Patrick Lei

All notes

FGA explained

Talks by Caleb Ji, Columbia, Fall 2020.

FGA Explained 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 Caleb (Oct 16): Representable Functors and Grothendieck Topologies

1.1 Representable Functors

We will always denote categories by C.

Definition 1.1.1.

Given an object x∈C, define the functor hX:Cop→𝖲𝖾𝗍 by hx=Hom⁡(−,X).

Any morphism f:x→y induces a natural transformation hf:hx→hy. By the Yoneda lemma, this correspondence is bijective.

Lemma 1.1.2 (Yoneda Lemma).

Let x∈C and F:Cop→𝖲𝖾𝗍 be a functor. Then Hom⁡(hx,F)≃F⁡(x).

Proof.

Let θ:hx→F. This gives a map θx:hx⁢(x)→F⁡(x), and we can consider id→θx⁢(id). Now given t∈F⁡(x), we need hx⁢(U)→F⁡(U). Given U→x, then we have a map F⁡(x)→F⁡(U) and then t↦Ff⁢(t). We can check that these are inverses. □

Definition 1.1.3.

A functor F:C𝔬⁢𝔭→𝖲𝖾𝗍 is representable if it is naturally isomorphic to hx for some x.

Definition 1.1.4.

If F is a presheaf, a universal object for F is a pair (X,ξ) such that ξ∈F⁢X and for any (U,σ) where σ∈F⁢U, there exists a unique f:U→X such that Ff⁢(ξ)=σ.

Note that representability is equivalent to having a universal object.

Example 1.1.5.
  1. 1.

    For the first example, consider C=𝖲𝖼𝗁/R for some ring R. Then if F=Γ⁡(𝒪), then clearly this is isomorphic to h𝔸1 and the universal object is (𝔸1,x).

  2. 2.

    Let F⁡(X)={ℒ,s0,…,sn} where ℒ is a line bundle and s0,…,sn generate ℒ, then (ℙn,x0,…,xn) is a universal object.

1.2 Grothendieck Topologies

According to Wikipedia, this is supposed to be a pun on “Riemann surface.” We want to generalize the idea of a topology because the Zariski topology is awful. Instead of open sets, we will consider suitable maps (coverings).

Definition 1.2.1.

A Grothendieck topology on a category C is a specification of coverings {Ui→U} of U for each U∈C. Here are the axioms for coverings:

  1. 1.

    If V→U is an isomorphism, then {V→U} is a covering.

  2. 2.

    If {Ui→U} is a covering, for all V→U, the fiber products {Ui×UV→V} and form a covering of V.

  3. 3.

    If {Ui→U} is a covering and {Vi⁢j→Ui} are coverings, then {Vi⁢j→U} is a covering of U.

A category with a Grothendieck topology is called a site.

Example 1.2.2.

Here are some topological examples. Let X be a topological space.

  1. 1.

    The site of X is the poset category of open subsets of X. The fiber product is just the intersection, and a covering is a normal open covering.

  2. 2.

    (Global classical topology) Let C=𝖳𝗈𝗉. Here, the coverings are sets of open embeddings such that the union of the images covers the whole space.

  3. 3.

    (Global étale topology) Here, C=𝖳𝗈𝗉 and the coverings are now local homeomorphisms.

Returning to schemes, we have several examples of Grothendieck topologies.

  1. 1.

    (Global Zariski Topology). Let C=𝖲𝖼𝗁. The coverings are jointly surjective open embeddings.

  2. 2.

    (Big étale site over S) The objects are schemes over S and the morphisms are S-morphisms that are étale and locally of finite presentation.

  3. 3.

    (Small étale site) This the same as the big étale site, but with the added requirement that U→S is also étale.

  4. 4.

    (fppf topology) This stands for the French fidèlement plat et présentation finie. The morphisms are Ui→U flat and locally of finite presentation. A covering is a set of jointly surjective morphisms such that the map ⨆Ui→U is faithfully flat and of finite presentation. Note that flat and locally of finite presentation implies open.

  5. 5.

    (fpqc topology) This stands for the French Fidèlement plat et quasi-compacte. An fpqc morphism is a morphism X→Y that is faithfully flat and one of the following equivalent conditions:

    1. (a)

      Every quasicompact open subset of Y is the image of a quasicompact open subset of X.

    2. (b)

      There exists an affine open cover {Vi} of Y such that Vi is the image of a quasicompact open subset of X.

    3. (c)

      Given x∈X, there exists a neighborhood U∋x such that f⁡(U) is open in Y and U→f⁡(U) is quasicompact.

    4. (d)

      Given x∈X, there exists a quasicompact open neighborhood U∋x such that f⁡(U) is open and affine in Y.

    The fpqc topology is given by maps {Ui→U} such that ⨆Ui→U is an fpqc morphism.

To check that this is a topology, we have to do a lot of work. However, we will list some properties of fpqc morphisms and coverings.

Proposition 1.2.3.
  1. 1.

    The composition of fpqc morphisms is fpqc.

  2. 2.

