Patrick Lei

All notes

Schemes

Lectures by Giulia Saccà, Columbia, Spring 2021.

Algebraic Geometry
Spring 2021
Notes by Patrick Lei
Lectures by Giulia Saccà

Columbia University

Disclaimer

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

Giulia’s iPad died at the beginning of the final lecture, and the rest of the lecture was spent with a phone camera pointed at a piece of paper, but the phone locked itself and eventually died as well. Giulia sent notes, so the transcription of these has replaced the live-ed notes.

Acknowledgements

I would like to thank Nicolás Vilches for pointing out mistakes and typos in these notes.

1 Schemes

1.1 Affine Schemes

Let R be a commutative ring. We will define the scheme Spec⁡R as a set, a topological space, and finally as a locally ringed space. Our goal is for R to be the ring of functions on Spec⁡R.

Definition 1.1.1.

We will define the set Spec⁡R to be the set of prime ideals P⊂R. Here, note that R is not a prime ideal and that (0) is prime if R is a domain.

Example 1.1.2.

If R=ℤ, then Spec⁡ℤ is the set of prime numbers together with 0. If R=k is a field, then Spec⁡k={(0)}. If R=k⁡[t], then Spec⁡R is the set of irreducible polynomials.

We will place the Zariski topology on Spec⁡R by declaring the closed sets to be V⁡(S)={𝔭∣𝔭⊃S} for any subset S⊂R. Some easy properties of V⁡(S) are:

  1. 1.

    If S⊂T, then V⁡(S)⊃V⁡(T).

  2. 2.

    If 𝔞=(S)⊆R, then V⁡(S)=V⁡(𝔞).

  3. 3.

    V⁡(S)=∅ if and only if 1∈(S) and V⁡((0))=V⁡({0})=Spec⁡R.

  4. 4.

    Given an ideal 𝔞⊂R, we have V⁡(𝔞)=V⁡(𝔞).

  5. 5.

    We verify that this forms a topology:

    • •

      First, V⁢(⋃αSα)=⋂αV⁡(Sα).

    • •

      Second, V⁡(𝔞⋅𝔞′)=V⁡(𝔞∩𝔞′)=V⁡(𝔞)∪V⁡(𝔞′).

Proof of all of these is a simple exercise. If R is considered as the set of functions on Spec⁡R, then for f∈R and x=𝔭∈Spec⁡R, we need to define f⁡(x). For this, we consider the field of fractions k⁡(x)=k⁡(𝔭) of R/𝔭. This is called the residue field.

Example 1.1.3.

If R=ℤ and x=(p) for p≠0, then k⁡(p)=ℤ/p⁢ℤ. If x=(0), then we see that k⁡(0)=ℚ.

Now we define f⁡(x) to be the image of f under the map R→R/𝔭→K⁡(R/𝔭)=k⁡(x). Then clearly {x∣f⁡(x)=0} is the closed subset V⁡(f).

Definition 1.1.4.

Given X=Spec⁡R and f∈R, we define Xf=X∖V⁡(f)=Spec⁡R⁡[1/f]. These are called the principal (or distinguished) open subsets.

Lemma 1.1.5.

Principal open subsets form a basis for the Zariski topology and are closed under finite intersections.

Proof.

If U is open, then we can write U=Spec⁡R∖V⁡(S)=V⁢(∑f∈S(f))=⋂f∈SV⁡(f)=⋃f∈SSpec⁡R∖V⁡(f), as desired. The proof that principal open subsets are closed under finite intersection is clear. □

Lemma 1.1.6.

Let g,fi∈R. Then Xg⊆Xfi if and only if V⁡(g)⊃V⁡(𝔞)=V⁡(𝔞), where 𝔞=∑(fi).

Proof.

We know Xg⊆⋃Xfi if and only if V⁡(g)⊇⋂V⁡(fi), which is equivalent to the right hand side. □

Corollary 1.1.7.

If g=1, then X=⋃Xfi if and only if 1∈∑(fi). In particular, because 1=∑ai⁢fi is a finite sum, and therefore X is a finite union of some of the Xfi. This implies that Spec⁡R is a quasi-compact topological space.

Definition 1.1.8.

Let Y⊆Spec⁡R=X. Then define

Proposition 1.1.9.
  1. 1.

    For all ideals 𝔞⊂R, we have I⁡(V⁡(𝔞))=𝔞.

  2. 2.

    V and I define inverse bijections

  3. 3.

    If Y⊂Spec⁡R is a subset, then V⁡(I⁡(Y))=Y¯, the Zariski closure of Y.

Proof.
  1. 1.

    If f∈I⁡(V⁡(𝔞)), then f∈𝔭 for all 𝔭⊇𝔞 and thus f∈𝔞.

  2. 2.

    This is left as an exercise.

  3. 3.

    Note that V⁡(b)⊃Y if and only if b⊆⋂𝔭∈Y=I⁡(Y).

□

In particular, we see that in general Spec⁡R has points that are not closed.

1.1.1 A Bit About Classical Varieties

Let k be an algebraically closed field and R=k⁡[t1,…,tn].

Definition 1.1.10.

A closed algebraic subset of kn (or 𝔸n⁢(k)) is the common set of zeros V⁡(f1,…,fm) of a finite set of polynomials f1,…,fm. Here, all of the same properties of these vanishing sets from Spec⁡R hold.

Now recall the Nullstellensatz from commutative algebra, which says that if k is a field and B a finite k-algebra, then B is a field and a finite extension of k.

Corollary 1.1.11.

All maximal ideas of k⁡[t1,…,tn] are 𝔪=(t1−x1,…,tn−xn) for xi∈k.

Corollary 1.1.12.

There is a bijection between radical ideals of k⁡[t1,…,tn] and closed algebraic subsets of kn given by V and I.

1.1.2 Back to Affine Schemes

Examples 1.1.13.

If R is a PID, then we can write 0≠f=∏i=1rpini and therefore the closed subsets of Spec⁡R are either Spec⁡R or a finite union of closed points. If R is also a local ring, then Spec⁡R={(0),𝔪}. If 𝔞⊂A and consider R=A/𝔞. Then Spec⁡A/𝔞=V⁡(𝔞). If f≠0 is not nilpotent, then Spec⁡Rf is the set of prime ideals not containing f, which is (Spec⁡R)f.

Suppose 𝔭∈Spec⁡R. Then we know that 𝔭¯=V⁡(𝔭)≅Spec⁡R/𝔭. This tells us that x∈Spec⁡R is a closed point if and only if it corresponds to a maximal ideal.

Remark 1.1.14.

Note that if k is not algebraically closed, kn is different from Spec⁡k⁡[t1,…,tn].

Example 1.1.15.

Let R be a domain. Then we see that (0)∈Spec⁡R is a generic point (it is dense). We will see that it is the unique generic point.

Definition 1.1.16.

Let X be a topological space. A closed subset Z⊆X is called irreducible if it is not the union of two proper closed subsets.

Proposition 1.1.17.

A closed subset Y⊆Spec⁡R is irreducible if and only if I⁡(Y) is prime. Moreover, any closed irreducible subset has a unique generic point.

Proof.

Let Y=V⁡(𝔞) and suppose 𝔞=𝔞=I⁡(Y). Then if 𝔞=𝔭 then 𝔭¯=Y and thus Y is irreducible. In the other hand, if f⁢g∈I⁡(Y), then f⁢g⁢(x)=0 for all x∈Y, and this means either f⁡(x)=0 or g⁡(x)=0. This implies that f∈𝔭 for all 𝔭∈Y or g∈𝔭 for all 𝔭∈Y. Then we can write Y=(V⁡(f)∩Y)∪(V⁡(g)∩Y) and by irreducibility, we see that Y=V⁡(f)∩Y, which implies that f∈I⁡(Y).

To prove the uniqueness of the generic point, we see that if there is more than one, then their closures are the same, so they contain each other and thus must be the same. For the existence of the generic point, we know Y=V⁡(𝔭) for a prime ideal 𝔭 and thus 𝔭 is the generic point. □

Notation 1.1.18.

For an irreducible closed subset Y⊆Spec⁡R we will denote by ηY the generic point of Y.

Now recall that a ring R is Noetherian if it satisfies the ascending chain condition of ideals.

Definition 1.1.19.

A topological space X is called Noetherian if any of the following conditions hold:

  • •

    Closed subsets satisfy the descending chain condition.

  • •

    Open subsets satisfy the ascending chain condition.

  • •

    Every open subset is quasi-compact.

Lemma 1.1.20.

A ring R is Noetherian if and only if Spec⁡R is Noetherian, and this implies that all open subsets of Spec⁡R are quasi-compact.

1.1.3 The Structure Sheaf

Let X be a topological space and 𝒞 be a category.

Definition 1.1.21.

A presheaf on X is a functor from the opposite category of the poset category of open sets to 𝒞.

Definition 1.1.22.

A presheaf ℱ on X is called a sheaf if for every open subset U⊆X and for all open coverings {Ui} the sequence

given by

is exact. Of course, if 𝒞=𝖠𝖻, then the second arrow can be replaced by (si)↦si|Ui∩Uj−sj|Ui∩Uj. What this means is that given two sections s,s′∈ℱ⁡(U) that agree on the restrictions, they s=s′. Also, if there exist si∈ℱ⁡(Ui) such that si|Ui∩Uj=sj|Ui∩Uj then there exists s∈ℱ⁡(U) that globalizes the si.

Now suppose ℬ={Ui} is a basis of open sets of X. Given ℱ⁡(Ui) for all Ui∈ℬ, we need to check when this defines a (pre)sheaf. Here, on an arbitrary open set V, we will simply define

To check when this presheaf is actually a sheaf, then we only need to check the gluing condition for U,{Ui}∈ℬ.

Now we will define the structure sheaf on X=Spec⁡R. Here, we write 𝒪X⁢(Xf)=Rf and 𝒪X⁢(Xg)→𝒪X⁢(Xf) be given by choosing n such that fn=a⁢g and then writing bfk↦ak⁢bfn⁢k. Now we need to check the two gluing conditions. The second is left as an exercise, so we will check the first.

If bfk↦0 for all i, then there exist mi such that fimi⋅b=0 in R. But then because Xf is quasi-compact, we can assume the cover is finite and choose n=max⁡mi. But then because Xf=⋃Xfi, we can write 1=∑ai⁢fin and this implies b=∑ai⁢fin⁢b=0.

The sheaf 𝒪X that we have defined is called the structure sheaf of X.

Definition 1.1.23.

The pair (X,𝒪X) is called an affine scheme.

Example 1.1.24.

For a field k, the space Spec⁡k is a point, but 𝒪X is different for different fields.

Example 1.1.25.

If X=Spec⁡D for D a D⁢V⁢R with uniformizer t, then Xt={0}, we see that 𝒪X⁢(Xt)=Dt=K⁡(D).

