The Stacks project

Theorem 38.39.7. Let $p$ be a prime number. Let $Y$ be a quasi-compact and quasi-separated scheme over $\mathbf{F}_ p$. Let $f : X \to Y$ be a proper, surjective morphism of finite presentation with geometrically connected fibres. Then the functor

\[ \mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(Y) \longrightarrow \mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(X) \]

is fully faithful with essential image described as follows. Let $\mathcal{E}$ be a finite locally free $\mathcal{O}_ X$-module. Assume for all $y \in Y$ there exists integers $n_ y, r_ y \geq 0$ such that

\[ F^{n_ y, *}\mathcal{E}|_{X_{y, red}} \cong \mathcal{O}_{X_{y, red}}^{\oplus r_ y} \]

Then for some $n \geq 0$ the $n$th Frobenius power pullback $F^{n, *}\mathcal{E}$ is the pullback of a finite locally free $\mathcal{O}_ Y$-module.

Proof. Proof of fully faithfulness. Since vectorbundles on $Y$ are locally trivial, this reduces to the statement that

\[ \mathop{\mathrm{colim}}\nolimits _ F \Gamma (Y, \mathcal{O}_ Y) \longrightarrow \mathop{\mathrm{colim}}\nolimits _ F \Gamma (X, \mathcal{O}_ X) \]

is bijective. Since $\{ X \to Y\} $ is an h covering, this will follow from Lemma 38.38.2 if we can show that the two maps

\[ \mathop{\mathrm{colim}}\nolimits _ F \Gamma (X, \mathcal{O}_ X) \longrightarrow \mathop{\mathrm{colim}}\nolimits _ F \Gamma (X \times _ Y X, \mathcal{O}_{X \times _ Y X}) \]

are equal. Let $g \in \Gamma (X, \mathcal{O}_ X)$ and denote $g_1$ and $g_2$ the two pullbacks of $g$ to $X \times _ Y X$. Since $X_{y, red}$ is geometrically connected, we see that $H^0(X_{y, red}, \mathcal{O}_{X_{y, red}})$ is a purely inseparable extension of $\kappa (y)$, see Varieties, Lemma 33.9.3. Thus $g^ q|_{X_{y, red}}$ comes from an element of $\kappa (y)$ for some $p$-power $q$ (which may depend on $y$). It follows that $g_1^ q$ and $g_2^ q$ map to the same element of the residue field at any point of $(X \times _ Y X)_ y = X_ y \times _ y X_ y$. Hence $g_1 - g_2$ restricts to zero on $(X \times _ Y X)_{red}$. Hence $(g_1 - g_2)^ n = 0$ for some $n$ which we may take to be a $p$-power as desired.

Description of essential image. Let $\mathcal{E}$ be as in the statement of the proposition. We first reduce to the Noetherian case.

Let $y \in Y$ be a point and view it as a morphism $y \to Y$ from the spectrum of the residue field into $Y$. We can write $y \to Y$ as a filtered limit of morphisms $Y_ i \to Y$ of finite presentation with $Y_ i$ affine. (It is best to prove this yourself, but it also follows formally from Limits, Lemma 32.7.2 and 32.4.13.) For each $i$ set $Z_ i = Y_ i \times _ Y X$. Then $X_ y = \mathop{\mathrm{lim}}\nolimits Z_ i$ and $X_{y, red} = \mathop{\mathrm{lim}}\nolimits Z_{i, red}$. By Limits, Lemma 32.10.2 we can find an $i$ such that $F^{n_ y, *}\mathcal{E}|_{Z_{i, red}} \cong \mathcal{O}_{Z_{i, red}}^{\oplus r_ y}$. Fix $i$. We have $Z_{i, red} = \mathop{\mathrm{lim}}\nolimits Z_{i, j}$ where $Z_{i, j} \to Z_ i$ is a thickening of finite presentation (Limits, Lemma 32.9.4). Using the same lemma as before we can find a $j$ such that $F^{n_ y, *}\mathcal{E}|_{Z_{i, j}} \cong \mathcal{O}_{Z_{i, j}}^{\oplus r_ y}$. We conclude that for each $y \in Y$ there exists a morphism $Y_ y \to Y$ of finite presentation whose image contains $y$ and a thickening $Z_ y \to Y_ y \times _ Y X$ such that $F^{n_ y, *}\mathcal{E}|_{Z_ y} \cong \mathcal{O}_{Z_ y}^{\oplus r_ y}$. Observe that the image of $Y_ y \to Y$ is constructible (Morphisms, Lemma 29.22.2). Since $Y$ is quasi-compact in the constructible topology (Topology, Lemma 5.23.2 and Properties, Lemma 28.2.4) we conclude that there are a finite number of morphisms

\[ Y_1 \to Y,\ Y_2 \to Y,\ \ldots ,\ Y_ N \to Y \]

of finite presentation such that $Y = \bigcup \mathop{\mathrm{Im}}(Y_ a \to Y)$ set theoretically and such that for each $a \in \{ 1, \ldots , N\} $ there exist integers $n_ a, r_ a \geq 0$ and there is a thickening $Z_ a \subset Y_ a \times _ Y X$ of finite presentation such that $F^{n_ a, *}\mathcal{E}|_{Z_ a} \cong \mathcal{O}_{Z_ a}^{\oplus r_ a}$.

