Patrick Lei

All notes

Deformation theory

Columbia, Fall 2021.

Deformation Theory Graduate Student Seminar
Fall 2021
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 Johan (Sep 24): Schlessinger’s paper

The paper by Schlessinger is titled Functors of Artin Rings. Throughout this lecture, k is a field, 𝒞 is the category of Artinian local k-algebras A,B,C,… with residue field k, and 𝒞̂ is the category of Noetherian complete local k-algebras R,S,… with residue field k.

Remark 1.0.1.

Every R∈𝒞̂ is of the form k⁡[[x1,…,xn]]/(f1,…,fm) by the Cohen structure theorem. Then R∈𝒞 if and only if and only if (f1,…,fm) contains (x1,…,xn)N for some N.

Remark 1.0.2.

In the paper, there is a more general setup, where Λ is a complete local Noetherian ring with residue field k. Then 𝒞Λ,𝒞̂ are defined analogously, which will allow things like Λ=ℤp.

The idea of deformation theory is to look at functors F:𝒞→𝖲𝖾𝗍.

Example 1.0.3.

Given R∈𝒞̂, we set hR:𝒞→𝖲𝖾𝗍 sending A↦Hom𝒞̂⁡(R,A). This is not necessarily representable because R∉𝒞 in general, but it is pro-representable.

Definition 1.0.4.

A functor F is pro-representable if F≃hR for some R∈𝒞̂.

Example 1.0.5.

Let M be a variety over k and m∈M⁡(k). Then define

It is easy to see that DefM,m⁢(A) is pro-representable by 𝒪̂M,m.

Observe that hR⁢(k)={∗} is a singleton. Also note that hR⁢(A×BC)=hR⁢(A)×hR⁢(B)hR⁢(C). Here, A×BC is the fiber product of rings and not the tensor product.

Now consider the following conditions on F: let A→B←C be a diagram in 𝒞 and consider the morphism

  • •

    (H1) The morphism (∗) is surjective if C↠B;

  • •

    (H2) The morphism (∗) is bijective if C=k⁡[ε]↠k=B;

  • •

    (H3) dimk(tF)<∞ (later, we will see that we need H2 for formulate this). Here, tF is the tangent space to F;

  • •

    (H4) The morphism (∗) is bijective if C↠B.

Example 1.0.6.

Fix a group G and a representation ρ0:G→G⁢Ln⁢(k). Now define

Better, we will define

In general these functors fail (H4) and Defρ0naive even fails (H2).

Namely, if H=ℤ and ρ0 is the trivial representation, then for Defρ0naive, we are looking at subsets of

This morphism is always surjective, but in general it is not injective.

For example, if A=k⁡[ε1],B=k,C=k⁡[ε2], we can look at elements of the form 1+ε1⁢T1+ε2⁢T2 and see that on the left we can only conjugate together, while on the right we can conjugate both T1,T2 arbitrarily. Here A×BC=k⁡[ε1,ε2]=k⁡[x1,x2]/(x12,x1⁢x2,x22).

Definition 1.0.7.

A natural transformation t:F→G of functors on 𝒞 is smooth if for all surjections B↠A the map F⁡(B)→F⁡(A)×G⁡(A)G⁡(B) is surjective.

Note that this is equivalent to the existence of a lift in the diagram below:

This definition is motivated by the following example: let f:M→N be a morphism of varieties over k. Let m∈M⁡(k),n=f⁡(m)∈N⁡(k). Then the following are equivalent:

  1. 1.

    DefM,m→DefN,n is smooth.

  2. 2.

    f is smooth at m.

Definition 1.0.8.

We say F has a hull if and only if F⁡(k)={∗} and there exists a smooth t:hR→F for some R∈Ĉ which induces an isomorphism tR≅tF.

Now we will say a bit about tangent spaces.

  1. 1.

    When F⁡(k)={∗}, then tF=F⁡(k⁡[ε]).

  2. 2.

    If F satisfies (H2) and F⁡(k)={∗}, then tF has a natural k-vector space structure. Here, H2 gives F⁡(k⁡[ε1,ε2])→F⁡(k⁡[ε])×F⁡(k⁡[ε]) is a bijection, and then we take ε1↦ε,ε2↦ε, which defines addition.

  3. 3.

    tR=Homk⁡(𝔪R/𝔪R2,k)=HomĈ⁡(R,k⁡[ε])=hR⁢(k⁡[ε])=t(hR).

Theorem 1.0.9 (Schlessinger).

Asssume that F⁡(k)={∗}. Then the conditions (H1),(H2),(H3) hold for F if and only if F has a hull. In addition, (H3) and (H4) hold if and only if F is pro-representable.

Very rough idea of proof of ⇒ for the hull case.

Let n=dimk(tF). Then (H2) and n<∞ imply the following: Let S=k⁡[[x1,…,xn]] and 𝔪=𝔪S=(x1,…,xn). We can find ξ1∈F⁡(S/𝔪2) such that

is an isomorphism.

Next, we will choose q≥2 and consider pairs (J,ξ) where 𝔪q+1⊂J⊂𝔪2 and ξ∈F⁡(S/J) such that ξ↦ξ1∈F⁡(S/𝔪2). Say that (J,ξ)≤(J′,ξ′) if J⊂J′ and ξ↦ξ′. Choose a minimal pair (J,ξ) for this ordering. We can choose Jq so that 𝔪q+1+Jq+1=Jq and ξq+1 maps to ξq for bookkeeping purposes.

Choose R=limS/Jq, which is a quotient of S. Set t:hR→F given by sending φ:R→A to the following: choose q such that φ factors as R→S/Jq→φqA and take ξq↦F⁡(φq)⁢(ξq)∈F⁡(A).

Finally, we must show that t is smooth. Consider the diagram

with B∋ξ~↦ξ∈A and S/Jq∋ξq↦ξ. First, choose φ:S→B making the diagram commute. We may increase q such that φ⁡(𝔪q+1)=0, so we now have φ¯:S/𝔪q+1→B. Now consider the fiber product S/Jq×AB and p⁢r1:S/Jq×AB→S/Jq, so we obtain φ¯′:S/𝔪q+1→S/Jq×AB. By (H1), we obtain some ξ~~∈F⁡(S/Jq×AB) mapping to ξ~ and ξq. We may now assume that B→A is a small extension, which means that dimkker⁡(B→A)=1, and thus p⁢r1 is a small extension. Therefore, either φ¯′ is surjective or its image maps isomorphically via p⁢r1 to S/Jq., so we have ψ which gives R→B lifting our given r→A.

The tricky part is to show that F⁡(ψ)⁢(ψq)=ξ~~, and this step is deliberately omitted. □

A generalization of this is as follows. Consider a functor ℱ:𝒞→𝖦𝗋𝗉𝖽. We say that ℱ satisfies the Rim-Schlessinger condition (RS) if

is an equivalence whenever C↠B. Let x0∈ℱ⁡(k) and set

where (x,α)≅(x′,α′) means that φ:x→x′ such that the diagram

commutes.

Theorem 1.0.10.

If ℱ has (RS) then ℱ¯x0 has (H1) and (H2). Therefore, if dimtℱ¯x0<∞ then ℱ¯x0 has a hull.

In this situation, ℱ¯x0 has (H4) if and only if AutA⁡(x)↠AutB⁡(x|B) whenever A↠B and x∈ℱx0⁢(A).

Example 1.0.11.

Let ℱ⁡(A) be the category of representations G↷A⊕n with morphisms being isomorphisms of representations. This has (RS).

Example 1.0.12.

Let ℱ⁡(A) be the category of smooth projective families of curves of genus g over A with morphisms being isomorphisms. This has (RS).

Returning to the example of representations, it turns out that tDefρ0=H1⁢(G,Mn×n⁢(k)), where G acts on Mn×n⁢(k) via ρ0 by conjugation.

Example 1.0.13.

Consider G=ℤ⊕ℤ and ρ0 to be the trivial representation on k⊕2. Then tDefρ0=H1⁢(ℤ2,M2⁢(k))=M2⁢(k)⊕M2⁢(k). Given two matrices A,B, we have the representation

We get a hull R with hR→Defρ0. We know that R is a some quotient of k⁡[[a11,…,a22,b11,…,b22]] with ρ looking like

and of course R is the quotient of the power series ring by the ideal generated by the coefficients of A⁢B−B⁢A.

2 Ivan and Cailan (Oct 1): Deformations of schemes

2.1 Deformations of affine schemes

We are looking for a Cartesian diagram of schemes

where π is flat and surjective and S is surjective. This is called an deformation of X over S. For the beginning of this lecture (the part given by Ivan), we are interested in S=Spec⁡A, where A∈𝒞∗ (this category was defined in the previous lecture). This case is called a local deformation, and in the face where A is Artinian, it is called an infinitesimal deformation.

For the ring theorists, we will make the following digression. Let A be a ring and I⊂A be an ideal with I2=0. Suppose that B¯ is an A/I-algebra, J is an B¯-module, and h:I→J is an A-module map. Then we are interested in a diagram

which we will call a deformation of A. Here are some interesting questions:

  1. 1.

    Is such a deformation unique?

  2. 2.

    If B¯ is flat over A/I, does that mean that B is flat over A?

Returning to the case of schemes, we will say that two deformations 𝒳,𝒳′ of X over S are isomorphic if there exists an S-isomorphism ϕ:𝒳′→𝒳 commuting with the inclusions of the central fibers X→𝒳,𝒳′.

Example 2.1.1.

The most basic example of a family is the trivial deformation

Definition 2.1.2.

A scheme X is rigid if all deformations of X are isomorphic to the trivial deformation.

Theorem 2.1.3.

If X is a smooth affine k-scheme and S=Spec⁡A for some local Artinian ring, then X is rigid.

Definition 2.1.4.

A closed immersion i:S0↪S of schemes is called a first (resp. nth) order thickening if the ideal sheaf ℐ=ker(i♭:𝒪S→𝒪S0) satisfies I2=0 (resp. In+1=0).

Definition 2.1.5.

A morphism f:X→S is called formally smooth (resp. unfamified, resp. étale) if for all first order thickenings i:T0→T of affine schemes and diagrams

there exists a lift u0~ (resp. there is at most one such u0~, resp. there exists a unique u~0).

Example 2.1.6.
  1. 1.

    Open immersions are formally étale. This is cleaer because T0,T have the same underlying topological space.

  2. 2.

    Closed immersions are formally unramified. This is clear because X→S induces an injection on T-points.

  3. 3.

    𝔸Sn→S is formally smooth. To see this, assume S=Spec⁡R is affine and then consider the corresponding lifting problem in commutative algebra.

Proposition 2.1.7.

The classes of formally smooth (resp. étale, resp. unramified) morphisms are closed under base change, composition, and products and local on both source and target.

Definition 2.1.8.

A f:X→S is smooth if it is formally smooth and locally of finite presentation.

We will now consider differentials. Let X=Spec⁡A be an affine scheme over k and choose a k-point and consider the diagram

If X is smooth, then there exists a lift Spec⁡k⁡[ε]→X. But this is given by a morphism

This motivates the following definition:

Definition 2.1.9.

