# Vanishing of Higher Direct Images

Let $X$ and $Y \$ be smooth varieties over a field or - depending on the answers - more general nice schemes (I don't know what one needs exactly as conditions). Let $p: X\times Y \rightarrow X$ be the projection morphism. It has constant fiber dimension $n:=dim(Y)$ over$X$, so is smooth of relative dimension n.

Is it then right that one has a vanishing of the higher direct images for any quasicoherent sheaf $\mathcal F$ on $X\times Y$:

$\mathcal R^i \mathcal F =0$ for all $i > n$ ?

Note that by Grothendieck's Vanishing Theorem one surely has it for $i>n+dim(X)$.

Also I really want Quasicoherence, not Coherence, so Semicontinuity Arguments don't seem to be at hand.

Furthermore, how far can one generalize this?

-
It is true, and in a very general context. Note that the fibers are supported on a $n$-dimensional subset, hence the higher direct images vanish (recall that these are just the direct images of the direct image functor, which calculates $\Gamma(p^{-1}U,F)$.) – Bonanza Dec 5 '11 at 16:48
@hilbert: But $p^{-1}(U)=U\times Y$. And this has dimension greater than n. So I don't see what exactly you mean. – Cyril Dec 5 '11 at 19:08
Do you accept $Y$ to be projective ? – user18119 Dec 6 '11 at 0:21
I would accept it proper. Does this help? – Cyril Dec 6 '11 at 7:58
There is a fairly elementary proof for quasi-projective morphisms. For proper morphisms over a noetherian scheme, you could reduce to coherent sheaves considering a quasi-coherent sheaf as inductive limit of coherent sheaves. – user18119 Dec 6 '11 at 9:28

I am not sure what you call semicontinuity theorems. Your question is about vanishing of $R^i$'s.
Let $f : X\to S$ be a proper morphism to a locally noetherian scheme $S$ such that all its fibers have dimension $\le n$. Let $F$ be a quasi-coherent sheaf on $X$. Then $R^if_*F=0$ for all $i> n$.
Proof (admitting the result for coherent sheaves on $X$). The question is local on $S$, so we can suppose $S$ is affine. Then we have to show that $H^i(X, F)=0$ for $i>n$. Fix a finite affine covering $\mathcal U$ of $X$. We will compute $H^i(X,F)$ by Cech cohomology $H^i(\mathcal U, F)$. Let $$c\in \mathrm{ker}(C^{i}(\mathcal U, F)\to C^{i+1}(\mathcal U, F)).$$ By EGA I.9.4.9, $F$ is the inductive limit of its coherent submodules $(F_{\alpha})_{\alpha}$. As $c$ involves finitely many sections, there exists $\alpha$ such that $$c\in\mathrm{ker}(C^{i}(\mathcal U, F_\alpha)\to C^{i+1}(\mathcal U, F_\alpha))=\mathrm{Im}(C^{i-1}(\mathcal U, F_\alpha)\to C^{i}(\mathcal U, F_\alpha)).$$ Hence $$c \in \mathrm{Im}(C^{i-1}(\mathcal U, F)\to C^{i}(\mathcal U, F))$$ and $H^i(X,F)=H^i(\mathcal U, F)=0$.