    Given f:X→Y, if f−1⁢(Vi)→Vi is fpqc, then f is fpqc.

  3. 3.

    Open and faithfully flat implies fpqc. Moreover, faithfully flat and locally of finite presentation implies fpqc. This means that fppf implies fpqc.

  4. 4.

    Base change preserves fpqc morphsisms.

  5. 5.

    All fpqc morphsism are submersive. Thus f−1⁢(V) is open if and only if V is open.

Note that Zariski is coarser than étale is coarser than fppf is coarser than fpqc.

1.3 Sheaves on Sites

Recall that a presheaf on a space is a functor Xclop→𝖲𝖾𝗍. Similarly, if C is a site, then a presheaf is a functor Cop→𝖲𝖾𝗍.

Definition 1.3.1.

A presheaf on a site C is a sheaf if

  1. 1.

    Given a covering {Ui→U} and a,b∈F⁢U such that pi∗⁢a=pi∗⁢b, then a=b.

  2. 2.

    Given a covering {Ui→U} and ai∈F⁢Ui such that pi∗⁢aj=pj∗⁢ai (in the fiber product) for all i,j, there exists a unique a∈F⁢U such that pi∗⁢a=ai.

An alternative definition of a sheaf is that F⁢U→∏F⁢Ui⇉F⁡(Ui×UUj) is an equalizer.

Theorem 1.3.2 (Grothendieck).

A representable functor on 𝖲𝖼𝗁/S is a sheaf in the fpqc topology.

This means that given any fpqc cover {Ui→U}, then applying hX, if we have fi:Ui→X that glue on Ui×XUj→X, then the sheaf condition says we can glue to a unique f:U→X. In the Zariski topology, this is trivial. This also means that the fpqc topology is subcanonical, which means that hX are all sheaves.

We will prove this result by reducing to the category of all schemes. Note that the topology on 𝖲𝖼𝗁/S comes from the topology on 𝖲𝖼𝗁. Then we can show that if C is subcanonical, then C/S is subcanonical. Then we use the following lemma.

Lemma 1.3.3.

Let S be a scheme and F:𝖲𝖼𝗁/Sop→𝖲𝖾𝗍 be a presheaf. If F is a Zariski sheaf if V→U is a faithfully flat morphism of affine S-schemes, then F⁢U→F⁢V⇉F⁡(V×UV) is an equalizer, then F is an fpqc sheaf.

Proof.

Given {Ui→U} an fpqc covering, let V=⨆Ui. Then consider the diagram

the columns are bijective, so it suffices to check this for single coverings.

Now if {Ui→U} are finite and all affine and the second assumption holds, we have the diagram

Then the middle row is an equalizer. □

Proof of Theorem 1.3.2.

If X,U,V are affine, then we know that Hom⁡(R,−) is left exact, so the result follows from commutative algebra. Now it suffices to check the general case for single covers. If X=⋃Xi is a union of affines, then separatedness follows by restricting to the Xi and using the affine case.

Please read the rest of this yourself. □

2 Caleb (Oct 23): Sieves and Fibered Categories

Recall that a Grothendieck topology on a site C is a collection of coverings {Ui→U} such that isomorphisms are coverings, pullbacks preserve coverings, and if {Ui→U} and {Ui⁢j→Ui} are coverings, then {Ui⁢j→Ui} is a covering.

Also recall from last time that two important examples of Grothendieck topologiyes are the global Zariski topology and the fpqc topology. We also defined a sheaf and proved that representable functors are sheaves in the fpqc topology.

2.1 Sieves

A sieve is a way to have “your barrel full and your wife drunk.” The motivating question is:

Question 2.1.1.

When do two Grothendieck topologies give rise to the same sheaves?

Definition 2.1.2.

A subfunctor G of F:Cop→𝖲𝖾𝗍 has G⁡(u)⊂F⁡(u) and the morphisms are just restrictions.

Definition 2.1.3.

A sieve on U∈Ob⁡(C) is a subfunctor of hU.

Example 2.1.4.

Given a cover 𝒰={Ui→U}, then h𝒰⁢(T) are arrows that factor through the covering.

Definition 2.1.5.

A sieve S⊂hU belongs to a topology T if h𝒰⊂S for some covering 𝒰 in the topology T.

Definition 2.1.6.

A covering 𝒱 is a refinement of 𝒰 if h𝒱⊂h𝒰.

This gives a poset structure on Grothendieck topologies, where T≺T′ if every covering in T has a refinement that is a covering in T′. For schemes, we have

Then we know that T1,T2 are equivalent if T1≺T2≺T1.

Proposition 2.1.7.

Equivalent topologies have the same sheaves.

Proof.

Note that F:Cop→𝖲𝖾𝗍 is a sheaf if and only for S belonging to T, the map F⁢U≃Hom⁡(hU,F)→Hom⁡(S,F) is bijective. □

A real world application of this is that we can construct the sheafification of a functor F:Cop→𝖲𝖾𝗍 by

where Si ranges over sieves belonging to T.