Let R→A be a morphism of rings and M be an A-module. A derivation d:A→M is an A-linear map satisfying the Leibniz rule.

Proposition 2.1.10.

There exists an A-module ΩA/k1 equipped with a derivation d:ΩA/k1 that is universal among derivations from A. This means that all derivations d~:A→M factor through d, and formally, we have an identity

Definition 2.1.11.

For an A-module M with derivation d:A→M, define the ring A⁡[M] as the module A⊕M with the multiplication

There is a sequence ϕ:A→A⁡[M]→A.

Proposition 2.1.12.

Let S←R→A→B be a diagram of rings. Then

  1. 1.

    ΩA⊗S/S1≃ΩA/R1⊗RS;

  2. 2.

    The sequence ΩA/R1⊗AB→ΩB/R1→ΩB/A1→0 is exact.

  3. 3.

    If B=A/I for some ideal I, we have an exact sequence

  4. 4.

    For all f∈A, we havae ΩA⁡[f−1]/R1≃ΩA/R1⊗AA⁡[f−1].

Remark 2.1.13.

If J=ker⁡(A⊗RA→A), then ΩA/R1=J/J2.

Theorem 2.1.14.

Let f:X→S be locally of finite presentation. The following are equivalent:

  1. 1.

    f is smooth;

  2. 2.

    f is flat with smooth fibers;

  3. 3.

    f is flat and has smooth geometric fibers.

We will finally return to deformation theory.

Lemma 2.1.15.

Let Z0 be a closed subscheme of Z determined by a nilpotent ideal sheaf N. If Z0 is affine, then so is Z.

Proof of this result can be found in EGA, Chapter I.5.9.

Proof of Theorem 2.1.3.

Recall that we have a diagram of the form

where A→B is flat and B0≃B⊗Ak is a smooth k-algebra. We need to prove that B0≃B⊗Ak. The first step is to prove this result for first-order deformations. Suppose that A=k⁡[ε] is a square-zero extension.

Lemma 2.1.16.

For a ring R with M,N flat over R, nilpotent ideal I⊂R, and f:M→N, then if f⊗RR/I is an isomorphism, then so is f.

To prove the lemma, note that the cokernel of f is preserved by I, so it must vanish. Returning to our case, we know that B is a smooth k⁡[ε]-algebra. Now we obtain a square-zero extension B0⁢[ε] of B0 and a diagram

with a lift B→B0⁢[ε]. But now by the lemma, we have B⊗k⁡[ε]k=B0⁢[ε]⊗k⁡[ε]k. The rest of the proof follows using an inductive argument that was verbalized but now written down. □

2.2 Deformations of schemes

The main theorem of this section is

Theorem 2.2.1.

Assume X is a smooth R-scheme. Then there is a bijection

Proof.

Let 𝒳′ be a smooth deformation over k⁡[I]. Then the diagram

is cartesian. Then if Uk=Spec⁡Bk is an affine cover of X and Uk′=Spec⁡Dk is an affine cover of X′, we have a k⁡[I]-linear ring isomorphism

Modulo I, φk is the identity on Bk. Without loss of generality, we may assume that Uk⁢j=Uk∩Uj is a distinguished open for both Uk and Uj, so let Uk⁢j=Spec⁡Bk⁢j and Uk⁢j′=Spec⁡Dk⁢j. Now note that both

induce the identity on Bk⁢j modulo I. Now we have the commutative diagram

Lemma 2.2.2.

The morphism g=φj−1⁢φk must be of the form

where δ:Bk⁢j→I is a derivation.

In particular, this means that φj−1∘φk⁢(b,b′)=(b,αk⁢j⁢(b)+b′), where αk⁢j:Bk⁢j→I⊗kBk⁢j is a derivation.

By definition, we have

Therefore, αk∈H0⁢(Bk⁢j,TX⊗kI). Note that

which implies that

and thus {αk⁢j}∈Z1⁢({Uk},TX⊗kI).

If two deformations are the same, note that φk is defined using a ring section sk:Bk→Dk of the canonical map πk:Dk→Bk. If φk′ is defined using another section sk′, then define θk=sk′−sk∈Der⁡(Bk,I⊗kBk). We now compute

and thus the two differ by the desired coboundaries. □

We will now consider some obstructions. We are looking for a diagram of the form

for each pair (j,k), we have a isomorphism ψj⁢k:Vj′→Vk′ and a cocycle

This induces Bj⁢k⁢ℓ∈DerA⁢(Dj⁢k⁢ℓ,J⊗ADk⁢ℓ)=Z2⁢(U,TX′/A⊗AJ).

Now we will discuss some examples.

Theorem 2.2.3.

Let C be a smooth projective curve, T=TC, and K=ΩC1. We have the following table:

Table 2.1: Cohomology
degree h0 h1 h2
K 2⁢g−2 g 1 0
T 2−2⁢g ε ε+3⁢g−3 0

where ε=0 where g≥2, ε=1 if g=1, and ε=3 if g=0. For g≥2, deg⁡T<0, and by Riemann-Roch and Serre duality, we have h1⁢(C,TC)=3⁢g−3.

Theorem 2.2.4.

ℙn has no infinitesimal deformations.

Proof.

Consider the Euler sequence

and use the long exact sequence in cohomology. Because positive degree line bundles have no higher cohomology, we have H1⁢(Tℙn)=0. □

3 Kevin (Oct 08): Deformations of coherent sheaves

There will be no mixed characteristic funny business during this lecture. Let X be a projective k-scheme (proper might be fine, but this makes certain facts more true) and ℱ be a coherent sheaf on X. Consider the deformation functor

We want to study the properties of this functor, which means we will check Schlessinger’s conditions:1 Let A→B←C be a diagram in 𝒞 and consider the morphism

  • •

    (H1) The morphism r is surjective if C↠A;

  • •

    (H2) The morphism r is bijective if C=k⁡[ε]↠k=A;

  • •

    (H3) dimk(tD)<∞ (later, we will see that we need H2 for formulate this). Here, tD is the tangent space to D;

  • •

    (H4) The morphism r is bijective if C↠A.

Recall from Johan’s lecture that (H1),(H2),(H3) are equivalent to the existence of a hull and (H3),(H4) are equivalent to D being pro-representable.

We only need to check (H1) for small extensions, which are extensions by a k-vector spaace

where I is killed by the maximal ideal of C.

Theorem 3.0.1.

The functor Dℱ admits a hull.

Lemma 3.0.2.

Let (A,𝔪) be a local Artinian ring.

  1. 1.

    If 𝔪⁢M=M, then M≅0.

  2. 2.

    If M→N induces an isomorphism M/𝔪⁢M≅N/𝔪⁢N and N is flat over A, then M≅N.

  3. 3.

    If M is flat, then M is free.

Proof.

We know that 𝔪d=0, so 𝔪d⁢M=0, and thus M=𝔪⁢M=𝔪2⁢M=⋯=𝔪d⁢M=0. Next, suppose M→N induces an isomorphism after killing 𝔪. Then we know that the kernel and cokernel vanish because they are killed by 𝔪, so M→N must be an isomorphism. The last part is left as an exercise. □

Proof of theorem.

We will simply prove (H1),(H2),(H3):

  1. 1.

    Suppose that C↠A is a small extension and consider a pair (ℱB,ℱC)∈D⁡(B)×D⁡(A)D⁡(C). We know that we have isomorphisms ℱB|XA≅ℱA,ℱc|XA≅ℱA, and so we take the fiber product

    We only need to show that our sheaf is flat over B×AC because it clearly restricts to ℱB and ℱC. We can consider each sheaf as a module M, and so we know MB is free over B by the lemma. Choose a basis {ei}. Also consider the diagram

    Then MA has A-basis u⁡(ei). Because MC surjects onto MA, we can lift the u⁡(ei) to fi∈C, and these form a C-basis for MC. This all implies that MB×MAMC is free with basis (ei,fi).

  2. 2.

    It suffices to prove injectivity. Suppose 𝒢∈D⁡(B×kk⁡[ε]) maps to (ℱB,ℱk⁡[ε])∈D⁡(B)×D⁡(k⁡[ε]), and so we have morphisms

    We will prove that this diagram is Cartesian. By the lemma, the morphism 𝒢→ℱB×ℱℱk⁡[ε] is an isomorphism.

  3. 3.

    We will prove that TD=ExtX1⁡(ℱ,ℱ). We will only prove this in the case where ℱ is a vector bundle ℰ of rank r. In this case, we have ExtX1⁡(ℰ,ℰ)=H1⁢(X,End⁡(ℰ)). Now we will associate cocycles to deformations. To each ℰk⁡[ε], we will associate an open cover (Uj) and

    and we write gi⁢j+ε⁢fi⁢j, where gi⁢j∈Aut𝒪X⊕r⁡(Ui⁢j) and fi⁢j∈End⁡(𝒪X⊕r)⁢(Ui⁢j). The cocycle condition is that

    which is the same as

    which is exactly the Čech 1-cocycle condition. Proving that equivalent cocycles give the same deformation is easy. □

Theorem 3.0.3.

The condition (H4) holds when ℱ is simple, which means that k≃EndX⁡(ℱ).

3.1 Tangent-obstruction theory

Suppose D is a deformation functor. Then a tangent-obstruction theory for D is given by finite-dimensional k-vector spaces (T1,T2). Suppose we have a small extension

Then we have another exact sequence

which means that

  1. 1.

    ξA∈D⁡(A) lifts to D⁡(B) if and only if ob⁡(ξA)=0;

  2. 2.

    T1⊗I acts transitively on the fibers of D⁡(B)→D⁡(A);

  3. 3.

    If A=k, then the action of T1⊗I acts simply transitively on D⁡(B).

Note that because T1 acts simply transitively on D⁡(k⁡[ε]), we must have T1=D⁡(k⁡[ε]). On the other hand, T2 is not canonical.

Theorem 3.1.1.

The deformations Dℱ admits a tangent-obstruction theory with T1=ExtX1⁡(ℱ,ℱ) and T2=ExtX2⁡(ℱ,ℱ).

Proof.

We claim that if D satisfies (H1) and (H2), then D⁡(k⁡[ε])⊗I naturally acts on D⁡(B) for small extensions 0→I→B→A→0. To see this, note that D⁡(k⁡[ε])⊗kI=D⁡(k⁡[I]). We also note that by (H2), D⁡(k⁡[I])×D⁡(B)=D⁡(k⁡[I]×kB). Now define α:k⁡[I]×kB→B by α⁡(1+i,b)=1+b, and this gives us an action of D⁡(k⁡[I])×D⁡(B)=D⁡(k⁡[I]×kB)→α∗D⁡(B). To prove transitivity, apply (H1) to the diagram

Now we will consider obstructions. We will assume again that ℱ is a rank r vector bundle, which we will call ℰ. Let

be a small extension, so we will consider H2⁢(X,End⁡(ℰ)). Consider an open cover (Ui) and gi⁢j∈Aut⁡(𝒪X⊕r⊗kA)⁢(Ui⁢j). We want to lift these to hi⁢j∈Aut⁡(𝒪X⊕r⊗kB)⁢(Ui⁢j). If this is possible, we have a cocycle

and the cocycle condition is satisfied when hi⁢j−1⁢hi⁢j⁢hi⁢k=1. If any other hi⁢j′=hi⁢j+si⁢j, then we note that

and this gives us a class in H2⁢(X,End⁡(ℰ))⊗I. □

Remark 3.1.2.

