, such that for every object $U$, there exists a covering $U_i$ of $U$ such that $F_{U_i}$ is non-empty for every $i$ and for any two objects $x_1$, $x_2$ in $G_{U}$, there exists a covering $U_i$ of $U$ such that $x_1|_{U_i}$ and $x_2|_{U_i}$ are isomorphic (i.e. objects exist locally and they are locally isomorphic).

My question is that if these two notions are related or if it is just the same name for completely different things. In particular: Are gerbes on a manifold a special stack on the small site of that manifold? Is there a fully faithful functor of $2$-categories that sends gerbes over $M$ to stacks over (the small site of) $M$?

• Your definition of gerbe on a manifold is only for "Lien $k^*$" if $k$ is your field. In your example you can think that on each $U_i$ you have trivialized your gerbe to obtain the Picard groupoid of line bundles on $U_i$, then on intersections you have the equivalence of groupoids which consists on tensoring with $L_{ij}$. So gerbes are a particular kind of stack in that way. Giraud's book is a fantastic read, Hitching notes are also very good. I personally like $\S 5$ of Deligne's "Le Symbole modéré" cause it's concise and precisely what you need. Nov 12, 2014 at 10:19
• Please always try to use at least one toplevel tag (those with two-letter prefix, corresponding to arXiv categories).
– user9072
Nov 12, 2014 at 11:09
• mathoverflow.net/questions/307383/… this might be of some relevance.. Apr 26, 2020 at 5:48

There is a canonical equivalence of $2$-categories

$$St\left(Man/M\right) \simeq St\left(Man\right)/M$$ between stacks on the large site of $M$ and stacks on the site of manifolds equipped with a map to $M$ (regarding $M$ as a representable sheaf). Given a map $\pi:\mathscr{Y} \to M$ for $\mathscr{Y}$ some stack on manifolds, it corresponds to the stack $\Gamma(\mathscr{Y})$ on $Man/M$ which assigns a map $f:N \to M$ the groupoid of sections $N \to \mathscr{Y}$ of $\pi$ over $f.$ Suppose that there is a cover of $U_i$ of $M$ such that each $U_i \times _M \mathscr{Y}\simeq U_i \times BU(1)$ (or if you prefer $U_i \times BGL(1)$). Then $\Gamma(\mathscr{Y})$ is easily seen to be a gerbe on the large site for $M$. By Dan Peterson's answer, we see that from the data of a bundle gerbe, one gets a stack $\pi:\mathscr{Y} \to M$ with this property. In fact, it is not hard to show that these are equivalent data, that is, given $\pi:\mathscr{Y} \to M$ such that there is a cover $U_i$ such that $U_i \times _M \mathscr{Y}\simeq U_i \times BU(1)$ is the same as giving a bundle gerbe on $M$. By taking each bundle gerbe $\pi:\mathscr{Y} \to M$ and sending it to $\Gamma(\mathscr{Y})$, one gets a fully faithful embedding of the $2$-category of bundles gerbes over $M$ into the $2$-category of gerbes over the large site of $M$ (which furthermore embeds fully faithfully into stacks on the large site of $M$). The essential image is precisely those gerbes on the site $Man/M$ which are banded by $U(1)$, as pointed out by Reimundo Heluani. It doesn't embed into the $2$-category of stacks on the small site of $M$ however.

• I guess I should mention that, in reference to Konrad's answer, it depends on that morphisms you allow between bundle gerbes. If you treat them as principle bundles for $BU(1)$ then the maps need to be equivariant for the $BU(1)$-action, which would mean the functor into gerbes on the large site would only be $2$-categorically faithful, but not full. Nov 12, 2014 at 21:21

Your question has been answered by Reimundo Heluani but let me spell out things in full detail. Suppose you have a gerbe on a smooth manifold $M$ in your sense, given by $(U_i,L_{ij},\alpha_{ijk})$. Here is a procedure for cooking up a stack out of this data:

i) Let $P_i = U_i \times \mathrm{BGL}(1)$ for all $i$. A map $X \to P_i$ is the same as a map to $U_i$ and a line bundle on $X$.

ii) For each pair of indices $i,j$, a map $X \to U_i \times_M P_j$ is a map $f \colon X \to U_i \cap U_j$ and a line bundle $L$ on $X$. Now define a $1$-morphism $\phi_{ij} \colon U_i \times_M P_j \to U_j \times_M P_i$ by the rule $(f,L) \mapsto (f,L \otimes f^\ast L_{ij})$.

iii) For each triple of indices $i,j,k$ there are two natural isomorphisms $U_i \times_M U_j \times_M P_k \to U_k \times_M U_j \times_M P_i$ which we may abusively denote $\phi_{ik}$ and $\phi_{ij}\circ \phi_{jk}$, and the isomorphism $\alpha_{ijk}$ defines a $2$-morphism between these.

iv) For each quadruple overlap, the cocycle condition for the $\alpha$'s ensures that these $2$-morphisms commute strictly.

Thus we obtain gluing data that allows us to glue together all the $P_i \to U_i$ to a stack $P \to M$, which is locally a "$\mathrm{BGL}(1)$-bundle".

• Ok, so this becomes a stack (which is a gerbe as in the stack definition) on the big site of $M$. There is one object in each fiber for each index $i$ and one morphism from $i$ to $j$ if $U_{ij} \neq \emptyset$. Here I strictified the composition. Am I right? Nov 12, 2014 at 16:13
• I don't really understand you so at least one of us is confused. The fiber over a point of $U_i$ is a groupoid with one object and morphism set $BGL(1)$. Nov 12, 2014 at 19:57

Here is one way to see it: gerbes on $M$ form a bigroupoid (= bicategory all of whose morphisms and 2-morphisms are invertible).

In particular, if $\mathcal{G}$ is a gerbe over $M$, and $\mathcal{I}_M$ denotes the trivial gerbe over $M$ (trivial cover $\{M\}$, trivial line bundles, trivial isomorphism) then we have a groupoid $Hom(\mathcal{G},\mathcal{I}_M)$ of "trivializations".

Define $$U \mapsto Hom(\mathcal{G}|_U,\mathcal{I}_U).$$ This is your stack on the site of open sets of $M$.