The Stacks project

18.30 Towards constructible modules

Recall that a quasi-compact object of a site is roughly an object such that every covering of it can be refined by a finite covering (the actual definition is slightly more involved, see Sites, Section 7.17). It turns out that if every object of a site has a covering by quasi-compact objects, then the modules $j_!\mathcal{O}_ U$ with $U$ quasi-compact form a particularly nice set of generators for the category of all modules.

Lemma 18.30.1. Let $(\mathcal{C}, \mathcal{O})$ be a ringed site. Let $\{ U_ i \to U\} $ be a covering of $\mathcal{C}$. Then the sequence

\[ \bigoplus j_{U_ i \times _ U U_ j!}\mathcal{O}_{U_ i \times _ U U_ j} \to \bigoplus j_{U_ i!}\mathcal{O}_{U_ i} \to j_!\mathcal{O}_ U \to 0 \]

is exact.

Proof. For any $\mathcal{O}$-module $\mathcal{F}$ the functor $\mathop{\mathrm{Hom}}\nolimits _\mathcal {O}(-, \mathcal{F})$ turns our sequence into the exact sequence $0 \to \mathcal{F}(U) \to \prod \mathcal{F}(U_ i) \to \prod \mathcal{F}(U_ i \times _ U U_ j)$, see (18.19.2.1). The lemma follows from this and Homology, Lemma 12.5.8. $\square$

Lemma 18.30.2. Let $(\mathcal{C}, \mathcal{O})$ be a ringed site. Let $\mathcal{U} = \{ U_ i \to U\} _{i \in I}$ be covering of $\mathcal{C}$. If $U$ is quasi-compact, then there exist a finite subset $I' \subset I$ such that the sequence

\[ \bigoplus \nolimits _{i, i' \in I'} j_{U_ i \times _ U U_{i'}!}\mathcal{O}_{U_ i \times _ U U_{i'}} \to \bigoplus \nolimits _{i \in I'} j_{U_ i!}\mathcal{O}_{U_ i} \to j_!\mathcal{O}_ U \to 0 \]

is exact.

Proof. This lemma is immediate from Lemma 18.30.1 if $U$ satisfies condition (3) of Sites, Lemma 7.17.2. We urge the reader to skip the proof in the general case. By definition there exists a covering $\mathcal{V} = \{ V_ j \to U\} _{j \in J}$ and a morphism $\mathcal{V} \to \mathcal{U}$ of families of maps with fixed target given by $\text{id} : U \to U$, $\alpha : J \to I$, and $f_ j : V_ j \to U_{\alpha (j)}$ (see Sites, Definition 7.8.1) such that the image $I' \subset I$ of $\alpha $ is finite. By Homology, Lemma 12.5.8 it suffices to show that for any sheaf of $\mathcal{O}$-modules $\mathcal{F}$ the functor $\mathop{\mathrm{Hom}}\nolimits _\mathcal {O}(-, \mathcal{F})$ turns the sequence of the lemma into an exact sequence. By (18.19.2.1) we obtain the usual sequence

\[ 0 \to \mathcal{F}(U) \to \prod \nolimits _{i \in I'} \mathcal{F}(U_ i) \to \prod \nolimits _{i, i' \in I'} \mathcal{F}(U_ i \times _ U U_{i'}) \]

This is an exact sequence by Sites, Lemma 7.8.6 applied to the family of maps $\{ U_ i \to U\} _{i \in I'}$ which is refined by the covering $\mathcal{V}$. $\square$

Lemma 18.30.3. Let $\mathcal{C}$ be a site. Let $W$ be a quasi-compact object of $\mathcal{C}$.

  1. The functor $\mathop{\mathit{Sh}}\nolimits (\mathcal{C}) \to \textit{Sets}$, $\mathcal{F} \mapsto \mathcal{F}(W)$ commutes with coproducts.

  2. Let $\mathcal{O}$ be a sheaf of rings on $\mathcal{C}$. The functor $\textit{Mod}(\mathcal{O}) \to \textit{Ab}$, $\mathcal{F} \mapsto \mathcal{F}(W)$ commutes with direct sums.

Proof. Proof of (1). Taking sections over $W$ commutes with filtered colimits with injective transition maps by Sites, Lemma 7.17.7. If $\mathcal{F}_ i$ is a family of sheaves of sets indexed by a set $I$. Then $\coprod \mathcal{F}_ i$ is the filtered colimit over the partially ordered set of finite subsets $E \subset I$ of the coproducts $\mathcal{F}_ E = \coprod _{i \in E} \mathcal{F}_ i$. Since the transition maps are injective we conclude.