Let R be the hull of D, which means we have a morphism hR→D. Then we know R=k⁡[[t1,…,td1⁢1]]/(f1,…,fd2). We also know that d1−d2≤dimR≤d1.

Example 3.1.3 (Good example).

Let X be a smooth projective curve and ℰ be a rank r vector bundle. Then we know that

so deformations of ℰ are unobstructed. Now assume that ℰ is simple. Then H0⁢(X,End⁡(ℰ))=k by definition, and we also know that Dε is pro-represented by some ring R with

by Riemann-Roch.

Example 3.1.4 (Bad example).

Let X be a smoorh projective variety and ℰ be a rank r vector bundle on X. Let ℰ1,ℰ2=Dℰ⁢(k⁡[ε]). Then (ℰ1,ℰ2)∈Dℰ⁢(k⁡[ε1,ε2]/(ε12,ε1⁢ε2,ε22)), and we would like to lift to Dℰ⁢(k⁡[ε1,ε2]/(ε12,ε22)).

We will compute the obstruction explicitly. We know ℰ1,ℰ2 give us classes u1,u2∈H1⁢(X,End⁡(ℰ)), and after some magical computation, the obstruction to lifting is given by

where the cup product comes from the algebra structure on End⁡(ℰ).

Now let X=C1×C2 be a product of curves. Then H1⁢(X,𝒪X)=H1⁢(C,𝒪C1)⊕H1⁢(C2,𝒪C2) and H2⁢(X,𝒪X)=H1⁢(C,𝒪C1)⊗jH2⁢(C2,𝒪C2). Suppose that α1∈H1⁢(C1,𝒪C1) and α2∈H1⁢(C2,𝒪C2) with nonzero cup product. Then we simply set ℰ=𝒪X⊕𝒪X and

this gives us our obstructed deformation.

4 Patrick (Oct 15): Deformations of singularities

We begin by fixing some notation. Let k be a field and R=P/I, where P=k⁡[x1,…,xn] and I=(f1,…,fr) is an ideal. Throughout this lecture, we will denote local Artinian rings with residue field k by A,B,C,… and rings by R,S,T,… Finally, denote Z=Spec⁡R.

4.1 Explicit criteria for flatness

We will study (embedded) deformations of singular affine schemes embedded in 𝔸n. The first thing we want to understand is to explicitly understand flatness of some RA over A, where RA⊗Ak=R. We will write RA=PA/IA, where PA=A⁡[x1,…,xn]=A⊗kP. Recall that over a Noetherian local ring S with residue field k, a module M is flat if and only if it is free, and this is equivalent to Tor1S⁡(M,k)=0 by standard results in commutative algebra.

Now consider the exact sequence

After tensoring with k, we have

Therefore, we know that RA is flat over A if and only if IA⊗Ak=I. We would like to understand this statement.

Consider a presentation

of IA. Then we know RA is flat over A if and only if after tensoring with k, we obtain an exact sequence

Note that to give this presentation Ps→Pr→I→0 is the same as giving a complete set of relations among the generators of I.

Proposition 4.1.1.

Suppose that

is exact and

is a complex such that PAr→RA→RA→0 is exact and tensoring (2) with k gives (1). Then RA is flat over A.

Proof.

Note that the hypotheses are equivalent to the fact that all relations in I can be lifted to IA. Now given g1′,…,gr′∈PA such that

this clearly descends to a relation in I by killing the maximal ideal of A. But now if we choose a complete set of relations for IA, this descends to a complete set of relations in I, so we may in fact assume that (2) is exact.

In this case, there exists some LA such that the sequence splits as

By right exactness of the tensor product, we know PAs⊗k→LA⊗k→0 is exact. We also know that

is exact, again by right exactness. But this means that IA⊗k is the cokernel of Ps→Pr, and therefore IA⊗k=I. This means that RA is flat. □

Corollary 4.1.2.

Let R=P/I and RA=PA/IA, where I=(f1,…,fr) and IA=(f1′,…,fr′) such that fi′ is a lift of fi. Then RA is flat over A if and only if every relation among the fi lifts to a relation among the fi′.

Remark 4.1.3.

This result essentially gives us that first-order embedded deformations of Spec⁡R⊂𝔸n are given by Hom⁡(I,R). The first-order (not embedded) deformations of Z are given by the cokernel of

which arises from the exact sequence

and this is supported on the singular points of X, so when X has isolated singularities, this is finite-dimensional.

Note that if Spec⁡R⊂𝔸n is a complete intersection, then I is generated by a regular sequence, so in particular the Koszul complex is a free resolution of R and therefore there are only trivial relations among the fi (this means the relations are generated by fi⁢fj−fj⁢fi=0). Clearly, because we are only considering commutative rings (after all, this is normal algebraic geometry), this means that all deformations of Spec⁡R are unobstructed.

4.2 Hilbert schemes of smooth surfaces

We will prove that deformations of finite length closed subschemes of 𝔸2 are unobstructed. In particular, this will imply that the Hilbert scheme Hilb⁡(𝔸2,n) is smooth.

Let Z⊂𝔸2 be a closed subscheme of dimension 0. Then because P=k⁡[x,y] has dimension 2, there exists a free resolution

of R. In this case it is possible to understand the matrix (gi⁢j), and in fact this is the special case of a more general result. First, when we study the local behavior, we have the following result.

Theorem 4.2.1 (Hilbert, Burch).

Let P be a regular local ring of dimension n and R=P/I be a Cohen-Macaulay quotient of codimension 2. Then there exists an (r−1)×r matrix G=(gi⁢j) whose maximal minors f1,…,fr minimally generate I, and there is a free resolution

Proof.

Note that the fact that the free resolution has this length is a corollary of the Auslander-Buchsbaum formula, which says that for a ring R and module M, we have

and the fact that depth equals dimension for Cohen-Macaulay things. Thus we have a free resolution

where a1,…,ar are a minimal set of generators for I. Let fi is (−1)i times the determinant of the i-th minor of gi⁢j. We will prove that the map (fi) is the same as the map (ai); clearly

is a resolution. This is because at the generic point of P, we know (gi⁢j) is injective, so at least one fi is nonzero. But then we know coker⁡(gi⁢j) is torsion-free (because I is torsion-free), and so it in fact must vanish by rank reasons. Thus (a1,…,ar) and (f1,…,fr) are isomorphic as P-modules.

At a codimension 1 point in Spec⁡P, note that 0→Pr−1→Pr→(ai)P→B→0 is split exact (because I has codimension 2). This implies that at least one fi is a unit, and thus (f1,…,fr) has codimension at least 2. But then the isomorphism I≅(f1,…,fr) is given by multiplication by some nonzero element of P which is a unit away from codimension 2. But this means it is a unit everywhere. □

Considering the global picture in 𝔸n, we obtain the following result.

Theorem 4.2.2 (Hilbert, Schaps).

Let Z=Spec⁡R⊂𝔸n be a Cohen-Macaulay closed subscheme of codimension 2. Then R=P/I has a free resolution of the form

where the fi are the maximal minors of the matrix (gi⁢j).

This result in fact holds over any Artinian local ring A, which we will use later.

Next, we want to understand what happens if we choose some Artinian local ring with residue field k and lift the gi⁢j to gi⁢j′, where gi⁢j′∈PA.

Theorem 4.2.3 (Schaps).

If A is a square zero extension of k, then the sequence

is exact. Moreover, any lifting of R over A arises by lifting the matrix (gi⁢j).

Proof.

We know that

is a complex. This is because composing the two maps amounts to evaluating determinants with a repeated column. Because PA is free (and therefore flat), we can tensor with the exact sequence

to obtain an exact sequence of complexes

Note that

In particular, this term is exact by Hilbert-Schaps. In addition, clearly LA∙⊗A𝔪A=L∙⊗k𝔪A because A→k is a square zero extension, so the complex LA∙⊗A𝔪A is exact. By the long exact sequence in homology, we know that LA∙ is exact. Note that L∙ extends to an exact sequence

and LA∙ extends to an exact sequence

However, by the homology long exact sequence, we have an exact sequence

But this implies that RA⊗Ak=R. Finally, by the local criterion for flatness, we see that RA is flat over A.

Let RA=PA/IA be a lifting of R over A. Lift fi∈I to hi∈IA. By Nakayama, these generate IA, so we obtain a free resolution

where gi⁢j′ lift the gi⁢j. However, we already have a lift

and so we must show RA=RA′. But we know that the ideals IA=(h1,…,hr) and IA′=(f1′,…,fr′) are isomorphic as PA-modules. But then if we restrict this isomorphism to 𝔸An∖supp⁡B, we obtain a unit in H0⁢(𝔸An∖supp⁡B,𝒪𝔸An). Because functions extend over codimension 2, we have H0⁢(𝔸An∖supp⁡B,𝒪𝔸An)=PA, so this is a global unit. This gives the desired result. □

This result holds if we replace A→k with any square-zero extension of Artinian local rings B→A and P,PA with flat things, and so we see that (embedded) deformations of codimension 2 Cohen-Macaulay subschemes of 𝔸n are unobstructed. In particular, any dimension 0 closed subscheme Z⊂𝔸2 is automatically Cohen-Macaulay (because it is dimension 0), so its embedded deformations are unobstructed. By some cohomological argument, the tangent space to Hilb⁡(𝔸2,n) is isomorphic to Hom⁡(R,R) and has dimension 2⁢n, so

4.3 An obstructed deformation

Let R=k⁡[x,y,z]/(z2,x⁢y,x⁢z,y⁢z). Note that this scheme has an embedded point at the origin, so in particular it is not Cohen-Macaulay.

z
Figure 4.1: Drawing of Spec⁡R

We will study embedded deformations of Spec⁡R and see that they are obstructed. In particular, we will choose two deformations of R over k⁡[ε] that cannot be simultaneously lifted. We claim that a complete set of relations (using the ordering (x⁢y,x⁢z,y⁢z,z2) for the generators of I) is given by the matrix

Now a first-order deformation of Spec⁡R is given by lifting (x⁢y,x⁢z,y⁢z,z2) over k⁡[ε], and the first candidate is to consider Iε1=(x⁢y+ε1⁢y,x⁢z,y⁢z,z2). Then we note that

and we can lift G to kill this vector with the matrix

Next consider the deformation given by Iε2=(x⁢y,x⁢z,y⁢z+ε2⁢z,z2). We note that

and we can lift G to kill this vector with the matrix

Now we consider Iε12,ε22,ε1⁢ε2=(x⁢y+ε1⁢y,x⁢z,y⁢z+ε2⁢z,z2) and attempt to lift this deformation to k⁡[ε1,ε2]/(ε12,ε22). Note that

and clearly z∉I, so in fact we cannot lift this deformation to k⁡[ε1,ε2]/(ε12,ε22). This proves obstructedness.

