Patrick Lei

All notes

Geometric invariant theory

Columbia, Fall 2020.

Geometric Invariant Theory Learning Seminar
Fall 2020
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

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

1 Anna (Sep 25): Three Approaches to GIT: Overview and Examples

Let X be an algebraic variety and suppose G acts on X. We want to produce a quotient X/s⁢s⁢l⁢a⁢s⁢h⁢G, which is an algebraic variety. However, this is not the set of orbits of G.

We will assume that G is a reductive group. Some examples of reductive groups are G⁢Ln,S⁢Ln, etc, and an example of non-reductive groups is 𝔾a,

1.1 Some Basic Examples

Example 1.1.1.

Consider the action of 𝔾m on 𝔸n. Then we expect 𝔸n⫽𝔾m=ℙn−1, but of course we know that ℙn−1 is not a geometric quotient of 𝔸n. However, it is the quotient (𝔸n∖{0})/𝔾m.

This example can be generalized to toric varieties, which are varieties with a torus action such that there exists a dense orbit. To construct toric varieties, choose a polytope P⊂V in a real vectir space of dimension d. We know that P is given by inequalities ai⁢(v)+bi≥0 where ai∈V∗ and bi∈ℝ. We want the covectors to be “integral” so we will choose a lattice Vℤ⊂V and ask that ai⁢(Vℤ)⊂ℤ.

Consider the exact sequence 0→K→ℝn→V∗→0, where n is the number of the ai. Taking the lattice ℤn⊂ℝn, we can take 0→Kℤ→ℤn→Vℤ∗→0. Taking quotients, we obtain

If we complexify this picture, we obtain an exact sequence

Consider the action of 𝔾mn on 𝔸n. We can restrict the action to Kℂ, so we will define the toric variety corresponding to the polytope P to be 𝔸n⫽Kℂ.

Our second example will be the moduli of n points in ℙ1. We want an algebraic variety which parameterizes subsets of n points of ℙ1 up to P⁢S⁢L2⁢(ℂ). When n=2,3, all sets of n points are equivalent (by basic facts about projective space). The first interesting case is n=4.

Each n points in ℙ1 define a point in Sn⁢ℙ1=ℙn, where Sn⁢ℙ1 denotes the symmetric product. We will construct the moduli as ℙn⫽S⁢L2.

1.2 Approaches to Taking Quotients

We will discuss various approaches to taking quotients.

1.2.1 Invariant Theory

Suppose that X=Spec⁡A and that G acts on X. Then G acts on A, so we can consider the ring of invariants AG. Then we will define the quotient X⫽G≔Spec⁡AG.

Example 1.2.1.

Suppose 𝔾m acts on 𝔸n with weight −1, Then λ acts on a monomial by

Here, the invariants are only the constants, so the quotient is simply a point.

Theorem 1.2.2.

Suppose X=Spec⁡A is of finite type over a field. Then X⫽G is also of finite type over a field. In addition, the map X→X⫽G is surjective and sends G-invariant closed subsets to closed sets. Finally, disjoint closed invariant subsets are sent to disjoint closed sets. In particular, if Z1,Z2 are closed G-orbits, they are sent to distinct points.

Remark 1.2.3.

This says morally that the quotient parameterizes closed orbits.

Returning to the original example of 𝔾m acting on 𝔸n, the only closed orbit is the origin. If we remove the origin, the other orbits become closed, so we obtain projective space.

Now suppose that X=Proj⁡A and that the action of G on X extends to an action on A that preserves the grading. Then we will define X⫽G=Proj⁡AG. This quotient map is generally not a morphism and is only a rational map.

Define 𝔸n=Proj⁡ℂ⁡[x1,…,xn,t] where xi have degree 0 and t has degree 1. Then λ⋅xi=λ−1⁢xi and λ⋅t=λ⁢t. Then invariants are given by t∑di⁢∏xidi. Thus the ring of invariants is generated by monomials of the form xi⁢t and so 𝔸n⫽𝔾m=Proj⁡ℂ⁡[x1⁢t,…,xn⁢t], which is clearly projective space.

To construct toric varieties, let λ⋅xi=λi−1⁢xi and λ⋅t=∏λibi⋅t, where the bi are from the definition of the polytope. We want to compute the invariants with respect to Kℂ.

Proposition 1.2.4.

(λ1,…,λn)∈Kℂ if and only if for all v∈Vℤ, ∏λiai⁢(v)=1. Differentiating at the origin, we see that (αi)∈Kℂ if and only if for all v∈V, ∑αi⁢ai⁢(v)=0.

A monomial x1d1⁢⋯⁢xndn⁢tr is invariant if and only if

for all (λi)∈Kℂ. In particular, this is equivalent to di=ai⁢(v)=bi⁢r for some v∈Vℤ.

For the moduli of n points, consider ℙn=Proj⁡K⁡[xn,xn−1⁢y,…,x⁢yn−1]. When n=1, there are no invariants, so ℙ1⫽S⁢L2 is empty. For n=2, note that the invariants are given by v0⁢v2=v12, where v0=x2,v1=x⁢y,v2=y2 and thus ℙ2⫽S⁢L2 is a point. When n=3, there is also only one invariant.

When n=4, we have more interesting behavior. Recall that we can define the cross-ration of four points in ℙ1. Thus there exists a unique automorphism of ℙ1 sending p1→0,p2→1,p3→∞, and λ is the cross-ratio. Clearly this depends on the order of the four points. We can fix this by defining the j-invariant. There is another invariant called the i-invariant, so ℙ4⫽S⁢L2=ℙ1.

In general, when n≤10, we can describe the invariants, but when n>10, nobody knows.

1.2.2 (Semi)stability

We have the following problem:

Question 1.2.5.

Describe the maximal open U⊂X such that U→X⫽G is a morphism.

Definition 1.2.6.

x∈X is called semistable if there exists a G-invariant s∈An such that s⁡(x)≠0.

Note this equivalent to the rational map X⇢X⫽G being defined at x. Usually, it is too difficult to G-invariants, so we would like another way to describe semistable points.

Here is an analytic description. Let X be a quasiprojective variety and consider the pullback of the Fubini-Study form (when X is projective) or the constant symplectic form when X is affine. Then (X,ω) is a symplectic manifold. Choose a maximal compact subgroup K⊂G that preserves ω. Then any ξ∈𝔨 defines a vector field on X, and we can rewrite this as

Therefore, we see that ιξ⁢ω is closed. Often, this is even exact. This motivates the following definition:

Definition 1.2.7.

Let (X,ω) be a symplectic manifold. Then an action of K on X that preserves ω is called Hamiltonian if for all ξ∈𝔨 there exists some function μξ such that d⁢μξ=ιξ⁢ω. Moreover, there exists μ:X→𝔨∗ such that μξ=⟨μ,ξ⟩. Finally, require μ⁡(g⁢x)=Ad∗⁢g⁢μ⁢(x).

Consider the action of 𝔾m on 𝔸n. Then K=U⁡(1)=S1. We want to construct μ:𝔸n→ℝ satisfying d⁢μ=ιξ⁢ω, where ξ is tangent to the action of U⁡(1). It is easy to see that

and that

Then define Z=μ−1⁢(0). This is K-invariant and is a sphere S2⁢n−1, so we define Xred=Z/K. We claim that ω|Z is horizontal, and therefore ω descends to the Z/K.

Theorem 1.2.8 (Marsden-Weinstein).
  1. 1.

    Assume that K acts on μ−1⁢(0) freely. Then μ−1⁢(0)/K is a symplectic manifold.

  2. 2.

    If X is Kähler, then μ−1⁢(0)/K inherits the Kähler structure.

In our example, μ−1⁢(0)=S2⁢n−1, and S2⁢n−1/U⁡(1)=ℙn−1. This is the same as the quotient from our algebraic procedure.

We conclude with another definition of semistability.

Definition 1.2.9.

The point x∈X is called semistable if G⁢x¯∩μ−1⁢(0)≠∅. Here, G is an algebraic group (not necessarily compact).

In conclusion, our algebraic methods of describing invariants and the analytic method of moment maps give a quotient.

2 Caleb (Oct 2): GIT Preliminaries

Today, we will discuss preliminaries for GIT. These may be familiar from other contexts, but they may be different in the category of schemes. We will assume the reader knows what a scheme is.

We will begin with some notation.

  1. 1.

    By a scheme X/S, we mean a morphism of schemes X→S.

  2. 2.

    An S=valued point is a morphism S→X. In particular, when S=Spec⁡k for an algebraically closed field k, this is called a geometric point.

  3. 3.

    A morphism os S-schemes X/S,Y/S is a morphism of schemes X→𝑓Y that commutes with the maps to X.

  4. 4.

    The product of two morphisms is simply the fiber product.

Definition 2.0.1.

A group scheme G/S is a scheme with multiplication m:G×G→G, inverse β:G→G, and identity e:S→G that satisfies the group axioms. This is simply a group object in the category of schemes.

Definition 2.0.2.

An algebraic group over k is a smooth separated group scheme over k.

Definition 2.0.3.

A group G acts on X/S if there exists a morphism σ:G×X→X such that the diagram

commutes.

Given a group action σ:G×X→X, let ψ:G×X→X×X be ψ=σ×p2. Then the orbit is the image of ψ and the stabilizer is the pullback of

Given a map f:T→X, we can define ψf:G×T→X×T to be ψf=(σ∘f,p2).

Definition 2.0.4.

A categorical quotient of σ:G×X→X is a pair (Y,ϕ) that is the pushout of X←p2G×X→𝜎X.

Definition 2.0.5.

A geometric quotient of σ:G×X→X is a pair (Y,ϕ) such that the diagram

commutes and

  1. (1)

    Φ is surjective and the image of ψ is X×YX.

  2. (2)

    Φ is submersive, or ϕ−1⁢(U) is open if and only if U is open.

  3. (3)

    𝒪Y is the subsheaf of invariant functions of ϕ∗⁢𝒪X.

The second condition means that the preimage of a point in Y is a single orbit, and the last condition says that functions on Y are G-invariant functions on X.

2.1 Properties of Quotients

Proposition 2.1.1.

The geometric quotient is a categorical quotient.

Proof.

We need to prove that a geometric quotient Y is the pushout. First, cover Z=⋃Vi by affines. If a map χ exists, we will prove it is unique. First, we will show that if ϕ⁡(x)=ϕ⁡(y), then ψ⁡(x)=ψ⁡(y). This follows from surjectivity and condition (1) of the geometric quotient. Thus ψ−1⁢(Vi)=ϕ−1⁢(Ui), so Ui is open. Then we have a diagram

Here, Ui is the cover of Y pulled back from Z. Also, ϕ∗ is injective. But then hi∗ is unique.

