Suppose $X,Y$ are smooth varieties over $\mathbb{C}$, and let $K_i \in D^b(X \times Y), i=1,2$ be objects in the derived category of bounded complex of coherent sheaves on $X \times Y$. Then there are Fourier-Mukai functors (FM functors) $\Phi_{K_i}$ from $D^{b}(X)$ to $D^b(Y)$ associated to kernel $K_i$:

$$F \mapsto q_{*}(K_i\otimes p^*(F)).$$ (here pullback, tensor and pusshfoward are all in the derived sense, and $p: X\times Y \to X$, $q: X \times Y \to Y$ are projections.)

My question is: if there is a natural transformation $G: \Phi_{K_1} \to \Phi_{K_2}$ respect derived categories, then does $G$ necessary come from a morphism of kernel $K_1 \to K_2$?

Somehow, I feel if one can write $\Phi_{K_1} \circ \Phi^{-1}_{K_2}$ (I use $\Phi^{-1}_{K_2}$ to denote some adjunct functor of $\Phi_{K_2}$), then it is also a FM transform, and by the uniqueness of FM kernel, one should get some morphism between the kernels. But I am not quite sure...

  • $\begingroup$ The FM kernel is not always unique for functors which are not fully faithful. See, for example, the article arxiv.org/abs/1009.5577. $\endgroup$
    – naf
    May 4 '14 at 7:54
  • 2
    $\begingroup$ The functor $\Phi:D^b(X\times Y) \to Fun(D^b(X),D^b(Y))$ is known to be neither faithful nor full. $\endgroup$
    – Sasha
    May 4 '14 at 8:36
  • 5
    $\begingroup$ However, the corresponding statement at the level of dg-categories is true: $\mathbf{D}(X \times Y) = \underline{\mathbf{Hom}}(\mathbf{D}(X), \mathbf{D}(Y))$, so that morphisms of integral kernels correspond to morphisms of the induced dg-functors (note though that the Hom means the internal Hom in the localization of the category of dg-categories with respect to the quasi-equivalences). $\endgroup$
    – AAK
    May 4 '14 at 8:45

This is somewhat sad, but I think (part of) what we've learned from the whole triangulated-vs-dg story is the following pseudo-statement: the bare category of functors Fun(D(X),D(Y)) is the wrong thing to take -- it should be replaced the category D(X x Y).

There is a nice example, due to the usual Bondal-Orlov-Bridgeland people (Example 6.5 in Caldararu's notes), which illustrates this perfectly.

Thus, if you find something weird happening in Fun(D(X),D(Y)) you shouldn't be discouraged -- it's not your fault.

Of course, there are big names in the field (Canonaco and Stellari to name two) who might disagree with my initial pseudo-statement. It really depends on what you care about.

  • $\begingroup$ Don't you rather want to say something like the (triangulated) functor category Fun(D(X), D(Y)) should be replaced by the category of dg-functors? $\endgroup$
    – AAK
    May 25 '14 at 8:43
  • 1
    $\begingroup$ @Adeel: well, looking at your posts it's obvious that you are well versed in the dark $\infty$-arts. People like me tend to use a differet point of view. I am aware that $D(X)$ is actually the shadow of a stable $\infty$-category and that problems like descent and uniqueness of cones are solved if you take into account this higher structure. However, when I work on a problem I try to be as 1-categorical as I can. Since Toen has proved that $dg-Fun(D(X),D(Y))=D(X \times Y)$, I might as well think of everything as a kernel. That way I can bypass dg-land entirely. (but you are quite right) $\endgroup$ May 25 '14 at 12:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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