32

Suppose $\mathcal{F}$ and $\mathcal{G}$ are sheaves on $X$. The sheaf hom from $\mathcal{F}$ to $\mathcal{G}$ is defined by $U \mapsto $ Hom($\mathcal{F}|_{U}$,$\mathcal{G}|_{U}$), where the Hom is taken in the category of presheaves, i.e., Hom($\mathcal{F}|_{U}$,$\mathcal{G}|_{U}$) is the set of all natural transformations from $\mathcal{F}|_{U}$ to $\mathcal{G}|_{U}$.

To verify the sheaf hom is a sheaf , I have to show that it is a presheaf. So I need to define a restriction map from Hom($\mathcal{F}|_{U}$,$\mathcal{G}|_{U}$) to Hom($\mathcal{F}|_V$,$\mathcal{G}|_V$) if $V$ is an open subset of $U$. There seems a natural restriction map by using the restriction maps for $\mathcal{F}$ and $\mathcal{G}$. But how can you describe it explicitly?

Bogdan
  • 1,904
user45955
  • 1,055
  • 9
  • 16

1 Answers1

45

Step $-1$ (unnecessary). Show that the set of sections Hom$(\mathscr F|_U,\mathscr G|_U)$ over an open subset $U\subset X$ is an abelian group (so that $\mathcal Hom(\mathscr F,\mathscr G)$ will be a sheaf of abelian groups). This is easy.

Step $0$. $U\mapsto \textrm{Hom}(\mathscr F|_U,\mathscr G|_U)$ is a presheaf (note: as pointed out in a comment, a section of this presheaf is a morphism of sheaves!). The restriction is defined as follows: for fixed $U$, and an open subset $V\subset U$, a section $\sigma\in\textrm{Hom}(\mathscr F|_U,\mathscr G|_U)$ goes to $\sigma|_V\in \textrm{Hom}(\mathscr F|_V,\mathscr G|_V)$, where $\sigma|_V$ is the morphism of sheaves on $V$ defined by $\sigma|_V(W)=\sigma(W):\mathscr F(W)\to\mathscr G(W)$ for any open subset $W\subset V$ (which is also open in $U$! for this reason, the squares that must commute, over $V$, do commute because they already commuted over $U$).

Step $1$. The first sheaf axiom. Let $U=\bigcup_{i\in I} U_i$ be an open covering of an open subset $U\subset X$. Let $\sigma: \mathscr F|_U\to\mathscr G|_U$ be a section such that $\sigma_i:=\sigma|_{U_i}=0$ for all $i\in I$. We want to show that $\sigma=0$.

Let $g\in\mathscr F(U)$ be a fixed section. Then look at the (zero!) morphisms of abelian groups $$ \sigma_i(U_i):\mathscr F(U_i)\to\mathscr G(U_i) \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,g_i\mapsto 0. $$ Now, because $\mathscr{G}$ is a sheaf and the image of the section $g$ under $\sigma(U)$ restricts to zero on every open set in an open cover $\{U_i \}$ (recall the usual commutative diagram for morphisms of (pre)sheaves/natural transformations), it has the property $$ \sigma(U)(g)=0. $$ Because this holds for every $g\in\mathscr F(U)$, we conclude that $\sigma(U)=0$, hence $\sigma=0$, as claimed.

Step 2. The second sheaf axiom. Let again $U=\bigcup_{i\in I} U_i$ be an open covering of an open subset $U\subset X$, and let $\{\phi_i:\mathscr F|_{U_i}\to\mathscr G|_{U_i}\}_{i\in I}$ be a family of sections such that $\phi_i=\phi_j$ on $U_{ij}$. We want a global $\phi$ (section over $U$) such that $\phi|_{U_i}=\phi_i$.

If $V\subset U$, then $A_i:=U_i\cap V$ cover $V$. So let us fix a section $g\in \mathscr F(V)$ and let us set $g_i:=g|_{A_i}$. We can give a name (say $t_i$) to the image of $g_i$ under $\phi(A_i)$, namely $$ \phi_i(A_i):\mathscr F(A_i)\to\mathscr G(A_i) \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,g_i\mapsto t_i. $$ The compatibility of the $\phi_i$'s implies that of the $t_i$'s, and since $\mathscr G$ is a sheaf there exists a global section $t\in \mathscr G(V)$ such that $t|_{A_i}=t_i$ for every $i$. We can define the $\phi$ that we are looking for by $$ \phi(V):\mathscr F(V)\to\mathscr G(V) \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, g\mapsto t. $$ for every $V\subset U$. In this way, by construction, $\phi|_{U_i}=\phi_i$, as wanted.

ggg
  • 1,283
  • 9
  • 26
Brenin
  • 14,072
  • Can you provide a proof of step $0$? How do you define the restriction map from Hom($\mathcal{F}|{U}$,$\mathcal{G}|{U}$) to Hom($\mathcal{F}|{V}$,$\mathcal{G}|{V}$)? – user45955 Feb 04 '13 at 22:59
  • @user45955: I edited step $0$. – Brenin Feb 04 '13 at 23:43
  • Shouldn't we check naturality for the last step? – Gil Jan 25 '14 at 00:49
  • 2
    @user123412 Yes, in some sense you should. However, since the restriction maps on the morphisms are just "using the same maps but on a smaller domain" then naturality follows from the fact that any $\sigma$ is a morphism, and so satisfies naturality. It is also possibly worth checking in the last step that $\phi(V)$ is a homomorphism, depending on how happy you are with that it really is true. – Tom Oldfield Dec 29 '14 at 10:17
  • 1
    Can you really say for every $g\in\mathscr F(U)$, we conclude that $\sigma(U)=0$, hence $\sigma=0$? Don't you need to check every $g\in\mathscr F(U \cap W)$ where $W$ is the open subset of $X$? Maybe it needs to be changed to "Let $g\in\mathscr F(U \cap W) $ be a fixed section"? – WWK Nov 07 '15 at 15:37
  • 1
    Yes, that could be more precise, thanks. But do you agree that if $\sigma(U)=0$ then $\sigma(U\cap W)=0$ as well? Hence $\sigma=0$. – Brenin Nov 10 '15 at 22:15
  • @Brenin I'm sorry I'm late to the party, but can you explain why we need to consider $V\subset U$ in step 2 and not just apply the entire argument on $U$ instead of $V$? – BlackAdder Jul 11 '16 at 06:09
  • 1
    @BlackAdder: never too late! So $U$ was fixed in order to check that local sections defined on a covering $(U_i)$ glue to global sections (over $U$). This requires working locally on $U$, so if I'm to define $\phi$ (global section of Hom on $U$), I need to give it to you on each open $V\subset U$. – Brenin Jul 17 '16 at 18:21
  • 1
    @Brenin Sorry...yet another late comment. In step 1, when you say "$\mathcal{F}$ is a sheaf," you really mean that $\mathcal{G}$ is a sheaf correct? I feel like this result doesn't rely on $\mathcal{F}$ being a sheaf. – Benighted Jan 10 '17 at 19:15
  • 5
    @spietro The result certainly holds if $\mathcal{F},\mathcal{G}$ are both sheaves. However, it suffices that $\mathcal{F}$ is only a presheaf. – mathematics2x2life Jan 18 '17 at 05:08
  • For the first axiom, you need to assume it is a sheaf of abelian groups, right? Because the first axiom as I recall was if f,g are two sections on U and they agree on restriction to an open cover, then f and g are equal. To deduce the definition I said above by Step 1 it requires there to be some sort of subtraction between section elements, am I right? So what about just sheaf of sets, rather than of abelian groups? – Hyacinth Aug 08 '20 at 00:05