To prove that hi exists, note that hi exists if ψ∗⁢(g) is an invariant function for g∈𝒪Z. Then the diagram

commutes, and then checking gluing is just the sheaf property. □

Example 2.1.2.

Let Gm=Spec⁡k⁡[t,t−1] act on 𝔸n. Then the categorical quotient is Spec⁡k, which means that there cannot be a geometric quotient. If we remove the origin from 𝔸n, then we obtain ℙn−1 as the geometric quotient.

Definition 2.1.3.

A universal geometric (resp. categorical) quotient is a geometric (resp. categorical) quotient that is preserved by base change under S′→S. A quotient is called uniform if it is preserved by flat base change.

Remarks 2.1.4.
  1. 1.

    The base scheme S doesn’t matter.

  2. 2.

    If (Y,ϕ) is a categorical quotient of G×X→X, then X is reduced (resp. connected, irreducible, etc) if and only if Y is. However, the property of being Noetherian is not preserved (Hilbert’s 14th problem).

  3. 3.

    Let (Y,ϕ) be a universal categorical quotient. Then condition (3) of the geometric quotient is satisfied.

  4. 4.

    Read the next remark yourself in Mumford’s book. This is used to give a criterion for geometric quotients.

  5. 5.

    The property of being a universal categorical quotient can be checked on affines.

  6. 6.

    Read the next remark yourself. It is not very important.

  7. 7.

    Suppose we have a geometric quotient (Y,ϕ). When is it universal? The diagram always commutes and (1) holds under any extension. Then (2) does not always hold and (3) holds for flat extensions. To see this, we can rephrase (3) as exactness of

    Then a flat extension is tensoring with a flat algebra.

  8. 8.

    If Y′→Y is faithfully flat and quasicompact, then (Y′,ϕ′) is universal if and only if Y is. This is proved in SGA 8.

Definition 2.1.5.

Suppose G×X→𝜎X is a group action. This action is closed if for all geometric points x, the orbit of x in X×Spec⁡k¯ is closed.

Definition 2.1.6.

A group action σ is separated if the image of ψ=(σ,p2) is closed.

Definition 2.1.7.

A group action σ is proper if ψ is proper. Here, recall that proper means separated, universally closed, and of finite type.

Definition 2.1.8.

A group action σ is free if ψ is a closed immersion. In particular, it is set-theoretically free.

Example 2.1.9.

Here, we will have a separated action that is set-theoretically free with a geometric quotient that is not proper. Let G=S⁢L2 and define Vn=Spec⁡ℂ⁡[xn,xn−1⁢y,…,yn]. Then G acts on Bn by matrix multiplication on (x,y).

Now define X⊂V1×V4 to be the set of points (F1,F4) where F1=0 and F4 is the equare of a homogeneous quadratic form with discriminant 1. Then (𝔸1,ϕ) is a geometric quotient, but ψ is not free or closed. This example shows that things do not necessarily work the way we expect set-theoretically.

Proposition 2.1.10.

Let (Y,ϕ) be a geometric quotient. Then σ is closed. This means that the orbit of any geometric point is closed. In addition, Y is separated if and only if σ is separated.

Proof.

We have the diagram

It remains to check that Orb⁡(x)=ϕ¯−1⁢ϕ¯⁢(x). Then if X lives over Spec⁡Ω for a field Ω, then any closed point is a maximal ideal.

Then we show that G×X→X×X is closed. If we have maps X1,X2→Y→Z, then the diagram

is a fiber product. Applying this to X1=X2=X,Z=S, we obtain the desired result because ϕ is universally submersive. The rest is omitted, and is Lemma 0.6 in Mumford. □

Remark 2.1.11.

Some of the definitions given may not be the current conventions.

Proposition 2.1.12.

G is affine and σ is proper if and only if ϕ is affine.

Definition 2.1.13.

A principal fiber bundle with group G is a geometric quotient (Y,ϕ) such that ϕ is flat and finite type, and ψ is an isomorphism.

Let Gm act on 𝔸n∖{0} with weights (2,2,…,2). Then ℙn−1 is a geometric quotient but not a principal fiber bundle.

Remark 2.1.14.

The topics at the end of this chapter include Hilbert schemes and Picard schemes. To learn more about these, attend Caleb’s seminar on FGA Explained.

3 Alex (Oct 9): Actions of Reductive Algebraic Groups

In this talk, we will let k be a field that is not necessarily algebraically closed. All of our schemes will live over k. Sometimes k will be embedded in some algebraically closed field Ω, and we will write X¯=X×kΩ. We will denote algebraic groups by G and invertible sheaves by L.

3.1 Basic Definitions

Definition 3.1.1.

A representation of an algebraic group G is a group homomorphism ρ:G→G⁢L⁢(n).

A representation of a group G can be an action of G on 𝔸n as a scheme, an action of G on kn as a vector space, and a third action, called the dual action. Here, we let S=Γ⁡(G,𝒪G). Then S is a coalgebra. Denote the multiplication by α̂ and the counit by β̂. Then any representation V of G is a comodule for S. Define the coaction by σ̂:V→S⊗V.

Definition 3.1.2.

Let σ̂ be a dual action. Then W⊆V is invariant if σ~⁢(W)⊆S⊗W, and a vector v is invariant if σ̂⁢(v)=1⊗v.

Lemma 3.1.3 (Cartier).

V is a union of finite-dimensional invariant subspaces.

Proof.

Let {ei} be a basis for V. Then write σ̂⁢(ei)=∑jai⁢j⊗vi⁢j. Then define γ:S∗⊗V→S∗⊗S⊗V→V. At this point, we cut the proof short because. □

Definition 3.1.4.

An algebraic group G is reductive if the radical R⁢G is a torus. Here, the radical is the unique largest connected normal solvable subgroup.

Definition 3.1.5.

A group G is linearly reductive if every representation is completely reducible.

Definition 3.1.6.

Let G be a linearly reductive group and S=Γ⁡(G,𝒪G). Let σ̂:V→S⊗V. Then E:V→V is a Reynolds operator if

  1. 1.

    σ̂∘E=(1S⊗E)∘σ̂;

  2. 2.

    E2=E;

  3. 3.

    E⁢x=x if and only if σ̂⁢(x)=1⊗v.

The important thing to know is that Reynolds operators always exist and that if V is a k-algebra, then E⁡(x⁢y)=E⁡(x)⁢y.

Remark 3.1.7.

There is a classification of linear reductive algebraic groups due to Nagata. In characteristic p>0, we have G0=(Gm)r and |π0⁢(G)| is prime to p. If char=0, then linearly reductive is equivalent to reductive and G=T⋉G′, where G′ is the commutator subgroup.

3.2 Affine Case

Fix char⁡k=0. We will consider the action of a reductive group G on an affine scheme X. Denote the action by σ.

Theorem 3.2.1.

There exists a uniform categorical quotient (Y,φ) where φ is universally submersive. If X is finite type (resp. Noetherian), so is Y. In characteristic 0, the quotient is actually universal.

Recall that σ induces a dual action of S=Γ⁡(G,𝒪G) on R=Γ⁡(X,𝒪X). Define R0⊂R to be the set of invariant elements. Then Y=Spec⁡R0 and φ is induced by the inclusion.

Remark 3.2.2.

From last time, it suffices to show that 𝒪Y is the invariant subsheaf of 𝒪X and that φ is closed. Also, we need to show that φ⁡(⋂Wi)=⋂φ⁡(Wi).

  1. 1.

    If we bet S0 be an R0-algebra, then S0 is the invariant ring in R⊗R0⁢S0.

  2. 2.

    If 𝔄i⊂R is a set of invariant ideals, then (∑𝔄i)∩R=∑(𝔄i∩R).

To prove that Y is Noetherian if X is, we can choose an ideal 𝔄⊂R0. Then (𝔄⁢R)∩R0=𝔄, so we have an ideal correspondence between R and R0.

To prove that being of finite type is preserved, note that if R is a graded k-algebra, then R0 is a graded subalgebra. Then R is finite type, so it is finite type, so then R0 is also Noetherian, so it is finite type. In general, there is a procedure to put a grading on any ring so that this works.

Theorem 3.2.3.

The categorical quotient (Y,φ) is a geometric quotient if and only if σ is a closed action. In characteristic 0, this is a universal quotient.

3.3 Linearization

Definition 3.3.1.

Let L be an invertible sheaf over X. Then a G-linearization is a map φ:σ∗⁢L→≅p2∗⁢L satisfying a nasty cocycle condition Alex didn’t bother to write down.

We can interpret a linearization in the following ways:

  1. 1.

    For α,β∈G, we have τα∗⁢L≅L. Also, we have a commutative diagram

  2. 2.

    φ extends the G-action to the G-action to the line bundle 𝕃 defined by L.

Example 3.3.2.

If X=Spec⁡k and L=𝒪, then 𝕃=𝔸1. A G-linearization will be a character χ:G→Gm.

This χ defines three actions:

  1. 1.

    An action Σ on 𝕃 defined by Σ⁡(a,z)=χ⁡(a)⋅z.

  2. 2.

    For χ¯∈Γ⁡(G,𝒪G∗), write φ⁡(f)=χ¯−1⁢f for f∈Γ⁡(G,𝒪G).

  3. 3.

    A dual action

    which is the composition

We observe that if L1,L2 are G-linearized invertible sheavs, then L1⊗L2 and L1−1 also carry a G-linearized structure. Thus we can define a group PicG⁡(X) inside the Picard group.

Remark 3.3.3.

If we have a morphism f:X→Y, then we can pull back φ∗:PicG⁡(Y)→PicG⁡(X). If G acts trivially on Y, then this is a map from Pic⁡Y. If Y is a geometric quotient of X, then pullback is an isomorphism of G-linearized Picard groups.

Example 3.3.4.

Let G=P⁢G⁢L⁢(n+1) acting on ℙn. Then define σ:G×X→X with

and

where xi are the coordinates on ℙn and ai⁢j are the homogeneous coordinates on P⁢G⁢L⁢(n+1).

It is known that 𝒪⁡(1)∈Pic⁡(P⁢G⁢L) has order n+1. If 𝒪⁡(1)∈PicG⁡(ℙn), then it would have order dividing n+1. However, if G=S⁢L⁢(n+1), then 𝒪⁡(1) has a G-linear structure. Then the total space of 𝒪⁡(1)=Bl0⁡𝔸n+1 and clearly there is an action here.

Proposition 3.3.5.

If G is connected and of finite type and X is of finite type, there exists no nontrivial character of G¯, and X is geometrically reduced, then L∈Pic⁡X has at most one G-linearization.

Corollary 3.3.6.