Formulated in this way, the condition descends to an absolute Noetherian approximation. We stronly urge the reader to skip this paragraph. First write $Y = \mathop{\mathrm{lim}}\nolimits _{i \in I} Y_ i$ as a cofiltered limit of schemes of finite type over $\mathbf{F}_ p$ with affine transition morphisms (Limits, Lemma 32.7.2). Next, we can assume we have proper morphisms $f_ i : X_ i \to Y_ i$ whose base change to $Y$ recovers $f : X \to Y$, see Limits, Lemma 32.10.1. After increasing $i$ we may assume there exists a finite locally free $\mathcal{O}_{X_ i}$-module $\mathcal{E}_ i$ whose pullback to $X$ is isomorphic to $\mathcal{E}$, see Limits, Lemma 32.10.3. Pick $0 \in I$ and denote $E \subset Y_0$ the constructible subset where the geometric fibres of $f_0$ are connected, see More on Morphisms, Lemma 37.28.6. Then $Y \to Y_0$ maps into $E$, see More on Morphisms, Lemma 37.28.2. Thus $Y_ i \to Y_0$ maps into $E$ for $i \gg 0$, see Limits, Lemma 32.4.10. Hence we see that the fibres of $f_ i$ are geometrically connected for $i \gg 0$. By Limits, Lemma 32.10.1 for large enough $i$ we can find morphisms $Y_{i, a} \to Y_ i$ of finite type whose base change to $Y$ recovers $Y_ a \to Y$, $a \in \{ 1, \ldots , N\} $. After possibly increasing $i$ we can find thickenings $Z_{i, a} \subset Y_{i, a} \times _{Y_ i} X_ i$ whose base change to $Y_ a \times _ Y X$ recovers $Z_ a$ (same reference as before combined with Limits, Lemmas 32.8.5 and 32.8.15). Since $Z_ a = \mathop{\mathrm{lim}}\nolimits Z_{i, a}$ we find that after increasing $i$ we may assume $F^{n_ a, *}\mathcal{E}_ i|_{Z_{i, a}} \cong \mathcal{O}_{Z_{i, a}}^{\oplus r_ a}$, see Limits, Lemma 32.10.2. Finally, after increasing $i$ one more time we may assume $\coprod Y_{i, a} \to Y_ i$ is surjective by Limits, Lemma 32.8.15. At this point all the assumptions hold for $X_ i \to Y_ i$ and $\mathcal{E}_ i$ and we see that it suffices to prove result for $X_ i \to Y_ i$ and $\mathcal{E}_ i$.

Assume $Y$ is of finite type over $\mathbf{F}_ p$. To prove the result we will use induction on $\dim (Y)$. We are trying to find an object of $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(Y)$ which pulls back to the object of $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(X)$ determined by $\mathcal{E}$. By the fully faithfulness already proven and because of Proposition 38.39.4 it suffices to construct a descent of $\mathcal{E}$ after replacing $Y$ by the members of a h covering and $X$ by the corresponding base change. This means that we may replace $Y$ by a scheme proper and surjective over $Y$ provided this does not increase the dimension of $Y$. If $T \subset T'$ is a thickening of schemes of finite type over $\mathbf{F}_ p$ then $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(T) = \mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(T')$ as $\{ T \to T'\} $ is a h covering such that $T \times _{T'} T = T$. If $T' \to T$ is a universal homeomorphism of schemes of finite type over $\mathbf{F}_ p$, then $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(T) = \mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(T')$ as $\{ T \to T'\} $ is a h covering such that the diagonal $T \subset T \times _{T'} T$ is a thickening.

Using the general remarks made above, we may and do replace $X$ by its reduction and we may assume $X$ is reduced. Consider the Stein factorization $X \to Y' \to Y$, see More on Morphisms, Theorem 37.53.4. Then $Y' \to Y$ is a universal homeomorphism of schemes of finite type over $\mathbf{F}_ p$. By the above we may replace $Y$ by $Y'$. Thus we may assume $f_*\mathcal{O}_ X = \mathcal{O}_ Y$ and that $Y$ is reduced. This reduces us to the case discussed in the next paragraph.

