Convention. geometric morphisms and inverse image [fgap-000S]
AGENTDRAFTED
For a geometric morphism
\[
f:\mathcal {E}\longrightarrow \mathcal {F},
\]
we write
\[
f^*:\mathcal {F}\longrightarrow \mathcal {E},
\qquad
f_*:\mathcal {E}\longrightarrow \mathcal {F},
\qquad
f^*\dashv f_*.
\]
Thus the morphism and its inverse-image functor point in opposite directions.
The functor \(f^*\) preserves finite limits.
Two examples fix the convention:
\[
\begin {array}{c|c|c}
\text {input map}&\text {geometric morphism}&\text {inverse image}\\ \hline
i:\{x\}\hookrightarrow X&
\mathsf {Set}\to \mathsf {Sh}(X)&
i^*F=F_x=\displaystyle \varinjlim _{x\in U}F(U)\\[3pt]
\varphi :H\to G&
\mathsf {B}H\to \mathsf {B}G&
\operatorname {Res}^G_H:\mathsf {B}G\to \mathsf {B}H.
\end {array}
\]
In the first row, the stalk is the filtered colimit of all
neighborhood sections, not the value on one chosen neighborhood.
Mac Lane and Moerdijk construct inverse-image sheaves through pulled-back
étale spaces in [maclane1992sheaves, sec. II.9], especially printed
pp. 99--101.
In the second row, right actions are presheaves. Precomposition with
\(B\varphi ^{\mathrm {op}}\) is restriction of actions. It has both
Kan-extension adjoints, so it is the inverse-image part of the displayed
geometric morphism. The homomorphism and geometric morphism point from \(H\)
to \(G\); restriction points from \(G\)-objects to \(H\)-objects.