Proposition 1.1.26.

Let X=Spec⁡R.

  1. 1.

    The stalks of 𝒪X at 𝔭=x∈X are given by 𝒪X,x=R𝔭.

  2. 2.

    For any U⊆X open, we define 𝒪X⁢(U) to be the the set of s𝔭∈∏𝔭∈UR𝔭 such that whenever U=⋃Xfi, there exist si∈𝒪X⁢(Xfi) mapping to s𝔭 whenever 𝔭∈Xfi.

Proof.

First, the stalk ℱx=limU∋xℱ⁡(U) and thus the stalk of the structure sheaf is easily computed to be the localization. For the second part, we note that

□

Remark 1.1.27.

The same method used to construct 𝒪X can be used to associate a sheaf for every R-module M. Here, we will define M~⁢(Xf)=Mf=M⊗RRf. Here, M~ is a sheaf of 𝒪X-modules. This means that for all U⊆X, M~⁢(U) is an 𝒪X⁢(U)-module and the diagram

commutes whenever V⊆U.

Proposition 1.1.28.

HomR⁡(M,N)≃Hom𝒪X⁡(M~,N~).

Proof.

Let M→𝜑N be a map of R-modules. Now on Xf, we have a map Mf→φfNf by functoriality of localization, and then we can take limits to get a map on every open set.

In the other direction, let f:M~→N~ be a map of sheaves. Then we simply apply the global sections functor to obtain a map M→N. Checking that the two maps defined are inverses is easy and uses naturality of localiation. □

1.2 General Schemes

Definition 1.2.1.

A scheme is a locally ringed space (X,𝒪X) such that there exists an open cover {Ui} of X such that (Ui,𝒪X|Ui) is an affine scheme.

Lemma 1.2.2.

Let R be a ring and X=Spec⁡R. Then for any f∈R, the schemes (Xf,𝒪X|Xf),(Spec⁡Rf,𝒪Spec⁡Rf) are isomorphic.

Proof.

We check that the structure sheaves agree on principal open subsets. □

Proposition 1.2.3.

Let (X,𝒪X) be a scheme. Then for any open subset U⊆X, the pair (U,𝒪X|U) is also a scheme.

Proof.

We need to show there exists an open affine covering of U. It suffices to check for X an affine scheme, but then U is covered by principal open subsets. □

1.2.1 Morphisms of Schemes

We will now define morphisms of schemes. Here, this will be a map of topological spaces that is compatible with the structure sheaves. From this, we will obtain a locally ringed space. In the category of topological spaces, smooth manifolds, or complex manifolds, then f:X→Y is a regular function if and only if the pullback of regular functions is regular. This tells us that we have a morphism of sheaves 𝒪Y→f∗⁢𝒪X. In other words, we obtain a morphism 𝒪Y⁢(V)→𝒪X⁢(f−1⁢(V)) for any open V⊆Y.

Definition 1.2.4.

A morphism f:(X,𝒪X)→(Y,𝒪Y) of schemes is the data of a continuous map f:X→Y and a morphism of sheaves 𝒪Y→f∗⁢𝒪X such that for every point y∈Y, the map 𝒪Y,y→(f∗⁢𝒪X)y→𝒪X,x is a morphism of local rings. What this means is that the maximal ideal of 𝒪Y,y is sent to the maximal ideal of 𝒪X,x. In particular, we obtain an extension k⁡(y)↪k⁡(x).

Theorem 1.2.5.

Let X be a scheme and R a ring.

  1. 1.

    The assigment f:X→Spec⁡R↦Γ⁡(f∗):R→Γ⁡(X,𝒪X) determines a bijection

  2. 2.

    In particular, when X=Spec⁡B this determines an anti-equivalence between the category of affine schemes and the category of commutative rings

Proof.

First we will show that this assignment is injective. First, we will show that f is determined (set-theoretically) by Γ⁡(f∗) and then we will show that f∗:𝒪Y→f∗⁢𝒪X is determined by this.

First, for x∈X, we recall that I⁡(f⁡(x))={h∈R∣f⁡(h⁡(x))=0}=(fx∗)−1⁢𝔪x and this gives us a prime ideal in R. To find the morphism of sheaves, we will simply consider principal open subsets Spec⁡Rh. Here, we have ring maps

and thus this is uniquely determined.

Now given a map R→Γ⁡(X,𝒪X), we want to construct a map of schemes. First, we will reduce to the affine case and then prove the theorem in the affine case. Cover X=⋃Uα by affines Uα=Spec⁡Aα. Then given R→Γ⁡(𝒪X)→Γ⁡(Uα,𝒪X)=Aα, we will prove the reduction to the affine case. For maps R→φα we obtain maps Spec⁡Aα→fαSpec⁡R, so we want to show that these glue. It suffices to show that the diagram

commutes, which is obvious becayse these maps are all induced by R→Γ⁡(X,𝒪X)→Γ⁡(U,𝒪X).

Now given φ:A→B, we will construct a map f:Spec⁡B→Spec⁡A. This is given by 𝔭↦φ−1⁢(𝔭). This is continuous because

Then we regard B as an A-module via φ:A→B, so B~=f∗⁢𝒪X and we simply choose the map of sheaves to be the map of sheaves A~=𝒪X→B~ we defined previously. □

Corollary 1.2.6.

Spec⁡ℤ is the terminal object in the category of schemes. This means that every scheme X has a unique morphism X→Spec⁡ℤ.

Proof.

Maps X→Spec⁡ℤ are determined by maps of rings ℤ→Γ⁡(X,𝒪X), and clearly there is a unique such map of rings. □

Remark 1.2.7.

There is an important variant. Let S be a scheme. Then a scheme X/S is a scheme X together with a morphism X→S. If S=Spec⁡R, then we can also write X/R. Of course, we can define the category of schemes over S, and the terminal object is S.

Proposition 1.2.8.

Let X be a scheme and x∈X. Then

  1. 1.

    There exists a canonical morphism Spec⁡𝒪X,x→ixX.

  2. 2.

    Let X be a local domain. Then any morphism Spec⁡R→X that sends 0↦x factors uniquely via ix.

Proof.

Let 𝔭=x∈U⊆∈X and U=Spec⁡A. Then we have a map A→A𝔭 and clearly in the category of schemes, we have a commutative diagram

Of course, we should check that this is independent of the choice of open affine.

For the second part, we have a map 𝒪X=j∗⁢𝒪Spec⁡R, which is a map 𝒪X,x→𝔒Spec⁡R,𝔪=R𝔪=R. The other part of this is an exercise. □

Corollary 1.2.9.

Let k⁡(x) be the residue field of x. Then there exists a map Spec⁡K→X given by 0↦x if and only if k⁡(x)↪K.

Remark 1.2.10.

The set Hom𝖲𝖼𝗁⁡(Spec⁡k⁡[ε]/ε2,(X,x)) is in bijection with the Zariski tangent space.

Now we will consider some examples. First, let X=Spec⁡A and let 𝔞⊆A. Then Z=Spec⁡A/𝔞→Spec⁡A is a homeomorphism onto V⁡(𝔞), and the map A→A/𝔞 corresponds to 𝒪X→i∗⁢𝒪Z.

Next, we can consider the ideal 𝔞n. Here, we note that V⁡(𝔞)=V⁡(𝔞n), but the structure sheaves differ and so we can view Spec⁡A/𝔞m→Spec⁡A/𝔞m+1 as a closed subscheme.

1.2.2 Gluing Schemes

Suppose we are given the following data:

  • •

    A set I.

  • •

    For i∈I, a scheme Ui.

  • •

    For all i,j∈I an open subset Ui⁢j⊆Ui

with compatibility conditions in the form of isomorphisms φi⁢j:Ui⁢j→Ui⁢j with φi⁢i=id. We will also have triple compatibility conditions (cocycle condition).

Proposition 1.2.11.

Given the data above, there exists a scheme X and morphisms Ui→ψiX that are isomorphisms onto open subsets of X such that ψi⁢(Ui⁢j)=ψi⁢(Ui)∩ψj⁢(Uj)=ψj⁢(Uj⁢i) and X=⋃ψi⁢(Ui).

Example 1.2.12.

Let R be a ring. Then we will denote by 𝔸Rn≔Spec⁡R⁡[t1,…,tn]. Here, we will take U1=𝔸R1⊃U12=𝔸R1∖{0}=U21⊆U2=𝔸R1 and φ12=id. The scheme X=U1∪U2 is known as the affine line with double origin.

Example 1.2.13.

Let R be a ring. Then consider Ui≔Spec⁡R⁢[x0xi,…,xixî,…,xnxi] and Ui⁢j={xjxi≠0}⊆Uj⁢i. Then the scheme is X=ℙRn.

Definition 1.2.14.

Let 𝒞 be a category and S∈𝒞 with morphisms X→𝑓S,Y→𝑔S. Then the fiber product Z=X×SY is the limit of the diagram X→𝑓S←𝑔Y.

Remark 1.2.15.

If 𝒞=𝖲𝖾𝗍, then we can write Z={(x,y)∈X×Y∣f⁡(x)=g⁡(y)}.

Theorem 1.2.16.

Fiber products exist in the category of schemes.

Lemma 1.2.17.

If X,Y,S are affine, then X×SY exists.

Proof.

Write X=Spec⁡A,Y=Spec⁡B,S=Spec⁡C. Then A⊗CB is the pushout of A←C→B, and then we use the fact that affine schemes are opposite to commutative rings. Now we need to prove that the universal property holds for all schemes. But this simply reduces to the case of affine schemes by the next exercise. □

Exercise 1.2.18.

For all schemes T, there exists an affine scheme Aff⁡(T) that is universal with respect to morphisms T→Spec⁡A.

Proof of Theorem.

First, we note that if U⊆X, then if X×SY exists, then for the map p:X×YS, the preimage p−1⁢(U) is the fiber product U×SY. On the other hand, if X=⋃Ui and Ui×SY exist for all i, then X×SY exists. To see this, we simply use gluing.

Next, suppose S=Spec⁡C,Y=Spec⁡B are affine. Then if X=⋃Ui is a cover by open affines then Ui×SY exist for all i, so X×SY exists. Third, we cover Y=⋃Vi by open affines and then we now have X×SY for general X,Y.

Finally, we cover S=⋃Wi by open affines. Then if we consider Xi=f−1⁢(Wi),Yi=g−1⁢(Wi), the fiber products Xi×WiYi=Xi×SYi exist, so by gluing twice, we obtain the fiber product X×SY. □

Remark 1.2.19.

X×SY has an affine open cover by open subsets of the form Spec⁡A⊗CB.

Example 1.2.20.

We have an identification 𝔸Rn=𝔸ℤn×Spec⁡ZSpec⁡R. Similarly, we have 𝔸Rn+m=𝔸Rn×Spec⁡R𝔸Rm.