5 Avi (Oct 22): Local-global methods

5.1 Curves with isolated singularities

Here, we will consider the same lifting problems that we have consider before. Let X be a curve over k and consider p:DefX→𝒞 be the deformations of X as a category cofibered over 𝒞.

Definition 5.1.1.

A functor F:D1→D2 is smooth if given φ:A′↠A and (Z,A′)→F⁡(Y,Z) over φ over φ and (Y′,A′)→(Y,A) over φ in D1, then there exists some F⁡(Y′,A′)→(Z,A′) over the identity in A′.

There is an absolute notion, for D1 to be smooth, where we set D2=Defk to be the category of trivial deformations. In this case, we also called D1 unobstructed, and this corresponds to being unobstructed in the tangent-obstruction theory.

Theorem 5.1.2 (Local-to-global).

Let X/k be separated of dimension at most 1 and smooth away from finitely many points. At each singularity p1,…,pn, consider the inclusions

into the henselizations and completions, respectively. Then the functors

are all smooth and (2),(3) induce isomorphisms on tangent spaces.

This means that we only have to check unobstructedness at the completions of the local rings at each singular point.

Proof.

First, we can reduce to the affine case.1 If X is a curve, then X=U1∪U2 can be covered by two affine opens such that U1,U1∩U2 are smooth. Because deformations of smooth affine schemes are unobstructed, we can essentially ignore U1. To do this, we will prove that DefX→DefU2 is smooth.

Because smoothness (roughly) respects products, the functor

is smooth by formal reasons. Because U1,U1∩U2 are smooth, we can ignore DefU1,DefU1∩U2 (in fact, the arrow is an equivalence) and project down to DefU2.

Now U2=Spec⁡P is affine. Then each pi gives a maximal ideal 𝔪i, and so we set J=⋂i𝔪i. If we consider the completion P̂J with respect to J, this decomposes as

Applying Lemma 91.12.5 in the Stacks Project, the functor

is smooth and an isomorphism on tangent spaces. The henselization step is similar. □

Example 5.1.3.

Let X=Spec⁡k⁡[x,y]/(x⁢y). Then ÔX,0=k⁡[[x,y]]/(x⁢y). We will show that all deformations look like k⁡[x,y]/(x⁢y−t) in some sense. Really, we will show that DefX has a hull, which is given by k⁡[[t]] with universal deformation

We want T such that Hom⁡(T,−)→DefX is smooth and induces an isomorphism on tangent spaces. Given a morphism

we obtain a deformation

of k⁡[[x,y]]/(x⁢y). We can check that this is smooth, and so we need to check that we have an isomorphism of tangent spaces.

We want to compute DefX⁡(k⁡[ε]). If S∈DefX⁡(k⁡[ε]), set R=k⁡[[x,y]]/(x,y), and we have a diagram

If we consider the exact sequence

and tensor with S, by flatness we have an exact sequence

But because k⊗k⁡[ε]S=R, we have a decomposition S=R⊕ε⁢R. If we choose lifts x~,y~ of x,y, the product x~⁢y~ lifts x⁢y=0, so x~⁢y~=(f⁡(x)+g⁡(y))⁢ε. If we choose different lifts x~′=x~+ε⁢h,y~′=y~+ε⁢h, we see that

But now x⁢j+y⁢h∈(x,y), and so if we set t=f⁡(0)+g⁡(0), then we can write x~⁢y~∈t⁢ε+(x,y)⁢ε, and so there is a canonical choice of the element of (x,y)⁢ε, namely 0. Thus we can set

and now we have a diagram

By the five lemma, St≃S, and so we are done.

5.2 Smooth curves with a finite group action

In this section, we will consider lifting from characteristic p>0 to characteristic 0. This means we will have to redefine what lifting. Here, we will suppose k=k¯ with characteristic p. Let Λ be a Noetherian complete local ring with residue field k. We define 𝒞Λ be the category of local Artinian Λ-algebras with residue field k.2

Forgetting about the requirement that k is algebraically closed, if k=𝔽p, then we can set Λ=ℤp. Let W=W⁡(k) be the ring of Witt vectors in k. This is apparently some universal Noetherian local ring lifting k, and for example, W⁡(𝔽p)=ℤp, and W⁡(𝔽¯p)=𝒪ℚˇp. We will consider some extension 𝒪/W of rings and write K for the fraction field of 𝒪.

We are interested in deforming smooth proper curves X/k with the action of some finite group G, conveniently denoted as a pair (X,G).

Definition 5.2.1.

The pair (X,G) lifts to characteristic 0 if there exists 𝒪 and 𝒳/𝒪 a projective smooth curve with an action of G such that (𝒳×0k,G)≃(X,G).

Given the action of G on X, we can construct the quotient X/G. To remember the data of the action, we remember the ramified Galois cover π:X→X/G.3

Consider the action of PGL2⁢(𝔽q) on ℙk1. This includes in ℙ𝒪1, which has an action of PGL2⁢(𝒪)⊂PGL2⁢(K). We want an embedding PGL2⁢(𝔽q)⊆PGL2⁢(K), but generally this is not possible. For example, set q=9 and consider the matrices

which generate a (ℤ/3)3. But then all finite subgroups of PGL2⁢(K) are cyclic, dihedral, A4, A5, or S4, so this action clearly cannot lift.

Proposition 5.2.2.

Suppose (X,G) is such that for all x∈X, the stabilizer Gx has order |Gx| prime to p. Then (X,G) lifts to characteristic 0.

Note that this is some local statement, so we need some kind of local-to-global method. Recall that 𝒪̂X,x=k⁡[[t]], and this has an action of Gx. Thus we define a local action as some action of a finite group G on k⁡[[t]], and this lifts if there exists 𝒪 with an action of G on 𝒪⁡[[t]] specializing to the original action on k⁡[[t]].

Theorem 5.2.3.

For (X,G), suppose that the local action of Gx on 𝒪̂X,x lifts to characteristic 0. Then (X,G) lifts to characteristic 0.4

It is a fact that if G acts faithfully on k⁡[[t]], then G=P⋊C, where P is P-Sylow and C is cyclic.

Proof of proposition.

We will simply prove that if |Gx| is prime to p, then the local action lifts. Then we know that Gx=G is cyclic of order n with generator σ. If we write

then of course if a0,i≠0, then σi⁢t is a unit and hence t is a unit, so a0,i=0. Our goal is to replace t by some generator with a more explicit action of σ. Let V be the vector space spanned by σi⁢t. Then on our designated basis, we have

This has eigenvalues the n-th roots of 1, so we choose z∈V such that σ⁢z=ζn⁢z and replace k⁡[[z]]=k⁡[[t]]. But now the action of μn clearly lifts to characteristic 0, so we are done. □

We have essentially proved that local actions of cyclic groups of order prime to p lift to characteristic 0. Here is a mild generalization:

Conjecture 5.2.4 (Oort5).

The local action on k⁡[[t]] by the action of any cyclic G lifts to characteristic 0.

This is now a theorem due to (among others?) Obus, Wewers, and Pop. Because of this, if the stabilizers Gx are all cyclic, then (X,G) lifts. However, this is not a necessary condition for lifting. If we consider an action of (ℤ/2)2 on ℙk1, this does lift (because D4=(ℤ/2)2).

6 Caleb and Morena (Oct 29): Lemma 0E3X and applications to contracting curves

6.1 One lemma in the Stacks Project

Lemma 6.1.1.

In Example 0DY7 let f:X→Y be a morphism of schemes over k. If f∗⁢𝒪X=𝒪Y and R1⁢f∗⁢𝒪X=0, then the morphism of deformation categories

is an equivalence.

Let A be an Artinian local ring with residue field k. Remember that DefX⁢(A) is the set of isomorphism classes of diagrams

with α flat. Also, DefX→Y consists of diagrams of the form

with α,β flat.

Lemma 6.1.2 (Lemma 063Y).

Let (f,f′):(X,X′)→(S,S′) be a morphism of first order thickenings such that f is flat. Then the following are equivalent:

  1. 1.

    f′ is flat and X=S×S′X′;

  2. 2.

    The canonical map f∗⁢CS/S′→CX/X′ is an isomorphism, where C is the conormal sheaf.

Proof.

In the affine case, write X=Spec⁡B,X′=Spec⁡B′,S=Spec⁡A,S′=Spec⁡A′. Then we are looking for a diagram of the form

The two conditions become

  1. 1.

    f is flat and B=B′⊗A′A;

  2. 2.

    I/I2⊗AB=J/J2 and I⊗AB=J.

To begin, note that B=B′⊗A′A is equivalent to B′/J=B′⊗A′A′/I, and this implies that J=I⁢B′ and thus that I⊗AB′→J is surjective. To prove injectivity, by flatness of B′, the map 0→I→A′ remains injective after tensoring with B′, so I⊗A′B′→B′ is injective.

In the other direction, we may cite Lemma 051C. Alternatively, we give the following argument. Assuming that I⊗AB→J is an isomorphism, we know J=I⁢B′, and thus B=B′⊗A′A. To prove that B′ is flat over A′, we know that B′/I⁢B′ is flat over A because B/A is flat and J=I⁢B′. We will prove that if 𝔞⊂A′ is an ideal, then 𝔞⊗A′B′→B′ is injective.

By some inexplicable brilliancy, we simply need to fill in the diagram

By diagram chasing reasons, we will have exactness. Consider the exact sequence

After tensoring with B′, we obtain a right exact sequence

This gives us the desired items in the question marks.

Now we want to prove that (I∩𝔞)⊗A′B′→I⁢B′ is injective. If we consider 0→I∩𝔞→I and tensor with B′/I⁢B′, we obtain

but this is clearly actually

For the other part, we simply take

and tensor with B′/I⁢B′. □

Proof of Lemma 0E3X.

We need to prove that β:(Y,f∗⁢𝒪XA)→Spec⁡A is flat. We can compose any thickening as a sequence

Now we apply Lemma 063Y to each square and obtain

Now if we consider the exact sequence

Applying f∗ and the assumption that R1⁢f∗⁢𝒪X=0, we obtain an exact sequence

Now if we consider the diagram

we want to prove that β is flat starting with β1 being flat. But here, we know

and therefore we have

Now we may apply Lemma 063Y repeatedly to obtain flatness of β. □

6.2 Application to moduli of curves

In this part of the lecture, we wish to define a contraction map ℳ¯g,n+1→ℳ¯g,n that deletes a marked point. This was used by Knudsen to prove that M¯g,n is a smooth and proper stack over Spec⁡ℤ. Other applications include relations between ℳ¯G and ℳ¯g+1 and their Chow groups. The roadmap for this section is:

  1. 1.

    We will discuss stable curves over an algebraically closed field k and define ℳ¯g,n.

  2. 2.

    We will define ℳ¯g,n+1→ℳ¯g,n in the wrong way.

