The Stacks project

59.88 Base change for higher direct images

This section is the analogue of Section 59.87 for higher direct images. This section is preliminary and should be skipped on a first reading.

Remark 59.88.1. Let $f : X \to S$ be a morphism of schemes. Let $n$ be an integer. We will say $BC(f, n, q_0)$ is true if for every commutative diagram

\[ \xymatrix{ X \ar[d]_ f & X' \ar[l] \ar[d]_{f'} & Y \ar[l]^ h \ar[d]^ e \\ S & S' \ar[l] & T \ar[l]_ g } \]

with $X' = X \times _ S S'$ and $Y = X' \times _{S'} T$ and $g$ quasi-compact and quasi-separated, and every abelian sheaf $\mathcal{F}$ on $T_{\acute{e}tale}$ annihilated by $n$ the base change map

\[ (f')^{-1}R^ qg_*\mathcal{F} \longrightarrow R^ qh_*e^{-1}\mathcal{F} \]

is an isomorphism for $q \leq q_0$.

Lemma 59.88.2. With $f : X \to S$ and $n$ as in Remark 59.88.1 assume for some $q \geq 1$ we have $BC(f, n, q - 1)$. Then for every commutative diagram

\[ \xymatrix{ X \ar[d]_ f & X' \ar[l] \ar[d]_{f'} & Y \ar[l]^ h \ar[d]^ e \\ S & S' \ar[l] & T \ar[l]_ g } \]

with $X' = X \times _ S S'$ and $Y = X' \times _{S'} T$ and $g$ quasi-compact and quasi-separated, and every abelian sheaf $\mathcal{F}$ on $T_{\acute{e}tale}$ annihilated by $n$

  1. the base change map $(f')^{-1}R^ qg_*\mathcal{F}\to R^ qh_*e^{-1}\mathcal{F}$ is injective,

  2. if $\mathcal{F} \subset \mathcal{G}$ where $\mathcal{G}$ on $T_{\acute{e}tale}$ is annihilated by $n$, then

    \[ \mathop{\mathrm{Coker}}\left( (f')^{-1}R^ qg_*\mathcal{F}\to R^ qh_*e^{-1}\mathcal{F} \right) \subset \mathop{\mathrm{Coker}}\left( (f')^{-1}R^ qg_*\mathcal{G}\to R^ qh_*e^{-1}\mathcal{G} \right) \]
  3. if in (2) the sheaf $\mathcal{G}$ is an injective sheaf of $\mathbf{Z}/n\mathbf{Z}$-modules, then

    \[ \mathop{\mathrm{Coker}}\left((f')^{-1}R^ qg_*\mathcal{F}\to R^ qh_*e^{-1}\mathcal{F} \right) \subset R^ qh_*e^{-1}\mathcal{G} \]

Proof. Choose a short exact sequence $0 \to \mathcal{F} \to \mathcal{I} \to \mathcal{Q} \to 0$ where $\mathcal{I}$ is an injective sheaf of $\mathbf{Z}/n\mathbf{Z}$-modules. Consider the induced diagram

\[ \xymatrix{ (f')^{-1}R^{q - 1}g_*\mathcal{I} \ar[d]_{\cong } \ar[r] & (f')^{-1}R^{q - 1}g_*\mathcal{Q} \ar[d]_{\cong } \ar[r] & (f')^{-1}R^ qg_*\mathcal{F} \ar[d] \ar[r] & 0 \ar[d] \\ R^{q - 1}h_*e^{-1}\mathcal{I} \ar[r] & R^{q - 1}h_*e^{-1}\mathcal{Q} \ar[r] & R^ qh_*e^{-1}\mathcal{F} \ar[r] & R^ qh_*e^{-1}\mathcal{I} } \]

with exact rows. We have the zero in the right upper corner as $\mathcal{I}$ is injective. The left two vertical arrows are isomorphisms by $BC(f, n, q - 1)$. We conclude that part (1) holds. The above also shows that

\[ \mathop{\mathrm{Coker}}\left( (f')^{-1}R^ qg_*\mathcal{F}\to R^ qh_*e^{-1}\mathcal{F} \right) \subset R^ qh_*e^{-1}\mathcal{I} \]

hence part (3) holds. To prove (2) choose $\mathcal{F} \subset \mathcal{G} \subset \mathcal{I}$. $\square$