Proof of (2). Let $\mathcal{F}_ i$ be a family of sheaves of $\mathcal{O}$-modules indexed by a set $I$. Then $\bigoplus \mathcal{F}_ i$ is the filtered colimit over the partially ordered set of finite subsets $E \subset I$ of the direct sums $\mathcal{F}_ E = \bigoplus _{i \in E} \mathcal{F}_ i$. A filtered colimit of abelian sheaves can be computed in the category of sheaves of sets. Moreover, for $E \subset E'$ the transition map $\mathcal{F}_ E \to \mathcal{F}_{E'}$ is injective (as sheafification is exact and the injectivity is clear on underlying presheaves). Hence it suffices to show the result for a finite index set by Sites, Lemma 7.17.7. The finite case is dealt with in Lemma 18.3.2 (it holds over any object of $\mathcal{C}$). $\square$

Lemma 18.30.4. Let $(\mathcal{C}, \mathcal{O})$ be a ringed site. Let $U$ be a quasi-compact object of $\mathcal{C}$. Then the functor $\mathop{\mathrm{Hom}}\nolimits _\mathcal {O}(j_!\mathcal{O}_ U, -)$ commutes with direct sums.

Proof. This is true because $\mathop{\mathrm{Hom}}\nolimits _\mathcal {O}(j_!\mathcal{O}_ U, \mathcal{F}) = \mathcal{F}(U)$ by (18.19.2.1) and because the functor $\mathcal{F} \mapsto \mathcal{F}(U)$ commutes with direct sums by Lemma 18.30.3. $\square$

In order to state the sharpest possible results in the following we introduce some notation.

Situation 18.30.5. Let $\mathcal{C}$ be a site. Let $\mathcal{B} \subset \text{Ob}(\mathcal{C})$ be a set of objects. We consider the following conditions

  1. Every object of $\mathcal{C}$ has a covering by elements of $\mathcal{B}$.

  2. Every $U \in \mathcal{B}$ is quasi-compact (Sites, Section 7.17).

  3. For a covering $\{ U_ i \to U\} $ with $U_ i, U \in \mathcal{B}$ the fibre products $U_ i \times _ U U_ j$ are quasi-compact.

Lemma 18.30.6. In Situation 18.30.5 assume (1) holds.

  1. Every sheaf of sets is the target of a surjective map whose source is a coproduct $\coprod h_{U_ i}^\# $ with $U_ i$ in $\mathcal{B}$.

  2. If $\mathcal{O}$ is a sheaf of rings, then every $\mathcal{O}$-module is a quotient of a direct sum $\bigoplus \nolimits j_{U_ i!}\mathcal{O}_{U_ i}$ with $U_ i$ in $\mathcal{B}$.

Lemma 18.30.7. In Situation 18.30.5 assume (1) and (2) hold.

  1. Every sheaf of sets is a filtered colimit of sheaves of the form

    18.30.7.1
    \begin{equation} \label{sites-modules-equation-towards-constructible-sets} \text{Coequalizer}\left( \xymatrix{ \coprod \nolimits _{j = 1, \ldots , m} h_{V_ j}^\# \ar@<1ex>[r] \ar@<-1ex>[r] & \coprod \nolimits _{i = 1, \ldots , n} h_{U_ i}^\# } \right) \end{equation}

    with $U_ i$ and $V_ j$ in $\mathcal{B}$.

  2. If $\mathcal{O}$ is a sheaf of rings, then every $\mathcal{O}$-module is a filtered colimit of sheaves of the form

    18.30.7.2
    \begin{equation} \label{sites-modules-equation-towards-constructible} \mathop{\mathrm{Coker}}\left( \bigoplus \nolimits _{j = 1, \ldots , m} j_{V_ j!}\mathcal{O}_{V_ j} \longrightarrow \bigoplus \nolimits _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i} \right) \end{equation}

    with $U_ i$ and $V_ j$ in $\mathcal{B}$.