2.2 Fibered Categories

We will consider functors pF:F→C. Our notation will be

Definition 2.2.1.

A morphism ϕ:ξ→η in F is Cartesian if for all ψ:ζ→η and h:pF⁢ζ→pF⁢ξ in C with pF∘h=pF⁢ψ, then there exists a unique θ:ζ→ξ such that pf⁢θ=h and ϕ∘θ=ψ.

If ϕ is a Cartesian arrow, then we say that ξ is a pullback of η. Pullbacks (with a fixed map in C) are unique up to isomorphism.

Definition 2.2.2.

A fibered category is a functor F→C such that for any f:U→V in C and η∈Ob⁡(F) with pF⁢η=V, then we have a cartesian arrow ϕ:ξ→η with pF⁢ϕ=f.

Definition 2.2.3.

Note that the fiber F⁡(U) is a category. The objects are ξ such that pF⁢(ξ)=U and morphisms are arrows h that map to the identity of U (i.e. pF⁢h=idU).

Definition 2.2.4.

A morphism of fibered categories over C is a functor H:F→G such that HU:F⁡(U)→G⁡(U) is a functor.

Definition 2.2.5.

A cleavage of F→C is a class K of cartesian arrows such that for all f:U→V and η∈F⁡(U), there exists a unique morphism in K mapping to f.

Question 2.2.6.

Does a cleavage give a functor from C to the category of categories?

Unfortunately, the answer is no. This is not a functor, but it does give a pseudofunctor, which is the same as a lax 2-functor. The idea is that idU∗ may not be the identity and f∗⁢g∗ may not be (g∘f)∗. However, they are canonically isomorphic.

Definition 2.2.7.

A pseudofunctor Φ on C (from Cop to 𝖢𝖺𝗍) is an assignment such that

  1. 1.

    For all objects U of C, Φ⁢U is a category.

  2. 2.

    For each f:U→V, f∗:Φ⁢V→Φ⁢U is a functor.

  3. 3.

    For all U of C, we have an isomorphism ϵU:idU∗≃idΦ⁢U.

  4. 4.

    For all U→V→W we have an isomorphism αf,g:f∗⁢g∗≃(g⁢f)∗ such that given f:U→V and η∈Φ⁢U, we have αidU,f⁢(η)=ϵU⁢(f∗⁢η), and αf,idV⁢(η)=f∗⁢ϵU⁢(η). In addition, we require that for all morphisms U→𝑓V→𝑔W→ℎT and θ∈F⁡(T), the diagram

    commutes.

Definition 2.2.8.

A cleavage is a splitting if it gives an honest functor.

Apparently every fibered category is equivalent to a split fibered category.

2.3 Examples

  1. 1.

    If a category C has fiber products, then the category of arrows is a fibered category over C, and the Cartesian arrows are the Cartesian diagrams.

  2. 2.

    If G is a topological group, then there is a “classifying stack” given by principal G-bundles.

  3. 3.

    Consider the category of sheaves on a site C. Given X an object, denote sheaves on X as sheaves in C/X. Then given f:X→Y, we have f∗:𝖲𝗁⁢Y→𝖲𝗁⁢X.

  4. 4.

    Let C=𝖲𝖼𝗁/S. Then given f:U→V, we have f∗:𝖰𝖢𝗈𝗁→𝖰𝖢𝗈𝗁⁡(U). However, (g⁢h)∗≃f∗⁢g∗ is not an equality. In the affine case, if f:A→B is a morphism of rings, then recall that quasicoherent sheaves are modules. Then f∗⁢M=M⊗AB. The problem is that M⊗BB≃M but they are not the same object. However, note that f∗⁢g∗=(f⁢g)∗ and that f∗,f∗ are adjoint.

3 Caleb (Oct 30): Étale morphisms and the étale fundamental group

Today we will take a break from the abstract nonsense.

3.1 Flat Morphisms

Definition 3.1.1.

Let f:X→Y be a map of schemes. Then f:X→Y is flat at x∈X if the map f♯:𝒪Y,f⁡(x)→𝒪X,x is flat. Then f is flat globally if it is flat at all x.

Note if f:Spec⁡B→Spec⁡A is flat if and only if A→B is flat. The geometric intuition behind flatness is that the fibers form a continuous family.

Example 3.1.2.

Consider the map Spec⁡k⁡[t]→Spec⁡k⁡[x,y]/(y2−x3−x2) given by x↦t2−1,y↦t3−t. This is the normalization of the nodal cubic, and we see that the integral closure of the second ring is given by adjoining the element yx.

Outside t=±1, the map of stalks is an isomorphism, but at 0, we see that

and thus the map is not flat.

Example 3.1.3.

Let A⊂B be integral domains with the same fraction field. Then Spec⁡B→Spec⁡A is faithfully flat if and only if A=B.

Proof.

The idea is to use that for any ideal I of A, we have I⁢B∩A=I. Then go read your commutative algebra homework (or your favorite commutative algebra text). □

Remark 3.1.4.

Normalization is flat if and only if it is an isomorphism.

