Suppose $X$ is a smooth scheme, $E=O_X^{\oplus n}$ and $\varphi\in SL_n(E)$, i.e. $\varphi$ has trivial determinant and is an isomorphism. Is the morphism $$\mathbb{P}(\varphi)^*:CH^{\bullet}(\mathbb{P}(E))\longrightarrow CH^{\bullet}(\mathbb{P}(E))$$ equal to the identity map?
-
1$\begingroup$ Every connected, linear algebraic group is rational (cf. the Bruhat decomposition). Thus the action on the Chow group is trivial. $\endgroup$– Jason StarrCommented Nov 8, 2020 at 9:57
-
1$\begingroup$ I should add one qualification: “rational” should be “geometrically rational”. Over any base, the special linear group is rational. $\endgroup$– Jason StarrCommented Nov 8, 2020 at 10:06
-
1$\begingroup$ @JasonStarr I see what you said. Is there a similar result for general vector bundles? $\endgroup$– Nanjun YangCommented Dec 6, 2020 at 12:32
1 Answer
This is true for any smooth variety $X$ (i.e. separated scheme of finite type over a field) and any vector bundle $E$ if and only if $\mathbb{P}(\varphi)^*(\mathcal{O}_{\mathbb{P}(E)}(1))=\mathcal{O}_{\mathbb{P}(E)}(1)$.
For one direction, note that if this equality doesn't hold, then the automorphism induces a morphism that's different on the Picard group which is a subgroup of the Chow ring. Below, I'll show the converse.
Claim 1: Let $X$ be any variety with irreducible components $X_i$. Let $\varphi: E\rightarrow E$ be any $X$-automorphism of a vector bundle $E$ over $X$. Let $\varphi_i$ be the restriction of $\varphi$ to $E|_{X_i}$. If $\mathbb{P}(\varphi_i)^*$ is the identity for all $i$, then $\mathbb{P}(\varphi)^*$ is the identity.
Proof. Let $\rho_i:\mathbb{P}(E|_{X_i})\rightarrow \mathbb{P}(E)$ be the inclusion. Then $\rho_{i*}\circ \mathbb{P}(\varphi_i)^*= \mathbb{P}(\varphi^*)\circ \rho_{i*}$ by Proposition 1.7 of Fulton. The result follows since the images of $\rho_{i*}$ generate $\mathrm{CH}(\mathbb{P}(E))$. $\square$
Claim 2: Let $X$ be any variety and let $\varphi:E\rightarrow E$ be any $X$-automorphism of a vector bundle $E$ over $X$. If $\mathbb{P}(\varphi)^*(\mathcal{O}_{\mathbb{P}(E)}(1))=\mathcal{O}_{\mathbb{P}(E)}(1)$, then $\mathbb{P}(\varphi)^*$ is the identity.
Proof. We'll proceed by Noetherian induction on the dimension of the irreducible components of $X$. For this purpose, let $X$ be any variety with irreducible components $X_i$ having dimension $\max_i\{ \mathrm{dim}(X_i)\}\leq n$.
For the base case, we observe that if $X$ is a point, then any automorphism of $\mathbb{P}^m_X$ has to take $\mathcal{O}(1)$ to itself in $\mathrm{Pic}(X)$ and this generates the ring $\mathrm{CH}^*(\mathbb{P}^m_X)$.
Assume that the result is true for all $X$ with irreducible components $X_i$ with $\max_i\{\mathrm{dim}(X_i)\}\leq n-1$. By Claim 1, it suffices to show that this implies the result when $X$ is irreducible. Then for any open $U\subset X$ there is a commuting ladder with exact rows
$$\begin{matrix} \mathrm{CH}(\mathbb{P}(E|_Z))&\rightarrow& \mathrm{CH}(\mathbb{P}(E))&\rightarrow& \mathrm{CH}(\mathbb{P}(E|_U))&\rightarrow & 0 \\ \downarrow && \downarrow && \downarrow\\ \mathrm{CH}(\mathbb{P}(E|_Z))&\rightarrow& \mathrm{CH}(\mathbb{P}(E))&\rightarrow& \mathrm{CH}(\mathbb{P}(E|_U))&\rightarrow & 0\end{matrix}$$ where the vertical morphisms are induced by $\mathbb{P}(\varphi)^*$. By our induction hypothesis, the left-most arrow is the identity whenever $U$ is nonempty. Taking a limit over the inductive system of closed subschemes $Z\subset X$, we get a commutative ladder with exact rows given by localization $$\begin{matrix} \varinjlim_Z \mathrm{CH}(\mathbb{P}(E|_Z))&\rightarrow& \mathrm{CH}(\mathbb{P}(E))&\rightarrow& \mathrm{CH}(\mathbb{P}(E|_{\eta}))&\rightarrow & 0 \\ \downarrow && \downarrow && \downarrow\\ \varinjlim_Z \mathrm{CH}(\mathbb{P}(E|_Z))&\rightarrow& \mathrm{CH}(\mathbb{P}(E))&\rightarrow& \mathrm{CH}(\mathbb{P}(E|_{\eta}))&\rightarrow & 0\end{matrix}$$ where $\eta$ is the generic point of $X$. By our induction, the rightmost vertical arrow is the identity. These sequences are (canonically) split by Claim 3 below. A diagram chase finishes the proof. $\square$
Claim 3: Let $L=\mathcal{O}_{\mathbb{P}(E)}(1)$. If $\mathbb{P}(\varphi)^*(L)=L$, then the map $p^*:\mathrm{CH}_i(\mathbb{P}(E))\rightarrow \mathrm{CH}_{i-\mathrm{dim}(X)}(\mathbb{P}(E|_\eta))$ is right-split by the map $s:\mathrm{CH}_{i-\mathrm{dim}(X)}(\mathbb{P}(E|_\eta))\rightarrow \mathrm{CH}_i(\mathbb{P}(E))$ sending $c_1(L|_\eta)^j$ to $c_1(L)^j$. Moreover, there is a commutative square $s\circ\mathbb{P}(\varphi|_\eta)^*=\mathbb{P}(\varphi)^*\circ s$.
