4
$\begingroup$

Let $f:X\to Y$ be a morphism between smooth algebraic varieties over $\mathbb{C}$.

We have natural functors $f^!:\mathsf{D}_{\text{qc}}(\mathcal{D}_Y)\to \mathsf{D}_{\text{qc}}(\mathcal{D}_X)$, $f_*:\mathsf{D}_{\text{qc}}(\mathcal{D}_X)\to \mathsf{D}_{\text{qc}}(\mathcal{D}_Y)$, and $\mathbb{D}_X:\mathsf{D}_{\text{qc}}(\mathcal{D}_X)\to \mathsf{D}_{\text{qc}}(\mathcal{D}_X)^\text{op}$ between the (unbounded, but feel free to think about the bounded case if you prefer) derived categories of left D-modules (i.e. the objects are complexes of D-modules with quasi-coherent cohomology). Naturally, we can form the diagrams below.

(The vertical arrows are simply the functors which forgets the D-module structure, and the arrows below are the functors in Grothendieck duality.)

My main question is: do these diagrams commute? If the third diagram does not commute, does it commute when restricted to $\mathsf{D}^b_{\text{coh}}$? Also, what happens for right D-modules?

$\endgroup$
3
  • $\begingroup$ The diagram for $f^!$ commutes if one works in the category of right $D$-modules (it can't commute for left $D$-modules because a $f^! \mathcal{O}_Y$ is a shift of $\mathcal{O}_X$ essentially by definition). The diagram for $f_*$ basically never commutes because $f_*$ is neither left nor right $t$-exact on the $D$-module side. You can make a commutative diagram for $f_*$ if you consider the induced functor $\mathrm{D}(\mathcal{O_X}) \to \mathrm{D}(\mathcal{D}_X)$ given by tensor product. Then the diagram would commute for right $D$-modules again. Probably the same happens for ... $\endgroup$
    – gdb
    Dec 24, 2021 at 1:56
  • $\begingroup$ ... duality functor if one restricts to coherent sheaves. $\endgroup$
    – gdb
    Dec 24, 2021 at 1:56
  • $\begingroup$ @gdb, would you mind explaining why the diagram for $f^!$ commutes if we work with right D-modules? I also didn't quite understand how you made the diagram for $f_*$ commutative. $\endgroup$
    – Gabriel
    Dec 24, 2021 at 12:52

1 Answer 1

3
+200
$\begingroup$

[All functors in this answer are assumed to be derived]

These commutativities basically boil down to unraveling all definitions. The actual computations are quite annoying (but totally possible), so I will only explain the main points.

  1. The diagram commutes for $f^*$ and left $\mathcal{D}$-modules. This basically follows from the definition (do not confuse it with the other $f^*_{\mathcal{D}}$ functor that is defined only on holonomic $\mathcal{D}$-modules).

  2. The functor $f^!$ on left $\cal{D}$-modules is defined as $f^*[\mathrm{dim}_X -\mathrm{dim}_Y]$ for a morphism of smooth varieties $f\colon X \to Y$. In particular $f^!(\mathcal{O}_Y) \simeq \mathcal{O}_X[\mathrm{dim}_X - \mathrm{dim}_Y]$, so there is no chance that the first diagram in your post commutes.

  3. Before we discuss what happens for right $\mathcal{D}$-modules, we need to recall the so-called "left-right switch". This is a canonical equivalence between left and right $\mathcal{D}$-modules on any smooth variety $X$. Namely, it is given by the tensor product functor $-\otimes_{\mathcal{O}_X} \omega_X\colon \mathrm{D}^b_{qc}(\mathrm{LMod}_{\mathcal{D}_X}) \to \mathrm{D}^b_{qc}(\mathrm{RMod}_{\mathcal{D}_X})$. One needs to actually make sense of this, roughly the point is that the canonical bundle $\omega_X$ has a canonical structure of a right $\mathcal{D}$-module via Lie derivatives. Then a tensor product over the structure sheaf of a left $\mathcal{D}$-module and a right $\mathcal{D}$-module has a canonical structure of a right $\mathcal{D}$-module.

