This question comes after reading the last paragraph of Vakil's FOAG, p. 400.
We consider the set of $\{(\mathcal L, s) \}$, where $\mathcal L$ is an invertible sheaf on a Noetherian, reduced, regular in codimension 1 (in case any of that matters) scheme $X$, and $s$ is a nonzero rational section of $\mathcal L$.
The claim is that once you mod out by isomorphism, this set is an abelian group under $\otimes$, with inverse $\{(\mathcal L^*, 1/s) \}$. Thematically, this all works out well given that we know $\mathcal L \otimes \mathcal L^* \simeq \mathcal O_X$, but why is $1/s$ a nonzero rational section of the dual?
The best I can say is that both sheaves are locally isomorphic to $\mathcal O_X$, so perhaps we mean to consider $s$ and $1/s$ as rational sections of $\mathcal O$, and then we glue to produce $s$ and $1/s$ as rational sections of $\mathcal L$?