Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Assume we have a locally free sheaf $R$ of associative $O_S$-algebras of rank $r^2$, possibly noncommutative, where $S=\mathbb{P}^2$ over $\mathbb{C}$. Furthermore given a locally free $O_S$-module $E$ also of rank $r^2$, which is a simple $R$-module, i.e. $End_R(E)=\mathbb{C}$.

We also know that all bundles come from a cyclic degree $r$ covering $f:X\rightarrow S$ ramified on a smooth curve of degree $nr$, i.e. $E$ is also a locally free $O_X$-module of rank $r$, and $R$ also an $O_X$-algbera of rank $r$ and $E$ is an $R$-module.

Now if $L$ is a cohomologically trivial $O_X$-line bundle, i.e. $H^i(X,L)=0$ for all $i$, what can we say about $Hom_R(E,L\otimes_{O_X} E)$? Does it vanish? Is it isomorphic to $\mathbb{C}$? Can it have bigger dimension?

If this is too broad, here is the situation im especially interested in: $X$ is a $K3$-surface with two disjoint (-2)-curves $C_1,C_2$, constructed as a double cover of $\mathbb{P}^2$ ramified on a sextic $C$. The line bundle $L$ is given by $O_X(C_1-C_2)$ and $R$ and $E$ are rank 2 locally free sheaves on $X$.

I was hoping that there are no such morphisms: my idea was to construct a map $\alpha: L\otimes_{O_X}E\rightarrow E$, such that if there were a map $\beta: E \rightarrow L\otimes_{O_X} E$, one could look at $\alpha\circ \beta=c id_E$ by simplicity. Then one had to show that a multiple of the identity can not factor in this way. But i couldn't make this argument work.

Another idea was to use the determinant bundle: if i had a map $E\rightarrow L\otimes_{O_X} E$, this would give me a map $det(E)\rightarrow L^2\otimes_{O_X} det(E)$, so a map $O_X\rightarrow L^2$, but i couldn't see if there can be such maps. If $L$ is cohomologically trivial, especially $H^0(L)=0$ does this imply $H^0(L^2)=0$?

Maybe there can be nontrivial elements in $Hom_R(E,L\otimes_{O_X} E)$? Or are there other ways to attack this problem? Any idea is welcome.

Edit: We may restrict to those bundles $E$ with first Chern class $c_1(E)=C_1-C_2+f^{\*}(nH)$ on $X$, with $H$ a line in $\mathbb{P}^2$, where $n=0$ and $n=1$ would be enough, so $c_1(E)=C_1-C_2$ and $c_1(E)=C_1-C_2+f^{\*}H=C_1-C_2+C_2+f(C_2)=C_1+f(C_2)$. We may also assume $E$ is a semistable $O_X$-module.

share|improve this question

1 Answer 1

up vote 3 down vote accepted

If you are specifically interested in the case where $L=\mathcal{O}_{X}(C_{1}-C_{2}),$ all that is needed for your determinant bundle argument to work is the vanishing $H^{0}(\mathcal{O}_{X}(2C_{1}-2C_{2}))=0.$ But this follows from taking cohomology in the exact sequences $$0 \rightarrow \mathcal{O}_{X}(-2C_{2}) \rightarrow \mathcal{O}_{X}(C_{1}-2C_{2}) \rightarrow \mathcal{O}_{C_{1}}(C_{1}-2C_{2}) \rightarrow 0$$ $$0 \rightarrow \mathcal{O}_{X}(C_{1}-2C_{2}) \rightarrow \mathcal{O}_{X}(2C_{1}-2C_{2}) \rightarrow \mathcal{O}_{C_{1}}(2C_{1}-2C_{2}) \rightarrow 0 $$

EDIT: Up to now we have only established that the elements of ${\rm Hom}(E,E \otimes L)$ have rank at most 1 on each fiber. If the restriction of $E$ to $C_{1}$ is $\mathcal{O}_{C_1}(a) \oplus \mathcal{O}_{C_1}(b)$ for $a,b \in \mathbb{Z}$ satisfying $|a-b| \leq 1,$ then the simplicity of $E$ and the sequence $$0 \rightarrow E \otimes E^{\ast}(-C_2) \rightarrow E \otimes E^{\ast} \otimes L \rightarrow E \otimes E^{\ast} \otimes \mathcal{O}_{C_1}(-2) \rightarrow 0 $$ give you what you want; I don't see right away how to proceed without extra assumptions on $E.$

share|improve this answer
1  
Ah, thanks, so one needs two sequences to show the vanishing of $H^0(L^2)$. I added some extra assumptions one can impose on $E$ at the end of the question, maybe that will help? –  TonyS Feb 6 '12 at 19:51

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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