5
$\begingroup$

So I would like to have a few simple examples where the presheaf associated to higher direct image of sheaf fails to be sheaf. So I'm looking for two (natural and simple) topological spaces $X$ and $Y$, a continuous map $f:X\rightarrow Y$ and a sheaf of abelian groups $\mathcal{F}$ on $X$ such that the presheaf $f_*^{i}(\mathcal{F})$ given by $$ U\mapsto H^i(f^{-1}(U),\mathcal{F}|_{f^{-1}(U)}) $$ (for some $i>0$) fails to be a sheaf on $Y$.

I'd like to see the two possible obstructions for $f_*^{i}(\mathcal{F})$ to be a sheaf namely

(i) the impossibility of gluing local sections of $f_*^{i}(\mathcal{F})$ to a global one,

(ii) the existence of a non-zero global section of $f_*^{i}(\mathcal{F})$ which is trivial at each stalk $y\in Y$.

$\endgroup$

1 Answer 1

12
$\begingroup$

For (i), take $X=S^3$, $Y= S^2$, $f$ the Hopf fibration. There is a local $H^1$ everywhere, coming from the fact that the fibers are circles. This local section is defined consistently everywhere (because there is no monodromy since $Y$ is simply-connected), but it does not glue to a global section because $H^1(X,\mathbb Z)=0$.

For (ii), simply take $X=Y$ a manifold space with nontrivial cohomology and $\mathcal F$ the constant sheaf. Then there is a global section corresponding to the higher cohomology classes. This section vanishes on every open ball, so it vanishes on every stalk.

$\endgroup$
12
  • $\begingroup$ As to the first case, why don't the local sections glue to a global one? What's wrong with $H^1(X,\mathbb{Z})$? $\endgroup$
    – user2013
    Dec 17, 2013 at 22:55
  • 2
    $\begingroup$ Because the Leray spectral sequence does not degenerate at the second page. The $H^1(X,\mathbb Z)$ coming from $H^0 R^1$ cancels with the $H^2(X,\mathbb Z)$ coming from $H^2 R^0$, leading to vanishing cohomology groups. Less abstractly, the nontrivial $H_1$ class of a fiber is represented by the full fiber, a circle. This circle is a trivial homology class because it bounds a disc in $S^3$. This disc wraps all the way around $S^2$, becoming the nontrivial homology class in $S^2$, but in $S^3$ one bounds the other so they cancel each other out. $\endgroup$
    – Will Sawin
    Dec 17, 2013 at 22:59
  • $\begingroup$ For (i), $f$ is a fibration with fibre $S^1$ and so $f^1_*(\underline{\mathbb{Z}})$ is a locally constant sheaf with fibre $\mathbb{Z}$. As $S^2$ is simply connected, we have $f_*^1$ is the constant sheaf, so I don't agree that (i) is an example. $f^3_*$ would however give an example for (ii). $\endgroup$ Dec 17, 2013 at 23:13
  • $\begingroup$ The pushforward presheaf is locally constant but not a sheaf. The sheafification, as you point out, is the constant sheaf. So it has a global section. But the presheaf does not have any sections on all of $Y$, because these are just defined to be $H^1(X, \mathbb Z)$. $\endgroup$
    – Will Sawin
    Dec 17, 2013 at 23:18
  • $\begingroup$ Sorry, I misread what you were saying. I agree with (i) now. $\endgroup$ Dec 17, 2013 at 23:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.