Example 3.1.5.

A closed embedding is flat if and only if it is also open. In particular, closed embeddings are generically not flat.

Proposition 3.1.6.

If f:X→Y is a flat map between irreducibles, then f⁡(U) is dense in Y for all nonempty U⊂X.

Proof.

Reduce to the case Spec⁡B→Spec⁡A. We simply need η=0 to be in the image of f. Writing the diagram

we see that Uη=Spec⁡B⊗Ak⁡(η). This contains the ring B/η⁢B by flatness. If B=η⁢B, then B is nilpotent and this implies U=∅. □

Theorem 3.1.7.

Let f:X→Y be a morphism of locally Noetherian schemes. Then

with equality if f is flat.

Sketch of Proof.

By schematic nonsense, replace Y by A=Spec⁡𝒪Y,n, so Y is Noetherian local. We can also assume Y is reduced. We will induct on dimY. Recall that if t∈A, then dim(A/t⁢A)=dimA−1 for t not a zero divisor or a unit. □

Corollary 3.1.8.

If f:X→Y be a faithful flat map of algebraic varieties, then Xy is equidimensional and dimXy=dimX−dimY.

3.2 Étale morphisms

From now on we will assume f:X→Y is finite type and locally Noetherian.

Definition 3.2.1.

A morphism f of schemes is unramified at x if 𝒪Y,y→𝒪X,x satisfies 𝔪y⁢𝒪X,x=𝔪x and k⁡(x)/k⁡(y) is separable.

Definition 3.2.2.

A morphism is étale if it is flat and unramified.

Example 3.2.3.

Let L/K be an extension of fields. Then Spec⁡L→Spec⁡K is étale if and only if L/K is separable.

Example 3.2.4.

Let L/K be a field extension. Then f:Spec⁡𝒪L→Spec⁡𝒪K is flat and the separability condition holds. Then f is unramified (and hence étale) if and only if it is unramified in the sense of number theory. Here, we consider the product

and f is unramified if all ei=1. For example, for the extension ℤ→ℤ⁡[i], we see that (2)=(1+i)2 and thus this map is not unramified.

Example 3.2.5.

Consider Spec⁡k⁡[t]/(P)→Spec⁡k. Then prime ideals are irreducibles Q dividing P. Then f is étale at Q if and only if Q is separable and is a simple factor of P.

More generally, a standard étale morphism for a monic P⁡(T)∈A⁡[T] and b∈B=A⁡[T]/P⁡(T) such that P′⁢(T) is a is a unit in Bb, then ϕb:Spec⁡Bb→Spec⁡A is standard étale.

Example 3.2.6.

Consider Spec⁡k⁡[x,y]/(x2−y)→Spec⁡k⁡[y]. This map is not étale at 0. To be explicit, note that 2⁢x is not a unit in k⁡[x,y]/(x2−y), but it is a unit after localization at x, and thus Spec⁡Bx→Spec⁡k⁡[y] is étale.

Theorem 3.2.7.

Every étale morphism is locally standard étale.

The proof uses the following theorem:

Theorem 3.2.8 (Zariski’s Main Theorem).

Let ϕ:X→Y be a quasi-finite morphism of finite type between Noetherian schemes. Then ϕ can be factored into X→𝑓X′→𝑔Y, where f is an open embedding and g is finite.

Definition 3.2.9.

A quasi-finite morphism is a morphism with finite fibers.

Remark 3.2.10.

If ϕ:X→Y is a finite-type morphism of schemes of finite type over l, then ϕ is étale if and only if ϕ̂x:𝒪̂Y,y→𝒪̂X,x is an isomorphism. Geometrically, this means that the formal neighborhoods are isomorphic.

3.3 Étale fundamental group

Recall that if X is a nice topological space, there is an equivalence of categories

This functor is represented by the universal covering space X~→X, and we have π1⁢(X,x)=Aut⁡(X~,X~).

Now we want to make this work for schemes, so we take finite étale morphisms to be the covers. Then we define a functor

by F⁡(Y)=HomX⁡(x¯,Y), where X¯ is a geometric point of X. We see that F is not representable, but it is a projective limit of representables, so we define the étale fundamental group

Example 3.3.1.

Let X=Spec⁡k. Then π1⁢(Spec⁡k,z¯)=Gal⁡(ksep/k). Similarly, π1⁢(Spec⁡ℤ)=π1⁢(𝔸ℂ1)=1.

If X is a variety over ℂ, then π1et⁢(X)=π1⁢(Xan)̂, where Ĝ is the profinite completion. In particular, π1et⁢(ℙℂ1∖{0,1,∞})=F̂2. If X=ℙℚ1∖{0,1,∞}, then we have an exact sequence

4 Caleb (Nov 6): More on the étale fundamental group and fibered categories

Recall that étale is the same as flat and unramified. Consider the category of finite étale morphisms over X and choose a geometric point x¯∈X⁡(k¯). Then consider the functor F⁡(Y)=HomX⁡(x¯,Y). Unfortunately, the morphism Spec⁡k¯→X is not étale, so this functor is not representable. If it was representable, it would be represented by the universal cover, which does not necessarily exist. We will now construct a Galois theory for schemes.