If X is a normal variety, then there exists n∈ℕ such that Ln is G-linearizable.

Proposition 3.3.7.

Let L be basepoint-free. Then there exists an action of G on ℙn and a G-linearization of 𝒪⁡(1) such that the map I:X→ℙn defined by L satisfies I∗⁢𝒪⁢(1)=L. If X is proper, then I is induced from H0⁢(X,L). If L is very ample, I is an immersion.

Proof.

Mumford does some setup and then uses some big fact with citation from three Japanese guys whose names I don’t know. Let V1⊂H0⁢(X,L) be a basepoint free finite-dimensional subspace. Then we can extend V1⊂V to a a finite-dimensional invariant subspace.

Then we use the fact that there exists a natural equivalence between dual actions of G on V=H0⁢(ℙ⁡(V),𝒪⁡(1)) and actions of G on ℙ⁡(V) with G-linearizations of 𝒪⁡(1). □

3.4 The General Case

Definition 3.4.1.

Let x∈X be a geometric point. Then x∈XS⁢(p⁢r⁢e) is a prestable point if there exists an invariant affine neighborhood U∋x such that G acting on U is a closed action.

Definition 3.4.2.

A geometric point x is a semistable point (x∈Xs⁢s⁢(L)) if there exists an invariant s∈H0⁢(X,Ln) such that s⁡(x)≠0 and Xs is affine. The point x is stable if it is semistable and the action of G on Xs is closed.

If the action of G on U is closed, the dimension of the stabilizers is constant over U. Then XS⁢(P⁢r⁢e)=⋃i=1nX(i)S⁢(P⁢r⁢e) for opens X(i). A similar statement holds for XS⁢(L). Finally, X(0)S is “properly stable.’

Remark 3.4.3.

We can form a graded ring V=⨁nH0⁢(X,Ln)G with invaraint sections such that Xs is affine.

Proposition 3.4.4.

A uniform geometric quotient (Y,φ) of XS⁢(P⁢r⁢e) exists and φ is an affine map. Also, Y is algebraic. Conversely, if U⊂X has a geometric quotient (Z,ψ) with ψ affine, then U is contained in the prestable locus.

There is no proof of this because it is unimportant.

Theorem 3.4.5.

A universal categorical quotient (Y,φ) of Xs⁢s⁢(L) exists.

  1. 1.

    The map φ is universally submersive.

  2. 2.

    There exists an ample line byndle M over Y such that φ(M)=Ln.

  3. 3.

    There exists Y~⊂Y open such that XS⁢(L)=φ−1⁢(Y~). Then Y~ is a uniform geometric quotient (and is universal in the charateristic 0 case).

Proof relies on doing everything into the affine case and gluing the affines together.

Proposition 3.4.6.

Let x∈Xs⁢s⁢(L) be a geometric point. The following are equivalent:

  1. 1.

    x∈XS⁢(L) is a stable point.

  2. 2.

    x is regular for the action and the orbit of x is closed.

  3. 3.

    x is regular and there exists an invariant section s∈H0⁢(X,LN) such that s⁡(x)≠0, Xs is affine, and the orbit of x is closed.

Moreover, if x1,…,xn∈Xs⁢s⁢(L) are geometric points, there exists an invariant s∈H0⁢(X,Ln) such that Xs is affine and s⁡(xi)≠0.

There is a converse: If (Y,φ) is a categorical quotient of X and φ is affine and Y is quasiprojective, then there exists L∈PicG⁡(X) such that X=Xs⁢s⁢(L). If Y is a geometric quotient, then X=XS⁢(L). This is not quite good enough because if U⊂ has a categorical quotient, we only obtain an invertible sheaf on U.

Here is a better converse. Let X be connected, algebraic, and smooth. Let G be connected, reductive, and algebraic. Then let U⊂X be invariant open. The following are equivalent:

  1. 1.

    There exists L∈PicG⁡(X) such that U=XS⁢(L).

  2. 2.

    There exists a geometric quotient (Y,φ) of U such that φ is affine and Y is quasiprojective.

  3. 3.

    The action of G on U is proper and there exists a quasiprojective geometric quotient (Y,φ).

4 Patrick (Oct 16): Stability

Note: These are the speaker’s notes.

Throughout this talk, we will let k be an algebraically closed field. All schemes X will live over k. Unless otherwise stated, all schemes are proper over k. The goal of this talk is to give a numerical criterion for stability and to describe where stability breaks down, and then present examples.

4.1 Hilbert-Mumford Stability Criterion

Definition 4.1.1.

Let G be an algebraic group. Then a 1-parameter subgroup of G is a map λ:Gm→G.

Then for any point x∈X and 1-parameter subgroup λ of G, we obtain a morphism ψx∘λ:Gm→X (action of Gm on x via λ). By the valuative criterion for properness, there is a unique extension to a morphism f:𝔸1→X. Clearly f⁡(0) is fixed under the Gm-action on X induced by λ. Now if L∈PicG⁡(X), then λ induces a Gm-linearization of L. Restricting to f⁡(0), Gm acts on the fiber of L (as a line bundle) by some character χ∈Hom⁡(Gm,Gm)=ℤ∋r, where r is the integer associated to χ under the isomorphism id↦1.

Definition 4.1.2.

Let X be proper over k with an action of G. Then if x∈X is a closed point, λ is a 1-parameter subgroup of G, and L∈PicG⁡(X), then we define

This satisfies the following properties:

  1. 1.

    For α∈G, we have μL⁢(σ⁡(α,x),λ)=μL⁢(x,α−1⁢λ⁢α).

  2. 2.

    Fix x and λ. Then μL⁢(x,λ) defines a group homomorphism PicG⁡(X)→ℤ.

  3. 3.

    Let f:X→Y be a G-equivariant morphism, L∈PicG⁡(Y), and x∈X. Then

  4. 4.

    If σ⁡(λ⁡(α),x)→y as α→0, then μL⁢(x,λ)=μL⁢(y,λ).

This is the main theorem of this talk:

Theorem 4.1.3 (Hilbert-Mumford Stability Criterion).

Let X be proper over k with an action of a reductive group G. Let L∈PicG⁡(X) be ample. Then for any x∈X, the following hold:

  1. 1.

    x∈Xs⁢s⁢(L) (x is semistable) if and only if μL⁢(x,λ)≥0 for all 1-parameter subgroups λ.

  2. 2.

    The inequality is strict for all λ if and only if x is stable.

To prove this result, we will choose a G-equivariant immersion X⊂ℙn−1 and reduce to the case of projective space. In Mumford, the proof is broken up into two propositions:

Proposition 4.1.4.

A point x∈ℙn−1 is semistable if and only if (0) is not in the closure of the orbit O⁡(x~) for all points x~∈𝔸n such that x=[x~]∈ℙn−1. Similarly, x is properly stable if and only if ψx is proper for all x~∈𝔸n living over x.

Proof of this fact is technical computation with the definitions.

The next step is to interpret the function μ when X=ℙn−1.

Proposition 4.1.5.

Let x∈ℙn−1 and λ a 1-parameter subgroup of G. Suppose that σ⁡(λ⁡(α),x)→y as α→0. Then let x~∈𝔸n live over x. Finally, suppose that Gm acts on 𝔸n with weights r1,…,rn. Then

In addition, σ~⁢(λ⁡(α),x~) has (no specialization, some specialization, specialization (0)) in 𝔸n when α→0 if and only if (μ⁡(x,λ)>0, μ⁡(x,λ)=0, μ⁡(x,λ)<0).

Proof.

The idea is to prove that μ⁡(x,λ)=max⁡{−ri∣x~i≠0}. We will omit the technical computation. □

Remark 4.1.6.

There is an alternative method of proof. If we consider the orbit O⁡(x), then we can normalize its closure and reduce to the case when O⁡(x) is dense in X.

We will now state a result whose motivation is completely unknown to me. However, there are consequences that fit into the philosophy of determining properties of an action of G on X using 1-parameter subgroups.

Let R=k⁡[[t]] and K=k⁡((t)). Then if λ:Gm→G is a 1-parameter subgroup, then we can define a point ⟨λ⟩∈G⁡(K) via the morphism Spec⁡K→Gm→G.

Theorem 4.1.7 (Iwahori).

Let G be a semisimple algebraic group over k of adjoint type. Every double coset of G⁡(K) with respect to G⁡(R) is represented by a point ⟨λ⟩.

Here, a semisimple group is of adjoint type if its center is trivial. Somehow we have the following corollaries that are of interest to us.

Corollary 4.1.8.

Let G be a reductive algebraic group and σ:G×X→X be an action of G on a separated algebraic scheme X over k. Then σ is proper if and only if for every nontrivial 1-parameter subgroup λ:Gm→G, the induced action of Gm on X is proper.

Corollary 4.1.9.

Let G be a reductive algebraic group acting on an algebraic scheme X and L∈PicG⁡(X). Then G acts properly on X(0)s⁢(L).

4.2 Flag Complex

When we construct moduli of subschemes of projective space, we will explore relationships between stability and flags in projective space. Here, we will simply state various propositions. In modern terminology, this is the Tits building of G.

Definition 4.2.1.

Let G be a reductive group and λ be a 1-parameter subgroup. Then there is a unique subgroup P⁡(λ)⊂G (which is parabolic) such that γ∈P⁡(λ) if and only if λ⁡(α)⁢γ⁢λ⁢(α−1) has a specialization in G whenever α→0. In addition, λ lies in the radical of P⁡(λ) and the specialization of λ⁢γ⁢λ−1 centralizes λ.

Proposition 4.2.2.

Let G be a reductive group and σ:G×X→X be an action of G on a proper separated scheme X. Then for all x∈X,L∈PicG⁡(X),λ:Gm→G,

if γ∈P⁡(λ).

We are now ready to define the Tits building for G.

Definition 4.2.3.

Let G be a reductive group. The rational flag complex Δ⁡(G) is the set of non-trivial λ:Gm→G modulo the following equivalence relation:

λ1∼λ2 if there exists n1,n2∈ℤ>0 and y∈P⁡(λ1) such that λ2⁢(αn2)=γ−1⁢λ1⁢(αn2)⁢γ.

In the rest of the chapter, it is shown that Δ⁡(G) is the set of rational points in a geometric realization of a simplicial complex (which is the spherical Tits building). On this geometric realization, we may construct various metric space structures. This Tits building is supposed to generalize the notion of a Riemannian symmetric space to groups over an arbitrary field, which I will now explain.

Definition 4.2.4.

Let M be a Riemannian manifold such that for all p∈M, there exists an isometric involution sp that fixes p and reverses all geodesics through p. Then M is called a symmetric space.