Definition 1.2.21.

Let X,S′ be schemes over S. Then the fiber product X×SS′→S′ is called the base change of X/S to S′.

Example 1.2.22.

Suppose k⊂K is a field extension and X/k is a k-scheme. Then XK=X×Spec⁡kSpec⁡K is a K-scheme.

Fiber products allow us to consider the notion of the preimage of a closed subset. For s∈S and morphism X→S, then the fiber product X×SSpec⁡k⁡(s)→Spec⁡k⁡(s) is the fiber of X→S over s.

Example 1.2.23.

Consider a closed subscheme Spec⁡A/𝔞=Z↪S=Spec⁡A. Then we may consider f−1⁢(Z)=X×SZ for some X→S. We may also consider the intersection of two closed subschemes Z=Spec⁡A/𝔞,W=Spec⁡A/𝔞′, which is simply the fiber product Z×SW=Spec⁡A/(𝔞+𝔞′).

Example 1.2.24.

Let k=k¯ and char⁡k≠2 and consider the morphism Spec⁡K⁡[x,y,t]/(x2−y⁢t)=X→S=Spec⁡k⁡[t]. So now for s=(t−a)∈Spec⁡k⁡[t], we see that Xs=Spec⁡k⁡[x,y]/(x2=a⁢y), and in particular X0=Spec⁡k⁡[x,y]/x2 is non-reduced.

On the other hand, if we consider X→Spec⁡k⁡[x], we see that X0 is the union of two copies of 𝔸1 intersecting at a point.

1.3 Quasicoherent Sheaves and Relative Spec

We will relativize the construction of Spec⁡R from a ring R. To do this, we will replace R with a sheaf of 𝒪X-algebras. Recall that if X=Spec⁡R and M is an R-module, then the ℬ-sheaf Xf↦Mf defines a sheaf M~ on X. Then we know that for two R-modules M,N,

This gives us a fully faithful exact functor (⋅)~ from R-modules to 𝒪X-modules.

Theorem 1.3.1.

The functor M↦M~ commutes with kernels and cokernels. In particular, it is exact.

Proof.

Recall that localization is exact. This implies that if K is the kernel of M→N, then K~ is the kernel of M~→N~. Next, for the cokernel of M→N, we note that C~ and coker⁡(M~→N~) are both sheaves extending the same presheaf. □

Definition 1.3.2.

A R-module M is called finitely presented if there is an exact sequence Rp→Rq→M→0 for p,q≥0.

Proposition 1.3.3.
  1. 1.

    If M is finitely presented, then ℋ⁢o⁢m𝒪X⁢(M~,N~)=HomR⁡(M,N)~.

  2. 2.

    The functor (⋅)~ commutes with arbitrary direct sums.

Definition 1.3.4.

Let X be a scheme. Then a sheaf ℱ of 𝒪X-modules is called quasi-coherent if for all x∈X there exists U⊆X and an exact sequence

Proposition 1.3.5.

Let X be a scheme and ℱ and 𝒪X-module. Then the following are equivalent:

  1. 1.

    ℱ is quasicoherent;

  2. 2.

    For all affine open U⊆X, ℱ|U=M~ for some 𝒪X⁢(U)-module M

  3. 3.

    there exists an affine open cover {Uα} such that ℱ|Uα=M~α for some 𝒪X⁢(Uα)-module Mα.

Proof.

Clearly we see that 2 implies 3, so we show that 3 implies 2. Let U⊆X be an open affine. Now we apply the exercise below to get a covering {Ui} such that Ui⊆U and Ui⊆Uα is principal in both U,Uα. Therefore ℱ|Ui=N~i for some 𝒪X⁢(Ui)-module Ni. Now if we write U=Spec⁡R and Ui=Spec⁡Ri, we see that if ji:Ui↪U, then (ji)∗⁢ℱ|Ui=N~i.

Next, the sequence

is exact by the sheaf axioms, so we are done because this is really an exact sequence

The implications 3 implies 1 and 1 implies 2 are trivial. □

Exercise 1.3.6.

Let X be a scheme, x∈X, and x∈U,V open subsets. Then there exists an open x∈W⊆U∩V such that W is principal in both U and V.

Example 1.3.7.

We will consider quasicoherent sheaves on Spec⁡R for R a discrete valuation ring. Then a sheaf on X=Spec⁡R is a map ℱ⁡(X)→resℱ⁡(X∖{0}). Now recall that ℱ is quasicoherent if and only if it comes from an R-module M, so we see that ℱ⁡(X)=M and ℱ⁡(X∖{0}) is M⊗K.

Remark 1.3.8.

Let f:X→Y be a morphism of schemes. Then f∗⁢𝒪X is a 𝒪Y-algebra.

Exercise 1.3.9.

If X is Noetherian or f is quasicompact and ℱ is quasicoherent, then f∗⁢ℱ is quasicoherent.

Example 1.3.10.

If f:Spec⁡A→Spec⁡B is a morphism of affine schemes, then f∗⁢M~=M~B and is thus quasicoherent.

Theorem 1.3.11.

Let Y be a scheme and ℛ be a quasicoherent sheaf of 𝒪Y-algebras. Then there exists a scheme X=Spec𝒪Y⁡ℛ→𝜋Y such that π∗⁢𝒪X=ℛ and for any f:Z→Y and morphism α:ℛ→f∗⁢𝒪Z, there exists a unique g:Z→X such that ℛ=π∗⁢𝒪X→𝛼π∗⁢g∗⁢𝒪Z=f∗⁢𝒪Z.

Proof.

If Y=Spec⁡A, then write ℛ=R~ and set X=Spec⁡R, and this has a natural morphism to Spec⁡A.

In general, cover Y=⋃Uα by open affines. Then write ℛ|Uα=R~α for some 𝒪Y⁢(Uα)-module. Then set Xα=Spec⁡Rα. To construct a transition map between Xα,Xβ, we simply consider the restriction Rβ=ℛ⁡(Uβ)→Γ⁡(Uα∩Uβ,ℛ), and this gives a morphism πα−1⁢(Uα∩Uβ)→Xβ and this factors through the πβ−1⁢(Uα∩Uβ) because the latter is the fiber product of Uα∩Uβ and Xβ over Y. The rest is obvious. □

Definition 1.3.12.

A morphism f:X→Y of schemes is called affine if for every U⊆Y open affine, the preimage f−1⁢(U)⊆X is affine.

Example 1.3.13.

The morphism Spec𝒪Y⁡ℛ→Y is affine.

Proposition 1.3.14.

The following are equivalent:

  1. 1.

    f:X→Y is affine.

  2. 2.

    There exists an open covering Y=⋃Uα such that f−1⁢(Uα) is affine.

  3. 3.

    f:X→Y can be written as Spec𝒪Y⁡ℛ→Y.

Proof.

The first implication is by definition and 3 implies 1 by the construction, so assume there exists an open covering Y=⋃Uα by affines such that f−1⁢(Uα) is affine. Set ℛ=f∗⁢𝒪X. By assumption, this is quasicoherent. Therefore there exists a morphism g that makes

commute. But there exists a covering (the preimages of the Uα) where g is an isomorphism, so g is an isomorphism. □

Theorem 1.3.15.

A scheme X is locally noetherian if for all x∈X, there exists an affine neithborhood Spec⁡R∋x such that R is a Noetherian ring.

If X is locally noetherian and quasicompact, then it is Noetherian.

Proposition 1.3.16.

X is locally Noetherian if and only if for any affine open Spec⁡R⊆X, R is Noetherian. Let X=⋃Spec⁡Rα where each Rα is Noetherian. Then if Spec⁡R⊆X is any affine open, we want to show that R is Noetherian. But here, we can choose V⊆Spec⁡R∩Spec⁡Rα, which is a principal open subset. Then V=Spec⁡(Rα)gα is Noetherian, so we can cover Spec⁡R=⋃Spec⁡Rfα by Noetherian schemes. But then affines are quasicompact, so this becomes a finite cover and thus Spec⁡R is Noetherian.

Proposition 1.3.17 (Affine Communication Lemma).

Let 𝒫 be a property enjoyed by affine schemes. Suppose that

  1. 1.

    If A has 𝒫, then Af also has 𝒫 for all f∈A.

  2. 2.

    If fi∈A such that (f1,…,fn)=A, then if Afi have 𝒫, so does A.

Then for any scheme X, if 𝒫 holds for one affine open cover, it holds for all affine open covers.

Proof.

Let X=⋃Spec⁡Ai where the Ai have 𝒫. Then there exists V⊆Spec⁡A∩Spec⁡Ai such that V is a principal open subset in both. □

Definition 1.3.18.

A morphism f:X→Y of schemes is called locally of finite type if there exist an open affine cover X=⋃Uα and open subsets Vα⊆Y such that f⁡(Uα=Spec⁡Aα)⊆Vα=Spec⁡Bα and Aα is a finitely generated Bα-algebra.

Proposition 1.3.19.

A morphism f:X→Y is locally of finite type if and only if for every pair of affine open sets U⊆X,V⊆Y such that f⁡(U)⊆V, 𝒪X⁢(U) is a finitely generated 𝒪Y⁢(V)-algebra.

Proof.

If A is a finitely generated B-algebra, then for all f∈A, Af=A⁡[1/f] is also a finitely-generated B-algebra. Next, if Afi are finitely generated B algebras and (f1,…,fn=A), we will show that A is a finitely-generated B-algebra. Suppose the Afi are generated by ai⁢jfikj and ∑ci⁢fi=1. We will show that the fi,ci,ai⁢j generate A as a B-algebra.

Let r∈A. Then in Ari, we see that r=pi⁢(ai⁢j)fiN, so by finiteness, we can assume that there exists M≥0 such that fiN+M⁢r=fiM⁢pi⁢(ai⁢j) for all i,j. Now we can write

and therefore

as desired. □

Definition 1.3.20.

Let X be a scheme and ℱ sheaf of 𝒪X-modules is called

  1. 1.

    Locally of finite type if for all x∈X, there exists U∋x and a surjection 𝒪Xn|U→ℱ|U→0.

  2. 2.

    Locally of finite presentation if for all x∈X, there exists U∋x open and an exact sequence

  3. 3.

    Locally free if for all x∈X, there exists U∋x and an isomorphism 𝒪Xn|U≃ℱ|U.

Remark 1.3.21.

If U=Spec⁡A and ℱ=M~, then ℱ is locally of finite type if and only if M is a finitely-generated A-module, locally of finite presentation if and only if M is finitely presented, and locally free if and only if M≃An.

Definition 1.3.22.

Let X be a scheme. An 𝒪X-module ℱ is called coherent if ℱ is locally of finite type and for all U open and morphisms 𝒪Xn|U→𝛼ℱ|U, ker⁡α is of finite type.

