Disclaimer: This was first asked here on math.stackexchange with no answers.

Let $F,G$ be quasicoherent sheaves of modules on a scheme $X$, then is the sheaf $\mathcal{Ext}^i(F,G)$ equal to the sheafification of the presheaf $$E^i(F,G) : U\mapsto Ext^i_U(F|_U,G|_U)?$$

It seems to me that (maybe modulo some technical points?), this is probably true.

Indeed, let $G\rightarrow I^\bullet$ be a injective resolution for $G$. By Lemma III.6.1 in Hartshorne's Algebraic Geometry, we have that for any open $U\subset X$, $I^\bullet|_U$ is an injective resolution for $G|_U$. (Actually this result seems somewhat suspicious to me, since I have heard multiple times that "localization does not always preserve injectives").

But, assuming Hartshorne's lemma is correct, the cohomology of the complex $Hom_U(F|_U,I^\bullet|_U)$ gives precisely the groups $Ext^i_U(F|_U,G|_U)$.

On the other hand for any inclusion of opens $U\subset V$, localizing yields a morphism of complexes $$Hom_V(F|_V,I^\bullet|_V)\rightarrow Hom_U(F|_U,I^\bullet|_U)$$ and hence a map on cohomology, which is clearly functorial w.r.t. further localization. Thus, for any inclusion of opens $U\subset V$, this gives restriction maps

$$Ext^i_V(F|_V,G|_V)\rightarrow Ext^i_U(F|_U,G|_U)$$ which shows that the rule $U\mapsto Ext^i_U(F|_U,G|_U)$ defines a presheaf.

Now, we want to define a map of presheaves $E^i(F,G)\rightarrow \mathcal{Ext}^i(F,G)$. To do this, we note that we have an equality of complexes $$Hom_U(F|_U,I^\bullet|_U) = \Gamma(U,\mathcal{Hom}_U(F|_U,I^\bullet|_U))$$ If $\partial$ denotes the differential of the complex $\mathcal{Hom}_U(F|_U,I^\bullet|_U)$, and $d$ denotes the differential of the complex $Hom_U(F|_U,I^\bullet|_U)$, then we get a map $$\ker d^i\stackrel{=}{\longrightarrow}\Gamma(\ker\partial^i)\rightarrow\Gamma(\ker\partial^i/\text{im }\partial^{i-1})$$ which kills $\text{im }d^{i-1}$, and hence for every open $U$, we get compatible maps $$Ext^i_U(F|_U,G|_U)\rightarrow\Gamma(U,\mathcal{Ext}^i(F,G))$$ whence a morphism of presheaves $$\varphi : E^i(F,G)\rightarrow\mathcal{Ext}^i(F,G)$$ which seems to induce isomorphisms on open affines (is there anything to check here? must we impose any finiteness conditions like $X$ noetherian or $F$ coherent?).

If the above is correct, then $\varphi$ should induces an isomorphism on stalks, and hence identifies $\mathcal{Ext}^i(F,G)$ with the sheafification of $E^i(F,G)$.

Can someone check this proof and/or confirm the result? References would be appreciated as well. On the one hand it seems like a nice statement to have in mind, and yet on the other hand if it were true, I would have expected to have seen/heard it somewhere.

  • 1
    $\begingroup$ One thing to be said about Hartschorne's lemma is that his injectives are injectives in the category of $O_X$ modules and not in $QCoh(X)$. So restricting to smaller open is not localization of a module. $\endgroup$ Jun 16, 2017 at 19:57
  • $\begingroup$ @user45150 Hmm, are you saying that the restriction (in $Mod(X)$) of a quasicoherent $O_X$-module to an open $U\subset X$ is not necessarily quasicoherent? Do you have an example in mind? $\endgroup$ Jun 16, 2017 at 20:10
  • 1
    $\begingroup$ No. I am saying that when Hartschorne says injective sheaf $\mathcal{I}$, $\mathcal{I}$ is not assumed to be quasi-coherent, but only an $O_X$-module $\endgroup$ Jun 16, 2017 at 20:53
  • 1
    $\begingroup$ @user45150 I'm thinking about this again, and I'm confused now: If the restriction of a quasicoherent injective module to $U$ is also quasicoherent, then doesn't Hartshorne's Lemma 6.1 "restrictions of injective objects in $Mod(X)$ is an injective of $Mod(U)$" imply that the restriction of a quasicoherent injective is also quasicoherent injective? $\endgroup$ Jul 7, 2017 at 21:37
  • 1
    $\begingroup$ the injective objects are in $Mod(X)$ not in $QCoh(X)$. He never considers "quasicoherent injectives" $\endgroup$ Jul 8, 2017 at 15:44

1 Answer 1


This is proven in the Stacks Project as pointed out in the comments on math.stackexchange by essentially the same proof you suggested.

  • $\begingroup$ Ah, how embarassing. I had thought for some reason that this question was never resolved, so I didn't think to reread the comments. Oops. $\endgroup$ Jun 16, 2017 at 20:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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