It turns out that the classification of (simply-connected) symmetric spaces is essentially the classification of Lie groups. First, there are three types: Euclidean (zero curvature), compact type (nonnegative curvature), and noncompact type (nonpositive curvature). Then every symmetric space is of the form G/K, where G is a Lie group and K is some compact subgroup. In the noncompact type case, K is always a maximal compact subgroup of G.

Example 4.2.5.

Some examples of symmetric spaces are the upper half plane, hyperbolic space, spheres, and ℝn.

4.3 Examples of Stability

From now on, we will work over ℂ. If you care about other ground fields, go find a reference and read yourself.

4.3.1 Points on the Line

The first example we will consider is the space of n points in ℙ!. Recall that any set of n points in ℙ1. We will describe the space of (semi)stable points with respect to the action of S⁢L2. In Anna’s lecture [chapter 1], we discussed the approach of computing invariants for this action, but this problem is extremely hard.

Recall that a set of n points in ℙ1 determines a degree n polynomial in the variables [x0,x1]. We will assume that n≥2 because the story for a single point in ℙ1 is very boring (really this should be n≥4). First, we recall some basic Lie theory. Every simple Lie algebra 𝔤 has a maximal solvable subalgebra 𝔟 and the Cartan subalgebra 𝔱=𝔟/[𝔟,𝔟]. Lifting to the situation for groups, this gives a Borel subgroup B and a maximal torus T⊂B. In particular, all Borels are conjugate and thus all maximal tori are also conjugate. For S⁢Ln, the Lie algebra is the set 𝔰⁢𝔩n of traceless matrices, 𝔟 is the set of upper triangular matrices, and 𝔱 is the set of diagonal matrices.

This means that WLOG, we may assume that for any λ:Gm→S⁢L2, that the image of λ is the maximal torus

Thus all 1-parameter subgroups are determined by their weight k. By Remark 6.12 in Hoskins, it suffices to check the Hilbert-Mumford criterion for primitive λ, or those with weight 1. This reduces the computation to checking one case.

Write our polynomial as P⁡(x,y)=∑i=0nai⁢xn−i⁢yi. Then it is easy to see that

and therefore, we have

In particular, μ⁡(F,λ) is determined completely by the multiplicity of the root [1:0]. We have

  1. 1.

    μ⁡(F,λ)≥0 if and only if [1:0] has multiplicity less than or equal to n/2.

  2. 2.

    μ⁡(F,λ)>0 if and only if [1:0] has multiplicity strictly less than n/2.

Generalizing to all roots of F, we see that:

  1. 1.

    F is semistable if and only if all of its roots have multiplicity at most n/2.

  2. 2.

    F is stable if and only if all of its roots have multiplicity strictly less than n/2.

The actual quotients were described in Anna’s lecture [chapter 1].

4.3.2 Plane Cubics

We would like to describe the space of plane cubics up to projective equivalence. First we will describe the reducible (or nonreduced) plane cubics. Every conic is equivalent to the conic given by y2−x⁢z=0, and so any union of a conic and a line is given by either

  • •

    (y2−x⁢z)⁢y=0 and the line meets the conic in two distinct points;

  • •

    (y2−s⁢z)⁢x=0, and the line is tangent to the conic.

These are the possibilities for the unions of three lines:

  • •

    y3=0 a triple line;

  • •

    x⁢y2=0 a double line and a distinct line;

  • •

    x⁢y⁢(x+y)=0 three concurrent lines;

  • •

    x⁢y⁢z=0 three distinct lines.

The singularities of all non-irreducible possibilities are: two double points, a single double point (with tangency 2), a line of triple points, a line of double points with a triple point, a single triple point, two double points.

Next, all irreducible plane curves are given by a Weierstrass form

Thus we have three cases: smooth (elliptic curve), nodal, and cuspidal. This gives a complete classification, but it does not tell us which curves are (semi)stable.

Theorem 4.3.1 ([H], Lemma 7.25).

A plane cubic curve C is semistable if and only if it has no triple point and no double point with a unique tangent. A plane cubic curve is stable if and only if it is smooth.

Proof.

We will represent the curve by a homogeneous polynomial F⁡(x,y,z)∈H0⁢(ℙ2,𝒪⁡(3)). In addition, all 1-parameter subgroups of S⁢L3 are conjugate to those of the form

where r0+r1+r2−0. Also, we may assume that r0≥r1≥r2. If we write

Then λ=(r0,r1,r2) acts by

Thus we have

Because we assumed r0≥r1≥r2, the monomials that do not appear are x3,x2⁢y,x⁢y2,x2⁢z,x⁢y⁢z if μ⁡(F,λ)<0. Then if the x⁢z2 term also vanishes, then [1,0,0] is a triple point. Otherwise, it is a double point with a unique tangent (if we affinize with x=1, then the lowest degree homogeneous term is z2, which means we have a double tangent).

Then if p is a singular point, we can assume p=[1:0:0] and then by the Jacobian criterion, we have a00=a10=a01=0. With weights (2,−1,−1), we can see that μ⁡(F,λ)≤0. On the other hand, if F is not stable, then there exists weights λ=(r0,r1,r2) such that μ⁡(F,λ)≤0. Then x3,x2⁢y have strictly negative weights. Then if a01 is nonzero, we have 2⁢r0+r2≤0. However, because r0≥r1≥r2, we have r1=r0, r2=−2⁢r0. This implies that

and because r0>0, we see that a20=a30=0. This implies that all terms without z vanish, so F is reducible and thus the curve is singular. □

There are three strictly semistable orbits: nodal cubics (y2=x3+x2), union of conic and secant line, and three non-concurrent lines. It turns out that the first orbit contains both of the others in its orbit, and this corresponds to the point at infinity in the compactification of the quotient of the stable locus. It is known that the quotient of the stable locus by S⁢L3 is isomorphic to ℋ/S⁢L2⁢(ℤ)≅𝔸1, so the compactification is ℙ1.

5 Nicolás (Oct 23): More Examples of Stability

Let G be a reductive algebraic group, X a smooth projective variety, and L an invertible sheaf over X with a G-linearization. Throughout this talk we work over an algebraically closed field of characteristic zero.

Definition 5.0.1.

x∈X is semistable if there exists s∈H0⁢(X,Ln) such that s⁡(x)≠0.

Then we have a categorical quotient (Y,ϕ) of Xs⁢s⁢(L). The semistable locus is dependent on the line bundle. Also, we can write Xs⁢s⁢(L)⊂X=Proj⁡R. Then Y=Proj⁡R0. Here, we have

To find Xs⁢s⁢(L), it suffices to compute R0.

5.1 Matrices and Conjugation

Let X=ℙ⁡(Mn⁢(ℂ))=Proj⁡ℂ⁡[Mn⁢(ℂ)]. Consider the action of G⁢Ln on X by conjugation. The action on functions is given by P⁢f⁢(A)=f⁡(P−1⁢A⁢P). To write the quotient X⫽G⁢Ln, we compute the ring of invariants. This is simply

where σi is the (n−i)-th coefficient of the characteristic polynomial.

Proof.

Let f∈ℂ⁢[Mn]G⁢Ln and let A∈Mn⁢(ℂ). We can replace A by its Jordan canonical form

Then if we multiply all δi by some ε≠0, we take ε→0 to see that f⁡(A)=f⁡(λ1,…,λn). But then

Now to compute the stable locus, we note that if σi⁢(A)=0 for all i, then A is nilpotent. Thus A is unstable if and only if it is nilpotent.

Recall the Hilbert-Mumford criterion and the following proposition from Patrick’s lecture [chapter 4]:

Proposition 5.1.1.

Let x∗∈ℙn−1, x∈𝔸n. Then x∈Xs⁢s if and only if 0∉G⋅X¯.

Now write T=(𝔾m)k⊂G. The following proposition is from the notes of Jenia Tevelev at http://people.math.umass.edu/~tevelev/moduli797.pdf.

Proposition 5.1.2.

The following are equivalent:

  1. 1.

    x∗ is unstable.

  2. 2.

    There exists y=G⋅x such that 0∈T⋅y¯.

This means that up to changing x in its orbit, we may assume G=T. Then recall that

If v∈𝔸n, then we write w⁢t⁢(v)={m∈ℤk∣vm≠0}. Then we will write w⁢t⁢(v)¯ for the convex hull.

Theorem 5.1.3.

([D], Theorem 9.2) v is semistable if and only if 0∈w⁢t⁢(v)¯.

We now return to the example of matrices and conjugation. Fix T to be the standard maximal torus in G⁢Ln. First, take the decomposition

Then the conjugation is

If A∈Mn, then there exists P∈G⁢Ln taking A to its Jordan form. Then if B=P⋅A, then

Then note that w⁢t⁢(Ei⁢i)={0} and w⁢t⁢(Ei,i+1)={(0,…,0,1,−1,0,…,0)}. If λi≠0 for some i, then 0∈w⁢t⁢(B)¯.

5.2 Subschemes of Projective Space

Let G=S⁢Ln+1 act on 𝔸n+1. This gives an action on ℂ⁢[z0,…,zn]d, which is the same thing as an action on degree d hypersurfaces in ℙn. Letting T be the standard maximal torus, we obtain the following grid for the values of w⁢t⁢(monomial):

X4X3⁢YX2⁢Y2X⁢Y3Y4X⁢Y2⁢ZX2⁢Y⁢ZX⁢Y⁢Z2Y3⁢ZX3⁢ZY2⁢Z2X2⁢Z2Y⁢Z3X⁢Z3Z4
Figure 5.1: Weights for plane quartics.

Then semistable monomials are given by sweeping out a line through the origin and seeing which monomials lie on one side. For example considering the line y=−x, the allowable monomials are x4,x3⁢y,x2⁢y2,x⁢y3,y4,x3⁢z,x2⁢y⁢z,x⁢y2⁢z,y3⁢z. If we have such a curve F, then if we set z=1, the point [0:0:1] has multiplicity at least 3.

The next option is to include x4 but not y4 or z4. One of the choices gives us the monomials x4,x3⁢y,x2⁢y2,x⁢y3,x3⁢z,x2⁢y⁢z,x2⁢z2, and thus the curve is reducible.

Now we will consider points in projective space. We will construct (ℙn)m⫽S⁢Ln+1. We will choose the line bundle L=𝒪⁡(k1,…,km). We know that L is ample if and only if ki>0 for all i. We say that k=(k1,…,km) is democratic if k1=⋯=km.1

Choosing a 1-PS with weights q0≥⋯≥qn, fix P=(p1,…,pm)∈(ℙn)m. Writing pi=[ai⁢0,…,ai⁢1,…,ai⁢vi,0,…,0] with vi≠0, we see that

Using some basic manipulation, this becomes

Theorem 5.2.1.