Assume $Y$ is reduced and $f_*\mathcal{O}_ X = \mathcal{O}_ Y$ over a dense open subscheme of $Y$. Then $X \to Y$ is flat over a dense open subscheme $V \subset Y$, see Morphisms, Proposition 29.27.2. By Lemma 38.31.1 there is a $V$-admissible blowing up $Y' \to Y$ such that the strict transform $X'$ of $X$ is flat over $Y'$. Observe that $\dim (Y') = \dim (Y)$ as $Y$ and $Y'$ have a common dense open subscheme. By More on Morphisms, Lemma 37.53.7 and the fact that $V \subset Y'$ is dense all fibres of $f' : X' \to Y'$ are geometrically connected. We still have $(f'_*\mathcal{O}_{X'})|_ V = \mathcal{O}_ V$. Write

\[ Y' \times _ Y X = X' \cup E \times _ Y X \]

where $E \subset Y'$ is the exceptional divisor of the blowing up. By the general remarks above, it suffices to prove existence for $Y' \times _ Y X \to Y'$ and the restriction of $\mathcal{E}$ to $Y' \times _ Y X$. Suppose that we find some object $\xi '$ in $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(Y')$ pulling back to the restriction of $\mathcal{E}$ to $X'$ (viewed as an object of the colimit category). By induction on $\dim (Y)$ we can find an object $\xi ''$ in $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(E)$ pulling back to the restriction of $\mathcal{E}$ to $E \times _ Y X$. Then the fully faithfullness determines a unique isomorphism $\xi '|_ E \to \xi ''$ compatible with the given identifications with the restriction of $\mathcal{E}$ to $E \times _{Y'} X'$. Since

\[ \{ E \times _ Y X \to Y' \times _ Y X, X' \to Y' \times _ Y X\} \]

is a h covering given by a pair of closed immersions with

\[ (E \times _ Y X) \times _{(Y' \times _ Y X)} X' = E \times _{Y'} X' \]

we conclude that $\xi '$ pulls back to the restriction of $\mathcal{E}$ to $Y' \times _ Y X$. Thus it suffices to find $\xi '$ and we reduce to the case discussed in the next paragraph.

Assume $Y$ is reduced, $f$ is flat, and $f_*\mathcal{O}_ X = \mathcal{O}_ Y$ over a dense open subscheme of $Y$. In this case we consider the normalization $Y^\nu \to Y$ (Morphisms, Section 29.54). This is a finite surjective morphism (Morphisms, Lemma 29.54.10 and 29.18.2) which is an isomorphism over a dense open. Hence by our general remarks we may replace $Y$ by $Y^\nu $ and $X$ by $Y^\nu \times _ Y X$. After this replacement we see that $\mathcal{O}_ Y = f_*\mathcal{O}_ X$ (because the Stein factorization has to be an isomorphism in this case; small detail omitted).

Assume $Y$ is a normal Noetherian scheme, that $f$ is flat, and that $f_*\mathcal{O}_ X = \mathcal{O}_ Y$. After replacing $\mathcal{E}$ by a suitable Frobenius power pullback, we may assume $\mathcal{E}$ is trivial on the scheme theoretic fibres of $f$ at the generic points of the irreducible components of $Y$ (because $\mathop{\mathrm{colim}}\nolimits _ F \textit{Vect}(-)$ is an equivalence on universal homeomorphisms, see above). Similarly to the arguments above (in the reduction to the Noetherian case) we conclude there is a dense open subscheme $V \subset Y$ such that $\mathcal{E}|_{f^{-1}(V)}$ is free. Let $Z \subset Y$ be a closed subscheme such that $Y = V \amalg Z$ set theoretically. Let $z_1, \ldots , z_ t \in Z$ be the generic points of the irreducible components of $Z$ of codimension $1$. Then $A_ i = \mathcal{O}_{Y, z_ i}$ is a discrete valuation ring. Let $n_ i$ be the integer found in Lemma 38.39.5 for the scheme $X_{A_ i}$ over $A_ i$. After replacing $\mathcal{E}$ by a suitable Frobenius power pullback, we may assume $\mathcal{E}$ is free over $X_{A_ i/\mathfrak m_ i^{n_ i}}$ (again because the colimit category is invariant under universal homeomorphisms, see above). Then Lemma 38.39.5 tells us that $\mathcal{E}$ is free on $X_{A_ i}$. Thus finally we conclude by applying Lemma 38.39.6. $\square$


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 0EXH. Beware of the difference between the letter 'O' and the digit '0'.