10
$\begingroup$

Given a graded ring $S$ and a quasi-coherent sheaf $\mathcal{F}$ on $Proj\ S$, does there exist a graded $S$-module $M$ such that $\mathcal{F}\cong \widetilde{M} $?

I know the result is true when $S$ is finitely generated by $S_1$ as an $S_0$-algebra (and so I am guessing the result to be false if we remove this hypothesis), but two cases I would like to know is when $S$ is generated by $S_1$, but not finitely and the second is when $S$ is not generated by $S_1$ at all.

$\endgroup$
7
+50
$\begingroup$

This may not be a full definitive answer to your question, but I hope it can shed some light on how these questions you ask can be a little more complicated than they might initially seem. If someone else has a good resolution to the non-quasicompact case discussed at the end, I would welcome reading about it.

The first thing to note is that the construction of $M$, if it exists, should be $M:= \bigoplus_{n\geq 0} \Gamma(\operatorname{Proj} S, \mathcal{F}(n))$. There's a canonical map $$\widetilde{M}\to \mathcal{F} \qquad (*)$$ regardless of whatever's going on with $\operatorname{Proj} S$. (See Stacks for the proof.) We're interested in when this is an isomorphism - as you state in your post, the condition that $S$ is finitely generated as an $S_0$ algebra by $S_1$ is sufficient. We'll see that this is far from a necessary condition, though, by demonstrating cases where $S$ is not necessarily finitely generated nor necessarily generated in degree one where this map is an isomorphism.


One funny thing about the Proj construction is that it does a really bad job at distinguishing between graded rings. That is, if $\operatorname{Proj} S\cong \operatorname{Proj} S'$, the graded rings $S$ and $S'$ can be very different. Here are two constructions that change $S$ significantly, but do not alter $\operatorname{Proj} S$:

  • Replacing $S_0$ with another ring $A_0$ with a homomorphism $f:A_0\to S_0$ and defining the action of $A_0$ on an element $s\in S$ by $a\cdot s = f(a)S$.

  • Replacing $S$ with $S^{(d)}:=\bigoplus_{n\geq 0} S_{dn}$ for $d>0$, a thinned out version of $S$.

Consider applying the first construction to $\Bbb C[x,y]$, replacing the copy of $\Bbb C$ in degree zero by $\Bbb Z$ and calling this new ring $S'$. Then $\operatorname{Proj} \Bbb C[x,y] \cong \operatorname{Proj} S' \cong \Bbb P^1_{\Bbb C}$, but the first is Proj of an algebra finitely generated in degree one and the second is not ($S'$ is uncountable, but any finitely-generated $\Bbb Z$-algebra is countable). On the other hand, the procedure which determines $M$ from the data of a sheaf $\mathcal{F}$ on this scheme doesn't see the difference between $\Bbb C[x,y]$ and $S'$. This means that asking about whether $S$ is finitely generated by $S_1$ as a $S_0$-algebra is not always the right question.

We may also apply the second construction to reduce any situation in which there's a $n>0$ so that all the generators of $S$ as an $S_0$-algebra are of degree $\leq n$ to a situation where $S$ is generated in degree 1 without changing Proj: replacing $S$ with $S^{(m)}$ for an appropriately-chosen $m$ (this should be the LCM of all degrees in which there's a generator), we may replace $S$ with a ring which is in fact generated in degree one without changing the projective space. For an example involving this construction, consider ring $S=\Bbb C[x,y,z]$ with $x$ in degree 1, $y$ in degree 2, and $z$ in degree 3. Let $S^{(6)}=k[x^6,x^4y,x^3z,x^2y^2,xyz,y^6,z^6]$ where each of the generating monomials is in degree one. Then $\operatorname{Proj} S=\operatorname{Proj} S'$, but one is generated in degree one and one is not. Since the construction of $M$ from $\mathcal{F}$ doesn't see this change, this means that asking about whether $S$ is generated in degree one is not always the right question either.


As mentioned in the comments by Ben, Stacks 0AG5 shows that this map $(*)$ from $\widetilde{M}\to \mathcal{F}$ is an isomorphism when $\operatorname{Proj} S$ is quasicompact. In particular, asking about quasicompactness of $\operatorname{Proj} S$ is a good way to eliminate the "tricks" from the previous section - rephrasing our conditions to depend on the scheme instead of the ring means we won't get fooled by the examples that violated the "$S$ is finitely generated as an $S_0$-algebra by $S_1$" condition if there's another choice of $S'$ which gives the same scheme after taking Proj and satisfies the quoted condition.

