2
$\begingroup$

Let $Y$ be a smooth algebraic variety and $i: X\hookrightarrow Y$ be a smooth divisor. We consider the derived functors $i^*: D^b_{coh}(Y)\to D^b_{coh}(X)$ and $i_*: D^b_{coh}(X)\to D^b_{coh}(Y)$. By adjunction, for any $\mathcal{F}\in D^b_{coh}(X)$ we have a canonical morphism $i^*i_*\mathcal{F}\to \mathcal{F}$ in $D^b_{coh}(X)$.

My question is: Can we complete $i^*i_*\mathcal{F}\to \mathcal{F}$ to an exact triangle in $D^b_{coh}(X)$? I am particularly interest in the case that $Y$ is Calabi-Yau and I guess that in this case the expected exact triangle should be $$ i^*i_*\mathcal{F}\to \mathcal{F}\to \mathcal{F}\otimes\omega_X^{-1} $$ but I cannot prove or disprove it.

$\endgroup$

1 Answer 1

7
$\begingroup$

The cone of $i^*i_*\mathcal{F} \to \mathcal{F}$ is isomorphic to $\mathcal{F} \otimes \mathcal{O}_X(-X)[2]$.

EDIT. Let me write an argument for a sheaf $F$. Consider the distinguished triangle $$ i^*i_*F \to F \to F'. $$ We need to identify $F'$. Applying $i_*$ we obtain $$ i_*i^*i_*F \to i_*F \to i_*F'.\tag{*} $$ By projection formula and the Koszul resolution of $i_*\mathcal{O}_X$, we have $$ i_*i^*i_*F \cong i_*F \otimes i_*\mathcal{O}_X \cong \mathrm{Cone}(i_*F \otimes \mathcal{O}_Y(-X) \to i_*F), $$ and since the map in the right-hand side is obviously zero, we conclude that $$ i_*i^*i_*F \cong i_*F \oplus i_*F \otimes \mathcal{O}_Y(-X)[1]. $$ Now the first map in $(*)$ is the projection to the direct summand, hence $$ i_*F' \cong i_*F \otimes \mathcal{O}_Y(-X)[2] \cong i_*(F \otimes \mathcal{O}_X(-X)[2]). $$ Since $i_*$ is fully faithful on the category of sheaves, it follows from this that $$ F' \cong F \otimes \mathcal{O}_X(-X)[2]. $$

$\endgroup$
10
  • $\begingroup$ Is it well-known? How to see it? $\endgroup$ Commented Jun 30, 2022 at 4:22
  • $\begingroup$ I think you can find this in the book of Huybrechts (FM transforms ...). $\endgroup$
    – Sasha
    Commented Jun 30, 2022 at 4:26
  • $\begingroup$ I think his Corollary 3.40 and Example 3.41 are closed to the result I want but they are still on different topic. Actually they compute $i^!$ but I am not sure if it implies the result you mentioned. $\endgroup$ Commented Jun 30, 2022 at 4:40
  • $\begingroup$ I am sure you can use for $i^*$ the same argument that is used for $i^!$. Alternatively, you can use the fact that $i^!$ differs from $i^*$ by a line bundle twist and shift, and applying this to the triangle for $i^!i_*$, obtain the triangle for $i^*i_*$. $\endgroup$
    – Sasha
    Commented Jun 30, 2022 at 7:04
  • 1
    $\begingroup$ @ZhaotingWei: I added an argument to the answer. $\endgroup$
    – Sasha
    Commented Jul 1, 2022 at 10:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .