This question arise from the comparision of the reconstruction theorems of Bondal-Orlov and Balmer and is inspired by Shizhuo Zhang's mathoverflow question: How to unify various reconstruction theorems (Gabriel-Rosenberg, Tannaka,Balmers)

On the one hand, A. Bondal and D. Orlov in http://arxiv.org/abs/alg-geom/9712029 proved their celebrated reconstruction theorem: Let $X$ be a smooth projective variety such that the canonical bundle $\omega_X$ is either ample or anti-ample, and let $Y$ be any projective variety. If $D^b_{\text{coh}}(X)\cong D^b_{\text{coh}}(Y)$ as triangulated categories, then $X\cong Y$.

In fact, they reconstruct $X$ from $D^b_{\text{coh}}(X)$ and their construction heavily use the $\textit{Serre functor}$ S. In general, for a $k$-linear category $\mathcal{C}$, an equivalence $S: \mathcal{C}\rightarrow \mathcal{C}$ is called a Serre functor for $\mathcal{C}$ if their is a natural, bifunctorial isomorphisms $$ \varphi_{A,B}: \text{Hom}_{\mathcal{C}}(A,B)\xrightarrow{\sim} \text{Hom}_{\mathcal{C}}(B,SA)^{\vee} $$ for every $A,B$. Here $(\bullet)^\vee$ denotes the $k$-dual. When $\mathcal{C}=D^b_{\text{coh}}(X)$, the Serre functor $S_X$ is given by $$ S_X\mathcal{E}=\mathcal{E}\otimes \omega_X[\dim X] $$ i.e tensor product with the canonical bundle and shift by $\dim X$ in $D^b_{\text{coh}}(X)$. We can refer to Section 4 of A. Caldararu's lecture notes http://arxiv.org/abs/math/0501094 for an excellent introduction. We notice that Bondal-Orlov reconstruction does not involve the tensor structure of $D^b_{\text{coh}}(X)$ besides the definition of the Serre functor.

(As pointed out by Piotr and Qiaochu in the comments, the Serre functor is unique, hence is part of the data of $D^b_{\text{coh}}(X)$ and is not an extra data.)

On the other hand, P. Balmer in http://arxiv.org/abs/math/0111049 proved another reconstruction theorem: Let $X$ be a noetherian scheme, then we can reconstruct $X$ from the $\textit{tensor triangulated category}$ $(D^{\text{perf}}(X),\otimes^L_{\mathcal{O}_X})$. This reconstruction uses the extra structure of $D^{\text{perf}}(X)$: the tensor product, and it applies for more general $X$ than Bondal-Orlov.

Now we can look at the case when the two theorems overlap: If $X$ is smooth projective with $\omega_X$ ample or anti-ample, then $D^{\text{perf}}(X)\cong D^b_{\text{coh}}(X)$. Now from $D^b_{\text{coh}}(X)$ either using the Serre functor or using the tensor structure. It seems that there are some redundancy here. In fact we have $$ D^b_{\text{coh}}(X) \xrightarrow [\text{reconstruction}]{\text{Bondal-Orlov}}X\rightarrow (D^b_{\text{coh}}(X),\otimes^L_{\mathcal{O}_X})\xrightarrow [\text{reconstruction}]{\text{Balmer}} X\rightarrow D^b_{\text{coh}}(X) \ldots $$ Hence one may expect a direct construction of the tensor structure $(D^b_{\text{coh}}(X),\otimes^L_{\mathcal{O}_X})$ just from $D^b_{\text{coh}}(X)$ itself, considered as a triangulated category.

$\textbf{My question}$ is: If $X$ is smooth projective variety with $\omega_X$ ample or anti-ample, could we define the tensor product structure on $D^b_{\text{coh}}(X)$ just from the triangulated cateogry structure on $D^b_{\text{coh}}(X)$?

In the other direction we have $\textbf{a related question}$: What is the role of the Serre functor $S_X$ and the canonical bundle $\omega_X$ in the the tensor triangulated category $(D^b_{\text{coh}}(X),\otimes^L_{\mathcal{O}_X})$?