  3. 3.

    We will define contraction of rational tails and bridges over an algebraically closed field correctly.

  4. 4.

    Finally, we will use Lemma 0E3X to define contraction over any scheme.

6.2.1 Stable curves

We will define n-marked genus g stable curves C over k=k¯. Here, we will take C to be a connected 1-dimensional scheme of finite type over k. We will work only with nodal curves, which are curves where for every x∈C⁡(k) either x is smooth and 𝒪C,x is a regular local ring and thus a UFD by Auslander-Buchsbaum or x is a node and

Remark 6.2.1.

In fact, 𝒪̂X,x is reduced and thus 𝒪X,x is reduced, so every nodal curve is reduced.

Now let

be the normalization. We know that ν−1⁢(node)=Spec⁡k⊔Spec⁡k. To check this, recall that the normalization of a reduced scheme is constructed by gluing the local maps

But now we know that

However, we know that

This implies that 𝒪C,p¯Q⁡(𝒪C,p) is not local because k⁡[[x]]×k⁡[[y]] is not local.

Proposition 6.2.2.

The arithmetic genus is given by

Proof.

Recall that ν−1⁢(node)=Spec⁡k⊔Spec⁡k. Then we have the exact sequence

Taking the long exact sequence in cohomology, we obtain

Because ν is finite, the Leray spectral sequence computing H1⁢(C~,𝒪C~) degenerates at the E2-page, and thus

But then we know H1⁢(C,ν∗⁢𝒪C~)=⨁H1⁢(𝒪C~i)=∑g⁡(C~i). By connectedness of C, we know that H0⁢(C,𝒪C)=k and H0⁢(C,ν∗⁢𝒪C~)=k#⁡(components). □

We now need to add marked points.

Definition 6.2.3.

A n-marked, genus g nodal curve C over k is a nodal curve of genus g with n smooth points x1,…,xn∈C⁡(k).

This definition allows too many curves, so we want to define a notion of stability.

Definition 6.2.4.

A genus g nodal curve C with n marked points is stable if for all irreducible components C~i, we have

where a special point is a node or a marked point.

Example 6.2.5.

Over ℂ, consider a genus 2 curve with two marked points x,y, an elliptic curve with marked point z, and a nodal cubic with no marked points all intersecting in a triangle. Here, we have three marked points and four nodes.1 By observation, this curve is stable.

Remark 6.2.6.

All of the C~i are stable if and only if g⁡(C~i)≥2, g⁡(C~i)=1 and C~i has at least one special point, or if g⁡(C~i)=0 and there are at least three special points.

6.2.2 Contraction done wrong

We will denote by ℳ¯g,n be the fibered category whose objects over a scheme S are given by the following data:

  • •

    A map f:𝒞→S proper and flat with sections σ1,…,σn:S→𝒞 and for any geometric point s∈S, the map fs:𝒞s→k⁡(s) and sections σ1,s,…,σn,s form a stable curve of genus g.

  • •

    Morphisms are given by cartesian diagrams

    that respect the sections.

Clearly the forgetful functor simply forgets the last section. Unfortunately, this does not actually respect the stability condition. The problem is when the component containing σn+1⁢(s) satisfies 2⁢gi−2+#⁡(special)=1, because when we delete σn+1 we lose stability. This only happens when g⁡(C~i)=1 and there is exactly one marked point (which is actually smooth and integral, so is an elliptic curve) or when g⁡(C~i)=0 and there are exactly three special points. In the second case, we may have ℙ1 with three marked points, two marked points and a node (rational tail), or one node and two marked points (rational bridge). Also, we may have a nodal cubic with one marked point.

6.2.3 Contraction done right

In this part, we construct from a prestable curve 𝒞k⁡(s) a stable curve C~k⁡(s). In the rational tail case, we are contracting the entire ℙ1 to the node and in the rational bridge case we identify the two nodes and collapse the ℙ1 to that point, leaving the two other components intersecting at a node (locally; globally this can be either one or two components). Of course, we need to check that f∗⁢𝒪Ck⁡(s)=𝒪C~k⁡(x) and R1⁢f∗⁢𝒪Ck⁡(s)=0. Next, we need to check that contraction can be extended to a neighborhood in a canonical way when S=Spec⁡𝒪S,sh.

7 Morena (Nov 05): Contraction morphisms between moduli stacks of curves

7.1 Recap of last time

Recall that for f:X→Y, if f∗⁢𝒪X=𝒪Y and R1⁢f∗⁢𝒪X=0, then we have an equivalence of categories

We also attempted to construct a morphism ℳ¯g,n+1→ℳ¯g,n using the forgetful functor, but this does not work. Remember the bad cases were called the rational tail and rational bridge. Also recall that the stability condition for every component was 2⁢g−2+#⁡(special points)>0, where special points are nodes and marked points.

7.2 Contraction of rational tails and bridges

In the rational tail case, our stable curve should be the scheme-theoretic closure (so just topological closure) C/Ci¯, where Ci is the rational tail. Because C is reduced, we see that C is the pushout

Of course, this gives us a map C→C∖Ci¯ by contracting Ci. Now we need to check that C∖Ci¯ is a prestable curve of genus g which is actually stable and that if c:C→C∖Ci¯, then c∗⁢𝒪C≃𝒪C∖Ci¯ and R1⁢c∗⁢𝒪C=0.

To check the sheafy conditions, we have an exact sequence

This gives us a longer exact sequence

To see that R1⁢c∗⁢𝒪C=0, we only need to check that R1⁢c∗⁢j∗⁢𝒪C∖Ci¯=0 because R1⁢c∗⁢ji⁣∗⁢𝒪Ci=H1⁢(Ci,𝒪Ci)=0. But here we have c∗⁢j∗⁢𝒪C∖Ci¯=0 because c∗⁢j∗=0, and so we need to prove surjectivity of the direct sum onto the skyscraper sheaf. This is clear.

Considering

we see that c∗⁢𝒪C≃𝒪C~ and

To check stability of C∖Ci¯, we simply note that stability condition of the component Cj attached to Ci is unchanged, and everything else was untouched, so we have a stable curve.

We now check the rational bridge case. Here we consider the pushout

This exists and has nice properties. What this really means is that we contract Ci and introduce a new self-intersection at x1=x2. Now the contraction morphism is given by considering the total pushout of

Restricting to an affine piece, where C¯=Spec⁡A and C∖Ci¯=Spec⁡A′, we obtain a fiber product diagram of rings

We can now tensor with Ay, and we need to show that that Ây=k⁡[[x,y]]/(x⁢y). Because completions are exact, we can consider the fiber product diagram

Because A′⊗AÂy=Âx1′×Âx2′=k⁡[[t1]]×k⁡[[t2]], we see that

and thus y is a node.

7.3 Contraction over any base scheme

Theorem 7.3.1.

Let (S,f:𝒞→S,σ1,…,σn+1)∈ℳ¯g,n+1(S). Then there exists a contraction such that

and the following conditions hold:

  1. 1.

    (S,g:𝒞¯→S,c∘σ1,…,c∘σn)∈ℳ¯g,n(S).

  2. 2.

    c∗⁢𝒪𝒞≃𝒪𝒞¯ and R1⁢c∗⁢𝒪𝒞=0, and this is stable under base change. In addition, for all geometric points s¯→S, cs¯ is either an isomorphism or a contraction of a rational tail or rational bridge.

Moreover, c:𝒞→𝒞¯ is unique up to unique isomorphism.

Corollary 7.3.2.

The morphism ℳ¯g,n+1→ℳ¯g,n is defined over ℤ.

Proof.

We know what happens to objects. We know that morphisms are cartesian diagrams

But now we have two candidate contractions of 𝒞′, namely 𝒞′¯ and 𝒞¯×SS′. But these have a unique isomorphism. These fit into the diagram

This gives us compatibility with morphisms. □

To prove the theorem, we want to work étale locally over S, with a cover (Si→S). Then there exists a factorization 𝒞i→𝑎𝒞¯i→Si, and then we need to check that the cocycle condition works on the overlaps. The existence of a global factorization means that our data is effective, and this is the same as ℳ¯g,n is a stack.

To prove existence of a factorization étale locally, over s¯∈Si, we have a contraction cs:Cs→C¯s. We want to prove that there exists an étale neighborhood (U,u)→(S,s) such that there exists a factorization CU→C¯U extending cs. In order to do this, we need to do some reduction. If S=Spec⁡A is affine is of finite type over ℤ, we want Ai⊆A such that A=lim→⁡Ai, and thus S=Spec⁡A=lim←⁡Spec⁡Ai.

Theorem 7.3.3 (0E6U, 0E6V, 0DSS, 0CMV).

The fibered category 𝖢𝗎𝗋𝗏𝖾𝗌gprestable is limit-preserving.

Lemma 7.3.4.

The category of schemes of finite presentation over S is the colimit of the categories of schemes of finite presentation over Si.

Now if S=Spec⁡A, then 𝒪S,sh=colim(U,u)→(S,s)⁡𝒪U⁢(U). Thus we may consider S=Spec⁡(𝒪S,sh). Write Λ for this henselization. Then if we consider Λ→Λ/𝔪n⁢Λ, we have fiber product diagrams

This gives us a formal object of DefCs⁢(Λ̂). But now we know that

and thus we actually have deformations of Cs→C¯s over Λ/𝔪n⁢Λ. But this gives us a formal element of

Now using 01W0, we have 𝒞¯n∈𝖢𝗎𝗋𝗏𝖾g.

Theorem 7.3.5.

The data 𝖢𝗎𝗋𝗏𝖾g is effective.

For a sketch of this, note that the first stability condition is an open condition. If p1,…,pn are the marked points, then write D=p1+⋯+pn. Stability is equivalent to ωC⁢(D) being ample, and so over Cs we have an ample line bundle ωCs⁢(D). We can lift this to (𝒞¯n,ℒn) because obstructions live in

Using the Grothendieck algebraization theorem, we see that if Xi→Si is proper and ℒi is ample, there exists a proper morphism X→S and ℒ an ample line bundle such that base change to Sn recovers (Xn,ℒn).

Here, we know that 𝒞¯→Spec⁡Λ̂ is finite type and separated while 𝒞→Spec⁡Λ̂ is proper, and so there exists a unique 𝒞→𝒞¯. We now need to return back to Spec⁡Λ. But now we know that Λ̂ is the direct limit of its finitely-generated (over Λ) subalgebras. This gives us Λ⊆Λ1⊆Λ̂. We know that

is regular, it is flat, and thus Λ̂/𝔪⁢Λ̂ is noetherian and geometrically regular over k⁡(𝔪) by Popescu. We have the diagram

Because Λ is henselian, we have a section Λét→Λ, and thus we have a base change to Λ.

8 Baiqing (Nov 12): Deformations of group schemes

8.1 Deformations of abelian schemes

Definition 8.1.1.

An abelian scheme p:X→S is a proper smooth group scheme over a connected scheme S with geometrically connected fibers.

If S=Spec⁡k is a field, then X is projective. However, this is not true in general.

Definition 8.1.2.

The deformation functor DefXAS:ΛW→𝖲𝖾𝗍 is defined for an abelian variety X over k and ΛW the category of local Artinian W-algebras with residue field k. It is given by

where (X1′,φ1′)∼(X2′,φ2′) if there exists an isomorphism ψ:X1′→X2′ of abelian schemes such that φ2∘ψk=φ1.

Theorem 8.1.3.

Let X be an abelian variety of dimension g over k. Then the deformation functor DefXAS is pro-representable by W⁡(k)⁢[[t1,t2,…,tg2]].

Recall Schlessinger’s criterion from Johan’s lecture.1 Note that if F is formally smooth and F⁡(R′)→F⁡(R) is surjective for all R′↠R, then F is pro-representable by a power series ring W⁡[[t1,…,td]], where d=dimkF⁡(k⁡[ε]).

We will prove the conditions (H3),(H4) and then prove that the deformation functor is formally smooth. Before this, we will consider the geometry of abelian schemes.

Lemma 8.1.4 (Rigidity lemma).

Given a diagram

suppose that S is connected, p is flat and proper, and H0⁢(Xs,𝒪Xs)≃k⁡(s) for all s∈S. If for some point s∈S, f⁡(Xs) is set-theoretically a single point, then there exists a section η:S→Y such that f=η∘p.

Corollary 8.1.5.

Let X,Y be abelian schemes over S and f:X→Y. If f∘εX=εY, then f is a homomorphism. Here, εX refers to the identity section.

This result holds in general if we replace Y be an arbitrary group scheme.

Sketch of proof.

Consider the diagram

Consider the morphisms X×SX→G×SX given by ψ1⁢(x1,x2)=(f⁡(x1⁢x2),x2) and ψ2⁢(x1,x2)=(f⁡(x1),x2). Now we argue that f⁡(g1⁢g2)=f⁡(g1)⁢f⁡(g2). □

Also, it is clear that abelian schemes are commutative because inversion takes the identity to itself and thus is a homomorphism. If ϕ1,ϕ2:X→Y are two homomorphisms and (ϕ1)s=(ϕ2)2 for some s∈S, then ϕ1=ϕ2. Next, if f:X→Y is a morphism of schemes between abelian schemes, then f−f⁡(0)=f−f∘ε∘p is a homomorphism.

We will now prove (H4) for DefXAS. For a small extension R′′→R, we show that

is a bijection. This is equivalent to proving that in the diagram

if (𝒳′,φ′)↦(𝒳,φ) along the bottom arrow, then π′−1⁢((𝒳′,φ′))=π−1⁢((X,φ)).

Proposition 8.1.6.

π−1⁢((𝒳,φ)) is the set of isomorphism classes of (𝒳′′,φR′′) such that 𝒳′′ is flat over R′′ and 𝒳R′′→𝒳 is an isomorphism as schemes over R.

Choose (𝒳′′,φ′′)∈π−1⁢((𝒳,φ)). Then there exists an isomorphism ϕ:𝒳R′′→𝒳 of abelian schemes. Now we will send (𝒳′′,φ′′)↦(𝒳′′,ϕ). Checking uniqueness is easy with the diagram

Now we check that if (𝒳1′′,φ1′′)∼(𝒳2′′,φ2′′), then they are sent to the same thing. This was erased from the board by Baiqing before I could process it.

Now we prove injectivity. If (𝒳1′′,φ1′′),(𝒳2′′,φ2′′) map to equivalent (𝒳1′′,ϕ1),(𝒳2′′,ϕ2), then there exists ψ:𝒳1′′≃𝒳2′′ such that ϕ2∘ψR=ϕ1. This implies that φ2′′∘ψk=φ1′′. Unfortunately, ψ is not an isomorphism of abelian schemes, so we can replace it by ψ~=ψ−ψ⁡(0).

Finally, we prove surjectivity, which is the difficult part of the argument.

Proposition 8.1.7.

Let S=Spec⁡A, where A is an Artinian local ring, 𝔪⊂A be the maximal ideal, and I⊂A such that 𝔪⋅I=0. Let π:X→S be proper and smooth, ε:S→X be a section of π, S0=Spec⁡A/I, and X0=X×SS0. Then if X0 is an abelian scheme with identity ε|S0, then X is an abelian scheme with identity ε.

Theorem 8.1.8.

Let X,Y/R are smooth schemes. Suppose that π:R′↠R is a small extension. Let X′∈DefX⁢(R′) and Y′∈DefY⁢(R′). For any f:X→Y over R, there is a canonically associated class o⁡(f)∈H1⁢(Xk,fk∗⁢TYk/k)⊗kker⁡π. If o⁡(f)=0, then H0⁢(Xk,fk∗⁢TYk/k)⊗kker⁡π.

Proof of proposition.

Given (m,ε,ι), we want (μ,ε), where μ⁡(x,y)=x−y. We have μ on X0, and so we want to deform this to μ′ on X. We know that

but because Xk is an abelian variety, we have

We want to show that o⁡(μ)=0. Consider g1=Δ:X0→X0×S0⁢X0 and g2=(id,ε⁢p):X0→X0×S0X0. Then μ∘g1=ε0∘π, and thus o⁡(μ∘g1)=0. Also, μ∘g2=1X0 and o⁡(μ∘g2)=0. We know that

Consider the morphisms (g1)k∗:H1⁢(Xk×Xk,μk∗⁢TX/k)→H1⁢(Xk,(ε∘π)∗⁢TX/k) and (g2)k∗:H2⁢(Xk×Xk,μk∗⁢(TX/k))→H1⁢(Xk,(ε∘π)∗⁢TX/k). But now we see that they are given by (g1)k∗:(x,y)⊗v↦(x+y)⊗v and (x,y)⊗v↦x⊗v, and so o⁡(μ)=0.

Now consider H0⁢(Xk×kXk,μk∗⁢TXk/k)⊗kker⁡π≅𝔱⊗kI and let μ′ be a deformation of μ. Then

is a deformation of ε0. Under the identification

and using what we have done previously, we have μ′∘(ε,ε)=ε and therefore an abelian scheme structure. □

Finally we prove that DefXAS⁡(R′)→𝖣𝖾𝖿XAS⁢(R) is surjective for R′↠R. Let (𝒳,φ)∈DefXAS⁡(R). Then o⁡(𝒳)∈H2⁢(X,TX/k)⊗kker⁡π=𝔱⊗k(𝔱∗∧𝔱∗). We want to prove that π−1⁢((𝒳,φ)) is nonempty. We know that ι∗ induces −1 on 𝔱,𝔱∗, and so o⁡(𝒳)=−o⁡(𝒳), and if we are working not in characteristic 2, o⁡(𝒳)=0.

To compute the dimension of the tangent space, we want to compute DefXAS⁢(k⁡[ε]), but this is H1⁢(X,TX/k)≃𝔱⊗𝔱∗, which has dimension g2.

8.2 Deformations of smooth affine group schemes

We will now discuss deformations of 𝔾m and 𝔾a. Let G/k be an affine smooth algebraic group scheme. We will consider the deformation problem

For example, for 𝔾m, we know that Spec⁡k⁡[ε]⁢[t,t−1] is a deformation of 𝔾m. We want to deform the multiplication, and so we have m′:T→T1⁢T2⁢(1+ε⁢Δ⁢(T1,T2)). To check associativity, we obtain

We will also consider Δ,Δ′ equivalent if there exists f such that Δ′⁢(T1,T2)=Δ⁡(T1,T2)+f⁡(T1⁢T2)−f⁡(T1)−f⁡(T2). Therefore we have

Proposition 8.2.1.

In fact, we have Def𝔾m⁡(k⁡[ε])=0.

Proof.

let f⁡(T)=∑ai⁢Ti. Then f⁡(T1⁢T2)−f⁡(T1)−f⁡(T2)=∑ai⁢(T1i⁢T2i−T1i−T2i). If we set

adjust Δ by f such that λi⁢0=0. Now we have the equation

and thus λi⁢j=0. □

For the additive group, our deformed multiplication is given by T→T1+T2+ε⁢Δ⁢(T1⁢T2). Associativity is equivalent to the condition

The trivial deformations are given by f⁡(T1+T2)−f⁡(T1)−f⁡(T2), and thus we have

This deformation space vanishes in characteristic 0 and is infinite-dimensional in positive characteristic. If we apply ∂∂T3 to the associativity relation, we obtain

and if we apply ∂∂T1, we obtain

and therefore Δ12⁢(T1,T2)=f⁡(T1+T2). In characteristic 0, f has a primitive, and so taking the primitive twice we obtain a desired F~. Primitives do not exist in positive characteristic, so this proof does not work. However, it is possible to work out the following:

Proposition 8.2.2.

In positive characteristic p, Δ is a linear combination of Bp⁢(T1,T2)pn for n≥0 and T1pn⁢T2pm for m≥n+1, where Bp⁢(T1,T2)=(T1+T2)p−T1p−T2pp∈ℤ⁡[T1,T2].

9 Haodong (Nov 19): Artin’s axioms

9.1 Prestacks and stacks

First, we will generalize the Zariski topology a little bit.

Definition 9.1.1.

Let S be a category. Then a Grothendieck topology on S is given by a set Cov⁡(X) for each x∈X, where each element in Cov⁡(X) is a collection of morphisms {xi→x}I in S. We require that

  1. 1.

    All isomorphisms x′→∼x are in Cov⁡(x).

  2. 2.

    If {xi→x}I∈Cov⁡(X) and y→x, then xi×xy exist and {xi×xy→y}I∈Cov⁡(y).

  3. 3.

    If {xi→x}I is a covering of X and {xi⁢j→xi}Ji is a covering of xi for all i, then {xi⁢j→x} is a covering of X.

A site is a category with a Grothendieck topology.

Example 9.1.2.

Let S be a scheme and let 𝒞=𝖲𝖼𝗁/S. For every T→S, we say that {fi:Ti→T} is a covering if all fi are open immersions and the fi are jointly surjective. This gives the big Zariski site of S. If we replace open immersion with étale morphism, then we obtain the big étale site.

9.1.1 Prestacks

Definition 9.1.3.

Let 𝒮 be a category and p:𝒳→𝒮 be a functor. We will denote objects of 𝒳 by a,b,… and objects in 𝒮 by S,T,…. We will denote morphisms in 𝒳,𝒮 by α,f respectively. We say that p is a prestack if the following conditions are satisfied:

  1. 1.

    If p⁡(b)=T and S→T is a morphism, there exists a and a morphism a→b such that p⁡(a)=S filling the diagram

  2. 2.

    If we have

    there exists a unique morphism a→b filling in the above arrow.

Exercise 9.1.4.

For every s∈𝒮, the category 𝒳⁡(s) over s has morphisms those lying over the identity and is thus a groupoid.

Example 9.1.5.

Let F:𝒮→𝖲𝖾𝗍 be a functor. We will define 𝒳F to have objects (a,s), where s∈𝒮 and a∈F⁡(s). Then morphisms (a,s)→(a′,s′) are morphisms f:s→s′ such that F⁡(f)⁢(a′)=a. We will define p:(a,s)↦s, and this defines a prestack, which we will simply call F.