Lemma 59.88.3. With $f : X \to S$ and $n$ as in Remark 59.88.1 assume for some $q \geq 1$ we have $BC(f, n, q - 1)$. Consider commutative diagrams

\[ \vcenter { \xymatrix{ X \ar[d]_ f & X' \ar[d]_{f'} \ar[l] & Y \ar[l]^ h \ar[d]^ e & Y' \ar[l]^{\pi '} \ar[d]^{e'} \\ S & S' \ar[l] & T \ar[l]_ g & T' \ar[l]_\pi } } \quad \text{and}\quad \vcenter { \xymatrix{ X' \ar[d]_{f'} & & Y' \ar[ll]^{h' = h \circ \pi '} \ar[d]^{e'} \\ S' & & T' \ar[ll]_{g' = g \circ \pi } } } \]

where all squares are cartesian, $g$ quasi-compact and quasi-separated, and $\pi $ is integral surjective. Let $\mathcal{F}$ be an abelian sheaf on $T_{\acute{e}tale}$ annihilated by $n$ and set $\mathcal{F}' = \pi ^{-1}\mathcal{F}$. If the base change map

\[ (f')^{-1}R^ qg'_*\mathcal{F}' \longrightarrow R^ qh'_*(e')^{-1}\mathcal{F}' \]

is an isomorphism, then the base change map $(f')^{-1}R^ qg_*\mathcal{F} \to R^ qh_*e^{-1}\mathcal{F}$ is an isomorphism.

Proof. Observe that $\mathcal{F} \to \pi _*\pi ^{-1}\mathcal{F}'$ is injective as $\pi $ is surjective (check on stalks). Thus by Lemma 59.88.2 we see that it suffices to show that the base change map

\[ (f')^{-1}R^ qg_*\pi _*\mathcal{F}' \longrightarrow R^ qh_*e^{-1}\pi _*\mathcal{F}' \]

is an isomorphism. This follows from the assumption because we have $R^ qg_*\pi _*\mathcal{F}' = R^ qg'_*\mathcal{F}'$, we have $e^{-1}\pi _*\mathcal{F}' =\pi '_*(e')^{-1}\mathcal{F}'$, and we have $R^ qh_*\pi '_*(e')^{-1}\mathcal{F}' = R^ qh'_*(e')^{-1}\mathcal{F}'$. This follows from Lemmas 59.55.4 and 59.43.5 and the relative leray spectral sequence (Cohomology on Sites, Lemma 21.14.7). $\square$

Lemma 59.88.4. With $f : X \to S$ and $n$ as in Remark 59.88.1 assume for some $q \geq 1$ we have $BC(f, n, q - 1)$. Consider commutative diagrams

\[ \vcenter { \xymatrix{ X \ar[d]_ f & X' \ar[d]_{f'} \ar[l] & X'' \ar[l]^{\pi '} \ar[d]_{f''} & Y \ar[l]^{h'} \ar[d]^ e \\ S & S' \ar[l] & S'' \ar[l]_\pi & T \ar[l]_{g'} } } \quad \text{and}\quad \vcenter { \xymatrix{ X' \ar[d]_{f'} & & Y \ar[ll]^{h = h' \circ \pi '} \ar[d]^ e \\ S' & & T \ar[ll]_{g = g' \circ \pi } } } \]

where all squares are cartesian, $g'$ quasi-compact and quasi-separated, and $\pi $ is integral. Let $\mathcal{F}$ be an abelian sheaf on $T_{\acute{e}tale}$ annihilated by $n$. If the base change map

\[ (f')^{-1}R^ qg_*\mathcal{F} \longrightarrow R^ qh_*e^{-1}\mathcal{F} \]

is an isomorphism, then the base change map $(f'')^{-1}R^ qg'_*\mathcal{F} \to R^ qh'_*e^{-1}\mathcal{F}$ is an isomorphism.

Proof. Since $\pi $ and $\pi '$ are integral we have $R\pi _* = \pi _*$ and $R\pi '_* = \pi '_*$, see Lemma 59.43.5. We also have $(f')^{-1}\pi _* = \pi '_*(f'')^{-1}$. Thus we see that $\pi '_*(f'')^{-1}R^ qg'_*\mathcal{F} = (f')^{-1}R^ qg_*\mathcal{F}$ and $\pi '_*R^ qh'_*e^{-1}\mathcal{F} = R^ qh_*e^{-1}\mathcal{F}$. Thus the assumption means that our map becomes an isomorphism after applying the functor $\pi '_*$. Hence we see that it is an isomorphism by Lemma 59.43.5. $\square$

Lemma 59.88.5. Let $T$ be a quasi-compact and quasi-separated scheme. Let $P$ be a property for quasi-compact and quasi-separated schemes over $T$. Assume

  1. If $T'' \to T'$ is a thickening of quasi-compact and quasi-separated schemes over $T$, then $P(T'')$ if and only if $P(T')$.

