Patrick Lei

All notes

Moduli spaces and hyperkähler manifolds

Lectures by Giulia Saccà, Columbia, Fall 2021.

Moduli Spaces and Hyperkähler Manifolds
Fall 2021
Notes by Patrick Lei
Lectures by Giulia Saccà

Columbia University

Disclaimer

Unless otherwise noted, these notes were taken during lecture using the vimtex package of the editor neovim. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the instructor. If you find any errors, please contact me at plei@math.columbia.edu.

Acknowledgements

I would like to acknowledge Kevin Chang, Kuan-Wen Chen, Che Shen, and Nicolás Vilches for pointing out mistakes in these notes.

1 Hyperkähler Manifolds

Some useful references for K3 surfaces are the book by Huybrechts and the Barth-Peters-van de Ven book Compact Complex Surfaces and another book. For Hilbert schemes some references are Chapter 7 of FGA Explained, Huybrechts-Lehn, some lectures of notes of Lehn, and Nakajima’s Lectures on Hilbert Schemes. For Hilbert schemes of K3 surfaces and abelian varieties, there is Beauville’s Varietés Kahlerienne dont la premiere classe de Chern est nulle.

1.1 Motivation

Giulia believes that hyperkähler manifolds are some of the most interesting objects in algebraic geometry because one can actually prove results about high-dimensional hyperkähler varieties, unlike the usual situation in algebraic geometry. Because these objects are of a differential-geometric nature, through the course we will work over ℂ.

Recall that in order to classify curves, for a given curve C, we want to consider the positivity of the canonical bundle. In the first case, we know ωℙ1=𝒪⁡(−2)<0, in the second case of an elliptic curve, we have ωC=𝒪C, and finally for a higher genus curve the canonical sheaf ωC>0 is ample.

In higher dimension, let X be a smooth projective variety. Then there exists an integer κ⁡(X), the Kodaira dimension of X such that

for m≫0 sufficiently divisible. Of course, this is a birational invariant.

There is a classification of surfaces. Each smooth surface is birational to a minimal surface. Here, a surface S is minimal if any birational morphism from S to a smooth surface is a birational curve. By Castelnuovo, we know that S is minimal if and only if it does not contain a (−1)-curve. Also, any surface dominates a minimal surface.

1.2 Hyperkähler manifolds

Example 1.2.1.

If S is a surface, then κ⁡(S)=−∞ if and only if

for all m≥1. Some examples of these are ℙ2 and ℙ1×ℙ1.

Example 1.2.2.

If S is a surface, then κ⁡(S)=0 if and only if Pm⁢(S)=0 generically and there exists m such that Pm⁢(S)=1. In this case, there are two cases. First, h0⁢(ωS)=1, in which case ωS=𝒪S, and thus either h1⁢(𝒪S)=0 (in which case we have a K3 surface) or 𝔥1⁢(𝒪S)=2 (in which case we have an abelian surface).

Alternatively, we may have h0⁢(ωS)=0, in which case there exists m≥2 (where m∈{2,3,4,6}) such that ωS⊗m=𝒪S. Either h1⁢(𝒪S)=0, in which case we have an Enriques surface, or h1⁢(𝒪S)=1, in which case we have a bi-elliptic surface.

If T is an Enriques surface, then there exists a K3 surface S with a 2:1 étale cover S→T. On the other hand, any bi-elliptic surface has an m:1 étale cover from an abelian surface.

Theorem 1.2.3 (Beauville-Bogomolov).

Let M be a compact Kähler manifold with c1⁢(ωM)=0. The there exists a finite étale cover of M by a product

where Tn is a complex torus, the Yi are strict Calabi-Yau, and the Xi are irreducible holomorphic symplectic (or hyperkähler).

Definition 1.2.4.

Let Y be a compact Kähler manifold. Then Y is strict Calabi-Yau if π1⁢(Y)=1 and H0⁢(ΩYP)=ℂ when p=0,dimY and H0⁢(ΩYP) vanishes elsewhere.

Definition 1.2.5.

A compact Kähler manifold X is irreducible holomorphic symplectic if π1⁢(X)=1 and H0⁢(ΩX2)=ℂ⁢σX, where σX is an irreducible symplectic form. In particular, σn is a nonzero top form and thus trivializes the canonical bundle. In addition, σ induces an isomorphism of holomorphic vector bundles ΩX≃TX.

1.3 Some surfaces

Returning to the simplest case, we will define K3 surfaces.

Definition 1.3.1.

A smooth projective surface S is a K3 surface if ωS=0 and h1⁢(𝒪S)=0.

It follows from the definition that K3 surfaces are simply connected, so they are in fact both strict Calabi-Yau and irreducible holomorphic symplectic. Later in the course, we will see that irreducible holomorphic symplectic varieties are the true higher-dimensional analogues of K3 surfaces.

Lemma 1.3.2.

Let S be a K3 surface and f:S→C be a dominant morphism to a smooth projective curve C with connected fibers. Then C=ℙ1 and the general fiber of f is an elliptic curve.

Proof.

The proof is left as an exercise to the reader. □

Any K3 surface S with a dominant map to a curve is called an elliptic K3. As a consequence, any surjective map f:S→B where B is not a point and f has connected fibers has either B=ℙ1 or B is a singular K3. This is generalized by the following remarkable result:

Theorem 1.3.3 (Matsushita).

Let X2⁢n be an irreducible holomorphic symplectic manifold and f:X→B be a proper surjective morphism with connected fibers with B a normal variety. If B is not a point, then either dimB=n and f is a Lagrangian fibration where the general fiber is an abelian n-fold or dimB=2⁢n and B is a singular symplectic variety if f is not an isomorphism. In the second case, f is called a symplectic resolution.

Remark 1.3.4.

There is another extremely difficult result of Hwang, which says that if B is smooth, then B=ℙn (if dimB=3, apparently B is a ℚ-factorial Fano threefold with klt singularities).

Now we will consider some examples. Beginning in the simplest case, consider a general section f4∈|𝒪ℙ3⁢(4)|. By the Bertini theorem, the general S=(f4=0) is smooth, and by the adjunction formula, ωS=𝒪S. Then we consider the exact sequence

and by the long exact sequence of cohomology and the known values of cohomology for projective space, we have H1⁢(𝒪S)=0.

Example 1.3.5.

A concrete example of this is the Fermat quartic, which has the equation

We will see that this is an elliptic K3. The first step is to see that S contains a line ℓ⊆S⊆ℙ3, so we choose a primitive ζ8 and set x0=ζ8⁢x1 and x2=ζ8⁢x3. Now we project S from ℓ, and considering planes that contain ℓ, we obtain a rational map S⇢ℙ1. This extends over ℓ. Finally, we know that S∩ℙ2 is a quartic curve containing a line ℓ, so in fact the generic fiber of this map is an elliptic curve.

Similarly, we may consider other complete intersections, such as the (2,3) complete intersection in ℙ4 (intersection of a quadric and a cubic) and the (2,2,2) complete intersection in ℙ5. In higher dimensions, any degree (n+1) hypersurface Y in ℙn has ωY=𝒪Y. By the Lefschetz hyperplane theorem, this is a strict Calabi-Yau.

Example 1.3.6.

Let Γ∈|𝒪ℙ2⁢(6)| be a general sextic and S be a 2:1 cover of ℙ2 branched along Γ. We will use the covering trick, which holds for any variety X, line bundle L, and 0≠s∈H0⁢(L⊗m) for some m≥1. Then if we set D=(s=0), there exists a finite flat morphism f:Y→X that is a ℤ/m-cover away from D and ramified along D. In this case, f∗⁢L has a section t such that (t=0)≃D. Finally if X and D are smooth, so is Y, and ωY=f∗⁢ωX⁢((m−1)⁢(t=0)).

In our example, we have ωS=f∗⁢ωℙ2⊗𝒪X⁢(y2=Γ)=𝒪S, so S is a K3 surface.

Example 1.3.7 (Kummer K3 surfaces).

Let A be an abelian surface. It has an involution −1 with fixed locus A⁡[2]. Thus A/±1 has 16 singular points that look like ℂ2/±1=Specℂ[x2,xy,y2]=Specℂ[a,b,c]/(ab=c2) (the A1 singularity). Now the surface S=BlA⁡[2]A/±1 is a K3 surface.

It is easy to see that the smooth locus of A/±1 has a holomorphic symplectic form σA/±1. Then we can bull back f∗σA/±1 to a holomorphic symplectic form on f−1⁢(U)⊆S, and this form extends to S. The reason for this is that BlA⁡[2]⁡A still has the involution −1, and S is the quotient of BlA⁡[2]⁡A by this involution. If we denote this diagram by

and denote A~≔BlA⁡[2]⁡A, then we obtain

and therefore q∗⁢ωS=𝒪A~, so ωS=𝒪S. The morphism f is called a symplectic resolution.

Before we proceed, we will discuss crepant and symplectic resolutions. Let Y be a smooth variety, so ΩY1 is locally free. Then ωY≔⋀dimYΩY1 is called the canonical bundle. Then if f:X→Y is a birational morphism of smooth varieties, we have an exact sequence

Then we know ΩX/Y1 is supported on the exceptional locus of f. Because Y is smooth, then the exceptional locus is divisorial, and thus

Now suppose that Y is just normal with smooth locus U. Also suppose that Y∖U has codimension at least 2, so Weil divisors on U and Y are the same. There are two ways to extend ωU to Y. The first is to denote the inclusion j:U⊆Y and consider the sheaf j∗⁢ωU, which is generally not locally free. On the other hand, we can extend the Weil divisor KU to Y, which determines a Weil divisor KY on Y, called the canonical class.

Remark 1.3.8.

In general, the Weil divisor KY is not Cartier. In fact, KY is Cartier if and only if j∗⁢ωU is locally free.

Now let f:X→Y be a reolution of Y. This means f is proper and an isomorphism over U. We want a formula relating of the form KX=f∗⁢KY+∑ai⁢Ei. Unfortunately, we can only pull back Cartier divisors, so we will assume that KY is ℚ-Cartier, which means that there exists m≥1 such that m⁢KY is Cartier. We know that f−1⁢(U)≃U, so KX|f−1⁢(U)=f∗⁢KY|f−1⁢(U). Thus there exist integers ai such that

where the Ei are the divisorial components of X∖f−1⁢(U). Formally dividing by m, we have

Here, the ai are known as the discrepancies and if ai=0, then the resolution is called crepant.

Example 1.3.9.

One example of a crepant resolution is S→A/±1.

Example 1.3.10.

Consider Y=ℂ2⁢N/±1. This is the cone over the degree 2 Veronese embedding of ℙ2⁢N−1. Now we will write f:X=Bl0⁡Y→Y, and the exceptional divisor is a ℙ2⁢N−1. We know that X is the total space of 𝒪⁡(−2), so there is a projection X→ℙ2⁢N−1. Now we need to compute a in the formula

First note that KY=0. This is because the standard holomorphic symplectic form on ℂ2⁢N descends to the smooth locus U⊆Y, so we have a symplectic form on X∖E. Now by the adjunction formula, we have

and thus because E=ℙ2⁢N−1, we have

Finally, we see that 𝒪X⁢(E)|E=𝒪ℙ2⁢N−1⁢(−2), and thus a+1=N, so a=N−1. In particular, f is a crepant resolution if and only if N=1. For N≥2, f∗⁢ωU extends to X with a zero of order N−1 along E. Therefore the form

has a zero of order N−1 along E, so f∗⁢σU does not extend over E.1

1.4 Hilbert Schemes of points on surfaces

Let X be a smooth quasiprojective surface. Consider the functor

associating a scheme T to isomorphism classes of flat proper morphisms T×X⊇Z→T satisfying pZt⁢(t)=n.

Theorem 1.4.1 (Grothendieck).

The functor HilbXn is representable by a quasiprojective scheme X[n]. If X is projective, so is X[n].

Later in the course, we will sketch a construction of the Hilbert scheme, but for now we will simply assume that it exists. A fundamental resolt about Hilbert schemes is

Theorem 1.4.2 (Fogarty).

Let X be a smooth quasiprojective surface. Then X[n] is a smooth connected quasiprojective variety of dimension 2⁢n and there exists a morphism h:X[n]→X(n), called the Hilbert-Chow morphism,2 which is a resolution of singularities. Here, if Z is a length n subscheme of Z, we have

Example 1.4.3.

For n=2, we are looking for ideal sheaves I⊆𝒪X with quotient 𝒪Z of length 2. At a point p, we know I/𝔪2⊆𝔪/𝔪2, and thus subschemes of length 2 supported on p form a ℙ⁢𝔪/𝔪2=ℙ1.

Sketch of smoothness.

We need to compute the Zariski tangent space at a given point, so we have

By definition, these are flat proper families of length n subschemes 𝒵→Spec⁡ℂ⁡[ε] such that 𝒵|ε=0=Z, and by a computation (for example in FGA Explained) we have

To compute the dimension, we begin by considering the exact sequence

and applying the functor HomX⁡(−,𝒪Z), we have an exact sequence

and thus because HomX⁡(𝒪X,𝒪Z)→HomX⁡(IZ,𝒪Z) is the zero morphism, and ExtX1⁡(𝒪Z,𝒪Z)=H1⁢(𝒪Z)=0, we can simply compute the Ext group. Here, we have

We simply need to show that the Euler characteristic vanishes because

has dimension n, as does Hom⁡(𝒪Z,𝒪Z). To do this, we use Grothendieck-Riemann-Roch, which says that

and here we see that χ⁡(𝒪X,𝒢)=χ⁡(ℱ) where ℱ=O~X. Now because supp⁡(ℱ) has dimension 0, then ch⁡(ℱ)=[0,…,±ℓ⁡(ℱ)]. □

Exercise 1.4.4.

Prove that χ⁡(𝒪Z,𝒪Z)=0 using a locally free resolution in the first variable.

Now we will review some basic theory of Hilbert schemes for quasiprojective varieties. Here, if X is quasiprojective and p⁡(t)∈ℚ⁡[t] is some Hilbert polynomial, consider [Z]∈HilbXp⁡(t).

Proposition 1.4.5.

If I⊆𝒪X is the ideal sheaf of Z, then

Sketch of proof.

We know that T[Z]Hilb=Hom(Speck[ε],Hilb,0↦[Z]). This set of morphisms is the same as the set of 𝒵⊆X×Spec⁡k⁡[ε] flat over k⁡[ε]. And a module M is flat over k⁡[ε] if and only if M⊗(ε)≃ε⋅M. Now we want an ideal sheaf I~⊆𝒪X⁢[ε] such that

But now we can see that I=I~/ε⁢I⊆𝒪X⊕ε⁢𝒪Z, and thus giving I~ is the same as giving an element of HomX⁡(I,𝒪Z). □

Exercise 1.4.6.

Let X be a smooth quasiprojective curve. Show that X[n]=X(n) is smooth of dimension n.3

Theorem 1.4.7.

Let X be a quasiprojective variety. Then there exists a regular proper morphism

which is surjective and birational. By a result of Fogarty, the fibers of hn are connected, so if X is connected, so is X[n].

From now on, we will assume that X is projective. Therefore, for all Z⊆X with ℓ⁡(Z)=n, there exists an open affine neighborhood U⊆X containing Z. Therefore we have [Z]∈U(n)⊆X(n). Now if Z=∑αi⁢pi and Ui∋pi are open neighborhoods, then Z∈∏Ui(αi).

Remark 1.4.8.

If X is a smooth surface, then the local structure of X(n) at n⁢p is the same as the local structure of (𝔸n)(n) at n⋅{0}. In particular, when n=2, we have

where Δ is smooth of dimension 2 and Q is the quadric cone.

Now for any partition n=∑αi of n into positive integers with length k, write α¯=(αi). Then define

These Xα(n) give a stratification of X(n) into locally closed subsets, where the open stratum is X(1,1,…,1)(n) and the closed stratum is X(n)(n). It is easy to see that dimXα¯(n)=2⁢ℓ⁢(α). Another important stratum is X(2,1,…,1)(n), where exactly two points come together. Now note that

where the hαi−1 are the punctual Hilbert schemes Hilbαi⁢(𝒪X,zi)≃Hilbαi⁢(k⁡[x1,x2],0). For α=2, the punctual Hilbert scheme is simply ℙ⁢𝔪/𝔪2.4

Theorem 1.4.9 (Briançon).

The fiber hn−1⁢(n⁢z) is irreducible of dimension at most n−1.

In particular, this tells us that X(1,…,1)[n]→X(1,…,1)(n) has fibers of dimension 0 and is thus an isomorphism.

Proposition 1.4.10.

The exceptional locus of hn is an irreducible divisor E.

Proof.

Because X(n) is normal and ℚ-factorial,5 then any birational Y→X(n) from a smooth variety Y has divisorial exceptional divisor.