Example 9.1.6.

Let S be a scheme and 𝒞=𝖲𝖼𝗁/S. If T→S is an S-scheme, then Hom⁡(−,T) is a presheaf on 𝒞 and by the previous example defines a prestack over 𝒞, which we will call T.

Example 9.1.7.

Consider the functor ℳg→𝖲𝖼𝗁/ℂ where the objects are morphisms 𝒞→S, where S is a scheme over ℂ and 𝒞→S is smooth and proper with all geometric fibers connected curves of genus g. Morphisms are simply pairs α:𝒞→𝒞′,f:S→S′ such that 𝒞=𝒞′⊗S′S.

Definitions 9.1.8.

Let 𝒮 be a site.

  1. 1.

    A morphism of prestacks 𝒳→𝒴 is a functor f such that for all a∈𝒳, p𝒳⁢(a)=p𝒴⁢(f⁡(a)).

  2. 2.

    If f,g:𝒳→𝒴 are morphisms of prestacks, then a 2-morphism α:f→g is a natural transformation such that for all a∈𝒳, αa:f⁡(a)→g⁡(a) lies over the identity in 𝒮. In particular, α is an isomorphism.

  3. 3.

    A diagram

    is a 2-commutative diagram if there exists a 2-morphism α:f⁢g′→g⁢f′.

  4. 4.

    f:𝒳→𝒴 is called an equivalence if there exists g:𝒴→𝒳 such that f∘g→∼id and g∘f→∼id.

Lemma 9.1.9 (2-Yoneda lemma).

Let 𝒳→𝒮 be a prestack and s∈𝒮. Then Hom𝒮⁡(−,s) is a prestack on 𝒮 and the functor Hom⁡(s,𝒳)→𝒳⁡(s) given by f↦fs⁢(ids) is an equivalence of categories.

Now we want a fiber product of prestacks for 𝒳→𝒴,𝒴′→𝒴, which is simply the final object in all 2-commutative diagrams

Fortuantely, these do exist.

Example 9.1.10.

The product 𝒳×𝒳 exists and is a prestack. We also have a diagonal Δ:𝒳→𝒳×𝒳.

Example 9.1.11.

Let 𝒳 be a prestack and a:𝒴→𝒳,b:𝒴′→𝒳 be morphisms. Then there exists a 2-cartesian diagram

9.1.2 Stacks

Definition 9.1.12.

A prestack 𝒳 over a site 𝒮 is called a stack if for every {Ui→U}∈Cov⁡(U), we have

  1. 1.

    (morphisms glue) There exists a unique a→b filling in the diagram

    lying over

    Precisely, this means that given a,b∈𝒳⁡(U) and ϕi:a|Ui→b such that ϕi|Ui⁢j=ϕj|Ui⁢j, then there exists a unique ϕ:a→b such that ϕ|Ui=ϕi for all i.

  2. 2.

    (objects glue) Given ai,aj and isomorphisms αi⁢j:ai|Ui⁢j→aj|Ui⁢j satisfying the cocycle condition on Ui⁢j, there exists a∈𝒳⁡(U) and isomorphisms ϕi:a|Ui→ai such that αi⁢j∘ϕi|Ui⁢j=ϕj|Ui⁢j.

Example 9.1.13.

Consider the prestack 𝖲𝗁𝖾𝖺𝗏𝖾𝗌 over 𝖲𝖼𝗁 with objects (S,F), where F is a sheaf on S. Then (S,F)→(S′,F′) is a pair f:S→S′ and α:F′→f∗⁢F such that the adjoint of α is an isomorphism F≃f−1⁢F′. We know that sheaves and their morphisms can be glued in the Zariski topology, so the prestack 𝖲𝗁𝖾𝖺𝗏𝖾𝗌 is a stack over 𝖲𝖼𝗁Zar.

Example 9.1.14.

Consider the prestack 𝖲𝖼𝗁𝖾𝗆𝖾𝗌 over 𝖲𝖼𝗁 with objects (T→S) and morphisms (T→S)→(T′→S′) is a pair f:T→T′ and g:S→S′ such that the two compositions T→S′ agree. Of course schemes can be glued in the Zariski topology, so 𝖲𝖼𝗁 is a stack in 𝖲𝖼𝗁Zar.

Proposition 9.1.15.

ℳg is a stack over 𝖲𝖼𝗁ét for g≥2.

Remark 9.1.16.

The stackiness conditions mainly come from descent theory.

9.2 Algebraic stacks and Artin’s axioms

From now on, we will work in the category 𝖲𝖼𝗁/S of schemes over S.

9.2.1 Algebraic stacks

Definition 9.2.1.

A morphism 𝒳→𝒴 of prestacks is representable by schemes if for all schemes T→𝒴, the fiber product 𝒳×𝒴T is also a scheme. A representable 𝒳→𝒴 is surjective, smooth, etc if for all schemes T→𝒴, the morphism 𝒳×𝒴T→T is surjective, smooth, etc.

Definition 9.2.2.

An algebraic space is a sheaf F on (𝖲𝖼𝗁/S)ét such that there exists a scheme U and a surjective étale U→F which is representable by schemes.

Definition 9.2.3.

A morphism 𝒳→𝒴 is called representable if for all schemes T→𝒴, 𝒳×𝒴T is an algebraic space. Moreover, a representable f:X→𝒴 is called surjective, smooth, etc if U↠𝒳×YT→T is surjective, smooth, etc.

Definition 9.2.4.

A algebraic stack over (𝖲𝖼𝗁/S)ét is a stack such that there exists a scheme U with a morphism U→𝒳 that is smooth, surjective, and representable.

This is equivalent to the following conditions (taken together):

  1. 1.

    The diagonal 𝒳→Δ𝒳×𝒳 is representable.

  2. 2.

    There exists a scheme U and a smooth surjective morphism U→𝒳.

A useful fact that the product and fiber products of algebraic stacks exist (in the category of algebraic stacks), and this is the same as their fiber products as prestacks.

Definition 9.2.5.

We say that f:𝒳→𝒴 of algebraic stacks is locally of finite type if for all (for some) smooth presentations V→𝒴 and U→𝒳×𝒴V, the composition

is locally of finite type.

9.3 Artin’s Axioms

Definition 9.3.1.

Let 𝒳 be a stack over (𝖲𝖼𝗁/S)ét. Then 𝒳 is limit-preserving if

is an equivalence of categories. Explicitly, this means:

  1. 1.

    Every object on the right hand side comes from ai|Spec⁡B for some i and some ai∈𝒳⁡(Spec⁡Bi).

  2. 2.

    For a,b∈𝒳⁡(Spec⁡Bi), we have

This should be viewed as a finiteness condition because if f:X→S is a scheme, then Hom⁡(−,X) is limit-preserving if and only if f is locally of finite presentation.

Definitions 9.3.2.

Let 𝒳 be a prestack over 𝖲𝖼𝗁/k. A formal object of 𝒳 is (R,{ξn},{fn}), where R is a complete local Noetherian k-algebra, ξn:Spec⁡R/𝔪Rn→𝒳, and fn:ξn→ξn+1 are morphisms in 𝒳 lying over Spec⁡R/𝔪kn↪Spec⁡R/𝔪Rn+1.

A morphism of formal objects (R,{ξn},{fn})→(T,{ηn},{gn}) is a collection of morphisms αn:ξn→ηn such that

commutes. These define morphisms Spec⁡R/𝔪Rn→Spec⁡T/𝔪Tn that are compatible, and hence a morphism Spec⁡R→Spec⁡T.

There is a functor from 𝒳⁡(Spec⁡R) to the category of formal objects (R,…). We say that a formal object is effective if it is in the essential image of this functor.

Definition 9.3.3.

Let R be a complete local Noetherian k-algebra and ξ∈𝒳⁡(Spec⁡R). We say that ξ is versal if for any diagram

such that B′↠B is a surjection of local Artinian k-algebras and α:ξ⁢_⁢Spec⁡B→η′|Spec⁡B is an isomorphism, then there exists Spec⁡B′→Spec⁡R such that α′=ξ|Spec⁡B′≃η′ extending α.

A formal object (R,{ξn},{fn}) is versal if in the same diagram, if we replace ξ wuth ξn and R with R/𝔪n, we can find a lift to Spec⁡R/𝔪m for some m≥n.

Theorem 9.3.4 (Artin’s axioms).

Let 𝒳 be a stack over (𝖲𝖼𝗁/S)ét. Then 𝒳 is an algebraic stack locally of finite type over k if and only if

  1. 0.

    𝒳 is limit-preserving.

  2. 1.

    The diagonal Δ:𝒳→𝒳×𝒳 is representable.

  3. 2.

    (formal deformations) for every x:Spec⁡k→𝒳, there exists a complete local Noetherian k-algebra (R,𝔪) and a versal formal object (R,{ξn},{fn}) such that ξ1=x.

  4. 3.

    Every formal object is effective.

  5. 4.

    (openness of versality) Let ξU:U→𝒳, where U is a scheme of finite type over k, and u∈U be a k-point such that ξU|Spec⁡𝒪̂U,u is versal. Then ξU is versal at all k-points in an open neighborhood of u.

Remark 9.3.5.

Suppose we want to prove that Artin’s axioms imply that 𝒳 is an algebraic stack locally of finite type over k. Given a formal object (R,{ξn},{fn}), we have an actual object ξ∈𝒳⁡(Spec⁡R). By approximation and algebraization, we have U→𝒳 finite type and versal at X=U. Now U→𝒳 is smooth, so we have ⨆U→𝒳 smooth, surjective, and representable.

Remark 9.3.6.

In many modular problems, the condition about formal deformations follows from the Rim-Schlessinger condition that for a diagram

of Artinian local k-algebras, the map

is an equivalence of categories. Here, if x0∈𝒳⁡(Spec⁡k), define F𝒞,x0 by A↦x∈𝒳⁡(Spec⁡A) such that x0→x lies over Spec⁡k→Spec⁡A. Then T⁢F𝒞,x0=F𝒞,x0⁢(k⁡[ε]) is a k-vector space. If dimkT⁢F𝒞⁡(x0)<∞, then we have the condition on formal deformations. Then smoothness of U→F𝒞,x0 gives versality.

Remark 9.3.7.

The third condition follows from Grothendieck’s existence theorem, and a deformation-obstruction theory gives openness of versality.

10 Che (Dec 03): Stack of coherent sheaves

Let k be an algebraically closed field of characteristic 0, (𝖲𝖼𝗁/k)ét be the étale site of schemes over k, and X be a projective scheme over k.

Definition 10.0.1.

Let 𝖢𝗈𝗁X be the category defined as follows:

  • •

    Objects are tuples (T,ℱ) of a scheme T and ℱ is a quasicoherent sheaf on X×T of finite presentation flat over T.

  • •

    Morphisms from (T,ℱ) to (T′,ℱ′) are pairs (h,φ), where h:T→T′ is a morphism of schemes and φ:(h′)∗⁢ℱ′→ℱ is an isomorphism of 𝒪XT-modules. Here, h′:XT→XT′ is the morphism induced from h.