(D, Theorem 11.1) A point P∈Xs⁢s⁢(Lk) if and only if for all w⊆ℙn,

Proof.

Suppose q0=⋯=qs=n−s and qs+1=⋯=qn=−(s+1). The other direction is similar. We will neglect all of the explicit computations. □

6 Morena (Oct 30): GIT and Moduli Problems

6.1 Moduli Problems

One of the central problems in algebraic geometry is to classify various geometric objects: curves, schemes, sheaves. We want to find a space such that M⁡(K) is in bijection with isomorphism classes of our objects. However, this is trivially solved by taking a disjoint union of points, so we need a stronger notion of moduli problem. This is given by a functor

that sends a scheme T to the set of families of objects over T.

Example 6.1.1.

Consider the Hilbert functor HilbM/R that sends a scheme S to

Theorem 6.1.2.

The Hilbert functor is represented by a quasi-projected scheme.

Example 6.1.3.

Consider the functor ℙℤn that sends

This functor is represented by the scheme ℙn.

Recall that there is a functor 𝔸n+1∖0→ℙn. Then X→𝔸n+1∖0 is the same thing as specifying n+1 sections of basepoint-free line bundle. For U⊂X such that ℒ|U=𝒪|U, we obtain sections

Because projective space does not see constants, we see that these maps glue and we obtain a global map.

Conversely, consider X→𝑖ℙn and take the sections i∗⁢x1,…,i∗⁢xn∈i∗⁢𝒪⁢(1). These do not simultaneously vanish, so we obtain an invertible sheaf with generating sections.

An obstruction to the solution of a moduli problem is the existence of nontrivial automorphisms. Suppose we want to classify vector bundles with rank r over an algebraically closed field k. Call this functor

If Y represents this functor, then a curve with nontrivial vector bundle is a map C→Y. Then given a cover {Ui} of C by affines, then any vector bundle 𝒱 is trivial when restricted to Ui. But then the trivial bundle is a closed point y∈Y⁡(k), and so if we attempt to glue the maps Ui→y→Y, then we obtain the map C→y∈Y and thus 𝒱 must have been trivial in the first place. Therefore, there is no fine moduli space.

The next moduli problem we will consider is the moduli problem M1,1 of elliptic curves over k. These are curves of genus 1 with one marked point. To generalize this to families, we will take

Then a morphism (a,b):(S,E,e)→(S′,E′,e′) is a cartesian diagram

that commutes with the sections e,e′. Here, the criterion of being Cartesian means that defining the functor on morphisms is well-defined, because pullbacks are unique.

6.2 Dealing with Nontrivial Automorphisms

Unfortunately, there are nontrivial automorphisms of elliptic curves. Over ℂ (or any algebraically closed of characterstic p≠2,3), we can write elliptic curves with the equation y2=x⁢(x−1)⁢(x−λ). We then have an action of S3 on Spec⁡ℂ⁢[t](t⁢(1−t)), where we have

Then the orbit of λ is the set

Generically, this has six points, but sometimes there are fewer (for example λ=−1). Then we have the j-invariant

and when λ,λ′ are in the same orbit, we have j⁡(λ)=j⁡(λ′). If we solve for j⁡(x)=j⁡(λ), we obtain a polynomial of degree 6, and thus generically there are six roots. Then we can see that all solutions belong to the orbit of λ. Therefore, we have

This discussion can be found in more detail in Hartshorne, Chapter 3, Section 4. We then see that E≃E′ if and only if j⁡(E)=j⁡(E′). Therefore, we have a bijection M1,1⁢(ℂ)≃𝔸1⁢(ℂ). In fact, Silverman proves this identity for all fields.

However, we can show that M1,1 has no fine moduli space. Consider the elliptic curves Y2⁢Z=X3+t⁢Z3 and Y2⁢Z=X3+t⁢Z3 over ℂ⁡(t). We see that E and E′ are not isomorphic at the level of ℂ⁡(t)-points because ℂ⁡(t) contains no roots of t. However, if we base change to ℂ⁡(t1/6), then we obtain an isomorphism of the two curves by

This gives us a contradiction because for any field extension L/K and scheme X, we have X⁡(K)⊂X⁡(L).

There are two ways to resolve this problem of nontrivial isomorphisms:

6.2.1 Stacks

We will in some sense, allow automorphisms. We can enlarge the category of schemes to algebraic spaces or stacks, in which we can restate the same problem. However, if we work with stacks, we will have to consider functors valued in the 2-category of groupoids. For a functor M:𝖲𝖼𝗁/Sop→𝖲𝖾𝗍, we will consider the functor

where 𝒞 is the category with objects the geometric objects over T and morphisms the T-morphisms of such objects. The study of such things often becomes very topological. For elliptic curves, we can also rigidify our elliptic curves, we obtain a functor Rell:𝖲𝖼𝗁/Sop→𝖲𝖾𝗍. This can then be represented by a scheme.

In this case, a rigidified elliptic curve is an elliptic curve with nowhere vanishing section α∈H0⁢(E,ΩE/S). Then automorphisms of rigidified elliptic curves are always trivial. Then writing E in Weierstrass form, any map E→E′ over Spec⁡R extends to a map ℙ2→ℙ2. This is represented by a matrix

where u∈R×. Then we can find a global nowhere vanishing section ω=Z2⁢Y⁢d⁢(XZ). If A preserves ω, we see that

and thus u=1. Thus we have killed all automorphisms. In fact, Rell is represented by an affine scheme Spec⁡ℤ⁢[a4,A6]Δ, where Δ is the discriminant. Note here that a4,a6 are here because under A an automorphism, we make the substitution X=u2⁢X,Y=u3⁢Y, and a4,a6 become u4⁢a4,u6⁢a6.

Now consider the action of 𝔾m on Rell that scales the holomorphic 1-form by 1μ (weight −1). Then we can define the morphism Rell→𝜋M1,1 that simply forgets ω. Then we can define a quotient Rell/𝔾m and this gives a map to M1,1. In this case, this is not a scheme, so we should really write the quotient stack [Rell/𝔾m].

6.2.2 Coarse Moduli

Some of us are plebs who can only think about schemes.1 We will loosen our requirement for representability and find a scheme which approximates the problem as well as possible. We say a scheme X coarsely represents M if

  1. 1.

    ψ:M→hX is a bijection over algebraically closed fields.

  2. 2.

    For all ϕ:M→hY such that the first item holds, there exists a unique morphism hX→hY that makes the diagram commute.

There are several problems that may arise: either X may be non-separated or it may have infinitely many components. In addition, even if X is separated with finitely many components, it may not be a scheme. Here are some more problems:

  1. 1.

    If we take varieties with no extra data, we run into the problem of having many automorphisms. This can be fixed by adding the datum of a polarization, which rigidifies the problem. For X/k proper, we can consider Picτ⁡(X)⊂Pic⁡X composed of line bundles such that ℒm∈Pic0⁡(X) for some m. Then an inhomogeneous polarization is a coset formed by ample line bundles. For example, for C→k of genus at least 2, the dualizing sheaf ωC is ample, and then ωC3 is very ample.

In Mumford, if we follow references [29] and [135], we see that requiring smoothness makes the the coarse moduli space X is separated and has finitely many components. However, smoothness is an open condition, so our moduli space will not be compact and thus will not have a good intersection theory. For example, take the intersection of two lines in the plane. Two parallel lines do not intersect in 𝔸2, but they do intersect at infinity in ℙ2. Thus, we will define the space M¯g, which is the moduli space of stable curves, where we allow nodes. In stack form, this becomes a proper Deligne-Mumford stack..

7 Morena (Nov 6): GIT and Coarse Moduli Problems (Part 2)

7.1 Moduli of Curves

Recall that the functor Mg sends a scheme S to the set of isomorphism classes of smooth curves of genus g over S. More precisely, these are flat and proper morphisms f:C→S such that geometric fibers are curves of genus g. Then recall that morphisms are given by cartesian diagrams. We will now rigidify this problem to find a coarse moduli space.

Theorem 7.1.1 (Olsson, Lemma 8.4.6).

Let f:C→S be a curve of genus g⁢g⁢e⁢q⁢2. Then consider ωf=ΩC/S1.

  1. 1.

    f∗⁢ωf3 is locally free of rank 5⁢g−5.

  2. 2.

    The map f∗⁢f∗⁢ωf⊗3→ωf⊗3 is surjective and C↪ℙ⁡(f∗⁢ωf⊗3) is a closed embedding. Here, ℙ⁡(vector bundle) is simply the relative Proj of the sheaf over S.

Definition 7.1.2.

Define the functor H3:𝖲𝖼𝗁/ℤop→𝖲𝖾𝗍 of rigidified smooth curves by

where the map of projective spaces is an S-isomorphism. Here, morphisms are cartesian squares

such that

commutes and the right square is commutative.

Fortunately, there are no nontrivial automorphisms, so there is hope that this functor is represented by a scheme.

Theorem 7.1.3.

The functor H3 is represented by a locally closed subscheme of Hilb¯ℙ5⁢g−6p⁡(t), where

by the Riemann-Roch theorem. The embedding H3→Hilb¯p is given by

Now note that Hilbp⁡(t) is a projective scheme. Then the identity map corresponds to the universal closed subscheme ℨ↪ℙHilb5⁢g−6. However, 𝒵→Hilb is not smooth (but it is flat and proper), so we can take W⊂𝒵 the non-smooth locus, so then we can take Hilb∖f⁡(W) parameterizing smooth subschemes.

Now to impose geometrically connected fibers, we use Stein factorization (Stacks Project, 37.48.5). This says many things, but what we need is that if we have f:X→S proper, then we can find S′→S such that f′:X→S′ is proper and has geometrically connected fibers and such that S′→S is integral and is an isomorphism over an open subscheme of S′. This implies using the Hilbert polynomial that dim=1 and that the genus is g. Thus we have an open subscheme H′⊂Hilb parameterizing smooth curves.

Now the identity H′→H′ gives a universal smooth curve

To recover ℙ⁡(σ), we need this to factor through ℙ⁡(f∗⁢ωf⊗3). Therefore, we need the pushforwards of ΩC′/H′⊗3 and i∗⁢(𝒪ℙ5⁢g−5⁢(1)) to pushforward to the same sheaf under f∗′. Here, the two sheaves give morphisms H′⇉PicC′/H′ and the square

is cartesian. Then Pic is separated, so Δ is a closed embedding. Thus we have a closed subscheme H2⊂H′. Now we ask for a surjective map

This is the same as the cokernel vanishing, so now we have H3⊂H2 an open subscheme. This is also Proposition 5.1 in Mumford.

