Patrick Lei

All notes

A note about derived categories

Talks by Rachel Webb and Tudor Pădurariu, Cornell, Apr 2022.

Abstract.

These are my notes from the introductory talks at the Derived categories and moduli spaces conference at Cornell University, given by Rachel Webb and Tudor Pǎdurariu.

1 Reconstruction of some schemes from their derived categories (Rachel Webb)

Let X be a scheme and let D⁡(X) be the bounded derived category of coherent sheaves.

Question 1.1.

Let X and Y be schemes such that D⁡(X)≃D⁡(Y). When do we have X≃Y?

Theorem 1.2 (Bondal-Orlov).

If X is smooth projective and ωX or ωX−1 is ample, then D⁡(X) classifies X up to isomorphism.

Roughly speaking, Serre duality is an invariant of D⁡(X), so if D⁡(X)≃D⁡(Y), then

so if ωX and ωY are ample, we can take Proj to obtain X≃Y.

1.1 Serre duality.

Recall that if X is a smooth projective variety, then ωX=det⁡(TX∗). If ℰ is locally free, then

Moreover, if ℰ,ℱ∈D⁡(X), we have

When ℰ,ℱ are locally free, we see that

Definition 1.3 (Bondal-Kapranov).

Let C be a k-linear category. A Serre functor is an equivalence S:C→C together with an isomorphism

Example 1.4.

Let C=D⁡(X). Then we define S⁢A=A⊗ωX⁢[n].

Theorem 1.5 (Bondal-Kapranov).

Let C be as above. A Serre functor on C is unique, exact, and if F:C≃D is an equivalence and SC is a Serre functor on C, then D has a Serre functor SD and F∘SC=SD∘F.

Proof.

Clearly Serre functors are unique (Homs into S⁢A are uniquely described). If F:C≃C is an autoequivalence, we note that

It is not true in general that F:D⁡(X)≃D⁡(Y) implies that F⁡(ωX)=ωY. More closely, we actually have F⁡(𝒪X⊗ωX⁢[n])=F⁡(𝒪X)⊗ωY⁢[n].

1.2 Reconstruction.

This subsection is dedicated to the proof of the following theorem.

Theorem 1.6 (Bondal-Orlov).

Let X be a smooth projective variety such that ωX is ample (or anti-ample). Let Y be any smooth quasiprojective variety. Then if D⁡(X)≃D⁡(Y), X≃Y.

There is a closely related result that is of interest.

Theorem 1.7 (Bondal-Orlov).

Let X be as above. Then up to natural transformation, every autoequivalence of D⁡(X) is a composition of shifts, twists, and automorphisms of X.

Example 1.8 (Mukai).

Let E be an elliptic curve. Then there exists Φ:D⁡(E)→∼D⁡(E) such that Φ∘Φ=ι⁡[−1].

We will need two key lemmas, describing which objects in the derived category are points and which objects are line bundles.

Definition 1.9.

A object P∈D⁡(X) is a point object of codimension s if

  1. (i)

    SD⁢(P)≃P⁡[s], were SD is the Serre functor on D⁡(X);

  2. (ii)

    Hom⁡(P,P⁡[i])=0 for i<0;

  3. (iii)

    Hom⁡(P,P)=k⁡(P) is some field.

Definition 1.10.

Let X be a smooth projective variety of dimension n and suppose one of ωX±1 is ample. Then point objects in D⁡(X) are 𝒪x⁢[i].

Definition 1.11.

An object L∈D⁡(X) is invertible if for all point objects P∈D⁡(X), there exists s∈ℤ such that

  1. (1)

    Hom⁡(L,P⁡[s])=k⁡(P);

  2. (2)

    Hom⁡(L,P⁡[i])=0 for all i≠s.

Lemma 1.12.

Let X be a smooth variety such that all point objects are 𝒪x⁢[i]. Then invertible objects are shifts of line bundles ℒ⁡[i].

Sketch of Theorem 1.7.

Let F:D⁡(X)→∼D⁡(X) be an autoequivalence. Then F⁡(𝒪) is invertible, so F⁡(𝒪)=ℒ⁡[i] for some line bundle ℒ. Now if we replace F by F⁡(−)⊗ℒ−1⁢[i], we can assume F⁡(𝒪)=𝒪. This implies that F⁡(ωX)=ωX. This gives an automorphism

Taking the proj, we obtain f∈Aut⁡(X). Finally, we can assume that F, fixes 𝒪, ωX and induces the identity on X. With some work, we can finally show that F is the identity functor. □

Sketchier summary of Theorem 1.6.

The proof proceeds in several steps.

  1. (1)

    The first step is to show that point objects in D⁡(Y) are precisely structure sheaves 𝒪y⁢[i]. This is because on X, if P,Q are point objects, either P=Q⁡[i] or Hom⁡(P,Q⁡[i])=0 for all i. But then if P∈D⁡(Y) is not 𝒪Y⁢[i], then for all y∈Y,i∈ℤ, Hom⁡(P,𝒪Y⁢[i])=0, so P=0.

  2. (2)

    The second step is to note that invertible objects on Y are shifts of line bundles ℒ⁡[i].