Definition 4.0.1.

For a finite étale morphism Y→X, define the degree [Y:X]≔|F⁡(Y)|. This is invariant of the choice of basepoint.

Proposition 4.0.2.

If Y is connected and we have a diagram

then θ is determined by where it sends a single geometric point of Y. In particular, |AutX⁡Y|≤[Y:X].

Corollary 4.0.3.

There is a faithful action of AutX⁡Y on F⁡(Y).

Definition 4.0.4.

A Galois cover Y→X is a finite étale morphism such that Y is connected and |AutX⁡Y|=[Y:X]. Equivalently, the action of AutY⁡X on F⁡(Y) is transitive.

Remark 4.0.5.

A transitive action is the same as normality in both Galois theory and topological covering spaces.

Example 4.0.6.

Let X=Spec⁡K. Then finite étale morphisms to X are schemes of the form ⋃Spec⁡Li where Li/K is finite separable. Then Spec⁡Li is Galois if and only if Li/K is Galois. For a degree n extension, the action of Aut⁡(L/K) on F⁡(Spec⁡L) is simply the action on the n embeddings K⊂L⊂K¯.

Lemma 4.0.7.

For connected Y∈𝖥⁢𝖤´⁢𝗍/X, there exists Z Galois over X such that the diagram

commutes. In fact, for any pair Y,Z, there exists a Galois W surjecting onto both.

4.1 Universal Cover

This universal cover is not a scheme, but is a projective limit of schemes. We will define the universal cover X~ of X with the following:

  1. 1.

    A poset I indexing Galois covers Xi→X.

  2. 2.

    For any Xi,Xj∈I, there k∈I such that k>i and k>j.

  3. 3.

    If i<j, then we have ϕi⁢j:Xj→Xi compactible with geometric points and composition.

Example 4.1.1.

Take all Galois covers and set i<j if and only if there exists Xj→Xi. To make everything work, we can adjust the map by an automorphism of Xi because the action of the Galois group is transitive.

Proposition 4.1.2.

Suppose Y is Galois over X and there is a finite étale map Y→Z over X. Then AutZ⁡Y⊂AutX⁡Y is a subgroup and Z is Galois if and only if AutZ⁡Y is a normal subgroup. In this case, AutX⁡Z≃AutX⁡Y/AutZ⁡Y.

In particular, if Y,Z are both Galois, then we have a map AutX⁡Y→AutX⁡Z.

Definition 4.1.3.

Take X~ to be the universal cover of X. Then define the étale fundamental group by

Aternatively, recall the action of AutX⁡(Y) on F⁡(Y). Then we obtain an action of π1⁢(X,x¯) on

Proposition 4.1.4.

F is the direct limit of the functors HomX⁡(Xi,−).

Theorem 4.1.5.

The functor

is an equivalence of categories.

By Yoneda, we may define π1⁢(X,x¯)=Aut⁡(F). Here F is the “fiber functor.”

  1. 1.

    Let X=Spec⁡k. Then if we take Galois coverings Li/K whose union is ksep. Then we have

  2. 2.

    Now consider a normal scheme X with function field K⁡(X). Then K⁢(X)un is the composition of all finite extensions K⁡(Y) such that Y→X unramified. Then we have

  3. 3.

    Because ℚ has no unramified extensions, we have π1ét⁢(Spec⁡ℤ)=1. By local class field theory, for any number field K, we have

    where IK is the ideal class group, H† is the maximum abelian extension unramified at finite primes (narrow Hilbert class field), and H is the Hilbert class field, which is unramified at all primes.

  4. 4.

    If X=𝔸ℂ1, then we can prove using some differential form that π1ét⁢(X)=1. In fact, all finite étale covers are isomorphisms. However, this is not true in positive characteristic.

  5. 5.

    Consider the nodal cubic Spec⁡k⁡[x,y]/(y2−x3−x2). Then finite étale morphisms over X are of the form Spec⁡Rn with AutX⁡Rn≃ℤ/n⁢ℤ and thus π1ét⁢(X)=ℤ̂.

  6. 6.

    Let Xk be finite over k with Xk¯ connected. Then we have an exact sequence

    To prove this, first note that the composition of the two middle maps is trivial because the map Xk¯→Spec⁡k factors through Spec⁡k¯ (by definition of base change). Then because Xk¯ is connected, we see that j is surjective by using the fact that for

    we have AutXk⁡(XL)≃Gal⁡(L/k). Finally, we use some magic to show that i is injective.

5 Caleb (Nov 13): Fibered categories

5.1 Review of Fibered Categories

Here is a motivating example for fibered categories. Recall that F:𝖲𝖼𝗁/Sop→𝖲𝖾𝗍 sending a scheme X to the set of isomorphism classes of elliptic curves over X is not representable. Instead, we will replace this by a fibered category

with objects (X,(E,e)) being a scheme and an elliptic curve over it and morphisms are (f,g) such that

is a Cartesian diagram. This means that E′≃E×XX′. In particular, when X=X′, the morphisms are the automorphisms.