  2. If $T' = \mathop{\mathrm{lim}}\nolimits T_ i$ is a limit of an inverse system of quasi-compact and quasi-separated schemes over $T$ with affine transition morphisms and $P(T_ i)$ holds for all $i$, then $P(T')$ holds.

  3. If $Z \subset T'$ is a closed subscheme with quasi-compact complement $V \subset T'$ and $P(T')$ holds, then either $P(V)$ or $P(Z)$ holds.

Then $P(T)$ implies $P(\mathop{\mathrm{Spec}}(K))$ for some morphism $\mathop{\mathrm{Spec}}(K) \to T$ where $K$ is a field.

Proof. Consider the set $\mathfrak T$ of closed subschemes $T' \subset T$ such that $P(T')$. By assumption (2) this set has a minimal element, say $T'$. By assumption (1) we see that $T'$ is reduced. Let $\eta \in T'$ be the generic point of an irreducible component of $T'$. Then $\eta = \mathop{\mathrm{Spec}}(K)$ for some field $K$ and $\eta = \mathop{\mathrm{lim}}\nolimits V$ where the limit is over the affine open subschemes $V \subset T'$ containing $\eta $. By assumption (3) and the minimality of $T'$ we see that $P(V)$ holds for all these $V$. Hence $P(\eta )$ by (2) and the proof is complete. $\square$

Lemma 59.88.6. With $f : X \to S$ and $n$ as in Remark 59.88.1 assume for some $q \geq 1$ we have that $BC(f, n, q - 1)$ is true, but $BC(f, n, q)$ is not. Then there exist a commutative diagram

\[ \xymatrix{ X \ar[d]_ f & X' \ar[d]_{f'} \ar[l] & Y \ar[l]^ h \ar[d]^ e \\ S & S' \ar[l] & \mathop{\mathrm{Spec}}(K) \ar[l]_ g } \]

where $X' = X \times _ S S'$, $Y = X' \times _{S'} \mathop{\mathrm{Spec}}(K)$, $K$ is a field, and $\mathcal{F}$ is an abelian sheaf on $\mathop{\mathrm{Spec}}(K)$ annihilated by $n$ such that $(f')^{-1}R^ qg_*\mathcal{F} \to R^ qh_*e^{-1}\mathcal{F}$ is not an isomorphism.

Proof. Choose a commutative diagram

\[ \xymatrix{ X \ar[d]_ f & X' \ar[l] \ar[d]_{f'} & Y \ar[l]^ h \ar[d]^ e \\ S & S' \ar[l] & T \ar[l]_ g } \]

with $X' = X \times _ S S'$ and $Y = X' \times _{S'} T$ and $g$ quasi-compact and quasi-separated, and an abelian sheaf $\mathcal{F}$ on $T_{\acute{e}tale}$ annihilated by $n$ such that the base change map $(f')^{-1}R^ qg_*\mathcal{F} \to R^ qh_*e^{-1}\mathcal{F}$ is not an isomorphism. Of course we may and do replace $S'$ by an affine open of $S'$; this implies that $T$ is quasi-compact and quasi-separated. By Lemma 59.88.2 we see $(f')^{-1}R^ qg_*\mathcal{F} \to R^ qh_*e^{-1}\mathcal{F}$ is injective. Pick a geometric point $\overline{x}$ of $X'$ and an element $\xi $ of $(R^ qh_*q^{-1}\mathcal{F})_{\overline{x}}$ which is not in the image of the map $((f')^{-1}R^ qg_*\mathcal{F})_{\overline{x}} \to (R^ qh_*e^{-1}\mathcal{F})_{\overline{x}}$.

Consider a morphism $\pi : T' \to T$ with $T'$ quasi-compact and quasi-separated and denote $\mathcal{F}' = \pi ^{-1}\mathcal{F}$. Denote $\pi ' : Y' = Y \times _ T T' \to Y$ the base change of $\pi $ and $e' : Y' \to T'$ the base change of $e$. Picture

\[ \vcenter { \xymatrix{ X' \ar[d]_{f'} & Y \ar[l]^ h \ar[d]^ e & Y' \ar[l]^{\pi '} \ar[d]^{e'} \\ S' & T \ar[l]_ g & T' \ar[l]_\pi } } \quad \text{and}\quad \vcenter { \xymatrix{ X' \ar[d]_{f'} & & Y' \ar[ll]^{h' = h \circ \pi '} \ar[d]^{e'} \\ S' & & T' \ar[ll]_{g' = g \circ \pi } } } \]