Now we have shown that the rigidified moduli problem H3 is representable. Then we consider the map π:H3→Mg that forgets ℙ⁡(σ), and so we have an action of P⁢G⁢L5⁢g−5 on H3 by postcomposition with ℙ⁡(σ). Now we define Mg′=H3/P⁢G⁢L=(S↦H3⁢(S)/P⁢G⁢L⁢(S)). Then we have a map H3/P⁢G⁢L→𝐼Mg compatible with π. Here I is a “principal bundle” and so I is injective and locally surjective, which means that for all S there exists α∈Mg⁢(S) and Zariski cover {Si→S} such that α|Si∈Im(I).

Then checking π⁡([f,ℙ⁡(σ)])=π⁡([f′,ℙ⁡(σ′)]), we use the large commutative diagram (7.1) to see that f,f′ are related by some element of P⁢G⁢L. Then there exists a geometric quotient H3/P⁢G⁢L and this is a coarse modulu space for Mg. Thus we have a bijection between maps Mg→N and P⁢G⁢L-invariant maps H3→N. This is equivalent to the orbits of H3⁢(Ω) to be bijective with N⁡(Ω), where N=H3/P⁢G⁢L.

Now recall from Theorem 3.4.5 that if X/k is of finite and G is a reductive group with ℒ∈PicG⁡(X), then there is a universal categorical quotient of Xss⁢(ℒ) and a uniform geometric quotient of Xs⁢(ℒ) (in characteristic 0, uniform is upgraded to universal). Now we want to express H3 as the locus of stable points of something, and in particular as X(0)s⁢(ℒ).

To do this, we will introduce the Chow scheme. The Chow variety 𝒞n,r,d parameterizes subvarieties of ℙn of dimension r and degree d. This is defined by

Then projecting Γ down to (ℙn)r+1, we obtain a hypersurface defined by the Cayley form FX. Then the coefficients of FX map to 𝒞⁡(X)∈ℙν.

Now we have a map ϕ:H3→𝒞5⁢g−6,1,6⁢g−6. Over k, we use a push-pull with respect to

where 𝒳 is defined by first defining H=∑xi⊗Ui⊂ℙn×ℙm, then pulling considering the maps p12,p13:ℙn×ℙn×ℙn→ℙn×ℙn, and taking 𝒳=p12−1⁢(H)×p13−1⁢(H).

Now recall from Proposition 1.14 of Mumford that X¯s⁢s⁢(ℒ¯)=Xs⁢s⁢(ℒ)¯ and similarly for the stable locus. Thus we can reduce to the case where k is algebraically closed, and by Proposition 1.18 in Mumford, if f:X→Y is G-equivariant, then f is quasi-affine. Thus f−1⁢(Y(0)s⁢(ℒ))⊆X(0)s⁢(f∗⁢ℒ). Then by Theorem 6.2 in SGA 8, an injective morphism of finite type is quasi-affine. If we require that f is finite, X is proper, and ℒ is ample, the inclusion becomes an equality by Proposition 1.19 in Mumford.

Now recall that we have ϕ:H3→𝒞5⁢g−6,1,6⁢g−6. Then ϕ is quasi-affine and ϕ−1⁢(𝒞(0)s⁢(𝒪𝒞⁢(1)))⊂(H3s)(0)⁢(ϕ∗⁢𝒪𝒞⁢(1)). Then by Theorem 4.5 of Mumford, we use the Hilbert-Mumford stability criterion to see that for any smooth curve γ, the point 𝒞⁡(γ) is stable if

  1. 1.

    γ is not contained in a hyperplane.

  2. 2.

    γ is defined by a complete linear system.

  3. 3.

    g≥1 and deg⁡γ≥3⁢g. This is equivalent to g≥2.

Thus the image of ϕ is contained in the properly stable locus and thus we have H3 is equal to the properly stable locus. Thus we obtain a geometric quotient.

8 Kuan-Wen (Nov 13): GIT and the moment map

8.1 Moment maps

Let (X,ω) be a symplectic manifold. Here, this means that X is a smooth manifold and ω is a closed nondegenerate 2-form on X. Now Let K be a compact Lie group acting on X symplectically, which means k∗⁢ω=ω for all k∈K. Define the Lie algebra of K to be 𝔨.

Definition 8.1.1.

The moment map μ:X→𝔨∗ is a map which is

  1. 1.

    Equivariant. Here, we use the coadjoint action on 𝔨∗.

  2. 2.

    The map dμ:T⁢X→𝔨∗ satisfies the identity

    for any x∈X,a∈𝔨 and ax=ddx⁡exp⁡(t⁢a)⁢x|t=0.

Now we will discuss existence and uniqueness of the moment map. For uniqueness, we know that

  1. 1.

    μ is unique up to addition of a constant in 𝔨∗.

  2. 2.

    If K is semisimple, then the only fixed point of the coadjoint action is 0, and thus μ is in fact unique.

  3. 3.

    If we require ∫Xμ⁢ωn=0 for X compact, then we obtain a unique moment map.

Now we will discuss existence. The following result is from reference [723] of Mumford’s book.

Theorem 8.1.2 (Marsden-Weinstein).

If K is semisimple, then the moment map μ exists.

Next, it is known that if H1⁢(X,ℚ)=0 and K is connected, then the moment map exists. If K is a torus, this comes from a differential equation, but the obstruction comes from H1⁢(X,ℚ) and so we are fine here. In general, we have an exact sequence

so by the previous discussion, we now have the existence of a moment map.

Example 8.1.3.

Consider the standard action of U⁡(n+1) on ℙn. Then the mment map μ:ℙn→𝔲⁢(n+1)∗ is given by

where x∗ is any lift of x in ℂn+1. This is unique up the the addition of the trace map tr⁡(−), which is the nontrivial fixed point of the coadjoint action. This follows from the fact that all characters of U⁡(n+1) are powers detm:U⁡(n+1)→S1 of the determinant map. To see invariance, note that

Using the Fubini-Study metric on ℙn and the fact tha tthe action is transitive, we will compute at the point [1:0:⋯:0] with coordinates [1:x1:⋯:xn]. Recall that the Fubini-Study form is given by

Now we can directly compute

This gives the desired result.

Now suppose X⊂ℙn and K acts on X with an action induced from a linear action on ℙn. Then we can assume that K acts unitarily, so then se can write the moment map as

This can be written explicitly as

where ρ:K→U⁡(n+1) is the linear action.

We will now discuss quotients in the symplectic category.

Theorem 8.1.4 (Marsden-Weinstein).

Let K act on (X,ω) and μ:X→𝔨∗ be the moment map. Let η∈𝔨∗ be a fixed point of the coadjoint action. If K acts freely and properly on μ−1⁢(η), then the quotient μ−1⁢(η)/K is a symplectic manifold of dimension dimX−2⁢dimK. If ι:μ−1⁢(η) is the inclusion and π is the projection, then we have the identity π∗⁢ωred=ι∗⁢ω. If η is not fixed, we can replace K by the stabilizer of η.

Proof.

We have the equality

This comes from the differential equation in the definition of the moment map, and then we also have

To get the dimension equality, we have a short exact sequence for any W⊆Tx⁢X given by

and thus dimTx⁢X=dimW+dimWωx. Now we replace W with Tx⁢(K⋅x) to obtain the desired result. On the other hand, we also have

By these computations, we know that η is a regular value if and only if the action of K on μ−1⁢(η) is locally free.

Now Tx⁢(K⋅x) is an isotropic subspace of Tx⁢X because we have

Next, for isotropic subspaces I of (V,ω), there is a symplectic form induced by ω on Iω/I. This gives a symplectic form on Tx⁢(K⋅x)ωx/Tx⁢(K⋅x)=Tx⁢μ−1⁢(η)/Tx⁢(K⋅x), and so we have a symplectic form on the quotient.

Finally, to prove closedness of this form, this follows from the fact that

and the fact that π∗ is injective. □

8.2 Comparison of symplectic quotients and GIT quotients

Now let G is reductive with a linear action ρ:G→G⁢L⁢(n+1) such that G acts on some projective variety X⊆ℙn. Now let K⊂G be a maximal compact subgroup.

Theorem 8.2.1.
  1. 1.

    A point x∈X is semistable if and only if OG⁢(x)¯∩μ−1⁢(0)≠∅.

  2. 2.

    There is an inclusion μ−1⁢(0)↪Xss (by Proposition 2.2 in Mumford). This induces a homeomorphism μ−1⁢(0)/K≃X⫽G with the GIT quotient.

Proof.

First, for all v∈ℂn+1, define pv⁢(g)=‖g⁢v‖2. Then from Kempf-Ness (reference [175] in Mumford), we have

  1. (i)

    All critical points of pv are minima.

  2. (ii)

    If pv attains a minimum, it does so on exactly one double coset in K\G/Gv.

  3. (iii)

    pv attains a minimum if and only if the orbit of v is closed in ℂn+1.

Also, because we can write the moment map explicitly, we have dpx∗⁢(g)=0 if and only if μ⁡(g⁢x)=0. Finally, we know that if OG⁢(x∗) is closed in ℂn+1, then x is semistable and OG⁢(x) is closed in Xss. To prove the second part, if y∈OG⁢(x)¯∩Xs⁢s, then there exists f invariant such that y∈Xf. This means x∈Xf, so we can choose x∗,y∗ such that f⁡(x∗)=f⁡(y∗)=1. Multiplying by roots of unity, we may assume that y∗∈OG⁢(x∗)¯=OG⁢(x∗). Projecting down, we have y∈OG⁢(x).

Using all of these observations, we can prove the two parts of the theorem.

  1. 1.

    If x∈Xss, then OG⁢(x)¯⊂ℂn+1∖{0} contains some closed orbit OG⁢(y). This implies that OG⁢(x)¯∩μ−1⁢(0)≠∅. The converse simply follows from the observations and Proposition 2.2 in Mumford.

  2. 2.

    We know μ−1⁢(0)/K→X⫽G is a continuous map from a compact space to a Hausdorff space. It suffices to show taht this is a bijection. Then recall that Xss→ϕX⫽G satisfies the condition that ϕ⁡(x)=ϕ⁡(y) if and only if their orbit closures intersect, so by the first part of this theorem, we have surjectivity.

    For injectivity, if μ⁡(x)=0=μ⁡(y) but x∉OK⁢(y), then we know that x∉OG⁢(y) by the previous discussion, but also OG⁢(x),OG⁢(y) are closed in Xss, which gives a contradiction. □

We conclude with some remarks. It is easier to compute the cohomology of μ−1⁢(0)/K than it is to compute the cohomology of X⫽G. To do this, if we consider the Morse function f⁡(x)=‖μ⁡(x)‖2, where the norm comes from the Killing form, then we have a Morse stratification {Cβ,m} and then we define the functions

This allows us to compute the cohomology.