Recall the definition of a Cartesian arrow and a fibered category from the second lecture.

Example 5.1.1.

Suppose a category C has fiber products. Then the arrow category of C is fibered over C under (X→Y)↦Y.

Definition 5.1.2.

In a fibered category F→U, the fiber F⁡(U) is the subcategory mapping down to U.

Recall that a cleavage is a class of Cartesian arrows such that there is one for every U→V and η∈F⁡(U). This does not give a functor from C to 𝖢𝖺𝗍, but it does give a pseudofunctor.

Example 5.1.3.

Recall that a group is a category with one object. If G→H is a surjective morphism, then we can think about G as being fibered over H. Then the sequence G→H→1 being split is equivalent to the existence of a split cleavage.

5.2 2-Yoneda Lemma

Recall that if F:Cop→𝖲𝖾𝗍 is a functor, then Hom⁡(hX,F)≃F⁡(X). This is the ordinary Yoneda lemma, and it tells us that we have an embedding C→Hom⁡(Cop,𝖲𝖾𝗍). Now the 2-Yoneda lemma embeds Hom⁡(Cop,𝖲𝖾𝗍) into the 2-category of fibered categories over C.

Given Φ:Cop→𝖲𝖾𝗍, we will construct a fibered category FΦ→C. The objects are pairs (U,ξ) where U∈C and ξ∈Φ⁡(U). This maps down to U. The morphisms (U,ξ)→(V,η) are maps U→V such that Φf⁢(η)=ξ.

If X∈C is an object, then hX is sent to the category C/X over X.

Lemma 5.2.1 (Weak 2-Yoneda).

For X,Y∈C, we have Hom⁡(X,Y)≃Hom⁡(C/X,C/Y)≃Hom⁡(hY,hX).

Lemma 5.2.2 (2-Yoneda lemma).

Let F→C be a fibered category. Then Hom⁡(C/X,F)≃F⁡(X).

5.3 Categories fibered in sets and groupoids

All moduli problems will be fibered in groupoids, and when they are representable by schemes, this is when the functor is fibered in sets.

Definition 5.3.1.

A category is fibered in sets if F⁡(U) is a set, which means there are only identity morphisms.

Now recall that we have embeddings

Theorem 5.3.2.

The functor from presheaves to categories fibered in sets is an equivalence.

Definition 5.3.3.

A category is fibered in groupoids if F⁡(U) is a groupoid.

Definition 5.3.4.

Recall the Grassmannian Gr⁡(n,k) represents the functor given by

where Q is a quotient bundle over X that is free of rank n−k.

6 Caleb (Nov 20): Stacks

6.1 Pursuing Stacks

We have been discussing fibered categories, and now we will be discussing fibered categories over a site. Our slogan to keep in mind is that a stack is a fibered category over a site with descent. Alternatively, a stack is a 2-sheaf, or a sheaf taking values in categories.

Remark 6.1.1.

Algebraic stacks are very special kinds of stacks and we have the following analogy: algebraic stacks are to stacks as schemes are to sheaves.

Example 6.1.2.

The basic prototype of a stack is the arrow category of 𝖳𝗈𝗉. Here, this is a fibered category with fiber F⁡(U)=𝖳𝗈𝗉/U. Next, it is easy to see that we can glue functions locally, and formally, this means that HomS⁡(X,Y) is a sheaf. We can also construct spaces locally, and this is a cocycle condition. Given an open cover {Ui} of U and maps fi:Xi→Ui with transition maps ϕi⁢j:fj−1⁢Ui⁢j→fi−1⁢Ui⁢j with ϕi⁢k=ϕi⁢j∘ϕj⁢k, then there exists a space f:X→U such that there exist isomorphisms fi−1⁢(Ui)≃Xi and ϕi⁢j=ϕi∘ϕj−1.

Definition 6.1.3.

We now define an object with descent data. This is a collection ({εi},{ϕi⁢j}) for a cover {Ui→U} such that εi∈F⁡(Ui) and ϕi⁢j:pr2∗⁢(εj)∼pr1∗⁢εi.

Remark 6.1.4.

The choices of pr1∗,pr2∗ depend on the cleavage chosen. This can also be phrased in terms of sieves, where

Here, we have the diagram

Given a covering {Ui→U}, define the category F⁡({Ui→U}) with the following data:

Objects:

These are simply objects with descent data.

Morphisms:

These are morphisms αi:εi→ηi such that the diagram

commutes.

Definition 6.1.5.

A functor F is a prestack if for each covering {Ui→U} the functor F⁡(U)→F⁡({Ui→U}) is fully faithful. If this functor is an equivalence, then F is a stack.

Note that this is a sheaf-like condition.

Definition 6.1.6.

An object with descent data is effective if it is isomorphic to the image of an object in F⁡(U).

Thus by definition, stacks are prestacks such that all objects with descent data are effective.

Example 6.1.7.

