1
$\begingroup$

Let $f:X\to Y$ be a flat, surjective, smooth morphism between smooth algebraic varieties (over $\mathbb C$). We assume that $f$ has relative dimension $n$ and we assume also that $\dim Y\ge 2$ (just to avoid the case of a curve that might be easier).

Let $L$ be a line bundle on $Y$, then we have a homomorphism in sheaf cohomology:

$$H^p(Y,L)\to H^p(X, f^\ast L) \quad\text{for } p=0,1,\ldots,\dim Y $$

Can we say anything about the injectivity of this map? Do we need some additional condition on $f$?

$\endgroup$

1 Answer 1

1
$\begingroup$

The unit of adjunction map $L\rightarrow f_\ast f^\ast L$ is an isomorphism if the fibres of $f$ are connected (I assume this in what follows). Then the Leray spectral sequence $E^{pq}_2=H^p(R^q f_\ast f^\ast L)\Rightarrow E^{p+q}=H^{p+q}(f^\ast L)$ has edge maps $e^p:E^{p0}\rightarrow E^p$ which are exactly the maps you are interested in. In particular, these maps are isomorphisms if $f$ is finite. In general $e^0$ is an isomorphism, $e^1$ is injective, but $e^2$ need not be injective (exact sequence of low degree terms). If $f$ is of the form $X\times Y\rightarrow Y$ then the maps $e^p$ are certainly injective (Künneth).

$\endgroup$
5
  • 1
    $\begingroup$ More generally, it happen quite often that $\mathcal{O}_Y\to Rf_* \mathcal{O}_X$ splits in the derived category, and then so does $L\to Rf_* f^* L$, so the pull-back maps $H^p(Y, L)\to H^p(X, f^* L)$ are injective. $\endgroup$ Apr 13, 2020 at 9:06
  • $\begingroup$ @Piotr Achinger, what do you mean by "quite often"? $\endgroup$
    – manifold
    Apr 13, 2020 at 22:10
  • $\begingroup$ Hello @Samuel, would you please provide a reference for the first statement? $\endgroup$
    – svelaz
    Oct 26, 2022 at 15:52
  • $\begingroup$ @svelaz 28.1.I. EXERCISE in Vakil's notes. $\endgroup$
    – ssx
    Oct 26, 2022 at 23:10
  • $\begingroup$ @Samuel I see. Thank you very much. I was hoping you were referring to some version of it without the proper assumption. $\endgroup$
    – svelaz
    Oct 27, 2022 at 13:02

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.