Example 1.3.23.

Let R=∏Rn, where Rn=k⁡[x0,…,xn]/(x02,x0⁢x1,…,x0⁢xn) and X=Spec⁡R. Then 𝒪X is not coherent. Indeed, the map 𝒪X→x0𝒪X is not of finite type.

Proposition 1.3.24.

A scheme X is locally Noetherian if and only if 𝒪X is coherent.

Proposition 1.3.25.

Let X be locally Noetherian. The following are equivalent for a sheaf ℱ of 𝒪X-modules:

  1. 1.

    ℱ is coherent.

  2. 2.

    ℱ is locally of finite presentation.

  3. 3.

    ℱ is quasicoherent and of finite type.

Proof.

Suppose ℱ is quasicoherent and finite type. Then let U=Spec⁡A be an open affine. Then A is Noetherian, so 𝒪Xn→ℱ corresponds to An→M on U, and this has finitely-generated kernel. □

Proposition 1.3.26.

Let X be locally Noetherian. Then kernels and cokernels of maps between coherent sheaves are coherent. This means that Coh⁡(X) is an abelian category.

Definition 1.3.27.

Let ℱ be a quasicoherent sheaf on a scheme X. For every point x∈X, the fiber of ℱ at x is the k⁡(x)-vector space ℱx⊗𝒪X,xk⁡(x)≕ℱ⁡(x). The rank of ℱ at x is dimk⁡(x)ℱ⁡(x)≕r⁡(x).

Example 1.3.28.

Let p∈X be a closed point. Then suppose i:Spec⁡(k⁡(p))→X and let ℱ=i∗⁢k⁢(p). Then ℱ⁡(x)=0 if and only if x≠p and ℱ⁡(p)=k⁡(p).

Lemma 1.3.29 (Nakayama).

Let X be a scheme and ℱ be a quasicoherent sheaf locally of finite type. If ℱ⁡(x)=0 at x∈X, then there exists U∋x open such that ℱ|U=0.

Proof.

Let Spec⁡A=V∋x be an affine open neighborhood. Then ℱ|V=M~. But this means that ℱx⊗𝒪X,xk⁡(x)=0, so 𝔪x⁢ℱx=0, and thus ℱx=0.

Now if m1,…,mk are generators of M as an A-module, then we see that mi∈ℱ⁡(V)→ℱx. By finiteness, up to restricting v, we can assume that mi∈ℱ⁡(V), and therefore mi=0, so ℱ|V=0. □

Corollary 1.3.30.

This tells us that Supp⁡(ℱ)⊆X is closed.

Corollary 1.3.31.

Let X be a scheme and ℱ quasicoherent and locally of finite type. Now choose x∈X and let s1,…,sk generate ℱ⁡(x) as a k⁡(x)-vector space. Then there exists and open U∋x and si~∈ℱ⁡(U) lifting the si that generate ℱ|U.

Proof.

Clearly, we can lift the sections, so we consider the cokernel of

We show that 𝒢⁡(x)=0. If we localize at x, then we obtain an exact sequence

and then by right-exactness of the tensor product, we have k⁢(x)n→ℱ⁡(x)→𝒢⁡(x)→0. But now the map k⁢(x)n→ℱ⁡(x) was surjective, so 𝒢⁡(x)=0. □

Proposition 1.3.32 (Upper Semicontinuity).

Let x∈X and ℱ be quasicoherent and locally of finite type. Then

  1. 1.

    The function rk:X→ℤ sending x↦rk⁡(ℱ⁡(x)) is upper semicontinuous.

  2. 2.

    If X is connected, reduced, and locally noetherian, then rk⁡(x)≡r if and only if ℱ is locally free of rank r.

Proof.
  1. 1.

    Let p∈X and rk⁡(p)≕r. Then there exists U∋p with a surjection 𝒪Xr|U↠ℱ|U, so by exactness of localization and right-exactness of tensor product, we obtain a surjection k⁢(x)r↠ℱ⁡(x) for x∈U. This tells us that rk⁡(ℱ⁡(x))≤r for all x∈U.

  2. 2.

    Assume that x↦rk⁡(ℱ⁡(x))≡r. Then for x∈X, we can choose Spec⁡A=U∋x, where A is Noetherian. Then the exact sequence

    corresponds to

    Now choose 𝔭∈Spec⁡A such that A𝔭 is a field and for some (a1,…,ar)∈N, at least one ai∉𝔭. Because A is Noetherian and X is reduced, there exist finitely many minimal primes, and now the sequence

    is exact and because rk⁡(ℱ⁡(p))=r, we see that N𝔭=0. □

Remark 1.3.33.

Passing to fibers does not preserve injections. For example, if we consider a field k, then the map 0→𝒪𝔸1→𝑡𝒪𝔸1→k⁡(0)→0 is exact.

Example 1.3.34.

Let X=Spec⁡k⁡[t]/t2, we can produce nontrivial sheaves with trivial fibers.

Now let X be a locally Noetherian scheme. Then if ℱ,𝒢 are coherent, then ℱ⊗𝒪X𝒢 and ℋ⁢o⁢m⁢(ℱ,𝒢) is coherent. In addition, any operation from multilinear algebra, in particular symmetric and exterior powers, can be performed on coherent sheaves.

Definition 1.3.35.

Let X be a scheme and ℱ be a quasicoherent sheaf of finite type. Then ℱ is called invertible if it is locally free of rank 1.

Example 1.3.36.

Let k be a field and consider 𝔸n. Then for f∈k⁡[t1,…,tn], the sheaf (f)~ is invertible.

The reason these are called invertible is because if ℱ is invertible, then there exists ℱ′ such that ℱ⊗𝒪Xℱ′≃𝒪X.

Definition 1.3.37.

We will denote the set (although we will see this is a group scheme with operation the tensor product) of isomorphism classes of invertible sheaves on X by Pic⁡X.

1.4 Functor of Points

We will begin a discussion of something that will eventually allow us to define moduli problems.

Example 1.4.1.

Let R be a ring and I=(f0,…,fn) be an ideal. We may consider the closed subscheme X≔Spec⁡R⁡[t1,…,tn]/I↪𝔸Rn. Then we know that

for any R-algebra A.

This generalizes to general schemes the idea that for A=k=k¯, then the closed points of X are the same as morphisms Spec⁡k→X. Can we recover a scheme X from the functor Hom⁡(−,X)?

Let 𝒞=𝖲𝖼𝗁/S for a fixed scheme X. Then for any X∈𝒞, consider the functor hX:𝒞op→𝖲𝖾𝗍 defined by hX⁢(−)=HomS⁡(−,X).

Remark 1.4.2.

We can perform this construction for any category 𝒞. For example, we can recover a group G as a set from Hom⁡(ℤ,G). Similarly, a smooth manifold can be recovered (as a set) from Hom⁡(pt,M).

Example 1.4.3.

Let X=𝔸ℤn. Then

Example 1.4.4.

Let X=Spec⁡R⁡[t,t−1]. Then we see that

Observe that for any T, X⁡(T) has the structure of a group, and this procedure will define a group scheme.

Example 1.4.5.

Fix a field k, let X/k, and let K/k be a field extension. Then

When K=k, then X⁡(k)={x∈X∣k=k⁡(x)}.

Example 1.4.6.

Let X→𝑓S be a scheme. Then XS⁢(S) is the set of sections of f. For example, if 𝒜→B is a family of abelian varieties over an integral scheme, then M⁢W⁢(π)=𝒜B⁢(K⁡(B)). Note taker: The most elementary example of this is an elliptic surface, for example a K3 surface or a rational elliptic surface.

Now we want to relate the functor of points to fiber products. By the universal property of the fiber product, we see that

Now observe that the assignment X↦hX=Hom⁡(−,X) is functorial in X! To see this, note that Hom⁡(−,−) is functorial in both arguments. This gives us a functor

After all of this discussion, we have the following question.

Question 1.4.7.

How much information is lost about X after passing to hX?

In fact, we lose no information because the functor X↦Hom⁡(−,X) is fully faithful.

Lemma 1.4.8 (Yoneda).

Let X,Y∈C. Then Hom⁡(X,Y)=Hom⁡(hX,hY). In fact, for any functor F:𝒞op→𝖲𝖾𝗍, we have Hom⁡(hX,F)≃F⁡(X).

Proof.

Let Y∈𝒞. Then consider the system of natural transformations ηY:hX⁢(Y)→F⁡(Y). In particular, if Y=X, we have ηX:hX⁢(X)→F⁡(X), and in particular the element ηX⁢(idX)∈F⁡(X).

Now given ξ∈F⁡(X). For every Y∈𝒞, we need to define ηY:hX⁢(Y)→F⁡(Y). Given f∈Hom⁡(Y,X), we have F⁡(f):F⁡(X)→F⁡(Y), so we make the assignment f↦F⁡(f)⁢(ξ). □

Corollary 1.4.9.

To give a morphism of schemes X→Y is the same as giving a natural transformation hX→hY, which is the same as giving compatible maps XS⁢(T)→YX⁢(T) for all T/S.

Remark 1.4.10.

In fact, it is enough to consider a scheme as a functor on affine schemes.

Now another natural question is the following:

Question 1.4.11.

Which functors in Hom⁡(𝒞op,𝖲𝖾𝗍) are of the form hX for some X∈𝒞?

Functors of this form are called representable.

Proposition 1.4.12.

A functor F is representable if and only if there exists X∈𝒞 and u∈F⁡(X) such that the map

is a bijection.

If F is representable, then X,u are unique up to unique isomorphism.

Proof.

Consider Z′→𝑔Z and suppose f∈F⁡(f). Then we assign (f∘g)↦F⁡(f∘g)⁢(u), and this will make the diagram

commute. □

Example 1.4.13.

Let X,Y∈𝖲𝖼𝗁/S. Then consider F:Z→HomS⁡(Z,X)×HomS⁡(Z,Y). This is represented by the fiber product X×SY with the two projections X×SY⇉X,Y.

We will now give examples of presheaves on the category of schemes over a fixed S.

Example 1.4.14.

Consider T↦Γ⁢(T,𝒪T)×. This is represented by Spec𝒪S⁡𝒪S⁢[t,t−1]≕𝔾m,S.

Example 1.4.15.

The functor T↦(Γ⁡(T,𝒪T))n is represented by 𝔸Sn=Spec𝒪S⁡𝒪S⁢[t1,…,tn].

Example 1.4.16.

The functor T↦G⁢Ln⁢(Γ⁡(T,𝒪T)) is represented by Spec𝒪S⁡𝒪S⁢[ti⁢j,det−1].

Example 1.4.17 (Projective Space).