Let C be a site and F:Cop→𝖲𝖾𝗍 a presheaf. Recall that F is a fibered category in 𝖲𝖾𝗍. Then F is a sheaf if and only if F is a stack.

We now give some examples of stacks.

  1. 1.

    Let C be a site. Then 𝖲𝗁/C→C is a stack. This example is called a topos. This was apparently one of Grothendieck’s most profound ideas.

  2. 2.

    Let S be a scheme. Then the category 𝖰𝖢𝗈𝗁/S→𝖲𝖼𝗁/S is a stack in the fpqc topology. Apparently this is important.

  3. 3.

    The category 𝖠𝖿𝖿/S→𝖲𝖼𝗁/S of affine arrows over S is a stack in the fpqc topology.

  4. 4.

    Because the speaker has been traumatized by combinatorics, we give the following: Let T be the category of trees with morphisms the embeddings and coverings the jointly surjective ones.1 Then fiber products are just intersections. We will define F⁡(T) to be the category of proper colorings of T.

    At risk of further infecting these notes with combinatorics, consider the chromatic symmetric function

    Then the conjecture is that XT defines T. It is easy to see that the category defined above is a stack.

7 Caleb (Dec 04): Jacobians I

7.1 Divisors and Picard Groups

Let X be Noetherian, integral, separated, and regular in codimension 1.

Definition 7.1.1.

The group Div⁡X of WEil Divisors is the group generated by codimension 1 closed integral subschemes.

Because X is integral, it has a function field K. If we choose f∈K∗, we can choose Y⊂X of codimension 1. Then at ηY, the stalk 𝒪ηY is a DVR with fraction field K. Then we can write (f)=∑YvY⁢(f)⁢Y. This is a finite sum by the Noetherian condition. Divisors of this form are called principal.

Example 7.1.2.

Let X=ℙ1. Then Weil divisors of ℙ1 are sums ∑ni⁢Pi, where Pi is a point. If Pi=[2:1], then VPi⁢(f) is the order of the zero (or pole) at 2.

Definition 7.1.3.

The Class group Cl⁡X is the group Div⁡X/K∗, where we identify K∗ with the principal divisors.

Definition 7.1.4.

The principal divisors of ℙ1 is all sums ∑ni⁢Pi where ∑ni=0. Thus Cl⁡ℙ1≃ℤ.

Now let X/k be a projective curve over an algebraically closed field k.

Definition 7.1.5.

We define the degree of ∑ni⁢Pi to be ∑ni⁢[k⁡(Pi):k]=∑ni.

Now if f∈K⁢(X)∗, we know that (f)=φ∗⁢({0}−{∞}) and thus deg⁡(f)=0. Therefore we have a map Cl⁡X↠ℤ, and the kernel is Cl0⁡X.

Example 7.1.6.

Define X=Proj⁡k⁡[x,y,z]/(y2⁢z−x3+z⁢x2). We will show that Cl0⁡X is in bijection with the set of closed points. In one direction, let p0=[0:1:0]. Then f⁡(p)=p−p0. Then the line (z=0) intersects C with multiplicity 3 at p0. Thus if L∩C={P,Q,R}, then (L/Z)=P+Q+R−3⁢P0. On the other hand, if D is a degree 0 divisor, we can eventually replace it by Pi−P0 because C is not rational.

Definition 7.1.7.

The Picard group Pic⁡X is defined to be the set of isomorphism classes of line bundles on X with group operation the tensor product.

If X is locally factorial, then Cl⁡X≃Pic⁡X.

Example 7.1.8.

Let X=Spec⁡ℤ⁡[−5]. This is not a UFD, so its class group is nontrivial. The line bundles are fractional ideals and we have Pic⁡X≃C2 with generator I≔(2,1+5). Over D⁡(2), we have I=1 and over D⁡(3), I becomes generated by 1+−5.

7.2 Jacobian Functor

Now we will always assume X is a smooth curve. Then we can define the degree of an invertible sheaf L=∑ni⁢[k⁡(pi):k] by translation to Weil divisor language. Next, χ⁡(C,Ln)=n⁢deg⁡L+1−g.

Definition 7.2.1.

Define the functor F⁡(T) by

However, we don’t want line bundles on T interfering, so we now define

Unfortunately, this is still not necessarily representable.

Theorem 7.2.2.

There exists an abelian variety J over k and a morphism of functors f:PC0→J that is an isomorphism if C⁡(T) is nonempty. In particular, if T=Spec⁡k and C has a rational point, then J represents PC0.

Now let G=Gal⁡(K′/K). We need the map PC0⁢(Spec⁡k)→PC0⁢(Spec⁡K′)G to be bijective for PC0 to be representable.

Proposition 7.2.3.

There exists an exact sequence

where Br⁡(k′/k) is the Brauer group.