Proof. Since the group $\mathrm{CH}_i(\mathbb{P}(E|_\eta))=\mathbb{Z}$, for the first claim it suffices to see that $p^*c_1(L)=c_1(L|_\eta)$ which is true by functorality of the pullbacks. To see the second claim, we need to check $$s\circ\mathbb{P}(\varphi|_\eta)^*(c_1(L|_\eta)^j)=\mathbb{P}(\varphi)^*\circ s(c_1(L|_\eta)^j).$$ The left hand side is equivalent to $$s\circ\mathbb{P}(\varphi|_\eta)^*(c_1(L|_\eta)^j)=s(c_1(\mathbb{P}(\varphi)^*(L|_\eta))^j)=s(c_1(L|_\eta)^j)=c_1(L)^j$$ while the right hand side is equivalent to $$\mathbb{P}(\varphi)^*\circ s(c_1(L|_\eta)^j)=\mathbb{P}(\varphi)^*(c_1(L)^j))=c_1(\mathbb{P}(\varphi)^*(L))^j$$ by functorality. It suffices then to observe that $\mathbb{P}(\varphi)^*(L)=L$. $\square$
Let $\pi_X:\mathbb{P}(E)\rightarrow X$ be the structure map and let $\mathbb{P}(\varphi):\mathbb{P}(E)\rightarrow \mathbb{P}(E)$ be an $X$-isomorphism. If $X$ is smooth, then by the projective bundle formula (and the relation between the divisor class group and the Picard group) we have $$\mathrm{Pic}(\mathbb{P}(E))=\pi_X^*\mathrm{Pic}(X)\oplus \mathbb{Z}$$ where $L$ generates the copy of $\mathbb{Z}$. The projection formula then yields \begin{align*}\pi_{X*}\mathbb{P}(\varphi)^*(L) &=\pi_{X*}(L\otimes \pi_X^*(M))\\ &= \pi_{X*}(L)\otimes M\\ &=E\otimes M\end{align*} with $M$ a line bundle on $X$.
Claim 4. If $E=\mathcal{O}_X^{\oplus n+1}$, then $\mathbb{P}(\varphi)^*$ is the identity.
Proof. The composition of a section $\rho: X\rightarrow X\times \mathbb{P}^n$ with $\varphi$ and the projection $\psi:X\times \mathbb{P}^n\rightarrow \mathbb{P}^n$ shows that $$ (\rho\circ\varphi\circ\psi)^*(\mathcal{O}(1))=M.$$ But we also have that the composition $X\rightarrow \mathbb{P}^n$ factors through a point in $\mathbb{P}^n$ so $M=\mathcal{O}_X$.$\square$
-
$\begingroup$ How did you perform the last step to prove the middle arrow is the identity? $\endgroup$ Commented Dec 6, 2020 at 6:48
-
$\begingroup$ Thanks. I think if $E=O_X^{\oplus n+1}$ is a trivial bundle, the $O(1)$ will be preserved. This is because the bundle $M$ you mentioned is the pullback of $O(1)$ along the composite $X\to X\times\mathbb{P}^n\to X\times\mathbb{P}^n\to\mathbb{P}^n$, which factors through $GL_{n+1}$. Then use $CH^1(GL_{n+1})=0$ to conclude that $M=O_X$. $\endgroup$ Commented Dec 7, 2020 at 0:46
-
-
$\begingroup$ Maybe the invariance of $O(1)$ is trivial. It's something like given a sheaf $R$ of graded rings and a graded automorphism $f$ of $R$, prove that the two $R$-module structures of $R[1]$ are isomorphic. But the isomorphism is just given by $f[1]$! The same argument maybe false if we replace $R[1]$ by a graded module! $\endgroup$ Commented Dec 7, 2020 at 4:04
-
1$\begingroup$ @abx ¸„.-•~¹°”ˆ˜¨ ~~!@@~!~✿✾~SỮ𝐫𝐫ᗴ𝐀Ĺ𝐢ⓈŦᶤ匚~~✴✽~!!~~^~ ¨˜ˆ”°¹~•-.„¸ $\endgroup$– EoinCommented Dec 8, 2020 at 19:12