Fix a positive integer n. Define the functor F:𝒞op→𝖲𝖾𝗍 by

To check that this is a functor, consider Z′→𝑓Z. Then pullback defines a map F⁡(Z)→F⁡(Z′) (by right-exactness). In fact, F is represented by ℙSn. The universal object is the line bundle 𝒪ℙn⁢(1). We will 𝒪⁡(1) as follows:

Let ℙn=⋃Uα where Uα=Spec⁡R⁡[x0/xα,…,xα/xα̂,…,xn/xα]. We will set

Then we see that multiplication by xβ/xα carries 𝒪⁡(1)|Uα to 𝒪⁡(1)|Uβ. Now we will study the global sections. For any homogeneous linear polynomial L⁡(x0,…,xn). Then on each open set we obtain a map of multiplication by L⁡(xi/xα). Gluing is obvious.

Conversely, suppose 𝒪Uα→sα𝒪Uα⁢(1) are morphisms that glue. Then the sα are rational functions, and we can show that they must come from a polynomial of degree 1.

Now choose a basis x0,…,xn of Γ⁡(ℙn,𝒪ℙn⁢(1)). Then these define a map 𝒪ℙnn+1→𝒪⁡(1). Then for any morphism Z→ℙn, we may consider 𝒪Zn+1=f∗⁢𝒪ℙnn+1↠f∗⁢𝒪⁢(1). Now given any 𝒪Zn+1→𝛼ℒ, we view α=(s0,…,sn). Then surjectivity implies that Z=⋃Zi for Zi={s⁡(x)≠0}. On each Zi, we see that 𝒪Zi→siℒ is surjective and is in fact an isomorphism. Now we will define

By construction, these maps glue, and so we obtain a morphism f:Z→ℙn. We can check that f∗⁢[𝒪ℙnn+1↠𝒪⁡(1)]=[𝒪Zn+1↠ℒ].

Example 1.4.18.

If we precompose 𝒪Zn+1→𝑔𝒪Zn+1→ℒ for some g∈G⁢Ln, we transform the map Z→ℙn by a projective transformation.

Example 1.4.19 (Grassmannian).

We can generalize ℙn to the functor

and obtain the Grassmannian Gr⁡(k,n+1).

Example 1.4.20 (Picard Functor).

Consider the “Picard functor” T/S↦Pic⁡(XT) for a given X/S. This functor is not representable. If it was representable, then F⁡(−)=Hom⁡(−,X), but we know that U⊆Z↦Hom⁡(U,X) is a sheaf of sets over Z. However, when we apply this to T↦Pic⁡XT, then there are nontrivial line bundles on XT that come from T. If ℒ is an invertible sheaf on T such that fT∗⁢ℒ≇𝒪XT, then let T=⋃Ui be an open cover such that ℒ|Ui=𝒪Ui. But then

sends fT∗⁢ℒ↦∏𝒪X𝒪Ui and so this sequence of sets is not exact.

Instead, we consider the relative Picard functor PicX/S, which is defined by

With additional assumptions on X/S (for example projective, integral, etc), we can show that this functor is representable.

Example 1.4.21 (Hilbert Scheme).

Later, we will define the correct notion of a closed subscheme. Then for a fixed X/S, we consider the functor

This is called the Hilb functor.

1.5 Properties of Schemes and Morphisms

Recall that if X is a scheme and U⊆X is open, then (U,𝒪X|U) is a scheme. We would like a similar definition for a closed subscheme.

Definition 1.5.1.

A morphism j:Y→X is called an open immersion if j is a homeomorphism onto an open subset U⊆X and the sheaf morphism 𝒪X→j∗⁢𝒪Y induces an isomorphism 𝒪X|U≃j∗⁢𝒪Y|U.

Example 1.5.2.

The maps 𝔸Rn→ℙRn onto the standard open subsets are open immersions. Similarly, if X=⋃Ui, then Ui→X is an open immersion.

Definition 1.5.3.

Let X be a scheme. Then a closed subscheme if a pair (Z,ℐ) of a closed subset Z⊆X and a sheaf of ideals ℐ⊆𝒪X supported on Z such that (Z,𝒪X/ℐ) is a scheme.

Example 1.5.4.

Let X=Spec⁡R and I⊆R be an ideal. Then if we take ℐ=I~, then (Z,𝒪X/ℐ) is a scheme isomorphic to Spec⁡R/I.

Remark 1.5.5.

Because 𝒪X/ℐ is an 𝒪X-module of finite type, we know that supp⁡𝒪X/ℐ is closed, and thus ℐ determines the closed subset Z⊆X.

Example 1.5.6.

If X is a scheme and ℐ is a quasicoherent sheaf of ideals, then Z=supp⁡𝒪X/ℐ is a scheme with structure sheaf 𝒪X/ℐ.

To see this, note that because ℐ is quasicoherent, then we can consider an open cover {Ui=Spec⁡Ai} such that ℐ|Ui=I~i for ideals Ii⊆Ai. But then we see that

and therefore the support is covered by affine open schemes Z∩Ui.

Example 1.5.7 (Non-example).

Let 0∈𝔸k1 be a closed point and let U=𝔸k1∖{0}↪𝔸k1 be the open immersion. Now if j:U↪X is open and ℱ is a sheaf on U, then we can define

One can check that this is a sheaf.

Now we see that j!𝒪U→𝒪X is a sheaf of ideals and 𝒪X/j!𝒪U is supported at 0. To see this, note that (j!𝒪U)x=𝒪X,x away from 0 and the stalk vanishes at 0. But now we know that Z/𝒪X/j!𝒪U is not a scheme because here Z={0}. If Z were a scheme, then Z would be affine, but Γ(Z,𝒪X/j!𝒪U)=k[t](t) is not a field.

The problem in the previous example is that j!𝒪U is not quasi-coherent!

Proposition 1.5.8.

Let ℐ⊆𝒪X be a sheaf of ideals and Z=supp⁡𝒪X/ℐ. If (Z,𝒪Z) is a closed subscheme, then ℐ is a quasicoherent sheaf of ideals.

Corollary 1.5.9.

Any closed subscheme of an affine subscheme is affine.

Remark 1.5.10.

Using the fact that quasicoherent sheaves form an abelian category, we see that ℐ⊆𝒪X is quasicoherent if and only if 𝒪X/ℐ is quasicoherent.

Proof of Proposition.

If U⊆X∖Z, then there is nothing to check. If x∈Z, then we pass to open affines x∈U⊆X. □

Definition 1.5.11.

A morphism f:Z→X of schemes is a closed immersion if

  1. 1.

    f is injective and a homeomorphism onto a closed subset of X.

  2. 2.

    The map 𝒪X→f∗⁢𝒪Z is surjective.

By definition, we have a bijection

Proposition 1.5.12.

Let i:Z→X be a closed immersion. For any U⊆X open affine such that U∩Z≠∅, the set i∗⁢(U)=Z∩U is an open affine subset of Z.

Proof.

Fix x∈Z and let x∈U1⊆X be open affine. Then let x∈V1⊆Z∩U be open affine. Now Z⊆V1 is a closed subset of Z (and of U1) and is disjoint from x∈Z. Now there exists α∈Γ⁡(U1,𝒪U1) that vanishes on Z∖V1 but not on x.1 But now (U1)α≕U is open affine, and therefore U∩Z=(U1)α∩Z=(V1)α. □

This means that if U=Spec⁡R, then i:U∩Z↪U is a map Spec⁡S→Spec⁡R, and thus (Z,𝒪Z)|V=(Spec⁡S,S~). This implies that i∗⁢𝒪Z=S~. Therefore ker⁡𝒪X→𝒪Z=I~=ker⁡(R→S)~ is quasicoherent and therefore we have proved the bijection between closed subschemes and closed subschemes.

Corollary 1.5.13.

The map f:Z→X is a closed immersion if and only if there exists an affine open covering {Ui} of X such that f−1⁢(Ui) is affine and Γ⁡(Ui,𝒪Ui)→Γ⁡(f−1⁢(Ui),𝒪f−1⁢(Ui)) is surjective.

Of course, given a closed subset Z⊆X, there may be many different quasicoherent sheaves of ideals that give Z different scheme structures.

Example 1.5.14.

Consider 0∈𝔸k1. Then the possible closed subschemes supported at 0 are given by Spec⁡k⁡[x]/x2 corresponding to ideals (x)⊇(t2)⊇⋯⊇(tn)⊇⋯. Note that these Artinian rings are relevant in deformation theory.

Definition 1.5.15.
  1. 1.

    Let X be a scheme. Then a subscheme of X is a pair (Y,𝒪Y) such that Y⊆X is locally closed and if U⊆X is the largest open subset of X such that Y⊆U is closed, then Y⊆U is a closed subscheme.

  2. 2.

    An immersion f:Y→X is a homeomorphism onto a locally closed subset such that for all y∈Y, the map 𝒪X,f⁡(y)→𝒪Y,y is surjective.

Now we may consider the image of a morphism of schemes. For example, we may have closed immersions (here, f⁡(Z) is a closed subscheme) and open immersions (here f⁡(Y) is an open subscheme).

Consider the map 𝔸k2→𝔸k2 given by (x,y)↦(x,x⁢y). Then the image of f is not locally closed, but it is a constructible set. Recall that if X is a topological space, then a constructible set S⊆X is a finite union of locally closed subsets.

Example 1.5.16.

Consider 𝔸k1 and let K=k⁡(t). Then we have the inclusion of the generic point Spec⁡K→𝔸K1.

Definition 1.5.17.

Let X,Y be integral schemes. Then a morphism f:X→Y is called dominant if f⁡(X)⊆Y is dense.

Example 1.5.18.

The morphism Spec⁡K→𝔸k1 is dominant. If U⊆Y is open, then the inclusion is dominant. Any surjective morphism is dominant. The map {x⁢y=1}⊂𝔸k2→𝔸1 is dominant.

Exercise 1.5.19.

Let f:X→Y is dominant if and only if f⁡(ηX)=ηY.

Example 1.5.20.

If A,B are domains, then Spec⁡A→Spec⁡B is dominant if and only if B→A is injective.

Now if f:X→Y is dominant, how bad can f⁡(X)⊆Y be? When does it contain an open subset?

Theorem 1.5.21 (Chevalley).

Let f:X→Y be a morphism of finite type and Y be Noetherian. Then for any constructible set S⊆X, f⁡(S)⊆Y is constructible.

Definition 1.5.22.

A morphism f:X→Y is of finite type if it is locally of finite type and quasi-compact.

Corollary 1.5.23.

If f:X→Y is of finite type, Y is Noetherian, and f is dominant, then f⁡(X)⊆Y contains an open subset.

Example 1.5.24.

If X is a Noetherian topological space, then C⊆X is constructble if and only if for all closed irreducible Z⊆X, Z∩C contains an open subset of Z or Z∩C¯⊊Z.