Using pullback maps we obtain a canonical commutative diagram

\[ \xymatrix{ (f')^{-1}R^ qg_*\mathcal{F} \ar[r] \ar[d] & (f')^{-1}R^ qg'_*\mathcal{F}' \ar[d] \\ R^ qh_*e^{-1}\mathcal{F} \ar[r] & R^ qh'_*(e')^{-1}\mathcal{F}' } \]

of abelian sheaves on $X'$. Let $P(T')$ be the property

  • The image $\xi '$ of $\xi $ in $(Rh'_*(e')^{-1}\mathcal{F}')_{\overline{x}}$ is not in the image of the map $(f^{-1}R^ qg'_*\mathcal{F}')_{\overline{x}} \to (R^ qh'_*(e')^{-1}\mathcal{F}')_{\overline{x}}$.

We claim that hypotheses (1), (2), and (3) of Lemma 59.88.5 hold for $P$ which proves our lemma.

Condition (1) of Lemma 59.88.5 holds for $P$ because the étale topology of a scheme and a thickening of the scheme is the same. See Proposition 59.45.4.

Suppose that $I$ is a directed set and that $T_ i$ is an inverse system over $I$ of quasi-compact and quasi-separated schemes over $T$ with affine transition morphisms. Set $T' = \mathop{\mathrm{lim}}\nolimits T_ i$. Denote $\mathcal{F}'$ and $\mathcal{F}_ i$ the pullback of $\mathcal{F}$ to $T'$, resp. $T_ i$. Consider the diagrams

\[ \vcenter { \xymatrix{ X \ar[d]_{f'} & Y \ar[l]^ h \ar[d]^ e & Y_ i \ar[l]^{\pi _ i'} \ar[d]^{e_ i} \\ S & T \ar[l]_ g & T_ i \ar[l]_{\pi _ i} } } \quad \text{and}\quad \vcenter { \xymatrix{ X \ar[d]_{f'} & & Y_ i \ar[ll]^{h_ i = h \circ \pi _ i'} \ar[d]^{e_ i} \\ S & & T_ i \ar[ll]_{g_ i = g \circ \pi _ i} } } \]

as in the previous paragraph. It is clear that $\mathcal{F}'$ on $T'$ is the colimit of the pullbacks of $\mathcal{F}_ i$ to $T'$ and that $(e')^{-1}\mathcal{F}'$ is the colimit of the pullbacks of $e_ i^{-1}\mathcal{F}_ i$ to $Y'$. By Lemma 59.51.8 we have

\[ R^ qh'_*(e')^{-1}\mathcal{F}' = \mathop{\mathrm{colim}}\nolimits R^ qh_{i, *}e_ i^{-1}\mathcal{F}_ i \quad \text{and}\quad (f')^{-1}R^ qg'_*\mathcal{F}' = \mathop{\mathrm{colim}}\nolimits (f')^{-1}R^ qg_{i, *}\mathcal{F}_ i \]

It follows that if $P(T_ i)$ is true for all $i$, then $P(T')$ holds. Thus condition (2) of Lemma 59.88.5 holds for $P$.

The most interesting is condition (3) of Lemma 59.88.5. Assume $T'$ is a quasi-compact and quasi-separated scheme over $T$ such that $P(T')$ is true. Let $Z \subset T'$ be a closed subscheme with complement $V \subset T'$ quasi-compact. Consider the diagram

\[ \xymatrix{ Y' \times _{T'} Z \ar[d]_{e_ Z} \ar[r]_{i'} & Y' \ar[d]_{e'} & Y' \times _{T'} V \ar[l]^{j'} \ar[d]^{e_ V} \\ Z \ar[r]^ i & T' & V \ar[l]_ j } \]

Choose an injective map $j^{-1}\mathcal{F}' \to \mathcal{J}$ where $\mathcal{J}$ is an injective sheaf of $\mathbf{Z}/n\mathbf{Z}$-modules on $V$. Looking at stalks we see that the map

\[ \mathcal{F}' \to \mathcal{G} = j_*\mathcal{J} \oplus i_*i^{-1}\mathcal{F}' \]

is injective. Thus $\xi '$ maps to a nonzero element of

\begin{align*} & \mathop{\mathrm{Coker}}\left( ((f')^{-1}R^ qg'_*\mathcal{G})_{\overline{x}} \to (R^ qh'_*(e')^{-1}\mathcal{G})_{\overline{x}} \right) = \\ & \mathop{\mathrm{Coker}}\left( ((f')^{-1}R^ qg'_*j_*\mathcal{J})_{\overline{x}} \to (R^ qh'_*(e')^{-1}j_*\mathcal{J})_{\overline{x}} \right) \oplus \\ & \mathop{\mathrm{Coker}}\left( ((f')^{-1}R^ qg'_*i_*i^{-1}\mathcal{F}')_{\overline{x}} \to (R^ qh'_*(e')^{-1}i_*i^{-1}\mathcal{F}')_{\overline{x}} \right) \end{align*}

by part (2) of Lemma 59.88.2. If $\xi '$ does not map to zero in the second summand, then we use

\[ (f')^{-1}R^ qg'_*i_*i^{-1}\mathcal{F}' = (f')^{-1}R^ q(g' \circ i)_*i^{-1}\mathcal{F}' \]