Proof. Proof of (1). By Lemma 18.30.6 every sheaf of sets $\mathcal{F}$ is the target of a surjection whose source is a coprod $\mathcal{F}_0$ of sheaves the form $h_{U}^\# $ with $U \in \mathcal{B}$. Applying this to $\mathcal{F}_0 \times _\mathcal {F} \mathcal{F}_0$ we find that $\mathcal{F}$ is a coequalizer of a pair of maps

\[ \xymatrix{ \coprod \nolimits _{j \in J} h_{V_ j}^\# \ar@<1ex>[r] \ar@<-1ex>[r] & \coprod \nolimits _{i \in I} h_{U_ i}^\# } \]

for some index sets $I$, $J$ and $V_ j$ and $U_ i$ in $\mathcal{B}$. For every finite subset $J' \subset J$ there is a finite subset $I' \subset I$ such that the coproduct over $j \in J'$ maps into the coprod over $i \in I'$ via both maps, see Sites, Lemma 7.17.7. (Details omitted; hint: an infinite coproduct is the filtered colimit of the finite sub-coproducts.) Thus our sheaf is the colimit of the cokernels of these maps between finite coproducts.

Proof of (2). By Lemma 18.30.6 every module is a quotient of a direct sum of modules of the form $j_{U!}\mathcal{O}_ U$ with $U \in \mathcal{B}$. Thus every module is a cokernel

\[ \mathop{\mathrm{Coker}}\left( \bigoplus \nolimits _{j \in J} j_{V_ j!}\mathcal{O}_{V_ j} \longrightarrow \bigoplus \nolimits _{i \in I} j_{U_ i!}\mathcal{O}_{U_ i} \right) \]

for some index sets $I$, $J$ and $V_ j$ and $U_ i$ in $\mathcal{B}$. For every finite subset $J' \subset J$ there is a finite subset $I' \subset I$ such that the direct sum over $j \in J'$ maps into the direct sum over $i \in I'$, see Lemma 18.30.4. Thus our module is the colimit of the cokernels of these maps between finite direct sums. $\square$

Proof. Let $\mathcal{F} = \mathop{\mathrm{Coker}}(\bigoplus j_{V_ j!}\mathcal{O}_{V_ j} \to \bigoplus j_{U_ i!}\mathcal{O}_{U_ i})$ as in (18.30.7.2). It suffices to show that the cokernel of a map $\varphi : j_{W!}\mathcal{O}_ W \to \mathcal{F}$ with $W \in \mathcal{B}$ is another module of the same type. The map $\varphi $ corresponds to $s \in \mathcal{F}(W)$. Since $\bigoplus j_{U_ i!}\mathcal{O}_{U_ i} \to \mathcal{F}$ is surjective, by (1) we may choose a covering $\{ W_ k \to W\} _{k \in K}$ with $W_ k \in \mathcal{B}$ such that $s|_{W_ k}$ is the image of some section $s_ k$ of $\bigoplus j_{U_ i!}\mathcal{O}_{U_ i})$. By (2) the object $W$ is quasi-compact. By Lemma 18.30.2 there is a finite subset $K' \subset K$ such that $\bigoplus _{k \in K'} j_{W_ k!}\mathcal{O}_{W_ k} \to j_{W!}\mathcal{O}_ W$ is surjective. We conclude that $\mathop{\mathrm{Coker}}(\varphi )$ is equal to

\[ \mathop{\mathrm{Coker}}\left( \bigoplus \nolimits _{k \in K'} j_{W_ k!}\mathcal{O}_{W_ k} \oplus \bigoplus j_{V_ j!}\mathcal{O}_{V_ j} \longrightarrow \bigoplus j_{U_ i!}\mathcal{O}_{U_ i} \right) \]

where the map $\bigoplus _{k \in K'} j_{W_ k!}\mathcal{O}_{W_ k} \to \bigoplus j_{U_ i!}\mathcal{O}_{U_ i}$ corresponds to $\sum _{k \in K'} s_ k$. This finishes the proof. $\square$