  4. Now we are ready to discuss what happens for the $f^!$ functor on the level of right $\mathcal{D}$-modules. In order to avoid any confusion, let us denote it by $f^{\dagger}$. It is defined via the left-right switch. Namely, $f^\dagger(M):= f^!(M\otimes_{\mathcal{O}_Y} \omega^{\vee}_Y)\otimes_{\mathcal{O}_X} \omega_X \colon \mathrm{D}^b_{qc}(\mathrm{RMod}_{\mathcal{D}_Y}) \to \mathrm{D}^b_{qc}(\mathrm{RMod}_{\mathcal{D}_Y})$ for a morphism $f\colon X \to Y$ of smooth varieties. Now we use steps $1$ and $2$ to compute this functor (on the level of underlying $\mathcal{O}$-modules) $$ f^\dagger(M) = f^!(M\otimes_{\mathcal{O}_Y}\omega^{\vee}_Y) \otimes_{\mathcal{O}_X} \omega_X= f^*(M) \otimes_{\mathcal{O}_X} f^*(\omega^\vee_Y)[\mathrm{dim}_X-\mathrm{dim}_Y]\otimes_{\mathcal{O}_X} \omega_X $$ $$ =f^*(M)\otimes_{\mathcal{O}_X} f^*(\omega^{\bullet, \vee}_Y) \otimes_{\mathcal{O}_X} \omega_X^\bullet $$ where $\omega_X^\bullet$ is the dualizing complex on $X$, so $\omega_X^\bullet = \omega_X[\mathrm{dim}_X]$, and the same for $Y$. Now any finite type morphism of regular scheme is of finite Tor-amplitude, so Tag0B6U gives us that $$ f^!(M)=f^*(M)\otimes_{\mathcal{O}_X} f^!(\mathcal{O}_Y) $$ and, essentially by definition, $$ \omega^\bullet_X = f^!(\omega^\bullet_Y) = f^*(\omega^\bullet_Y) \otimes_{\mathcal{O}_X} f^!(\mathcal{O}_Y). $$ Combining these two equalities, we get the desired formula $$ f^!(M) = f^*(M)\otimes_{\mathcal{O}_X} \omega^\bullet_X \otimes_{\mathcal{O}_X} f^*(\omega^{\bullet, \vee}_Y) = f^\dagger(M) $$

  5. The diagram for $f_*$ does not commute neither for left nor for right $\mathcal{D}$-modules because it is neither left nor right $t$-exact on the $\mathcal{D}$-module side (consider separately cases of a closed immersion and a smooth morphism) but it is left exact on the $\mathcal{O}$-module side. However, one can ask for commutativity of another diagram that relates $\mathcal{O}$-modules and $\mathcal{D}$-modules. Namely, given any object $K\in \mathrm{D}_{qc}^b(\mathcal{O}_Y)$, one can define an induced $\mathcal{D}$-module $K\otimes_{\mathcal{O}_Y}\mathcal{D}_Y \in \mathrm{D}^b_{qc}(\mathcal{D}_Y)$. Then the claim is that for the right $\mathcal{D}$-modules, there is a functorial isomorphism $$ f_*(K\otimes_{\mathcal{O}_X} \mathcal{D}_X) \simeq f_*(K) \otimes_{\mathcal{O}_Y} \mathcal{D}_Y $$ for a morphism $f\colon X \to Y$ of smooth varieties. The proof essentially boils down to the projection formula (and a bit tedious computations since you want to identify those objects as $\mathcal{D}$-modules and not merely $\mathcal{O}$-modules) once all definitions are given (but it is quite tricky to define $f_*$ on the $\mathcal{D}$-module side). A version of this result is Lemma 4.26 in Kashiwara's book "$\mathcal{D}$-modules and microlocal analysis" (actually formulated for left $\mathcal{D}$-modules).

  6. Commutativity of your last diagram depends on your definition of duality on both sides, I think.

$\endgroup$
1
  • $\begingroup$ Dear @gdb, I appreciate immensely your Christmas present. As for you last point, the duality functor on left D-modules is $\mathsf{R}\underline{\text{Hom}}_{\mathcal{D}_X}(-,\mathcal{D}_X)\otimes_{\mathcal{O}_X}\omega_X^\vee [\operatorname{dim}X]$, while the duality functor on O-modules is $\mathsf{R}\underline{\text{Hom}}_{\mathcal{O}_X}(-,\omega_X^\bullet)$, where I use your notation for the dualizing complex. $\endgroup$
    – Gabriel
    Dec 25, 2021 at 11:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.