  3. (3)

    We now want to define a morphism f:|X|→|Y| of topological spaces. Modifying F a little bit, we can enforce F⁡(𝒪x⁢[0])→𝒪Y⁢[0] by choosing a line bundle ℒX and F⁢ℒX≕ℒY.

  4. (4)

    Next, f is a homeomorphism. If ℳ,𝒩 are line bundles and we have 𝒩→ℳ, then

    is a closed set, and the complements of such sets form a basis for the topology.

  5. (5)

    Now we know that Y is a smooth projective variety of the same dimension as X, so using a topological argument, we see that ωY is ample. This gives us an isomorphism of pluricanonical rings, so X≃Y. □

2 The Fourier-Mukai transform (Tudor Pǎdurariu)

Following from the Bondal-Orlov reconstruction theorem, we now consider examples of nonisomorphic varieties X,Y such that Db⁢(X)≃Db⁢(Y). These are related by Fourier-Mukai transforms, and our strategy will be to start with X with some properties and construct Y as a moduli space of objects on X.

Let X,Y be smooth projective varieties and consider ℰ∈Db⁢(X×Y). We of course have the following diagram:

Then we will define the Fourier-Mukai transform with kernel ℰ

Examples 2.1.
  1. (1)

    The identity functor id:Db⁢(X)→∼Db⁢(X) is given by the Fourier-Mukai transform with kernel ℰ=𝒪Δ.

  2. (2)

    The shift [1]:Db⁢(X)→Db⁢(X) is given by the Fourier-Mukai transform with kernel 𝒪Δ⁢[1].

  3. (3)

    Let f:X→Y be a morphism and let Z⊂X×Y be the graph of f. Then f∗:Db⁢(X)→Db⁢(Y) is given by the Fourier-Mukai transform with kernel 𝒪Z and f∗ is given by the Fourier-Mukai transform in the opposite direction with the same kernel.

Fourier-Mukai transforms are interesting because of the following theorem.

Theorem 2.2 (Orlov).

Let X,Y be smooth and projective over ℂ and Φ:Db⁢(X)→∼Db⁢(Y) be a derived equivalence. Then there exists ℰ∈Db⁢(X×Y) such that Φ=ΦX→Yℰ.

Here are some properties of Fourier-Mukai transforms:

  1. (1)

    We can compose Fourier-Mukai transforms. If ℰ∈Db⁢(X×Y),ℱ∈Db⁢(Y×Z), we would like 𝒢∈Db⁢(X×Z) such that

    Consider the diagram

    Then we define

  2. (2)

    If ℰ∈Db⁢(X×Y), then ΦX→Yℰ has left adjoint ΦY→Xℰ∨⊗ωY⁢[dimY] and right adjoint ΦY→Xℰ∨⊗ωX⁢[dimX].

Now we will construct examples of derived equivalent varieties that are nonisomorphic.The first example is due to Mukai. Let A be an abelian variety. The dual abelian variety A∨ parameterizes degree 0 line bundles on A, so we have a universal line bundle 𝒫 on A×A∨ such that 𝒫|A×p=ℒp is the line bundle corresponding to p∈A∨.

Theorem 2.3 (Mukai).

The Fourier-Mukai transform ΦA→A∨𝒫:Db⁢(A)→Db⁢(A∨) is an equivalence.

The second example is also due to Mukai. Let S be a smooth projective surface with ωX≃𝒪X. Let M be a moduli space of stable sheaves on S (with no strictly semistable sheaves). Then of course there exists a universal sheaf 𝒰 on S×M, where a point p∈M corresponds to a stable sheaf 𝒰p on S. Then we have 𝒰|S×p=𝒰p.

Theorem 2.4 (Mukai).

If dimM=2, which says that if M parameterizes sheaves with Mukai vector (r,β,d)∈H0⁢(S,ℤ)⊕H2⁢(S,ℤ)⊕H4⁢(S,ℤ), then β2−2⁢r⁢d=0, then ΦM→S𝒰:Db⁢(M)→Db⁢(S) is a derived equivalence.

A useful lemma is the following:

Lemma 2.5 (Mukai, Bondal-Orlov, Bridgeland).

Let X,Y be smooth and projective, Φ:Db⁢(X)→Db⁢(Y), x∈X, and 𝒫x≔Φ⁡(𝒪x).

  1. (1)

    Φ is fully faithful if and only if

  2. (2)

    Φ is an equivalence if, in addition, 𝒫x⊗ωY≅𝒫x.

Proof of Theorem 2.4.

First, note that Φ𝒰⁢(𝒪p)=𝒰p. The second condition of the lemma is clear because S is Calabi-Yau, so we only need to check the first condition. Clearly Hom⁡(𝒰p,𝒰p)=ℂ because 𝒰p is stable, so for points p,q we need to consider Exti⁡(𝒰p,𝒰q). This vanishes for i≤1 or i≥3 because 𝒰p,𝒰q is an honest sheaf. We can see that

where the first equality is because 𝒰p,𝒰q are different stable sheaves, the second follows from Serre duality, and the third follows from Riemann-Roch. □