Lemma 18.30.9. In Situation 18.30.5 assume (1), (2), and (3) hold. Let $\mathcal{O}$ be a sheaf of rings. Assume given a map

\[ \bigoplus \nolimits _{j = 1, \ldots , m} j_{V_ j!}\mathcal{O}_{V_ j} \longrightarrow \bigoplus \nolimits _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i} \]

with $U_ i$ and $V_ j$ in $\mathcal{B}$, and coverings $\{ U_{ik} \to U_ i\} _{k \in K_ i}$ with $U_{ik} \in \mathcal{B}$. Then there exist finite subsets $K'_ i \subset K_ i$ and a finite set $L$ of $W_ l \in \mathcal{B}$ and a commutative diagram

\[ \xymatrix{ \bigoplus _{l \in L} j_{W_ l!}\mathcal{O}_{W_ l} \ar[d] \ar[r] & \bigoplus _{i = 1, \ldots , n} \bigoplus _{k \in K'_ i} j_{U_{ik}!}\mathcal{O}_{U_{ik}} \ar[d] \\ \bigoplus _{j = 1, \ldots , m} j_{V_ j!}\mathcal{O}_{V_ j} \ar[r] & \bigoplus _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i} } \]

inducing an isomorphism on cokernels of the horizontal maps.

Proof. Since $U_ i$ is quasi-compact, we may choose finite subsets $K'_ i \subset K_ i$ as in Lemma 18.30.2. Then since $\bigoplus _{i = 1, \ldots , n} \bigoplus _{k \in K'_ i} j_{U_{ik}!}\mathcal{O}_{U_{ik}} \to \bigoplus _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i}$ is surjective, we can find coverings $\{ V_{jm} \to V_ j\} _{m \in M_ j}$ with $V_{jm} \in \mathcal{B}$ such that we can find a commutative diagram

\[ \xymatrix{ \bigoplus _{j = 1, \ldots , m} \bigoplus _{m \in M_ j} j_{V_{jm}!}\mathcal{O}_{V_{jm}} \ar[d] \ar[r] & \bigoplus _{i = 1, \ldots n} \bigoplus _{k \in K'_ i} j_{U_{ik}!}\mathcal{O}_{U_{ik}} \ar[d] \\ \bigoplus _{j = 1, \ldots , m} j_{V_ j!}\mathcal{O}_{V_ j} \ar[r] & \bigoplus _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i} } \]

Since $V_ j$ is quasi-compact, we can choose finite subsets $M'_ j \subset M_ j$ as in Lemma 18.30.2. Set

\[ L = \left(\coprod \nolimits _{i = 1, \ldots , n} K'_ i \times K'_ i \right) \coprod \left(\coprod \nolimits _{j = 1, \ldots , m} M'_ j\right) \]

and for $l = (k, k') \in K'_ i \times K'_ i \subset L$ set $W_ l = U_{ik} \times _{U_ i} U_{ik'}$ and for $l = m \in M'_ j \subset L$ set $W_ l = V_{jm}$. Since we have the exact sequences of Lemma 18.30.2 for the families $\{ U_{ik} \to U_ i\} _{k \in K'_ i}$ we conclude that we get a diagram as in the statement of the lemma (details omitted), except that it is not yet clear that $W_ l \in \mathcal{B}$. However, since $W_ l$ is quasi-compact for all $l \in L$ we do another application of Lemma 18.30.2 and find finite families of maps $\{ W_{lt} \to W_ l\} _{t \in T_ l}$ with $W_{lt} \in \mathcal{B}$ such that $\bigoplus j_{W_{lt}!}\mathcal{O}_{W_{lt}} \to j_{W_ l!}\mathcal{O}_{W_ l}$ is surjective. Then we replace $L$ by $\coprod _{l \in L} T_ l$ and everything is clear. $\square$

Proof. Let $0 \to \mathcal{F}_1 \to \mathcal{F}_2 \to \mathcal{F}_3 \to 0$ be a short exact sequence of $\mathcal{O}$-modules with $\mathcal{F}_1$ and $\mathcal{F}_3$ as in (18.30.7.2). Choose presentations

\[ \bigoplus A_{V_ j} \to \bigoplus A_{U_ i} \to \mathcal{F}_1 \to 0 \quad \text{and}\quad \bigoplus A_{T_ j} \to \bigoplus A_{W_ i} \to \mathcal{F}_3 \to 0 \]

In this proof the direct sums are always finite, and we write $A_ U = j_{U!}\mathcal{O}_ U$ for $U \in \mathcal{B}$. Since $\mathcal{F}_2 \to \mathcal{F}_3$ is surjective, we can choose coverings $\{ W_{ik} \to W_ i\} $ with $W_{ik} \in \mathcal{B}$ such that $A_{W_{ik}} \to \mathcal{F}_3$ lifts to a map $A_{W_{ik}} \to \mathcal{F}_2$. By Lemma 18.30.9 we may replace our collection $\{ W_ i\} $ by a finite subcollection of the collection $\{ W_{ik}\} $ and assume the map $\bigoplus A_{W_ i} \to \mathcal{F}_3$ lifts to a map into $\mathcal{F}_2$. Consider the kernel

\[ \mathcal{K}_2 = \mathop{\mathrm{Ker}}(\bigoplus A_{U_ i} \oplus \bigoplus A_{W_ i} \longrightarrow \mathcal{F}_2) \]