9 Alex (Nov 20): GIT quotients of symplectic manifolds

9.1 Kähler and Hyperkähler quotients

Let (X,ω) be a compact Kähler manifold and G be a complex reductive group. Then K⊆G is a maximal compact subgroup and suppose that the action of K on X preserves ω. Let μ:X→𝔨∗ be the moment map. Recall the Marsden-Weinstein theorem from last time. Define

This should be thought of as the stable locus in the symplectic quotient.

Theorem 9.1.1.

There exist a Kähler structure on

where the equivalence here is diffeomorphism and X(0)min⊆X is open.

To sketch a proof of this, we first note that if x∈μreg−1⁢(0), then if we call the quotient Y, we have

This gives us an almost complex structure on the quotient, and by reference [624] of Mumford, this is integrable and compatible with the symplectic structure. Here, the second isomorphism comes from the identity

Now we also need to define this mysterious Xmin. Define the function f⁡(x)=‖μ⁡(x)‖2, where we fix an invariant inner product on 𝔨∗. Then define

This should be thought of as the normal bundle of μ−1⁢(0). In fact, we have Xmin=Xss and then we have X(0)min=X(0)s.

We now discuss the hyperkähler case.

Definition 9.1.2.

A manifold (X,I,J,K,g) is hyperkähler if I,J,K are complex structures and I⁢J=K=−J⁢I. This gives us an entire S2 of complex structures.1

The standard picture for a hyperkähler manifold is the space ℍn. Note that a hyperkähler structure gives three Kähler structures. Now if we have the corresponding moment maps μ1,μ2,μ3, the hyperkähler quotient is

Example 9.1.3.

An interesting hyperkähler quotient is the space of anti-self-dual instantons, or principal S⁢U⁢(ℓ) bundles over S4. Alternatively, this is the space of holomorphic S⁢L⁢(ℓ)-bundles over ℙ2 with c2=k. There is a construction of Atiyah, Drinfeld, Hitchin, and Manin (ADHM).

This is a hyperkähler quotient of ℍk⁢(k+ℓ). We consider elements of ℍk⁢(k+ℓ) as complex matrices α1,α2,a,b complex matrices of size k×k,k×k,ℓ×k,k×ℓ satisfying

  1. 1.

    [α1,α2]+b⁢a=0.

  2. 2.

    [α1,α1∗]+[α2,α2∗]+b⁢b∗−a⁢a∗=0.

  3. 3.

    The matrix

    is injective and the matrix (λ⁢Ik−α1α2−μ⁢Ikb) is surjective.

Then define the action of U⁡(k) on ℍk⁢(k+ℓ) by

All of this is in a paper by Donaldson.

9.2 Desingularization

There are two reasons that μ−1⁢(0)/K can have singularities:

  1. 1.

    x∈μ−1⁢(0) is not a regular point.

  2. 2.

    The action of K on μ−1⁢(0) is not a free action.

Remark 9.2.1.

If dμ⁢(x) is surjective for all x, then X(0)s=Xss.

Note that not all finite quotient singularities cannot be resolved by equivariant blowups.

Theorem 9.2.2.

If X(0)s≠∅, then there exists X~ which is a blowup of X along G-invariant nonsingular subvarieties with a suitable G-linearization such that X~ss=X~(0)s and such that X~⫽G→X⫽G is a birational morphism and X~⫽G only has finite quotient singularities.

To prove this, if we let R⊆G be a reductive subgroup, then we define ZRss=(Xss)R. Now if we define r=max⁡{dimR∣ZRss≠∅}, we define R⁡(r) to be a choice of representatives of conjugacy classes of reductive subgroups such that ZRss≠∅ and dimR=r. Then we outsource the fact that the set

is a disjoint union of G-invariant subvarieties and thus we can set Y to be the blowup of X along these subvarieties. Now this procedure decreases r, and so by a finite process we end up with only finite stabilizers.

Remark 9.2.3.

This quotient is sometimes quite nice for algebraic geometry but not for symplectic geometry. Let G=G⁢L1=ℂ∗ act on ℙ2 by

Then we know that ℙ2⫽G≅ℙ1 because the ring of invariant functions is generated by x⁢z2,y⁢z.

However, if we take the symplectic quotient, we have the moment map

and thus we have

Thus we have nontrivial stabilizers, where for example the point (1,0,2) is fixed by ℤ/3⁢ℤ and the point (0,1,1) is fixed by ±1. Therefore the symplectic quotient is an orbifold, and it can be resolved by using a Poisson bracket. To do this, the space μ−1⁢(0)/K has a stratification given by conjugacy classes of stabilizers such that each stratum is a symplectic manifold.

9.3 Yang-Mills functional

In some sense this is an infinite-dimensional version of what we were doing before. Here, (M,ω) is a compact Kähler manifold and E→M is a holomorphic vector bundle with a Hermitian metric.

Definition 9.3.1.

A connection A on E is a map dA:Ω0⁢(E)→Ω1⁢(E) satisfying the Leibniz rule:

Note that by extending the Leibniz rule, we can define dA:Ωp→Ωp+1 by

Note that if A,A′ are connections, then dA⁢(f⁢s)−dA′⁢(f⁢s)=f⁡(dA⁢s−dA′⁢s) and thus we have dA−dA′∈Ω1⁢(End⁡E). Also note that dA∘dA=FA∈Ω2⁢(End⁡E). This is called the curvature form. Now if A is a unitary connection, then everything lives in Ω∗⁢(𝔤E), where 𝔤=𝔲⁡(n). Furthermore, the space of connections is an affine space 𝒜. Now we define the Yang-Mills functional to be

Then the Yang-Mills equation is dA(∗FA)=0, where ∗:Ω∗⁢(𝔤)→Ω2⁢n−⁣∗⁢(𝔤) is the Hodge star. The gauge group is the group 𝒢=Aut⁡(E) of unitary automorphisms of E, and the corresponding Lie algebra is 𝖫𝗂𝖾⁡(𝒢)=Ω0⁢(𝔤E).

Now a complex structure on E gives us an operator ∂¯E:Ωp,q⁢(E)→Ωp,q+1⁢(E) with kernel the holomorphic sections. Then there exists a unique unitary connection A such that dA(0,1)=∂¯E. Because ∂¯E2=0, this implies that FA∈Ω1,1⁢(E). We have no identified holomorphic structures on E with the space 𝒜1,1 of unitary connections with (1,1) curvature. Finally, there exists a 𝒢ℂ action such that holomorphic structures are biholomorphic if and only if they are in the same orbit. Finally, the Hermitian Yang-Mills connections are those that minimize the Y-M functional and those exist if and only if the bundle is slope-stable.

10 Patrick (Dec 04): Everything you need to know about Morena’s lectures

Note: These are the speaker’s notes. Throughout this lecture, we work over an algebraically closed field k of characteristic 0.

10.1 Chow Varieties

Recall that the Chow variety 𝒞n,r,d parameterizes algebraic cycles of ℙn of dimension r and degree d. Denote the dual projective space by ℙ̂n. Now points of ℙ̂n are hyperplanes in ℙn, and so we define

Now projecting ΓX→(ℙ̂n)r+1, obtain a hypersurface defined by a Cayley Form FX. Now the coefficients of FX give us a point 𝒞⁡(X)∈ℙν.

Now we will discuss the stability of the Chow point of a nonsingular space curve. There are two key things we need to know:

  1. 1.

    The Chow form is compatible with specialization.

  2. 2.

    How to compute the Chow form of a cycle of linear subspaces.

For the second, we will do this explicitly. Let X0,…,Xn be coordinates on ℙn and Y0,…,Yn be coordinates on ℙ̂n. Now if LI is the linear subspace defined by Xi=0,i∈I, we can compute

Then if λ is the 1-PS of S⁢L⁢(n+1) given by diagonal matrices with weights r0,…,rn, we have

Now for a smooth curve C in ℙn, we will find the cycle Z that is the specialization of C by λ⁡(α) as α→0. This will typically be a sum of lines and then we can compute μ⁡(C,λ) easily. Assume that the weights of λ are r0>⋯>rn. We will assume that C is not contained in any hyperplane. Now recall that points on (xi=⋯=xn=0) but not on (xi−1=⋯=xn=0) specialize to [0,…,1,…,0], where we have xi−1=1 and xj=0 for j≠i−1. Thus almost all points of C specialize to [0,…,0,1] and therefore the specialization of C is determined by C∩(xn=0).

Now consider the composition Gm×C→S⁢L⁢(n)×ℙn→𝜎ℙn, which now defines a rational map 𝔸1×C⇢ℙn with basepoints {0}×(C∩(xn=0)). Suppose this map is given at the point pi∈C∩(xn=0) by (f0,…,fn)≕ℐi (abusing notation and treating this list as an ideal). Then if V is the normalization of the blowup of 𝔸1×C at the ideals ℐi. Now on V, the divisors {α}×C specialize to the divisor given by the zeroes of the function α on 𝔸1. Thus Z is the image of the union of the proper transform of {0}×C an the exceptional divisors ξi. Because 0×C is taken to a point, we have Z=∑iF′⁢(ξi), where F′:C→ℙn is the replacement of the rational map 𝔸1×C⇢ℙn.

Now we need to describe F′⁢(ξi). To do this, we will represent ι:C→ℙn by a power series. If ti is a local coordinate at pi, then suppose that ι is given by

Now at least one sk,i=0 and thus the rational map is given by

where α is the coordiante on 𝔸1.

Lemma 10.1.1.

Let V0 be a smooth surface and p∈V0. Let x,y are local coordinates and

for integers ri,si. Let V be the normalization of the blowup of V0 at 𝔘. If we plot the points (ri,si) on ℝ2 and construct the Newton polygon, then there is one exceptional divisor for each edge of the Newton polygon.

Now if we compute the Newton polygon of the (rk,sk,i), then the divisors correspond to slopes a=p/q such that for some e,f,

This allows us to compute

where le,f joins the two points (xi=0,i≠e) and (xj=0,j≠f). WE are now able to compute

By combinatorial magic, can be rewritten as

where sj,i′=mine≥k⁡se,i. Here is some more magic:

Lemma 10.1.2.

The number μ⁡(C,λ)>0 if the ej=∑i=1νsk⁡(j),i′ satisfy

  1. 1.

    e0=0,en=n+g,e0≤⋯≤en;

  2. 2.

    ei≤i for i=0,…,n−g;

  3. 3.

    ei≤i+g for i=n−g+1,…,n;

and if n≥2⁢g and g≥1 and r0≥⋯≥rn, ∑ri=0, and not all ri=0.

This then tells us that

Theorem 10.1.3.