Corollary 1.5.25.

If f is as above and dominant and X,Y are integral, then f⁡(X)⊇U for some open subset U⊆Y.

Proof of Chevalley.

Because f is of finite type and Y is Noetherian, there exists a finite cover X=⋃Spec⁡Ai⁢j and Y=⋃Spec⁡Bi, where f⁡(Spec⁡Ai⁢j)⊆Spec⁡Bi. Then we know Ai⁢j is a finitely-generated Bi-algebra. But then each f⁡(C∩Spec⁡Ai⁢j) is constructible, we can assume that X=Spec⁡R,Y=Spec⁡S are affine. Then we have a morphism of rings S→R=S⁡[t1,…,tk]/I. We may also assume that I=I because this is a topological statement. In addition, we may also assume that S is reduced.

Now X→Y factors through 𝔸Sn, where X↪𝔸Sn is a closed immersion. Therefore we can assume X=𝔸n. But then 𝔸n→Spec⁡S factors as

and therefore we can assume X=𝔸S1. Because Spec⁡S is Noetherian, it has finitely many irreducible components Zi, so now we may assume that S is a domain. After this, we apply the following lemma. □

Lemma 1.5.26.

Let S be a domain and f:𝔸S1→Spec⁡S. Then for all C0⊆C⊆𝔸S1 where C0⊆C is open and C⊆𝔸1 is closed and irreducible, there exists an open subset U⊆Spec⁡S such that f⁡(C0)⊇U or f⁡(C0)∩U=∅.

Proof.

Let Spec⁡S be integral and η∈Spec⁡S be the generic point. Then let K≔K⁡(η). Then we have a commutative diagram

Using the following exercise, we see that either C→Spec⁡S is dominant, in which case ηC∈Cη≠∅, or not, in which case f⁡(C)¯⊆Spec⁡S and thus there exists U⊆Spec⁡S such that f⁡(C0)∩U=∅. Now there are two cases:

  1. 1.

    f−1⁢(η)∩C=𝔸K1. In this case, choose C=𝔸S1⊇C0⊇(𝔸S1)g for some 0≠g=a0⁢tn+a1⁢tn−1+⋯. Therefore 0≠a0∈S, so now we show that f⁡(C0)⊇Ua0−Spec⁡Sa0. But here, for all x∈Spec⁡C, we have f−1⁢(x)=Spec⁡k⁡(x)⁢[t]=𝔸k⁡(x)1, so

    But then if x∈Ua0, then a¯0≠0, so g¯≠0. But now

    But this is nonempty and thus x∈f⁡(C0), so Ua0⊆f⁡(C0).

  2. 2.

    f−1⁢(η)∩C≕Cη∈𝔸K1 is a closed point. Then C⊂V⁡(𝔭) for some prime ideal, and then 𝔭⁢K⁢[t]=(g) for some irreducible g∈K⁡[t]. Up to inverting denominators, we may assume that g∈S⁡[t]. But then C0⊆C⊆V⁡(G)⊆𝔸S1. Now we see that

    But now V⁡(g)∖C0 is constructible, so we can write V⁡(g)∖C0¯=⋃Wi as a finite union of closed irreducible subsets, and f⁡(Wi)¯⊊Spec⁡S. Therefore f¯⁢(Wi)⊆V⁡(α) for some 0≠α∈S. Now consider Spec⁡S⊇Uα⁢a0∋x:

    1. (a)

      If α⁡(x)≠0, then x∉f⁡(Wi)¯, so f−1⁢(x)∩V⁡(g)=f−1⁢(x)∩C0.

    2. (b)

      If a0⁢(x)≠0, then g¯⁢(t)∈k⁡(x)⁢[t] is nonzero of positive degree, so V⁡(g¯)⊆𝔸k1 is a nonempty closed subset.

    Therefore, for x∈Uα,a0, we have f−1⁢(x)∩C0=f−1⁢(x)∩V⁡(g)≠∅, so Uα,a0⊆f⁡(C0). □

Exercise 1.5.27.

Let f:X→Y with X,Y integral and ηY∈Y the generic point. Then XηY is irreducible.

We will use this to study closed points of schemes X/k of finite type over a field k.

Corollary 1.5.28.

Let X be of finite type over a field k. Then x∈X is a closed point if and only if k⁡(x) is an algebraic extension of k.

Proof.

Suppose x∈X is closed. Then x∈U=Spec⁡R⊆X is constructible in U. Then x∈U⊆𝔸kn→𝔸k1, and we will denote the coordinates by U→fi𝔸k1. Therefore fi⁢(x) is a constructible set in 𝔸k1, so it must be a closed point. But then k⁡(fi⁢(x)) is an algebraic extension of k. But then the extension k⊆k⁡(x) by the fi⁢(x) and is thus algebraic.

Now suppose k⊆k⁡(x) is algebraic. If x∈X is not closed, then there exists x≠y∈{x}¯. Then we can choose U∋x,y open affine, so x is not closed in U. But now x=𝔭∈Spec⁡R, so k⊆R/𝔭⊆k⁡(x). But then R/𝔭 is a finitely generated integral extension of k, so L is a field and thus 𝔭 is maximal and x is closed. □

Remark 1.5.29.

If k is algebraically closed, then closed points are precisely those with residue field k.

Example 1.5.30.

Let A be a local Noetherian ring. If U=X∖𝔪, then U satisfies the descending chain condition for closed subsets, and therefore has closed points. However, none of these points are closed in X because X has a unique closed point.

Corollary 1.5.31.

Let X be a scheme of finite type over k. Then if U⊆X is open and x∈U, then x is closed in U if and only if x is closed in X.

Corollary 1.5.32.

Let X be of finite type over k. Then

  1. 1.

    For any S⊆X closed, the closed points of S are dense in S.

  2. 2.

    X can be reconstructed as a topological space from the set of its closed points.

Proof.

Let S⊆X be closed. It suffices to show that for all open U⊆X, U∩S contains a closed point. Assuming U=Spec⁡R is affine, then S∩U=V⁡(I) for some ideal I⊆R, and the desired result follows from the existence of maximal ideals. □

1.6 More on Varieties

Note: Notes were not taken in great detail for this section.

Let k be an algebraically closed field. We know 𝔸n⁢(k)=kn. Then we will define affine algebraic sets to be the common zero set of a set of polynomials. Of course, we can declare the Zariski topology on 𝔸n⁢(k). Of course there is a natural correspondence between radical ideals of k⁡[x1,…,xn] and closed subsets of 𝔸n⁢(k) giving a correspondence between maximal ideals and points.

Remark 1.6.1.

With the Zariski topology, 𝔸n⁢(k) is a Noetherian topological space.

A morphism of algebraic sets is a map 𝔸n⊇X→Y⊆𝔸m that is expressible in terms of polynomials f=(f1,…,fm)∈k⁢[x1,…,xn]n. Dually, this defines a map

References that use the language of varieties are Chapter 1 of Hartshorne, Shafarevich, Griffiths-Harris, etc.

Now if X=V⁡(I)⊆𝔸n⁢(k) for a radical ideal I, then we can define

Definition 1.6.2.

The affine coordinate ring of X is k⁡[X]=Hom⁡(X,𝔸1⁢(k))=k⁡[x1,…,xn]/I.

Proposition 1.6.3.

If X is an affine algebraic set, then k⁡[X] is a reduced finitely-generated k-algebra. In addition, X is irreducible if and only if k⁡[X] is a domain.

For any x∈X, we define the maximal ideal 𝔪x=ker⁡(k⁡[X]→evxk). More generally, if Z⊆X is closed, then I⁡(Z)={f∣f⁡(x)=0}=⋂x∈Z𝔪x. For an ideal I⊆K⁡[X], we can define the closed set V⁡(I). For example, if I=(f), we can define the principal open subsets Xf.

Note that there is an equivalence of categories between irreducible affine algebraic sets and finitely-generated k-algebras that are domains.

1.6.1 Rational Functions

Let X be an irreducible algebraic set in 𝔸n. We may consider the field of fractions k⁡(X) of k⁡[X], and this will be called the field of rational functions on X. If fg∈k⁡(X), we have a map

whatever ℙ1 means. Really, we have a map X⇢𝔸1⁢(k). This is of course regular on Xg.

Remark 1.6.4.

If f′=f⁢h,g′=g⁢h, then fg=f′g′ in k⁡(X). We see that Xg⊇Xg′ and for all x∈Xg′, the fractions f⁡(x)g⁡(x)=f′⁢(x)g′⁢(x) agree.

Lemma 1.6.5.

If there exists fg,f′g′∈k⁡(X) such that there exists U⊆X where f⁡(x)g⁡(x)=f′⁢(x)g′⁢(x) for all x∈U, then fg=f′g′.

Proof.

Up to multiplying by something in k⁡[X], we may assume g=g′. But this means that f′⁢(x)=f⁡(x) for all x∈U, which means V⁡(f−f′)⊇U. By irreducibility of X, we have X=V⁡(f−f′), so f=f′. □

We now define a sheaf of regular functions, and we can upgrade affine algebraic sets to ringed spaces.

Definition 1.6.6.

Define the sheaf 𝒪X by

It should be obvious what the restriction maps are, and they are injective.

Lemma 1.6.7.

If f∈k⁡[X], then Γ⁡(Xf,𝒪X)=k⁢[X]f.

Proof.

One direction is obvious. In the other direction, if we define 𝔞={h∣h⁢g∈k⁡[X]}⊆k⁡[X], then we want to show that f∈𝔞. If we choose representatives 𝔤=g1/g2∈𝒪X⁢(Xf), we see that 𝔤2∈𝔞. Thus g2∉𝔪X for x∈Xf, so if x∈Xf, then x∉V⁡(𝔞). □

Proposition 1.6.8.

A map f:X→Y is a morphism of irreducible affine algebraic sets if and only if for all g∈k⁡[Y], then g∘f∈k⁡[X]. Equivalently, f is continuous and induces a morphism of sheaves.

Definition 1.6.9.

A locally ringed space (X,𝒪X) is called a prevariety if X is connected and there exists a finite covering of X by irreducible affine algebraic sets.

Remark 1.6.10.

Any prevariety X is Noetherian and irreducible.

Definition 1.6.11.

We define the function field of a prevariety (X,𝒪X) to be the fraction field of 𝒪X⁢(U) for any open affine U⊆X.

In particular, K⁡(U) is independent of the affine open subset U⊆X. In particular, all restrictions are injective and 𝒪X⁢(U)∩𝒪X⁢(V)=𝒪X⁢(U∩V).

Definition 1.6.12.

If X,Y are prevarieties, then a morphism f:X→Y is a morphism of locally ringed spaces.

Remark 1.6.13.

If f:X→Y is a morphism of prevarieties, then we do not have a pullback of rational functions in general.

Example 1.6.14.