There is a functor p:𝖢𝗈𝗁X→(𝖲𝖼𝗁/k)ét given by (T,ℱ)↦T, and we want to prove that 𝒳≔𝖢𝗈𝗁X is an algebraic stack. Also, we will abuse notation and write h for h′. We will prove that 𝒳 satisfies Artin’s axioms. Recall that these are

  1. (0)

    𝒳 is a stack. This means that 𝒳 is a prestack and objects and morphisms glue.

  2. (1)

    Δ:𝒳→𝒳×𝒳 is representable by algebraic spaces.

  3. (2)

    𝒳 is limit-preserving.

  4. (3)

    𝒳 satisfies the Rim-Schlessinger condition.

  5. (4)

    The tangent spaces T⁢ℱ𝒳,X0 and Inf⁡(ℱ𝒳,X0) are finite-dimensional.

  6. (5)

    Every formal object is effective.

  7. (6)

    𝒳 satisfies openness of versality.

These will imply that there is a smooth surjective covering of 𝒳 by a scheme.

10.1 𝒳 is a stack

First, we will prove that 𝒳 is a prestack. We note that 𝒳⁡(T) has objects finitely presented ℱ on XT flat over T and morphisms (idT,φ), where φ:ℱ′→ℱ is an isomorphism. Thus 𝒳⁡(T) is a groupoid. We now want to prove that pullbacks exist, which is clear because given h:T→T′ and an object (T′,ℱ′), our pullback is simply (T,h∗⁢ℱ′). To show that pullbacks are universal, consider h1:T1→ℎT2→h2T. Now suppose that we have morphisms (T1,ℱ1)→(T,ℱ) and (T2,ℱ2)→(T,ℱ). Now we want to prove that there exists an isomorphism ℱ1≃h∗⁢ℱ2. But we know that

and so we are done.

Now we want to check that objects glue. Given T and a covering {ai:Ti→T}, suppose we have ℱi on XTi flat over Ti. Suppose that the ℱi are isomorphic on intersections and satisfy the cocycle condition. We want to construct a ℱ on XT of finite presentation, flat over T. This follows from étale descent, so we are done.

We now want to prove that morphisms glue. To do this, we need to introduce some new notions.

Definition 10.1.1.

Given X=(T,ℱ) and Y=(T,𝒢), define the Isom presheaf Isom𝒳⁢(X,Y) sending a scheme S→T to the set Hom⁡(ℱ|S,𝒢|S) in 𝒳⁡(S).

It is easy to see that morphisms glue if and only if this is a sheaf, and we will omit the proof.

10.2 Representability of the diagonal

We want to prove that for all schemes S over k, the stack 𝒴 given by pullback in the diagram

is an algebraic space. By the 2-Yoneda lemma, a map S→𝒳×𝒳 is given by ξ=(S,ℱ),η=(S,𝒢) in 𝒳⁡(S). If we compute the fiber product S×𝒳×𝒳𝒳, we actually obtain the Isom presheaf Isom𝒳⁢(ξ,η). We will use without proof the fact that Isom𝒳⁢(ξ,η) is a closed subfunctor of Hom¯⁢(ℱ,𝒢) (defined by T↦Hom⁡(ℱT,𝒢T)).

We will only prove that Hom¯⁢(ℱ,𝒢) is representable by an algebraic space when ℱ,𝒢 are locally free (and X is a point apparently).1 In this case,

But now this is clearly represented by the total space Spec⁡Sym⁡(ℱ⊗𝒢∨) because2

10.3 Preservation of limits

Given T=limTi, we want to prove that 𝒳⁡(T) is the colimit of the 𝒳⁡(Ti). Given ℱ on X×T of finite presentation and flat over T, we want to show that there exist Ti,ℱi such that ℱ=(X→Xi)∗⁢ℱi.

Proposition 10.3.1.

Let R,Ri be rings such that R is the colimit of the Ri. Let M be an R-module of finite presentation. Then there exists i and a finitely-presented Ri-module Mi such that M=Mi⊗RiR.

Proof.

We know that M is finitely presented, so we have an exact sequence

Because R is the colimit of the Ri, there exists i such that aj⁢k lift to Ri. If we define Mi by

this is clearly the desired Mi. □

Checking flatness is too hard, so we will not do it.

10.4 Rim-Schlessinger

We will not prove this, but we will see that this is a natural condition to satisfy. Given a pushout diagram

where U,U′,V,V′ are spectra of local Artinian rings of finite type over k and U→U′ is a closed embedding, we want the functor

to be an equivalence of categories.

10.5 Finiteness

We want to prove that the tangent space T⁢ℱ𝒳,X0 and the infinitesimal automorphisms Inf⁡(ℱ𝒳,X0) are finite-dimensional. Given X0:Spec⁡k→𝒳 (which is just a finitely presented sheaf ℱ on X), the tangent space is

In addition, we know that

Because X is projective, these Ext groups are finite-dimensional.

10.6 Formal objects are effective

We will now prove that formal objects are effective. Let R∈𝒞̂ be a complete Noetherian local ring. Recall that a formal object is {ξn}, where ξ∈𝒳⁡(Spec⁡R/𝔪Rn) along with fn:ξn→ξn+1 living over the natural inclusions. A formal object is effective if it comes from an actual object over R.

In our case, a formal object is given by ℱn on X×Spec⁡R/𝔪Rn flat and finitely presented and fn:in∗⁢ℱn+1≅ℱn. We want to show that there exists ℱ over Spec⁡R restricting to each ℱn.

Theorem 10.6.1 (Grothendieck existence theorem).

Let A be a Noetherian ring which is complete with respect to some ideal I. Let f:X→Spec⁡A be a proper morphism. Let ℐ=I⁢𝒪X. Then the functor

is an equivalence.

In our case, take A=R and I=𝔪, and now we are done.

11 Johan (Dec 10): Openness of versality

Let k be an algebraically closed field. In this lecture, all schemes live over k for simplicity.

11.1 Moduli of curves

Recall that ℳ¯g is a stack such that a morphism U→ℳ¯g is a family C→U of stable curves of genus g. Assume that U is of finite type over k and let u0∈U⁡(k).

Definition 11.1.1.

We say that U→ℳ¯g is versal at u0 if 𝒪̂U,u0 and the map

given by C|Spec⁡𝒪̂U,u0 are a hull.

Lemma 11.1.2.

U→ℳ¯g is versal if and only if U is smooth at u0 and Tu0⁢U→T⁢DefCu0 is surjective.

Proof.

Earlier, we discussed that deformations of Cu0 are unobstructed. Therefore any hull is a power series ring over k. Thus U must be smooth at u0. If U is smooth at u0, then look at 𝒪̂U,u0←R, where R is the deformation ring of Cu0. But now this is a map of power series rings, and therefore defines a smooth transformation of functors if and only if the map on tangent spaces is surjective. □

Lemma 11.1.3.

We have openness of versality for ℳ¯g and ℳg.

Proof for ℳg.

By the previous lemma, we may assume that U is smooth. Call f:C→U and consider the exact sequence

This gives TU=f∗⁢f∗⁢TU→R1⁢f∗⁢TC/U. Taking fibers aat u0, we obtain

Also, R1⁢f∗⁢TC/U is a vector bundle of rank 3⁢g−3 over U. By the lemma and the assumption of versality at u0, we see this is surjective. Thus this is surjective in an open neighborhood. □

Proof for ℳ¯g1.

Consider the exact sequence

We think of this as ΩC/U→f∗⁢ΩU/k⁢[1] in the derived category, and now if we tensor with the relative dualizing sheaf, we have

where f! is the right adjoint to R⁢f∗ in this case. This gives

By positivity properties, R⁢f∗⁢(ΩC/U⊗ωC/U)=f∗⁢(ΩC/U⊗ωC/U). Taking the fiber at u0, we have

By the same argument as before, we are done. □

11.2 Properties of cotangent complexes

Suppose that U is of finite type over k. Then there is a complex LU/k∈Dcoh≤0⁢(𝒪U) such that H0⁢(LU/k)=ΩU/k called the cotangent complex. Suppose we write

with n,m minimal. This implies that f1,…,fm∈(x1,…,xn)2. Then H0⁢(LU/k⊗κ⁡(u0)) is the cotangent space of U at u0 and thus has dimension n. Next we note that

and this has dimension m.

Now let g:U→V be a morphism of schemes of finite type over k. Write u0↦v0. Then if we consider the distinguished triangle

we see that g is smooth at u0 if and only if H0⁢(LU/k⊗κ⁡(u0))←H0⁢(LV/k⊗κ⁡(v0)) is injective and H−1⁢(LU/k⊗κ⁡(u0))←H−1⁢(LV/k⊗κ⁡(v0)) is surjective.

Remark 11.2.1.

Suppose that f:X→U is a proper flat morphism corresponding to U→𝒳, where 𝒳 is the prestack pramaterizing families of of flat proper schemes. Then we have LX/U→L⁢f∗⁢LU/k⁢[1], and tensoring with the dualizing complex, we have

We should consinder instead

Versality is then related to properties of Hi⁢(canX/U⊗𝕃κ⁡(u0)) for i=0,−1.

11.3 Coherent sheaves

Let X→Spec⁡k be proper. Then 𝖢𝗈𝗁X/k is a stack such that for all U/k of finite type, a morphism U→𝖢𝗈𝗁X/k is a coherent sheaf ℱ on X×U which is flat over U.

Lemma 11.3.1.

We have openness of versality for U→𝖢𝗈𝗁X/k.

Easy case.

We will assume that ℱu0 is a vector bundle and ExtX2⁡(ℱu0,ℱu0)=0. As before, we may assume that U is smooth and ℱ is a vector bundle. We then have the Atiyah extension

and a map ℱ⊗ΩX×U/k1→ℱ⊗p∗⁢ΩU/k. Then we have the Atiyah class

Dually, we have f∗⁢TU/k→Hom⁡(ℱ,ℱ)⁢[1],

etc. By similar arguments as before, we obtain the desired result. □

General case (terrible).

We have an Atiyah class

This yields

The adjunction gives us

We need to show that formation of the left hand side commutes with base change and that

and then we can use cohomology and base change. □

11.4 A trick

Sometimes we can get openness of versality for a prestack 𝒳. Here, openness of versality holds for the prestack 𝒳 if

  1. 1.

    𝒳→𝒳×𝒳 is representable by algebraic spaces.

  2. 2.

    We have the condition (R⁢S∗), which is a version of RS where the rings do not need to be Artinian.

  3. 3.

    𝒳 is limit-preserving.

  4. 4.

    The following effectiveness holds: If A=limAn where

    where each An→A−1 is surjective with square zero kernel, then 𝒳⁡(A)=lim𝒳⁡(An).

In the example on 𝖢𝗈𝗁X/k, we have ℱn on X⊗An, and then we can attempt to take the limit ℱ of the ℱn. This fails because the limit is not quasicoherent, but we can fix it.