The Quill 36: Towards Picard Group

Given any monoidal category $(\mathcal{C}, \otimes, 1)$, we write the Picard group as the group of invertible objects with respect to $\otimes$, where the multiplication is $[X] \cdot [Y] = [X \otimes Y]$. If we have a symmetric monoidal category, then Pic($\mathcal{C}$) is an abelian group. In geometry, the most interesting invertible objects are locally free sheaves of rank $1$, otherwise called line bundles. They are locally trivial; however, globally we can use them to construct and do topology.


Let $X$ be a scheme. A locally free sheaf of rank $n$ on $X$ is such that for any open cover $(\mathcal{U}_i)$ of $X$, we have  $\mathcal{L}|_{U_i}\cong\mathcal O_{U_i}^{\oplus n}$. A line bundle is a locally free sheaf of rank $1$. We say $\mathcal{L}$ is inverible if $\mathcal{L} \otimes \mathcal{L}^\vee (=\mathcal{L}^{-1}) \cong \mathcal{O}_X$ where $\otimes$ is tensor product on $\mathcal{O}_X$-modules. This defines for us the Picard group $Pic(X)$ as the group of invertible sheaves on $X$ up to isomorphism with tensor product multiplication. Indeed, the category of invertible sheaves, with $O_X$ being the unit and $\otimes$ the multiplication, is a symmetric monoidal category.

Now, let $\mathcal{U}_i$ be a trivializing open cover, we know that a line bundle is locally $\mathcal{L}|_{U_i}\cong\mathcal O_{U_i}$. But a trivializing open cover need not be globally agreed upon (you may think of the failure to define a connection [vector potential] globally on a $U(1)$-bundle). Moreover, we have transition functions $g_{ij}\in\mathcal O_X^\times(U_i\cap U_j): \mathcal{O}_{U_i}|_{U_i \cap U_j} \to \mathcal{O}_{U_j}|_{U_i \cap U_j}$, and a cocycle condition $g_{ij} g_{jk} g_{ki} =1$ which is a Cech 1-cocycle condition. Every local frame changes the cocycle by a coboundary, and thus we have the Cech cohomology $H^{1}(X,\mathcal{O}_X^\times)$. (An alternative definition of this Cech cohomology is to take the direct limit of cocycles over suitable covers.) This sheaf cohomology classifies line bundles up to isomorphism, and hence provides an equivalent definition of the Picard group. So in thie sense, the Picard group measures how the local information fails to glue globally.  

A different aspect of the Picard group is its equivalence to the divisors [let us keep the definitions for another time]. For a scheme $X$, regular, integral, Noetherian, Weil divisors (formal $\mathbb{Z}$-formal combination of codimension 1 closed subschemes) and Cartier divisors are equivalent descriptions. For $X$, we have a sheaf of rational functions $K_X$, then Cartier divisor is a collection $(U_i, f_i)$ on $K_X$ such that $\frac{f_i}{f_j} \in \mathcal{O}_X^\times(U_i \cap U_j)$ i.e. $f_i = f_j$ up to a multipication of section in $\mathcal{O}_X^\times$. These divisors are useful to understand the poles and zereos of a function. Now, we will say it is a fact that 'effective' Cartier divisors are just invertible ideal sheaves. Concretely, for $X$ integral, we have $Pic(X) = \text{\{Cartier Divisors on X\}}/ \text{\{Principal Divisors on X\}}$.  If you can trust this fact, then line bundles can be understood through zeroes and poles of their rational functions. 

Moreover, if one pushes to class field theory, then for a number field $K$ and its ring of integers $\mathcal{O}_K$, we have $\operatorname{Pic}(\operatorname{Spec}\mathcal O_K)\simeq\operatorname{Cl}(\mathcal O_K)$. We will come back to this and extend it with a proof, and push ourselves further into function fields and rank 1 geometric Langlands later.

To close this off, one can think of Pic$(\mathcal{C}$ as a decategorification of the Picard $2-$group (groupoid). In the Picard groupoid, we also remember isomorphisms (and then we also have the definitions of the Picard variety, Picard Schemes, and of course, the Picard stack!). 

[Acknowledgment: ChatGPT Free was used to make a good extraction recipe out of new grounds of coffee, which fueled this blog.]

This entry was posted in . Bookmark the post. Print it.

Leave a Reply