22
$\begingroup$

Let $X$ be a projective variety, and $E$ be a coherent sheaf on $X$. Grothendieck has proven that there is a scheme $\mathrm{Quot}_X(E)$ parametrizing arbitrary quotient sheaves of $E$. It is probably a well-known question, but I found nothing in Google:

$$\text{Why not a scheme $\mathrm{Sub}_X(E)$ parametrizing arbitrary subsheaves of $E$?}$$

Does such a scheme exist? If yes, is it purely a question of style, like Grothendieck's definition of the projectivization $\mathbb P(V)$ as the set of one-dimensional quotients of a vector space $V$? If not, is it because cokernels of sheaves require sheafification, while kernels coincide with kernels for pre-sheaves?

Moreover, if one considers a reflexive sheaf $E$ (that is the canonical morphism $E \to E^{**}$ is an isomorphism) and its reflexive subsheaves, then there is a Sub-scheme via dualizing and taking a Quot-scheme. Is there a less restrictive constraint?

$\endgroup$
  • 1
    $\begingroup$ I think that this is a good question for mathoverflow. By the way, quot schemes parametrize locally free quotients. I guess that the answer is then that there are more locally free quotients than locally free subobjects. $\endgroup$ – HeinrichD Nov 7 '16 at 10:37
  • 1
    $\begingroup$ @HeinrichD, I believe that Quot-schemes parametrize all quotients: if $E=\mathcal O_X$ and one considers a component with Hilbert polynomial of degree zero, it is isomorphic to Hilbert scheme of points. In particular, all such sheaves have zero-dimensional support. $\endgroup$ – evgeny Nov 7 '16 at 11:58
  • 22
    $\begingroup$ For standard universal properties, you need the scheme to behave well under base change, which in these cases would mean tensor products. Tensor product is right exact, so a quotient remain a quotient, not left exact, so a sub may not remain a sub. $\endgroup$ – Mohan Nov 7 '16 at 13:57
  • 5
    $\begingroup$ I'd say it is like the definition of $P^n$, but it is not purely a `matter of style'. This essentially rephrases Mohan's comment, but here it is: On the level of points, there is of course no difference between subs and quotients: we can just think of short exact sequences $0\to A\to E\to B\to 0$. Now as you consider families (say, over $S$), you really want sequences $$0\to A\to E\boxtimes O_S\to B\to 0$$ with $A$ and $B$ flat over $S$ (so you have a family of short exact sequences). But flatness of $B$ implies flatness of $A$, so we can forget $A$ and think of $B$ only. $\endgroup$ – t3suji Nov 7 '16 at 15:58
  • 5
    $\begingroup$ A fundamental sense in which Grothendieck's definition of $\mathbf{P}(V)$ is not a "convention" is that it gives a meaningful notion of $\mathbf{P}(\mathscr{E}) := {\rm{Proj}}({\rm{Sym}}(\mathscr{E}))$ in the relative situation over a scheme $S$ when the quasi-coherent $O_S$-module $\mathscr{E}$ is not locally free of finite rank (so passing to its dual is not a real option, and so "only one" option is available, namely Grothendieck's). Also, if you read the proof of existence of Quot-schemes you'll see exactly how quotients are better-suited than subsheaves (related to Mohan's comment). $\endgroup$ – nfdc23 Nov 7 '16 at 16:22
15
$\begingroup$

For standard universal properties, you need the scheme to behave well under base change, which in these cases would mean tensor products. Tensor product is right exact, so a quotient remain a quotient, not left exact, so a sub may not remain a sub.

$\endgroup$

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.