Now the exceptional locus E(2,1,…,1)⁢t⁢o⁢X(2,1,…,1)(n) has fibers ℙ1, while for a general α¯, we have

Because the strata are irreducible and so are the fibers, we obtain irreducibility for the exceptional divisor. □

Proposition 1.4.11.

Let X be a projective variety. Then there exists a birational surjective morphism h:X[n]→X(n).

Proof.

We will show that for all Z⊆T×X proper and flat over T with ℓ⁡(Zt)=n for all t, there exists a natural morphism T→X(n) given by

Fix t0∈T. Because X is projective, there exists U⊆X affine with Zt0∈U=Spec⁡A. Then because p:Z→T is proper, there exists t0∈V⊆T, where V=Spec⁡B is affine, and for all t∈V, Zt∈U. In conclusion, we have a family ZV⊆V×U. At the level of rings, we have a diagram

Now we need a map (A⊗n)Sn→B. Because Z→V is flat, C is a rank n projective B-module. Clearly we have a map A→EndB⁡(C) given by A↦φ⁡(1⊗a), and thus A⊗n acts on C⊗n. Then we obtain an action of (A⊗n)Sn on ⋀nC, which is just a map

which is the map we want. □

As an example, consider B=k. Then C=∏Ci is Artinian, hence a product of Artinian local rings Ci of length αi. Then the map we defined at the end of the proof factors through ∏Symαi⁢(Ci).

Theorem 1.4.12 (Beauville-Fujiki).

Let X be a smooth surface. Then the Hilbert-Chow morphism is a crepant resolution and if X has a holomorphic symplectic form, so does X[n].

Exercise 1.4.13.

If X is a smooth surface, prove that

Proof.

Consider X∗(n)=X(1,…,1)(n)∪X(2,1,…,1)(n) and define X∗n,X∗[n] similarly. In Xn consider Δ=⋃Δi⁢j, where the i-th and j-th points coincide. Now consider the diagram

which quite clearly commutes by the same argument showing that X[n] is the blowup of X(n) along the diagonal. Then the exceptional divisor ⋃Ei⁢j is fixed by S2, so it maps to E∗⊂X∗[n]. Now it suffices to prove that X∗[n]→X∗(n) is crepant because the complement has codimension 2. To see this, the quotients by Sn have simple ramification, and thus we have

so a=0 because π:X∗n→X∗(n) is étale away from codimension 2 and thus KX∗n=π∗⁢KX∗(n). Therefore the Hilbert-Chow morphism is a crepant resolution.

Now suppose that X has a holomorphic symplectic form ωX∈H0⁢(ΩX2). By codimension reasons, it is enough to produce a holomorphic symplectic form on X∗[n]. Clearly we have a symplectic form ω≔∑ipi∗⁢(ωX) on X∗n, which is clearly Sn-invariant. Therefore, we obtain a symplectic form σX(1,…,1)(n) on X(1,…,1)(n) and a symplectic form η∗⁢ω on BlΔ⁡X∗n, which is degenerate along ⋃Ei⁢j and Sn-invariant. This induces a holomorphic 2-form σX∗[n] on X∗[n] (as in there exists such a σ such that η∗⁢ω=ρ∗⁢σ). We know that σX∗[n] is generically nondegenerate.

We now show that σ≔σX∗[n] is symplectic. We know σn is a section of ωX∗[n], so the degeneracy locus of σ is the zero locus of σn. However, we know KX∗(n)=0 by the existence of ωX, and because h is crepant, we see that KX∗[n]=0, and thus σn must be nonzero everywhere. □

We will now discuss some invariants of X[n].

Proposition 1.4.14.

There is an isomorphicm of Hodge structures

Now we know that H2⁢(X(n))=H2⁢(Xn)Sn. By the Künneth formula, we have

and therefore

Now

Over ℤ, we can check in local coordinates that there exists a class δ∈H2(X[n],ℤ) such that 2⁢δ=E.7

Corollary 1.4.15.

If X is a K3 surface, then there is an isomorphism

as Hodge structures, and in particular,

We will now sketch the computation of the fundamental group of X[n]. One fact is that

is an isomorphism. In particular, if X is a K3 surface, then π1⁢(X[n])=0, so X[n] is irreducible holomorphic symplectic.

If A is an abelian surface, we know A has a holomorphic symplectic form σA, and thus A[n] has a holomorphic symplectic form σA[n], so ωA[n]=𝒪A[n]. However, we know that H1(A[n])=H1(A)=ℤ4, so it is not simply connected. But then we know that

so A[n] has larger H2,0. By Beauville-Bogomolov, we know that A[n+1] has an étale cover by a product of complex tori, irreducible holomorphic symplectics, and strict Calabi-Yaus. There exists a natural morphism

and an action of A on A[n+1] given by translation. Of course, this is not equivariant because ∑(zi+a)↦∑zi+(n+1)⁢a, and by generic smoothness all fibers are isomorphic and smooth. Now if we consider the diagram

we see that A×AA[n+1]=Kn⁢(A)×A. Later, we will show that Kn⁢(A) is irreducible holomorphic symplectic.

Example 1.4.16.

If n=1, then K1⁢(A) is the Kummer K3 surface associated to A.

Proposition 1.4.17.
  1. 1.

    ωKn⁢(A)=𝒪Kn⁢(A);

  2. 2.

    The restriction of teh holomorphic symplectic form σA[n+1]|Kn⁢(A) is a symplectic form.

  3. 3.

    H2⁢(Kn⁢(A))=H2⁢(A)⊕ℚ⁢F and π1⁢(Kn⁢(A))=1, where F is one of the fibers of E→A, where E⊂A[n+1] is the exceptional divisor of h.

We will prove a result that

Proposition 1.4.18.

There exists a (non-effective) line bundle ℒ on X[n] such that ℒ⊗2=𝒪X[n]⁢(E).

Proof.

Consider BlΔ⁡X∗n/An. This has simple ramification over E∗⊆X∗[n], and thus f∗⁢𝒪Z=𝒪X∗[n]⊕ℒ. This is the desired line bundle. □

Corollary 1.4.19.

If X is a K3 surface, then Pic⁡(X[n])=Pic⁡(X)+ℤ⁢δ, where δ=c1⁢(ℒ).

1.5 Generalized Kummers

Recall the construction of the varieties Kn⁢(A) for an abelian surface A. Recall the diagram

Also recall that π1⁢(A[n+1])=π1⁢(A). Using the long exact sequence of homotopy groups and the fact that A is a K⁡(ℤ4,1), we see that π1⁢(Kn⁢(A))=0.

Proposition 1.5.1.

Kn⁢(A) is an irreducible holomorphic symplectic manifold. In particular,

  1. 1.

    If σA[n+1] is the holomorphic symplectic form on A[n+1], then its restriction to Kn⁢(A) is a holomorphic symplectic form.

  2. 2.

    H2⁢(Kn⁢(A))=H2⁢(A)⊕Q⁢F, where F=E∩Kn⁢(A).

Proof.

Consider the Leray filtration on H2⁢(A[n+1]) induced by the map a. Here, we have

where invariants are taken with respect to the monodromy group of a, which is A⁡[n+1] because base change by A→n+1A trivializes a. We will show that the last inclusion is an equality. Also, note that

and that if α,β∈H1⁢(A), then

Next, write K∗n⁢(A) analogously to A∗[n] and let N=ker⁡(An+1→A). Then we have a diagram

Note that N has an action of Sn+1 and an action of A⁡[n+1] given by adding ε to all elements that preserves N and Δ. Then we know that

has an action of A⁡[n+1] has an action by translation, which is trivial in cohomology. Finally, we conclude that

and obtain the desired result. □

Now we have two examples of irreducible holomorphic symplectic manifolds. The first is K3[n] with b2=b2⁢(K3)+1=23 and the second is Kn⁢(A) with b2=b2⁢(A)+1=7.

Proposition 1.5.2.

Let f:𝒳→B be a smooth proper morphism of complex manifolds such that for some 0∈B, 𝒳0⁢(B) is a Kähler irreducible holomorphic symplectic manifold. Then there exists an analytic neighborhood 0∈V⊆B such that for all t∈U, 𝒳t is Kähler and holomorphic symplectic.

Proof.

By a result of Kodaira, being Kähler is an open condition, so there exists an open U⊆B such that for all t∈U, 𝒳t is Kähler. Therefore, for all t∈U, the map t↦hp⁢(𝒳t,Ω𝒳tq) is constant by Ehresman’s theorem that this family is topologically trivial and upper semicontinuity.

This implies that up to further restricting U, f∗⁢Ω𝒳/B2|U is free. This implies that σ0∈H0⁢(Ω𝒳02) extends locally to a section σ~∈H0⁢(𝒳U,Ω𝒳U/U2). To check that this is symplectic, we know that σ~n∈H0⁢(K𝒳/U) has closed zero locus which does not intersect the zero fiber, and so we obtain an open set where this form is nondegenerate. □

Proposition 1.5.3.

Let f:𝒳→B be a smooth proper family of Kähler manifolds. Then if 𝒳0 is irreducible holomorphic symplectic, so is 𝒳t for all t∈B.

Sketch of proof.

First, note that the relative canonical bundle K𝒳/B≅f∗⁢ℒ, where ℒ is a line bundle on B. By the same proof as before, there exists Z⊆B such that for all t∈B∖Z, 𝒳t is irreducible holomorphic symplectic.

Now suppose t0∈Z. Then K𝒳t0 is trivial and 𝒳t0 is simply connected, so 𝒳t0 is a product of irreducible holomorphic symplectic varieties and strict Calabi-Yau manifolds.

Now we will state without proof the fact that if X is a complex manifold with KX=𝒪X, then Def⁡(X) is smooth (as a germ of complex manifold). This is a nontrivial result of Bogomolov-Tian-Todorov. Note that if Xt0=∏Xi×∏Yi, then

because all Xi,Yi satisfy h1,0=0. Thus the splitting situation is impossible. □

It is known that if S is a K3 surface, then Def⁡(S) has dimension 20. Also, note that projective K3 surfaces are a 19-dimensional locus. Here, note that Def⁡(X)=h1⁢(TX)=h1⁢(ΩX1)=h1,1.

In the next case, if X=S[n], then Def⁡(X) has dimension 21, and there is a 20-dimensional locus of genuine Hilbert schemes of K3 surfaces. There are also higher-codimension loci parameterizing the spaces Mv⁢(S,h). Note that in both of these situations, the very general object is Kähler but not projective.

Now we will discuss some examples of Lagrangian fibrations.

Example 1.5.4.

Let f:S→ℙ1 be an elliptic K3 surface. Then we have a morphism

This is clearly a Lagrangian fibration.

Example 1.5.5.

Let A=E×F be the product of two elliptic curves and let φ:A→F be the second projection. Then we have a diagram

Here, we see that ε−1⁢(0)=ℙˇ2, and so in general there is a Lagrangian fibration Kn⁢(A)→ℙn.

1.6 Some operations

Now we will consider some birational transformations.

Example 1.6.1 (Atiyah flop).

Let f:𝒮→Δ be a family of quartic surfaces in ℙ3. Suppose that 𝒮t is smooth and 𝒮0 has one simple node p∈𝒮0. This simple node is given locally by x2+y2+z2=t.

Note that Blp⁡𝒮0≕𝒮~0 is a smooth K3 surface. We would like to modify the family such that we get smooth fibers for all t∈Δ. Now if we take a base change of Δ by t↦t2, locally at p∈𝒮¯ we have the equation x2+y2+z2=t2 is a singular point of 𝒮¯. But then 𝒳≔Blp⁡𝒮¯ is smooth, and 𝒳0=𝒮~0∪Q, where Q=ℙ1×ℙ1.

Unfortunately, the discrepancy of ν:𝒳→𝒮¯ is 1, so K𝒳=ν∗⁢K𝒮¯+Q, and so by adjunction we see that

and thus 𝒪X⁢(Q)=𝒪⁡(−1,−1). This tells us that we can contract Q along both of the factors and produce 𝒮+,𝒮− with maps to 𝒮¯. Then there is a birational map φ:𝒮+⇢𝒮− which is an isomorphism away from the central fiber.

We conclude that 𝒮0±=𝒮~0 and that φ is an isomorphism outside of the copies of ℙ1 that we contracted Q onto but does not extend over those copies of ℙ1. Also, note that 𝒳=Γφ and that 𝒳t=Γφt for all t≠0, and 𝒳0=𝒮~0∪ℙ1×ℙ1.

The next observation is that H2⁢(𝒮t0±)≃H2⁢(𝒮~0), but passing between the two identifications is actually reflection across the (−2)-curve produced from the Atiyah flop.

Here, we have used the following result of Nakano and Fujiki: Let M~ be a complex manifold and E⊆M~ be a smooth divisor that is a ℙn-bundle over some Z.. Then there exists a complex manifold M⊃Z and π:M~→M such that M~=BlZ⁡M if and only if 𝒪M~⁢(E)|E=𝒪X⁢(−1).

Another fact that we used to show that 𝒮0+ and 𝒮0− are isomorphic is that two birational K3 surfaces are isomorphic.

Now let X⊇ℙn be a holomorphic symplectic manifold of dimension 2⁢n. For example, some K3 surfaces contain (−2)-classes, which are isomorphic to ℙ1.

Lemma 1.6.2.

Any such ℙn⊆X is a Lagrangian submanifold of X. Moreover, if Z⊆X is any Lagrangian, NZ/X≅ΩZ1.

Proof.

The first part is clear because H0⁢(Ωℙn2)=0. Next, consider the exact sequencec

Note that the rightmost vertical morphism is generically injective with torsion kernel, but because NX|ℙn is torsion free, we have an isomorphism. □

Now consider Blℙn⁡X and let E be the exceptional divisor. Denote ℙn=ℙ⁢V for some vector space V.

Lemma 1.6.3.

We have an isomorphism E≃I⊆ℙ⁢V×ℙ⁢V∨, where I is the incidence subscheme. Moreover, we have 𝒪X~⁢(E)|E≅𝒪E⁢(−1,−1).

Proof.

We know that E=ℙ⁢Nℙn/X≃ℙ⁢Ωℙn1. Now if we consider the Euler sequence

we obtain an embedding

as the locus {(s,x)∣s⁡(x)=0}. Next, we use adjunction in X~ and in ℙ⁢V×ℙ⁢V∨ to see that

□

Now by the Nakano-Fujiki criterion, there exists X~′⊇ℙ⁢V and q′:X~→X′ such that we have the following diagram:

such that q′ takes E to ℙ⁢V∨.

Definition 1.6.4.

Such an X′ is called the Mukai flop of X at ℙn.

Remark 1.6.5.

We can perform the Mukai flop whenever we have Z⊆X such that there exists some ℙr-bundle structure Z→B and Z has codimension r in X. We also require that NZ/X≃ΩZ/B1.

Remark 1.6.6.

We have a diagram

Remark 1.6.7.

If X′ and X are isomorphic in codimension 2, they have isomorphic H2 and X′ is holomorphic symplectic.

The local structure of (X0,p) is isomorphic to that of the cone C∙⁢(I). In particular, X0 is not ℚ-factorial because the exceptional locus of π is ℙn, which is not a divisor. In addition, π is a crepant (symplectic resolution).

Proposition 1.6.8.

A birational map f:X⇢X′ of compact complex manifolds (or projective varieties) with trivial canonical bundles is an isomorphism in codimension 2. In particular, π1⁢(X)=π1⁢(X′) and H2⁢(X,ℤ)=H2⁢(X′,ℤ).

Proof.

Let Γ be the graph of f and consider the diagram

Then if E,F are the exceptional divisors of p,p′, we have KΓ=E=F up to linear equivalence. However, we know that H0⁢(m⁢F)=H0⁢(m⁢E)=H0⁢(m⁢KΓ)=1, but these h0⁢(ωX⊗m) are birational invariants, so E,F do not move in their equivalence class. In particular, we have an isomorphism X∖p⁡(E)≃X′∖p′⁢(F). □

Corollary 1.6.9.

Suppose that f:X↠S′ is a birational map of K3 surfaces. Then f is an isomorphism.

Proof.

Consider the graph Γ and diagram

We know that f is an isomorphism away from finitely many points. We know that f is not defined at x if and only if p−1⁢(x) is a curve. But then there exists a curve C′⊆S′ contracted by f−1, which is impossible. □

Example 1.6.10 (Beauville).

This example comes from the paper Some remarks on Kähler manifolds with c1=0 by Beauville.8 Let S⊆ℙ3 be a quartic K3 surface. Choose a length 2 point z∈S[2], which has linear span a line. But then ℓ∩S=z+w, and so we define a rational map φ:S[2]⇢S[2] given by z+w.

Proposition 1.6.11.
  1. 1.

    φ is regular at [Z]∈S[2] if and only if ℓ=⟨Z⟩⊈S.

  2. 2.

    If S⊇ℓ1,…,ℓk where the ℓi are disjoint lines, then φ is the Mukai flop at ℓ1[2],…,ℓk[2].

Proof.

We have a commutative diagram

and clearly p is finite over [ℓ]∈G if and only if ℓ⊈S. In particular, if ℓ⊆S, we have p−1⁢([ℓ])=ℓ[2]. Now consider the graph Γ and note that Γ⊆S[2]×GS[2]⊆S[2]×S[2]. Because S[2] is smooth, we know φ is regular at [Z] if and only if q1−1⁢(Z) is finite.

But now q1−1⁢(Z)⊆S[2]×S[2] is contained in [Z]×p−1⁢(ℓ). Thus, if p−1⁢(ℓ) is finite, so is q1−1⁢(Z). For dimension reasons, if ℓ⊆S is a line, then ℓ[2]×ℓ[2] is an irreducible component of S[2]×GS[2]. But then q1−1⁢(ℓ[2])=Γ∩ℓ[2]×ℓ[2], and then S[2]×GS[2]⊆S[2]×S[2] is a local complete intersection. But then irreducible components intersect in the correct dimension, so we are done.

It remains to show that f is the Mukai flop. We may assume that there is a unique line ℓ⊆S. The key technical lemma is that φ extends to Blℓ[2]⁡S[2], which means we have a map

But then φ~ takes E to itself, which means that it must swap the two rulings on E. But this means that the two copies of π contract E along the two rulings, as desired.

To prove the lemma, S[2]→G factors as S[2]→Z→G, where Z is normal and Z→G is finite. But then φ descends to an honest morphism φ¯:Z→Z, and thus if ℓ[2] is contracted to z0, φ¯ lifts to Blz0⁡Z=Blℓ[2]⁡S[2]. □

Proposition 1.6.12 (Huybrechts).

This proposition comes from the paper Birational symplectic manifolds and their deformations. Let ℙn⊆X2⁢n, where X is Kähler and symplectic and f:X⇢X′ be the Mukai flop. Then there exist two birational smooth proper families

such that ϕt is an isomorphism for all t≠0, 𝒳0=X, and 𝒳0′=X′.

Corollary 1.6.13.

There exists an isomorphism of Hodge structure H∗⁢(X)≃H∗⁢(X′).

Proof.

Let Γ⊆𝒳×Δ𝒳′ be the fiber product. Then we know Γt=Γφt⊆𝒳t×𝒳t′, and this implies that

is an isomorphism. But then we know H∗⁢(𝒳t′)≃H∗⁢(𝒳0′) and similarly for 𝒳. We also have a correspondence Γ0∗, and this is an isomorphism. □

Example 1.6.14.

Let S→ℙ2 be a degree 2 K3 surface. Then we obtain some ℙ2⊆S[2]. Then the Mukai flop of S[2] is a hyperkähler manifold M with a Lagrangian fibration over ℙˇ2. Here, if ℓ⊂ℙ2 is a line, we consider C∈|f∗⁢𝒪ℙ2⁢(1)|, and the fiber over [C]∈|C|=ℙ2ˇ is simply Pic2⁡(C). In addition, the Mukai flop takes z∈S[2] to the line bundle 𝒪C⁢(Z).

Proposition 1.6.15.

Let ℙn⊆X2⁢n be a Kähler holomorphic symplectic manifold and f:X⇢X′ be the Mukai flop. Then there exist 𝒳,𝒳′ over a disk Δ and ϕt:𝒳→𝒳′ such that ϕt is an isomorphism for t≠0, and (Γϕ)0=Γf+ℙn×ℙˇn.

Corollary 1.6.16.

There exists a universal deformation space for X (as a germ of complex manifold).

Proof.

We have an identity

Here, a deformation v is taken to the exact sequence

The first step in the proof is to show that there exists 𝒳→Δ such that Nℙn/𝒳=V∨⊗𝒪ℙn⁢(−1). To do this, note that H1⁢(X,TX)=H1⁢(X,ΩX1) and there is a sequence of maps

We also have a map H1⁢(X,ΩX1)→H1⁢(P,ΩP1), and the resulting diagram commutes. We need to find v∈H1⁢(X,TX) such that v|P≠0. Note that because X is Kähler, there exists a Kähler form ω that restricts to a nonzero form on P. Next, the exact sequence (1.2) remains exact after restricting to P, and therefore we have

Because vP≠0, the bottom sequence is not split, and because vP∈H1⁢(P,NP/X)=H1⁢(P,ΩP1) is contained in a 1-dimensional vector space, the sequence is actually the Euler sequence, and thus NP/𝒳=V∨⊗𝒪P⁢(−1).

Next, we consider the exceptional divisor ℙ⁡(V∨⊗𝒪⁡(−1))=ℙ⁢V∨×ℙ⁢V of BlP⁡X, and we can check that 𝒪⁡(E)|E=𝒪⁡(−1,−1). By Nakano-Fujiki, there exists a contraction of the exceptional divisor E onto the first factor. □

1.7 Deformations

For this part, we will follow notes by Voisin from a class in 2006-07. Let X be a compact complex manifold (or a reduced variety). We will see that

Fix Δn=Spec⁡ℂ⁡[t]/tn+1. By reducedness, for every 𝒳1→Δ1, we assign the exact sequence

Remark 1.7.1.

The sheaf Ω𝒳1 has torsion, but its restriction to the central fiber is locally free if X is smooth.

Conversely, given

we want to define an algebra 𝒪𝒳1 fitting in

Equivalently, we want a sheaf A and A→ℰ commuting with the inclusion of 𝒪 and the differentia d. We simply set

where r:ℰ→ΩX1 is the map in the exact sequence above. To define the algebra structure, we simply set

It remains to check that the kernel of A→𝒪X is a square zero ideal.

Next, we will consider global deformations of X a compact complex structure. We can consider the deformations of X over either germs of complex spaces or local Artinian rings.

Theorem 1.7.2 (Kuranishi).

If H0⁢(X,TX)=0, then there exists a universal family 𝒳→Def⁡(X) over a germ of (pointed) complex analytic spaces.

Alternatively, using the point of view of Schlessinger, because H0⁢(TX)=0, then DefX⁢(−) satisfies the axiom H4 and is thus pro-representable.

Remark 1.7.3.

There is generally no chance of having an algebraic family. Unfortunately, even if the central fiber is algebraic, there are arbitrarily small deformations that are not algebraic. If you want to keep everything algebraic, then we need to mark X with an ample line bundle.

Theorem 1.7.4 (Bogomolov, Tian, Todorov).

Let X be a compact Kähler Calabi-Yau manifold with H0⁢(TX)=0. Then the germ of space (Def⁡(X),0) is smooth (equivalently, the pro-representing ring R is a formal power series ring).

The proof of this result uses the T1-lifting principle, resting on the fact that by the infinitesimal lifting principle, smoothness of Def⁡(X) at 0 is equivalent to the fact that deformations can be lifted to any order. Before we do this, we need some notation and results.

Lemma 1.7.5.

Given an n-th order deformation fn:𝒳n→Δn, the sheaves Ω𝒳n1|𝒳n−1 and Ω𝒳n−1/Δn−11 are both locally free. Moreover, they fit into an exact sequence

Proof.

Note that ΩΔn1={dt∣tn⁢dt=0}. Then the sheaf ΩΔn1|Δn−1 is locally free of rank 1 generated by dt. Now because X is smooth, 𝒪̂X≃ℂ⁡[[x1,…,xm]], and in fact we have isomorphisms

for all n. This implies that Ω𝒳n1 is locally generated by dx1,…,dxm,dt with tn⁢dt=0. In particuar, after killing tn, we see that Ω𝒳n1|𝒳n−1 is locally free and generated by the dxi. We know the exact sequence

and restricting to 𝒳n−1, we obtain the desired result. □

Definition 1.7.6.

Given fn:𝒳n→Δn, set

This is called the Kodaira-Spencer class.

Remark 1.7.7.

By the lemma, we have

Theorem 1.7.8 (Ran).

Let X be a compact complex manifold. Given 𝒳n→Δn, there exists a lift fn+1:𝒳n+1→Δn+1 if and only if en lifts to some class en+1∈H1⁢(T𝒳n/Δn), where the map

is induced as follows: make the identification

Equivalently, given an extension

everything is locally free, so we can restrict to Δn−1 and use the identity Ω𝒳n/Δn1|𝒳n−1=Ω𝒳n−1/Δn−11.

Lemma 1.7.9.

The algebra 𝒪𝒳n+1 is determined by 𝒪Xn and the short exact sequence

As we did earlier, we construct A to fit into the exact sequence

Sketch of proof of theorem.

One direction is clear. If there is a lift, we construct the class en+1 and clearly, it must restrict to en by definition.

In the other direction, suppose there is a class en+1 lifting en. We want to construct 𝒪Xn+1. Given this en+1, we will find ℰ fitting into the sequence

Given any ℰ as above, there exists an isomorphism Ω𝒳n1|𝒳n−1≃ℰ|𝒳n−1 induced by a surjection r:ℰ→Ω𝒳n1. This will give us the desired algebra.

To prove that such an isomorphism exists, by the existence of en+1, we already have a surjection f1:ℰ→Ω𝒳n/Δn1. Then after restricting to 𝒳n−1, we have a sequence

We now claim that

Because g1∘f1=g2∘f2, we have our desired r:ℰ→Ω𝒳n1.

Next, we show that ker⁡r≃𝒪X. We have a short exact seqeunce defining ℰ, and then we obtain a diagram

Constructing A⊆ℰ⊕𝒪𝒳n as we did before, we are done. □

To prove that Def⁡(X) is smooth when X is a compact Calaby-Yau, we need to check the T1 lifting principle. In order to do this, we need some Hodge theory.

Lemma 1.7.10.

The sequence

is a resolution of the constant sheaf ℂ in the analytic topology.

Corollary 1.7.11.

Hk⁢(X,ℂ)=Hk⁢(X,Ω∙).

This gives us a filtration bête (for stupid filtration) Fp⁢ΩX∙=ΩX≥p, which by standard techniques leads to a spectral sequence where

which comes from the filtration Fp⁢Hk⁢(ΩX∙)=Im(Hk⁢(ΩX≥0)→Hk⁢(ΩX∙)).

If X is Kähler and compact, then the Hodge theorem implies that this spectral seqeunce (called the Frolicher spectral sequence) at the E1-page and that Fp⁢Hk⁢(X,ℂ) is the Hodge filtration. To see this, note that

Deligne in the paper Théorème de Lefschetz et Critères de Dégénérescence de Suites Spectrales shows that if X is any smooth proper scheme over ℂ, the Frolicher spectral sequence degenerates at E1.

Previously, we considered 𝒳→B smooth proper morphisms of complex manifolds. then if X0 is Kähler, the hp,q⁢(Xt) are locally constant, and in fact the degeneration of the Frolicher spectral sequence at E1 is enough. Thus if B is a complex manifold, the sheaves Rq⁢f∗⁢ΩX/Bp are locally free and satisfy base change.

Proposition 1.7.12.

Let f:𝒳→B be smooth and proper with B a scheme over ℂ (possibly of finite type). Then the higher direct images Rk⁢f∗⁢ΩX/B∙ and Rq⁢f∗⁢ΩX/Bp are locally free and satisfy base change. Moverover, there exists a filtration ℱp⁢Rk⁢f∗⁢Ω𝒳/B∙ whose successive quotients are locally free and whose associated graded components are Rq⁢f∗⁢Ω𝒳/Bp.

Lemma 1.7.13 (Deligne).

Let A be a local Artinian ring over ℂ and K∙ be a bounded above complex of free A-modules. Then ℓA⁢(Hn⁢(K∙))≤ℓ⁡(A)⋅ℓℂ⁢Hn⁢(K∙⊗Aℂ), and if equality holds, then base change holds in degree n,n+1, which means that for j=n,n+1, we have

where N is any B-module of finite type for some Artinian A-algebra B. In addition, Hn⁢(K∙) is a free A-module.

Proof of proposition.

We will reduce to the case of B=Spec⁡A, where A is an artinian ring over ℂ. Recall that we have an exact sequence

Then we have the relative de Rham complex which resolves

Now because B is affine, we note that

by A-linearity of the differential. Thus ℋk⁢Rk⁢f∗⁢ΩX/B∙ is a free A-module (in the general case, we obtain Rk⁢f∗⁢ΩX/B1=(Rk⁢f∗⁢ℂ)⊗𝒪B).

Now we consider the stupid filtration ℱp⁢ΩX/B∙=ΩX/B≥p and this induces a spectral sequence

where ℱp⁢Rk⁢f∗⁢ΩX/B∙=Im(Rk⁢ℱp⁢ΩX/B∙→Rk⁢ΩX/B∙). We now have

by the lemma, so all inequalities are equalities. Using the lemma again, the Rp⁢f∗⁢ΩX/Bq are free A-modules and satisfy base change. The remainder of the result is easy to see. □

Proof of Bogomolov-Tian-Todorov.

Given 𝒳n→Δn, we have a class en∈H1⁢(TXn−1/Δn−1), and we want to lift this to H1⁢(TXn/Δn). Suppose that X0 is a compact Calabi-Yau such that the Frolicher spectral sequence degenerates at E1 and dimX0=m. Then ΩXn/Δnm≅𝒪Xn, so there exists a perfect pairing

Thus TXn/Δn≃ΩXn/Δnm−1. If we consider

we are done. □

1.8 Some Hodge theory

Let f:𝒳→B be a proper surjective smooth morphism either of schemes over ℂ or with Kähler fibers Xb. Let

be the Hodge bundle. Then there is a decreasing filtration ℱp⁢ℋk⊂ℋk in subbundles. Now suppose that B is sufficiently small so that ℋk=Hk⁢(X0,ℂ)⊗𝒪B is free.

Theorem 1.8.1.

Fix k,p. Then there is a holomorphic map, called the period mapping,

where fkp=∑ℓ≥phℓ,k−ℓ⁢(X0). Pulling back the tautological seqeunce on the Grassmannian gives ℱp⁢ℋk⊆ℋk.

We would like to study φ, and a first step is to study its differentials. Note that φ factors via the universal deformation space Def⁡(X0), so we have a diagram

Now recall that T[W]⁢Gr⁢(j,V)=Hom⁡(W,V/W), and so the differential of the period map is a morphism

This factors through Hom⁡(Fp,Fp−1/Fp) by Griffiths transversality, and so we have the diagram

and the bottom term in the diagram is isomorphic to Hom⁡(Hk−p⁢(ΩXp),Hk−p+1⁢(ΩXp−1)).

Proposition 1.8.2.

The morphism H1⁢(X,TX)→Hom⁡(Hk−p⁢(ΩXp),Hk−p+1⁢(ΩXp−1)) is the morphism sending a tangent vector v to the morphism induced by contraction by v.

Returning to irreducible holomorphic symplectic manifolds, let X be irreducible holomorphic symplectic. Then the Hodge filtration here gives

Then the period map is a morphism

where we have made an identification ηt:H2⁢(Xt,ℂ)≅H2⁢(X0,ℂ).

Proposition 1.8.3.

The differential d℘0 has maximal rank.

Proof.

Write the morphism

It is enough to show that the morphism H1⁢(TX)→Hom⁡(H2,0,F1⁢H2/H2,0)=H1⁢(Ω1) is an isomorphism. But this map is given by contraction, and so it is precisely the isomorphism TX≃ΩX1 induced by σ. □

We conclude that ℘ is an isomorphism onto its image. There will be a local statement, where we consider small deformations, and a global statement, where we consider the entire moduli space. First, note that dimDef⁡(X)=b2−2 and dimℙ⁢H2⁢(X)=b2−1.

Proposition 1.8.4.

There exists a quadric hypersurface Q⊆ℙ⁢H2⁢(X,ℂ) such that Im℘⊆Q.

Proof.

First note that Im℘ is contained in a degree 2⁢n-hypersurface, where dimX=2⁢n. There is a natural degree 2⁢n polynomial on H2 given by the cup product of H2 with itself 2⁢n times. By type reasons9, because σXt is a (2,0)-form, we know σXt2⁢n=0, and thus σt2⁢n=0. Thus if F is the cup product polynomial, we see, Im℘⊆{F=0}⊆ℙ⁢H2⁢(X,ℂ). (Note that for K3 surfaces, we are done and that this quadric is defined over ℤ. In fact, the H2⁢(S,ℤ) is a unimodular lattice.)

Now we will prove that F=qn up to a constant. To do this, we will prove that F vanishes with order at least n on Im℘ and then write down a quadric explicitly. Write σt=σ0+t⁢α, and so we have

Thus F vanishes up to order n, as desired. □

Theorem 1.8.5 (Beauville, Bogomolov, Fujiki).

Let X be irreducible holomorphic symplectic manifold of dimension 2⁢n. There exists an integral, indivisible, quadratic form q:H2⁢(X,ℂ)→ℂ of signature (3,b2−3) and a constant cX∈ℚ>0 such that

for all α∈H2⁢(X).

Note that the relation, called the Fujiki relation, identifies q,cX with no ambiguity except when n is even, in which case we specify q⁡(ω)>0 for ω Kähler. Moreover:

  • •

    Im℘⊆Ω={q⁡(x)=0,q⁡(x,x¯)>0}⊆ℚ. In particular, the map ℘:Def⁡(X)→Ω is a local isomorphism.

  • •

    With respect to q, H1,1⟂H2,0⊕H2,0.

Proof.

We will normalize σ such that ∫(σ⁢σ¯)n=1. For a class α=a⁢σ+ω+b⁢σ¯, we will define

First we check that if ω is a Kähler form, then q⁡(ω)>0. In fact, we have

and now we use the Hodge-Riemann bilinear relations. Here, if M is compact and Kähler with dimM=m and ω a Kähler form, define

Then the bilinear form

is positive-definite. In our situation, note that σn−1 is a primitive form, and so we obtain the desired result. It is clear that q⁡(σ)=0,q⁡(Re(σ))=q⁡(Im(σ))>0, and q⁡(Re(σ),Im(σ))=0, and so q has rank at least 3 and is thus irreducible.

We now prove the Fujiki relation. By type reasons, we have σtn+1=0, and so if we write at⁢σ+ωt+bt⁢σ¯, and thus

and when we multiply by σ¯n−1, we have

Because at≠0 for sufficiently small t, we see q⁡(σt)=0, and thus (q)=I⁡(Im(℘))∋F, which gives us the Fujiki relation.

We will compute the signature of q. If we differentiate the equation

with respect to t under α+t⁢β, we have

Now if ω is Kähler form, then β is primitive if and only if q⁡(ω,β)=0. It is now enough to compute the sign of q|Hprim1,1. Differentiating again, we obtain

Choosing α=ω to be Kähler and β,γ primitive, we obtain

and by the Hodge-Riemann bilinear equations, the left-hand-side is negative.

Finally, we need to prove integrality. For all λ,α∈H2⁢(X,ℂ), we have

This is obtained from previous formulae by multiplying the derivative of the Fujiki repation by q⁡(α), using the Fujiki relation, and taking the derivative again. □

Corollary 1.8.6.

Up to multiplication by a nonzero constant, we can assume that if α∈H2⁢(X,ℚ), then q⁡(α)∈ℚ.

Proof.

We prove that there exists λ∈H2⁢(X,ℚ) such that q⁡(λ)≠0. The class σ+σ¯∈H2⁢(X,ℝ) and q⁡(σ+σ¯)>0, so by density of H2⁢(X,ℚ)⊂H2⁢(X,ℝ), we are done. □

Once we have this normalization, we have proven the integrality statement in Beauville-Bogomolov-Fujiki.

Remark 1.8.7.

The Fujiki relation implies that both q⁡(−) and cX∈ℚ>0 in the relation

are deformation invariants.

Using the Hodge-Riemann bilinear relations, we see that restricting q to H2,0⊕H0,2 (as a real vector space) is positive definite. In addition, a basis of H2,0⊕H0,2 is Reσ,Imσ, and in fact q⁡(Reσ,Imσ)=0.

Now the period map ℘:Def⁡(X)→Q lands in the set

This set Ω is called the period domain.

Remark 1.8.8.

The points of Ω parameterize Hodge structures on H2⁢(X,ℤ) of K3-type (1,b2−2,1). These structures have x∈H2⁢(X,ℂ) with q⁡(x)=0,q⁡(x,x¯)>0, and then H1,1 is the orthogonal complement of x.

We hope that for every Hodge structure, there is some manifold realizing the Hodge structure, but this is a highly nontrivial result of Huybrechts.

Proposition 1.8.9.

There exists a natural diffeomorphism Ω≃Gr+⁢(2,H2⁢(X,ℝ)), where the + means that for a subspace W, we have both an orientation and positive-definiteness of q|W given by

Now we consider the case of K3 surfaces. We know that if X=S is a K3 surface, then q is simply the cup product. To say more about this, first we will prove

Proposition 1.8.10.

We have an identity c2⁢(S)=24∈H4⁢(S,ℤ)=ℤ, and of course this means b2⁢(S)=22 and h1,1=20. In addition, we have H2⁢(S,ℤ)=E8⁢(−1)⊕2⊕U⊕3, where U=(0110).

Proof.

Recall Noether’s formula, which says that

In the case of a K3 surface, we know χ⁡(𝒪M)=2, so we get c2⁢(S)=24. Because c2⁢(S) is also the Euler class, we see that χtop⁢(S)=24, which gives us b2=22 and h1,1=20. The computation of H2⁢(S,ℤ) as a lattice follows from the abstract classification of lattices once we know that H2⁢(S,ℤ) is unimodular and even of indefinite signature. Recall that a lattice Λ is called even if for all α∈Λ, α2∈2⁢ℤ and unimodular if the matrix corresponding to the bilinar form has determinant ±1.10

To prove that H2⁢(S,ℤ) is even, we use Wu’s formula, which says that for all α∈H2⁢(M,ℤ) (for any compact complex surface M) we have c1⋅α≡α2(mod2). For a K3 surface, this clearly implies that H2⁢(S,ℤ) is even. □

We now consider Hirzebruch-Riemann-Roch on a K3 surface. If L is a line bundle and X is a surface, we have

For a K3 surface, we have χ⁡(S,L)=L22+2.

Corollary 1.8.11.
  • •

    If L2≥−2, then ±L is effective.

  • •

    If L2≥0, then either L=𝒪S or h0⁢(±L)≥2.

  • •

    If L=𝒪S⁢(C) for an irreducible curve C, then L2=2⁢g−2.

Example 1.8.12.

If C=R is a smooth rational curve, then R2=−2. If C=E is an elliptic curve, then E2=0.

Now for a compact complex manifold, define the Neron-Severi group NS⁡(X)=Pic⁡X/Pic0⁡X. This emerges from the exponential sequence

as the image of Pic⁡X→c1H2⁢(X,ℤ). By the Lefschetz theorem on (1,1)-classes, we know Imc1=H2⁢(X,ℤ)∩Hℝ1,1. In particular, if X is irreducible holomorphic symplectic, H1⁢(𝒪X)=0, so NS=Pic. The reason we care about this is that if we consider Def⁡(X)→Ω and identify H2⁢(X,ℝ)=H2⁢(Xt,ℝ), then the rank ρ⁡(X)=rk⁡Pic⁡X can vary.

We now return to consider consequences of the local Torelli theorem.

Proposition 1.8.13.

Let X be irreducible holomorphic symplectic. Then some small deformation of X is projective.

Proof.

It suffices to show that there exists a small deformation such that Xt contains a Kähler class that is rational. By the Kodaira embedding theorem, such a class is the first Chern class of an ample line bundle. To prove this, recall the identification Ω≅Gr+⁢(2,H2⁢(X,ℝ)). But then the set of planes W⊆H2⁢(X,ℝ) defined over ℚ are dense, so for 0∈Def⁡(X) and corresponding ℘⁡(0)∈Ω, we can choose a nearby point such that W is defined over ℚ. But then W⟂=H1,1 is also defined over ℚ. Thus if t∈Def⁡(X) satisfies ℘⁡(t)=W, then H1,1⁢(Xt) is also defined over ℚ. In particular, Hℝ1,1∩H2⁢(Xt,ℚ) is dense and has maximal rank. But then it must have nonempty intersection with the Kähler cone of Xt, so we are done. □

Now let X be irreducible holomorphic symplectic, L be a line bundle, and ℓ=c1⁢(L)∈H2⁢(X,ℤ). Set Ωℓ≔Ω∩ℓ⟂ and Def⁢(X)ℓ=℘−1⁢(Ωℓ)⊆Def⁡(X). Then if 𝒳→Def⁡(X) is the universal deformation, we will call 𝒳ℓ the base change to Def⁢(X)ℓ.

Proposition 1.8.14.

The space Def⁢(X)ℓ⊂Def⁡(X) is a smooth hypersurface and is the univeral deformation space of Def(X,L) (which means there exists a universal line bundle ℒ on 𝒳ℓ such that ℒ|X0=L, ℒ|Xt=Lt, and c1⁢(Lt)=ℓt). More generally, if L1,…,Lk∈Pic⁡X have ℓ1,…,ℓk linearly independent, then Def⁡(X,L1,…,Lk)⊆Def⁡(X)=℘−1⁢(Ω∩⟨ℓi⟩⟂) is smooth of codimension k.

Proof.

Note that ℓ⟂⊆ℙ⁢H2⁢(X,ℂ) is smooth if and only if q⁡(ℓ)≠0. On the other hand, if q⁡(ℓ)=0, the only singular point is ℓ∈ℓ⟂, but then such an ℓ cannot lie in Ω. Thus a necessary condition for Xt to have a line bundle Lt with c1⁢(Lt)=ℓt is that ℓt is a (1,1)-class, which is equivalent to q⁡(ℓt,σt)=0. Of course, this is equivalent to ℘⁡(t)∈Ωℓ. In particular, we know that if (𝒳,ℒ)→B is a deformation of (X,L), the Kodaira-Spencer map B→Def⁡(X) must factor through Def⁢(X)ℓ. Of course, by the Lefschetz (1,1)-theorem, this is also a sufficient condition. In particular, for all t∈Def⁢(X)ℓ, we have q⁡(σt,ℓt)=0, so ℓt∈H1,1∩H2⁢(X,ℤ), so there exists a unique line bundle Lt such that c1⁢(Lt)=ℓt. Therefore on Xℓ→Def⁢(X)ℓ, every fiber has a line bundle, so we prove that there exists a global line bundle and that such a line bundle is universal.

We show that Def⁡(X,L) is unobstructed with tangent space ker⁡[H1⁢(TX)→c⁡(L)H2⁢(X,𝒪X)]⊆H1⁢(TX). This is induced by the perfect pairing

Consider 𝒪X→dΩX1 and the corresponding map 𝒪X∗→ΩX1 given by u↦duu. This induces a map

Thus we have L↦c⁡(L)∈[0→𝒪X→ℰL→TX→0]. This gives us an exact sequence

We also have H1⁢(X,ℰL)=TDef⁡(X,L) and of course H1⁢(X,TX)=TDef⁡(X). In addition, we have H2⁢(X,ℰL)=Obs⁡(Def(X,L)), but deformations of X are unobstructed, so Obs⁡(X,ℰL)=0, and thus deformations of (X,L) are unobstructed. □

1.9 Noether-Lefschetz loci

Definition 1.9.1.

Let f:𝒳→B be a non-isotrivial family of irreducible holomorphic symplectic varieties over a connected B.11 Define

Then the Noether-Lefschetz locus12 of f is

Proposition 1.9.2 (Green).

The Noether-Lefschetz locus NL⁡(f)⊆B is dense in the analytic topology.

In fact, we will prove a stronger statement.

Proposition 1.9.3 (Oguiso).

Suppose that B is small enough such that there is an identification ηt:H2⁢(Xt,ℤ)≃H2⁢(X0,ℤ)≕Λ. In this case, there exists a period mapping ℘:B→ℙ⁢Λℂ. Then there exists a primitive sublattice Λ0⊆Λ of rank ρ0 such that for all t∈B, NS⁡(Xt)⊇Λ0.13

Proof.

Let I be the set of all possible primitive sublattices Λα⊆Λ. Then for all α∈I, define Bα={t∈B∣NS⁡(Xt)=Λα}. But then we see that B=⋃Bα and that ℘⁡(Bα)⊆ℙ⁡(Λα,ℂ)⟂. But now we know that

and thus there exists α0 such that B=℘−1⁢(Λα0,ℂ⟂) and therefore Λα0⊆NS⁡(Xt) for all t. □

Proof of Green.

We will assume that B=Δ is a disk and that ℘:Δ→ℙ⁢Λℂ is injective. Let ℋℂ2=R2⁢f∗⁢ℂ⊗ℂ𝒞∞⁢(Δ)≃H2⁢(X0,ℂ)×Δ (where the last identification is local). Here, we have a Hodge filtration ℱ∙⁢ℋℂ2, and we let ℋℝ2 be the real part of this bundle. If we intersect ℱ1⁢ℋ2∩ℋℝ2, we obtain precisely the bundle ℋℝ1,1.

The key claim is that the natural map ϕ:ℋℝ1,1→H2⁢(X0,ℝ) is an open immersion. Assuming this, we know H2⁢(X0,ℚ)⊆H2⁢(X,ℝ) is dense and Λ0⊆H2⁢(X,ℚ) has smaller rank, so H2⁢(X0,ℚ)∖Λ0 is dense. In particular, ϕ−1⁢(H2⁢(X0,ℚ)∖Λ0)⊆ℋℝ1,1 is dense. In particular, for any α, the set of t for which αt has type (1,1) is dense.

We omit the proof of the key claim because it uses the Gauss-Manin connection. The idea is that the differential of ϕℂ:ℋℂ2→H2⁢(X0,ℂ) is surjective. This is done by identifying it in terms of the Gauss-Manin connection and the differential of the period map. □

Definition 1.9.4.

A pair (X,H) of a complex manifold X and line bundle H is called a polarized complex manifold if H is ample (in particular this means X is projective). Here, H is called the polarization. A polarized family (𝒳,ℋ) over B has ℋ ample on every fiber.

Definition 1.9.5.

A family 𝒳→B of irreducible holomorphic symplectic manifolds is (locally) complete if B=Def⁢(X0)h.

Example 1.9.6.

The family of quartic K3 surfaces is locally complete.

For higher-dimensional irreducible holomorphic symplectic manifolds, it is in general very hard to construct such locally complete families. The known such constructions are EPW sextics, which are IHS fourfolds of K⁢3[n] type, fourfolds constructed by Debarre-Voisin, and various examples constructed from cubic fourfolds.

Theorem 1.9.7 (Matsushita).

Let X be irreducible holomorphic symplectic of dimension 2⁢n and f:X→B be surjective and proper with connected fibers and 0<dimB<2⁢n for B a Kähler manifold (alternatively it can be a projective variety). Then dimB=n, B is projective, b2⁢(B)=ρ⁡(B)=1, and B is Fano (alternatively B is ℚ-factorial and Fano with log-terminal singularities). Moreover, the general fiber is a complex torus and every component of a fiber is a Lagrangian subvariety.

Definition 1.9.8.

Let X be holomorphic symplectic. Then a subvariety V is called Lagrangian if dimV=12⁢dimX and for all resolutions ν:V~→V⊆X, ν∗⁢σX=0.

Proof.

Recall that if g:Y→Z is a surjective morphism of Kähler manifolds, then g∗:Hk⁢(Z,ℚ)→Hk⁢(Y,ℚ) is injective.14 This immediately gives H2,0⁢(B)=0, and thus H1,1⁢(B)=H2⁢(B), which immediately gives a rational Kähler class, so B is projective. Let H be a polarization on B. If m=dimB, then c1⁢(H)m≠0 but c1⁢(H)m+1=0, and because m<2⁢n, we know c1⁢(H)2⁢n=0, and by Fujiki, we have q⁡(f∗⁢H)=0.

Now let ω be a Kähler form on X. Because (f∗⁢H)m=[Xb], if we write L=f∗⁢H, then Lm∧ωX2⁢n−m≠0. Similarly, for k≤m, Lk∧ωX2⁢n−k≠0 because Lk=f−1⁢(H1∩⋯∩Hk). We need to prove that m=n. Here, we apply Fujiki to ω+t⁢L. This gives us

If we expand this, we obtain

Comparing coefficients in t, we observe that the coefficient of tn is nonzero, and thus n=m.

Now we prove that if Xb is smooth, then it is Lagrangian. Using the Hodge-Riemann bilinear relations, this is the same as showing that

If ω is the restriction of a Kähler form on X, then our integral simply becomes

Applying Fujiki to σ+σ¯+t⁢ω+s⁢L and using the fact that q⁡(σ,L)=q⁡(σ¯,L)=0, we obtain the desired conclusion. To conclude that all fibers are Lagrangian, we use a major result of Kollár, which says that if h:Y→Z is a proper and surjective morphism of smooth projective varieties, then Ri⁢h∗⁢ωY is torsion free for all i. In particular, for f:X→B, we see that Ri⁢f∗⁢𝒪X is torsion-free, so for σ¯∈H2⁢(𝒪X), this maps to a torsion section σ¯∈H0⁢(B,R2⁢f∗⁢𝒪X), which must vanish. Pulling back to H2⁢(Xb~,𝒪), Xb is Lagrangian. By linear algebra reasons, dimXb≤n, so f is equidimensional.

Next we prove that B is Fano. First, it is clear that H1⁢(𝒪B)=0, so Pic⁡B=NS⁡B. But then f∗:H2⁢(B)⊆H1,1⁢(X). For all α∈H2⁢(B), q⁡(f∗⁢α)=0, but because H1,1⁢(X) has signature (1,−), we see that dimH2⁢(B)=1. Thus NS⁡(B)=ℤ⁢H for some H. This implies that KB=m⁢H, and we want to show that m<0. If we consider the inclusion f∗⁢ΩB1↪ΩX1, ΩX1 is a slope-semistable bundle because c1⁢(X)=0. In particular, μ⁡(ΩB1)≤μ⁡(ΩX1)=0, so m≤0. To prove that m≠0, we see that if m≠0, then 𝒪B=f∗⁢KB↪ΩXn, so H0⁢(𝒪B)⊆H0⁢(ΩXn), and for type reasons, this is impossible. □

Remark 1.9.9.

A similar argument shows that if α∈H2⁢(X,ℤ) satisfies q⁡(α)=0, then αn≠0 but αn+1=0.

There is a result of Verbitsky that the kernel of q:S2⁢H2⁢(X,ℤ)→H∗⁢(X,ℚ) is given by ⟨αn+1∣q⁡(α)=0⟩. Also, when we proved that B is Fano, we used the following result.

Proposition 1.9.10 (Beauville).

Let X be irreducible holomorphic symplectic of dimension 2⁢n. then H0⁢(ΩX∗)=⟨σX⟩, where σX is the holomorphic form.

Idea of proof.

We consider the holonomy representation. We know that Hol⁡(g)=Sp⁡(n), where g is the hyperkähler metric. This has an action on ΩX,x0k. By compactness of X, holomorphic tensors are parallel. Coversely, parallel forms are holomorphic. But then we consider representations of Sp⁡(n) on ⋀kℂ2⁢n, but then by the representation-theoretic black box there exists a unique invariant if k is even and no invariants if k is odd. □

Remark 1.9.11.

There exists a singular definition of irreducible holomorphic symplectic varieties. This uses the algebra of reflexive holomorphic forms.

Now we want to see that the smooth fibers of a Lagrangian fibration X→B are complex tori. But it is clear that NXb/X=𝒪Xbn and that NXb/X=ΩXb1. Then we need to prove that a:X→Alb⁡(X) is an isomorphism, but we consider the sequence

and the map T→X is étale, the map T→Alb⁡(X) is surjective, and finally by considering the effect of a on H1, this is an isomorphism.

Alternatively, we may use the holomorphic version of Arnold-Liouville. This describes smooth compact fibers of a completely integrable system. If M is holomorphic symplectic and h=(h1,…,hn):M→ℂn has compact connected fibers and dh1,…,dhn linearly independent at every point, h is an integrable system if they Poisson commute. This all implies that the smooth fibers are biholomorphic to complex tori. In our case, each vector field Xhi defined by dhi=σM⁢(Xhi,−) acts infinitesimally on M and preserves the level set of hj for all j. Thus Xhi acts on each fiber. Because the fibers are compact, the action lifts to an action of ℂ on the fibers Mc. Of course, all of these actions commute, and we obtain an action of ℂn on each Mc. The orbits are open, and the fibers are connected, so there exists a unique orbit and the action is transitive. However, the kernel is discrete and has maximal rank, and thus we obtain Mc=ℂn/Λ.

In our case, the fact that [Xhi,Xhj]=0 is the same thing as our fibers being Lagrangian, and then we can just work locally.

1.10 An explicit computation

We will compute the Beauville-Bogomolov-Fujiki form for irreducible holomorphic symplectic varieties X of K3[n] type.

Proposition 1.10.1.

There exists an isomorphism of lattices

Moreover, the Fujiki constant is given by

Proof.

By deformation invariance, it is enough to perform this computation for X=S[n], where S is a K3 surface. Recall that

where h is the Hilbert-Chow morphism and we have the isomorphism H2⁢(S(n),ℤ)≃H2⁢(S,ℤ). Also, if E is the exceptional divisor of h, then 2⁢δ=c1⁢(E). Our strategy is the following:

  1. 1.

    We will prove that q|H2⁢(S,ℤ)=(−,−)S up to a constant.

  2. 2.

    We will prove that δ⟂H2⁢(S,ℤ).

  3. 3.

    We will compute q⁡(δ).

  4. 4.

    We will compute the Fujiki constant.

Let α∈H2⁢(S,ℤ). We will compute

Continuing, we have

because h⁡(E)=Δ2⁢n−2, and so the integral vanishes by dimension reasons. Now we have

for some λ∈ℤ, and because ∫α2⁢n=cn⁢q⁢(α)n=cX⁢(α,α)n, we have cX=cn⋅λn. Later, we prove that λ=1. To compute q⁡(δ), we compute

Note that h∗⁢(E|E)=−2 because E→Δ looks locally like the resolution of a quadric cone. Noting that we have a diagram

our integral becomes

On the other hand, we can differentiate the Fujiki relation twice, and we know that

Comparing the coefficients, we obtain

and after cancelling everything, we have

and therefore q⁡(δ)=−2⁢(n−1)⁢λ.15 Finally, we are forced to take λ=n−1, so we are done. □

In the case when n=2, there is a more explicit argument using intersection theory and Segre classes of the Hilbert scheme of 2 points.

Remark 1.10.2.

It is not a coincidence that q⁡(E)<0.

Definition 1.10.3.

If X is irreducible holomorphic symplectic, then E⊆X is called prime exceptional if it is integral and q⁡(E)<∞.

Theorem 1.10.4 (Markman,16 Druel).

E⊆X is prime exceptional if and only if there exists a birational f:X⇢X′ such that the strict transform E′ of E can be contracted.

Proposition 1.10.5.

Let f:x⇢X′ be a birational map between irreducible holomorphic symplectic varieties. Then the isomorphism

preserves qX′ and qX.

Proof.

If we consider the graph Γ~ of f with projections p,q, we can normalize first so that

and then we compute

□

Recall that if Z is a smooth projective variety and ℱ is a coherent sheaf, then

by Hirzebruch-Riemann-Roch. If ℱ=L is a line bundle, then we have

Proposition 1.10.6.

Let X be irreducible holomorphic symplectic. Then there exist qi∈ℚ depending only on the deformation class of X such that

Corollary 1.10.7.

For all isotropic L, we have χ⁡(X,L)=n+1, where X has dimension 2⁢n.

To prove the proposition, we need to prove the following Fujiki-like result.

Proposition 1.10.8.

Let β∈H4⁢ℓ⁢(X,ℝ) be a class that stays of type (2⁢ℓ,2⁢ℓ) for all small deformations of X. Then there exists cβ∈ℝ such that ∫β∧α2⁢n−2⁢ℓ=cβ⁢q⁢(α)n−ℓ for all α∈H2⁢(X,ℤ).

Proof of this is the same as the proof of Fujiki.

2 Moduli spaces

Our goal is to eventually prove the following theorem:

Theorem 2.0.1 (Mukai, O’Grady, Yoshioka, Huybrechts).

Let S be a projective K3 surface. Let v∈Halg∗⁢(S,ℤ)=H0⊕NS⁡(S)⊕H4 be a primitive Mukai vector. Let H be a general polarization and Mv,H be the moduli space of H-semistable coherent sheaves on S with Mukai vector

If v2≥−2, then Mv,H is a smooth projective irreducible holomorphic symplectic manifold of dimension v2+2 deformation equivalent to S[n], where n=v22+1. Moreover, if v2≥2, then there is a canonical isomorphism

which is an isomorphism of Hodge structures.

We’ll have to understand what a general polarization is, and here semistability will mean Gieseker stability, and of course we’ll need to prove smoothness, projectivity, and symplecticness. OF course just being nonempty and irreducible is a nontrivial result. This result also holds more generally for Bridgeland stability, which we will discuss in this course.

2.1 Strategy for deformation equivalence

Here is the strategy for deformation equivalence. Up to deforming S and considering birational maps among moduli spaces with different Mukai vectors, we can relate Mv,H⁢(S) to a Hilbert scheme. Because Yoshioka is notoriously hard to read, we apparently will sketch this. The missing piece is the following result:

Theorem 2.1.1 (Huybrechts).

Let f:X⇢X′ be a bimeromorphic map of irreducible holomorphic symplectic manifolds. Then X and X′ are deformation equivalent. This means that there exists a diagram

such that F is an isomorphism after restricting to Δ∗ and (ΓF)0=Γf.

This result can be found in a series of papers by Huybrechts titled:

  • •

    Birational symplectic manifolds and their deformations;

  • •

    Compact hyperkähler manifolds: basic results and erratum;

  • •

    Kähler cone…

Fix a lattice Λ of signature (3,ρ−3), where ρ=rk⁡Λ. Then the moduli space ℳΛ of Λ-marked hyperkähler varieties parameterizes pairs (X,φ) where X is irreducible holomorphic symplectic and φ:H2⁢(X,ℤ)→∼Λ is an isomorphism of lattices. Here, (X,φ)≃(X′,φ′) if there exists f:X→X′ such that φ∘f∗=φ′.

This is apparently some nonseparated complex manifold because of local Torelli. Here, we obtain a system of charts on MΛ that looks like

Lemma 2.1.2 (Tautological lemma).

Two points (X,φ),(X′,φ′)∈ℳΛ are non-separated if and only if there exist families 𝒳,𝒳′→Δ and V⊆Δ such that 0∈V¯ and 𝒳,𝒳′ are isomorphic over V with central fibers X,X′.

This result tells us that X,X′ have the same period map because the diagram

Proposition 2.1.3 (Huybrechts, Burns-Rapoport).

The tautological lemma implies that there exist families

that are isomorphic over Δ∗. Now define ΓF∗⊆𝒳×Δ∗𝒳′. The key input is a volume estimate, which says that the volume of Γt is bounded. By a result of Bishop, there exists a limit cycle Γ0⊆X0×X0′.

Let ωt,ωt′ be Kähler forms on 𝒳t,𝒳t′ varying continuously with t and consider the Kähler form pt∗ωt+pt′∗ωt′. Then we see that

This implies that

and therefore that the limit cycle Γ0⊆X×X′ exists. We want to show that there exists Γf⊆Γ0 inducing a birational map. Because the correspondences Γt∗ are isomorphic, so is Γ0∗. This also tells us that Γ0∗⁢[X′]=[X], and so Γ0 is dominant of degree 1 on both factors. We want to prove that there exists a component that is dominant of degree 1 on both components. We need to exclude the case where

where Z→X is dominant, Z′→X′ is dominant, and Yi are not dominant onto either factor. Now we have

By type reasons, the integrand vanishes on Z′,Yi and thus we only need to study the integral on Z. But now p′∗(σ′)|Z=0. This is because it comes from a smaller-dimensional subset of X′ which must be degenerate, but we know that holomorphic forms are a birational invariant, so this must be degenerate.

Remark 2.1.4.

In general, we have Γf⊊Γ0. Sometimes, even their action on H2 is different.

Proof of Huybrechts for projective case.

Let f:⇢X′ be birational. We know that X,X′ have L,L′ such that qX⁢(L)=qX′⁢(L′). Also, we know that H0⁢(X,L)=H0⁢(X′,L′). Now we use Riemann-Roch, and so there exists ai,ai′∈ℚ such that

Without loss of generality, we can assume that for t≫0, ∑ai⁢ti≥∑ai′⁢ti. For n≫0, this means that

Now choose L′ ample. If f is not an isomorphism, then L will not be nef (but it will be big). We know that

and so our map f:X⇢X′ will be induced by L (which is not nef and thus has a base locus, but is big and thus has many global sections).

Let (𝒳,ℒ) be a deformation of (X,L) such that for a very general t∈Δ, ρ⁡(Xt)=1. Now we use a theorem from the erratum of the second Huybrechts paper, which says that if X is irreducible holomorphic symplectic, then X is projective if and only if there exists L such that q⁡(L)>0. Here, we know that q⁡(L′)=q⁡(L)=q⁡(ℒt)>0, and thus ℒt is ample on Xt for very general t. This is an open condition, so we may assume ℒt is ample for t∈Δ∗. Thus for all n>0, Hi⁢(Xt,Ltn)=0.

The next claim is that π∗⁢ℒn is locally free for n≫0. Because the base is reduced, it is enough to prove constant fiber dimension. It is enough to show that t↦h0⁢(Xt,ℒtn) is constant for n≫0. But now we have

By upper semicontinuity, everything here must be equal. Taking the relative Proj construction and setting

we are done. □

2.2 Quot schemes

Let k be an infinite field with X⊆ℙkn. Alternatively, we may consider a projective morphism X→B with B quasiprojective over k. Fix p⁡(t)∈ℚ⁡[t]. We would like to construct a proper (projective?) separated scheme of finite type over k whose closed points are in bijection with isomorphism classes of sheaves on X whose Hilbert polynomial is p⁡(t). We would also like M to satisfy some universal property, namely representing some functor.

Definition 2.2.1.

A family {Fα}α∈A of (isomorphism classes of) coherent sheaves on X is called bounded if there exists a scheme S fo finite type over k and a coherent sheaf ℱ on S×X such that

Unfortunately, fixing the Hilbert polynomial is not in general sufficient to have a bounded family.

Example 2.2.2.

On ℙk1 consider the constant Hilbert polynomial p⁡(t)≡t+2. We will consider locally free sheaves on ℙ1 with this Hilbert polynomial. However, we have sheaves Fa=𝒪⁡(a)⊕𝒪⁡(−a). But then h0⁢(Fa)=a+1. This family cannot be bounded because for any scheme S and any m the subscheme

is closed, and thus S cannot be of finite type.

Example 2.2.3.

Let f:C→ℙ1 be a hyperelliptic curve with g⁡(C)≥2. Then Picg−1⁡(C) contains a theta divisor Θ, but here we have h0⁢(L)=0 generically and h0⁢(L)>0 on Θ. Then there exists a universal line bundle ℒ on Picg−1⁡(C)×C. Then we have a family of line bundles FL=f∗⁢L on ℙ1 which are 𝒪⁡(a)⊕𝒪⁡(−a−2) for a≥−1. If L∉Θ, we have FL=𝒪⁡(−1)⊕𝒪⁡(−1). However, if L∈Θ and h0⁢(L)=1, we have FL=𝒪⊕𝒪⁡(−2). This gives a nonseparated family of sheaves.

To achieve boundedness, we need to both fix the Hilbert polynomial and consider sheaves that are quotients of a fixed sheaf 𝒪XN. Our main technical tool is the following:

Definition 2.2.4.

A coherent sheaf F on ℙn is Castelnuovo-Mumford regular if for all i>0 we have Hi⁢(F⁡(−i))=0.

Theorem 2.2.5 (Castelnuovo, Mumford).

Let F be Castelnuovo-Mumford regular. Then

  1. 1.

    F is globally generated.

  2. 2.

    F⁡(ℓ) is Castelnuovo-Mumford regular for all ℓ≥0.

  3. 3.

    H0⁢(F)⊗H0⁢(𝒪X⁢(ℓ))↠H0⁢(F⁡(ℓ)).

This can be proved by inducting on the support of F and using the exact sequence

Some references are of course FGA Explained and Grothendieck’s Techniques des construction et théorème d’existence en geometrie algebrique IV: les schemas de Hilbert. Also there is a reference due to Mumford.

Theorem 2.2.6.

Fix (X,H) and some N≥0. Also fix p⁡(t)∈ℚ⁡[t] the Hilbert polynomial. Then there exists an integer ℓ0 sich that for all F a quotient of 𝒪Xn↠F, then

  1. 1.

    H0⁢(𝒪XN⁢(ℓ0))↠H0⁢(F⁡(ℓ0)) and Hi⁢(F⁡(ℓ0))=0 for all i>0.

  2. 2.

    G⁡(ℓ0) is globally generated and

    for all s≥0, where G is the kernel of 𝒪XN→F.

  3. 3.

    Hi⁢(G⁡(ℓ0+s))=0 for all s≥0 and i>0.

We now consider the functor 𝖰𝗎𝗈𝗍XN,p that takes a scheme B to the set of isomorphism classes of coherent sheaves ℱ on X×B flat over B such that pℱb⁢(t)=p⁡(t) and 𝒪X×BN↠ℱ. We will not need this, but it is possible to identify quotients if their kernels are the same as subsheaves of 𝒪X×BN.

Theorem 2.2.7 (Grothendieck).

The functor 𝖰𝗎𝗈𝗍XN,p is represented by a projective scheme QuotXN,p. Of course this means there is a universal coherent sheaf ℱ on the Quot scheme with a surjection 𝒪N↠ℱ.

Example 2.2.8.

Let r≤n be integers. Then Gr⁡(r,n) is the Quot scheme for X=pt if we consider quotients kn↠W to vector spaces of dimension n−r.

Of course, we use the Grassmannian to construct the Quot scheme, so we will construct the Grassmannian by hand. If [V]∈Gr⁡(r,n), we can choose a splitting kn=V⊕W. Then the open affines are given by Hom⁡(V,W) given by associating a map φ to its graph Γφ. There is also the tautological sheaf 𝒮 and exact sequence

and of course we know that H0⁢(det⁡𝒮∨)=⋀∨kn.

Our strategy is to define the Quot scheme as a set, then slap a scheme structure on it and prove that it is locally closed, and finally we prove that the universal family exists.

Let ℓ0 be as in the previous theorem, n=dimH0⁢(𝒪XN⁢(ℓ0)), and r=p⁡(ℓ0). We will consider Gr⁡(n−r,r). Given 𝒪XN↠F with pF⁢(t)=p⁡(t), let G=ker⁡(𝒪XN↠F). Then we know

and because the higher cohomology of G⁡(ℓ0) vanishes, the last arrow is surjective. We know that the middle term has dimension n and the last term has dimension r, so Kℓ≔H0⁢(G⁡(ℓ0)) has dimension n−r.

First we need to prove that this assignment is injective because Kℓ0 determines G and hence F. Indeed, by the theorem, we have a series of surjections

Thus it determines G as a graded module, and so in fact G is determined as a coherent sheaf. Alternatively, we can consider Kℓ0⊗𝒪X→𝒪XN⁢(ℓ0), and this surjects onto G. We now define the set

We know that φs,k is the fiber of a map of vector bundles. We know 𝒮⊆H0⁢(𝒪XN⁢(ℓ0))⊗𝒪Gr and so we have a map

Now the locus Qs where the rank of φs is p𝒪XN⁢(ℓ0+s)−p⁡(ℓ0+s) is localy closed.

Theorem 2.2.9.

Fix X,H,N,p,ℓ0. Then there exist finitely many Hilbert polynomials of quotients 𝒪XN↠F such that G⁡(ℓ0) is globally generated by n−r global sections.

Proof.

Consider Gr⁡(n−r,n)≕Gr. We of course have the tautological sequence and the two projections on Gr×X. Then we have

Call the kernel of the last morphism 𝒢⁡(ℓ0). By the flattening stratification and Noetherian induction, we have only finitely many possible Hilbert polynomials. □

Now let ℱ be as in the proof of the theorem and Q1,…,Qm be the possible Hilbert polynomials. Then let ℓ0′ be the maximum of the ℓ0⁢(Qi). For all s such that ℓ0+s≥ℓ0′, we know that

for all i>0. Thus pℱk⁢(ℓ0+s)=h0⁢(ℱk⁢(ℓ0+s)). Also, we know that Hi⁢(Gk⁢(ℓ0+s))=0, and thus

is surjective. Finally, using the original theorem, the map

is surjective.

Let γ∈ℕ be large enough such that if Qi⁢(t)=p⁡(t) for γ values, then Qi⁢(t)≡p⁡(t). We will consider K∈⋂s=1γQs+ℓ0=Q. Here, we observe that

has the correct rank for these γ values. Therefore ℱK has the correct Hilbert polynomial, and so we are done.

It remains to prove that Quot represents the functor and that it is a closed subscheme of Gr. First, we will construct a universal family on Quot. Recall on the Grassmannian that we have the tautological sequence

Now restricting to Quot and pulling back to Quot×X, we have

Now observe that ℱ is flat over Quot if and only if p1⁣∗⁢ℱ⁢(ℓ) is locally free for ℓ≫0. We will see that

gives the universal quotient on Quot×X. But the flatness condition follows from the vanishing result for ℓ≥ℓ0. Using the diagram (2.1), note that

recovers

We will now check the universal property of our family. Consider a quotient 𝒪X×BN↠ℱB. We want φ:B→Quot such that

But here we push forward p2⁣∗⁢(𝒪X×BN→ℱB)⊗𝒪⁡(ℓ0) and we now have

But now the first factor is locally free of the correct rank, and so we have a map B→Gr factoring through Quot.

Finally we prove that Quot⊆Gr is a closed immersion. Here, we will use the valuative criterion. Let R be a discrete valuation ring over k with fraction field K. Let U=Spec⁡K⊆Spec⁡R=C. We want to fill in the diagram

On U×X we have an exact sequence

and on C we have the exact sequence

We know ℱ~C is flat over C if and only if t:ℱ~C→ℱ~C is injective. Set ℱC≔ℱ~C/[[t∞]]. Then we have

But now ℱC is flat.

Remark 2.2.10.

We can replace 𝒪XN with any sheaf ℋ on X and consider quotients ℋ↠ℱ. Then Quotℋ,pX⊆Quot𝒪XN,pX is a closed subscheme. Here, we may assume that ℋ is a quotient of 𝒪XN for some large enough N. This also shows that HilbX⊆Hilbℙn is a closed subscheme for X⊆ℙn.

Remark 2.2.11.

We may also replace X/k with X→S a projective morphism with S quasi-projective. Then QuotX/S→S is a projective morphism. Also, we know (QuotX/S)s=QuotXs.

Wew will now compute the tangent space of the Quot scheme. Here we are following chapter 6 of FGA explained. We want to compute

Theorem 2.2.12.

The deformation functor Def𝒪XN↠F has a tangent-obstruction theory given by

where G=ker⁡(𝒪XN↠F).

We will prove the tangent part of the theorem. Consider the exact sequence

We have the diagram

We want G′,F′ flat over k⁡[ε] filling in the diagram. Remember flatness is equivalent to injectivity of multiplication by ε. First we define e∈ExtX1⁡(G,F). If e=0, then we can fill in the diagram.

Remark 2.2.13.

Let A→B→C be a split exact sequence. Then the splittings are the same as Hom⁡(C,A).

Therefore, the possible ways to fill the diagram are the same as splittings of e, which are the same as Hom⁡(G,F). We know that Im(α)⊆ker⁡(β)→G. In particular, we have an exact sequence

We will define e∈ExtX1⁡(G,F) to be the class of this extension. We will omit checking that multiplication by ε is zero on this exact sequence.

If e=0, then choose some splitting ξ:G→ker⁡(β)/Im(α). We will construct a filling of the diagram. We will define G′ by the diagram

We will define G′ to be the preimage of ξ⁡(G). This gives us G⁡(ε)↪G′↠G and of course F′ with F⁡(ε)→F′→F. Checking flatness is omitted.

Conversely, given G′,F′ filling the diagram, we know that G=G′/Im(α)⊆ker⁡(β)/Im(α), and this splits the surjection ker⁡(β)/Im(α)↠G. It is not hard to check that these two constructions are inverses of each other.

2.3 Semistable sheaves

Given p⁡(t)∈ℚ⁡[t], we want to know which sheaves F with pF=p are quotients of a given sheaf.

Remark 2.3.1.

A family {Fα} of sheaves is bounded if and only if there exists a sheaf ℋ such that ℋ↠Fα and there are only finitely many Hilbert polynomials.

Let X⊆ℙn be projective and F coherent with Supp⁡F⊆X.

Definition 2.3.2.

F is pure of dimension d if for all nonzero subsheaves E⊆F, dimSupp⁡E=d.

Note that this is equivalent to saying that all associated points of F have dimension d.

Example 2.3.3.

If Z⊆X is a subscheme of dimension 0, then 𝒪Z has pure dimension 0. Also, if j:C→X is a curve and L is a line bundle on C, then j∗⁢L is pure of dimension 1.

Example 2.3.4.

Let X be integral and dimX=d. Then F is pure of dimension d if and only if F is torsion-free.

Now define the torsion filtration of F as

where Ti⁢(F) is the maximal subsheaf of dimension at most i. Note that F/Td−1⁢(F) is pure of dimension d.

Let (X,H) be a polarization with X projective over k=k¯. If E is a coherent sheaf, recall that

where Hi∈|H| are generic. We know α0=χ⁡(E). If d=dimE, then αd⁢(E)>0.

Remark 2.3.5.

If E=𝒪X, then αd⁢(𝒪X)=Hd is the degree of X with respect to H. We may now define the generalized rank of E by αd⁢(E)αd⁢(𝒪X). If X is not irreducible, the generalized rank of E may depend on the polarization.

Remark 2.3.6.

Using Grothendieck-Riemann-Roch, if X is integral, then

Definition 2.3.7.

The reduced Hilbert polynomial p⁡(E) of E is defined by

If we consider the lexicographic ordering on polynomials, we can make the following definition:

Definition 2.3.8.

A coherent sheaf E on X of dimension dimE=d is called (semi)stable if E is pure and for all F⊊E, p⁡(F)<(≤)⁢p⁢(E).

Remark 2.3.9.

IT is enough to check semistability for so-called “saturated” subsheaves, which are those F⊆E with E/F pure of dimension d. Indeed, if F⊆E→G and G is not pure, we can consider T=Td−1⁢(G) and then G/T is pure. We can then replace F with F′, called the saturation of F in E. Because saturating F increases the Hilbert polynomial because αd⁢(F)=αd⁢(F′) and αd−1⁢(F)≤αd−1⁢(F′), we can check semistability for saturated subsheaves.

Lemma 2.3.10.

It is enough to check pure quotients. Namely, for all E↠G pure of dimension d, E is (semi)-stable if and only if p⁡(E)<(≤)⁢p⁢(G).

We can check that

Proposition 2.3.11.

Let F,G be pure of dimension d and semistable.

  1. 1.

    If p⁡(F)>p⁡(G), then Hom⁡(F,G)=0;

  2. 2.

    If p⁡(F)=p⁡(G) and there exists a nonzero f:F→G, then if F is stable, f is injective. If G is stable, then f is surjective.

Remark 2.3.12.

Even when X is integral, stability depends on H in general.

Corollary 2.3.13.

If E is stable and k=k¯, then E is simple (which means End⁡(E)=k). Otherwise, End⁡(E) is a finite-dimensional division algebra over k.

Proof.

Consider the image E of f. By semistability, we have p⁡(F)≤p⁡(E)≤p⁡(G), and so a nonzero map cannot exist. For the second part, if F is stable, consider the image E⊆G again. If f is not injective, then p⁡(F)<p⁡(E), and this is impossible. □

Example 2.3.14.

Let C be a curve of genus g≥1 and L a line bundle of degree d. Consider e∈H1⁢(𝒪C)=Ext1⁡(L,L). Then we have 0→L→E→L→0, and E is strictly semistable.

Example 2.3.15.

Let E be a vector bundle on D such that (deg⁡E,rk⁡E)=1. Then if E is semistable, then E is stable.

2.3.1 Some filtrations

We will now discuss the Harder-Narasimhan filtration

Lemma 2.3.16.

Let E be a coherent sheaf pure of dimension d. Then there exists F⊆E such that for all G⊆E, p⁡(F)≥p⁡(G) and if equality holds, then G⊆F. Moreover, F is semistable and uniquely determined.

Definition 2.3.17.

The F in the lemma is called the maximal destabilizing subsheaf of E.

Proof.

Consider the ordering on the set of pairs (F⊆E,p⁡(F)) where (F,p⁡(F))≤(F′,p⁡(F′)) if F⊆F′ and p⁡(F)≤p⁡(F′). This is clearly a nonempty set and every ascending subsequence of subsheaves has a maximal element. Among all maximal elements, choose one with minimal αd>0. We will prove that F is the desired maximal destabilizing subsheaf.

If not, there exists G⊆E such that p⁡(G)≥p⁡(F). If G⊈F, then F⊊F+G and G∩F⊊G. By the maximality of F, we can asssume that p⁡(F)>p⁡(F+G). We have the exact sequence

But now we have p⁡(F+G)<p⁡(F) by assumption, but this means that p⁡(F∩G)>p⁡(G)≥p⁡(F), and this is a contradiction. Thus every G′⊆E with p⁡(G′)>p⁡(F) defines a subsheaf G⊆F with p⁡(G)>p⁡(F).

Now let G⊆F be a maximal subsheaf with p⁡(G)>p⁡(F). Let G⊆G′⊆E be maximal with respect to the ordering. By maximality of G′, we know that p⁡(G′)≥p⁡(G)>p⁡(F). We will now prove that G′⊈F. By αd-minimality of F, if G′⊆F, then αd⁢(G′)<αd⁢(F), and thus G′⊈F by contradiction. This implies that F⊊F+G′ and G′∩F⊊G′. By maximality of F, p⁡(F)>p⁡(F+G′). We assumed that p⁡(F)<p⁡(G′), and thus p⁡(G′∩F)>p⁡(G′)≥p⁡(G), contradicting the maximality of G in F. □

Definition 2.3.18.

Let E be pure of dimension d. A Harder-Narasimhan filtraton of E is an increasing filtration

such that

  1. 1.

    The graded pieces H⁢Ni/H⁢Ni−1 are semistable with reduced polynomials pi.

  2. 2.

    We have p1>p2>⋯>pℓ.

Example 2.3.19.

If E is semistable, then 0⊆E1=E is a Harder-Narasimhan filtration.

Example 2.3.20.

Consider L1,L2 line bundles of degrees d1≥d2. Then Ext1⁡(L2,L1)=H1⁢(L1⊗L2∨)≠0. Considering a nontrivial extension

note that E is unstable and 0⊆L1⊆E is a Harder-Narasimhan filtration.

Theorem 2.3.21.

Let E be pure of dimension d. Then there exists a unique Harder-Narasimhan filtration.

Proof.

We induct on αd⁢(E). We may assume that E is unstable. Otherwise, let E1⊆E be the maximal destabilizing subsheaf. Then we consider E/E1 and αd⁢(E/E1)<αd⁢(E). By induction, we have a Harder-Narasimhan filtration

such that p⁡(G1)>p⁡(G2/G1)>⋯⁢p⁡(Gk/p⁡(Gk−1)). Set Ei+1=π−1⁢(Gi) under π:E→E/E1. But now we have p⁡(E2/E1)=p⁡(G1),p⁡(E3/E2)=p⁡(G2/G1),…, and this gives us

We only need to check that p⁡(E1)>p⁡(E2/E1). By the maximality of E1, we know that p⁡(E1)>p⁡(E2), and this gives the desired result.

Now let E∙,E∙′ be two Harder-Narasimhan filtrations. Then we know E1,E1′⊆E, and we may assume that p⁡(E1′)≥p⁡(E1). Let j be the minimum integer such that E1′⊆Ej. But now the map E1′→Ej/Ej−1 is nonzero, and source and target are both semistable, so p⁡(E1′)≤p⁡(Ej/Ej−1)<p⁡(E1), and the last inequality holds unless j=1. By assumption, we know E1′⊆E1 and p⁡(E1)=p⁡(E1′). Reversing the argument, we see that E1⊆E1′, so E1=E1′. This proves uniqueness of the maximal destabilizing subsheaf, and by induction, the two filtrations on E/E1 are the same, so E∙=E∙′. □

We will now move on to another filtration:

Definition 2.3.22.

Let E be a semistable sheaf pure of dimension d. Then a Jordan-Hölder filtration for E is an increasing filtration

such that gri⁢(E)=Ei/Ei−1 is stable with p⁡(gri⁢(E))=p⁡(E).

Example 2.3.23.

If 0→L→E→L→0 is a nontrivial extension of a line bundle L by itself, this gives a Jordan-Hölder filtration. Also, E=L⊕n shows that the filtration is not unique in general.

Proposition 2.3.24.

Jordan-Hölder filtrations exist and

is unique up to isomorphism.

Proof.

We induct on αd⁢(E)>0. If E is stable, we are done. Otherwise, there exists F⊆E such that p⁡(F)=p⁡(E) and αd⁢(F)≤αd⁢(E). This inequality must be strict because otherwise pF⁢(t)=pE⁢(t) and thus F=E. Therefore, there exists a subsheaf F⊆E with p⁡(F)=p⁡(E) and minimal αd⁢(F), and this must be stable. By induction, E/F is semistable with the same reduced Hilbert polynomial, and so by induction, it has a Jordan-Hölder filtration. Pulling back to E, we obtain the filtration on E.

The proof of uniqueness is omitted.1 The outline is similar to the case of the Harder-Narasimhan filtration, but the actual argument is different. □

2.3.2 Boundedness

We will prove boundedness of semistable sheaves with a fixed Hilbert polynomial in the case of a smooth projective surface (because we will be using this result for K3 surfaces only). Note that this proof will only work for surfaces.

Recall that for a smooth projective surface X with polarization H and a sheaf E of rank r, Riemann-Roch says that

This implies that

Thus the coefficients of p⁡(E) are determined by c1⁢(E)⋅Hr and χ⁡(E)r if E is pure of dimension 2.

Definition 2.3.25.

Let E be pure of dimension 2. Then define the slope of E by

This gives a notion of slope-stability. It is a fact that slope-stability implies Gieseker stability which implies Gieseker semistability, which implies slope-semistability.

Remark 2.3.26.

There exists a maximal destabilizing subsheaf F⊆E, and we denote μ⁡(F)≕μmax⁢(E).

Remark 2.3.27 (Rudakov).

We can define stability for any abelian category 𝒞 with a preorder ≤. Then we can define stability with respect to ≤, and in this case we obtain a Schur Lemma. Under certain finiteness assumptions (where 𝒞 is weakly Artinian or weakly Noetherian with respect to ≤), we obtain Harder-Narasimhan filtrations or Jordan-Hölder filtrations.

Example 2.3.28.

King-stability for quiver representations is an example of such a stability condition.

Theorem 2.3.29.

Let (X,H) be a smooth projective polarized surface and p⁡(t)∈ℚ⁡[t]. Then there exist N,ℓ only depending on (X,H) and p⁡(t) such that any slope-semistable F is a quotient 𝒪XN⁢(−ℓ)↠F.

Remark 2.3.30.

Note that twisting by 𝒪X⁢(ℓ⁢H) does not affect (semi)-stability. Therefore the theorem gives boundedness of slope-semistable sheaves with fixed Hilbert polynomial (because they live in the Quot scheme).

We will actually prove a slightly stronger result which allows us to use induction.

Theorem 2.3.31.

Let (X,H) and p⁡(t)∈ℚ⁡[t] as above and fix μ∈ℚ. Then there exists N,ℓ only depending on (X,H),p,μ such that for all c≥0 and any pure dimension 2 sheaf F with Hilbert polynomial pF=p+c and such that μmax⁢(F)≤μ, 𝒪XN↠F⁡(ℓ). Moreover, the set of possible c (and hence of possible pF) is finite.

Proof.

We will induct on the rank of F. Let F be such that pF=p+c for c≥0 and μmax⁢(F)≤μ. First we find a uniform ℓ such that h0⁢(F⁡(ℓ))≠0. Using Riemann-Roch, we will have

By Serre duality, h2⁢(F⁡(ℓ))=dimHom⁡(F,KX⁢(−ℓ)).

First, we claim that there exists a uniform ℓ such that dimHom⁡(F,KX⁢(−ℓ))=0. If we consider

where E has rank 1, then c1⁢(E)⋅H≤(KX−ℓ⁢H)⋅H, and thus

In particular, μ⁡(G)≤μmax⁢(F)<μ, so we can fix ℓ such that h0⁢(F⁡(ℓ))≠0. If we consider

we can saturate this and obtain

We know that ℋ is rank 1 and torsion free, so ℋ=ℒ⊗ℐZ, where dimZ=0 and ℒ is a line bundle. This is because 0→ℋ↪ℋ∨⁣∨, and reflexive sheaves are locally free in codimension 3 and the cokernel is supported in codimension 2.

Now we have h0⁢(ℋ⁡(ℓ))≠0, and thus H0⁢(ℒ⁡(ℓ))≠0. Write ℒ⁡(ℓ)=𝒪⁡(D) with D effective. Then

so there are only finitely many possibilities for c1⁢(D) (because we have bounded D⋅ample). In particular, this means there are only finitely many possibilities for c1⁢(G) and thus for μ⁡(G). In fact, we can uniformly bound μmax⁢(G) because for any G′⊆G, we can produce

and μ⁡(F′) is bounded.

Now G has rank r−1 and slope μmax⁢(G)≤μ′, where μ′ is uniform. Thus

for some p′,c′. By induction, there exist uniform N′,ℓ′ such that 𝒪XN′↠G⁡(ℓ′). Moreover, there exist finitely many possible c′ and thus finitely many deg⁡Z. Applying the theorem to ℋ (that there are finitely many c1⁢(ℋ) and deg⁡Z), we have a surjection 𝒪XN′′↠ℋ⁡(ℓ′). Because such ℋ form a bounded family, we can assume that H1⁢(ℋ⁡(ℓ′))=0. Thus we have

Now we lift the N′ sections on the right and consider the N′′ sections on the right, and we get 𝒪XN′′′↠F⁡(ℓ′).

It remains to prove the rank 1 case. We will prove that give (X,H),p⁡(t) (here μmax=μ because we are in the rank 1 case), there exist N,ℓ such that if ℋ has rank 1 and pℋ=p+c for c≥0, then 𝒪XN↠ℋ⁡(ℓ)→0. We know that ℋ=ℒ⊗ℐZ. By Riemann-Roch, there exists ℓ such that H0⁢(ℋ⁡(ℓ))≠0. Fixing such an ℓ, we know that

and so there are only finitely many possibilities for c1⁢(ℋ). Using Riemann-Roch again, we know

and therefore there are only finitely many possible deg⁡Z. □

Proposition 2.3.32.

The conditions of Gieseker (semi)stability and slope-semistability are open in families.

Sketch.

Consider ℱ on X×B. Then consider a destabilizing quotient ℱb↠G. Suppose you prove that there exist only finitely many possible Hilbert polynomials pi. We then consider the relative Quot schemes

which are proper over B. Studying this morphism, we obtain the desired result. □

2.4 Moduli of sheaves

Definition 2.4.1.

Two pure sheaves F,F′ of dimension d on (X,H) are called S-equivalent if the associated graded pieces grGH⁢(F)≃grJH⁢(F′) are the same. Here, the “S” apparently stands for Seshadri.

Definition 2.4.2.

A polystable sheaf is a direct sum ⨁Fi, where the Fi are stable and p⁡(Fi)=p⁡(Fj) for all i,j.

Note that we cannot separate S-equivalent sheaves. More precisely, consider the following example.

Example 2.4.3.

Suppose that F1,F2 are stable with p⁡(F1)=p⁡(F2) and suppose that there exists a nontrivial extension

Clearly F is semistable. We will show that there exists a flat family ℱ on 𝔸1×X such that for all t≠0, ℱt=F and for t=0, ℱ0=F1⊕F2. This will tell us that if there is a separated moduli space, then F1⊕F2 and F must be the same closed point.

Consider the second projection p2:𝔸1×X→X. Then if we consider i0:{0}×X↪𝔸1×X, the exact sequence

will define the desired family ℱ.

By boundedness of semistable sheaves with a fixed Hilbert polynomial, there exists m such that all such cheaves are m-regular (they become Castelnuovo-Mumford regular after twisting with 𝒪⁡(m)). We will now denote αd⁢(F) by r, and now all sheaves with 0<r′<r and the same reduced Hilbert polynomial are also m-regular. We will now fix m and set N=p⁡(m). Then we have a surjection 𝒪XN⁢(−m)↠F. We may now consider the Quot scheme

Recall that the map to the Grassmannian was determined by

which gave H0⁢(G⁡(ℓ))⊆H0⁢(𝒪XN⁢(ℓ−m)). In particular, we have

and so ℒℓ≔det⁡p2⁣∗⁢ℱ⁡(ℓ) is very ample. Now write ℋ=𝒪XN⁢(−m) for shorthand. Then ρ∈Quot is really a map ℋ→F. Now define

and consider Rs⊆R the locus of stable F. Let V≔kN. We know that R is preserved by the natural action of G⁢L⁢(V) because g acts by

and so g∘ρ=ρ∘g.

Lemma 2.4.4.

Let ρ∈Quot such that F⁡(m) is globally generated. Then if h0⁢(ρ⁡(m)) is an isomorphim, the stabilizer of ρ is actually Aut⁡(F)⊆G⁢L⁢(V).

We need to understand this G⁢L⁢(V)-action and why there is an natural linearization. On G⁢L⁢(V), there exists a “universal automorphism” τ:V⊗𝒪G⁢L⁢(V)→V⊗𝒪G⁢L⁢(V). On G⁢L⁢(V)×Quot, we have

and this gives another quotient, and thus defines a map σ:G⁢L⁢(V)×Quot→Quot. By the universal property, we have p2∗⁢ℱ=σ∗⁢ℱ and thus a linearization of the action. Because all natural constructions of G⁢L⁢(V)-linearized sheavs are naturally G⁢L⁢(V)-linearized, we know det⁡p1⁣∗⁢ℱ⁡(ℓ)=ℒℓ is also linearized. Now we have R¯ with the linearized line bundle ℒℓ for G=S⁢L⁢(V), so we can form the GIT quotient.

Theorem 2.4.5.

For ℓ sufficiently large,

  1. 1.

    ρ∈R¯ is GIT semistable if and only if F is semistable;

  2. 2.

    G⁢L⁢(V)⋅F¯∩G⁢L⁢(V)⋅F′¯≠∅ if and only if F,F′ are S-equivalent.

  3. 3.

    F is polystable if and only if S⁢L⁢(V)⋅ρ⊆R is closed.

Recall that for G a reductive group acting on projective X and a G-linearized ample line bundle L on X, the GIT quotient X⫽G is given by Xs⁢s⁢(L)/G. Recall that points of X⫽G correspond to equivalence classes of orbits and that X⫽G is projective. Also recall that one can explicitly determine GIT semistability using the Hilbert-Mumford criterion. This says that x∈X is semistable if and only if for all λ:𝔾m→G, the number m⁡(x,λ)≥0. This m⁡(x,λ) is determined by considering limt→0λ⁡(t)⋅x≕x0 and considering the weight of 𝔾m on the fiber above x0. Then m⁡(x,λ) is defined to be −weight.

Now consider ρ:V⊗𝒪X⁢(−m)→F∈R¯. Let λ:𝔾m→S⁢L⁢(V) be a one-parameter subgroup. This is determined by its weight decomposition V=⨁Vn. Define a filtration on V by V≤n=⨁m≤nVm. This determines a filtration

Lemma 2.4.6.
  1. 1.

    We have limt→0λ⁡(t)⁢ρ=ρ0, where

  2. 2.

    The weight of 𝔾m on (ℒℓ)ρ0 is given by ∑n⁢pFn⁢(ℓ).

Proof.

For ℓ≫0, formation of ℒℓ commutes with base change, and so

Assuming that there is no higher cohomology, this has dimension dimH0⁢(Fn⁢(ℓ))=pFn⁢(ℓ). But now the action of 𝔾m here is given by n⋅h0⁢(Fn⁢(ℓ)).

We will now sketch the first part. Consider the map 𝔾m→Quot given by λ⁡(t)⋅ρ. By properness, this extends to a map 𝔸1→Quot. Now ρ:V⊕𝒪X⁢(−m)→F defines a sheaf F⊗k⁡[t,t−1], and we want to extend this to 𝔸1.

First, define

This is the same as V⊗k⁡[t], where v⊗1↦v⊗tn if v∈Vn. Now multiplication by t on each piece is just the natural inclusion V≤n→V≤n+1. Thus the fiber of 𝒱 above 0 is naturally the direct sum ⨁V≤n+1/V≤n. Similarly, we define the sheaf

on 𝔸1×X, and it is easy to see that ℱ0=⨁Fn. □

We now assume that ∑n⁢dimVn=0. This tells us that

Now for all V′⊆V, define F′=ρ⁡(V′⊗𝒪X⁢(−m))⊆F, and for all F′⊆F, define V′=ρ−1⁢H0⁢(F′⁢(m)). We now want to reformulate the Hilbert-Mumford criterion.

Lemma 2.4.7.

ρ∈R¯ is GIT semistable if and only if for all V′⊆V, ϑ⁡(V′)≥0, where

The next step is to translate this in terms of subsheaves F′⊆F and their Hilbert polynomials. Here, we want

By a theorem of Le Potier characterizing semistability, we consider h0⁢(F⁡(m))r.

Theorem 2.4.8 (Le Potier).

Fix p∈ℚ⁢[t]d and a multiplicity r. Then there exists m≫0 such that for all sheaves F pure of dimension d with αd⁢(F)=r and pF=r⁢p, the following are equivalent:

  1. 1.

    F is semistable.

  2. 2.

    r⁢p⁢(m)≤h0⁢(F⁡(m)) and for all F′⊆F, h0⁢(F′⁢(m))r′≤p⁡(m).

Modulo actually proving anything, we now have semistability. It remains to discuss the orbit closures. Consider a semistable (F,ρ). Then we know that (JH⁡(F),JH⁡(ρ))∈S⁢L⁢(V)⋅ρ¯. It is enough to show that orbits of polystable sheaves are closed in R.

Lemma 2.4.9.

Consider a sheaf ℰ on C×X flat over C and suppose that F=⨁Fini is polystable. Suppose that ℰt=F for t≠0 and that ℰ0 is semistable. Then ℰ0=F.

To prove this lemma, consider the upper semicontinuous function t↦hom⁡(Fi,ℰt). But then the map

is injective, and this gives us the desired inequality.

Now consider Rs⊆R. We know that Rs⫽S⁢L⁢(V)⊆R⫽S⁢L⁢(V)≕M is open, and that M is projective by GIT quotient. We know that Rs→Rs⫽S⁢L⁢(V) is a geometric quotient and that R→R⫽S⁢L⁢(V) is a good quotient. Also, closed points are the same as S-equivalence classes of semistable sheaves while points of Rs are isomorphism classes of stable sheaves. Now recall the functor ℳ:𝖲𝖼𝗁/kop→𝖲𝖾𝗍 sending a scheme S to isomorphism classes of F on S×X flat over S with the correct Hilbert polynomial and semistable. Here, F∼F′ is there exists L∈Pic⁡(S) such that F=p1∗⁢L⊗F′.

Theorem 2.4.10.

The moduli spaces M (resp. Ms) universally corepresent ℳ (resp. ℳs).2

Remark 2.4.11.

If there exist strictly semistable sheaves, then M is not a fine moduli space for ℳ.

We now try to salvage existence of a universal family on Ms (which is not always possible).

Definition 2.4.12.

A sheaf ℰ on Ms×X flat over Ms is a universal family if for all S-flat families of stable sheaves F with Hilbert polynomial p, there exists L∈Pic⁡S such that ϕF∗⁢(ℰ)=F⊗p1∗⁢L. The family is quasi-universal if there exists a locally free W such that ϕF∗⁢(ℰ)=F⊗p1∗⁢W.

Now consider a coherent sheaf ℰ on Rs. This descends to Rs⫽G⁢L⁢(V) if and only if the action of 𝔾m⊆G⁢L⁢(V) is trivial on ℰx for all x∈Rs. We know that on Rs×X, we have the universal sheaf ℱ~. Clearly this does not descent because of the morphism Aut⁡(F)→G⁢L⁢(V), so the action is nontrivial. For ℓ≫0, the sheaf

is locally free and 𝔾m-linear if weight 0. Thus the sheaf Hom⁡(p1∗⁢Aℓ,ℱ~) descends to Ms×X. We claim that this is a quasi-universal family.

2.4.1 Determinantal line bundles

Let X be smooth and projective and let K⁡(X) be the Grothendieck group of coherent sheaves on X. Of course, this is a ring, and it has a quadratic form

We may also consider the numerical Grothendieck group Knum⁢(X) by quotienting out by the kernel of the quadratic form.

Now suppose that X→Y is a smooth projective morphism of relative dimension n and F is coherent on X and flat over Y. Then there exists a locally free resolution F∙→F of length n which computes higher direct images. What we mean by this is that

Also, we have ℋn−i⁢(Fn⁢f∗⁢F∙)=Ri⁢f∗⁢F. Then we have

Now consider ℰ on S×X flat over S and consider λℰ:𝒞⁢ℴ⁢𝒽⁢(X)→Pic⁡S given by

λℰ satisfies certain properties. The most important for us right now concerns a rank r locally free sheaf W on S. Here, we have

where c is the class of ℰs, for all u∈K⁡(X). This gives us a map

For a K3 surface, we can do this on integral cohomology, and then H2⁢(Mv,ℤ)=v⟂ as Hodge structures.

2.4.2 Mukai’s theorem

Theorem 2.4.13.

Let X be projective over k, where k is algebraically closed and of characteristic 0. If F is a stable sheaf on X and M is the moduli space containing F, then 𝒪̂M,F pro-represents DefF. Moreover,

is the trace map. In adddition, for any small extension

we have tr⁡(ob⁡(FA,A~,A))=ob⁡(det⁡FA,A~,A). Also, DefF has a tangent-obstruction theory given by T1=Ext1⁡(F,F) and T2=Ext2⁡(F,F)0.

Corollary 2.4.14.

If F is stable and X is a surface with ωX=𝒪X, then M is smooth at [F]∈M.

Proof.

Here, we know that Ext2⁡(F,F)≃H2⁢(X,𝒪X) is 1-dimensional, ans thus the trace morphism has no kernel, so deformations are unobstructed. □

Proof of theorem.

The result about the tangent-obstruction theory of a coherent sheaf is due to Maruyama, Mukai, and Artamkin, and there is a generalization of Lieblich for complexes and for a morphism X→S.