Restricting to quasicompact projective schemes should not be such a big deal if you're just starting out learning these things (I see this is your first question or answer in the algebraic-geometry tag, welcome!). Strange things can happen with non-quasicompact schemes - there are examples of such with no closed points, for instance. In general, once one leaves the confines of nice finiteness assumptions (noetherian/locally noetherian/quasicompact/quasiseparated are common such assumptions), a lot of machinery can no longer be guaranteed to work. Until you decide to work on a problem which needs the removal of these assumptions, it can be a lot easier to let sleeping dogs lie.


In an attempt to point out a meaningful example when $(*)$ is not an isomorphism, I would guess that a ring $S$ which does not satisfy "$S$ is generated by elements of degree $\leq n$ for some fixed integer $n$" cannot have $\widetilde{M}\to \mathcal{F}$ be an isomorphism. A key ingredient of the proof that $(*)$ is an isomorphism in the quasi-compact case is picking a $d$ so that $\mathcal{O}(d)$ is invertible, so that tensoring with $\mathcal{O}(d)$ is an isomorphism. By setting it up so that no such sheaf is invertible, we should be guaranteeing that we destroy information about $\mathcal{F}$ every time we tensor by $\mathcal{O}(d)$ which will cause $(*)$ to fail to be an isomorphism.

The reason having generators of unbounded degree should cause this to fail is given by Stacks 01MU which states that for $X=\operatorname{Proj} S$, the largest open set $W_d\subset X$ so that $\mathcal{O}_X(dn)|_{W_d}$ is invertible is exactly the union of all open sets $D_+(fg)$ for $f,g$ homogeneous with $\deg(f)=\deg(g)+d$. If we can't cover $X$ by such subsets for any $d>0$, this shows that no $\mathcal{O}_X(d)$ is invertible.

The case of "generated by $S_1$ but not finitely" and $\operatorname{Proj} S$ not quasicompact seems trickier. For instance, $\Bbb P^\infty = \operatorname{Proj} k[x_0,x_1,\cdots]$ with each $x_i$ in degree one doesn't seem so bad - there's a number of sub-results used in the proof that $(*)$ is an isomorphism which depend on quasicompactness which I wouldn't be surprised either way if they just went through because of the relatively nice situation we had, or instead broke in interesting ways. If anyone has treated this specific case, I would be interested in hearing about it - please leave a link in the comments.

$\endgroup$
  • $\begingroup$ I was thinking the same thing. Perhaps it would be easier to come up with an extraordinary module on $\mathrm{Proj}(k[x_1,x_2,...])$ with $x_i$ of degree $i$, but I haven't had any time to think about it any further. $\endgroup$ – Ben Aug 29 '19 at 19:32
  • $\begingroup$ Thank you. I believe this largely clears up my doubts. $\endgroup$ – Abhijit A J Aug 30 '19 at 5:15
  • $\begingroup$ My only problem (which I don't really believe affects your answer at all), is that you have written in the second paragraph that $M:= \Gamma(Proj\ S, \mathcal{F}(n))$. Did you mean $M:= \bigoplus\Gamma(Proj\ S, \mathcal{F}(n))$? Even so, we could have some other $M'$ which is not isomorphic to the mentioned module and still generate the same $\mathcal{O}_X$-module as shown by you later in the answer and also shown in (Hartshorne, Chapter 2, Excercise 5.9). It could be that I have misunderstood what you wrote, but it would be great if you could clarify. $\endgroup$ – Abhijit A J Aug 30 '19 at 5:26
  • $\begingroup$ Yes, that first one was a typo and it's been corrected. We definitely could have other $M'$ - any graded module $N$ which is nonzero in only finitely many degrees will give $\widetilde{N}=0$, so we can take a direct sum with one of those and that will change $M$ but not $\widetilde{M}$. I didn't really get in to this since the question was more focused on properties of $S$ (and my answer was already pretty long!), but I can definitely add this if you would like. $\endgroup$ – KReiser Aug 30 '19 at 5:30
  • $\begingroup$ Thanks a lot for the prompt reply $\endgroup$ – Abhijit A J Aug 30 '19 at 5:31

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.