Projective varieties are prevarieties.

Theorem 1.6.15.

Let k be algebraically closed. Then there is an equivalence of categories between integral schemes of finite type over k and prevarieties over k.

Proof.

Given a scheme (X,𝒪X), we will consider the prevariety (X⁡(k),𝒪X⁡(k)), where X⁡(k)=Homk⁡(Spec⁡k,X) is the set of closed points. Of course, if Spec⁡A=U⊆X is open affine, then U⁡(k) is an affine algebraic set. Now, we simply define 𝒪X⁡(k)⁢(U⁡(k))=𝒪X⁢(U), and this is a sheaf. We can view this as functions to k. Equivalently, we see that α−1⁢(𝒪X) coincides locally with the sheaf of regular functions because X is quasi-compact. Therefore we have defined a prevariety. To see that this is functorial, note that morphisms of schemes of finite type send closed points to closed points.

Now in the other direction, given X⁡(k), we will simply aff a generic point ηZ for every irreducible Z⊆X⁡(k). In fact, this defines a functor on topological spaces, called soberification. It is obvious what the topology should be. Of course, the inclusion α⁡(X⁡(k))→t⁡(X⁡(k)) is continuous and induces a bijection of open subsets. But now we note that if X⁡(k)⊆𝔸n is an affine algebraic set with A=k⁢[X⁡(k)], we note that t⁡(X⁡(k))=Spec⁡A. Therefore t⁡(X⁡(k)) is covered by finitely many open affines. Now it remains to check that α∗⁢𝒪X⁡(k) makes t⁡(X⁡(k)) a scheme. But this can be checked on affines. It is enough to show that if U⁡(k) is an affine algebraic set with A=k⁢(U⁡(k)), then Spec⁡A=U⊆X. But now, by definition, we have

so when U=Spec⁡A, we have 𝒪X⁢(U)=A. In addition, we have 𝒪X⁢(Uf)=𝒪U⁡(k)⁢(U⁢(k)f)=Af for all f∈k⁢[U⁡(k)], so we are done. To check functoriality, we note that soberification is functorial and then check that we obtain a map 𝒪Y→f∗⁢𝒪X from the map on k-points. □

Remark 1.6.16.

Note that this means that a morphism of (integral) schemes of finite type over k is determined by its value on closed points.

Remark 1.6.17.

Let X be an integral scheme of finite type over k and X⁡(k) be the corresponding prevariety. Then we have K⁡(X)=𝒪X,η and the field k⁢(X⁡(k))=Frac⁡(k⁢[U⁡(k)]) for any affine open k⁢(U⁡(k)), and these are the same field.

1.6.2 Comparison with GAGA

Let X be a projective scheme of finite type over ℂ. We may consider X⁡(ℂ)⊆ℙn⁡(ℂ). But now we may consider the analytic topology on ℙn⁡(ℂ), and so we may consider the analytic space Xan⁢(ℂ) with a sheaf of holomorphic functions 𝒪Xan. Of course, we may still consider the continuous map

Next, to a coherent sheaf ℱ, we may consider a coherent analytic sheaf ℱan=γ−1⁢ℱ⊗γ−1⁢𝒪X𝒪Xan. Now Serre’s GAGA tells us that this is an equivalence of categories.

1.6.3 Non-algebraically closed fields

We will now consider what happens if k⊊k¯. The most basic example is 𝔸k1=Spec⁡k⁡[t], and the closed points are given by (p⁡(t)) for irreducible polynomials p⁡(t). Then any finite extension of k that is generated by a single element is the residue field of a closed point. In particular, for any such field k′ we obtain a point in 𝔸k1⁢(k′). For another example, consider a field extension k⊊k′ with induced morphism X=Spec⁡k′→Spec⁡k. Then X has no k-points. Returning to affine space, any commutative diagram

gives us a k′-rational point of 𝔸kn×Spec⁡kSpec⁡k′=𝔸k′n. In particular, if X=V⁡(f1,…,fm)⊆𝔸kn, then X⁡(k′)={x∈(k′)n∣fi(x)=0,i=1,…,m}. We have a similar statement for ℙkn and closed subschemes.

Remark 1.6.18.

If X=⋃Ui is a union of open affines, then X⁡(k′)=⋃Ui⁢(k′).

Remark 1.6.19.

A k′-point x∈Xk⁢(X′) determines a field extension k⊆k⁡(x)⊆k′. If σ∈Aut⁡(k′/k), then we can precompose Spec⁡k′→𝜎Spec⁡k′→X to get another k′-point. In addition, we have Xk⁢(k′)σ=Xk⁢(kσ).

Example 1.6.20.

Let X be (locally) of finite type and k¯ be the algebraic closure of k. Then we have the map

From the characterization of the closed points, the image of Σ¯ consists of all closed points. Then we know that the absolute Galois group G=Aut⁡(k¯/k) of k¯ acts on Xk⁢(k¯). But then the G-orbits of this action are contained in the fibers of Σ and G acts transitively on the orbits, so the fibers are the G-orbits.

1.6.4 Classical projective geometry

We will now consider open and closed prevarieties.

  1. 1.

    An open subprevariety is simply an open subset with the structure sheaf restricted.

  2. 2.

    If X is a prevariety and Z⊆X is an irreducible closed subset, we can give Z the structure of a prevariety. Then if V⊆Z is open, we can define

    Note that if Z⊆X⊆𝔸n, the structure of Z as a subprevariety of X is the same as the one of Z⊆𝔸n⁢(k).

Now we will define projective varieties over algebraically closed fields. If we consider the scheme ℙkn, then ℙn⁡(k)=(kn+1∖{0})/k×. Then the open affine charts give us the structure of a prevariety. Now we define the sheaf of functions to be 𝒪ℙn⁢(U)=𝒪Ui⁢(U) when U⊆Ui and in particular if we homogeneize, we have

As a consequence, we have k⁢(ℙn⁡(k))=k⁡(Ui)=k⁡(x0,…,xn). Now we need to consider the global regular functions on ℙn. NOte that for a prevariety X and U,V open subsets, then 𝒪X⁢(U∪V)=𝒪X⁢(U)∩𝒪X⁢(V), so

Remark 1.6.21.

Note that if ℙn⁡(ℂ) is considered as a complex manifold, then Γ⁡(ℙn⁡(ℂ),𝒪Xan)=ℂ. More generally, there are no nonconstant global holomorphic functions on any compact complex manifold.

Now closed subsets of projective space are given by the vanishing of homogeneous polynomials Fi. This gives us the definition

Definition 1.6.22.

A prevariety is called a projective variety2 to a closed subprevariety of ℙn⁡(k).

Definition 1.6.23.

A prevariety is called quasiprojective if it is isomorphic to an open subset of a projective variety.

Remark 1.6.24.

The structure of a prevariety does not depend on the embedding in some ambient space.

We will now consider morphisms between (quasi)-projective varieties.

Proposition 1.6.25.

Let Y⊆ℙn⁡(k) be a quasi-projective variety.

  1. 1.

    Given f0,…,fm∈k⁡[x0,…,xn] homogeneous polynomials of the same degree such that V+⁢(f0,…,fm)∩Y=∅, then the map

    is a morphism of prevariety. Moreover, if g0,…,gm are homogeneous polynomails of the same degree with V+⁢(g0,…,gm)∩Y=∅, they define the same morphism if and only if gi⁢fj=gj⁢fi for all i,j.

  2. 2.

    Conversely, given φ:Y→ℙm⁡(k) a morphism of prevarieties, then φ is locally defined as above.

To prove this result, simply consider the affine open subsets of ℙm and the sets Y∩{fi≠0}.

Remark 1.6.26.

Compare this to the statement that ℙkn represents the functor taking a scheme X to the set of isomorphism classes of line bundles ℒ with linearly independent sections s0,…,sn.

Proposition 1.6.27.

If ∅≠V+⁢(f0,…,fm)=Z⊊Y, then there exists a morphism φ:U→ℙm⁡(k), where U=Y∖Z. This gives us a rational map Y⇢ℙn⁡(k).

Example 1.6.28.

The map

is a morphism of prevarieties. Given Z=V+⁢(I)⊆ℙn⁡(k), define the cone over Z to be C⁡(Z)=π−1⁢(Z)¯=V⁡(I)⊆𝔸n+1⁢(k). On the other hand, if I⊆k⁡[x0,…,xn] is a homogeneous ideal, then V⁡(I)=C⁡(V+⁢(I)).

Remark 1.6.29.

The cone over Z depends both on Z and on the embedding in projective space. On the cone over Z, the origin is usually a singular point. Sometimes (for example when Z is a point) it is not, but in general, the properties of the singularity depend on the properties of the embedding.

Example 1.6.30.

Consider the map ℙ1→ℙ2 given by [x,y]↦[x,y,0] with image W1={z=0} and C⁡(W1)=𝔸2⊂𝔸3. Consider the other map ℙ1→ℙ2 given by [x,y]↦[x2,x⁢y,y2] with image W2={x⁢z=y2} and cone C⁡(W2). But then W1≅W2≅ℙ1, but W2 has an A1 singularity at the origin while W1 is smooth.

Suppose g∈G⁢L⁢(n+1). Then the action of g on 𝔸n+1⁢(k)∖{0} is scaling-invariant, so it descends to ℙn⁡(k). Of course, we have an exact sequence

In fact, we will see that Aut⁡(ℙn)=P⁢G⁢L⁢(n+1).

Remark 1.6.31.

The Cremona group of birational transformations of ℙn is massive.

If I=(L0,…,Lm), then V+⁢(I)⊂ℙn is isomorphic to some ℙn−m. Therefore, P⁢G⁢L⁢(n+1) acts transitively on the set of m-dimensional linear subspaces in ℙn. Of course, this has the structure of an algebraic variety, the Grassmannian. This represents a functor 𝒪n+1↠ℒ, where ℒ has rank m.

A linear subspace is a hyperplane when it is defined by a single equation V+⁢(L), and is a line if it is isomorphic to ℙ1. We can also define the linear span of Z⊆ℙn to be

For two points p,q, the line containing p,q is denoted p⁢q¯.

Definition 1.6.32.

Points p1,…,pm∈ℙn are said to be in general position if no (k+1) of them lie on a (k−1)-plane.

Example 1.6.33.

Three points in ℙ2 are in general position if and only if they are not collinear.

Let H={x0=0}⊆ℙn be a hyperplane and suppose p=[1,0,…,0]∉H. For any closed subset Z⊆H, we can write p⁢Z¯=⋃z∈Zp⁢z¯⊆ℙn. This is a closed subset. Thus if Z=V+⁢(I) for some ideal in k⁡[x1,…,xn], then p⁢Z¯=V+⁢(I⋅K⁡[X0,…,Xn]).

Remark 1.6.34.