Next, if F is stable, then DefF is pro-representable. If F is only polystable, then DefF has a hull. Here, we have a map DefF→𝒪̂M,F and Quot⊇Rs→M2, and then we use Luna’s étale slice. First, there there exists F∈V⊆Quot that is Aut⁡(F)-invariant such that V⫽Aut⁡(F) is locally isomorphic to (M,F), and this constructs an inverse when F is stable.

We now want to consider the trace map. If E is locally free, then Exti⁡(E,E)=Hi⁢(End⁡(E)), and this clearly has a trace map to Hi⁢(𝒪X). We also have the inclusion

and this induces multiplication by the rank of E on cohomology. For any E, choose a locally free resolution E∙→E. Then

and now we have the maps

Now we want to compute deformation of F on k⁡[ε]. Here, we have a short exact sequence

But now there exists a splitting k→k⁡[ε], and so we have an element of ExtX1⁡(F,F). Conversely, given any extension

we declare the multiplication by ε to be ℱ→F→ℱ.

We now consider obstructions. Up to tensoring by 𝒪⁡(m), we can assume that Hi⁢(F)=0 for all i>0. Then declare V=H0⁢(F) and ℋ=V⊗𝒪X. Then we have a short exact sequence

Because ℋ is locally free and X is smooth, then d⁢h⁢(G)=max⁡{0,d⁢h⁢(F)−1}. Applying Hom⁡(−,F), we now have

Becuase Ext1⁡(ℋ,F)=0 and Ext1⁡(G,F) is the obstruction space for the Quot scheme, we have the desired expression for the obstruction. We will in fact give a different proof. For

we have the following diagram:

Given FA, we know that ob⁡(FA,A~,A)∈Ext2⁡(F,F⊗kI)=Ext2⁡(F,F)⊗I. Tensoring by FA, we now have

Now we know that

and

Applying Hom⁡(F,−), we now obtain

This tells us that the obstruction ob⁡(FA,A~,A)=fFA∪eFA.

Lemma 2.4.15.

If E is locally free, then ob(EA,A~,A)∈ker(tr:H2(End(E))→H2(𝒪X)).

To prove this, for some EA~, we get {Gα⁢β}∈H1⁢(End⁡(E)⊗A~) such that

and we can construct the obstruction by hand.

Now denote

We claim there exists ψ∈Ext1⁡(G,F) such that

and

We know that tr⁡(θ∪ψ)=0 if and only if tr⁡(ψ∪θ)=0. □

We will now construct a holomorphic form on the moduli space.

Theorem 2.4.16.

Let X be a K3 surface or an abelian surface. Then Ms is smooth and has a holomorphic symplectic form coming from

This is in fact induced from Serre duality.

The first step is to prove that there exists an isomorphism of sheaves

where Extf1⁡(F,−)=Ri⁢f∗∘Hom⁡(F,−). We need to consider the Kodaira-Spencer map and the Atiyah class. Given a coherent sheaf F on Y, we have

Considering p1⁣∗⁢(−)⊗p2∗⁢F, the Atiyah class of F is

Now given Ai⁢(F)∈Exti⁡(F,F⊗ΩYi), we will see that tr⁡(Ai⁢(F))=chi⁢(F).

Given F on S×X flat over S, we consider

Note that ΩS×X1=p1∗⁢ΩS1⊕p2∗⁢ΩX1. Now we can define K⁢SF:TS→Extp11⁡(F,F).

The next step is to see that if F is a family of stable sheaves on S×X, then p1⁣∗⁢Hom⁡(F,F)=H0⁢(𝒪X)⊗𝒪S. Moreover, we have a series of isomorphisms

We now apply this on the Quot scheme and descend, and we deduce that

Before we continue, we review some things about correspondences. Let X be smooth and projective. Suppose Γ∈CHk⁢(S×X), where S is smooth and quasiprojective. Then there exists a lift

But then there is a cycle map to H2⁢k⁢(S¯×X), and we obtain a cohomology class [Γ¯]. Then we obtain a map

Because [Γ¯] is a Hodge class, this is a morphism of Hodge structures. Recall that by the Künneth formula, we have

and by Poincaré duality we have H2⁢k−j⁢(X)=H2⁢n−2⁢k+j⁢(X)∨. Thus we have

In general, we will view Γ¯∗ in Hk⁢(S×X,ΩS×Xk). Recall that

and therefore Γ∈Hk⁢(ΩS×Xk)=⨁Hi⁢(ΩSj)⊗Hk−i⁢(ΩXk−j). By Serre duality, Hk−i⁢(ΩXk−j)≃Hn−k+i⁢(ΩXn−k+j)∨, and so we have Hn−k+i⁢(ΩXn−k+j)→Hi⁢(ΩSj). If dimX=2, we can choose k=2,i=0, and so we have

As a corollary, for all α∈H0⁢(ΩXj) a holomorphic j-form, Γ∗⁢(α) is a holomorphic j-form on S that extends to a j-form on any smooth projective compactification S¯ of S. In particular, Γ∗⁢(α) is closed.

Now let F be a sheaf on S×X that is a flat family of sheaves on X. Then call γ2⁢(F)=tr⁡(A2⁢(F)), where A2 is the Atiyah class. This was introduced by Atiyah for vector bundle and Illusie for coherent sheaves. Here, A⁡(F)∈Ext1⁡(F,F⊗ΩS×X). Thus γi⁢(F)∈Hi⁢(ΩS×Xi). If L is a line bundle, then γi⁢(L)∈H1⁢(ΩL1) is up to a multiple c1⁢(L). Also, γi⁢(F)i!=chi⁢(F). Now if we consider γ2⁢(F)∈H2⁢(ΩS×X2), considering the correct Künneth component gives us a morphism

We will now assume that F is a family of stable sheaves.

Lemma 2.4.17.

For all smooth points s∈S and all α∈H0⁢(ΩX2), the holomorphic 2-form τF⁢(α)∈H0⁢(ΩS2) induces the alternating bilinear pairing given by

Theorem 2.4.18.

Let M0s be the stable locus where the trace map is an isomorphism and ℰ be a quasi-universal family on M0S×X. Then τℰrk⁡ℰ:H0⁢(ΩX2)→H0⁢(ΩM0s2) is independent of ℰ. Moreover, for all α∈H0⁢(ΩX2), τ⁡(α) is nondegenerate at [F]∈M0s if and only if Ext1⁡(F,F)→𝛼Ext1⁡(F,F⊗ΩX2) is an isomorphism.

Proof.

First, let ℰ,ℰ′ be quasi-universal families. Then we know that ℰ⊗p1∗⁢W=ℰ′⊗p1∗⁢W′ for some W,W′. But now we have

When we consider the component H0⁢(ΩX2)→H0⁢(ΩS2), only the A⁡(ℰ) term contributes. Finally, if we consider the traces, we obtain the desired result.

The second part of this follows from Serre duality. Here, we have

This must commute, and thus our pairing must have been nondegenerate. □

Corollary 2.4.19.

Let X be a K3 surface or an abelian surface. Then Ms is smooth and has a holomorphic symplectic form.

We now want to resolve the following questions:

  1. 1.

    What is dimMs? Is it even non-empty?

  2. 2.

    When does Ms=M? Is it irreducible?

  3. 3.

    What happens on M∖Ms? When does there exist a symplectic resolution M~→M?

We will focus on the case when X is a K3 surface. Then dimMs=dimE⁢x⁢t1⁢(F,F), where F is a stable sheaf. But now dimHom⁡(F,F)=dimExt2⁡(F,F)=1, we have dimMs=−χ⁡(F,F)+2. Using Grothendieck-Riemann-Roch, we see that

Definition 2.4.20.

Define v⁡(F)≔ch⁡(F)⋅td⁡(X)∈H∗⁢(X,ℤ). This is called the Mukai vector of F and is equal to (r,c1,c122−c2+r).

For v=(a,b,c),w=(a′,b′,c′)∈H∗⁢(S,ℤ), we write v⋅w=(v,w∨)=b⁢b′−a⁢c′−a′⁢c, where v∨=(a,−b,c). Now we have χ⁡(F,F)=(v⁡(F),v⁢(F)∨)=−v2. Thus dimMs=v2+2.

We will now discuss v-generic polarizations. Let H be a polarization on X and let F be a μH-semistable sheaf on X. Let F′⊆F be such that μH⁢(F′)=μH⁢(F). Define ξF′=c1⁢(F)⋅r′−c1⁢(F′)⋅r. Then our condition that μH⁢(F′)=μH⁢(F) is equivalent to ξF′⋅H=0, and by the Hodge index theorem, ξF′2≤0. Of course, this means that ξF′=0, and in fact this is equivalent to ξF′⋅H=0.

Definition 2.4.21.

The locus {ξF′⋅x=0}⊆Amp⁡(X)ℝ (or in ℋ a cross-section) is called the wall associated to F′.

Theorem 2.4.22.

The walls of v⁡(F) are locally finite in ℋ.

Proof.

We have −r2⁢Δ⁢(F)4≤ξ2≤0, where Δ⁡(F)=c2⁢(End⁡(F)). This gives local finiteness. □

Now H is called generic if H-semistability implies H-stability. There may not always exist such a polarization, but if v∈H∗⁢(S,ℤ) is a primitive element, then a v-generic polarization does exist.

Theorem 2.4.23 (Yoshioka, O’Grady).

Let (X,H) be a polarized K3 surface and v a primitive Mukai vector such that v2≥−2. Then Mv,H is smooth, projective, nonempty, nonempty, irreducible, and connected of dimension v2+2 and is deformation to X[v2/2+1].

This is proved by first deforming the K3 surface to an elliptic K3 and then by using Fourier-Mukai transforms to get birational maps of moduli spaces.

Theorem 2.4.24 (O’Grady, KLS, PR).

Let (S,H) be a polarized K3 surface. Suppose that v=m⁢v0, where v0 is primitive. Suppose that H is v0-generic and v02≥2. Then Mm⁢v0 is a singular symplectic variety which has a resolution if and only if m=2 and v02=2. In this case, then the resolution M~2⁢v0 is irreducible holomorphic symplectic of OG10 type.

Remark 2.4.25.

If v02=0, then the moduli space is a symmetric power of a K3 and if v02=−2, the moduli space is a point.

Note that b2⁢(O⁢G⁢10)=24>b2⁢(S[5]).

3 Bonus: cubic fourfolds

Let X⊆ℙ5 be a smooth cubic hypersurface. By the Lefschetz hyperplane theorem, Hk⁢(ℙ5,ℤ)→Hk⁢(X,ℤ) is an isomorphism for k<4 and is injective for k=4. The only mystery is H4⁢(X,ℤ). Note that H4=3, so if k>2, then H2⁢k⁢(X,ℤ)=ℤ⁢Hk3. We know that H4⁢(X,ℤ) is torsion free and that the cup product is an odd unimodular lattice. But now we compute χtop⁢(X)=c4⁢(X), and here we simply use

and therefore c4⁢(X)=27. In particular, we have b4=27−4=23. Because X is Fano by the adjunction formula, we have

Proposition 3.0.1.

H1⁢(ΩX3) has dimension 1 and under the natural isomorphism between ΩX3≃TX⁢(−3), the generator corresponds up to a multiple to the extension class of

in Ext1⁡(𝒪⁡(3),TX)=H1⁢(TX⁢(−3)).

This is proved using Bott vanishing for Hp⁢(Ωℙnq⁢(k)) and standard exact sequences. We also obtain H3⁢(ΩX1)≃H4⁢(𝒪X⁢(−3)).

Now let Γ⊆S×X be a family of curves on X parameterized by S. Then for any smooth projective model S~ of S, there exists a holomorphic 2-form Γ~∗⁢(η)∈H0⁢(ΩS~2). Here, we have a map H4⁢(X)→H2⁢(S), and then H3,1↦H2,0 because this shifts the Hodge structure down by (1,1).

3.1 Fano variety of lines on X

The main theorem is the following

Theorem 3.1.1 (Beauville-Donagi).

Let X be a cubic fourfold. Then the Hilbert scheme F⁡(X) of lines in X is a smooth connected irreducible holomorphic symplectic fourfold deformation equivalent to K⁢3[2].

Let Xd⊆ℙn and write V=H0⁢(𝒪ℙn⁢(1))∨. Then F⁡(Xd)⊆Gr⁡(2,n+1). Using the tautological exact sequence

dualizing, and taking symmetric powers, we have

This takes X∈Symd⁡V∨ to f𝒮. thus F⁡(Xd)={f𝒮=0}. This has expected dimension 4.

We now compute the tangent space. If ℓ⊆X, we have the normal bundle

The terms are ⨁i=13𝒪⁡(ai), where ai≤1 and ∑ai=1, 𝒪⁢(1)⊕4, and 𝒪⁡(3). The first term could be either 𝒪⁡(1)⊕𝒪⊕𝒪 or 𝒪⁢(1)⊕2⊕𝒪⁡(1). The first case are lines of the first kind and form an open subset, and the lines of the second kind form a closed nonempty subset. Because h0⁢(Nℓ/X)=4 and h1⁢(Nℓ/X)=0, the deformations are unobstructed, and so F⁡(X) is smooth of the expected dimension. The next step is the following:

Proposition 3.1.2.

For any cubic fourfold ωF⁡(X)=𝒪F⁡(X). In the case when X is a cubic threefold, then ωF⁡(X)=hF⁡(X), where h is the Plücker restricted to F⁡(X).

To see this, we know that ωF⁡(X)=ωGr⊗det⁡NF/Gr. This becomes det⁡Hom⁡(𝒮,𝒬)⊗Sym3⁡𝒮∨, and in the dimension 4 case, this is 𝒪⁡(−6⁢h)⊗𝒪⁡(6⁢h)=𝒪F⁡(X).

To prove that F⁡(X) is connected, let Γ⊆F⁡(X)×X be the universal family of lines. Then p:Γ→F⁡(X) is a ℙ1-bundle if course, and we claim that q:Γ→X is a fibration in (2,3) complete intersections in ℙ3 and in particular, this means that F⁡(X) is connected.

We know that p−1⁢(x) is the locus of all lines containing x. Then the locus swept by ℓ∋x is the intersection of Qx∩Hx∩X, where Qx is a quadric cone and Hx is a hyperplane. Choosing local coordinates for X∩Hx⊆Hx=ℙ4, we may choose x=[0,0,0,0,1]. Then X is given by

and then X∩Hx={f2=f3=0}.

Remark 3.1.3.

The same shows that if X0 is a cubic fourfold with a node (A1 singularity) o, then the set of lines through o∈X0 is a (2,3) complete intersection in ℙ4 and is therefore a K3 surface.

We will finally prove that F⁡(X) is irreducible holomorphic symplectic and deformation equivalent to S[2]. Let 𝒳→Δ be a family such that Xt is a smooth cubic fourfold for t≠0 and X0 is a cubic fourfold with a single A1 singularity. Then we have F⁡(X0)⊆F⁡(𝒳/Δ).

Lemma 3.1.4.

Let S be the K3 surface of lines passing through the node. Then F⁡(X0)≃Sym2⁡(S).

Here, there is a unique plane passing through both a line not containing the node and the node, and the intersection of this ℙ2 with this line is part of a triangle of lines. Up to a 2:1 base change of Δ, we can resolve the central fiber. The upshot is that there exists F~→Δ such that F~t=F⁡(Xt) for t≠0 and F~0=S[2].

Now if ℳ→Δ is a projective morphism and ℳ0 is birational to a hyperkähler and the ℳt is hyperkähler, then up to a base change, we can resolve the central fiber such that ℳ~0 is smooth hyperkähler.

Theorem 3.1.5 (Beauville-Donagi).

The Abel-Jacobi map

is an isomorphism of Hodge structures over ℤ, and more specifically,

is an isomorphism of lattices.

Corollary 3.1.6.

The set of (F⁡(X),h) forms a codimension 1 locus in the moduli space of irreducible holomorphic symplectic manifolds of their deformation type.

For very general X, the Neron-Severi group of F⁡(X) is ℤ⁢h.

Proposition 3.1.7.

Let F→S×X be a flat family of sheaves. For all ω∈H1⁢(ΩX3), tr⁡(A3⁢(F)) induces a holomorphic 2-form on S which pointwise is

This is proved by considering the Künneth component H1⁢(ΩX3)→H0⁢(ΩS2)⊗H3⁢(ΩX1).

Now suppose that M is a moduli spaace of stable sheaves. Then a sufficient condition, due to Kuznetsov and Markusevich, for this form to be nondegenerate on a smooth point [F] is that H∙⁢(F)=H∙⁢(F⁡(−1))=H∙⁢(F⁡(−2))=2, and this is precisely the same as F being in the Kuznetsov component of Db⁢(X). This is because

Proposition 3.1.8 (Kuznetsov-Markusevich).

The pairing

is nondegenerate. Here, εG is the composition of idG⊗ν∈Ext1⁡(G⊗ΩX1,G⊗𝒪⁡(−3)) and A⁡(G)∈Ext1⁡(G,G⊗ΩX1).