EDIT: The original version regarding coker is false.

Consider a morphism of sheaves of abelian groups $\Phi:\mathscr{F}\to\mathscr{G}$.

It is not in general true that Im$_{pre}\Phi$ is a sheaf. However, if $\Phi$ is a monomorphism, then $\Phi_U$ is a monomorphism on open sets and Im$_{pre}\Phi$ seems to be a sheaf. Consider the explicit construction of Im$_{pre}\Phi$ where $$\text{Im}_{pre} \Phi(U) := \text{Im}\; \Phi_U \subset \mathscr{G}(U)$$ since we are dealing with sheaves of abelian groups, this seems to be a sheaf.

Is this true? What if we are dealing with a monomorphism $\Phi:\mathscr{F}\to\mathscr{G}$ of sheaves of some unknown abelian category, is it still true that Im$_{pre} \Phi$ is a sheaf? And if so what is a purely abstract argument?

  • $\begingroup$ One observation, and really I wouldn't dignify it by calling it anything more, is that the presheaf image is already a sheaf in this case. That's useful. $\endgroup$ – Hoot Jan 25 '15 at 20:45
  • 2
    $\begingroup$ To explain: the presheaf image here is isomorphic as a presheaf to $\mathscr{F}$, so it's a sheaf. $\endgroup$ – Hoot Jan 25 '15 at 22:54

Perhaps I've misunderstood what you wrote, but it seems to me that your statement about the presheaf cokernel being a sheaf is false.

E.g. take $X := \mathbb C^{\times}$ with its usual topology, let $\mathscr G := \mathscr O_X$ denote the sheaf of holomorphic functions on $X$, and let $\mathscr F$ denote the sheaf of locally constant functions on $X$ which take values in $2\pi i \mathbb Z.$

Then $\mathscr F \hookrightarrow \mathscr G$ (any locally constant function is holomorphic).

Let $\mathscr O_X^{\times}$ denote the sheaf of nowhere-zero holomorphic functions on $X$ (which is a sheaf of abelian groups via multiplication).

There is a natural isomorphism $\mathscr G/\mathscr F \buildrel \sim \over\longrightarrow \mathscr O_X^{\times}$, given via the morphism $f \mapsto e^f.$

This map is not surjective on global sections: there is no global section of $\mathscr G$ which maps to the global section $z$ of $\mathscr O_X^{\times}.$ (Concretely, there is no single valued branch of $\log z$ defined on $X$.)

Thus $\mathscr G(U) \to (\mathscr G/\mathscr F)(U)$ is not surjective, and in particular, the natural injection $$\mathscr G(U)/\mathscr F(U) \to (\mathscr G/\mathscr F)(U)$$ is not surjective.

(This failure of surjectivity is the source of sheaf cohomology. In the particular case I've described here, the related cohomological fact is that the $1$st cohomology of $X$ with integral coefficients is non-zero.)

| cite | improve this answer | |
  • $\begingroup$ This answer is indeed perfectly correct, contra mez's assertion. $\endgroup$ – Georges Elencwajg Jan 25 '15 at 20:07
  • $\begingroup$ You are right, I was always thinking about image and naively just reduced to cokernel. What about current version regarding image? $\endgroup$ – mez Jan 25 '15 at 22:49
  • $\begingroup$ @mez: If $\Phi$ is a monomorphism then it induces an isomorphism of $\mathscr F$ onto its image, and so yes, the presheaf and sheaf images coincide. $\endgroup$ – tracing Jan 26 '15 at 0:48

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.