Let C be a smooth curve in ℙn. Then C is stable with respect to P⁢G⁢L⁢(n+1) if

  1. 1.

    C is not contained in any hyperplane;

  2. 2.

    C↪ℙn is defined by a complete linear system;

  3. 3.

    g⁡(C)≥1 and deg⁡C≥3⁢g.

The proof of this result relies on Riemann-Roch and the Weierstrass Gap Theorem (about linear systems associated to points).

Remark 10.1.4.

If g≥1 and n=2,3, then any smooth curve spanning ℙn is stable. However, there are elliptic curves in ℙ4 that are unstable.

10.2 Construction of Mg

Recall that Chow points are divisors in (ℙ̂n)r+1 that come from sections of the sheaf ⨂ipi∗⁢(𝒪⁡(d)). Denote the projective space of such divisors by Divd,…,dn.

First we will construct a morphism ϕ:HilbℙnP⁡(x)→Divd,…,dn for any Hilbert polynomial of the form P⁡(x)=d⁢xrr!+O⁡(xr−1). This morphism will send a subscheme to its Chow point and will be P⁢G⁢L⁢(n+1)-equivariant. The construction will define a functor from the Hilbert functor to the functor represented by Div. Essentially, to a flat subscheme ℤ⊂ℙn×S→S with Hilbert polynomial P⁡(x), we will assign a flat effective Cartier divisor D⊂(ℙn)2×S→S which induces divisors of degree (d,d) on the fibers over S.

Define H⊆ℙx0,…,xnn×ℙy0,…,ynn be the hypersurface given by the vanishing of the polynomial x0⁢y0+x1⁢y1+⋯+xn⁢yn∈H0⁢(p1∗⁢(𝒪⁡(1))⊗p2∗⁢(𝒪⁡(1))). Now consider the variety ℙn×ℙ̂n×ℙ̂n and the morphisms

We will define ℋ≔p12∗⁢(H)∩p13⁢(H). We now have the following facts:

  1. 1.

    Points of ℋ are triples (x,h1,h2), where x∈h1,h2.

  2. 2.

    p1:ℋ→ℙn is a fiber bundle with fiber ℙn−1×ℙn−1.

Now consider the diagram

Now given a closed subscheme Z⊂ℙn×S flat over S, we will compute a Cartier divisor. To do this, let Z∗ be the fiber product of the following diagram:

Now we write D=Div⁡(q), where the Div of a morphism was defined in Section 5.3 of Mumford. Here if f:X→Y is a projective morphism of Noetherian schemes such that

  1. 1.

    f is of finite Tor-dimension. This means that for all x∈X, the ring 𝒪x,X is a 𝒪f⁡(x),Y-module of finite Tor-dimension, and this dimension is bounded over all x;

  2. 2.

    f−1⁢(x) is empty if x has depth 0 and finite if x has depth 1;

then there is a canonical effective Cartier divisor Div⁡(f) whose support is contained in f⁡(X).

Therefore, we need to check that

  1. 1.

    q is of finite Tor-dimension;

  2. 2.

    Div⁡(q) is a relative divisor over S.

  3. 3.

    When S=Spec⁡Ω for an algebraically closed field Ω, q has empty fibers over points of depth 0 and finite fibers over points of depth 1.

  4. 4.

    When S=Spec⁡Ω, the divisor Div⁡(q) has degree (d,d).

The first item follows from

Lemma 10.2.1 (Fogarty).

Let

be morphisms of finite type between Noetherian schemes. Suppose q is a smooth morphism with fibers of dimension n and that p is flat. Then f has Tor-dimension at most n.

Proof of this follows from SGA 2, Theorem 4.15. Next, 2 follows from 3. The proof of 3 and 4 is apparently the “classical theory of Chow forms.” Because dimZ=1, we know that there exists some set of hyperplanes h1,h2 such that h1∩h2∩Z=∅. Thus q−1⁢(y) is empty when y is the generic point. If y is the generic point of a divisor, then the set of hyperplanes h1,h2 such that h1∩h2∩Z=1 is the set of closed points on a subset W⊂ℙn×ℙn of codimension at least 2, and this is also classical.

Finally, to show that Div⁡(q) is of degree (d,d), we will show that it is the Chow form of the cycle of 1-dimensional components of Z.

Definition 10.2.2.

Let X be of finite type of dimension r over Spec⁡Ω. Let X1,…,Xk be the components of X and Zi the underlying set of Xi. Write xi be the generic point of Xi and 𝒪i=𝒪xi,Xi. This is an Artinian local ring of length ri. Now define

If X is an effective Cartier divisor, then this construction recovers the corresponding Weil divisor.

Now recalling the diagram

we define the Chow divisor 𝒞Z of Z by 𝒞Z=(p23)∗⁢p1∗⁢[Z]. The following lemma will complete the construction of our morphism ϕ.

Lemma 10.2.3.

Let X,Y be smooth varieties and f:X→Y be a projective morphism. Let Z⊂X be a closed subscheme such that dimZ=dimY−1. Let g=f|Z. Then f∗⁢[Z]=[Div⁡(g)].

The rest of the construction of Mg can be found in Morena’s second lecture.

11 Nicolás (Dec 11): The moduli space of stable curves

11.1 Introductions

Recall the functor ℳg⁢(S)={f:C→S∣flat, proper, fibers smooth curves of genus g}. We constructed a coarse moduli space Mg, which is a smooth quasiprojective variety contained in some ℙN⫽P⁢G⁢L⁢(n+1). This allows us to compactify our moduli space in a natural way. We want our compactification to have a geometric meaning, which means some meaning in terms of moduli problems.

Example 11.1.1.

Let us consider the moduli problem ℳ0,4 of distinct points (p1,p2,p3,p4)∈(ℙ1)4 up to projective isomorphism. Then we can embed M0,4⊂(ℙ1)4⫽P⁢G⁢L⁢(2). Recall that (P1,P2,P3,P4) are semistable as long as no three points are equal. Thus we can consider the semistable orbits in the GIT quotient as paramerizing tuples with repeated points.

11.2 Stable Curves

From here, we will follow the book Moduli of Curves by Harris and Morrison.

Definition 11.2.1.

A stable curve C is a complete connected curve with only finitely many nodal singularities and finitely many automorphisms.

Recall that if Γ⊆C is a component and g⁡(Γ)≥2, then we know Aut⁡(Γ)≤84⁢(g−1). When g⁡(Γ)=1, then Γ has infinitely many automorphisms, but if we choose a marked point, then Γ has finitely many automorphisms. When g⁡(Γ)=0, we need to fix three points. In the smooth case, this kills all automorphisms.

Theorem 11.2.2 (Deligne-Mumford-Knudsen).

There exists a coarse moduli space M¯g of stable curves for g≥2 that is a projective variety. Here, g is the arithmetic genus of the curve.

We may alternative define g as follows. If C=C1∪C2∪⋯∪Cr and there are δ nodes, we can define

The previous theorem is one of the most important results in the history of the algebraic geometry of curves.1 By the valuative criterion of properness, we can fill in families of curves with missing fibers.

Proposition 11.2.3 (HM, Prop 3.49).

Let B be a smooth curve and fix 0∈B. Let 𝒳→B∗=B∖0 be a family of stable curves. Then there exists B′→B totally ramified at 0 and a family 𝒳′→B′ which extends 𝒳×BB∗ such that all fibers are stable curves. If two such coverings exist, they are dominated by a third, so the central fiber is unique.

Remark 11.2.4.

A reason this theory is well-known is because the theory of surfaces is well-understood. If we want to perform analogous constructions for the moduli of surfaces, we need to understand the minimal model program for threefolds.

11.3 Some Examples

In our setup, we may have some non-allowed singularities. Alternatively, we may have multiple fibers. In the first scenario, we can resolve the problem by blowing up. If we have multiple fibers, we need to remove the multiplicity by taking a base change. There is an explicit algorithm for this, described in [HM, p.124–126].

  1. 1.

    Begin with X→B, and let D=∑ai⁢Di be the fiber at 0.

  2. 2.

    Fix some prime m such that m∣ai for some i.

  3. 3.

    Let B′→B be locally given by t↦tm and consider the fiber product X×BB′. Take the normalization X′→X×BB′.

  4. 4.

    The map X′→X ramifies D≡M=∑a¯i⁢Di, where a¯i≡aimodm with 0≤ai¯<m. The new fiber can be computed as follows:

    • •

      Take the preimage of D.

    • •

      Multiply by m all components contained in the branch locus.

    • •

      Divide all components by m.

Example 11.3.1.

Consider a family of curves 𝒳→𝔸1 with smooth fibers for t≠0 but a triple point over 0. Define 𝒳⊂𝔸1×ℙ2 by the equation (locally)

where g vanishes to degree at least 4 at (0,0). Let C be the singular fiber over 0. We now blow up at the singular point, and the new fiber is D=C~+3⁢E. To remove the multiplicity, we choose m=3. Then X′→X is ramified over C~ and the fiber of E is a generically 3 to 1 cover E′ of E ramified at the intersection points with C~. By Riemann-Hurwitz, we have

and thus g⁡(E′)=1. Finally, we see that the fiber is 13⁢(3⁢C~+3⁢E′)=C~+E′.

Remark 11.3.2.

There are not many curves with automorphisms of order 3, and so we have j⁡(E′)=0. This uniquely determines E′.

Example 11.3.3.

Let Q be a smooth conic and F a general smooth quartic. Consider the family {Q2+t⁢F}. When t≠0, this is a generic quartic, but when t=0, the fiber is a double conic. Assume that Q∩F has 8 points (so all intersections are transverse). Then the total space 𝒳 is not smooth because around each intersection, we have an intersection that is locally x2+t⁢y, which is an A1 singularity.

Now we have the fiber 2⁢Q over 0 with 8 marked points. We blow up each of the A1 singularities and obtain the new fiber 2⁢Q+E1+⋯+E8 (here, each Ei is a (−2)-curve). Now choose m=2 and follow the procedure, and we obtain a new family 𝒳′→B′ ramified at E1+⋯+E8. Now the fiber above Q is a 2:1 cover Q~ with 8 ramification points. By Riemann-Hurwitz, we compute g⁡(Q~)=3. The new fiber is Q~+E1+⋯+E8. Then the Ei are now (−1)-curves because 0=Ei⋅(Q~+E1+⋯+E8)=1+Ei2, so we can contract them to resolve the instability of the Ei by Castelnuovo’s criterion. Once we contract the Ei, the new fiber is Q~.

Thus the central fiber of the family is a genus 3 hyperelliptic curve ramified over the intersection points Q∩F.

Remark 11.3.4.

We know that a generic quartic curve has 28 bitangents. Under the limit t→0, the lines become the 28 lines joining two points in Q∩F.