  • 1
    $\begingroup$ Isn't the Serre functor unique if it exists? $\endgroup$ Oct 1, 2013 at 3:20
  • $\begingroup$ @PiotrAchinger Yes, the Serre functor, if exists, is unique up to a unique isomorphism of functors. But what is the relation between the uniqueness of Serre functor and my questions? $\endgroup$ Oct 1, 2013 at 3:36
  • 1
    $\begingroup$ @Zhaoting: Piotr's point, I think, is that the Serre functor isn't extra data, it's an extra property (namely the property that the derived category admits a Serre functor). So the "together with the Serre functor" part of your question is unnecessary. $\endgroup$ Oct 1, 2013 at 5:27
  • $\begingroup$ @QiaochuYuan En, you're right. Now the question remains: Could we define the tensor product just from the derived category if $X$ has a ample or anti-ample canonical sheaf? Maybe I need to edit the question. $\endgroup$ Oct 1, 2013 at 6:00
  • $\begingroup$ @PiotrAchinger I have made certain changes according to you and Qiaochu. Thank you guys for pointing out the problem! $\endgroup$ Oct 1, 2013 at 6:14

1 Answer 1


It seems like the answer to your question is no, at least without further clarification. If you could define the tensor product structure on $D^b(X)$ just from the triangulated structure and the Serre functor, then any $k$-linear derived autoequivalence $F:D^b(X)\rightarrow D^b(X)$, since it commutes with the Serre functor, would be a $\otimes$-autoequivalence as well. Now, consider $\mathbb{P}^1_k$ and the autoequivalences $F(n,k)(\mathscr{F})=\mathscr{F}(n)[k]$. These are not $\otimes$-autoequivalences unless $n=k=0$. More or less, they induce new $\otimes$-structures on $D^b(X)$ where the unit is $\mathscr{O}_{\mathbb{P}^1_k}(-n)[-k]$. So, I don't see how to make this work.

That being said, it's possible that you can reconstruct all of these $\otimes$-structures at once. Here is what I have in mind. Pick any invertible object $U$ in $D^b(X)$. Recall that the invertible objects are those complexes such that for every point-like object $P$ there exists $n_p\in\mathbb{Z}$ such that $Hom(U,P[i])=k(P)$ when $i=n_P$ and $0$ otherwise. Bondal and Orlov proved that the point-like objects are just the shifted sheaves $k(P)[n]$ for $P\in X$ (here on out I'll work over an algebraically closed field for simplicity). Then, they proved that the invertible objects are precisely the shifted line bundles $\mathscr{L}[n]$. So, our $U$ is equivalent to $\mathscr{L}[n]$ for some $\mathscr{L}$ and $n$. This can be ignored for time being; I'll come back to it later.

I claim that there is a tensor product $\otimes_U$ on $D^b(X)$ for which $U$ is the unit. Given two complexes $\mathscr{F}$ and $\mathscr{G}$, resolve $\mathscr{F}$ by terms of the form $S_X^n(U)$. The fact that we can do this is because the $S_X^n(U)$ form a spanning class of $D^b(X)$, which follows easily from the fact that the powers of $\omega_X$ form a spanning class, and the fact that $U=\mathscr{L}[n]$. Then, define $S_X^n(U)\otimes_U\mathscr{G}$ as $S_X^n(G)$. Then use these to define $\mathscr{F}\otimes_U\mathscr{G}$. Note that we need ampleness or anti-ampleness in order to perform this construction. It looks like the details work out: uniqueness follows from $S^n_X(U)\otimes_UU=S^n_X(U)$.

Now, recall that Bondal and Orlov also proved that the group of derived $k$-linear autoequivalences of $D^b(X)$ fits into an exact sequence $$0\rightarrow\mathbb{Z}\times Pic(X)\rightarrow Aut(D^b(X))\rightarrow Aut(X)\rightarrow 1.$$ We see that by fixing the invertible object $U$, i.e. the shifted line bundle $\mathscr{L}[n]$ we are picking exactly an element in $\mathbb{Z}\times Pic(X)$.

  • $\begingroup$ Thank you very much for your answer! I think the tensor product you constructed using $U$ is a kind of "twisted tensor product", which is analogous to the twisted D-module in D-module theory, isn't it? $\endgroup$ Oct 1, 2013 at 15:35

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.