3
$\begingroup$

Suppose $X$ is a smooth variety over $\mathbb{C}$, why do we have $K_X^*/O_X^*$ is a flasque sheaf? (Beauville "Complex Algebraic Surface" p.28)

(To show the surjection $K_X^*/O_X^*(X)\to K_X^*/O_X^*(U)$, given an element of the latter group, a collection of $(U_I,f_i)$ with $f_i/f_j\in O^*_{U_i\cap U_j}$. How to find the corresponding data for a global section? )

$\endgroup$
  • $\begingroup$ @GeorgesElencwajg Sorry for the confusion, but in the context he seemed to claim the sheaf $K_X^*/O_X^*$ be flasque for any smooth variety (p.28 Line -9), is there an example that $K_X^*/O_X^*$ not flasque? $\endgroup$ – user206517 Jan 8 '15 at 23:26
  • $\begingroup$ You are right: my apologies. I have upvoted you and deleted my first comment . If you delete yours, I will delete this one too. And, no, there is no example with $K_X^*/O_X^*$ not flasque: Beauville is perfectly right. $\endgroup$ – Georges Elencwajg Jan 9 '15 at 0:29
5
$\begingroup$

a) We have-completely tautologically- an exact sequence of sheaves in the Zariski topology $$ 1\to O_X^* \to K_X^* \to K_X^*/O_X^* \to 0 \quad (\bigstar) $$
It is clear that $K_X$ is flasque by definition of "rational function" (= regular function defined on some open subset of $X$ ), and thus that $K_X^*$ is flasque too.
But why is $K_X^*/O_X^*$ flasque?
The secret is to reinterpret that sheaf as the sheaf of Cartier divisors on $X$: $$(K_X^*/O_X^*)(U)=CaDiv(U)$$
And then to use that on a smooth variety Cartier divisors coincide with Weil divisors: $$CaDiv(U)=WDiv(U)$$
So finally we are reduced to show that $WDiv$ is a flasque sheaf on $X$.
But nothing is easier: given a prime divisor $Y\subset U$ we obtain a prime divisor $\bar Y\subset X$ on $X$ just by taking the Zariski closure $\bar Y$ of $Y$ in $X $.

b) Since flasque sheaves are acyclic, the exact sequence $(\bigstar)$ (extended by zeros) provides us with an acyclic resolution of $O_X^*$, which permits us to compute the cohomology of that sheaf.
Since all sheaves of index $\geq 2$ are zero in that resolution we conclude: $$ H^i(X, O_X^*)=0 \quad \operatorname {for all} i\geq 2 . $$

$\endgroup$
  • 1
    $\begingroup$ This is all nicely exposited on in the Tohoku paper, if anyone is curious. $\endgroup$ – Alex Youcis Jan 9 '15 at 22:46
  • $\begingroup$ Ah, interesting @Alex: I am curious. $\endgroup$ – Georges Elencwajg Jan 9 '15 at 23:22
  • $\begingroup$ I just checked Grothendieck's article: the argument Beauville uses and which I extended above is indeed in the Tohoku paper, and was probably to be found there for the first time. It is quite amusing to see Grothendieck express himself in a language that announces schemes but doesn't mention them yet : that would come about two years later ( think). $\endgroup$ – Georges Elencwajg Jan 9 '15 at 23:35
  • 1
    $\begingroup$ Yeah, I am moderately sure that is the first appearance. Indeed, in a letter to Serre (found in the famous Correspondence) Grothendieck discusses the subject with much excitement (dare I say pride). This is for the above theorems applciation to show that the canonical map $H^1(X,\text{GL}_n)\to H^1(X,\text{PGL}_{n-1})$ is surjective (since it's cokernel is $H^2(X,\mathcal{O}_X^\times)$!), and thus that every projective bundle comes from a vector bundle (at least on a factorial scheme). Beware though, if you go look for this letter: Grothendieck has it couched in non-modern language $\endgroup$ – Alex Youcis Jan 10 '15 at 0:50
  • 1
    $\begingroup$ he speaks of showing $H^2(X,k^\times)=0$ and discusses the application to showing $H^1(X,\text{GL}_n(k))\to H^1(X,\text{PGL}_{n-1}(k))$ is surjective. What he denotes by $k^\times$ is short hand for the sheaf of regular maps to $k^\times$, which is just $\mathcal{O}_X^\times$ (the same goes for the other two sheaves)! I think this notation is also explained in the Tohoku paper. Anywhoo, nice answer Georges! +1 $\endgroup$ – Alex Youcis Jan 10 '15 at 0:51

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.