The idea of the proof is that to show injectivity, let L,L′∈Pic⁡C. If they become isomorphisc over k′, we show the isomorphism descends to Pic⁡X. Choose σ∈G. If i:Lk′→Lk′′ is an isomorphism, then we can apply σ to get another isomorphism between the two. This differs from i by cσ∈Aut⁡(Lk′)=(k′)×, and then cσ∈H1⁢(G,(k′)×)=0 is a coboundary by Hilbert Theorem 90. Thus c0=σ⁡(α)α for some α∈k′. The upshow now is that α−1⁢i is Galois invariant, so this descends to an isomorphism L→L′ over k.

Now suppose L∈Pic⁡(Xk′)G that does not come from Pic⁡X. Here, for σ,τ∈G, we have iτ⁢σ=cτ⁢σ⁢(σ∗⁢iτ∘iσ), and thus cσ⁢τ∈H2⁢(G,(k′)×).

Proposition 7.2.4.

Suppose X⁡(k)≠∅, or equivalently X has a rational point. Then PC0 is an étale sheaf and it is representable.

7.3 Relation to Complex Analysis

Consider H0⁢(C,Ω1), which has dimension g. We have an integration pairing H1⁢(C,ℤ)→H0⁢(C,Ω1)∨. The quotient of this map is Jan=H0⁢(C,Ω1)∨/H1⁢(C,ℤ) is a torus. Via Poincare duality, we have a pairing H1⁢(C,ℤ)⊗H1⁢(C,ℤ)→ℤ satisfying the Riemann bilinear equations. These conditions are called a Riemann form on Jan and give it the structure of an abelian variety. Now we have a map f:H0⁢(J,ΩJ1)→H0⁢(C,ΩC1). This gives a map

with kernel H1⁢(C,ℤ) by Abel’s theorem and Jacobi inversion. This gives an isomorphism between the algebraic and analytic constructions.

8 Caleb (Dec 11): Jacobians II

8.1 Construction of the Jacobian

Let X be a curve over k. Recall the functor

If X has a k-point, then it is representable. Assume x∈X⁡(k). We will now construct the Jacobian representing the functor G. Recall that for an elliptic curve over ℂ, we have Jac⁡(X)≅X. For a higher genus curve X, the Jacobian is birational to X(g)=Xg/Sg.

We will use the correspondence between k′-points of X(g) and effective Cartier divisors of Xk′ of degree g. Here, a Cartier divisor is an element α∈Γ⁡(X,Kx∗/𝒪X) and a Cartier divisor is effective if we have charts Ui such that αui∈𝒪Ui. For example, if we consider the Weil divisor n⁡[0], this is the Cartier divisor (𝔸1,xn).

Recalling the Riemann-Roch formula ℓ⁡(D)−ℓ⁡(K−D)=deg⁡D−g+1, chose effective divisors D,D′ of degree g. Then ℓ⁡(D+D′−g⁡[x])≥1 and semicontinuity and an instruction gives that the set U⊂X(g)×X(g) such that equality holds is open and nonempty. Up to scaling, there exists a unique f such that D′′=(f)+D+D′−g⁡[x] is effective.

Now we may define the map U→X(g) by (D,D′)→D′′. This gives a rational map X(g)×X(g)⇢X(g). By general results of Weil, we can upgrade this to an abelian variety X(g)→J.

Theorem 8.1.1.

For any birational group V over k, there exists a unique group variety G over k and a birational f:V→G such that f⁡(a⁢b)=f⁡(a)⁢f⁡(b).

In our case, we have V=X(g) and J is proper (or complete), so X(g)→J is an honest morphism.

Theorem 8.1.2.

J represents the functor G.

The idea of the proof is to define a map f:Pic0⁡(X)→J. Let D∈Pic0⁡(X) and suppose D+g⁡[x] is effective. Then if we consider the diagonal map X→ΔX(g)→ℎJ, then we define f⁡(D)=h∘Δ⁡(D+g⁡[x]). Otherwise, we can take D′ such that D+D′+g⁡[x] and D′+g⁡[x] are effective, and we set f⁡(D)=f⁡(D+D′)−f⁡(D′).

8.2 Some results about the Jacobian

Here are some results about Jacobians. We are very far away from proving any of them.

Theorem 8.2.1.

For any abelian variety A over an infinitely field k, there exists a Jacobian J such that J↠A.

Theorem 8.2.2 (Torelli).

Let k be an algebraically closed field. Then a curve C over k is uniquely determined by its canonically polarized Jacobian (J,λ), where λ is a certain ample line bundle giving a map J→Ĵ. Over ℂ, λ is given by the image of

The proof allegedly involves complicated combinatorial arguments. For arithmetic geometers, there is the Shafarevich conjecture:

Conjecture 8.2.3 (Shafarevich).

Let K be a number field and S be a finite set of primes of K. Then there are finitely many isomorphism classes C/k of genus g with good reduction outside of S.

Using various facts about Jacobians, this implies Faltings’s theorem.

Theorem 8.2.4 (Faltings).

Let X be a curve of genus g≥2 over ℚ. Then X has finitely many rational points.

Finally, there is a very roundabout way to prove the Riemann Hypothesis for curves. We can prove it for abelian varieties and then deduce it for a curve using the Jacobian.