By the snake lemma this kernel surjects onto $\mathcal{K}_3 = \mathop{\mathrm{Ker}}(\bigoplus A_{W_ i} \to \mathcal{F}_3)$. Thus, arguing as above, after replacing each $T_ j$ by a finite family of elements of $\mathcal{B}$ (permissible by Lemma 18.30.2) we may assume there is a map $\bigoplus A_{T_ j} \to \mathcal{K}_2$ lifting the given map $\bigoplus A_{T_ j} \to \mathcal{K}_3$. Then $\bigoplus A_{V_ j} \oplus \bigoplus A_{T_ j} \to \mathcal{K}_2$ is surjective which finishes the proof. $\square$

Lemma 18.30.11. In Situation 18.30.5 assume (1), (2), and (3) hold. Let $\mathcal{O}$ be a sheaf of rings. Let $\mathcal{A} \subset \textit{Mod}(\mathcal{O})$ be the full subcategory of modules isomorphic to a cokernel as in (18.30.7.2). If the kernel of every map of $\mathcal{O}$-modules of the form

\[ \bigoplus \nolimits _{j = 1, \ldots , m} j_{V_ j!}\mathcal{O}_{V_ j} \longrightarrow \bigoplus \nolimits _{i = 1, \ldots , n} j_{U_ i!}\mathcal{O}_{U_ i} \]

with $U_ i$ and $V_ j$ in $\mathcal{B}$, is in $\mathcal{A}$, then $\mathcal{A}$ is weak Serre subcategory of $\textit{Mod}(\mathcal{O})$.

Proof. We will use the criterion of Homology, Lemma 12.10.3. By the results of Lemmas 18.30.8 and 18.30.10 it suffices to see that the kernel of a map $\mathcal{F} \to \mathcal{G}$ between objects of $\mathcal{A}$ is in $\mathcal{A}$. To prove this choose presentations

\[ \bigoplus A_{V_ j} \to \bigoplus A_{U_ i} \to \mathcal{F} \to 0 \quad \text{and}\quad \bigoplus A_{T_ j} \to \bigoplus A_{W_ i} \to \mathcal{G} \to 0 \]

In this proof the direct sums are always finite, and we write $A_ U = j_{U!}\mathcal{O}_ U$ for $U \in \mathcal{B}$. Using Lemmas 18.30.1 and 18.30.9 and arguing as in the proof of Lemma 18.30.10 we may assume that the map $\mathcal{F} \to \mathcal{G}$ lifts to a map of presentations

\[ \xymatrix{ \bigoplus A_{V_ j} \ar[r] \ar[d] & \bigoplus A_{U_ i} \ar[r] \ar[d] & \mathcal{F} \ar[r] \ar[d] & 0 \\ \bigoplus A_{T_ j} \ar[r] & \bigoplus A_{W_ i} \ar[r] & \mathcal{G} \ar[r] & 0 } \]

Then we see that

\[ \mathop{\mathrm{Ker}}(\mathcal{F} \to \mathcal{G}) = \mathop{\mathrm{Coker}}\left(\bigoplus A_{V_ j} \to \mathop{\mathrm{Ker}}\left( \bigoplus A_{T_ j} \oplus \bigoplus A_{U_ i} \to \bigoplus A_{W_ i}\right)\right) \]

and the lemma follows from the assumption and Lemma 18.30.8. $\square$


Comments (0)


Post a comment

Your email address will not be published. Required fields are marked.

In your comment you can use Markdown and LaTeX style mathematics (enclose it like $\pi$). A preview option is available if you wish to see how it works out (just click on the eye in the toolbar).

Unfortunately JavaScript is disabled in your browser, so the comment preview function will not work.

All contributions are licensed under the GNU Free Documentation License.




In order to prevent bots from posting comments, we would like you to prove that you are human. You can do this by filling in the name of the current tag in the following input field. As a reminder, this is tag 0933. Beware of the difference between the letter 'O' and the digit '0'.