(because $Ri_* = i_*$ by Proposition 59.55.2) and

\[ R^ qh'_*(e')^{-1}i_*i^{-1}\mathcal{F} = R^ qh'_*i'_*e_ Z^{-1}i^{-1}\mathcal{F} = R^ q(h' \circ i')_*e_ Z^{-1}i^{-1}\mathcal{F}' \]

(first equality by Lemma 59.55.3 and the second because $Ri'_* = i'_*$ by Proposition 59.55.2) to we see that we have $P(Z)$. Finally, suppose $\xi '$ does not map to zero in the first summand. We have

\[ (e')^{-1}j_*\mathcal{J} = j'_*e_ V^{-1}\mathcal{J} \quad \text{and}\quad R^ aj'_*e_ V^{-1}\mathcal{J} = 0, \quad a = 1, \ldots , q - 1 \]

by $BC(f, n, q - 1)$ applied to the diagram

\[ \xymatrix{ X \ar[d]_ f & Y' \ar[l] \ar[d]_{e'} & Y \ar[l]^{j'} \ar[d]^{e_ V} \\ S & T' \ar[l] & V \ar[l]_ j } \]

and the fact that $\mathcal{J}$ is injective. By the relative Leray spectral sequence for $h' \circ j'$ (Cohomology on Sites, Lemma 21.14.7) we deduce that

\[ R^ qh'_*(e')^{-1}j_*\mathcal{J} = R^ qh'_*j'_*e_ V^{-1}\mathcal{J} \longrightarrow R^ q(h' \circ j')_* e_ V^{-1}\mathcal{J} \]

is injective. Thus $\xi $ maps to a nonzero element of $(R^ q(h' \circ j')_* e_ V^{-1}\mathcal{J})_{\overline{x}}$. Applying part (3) of Lemma 59.88.2 to the injection $j^{-1}\mathcal{F}' \to \mathcal{J}$ we conclude that $P(V)$ holds. $\square$

Lemma 59.88.7. With $f : X \to S$ and $n$ as in Remark 59.88.1 assume for some $q \geq 1$ we have that $BC(f, n, q - 1)$ is true, but $BC(f, n, q)$ is not. Then there exist a commutative diagram

\[ \xymatrix{ X \ar[d]_ f & X' \ar[d] \ar[l] & Y \ar[l]^ h \ar[d] \\ S & S' \ar[l] & \mathop{\mathrm{Spec}}(K) \ar[l] } \]

with both squares cartesian, where

  1. $S'$ is affine, integral, and normal with algebraically closed function field,

  2. $K$ is algebraically closed and $\mathop{\mathrm{Spec}}(K) \to S'$ is dominant (in other words $K$ is an extension of the function field of $S'$)

and there exists an integer $d | n$ such that $R^ qh_*(\mathbf{Z}/d\mathbf{Z})$ is nonzero.

Conversely, nonvanishing of $R^ qh_*(\mathbf{Z}/d\mathbf{Z})$ in the lemma implies $BC(f, n, q)$ isn't true as Lemma 59.80.5 shows that $R^ q(\mathop{\mathrm{Spec}}(K) \to S')_*\mathbf{Z}/d\mathbf{Z} = 0$.

Proof. First choose a diagram and $\mathcal{F}$ as in Lemma 59.88.6. We may and do assume $S'$ is affine (this is obvious, but see proof of the lemma in case of doubt). By Lemma 59.88.3 we may assume $K$ is algebraically closed. Then $\mathcal{F}$ corresponds to a $\mathbf{Z}/n\mathbf{Z}$-module. Such a modules is a direct sum of copies of $\mathbf{Z}/d\mathbf{Z}$ for varying $d | n$ hence we may assume $\mathcal{F}$ is constant with value $\mathbf{Z}/d\mathbf{Z}$. By Lemma 59.88.4 we may replace $S'$ by the normalization of $S'$ in $\mathop{\mathrm{Spec}}(K)$ which finishes the proof. $\square$


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 0F01. Beware of the difference between the letter 'O' and the digit '0'.