Let L1,L2⊆ℙn be linear subspaces. Then ⟨L1,L2⟩=ℙ3 if and only if L1∩L2=∅.

Example 1.6.35.

With the same assumptions as the remark, let Z⊆L1. Then L2⁢Z¯=⋃z∈Z⟨L2,Z⟩⊆ℙn is a closed subset.

Definition 1.6.36.

Let F be a homogeneous polynomial of degree e in k⁡[x0,…,xn]. Then V+⁢(F)=X⊆ℙn is called a hypersurface of degree d. If d=2, then it is called a quadric.

Example 1.6.37.

In ℙ2, the quadrics are either conics (x2+y2+z2), a pair of lines (x2+y2), or a double line (x2).

In characteristic not equal to 2, we know that if Q=V+⁢(q) is a quadric, then q is given by a symmetric matrix. But then we know that P⁢G⁢L⁢(n+1) acts on k⁢[x0,…,xn]d and thus on quadrics in ℙn in a way that preserves the rank. This tells us that Q is irreducible if and only if q has rank different from 2. If r=rk⁡(q)<n, then Q is an (iterated) cone over a rank r quadric in ℙr.

Consider the map ℙ1→v3ℙ3 given by [x,y]↦[x3,x2⁢y,x⁢y2,y3]. Clearly the image of v2 satisfies the equations A⁢D=B⁢C,A⁢C=B2,C2=B⁢D. On the other hand, we can explicitly construct x,y from the equations in A,B,C,D, so on the open chart W∩UA, we have the map [A,B,C,D]↦[A,B]. Similarly, on W∩UD, we have the inverse [A,B,C,D]↦[C,D]. The image of v3 is called a twisted cubic. It is easy to see that ⟨v2⁢(ℙ1)⟩=ℙ3.3

More generally, the degree d Veronese embedding ℙn→ℙ(n+dd)−1 is given by

Up to P⁢G⁢L⁢(n+1), replacing x0,…,xd with a different basis of k⁢[x0,…,xn]d. In fact, the image is closed, and if we write vd⁢(ℙn)=V+⁢(𝔞), then we know that 𝔞 is a homogeneous prime ideal.

Example 1.6.38.

Consider the degree 2 Veronese embedding v2:ℙ2→ℙ5. Then if H is a hyperplane, H∩v2⁢(ℙ2) is the image of a conic in ℙ2.

Example 1.6.39.

Let F∈K⁢[x0,…,x1]d and let X=V+⁢(F). Then there exists a hyperplane in ℙ(n+dd)−1 such that H∩vd⁢(ℙn)=V+⁢(F). This tells us that ℙn∖X is affine.

Note the following facts:

  1. 1.

    If k is a field and X,Y are k-schemes (locally) of finite type over k, then X×kY is (locally) of finite type over k.

  2. 2.

    If k=k¯, then X,Y are integral if and only if X×kY are integral. At the level of closed points, we note that X⁡(k)×Y⁡(k)=(X×kY)⁢(k).

Now observe that the product of projective spaces is a projective variety. Define the map

This is called the Segre embedding. If the (n+1)×(m+1) matrix of coordinates is given by Zi⁢j, then the image of the Segre embedding is given by the vanishing of the 2×2 minors. For example, we note that the Segre embedding

has image {A⁢D=B⁢C}, which is a smooth quadric. In fact, the images of ℙ1×{p},{p}×ℙ1 give us two families of lines on ℚ.

1.7 Dimension

Dimension is a topological property.

Definition 1.7.1.

Let X be a topological space. Then we define the dimension of X to be

where each Xi is a closed irreducible subset of X.

We will define the dimension of a scheme to be the dimension of its underlying topological space. If X=∅, then we set dimX=−∞.

Warning 1.7.2.

Even for a Noetherian scheme X, we can have dimX=∞. There is an example of Nagata in Vakil’s notes or as tag 02JC in the Stacks project.

It is easy to see that if X=Spec⁡A, then dimX=dimA.

Example 1.7.3.

Let A be a principal ideal domain such that A is not a field. Then all chains of prime ideals are given by 0⊆(t), and thus dimA=1.

Example 1.7.4.

If A is a ring and 𝔭0⊊⋯⊊𝔭ℓ is a chain of prime ideals, then in A⁡[t] as have the chain of ideals

and thus dimA⁡[t]≥dimA+1. If A is Noetherian, then this is an equality.

Lemma 1.7.5.

Let X be a topological space.

  1. 1.

    If Y⊆X has the subspace topology, then dimY≤dimX. If X is irreducible and dimX<∞, then this inequality is strict.

  2. 2.

    If X=⋃Uα is an open covering, then dimX=supαdimUα.

  3. 3.

    If X=⋃Xi is a union of irreducible components, then dimX=supdimXi.

  4. 4.

    If X is a scheme, then dimX=supx∈Xdim𝒪X,x.

Lemma 1.7.6.

The first three properties are obvious. Now if X=⋃Spec⁡Aα is a cover by affines, then we know dimX=supαdimAα. But now we know that for any prime ideal 𝔭, ht⁡(𝔭)=dimA𝔭, so we are done.

Corollary 1.7.7.

If Y↪X is a closed immersion, then dimY≤dimX. If X is integral and Y⊊X, then dimY<dimX.

Example 1.7.8.

Let A be a ring. Then dimA=0 if and only if all prime ideals are maximal. If A is Noetherian, then this holds if and only if A is Artinian, which is equivalent to A being the product of its localizations.

Definition 1.7.9.

A morphism Spec⁡B→Spec⁡A is integral if A→B is an integral map of rings.

Recall that B is a finite A-module if and only if φ is integral and B is a finitely-generated A-algebra. Recall that for integral morphisms if 𝔮1⊆𝔮2 are prime ideals of B such that 𝔮1∩A=𝔮2∩A, then q1=q2. Of course, integral morphisms of rings also satisfy going-up and going-down. The geometric interpretation of this is

Proposition 1.7.10.

Let Spec⁡B→Spec⁡A be an integral morphism with A⊂B. We already know that f is closed and surjective. The three properties of integral morphisms of rings imply that dimB=dimA.

Definition 1.7.11.

Let X be a topological space and Z⊆X be closed and irreducible. Then we will define

where each Zi is a closed irreducible subset of X.

Example 1.7.12.

If X=Spec⁡A and Z=V⁡(𝔭), then codimX⁡(Z)=ht⁡(𝔭)=dimA𝔭.

Example 1.7.13.

Let X be a scheme and Z⊆X be closed and irreducible. Then

Remark 1.7.14.

We always have the inequality dimZ+codimX⁡Z≤dimX. Equality holds if all maximal chains of closed irreducible subsets have the same length.

Definition 1.7.15.

A topological space X is called catenary if all maximal chains of closed irreducible subsets have the same length.

Example 1.7.16.

If A is a DVR over a field k with 𝔪=(t), then dimSpec⁡A⁡[x]≥2. It is easy to see that the ideals 𝔪1=(t⁢x−1) and 𝔪2=(t,x) are both maximal, so if Zi=V⁡(mi), we know dimZi=0. However, we see that codimX⁡(Z2)≥2 while codimX⁡(Z1)=1 because 𝔪1 is principal.

Theorem 1.7.17.

Let X be an integral scheme locally of finite type over k.

  1. 1.

    There is an equality dimX=tr.deg⁡k⁡(η). Moreover, X is catenary.

  2. 2.

    For all closed points x∈X, dimX=dim𝒪X,x.

  3. 3.

    If X,Y are finite type over k with Y integral and f:Y→X is dominant, then dimY≥dimX.

  4. 4.

    If f:Y→X is a quasi-finite morphism of schemes of finite type over k, then dimY≤dimX.

Remark 1.7.18.

Let A be a discrete valuation ring over k and X=Spec⁡A. Then Spec⁡k⊔Spec⁡k⁡(ηX)→Spec⁡A is a bijection on points, but dim(Spec⁡k⊔Spec⁡k⁡(ηX))=0<dimX=1.

The proof of the Theorem will require Noether normalization:

Theorem 1.7.19 (Noether normalization lemma).

Let A≠0 be a finitely generated integral k-algebra. Then there exists t1,…,td∈A such that the morphism k⁡[t1,…,td]→A is injective and integral.

Proof of Theorem 1.7.17.
  1. 1.

    Let X=⋃Spec⁡A where A is a finitely generated k-algebra. Now we only need to prove the statements for Spec⁡A. By Noether normalization, there exists t1,…,td such that k⁡[t1,…,td]↪A and X=Spec⁡A→𝔸kd is a finite morphism. Therefore dimX=dim𝔸kd, so we need to prove that dim𝔸d=d. Clearly we have dim𝔸n≥n, so suppose we have a maximal chain

    Choose some nonzero f∈𝔭1. Up to passing to an irreducible factor, we may assume that f is irreducible. Therefore we can replace 𝔭1 with (f). Now we may consider X=Spec⁡k⁡[x1,…,xn]/(f), and then its fraction field has transcendence degree n−1, so by induction we may assume that dimX=n−1.

    Now we will prove that X is catenary. It suffices to prove that if Z⊊X=Spec⁡A is a maximal proper closed irreducible subset, then dimZ=n−1. If X→𝜋𝔸kn is the morphism obtained from Noether normalization, then we consider π⁡(Z)⊊𝔸n. By going-down, we know π⁡(Z) is maximal, so we have now reduced to proving the statement for 𝔸n. But then W=V⁡(𝔔) for some prime ideal 𝔔, so let f∈𝔔 be nonzero. If g is an irreducible factor, then maximality of W implies that 𝔔=(g). But then the desired statement about dimension is simply a statement about the transcendance degree of k⁡[x1,…,xn]/f.

  2. 2.

    We can reduce to the case where X=Spec⁡A. Then we know dimX=dimA=dimA𝔪 for any maximal ideal x=𝔪∈Spec⁡A because X is catenary.

  3. 3.

    Let y∈Y satisfy f⁡(y)=ηX and set Z≔{y}¯. Therefore Z→X is dominant, so ηZ→ηX, so we have an extension k⁡(ηX)⊆k⁡(ηZ) and therefore dimX≤dimZ≤dimY.

  4. 4.

    We reduce to the affine case. Up to passing to the closure of the image, we may assume f is dominant. Then we have f:Spec⁡A→Spec⁡B and then f−1⁢(ηY) is a finite set, and in fact is a 0-dimensional scheme of finite type over k⁡(ηY). In particular, ηX is a closed point of ηY, and thus the field extension k⁡(ηY)⊆ηX is finite. This means that dimY≤dimX. □

Here is a general statement about unique factorization domains A with Spec⁡A=X. If Z⊆X is a closed subset all of whose components have codimension 1, then Z=V⁡(f) for some f∈A. The converse also holds. If we remove the assumption of unique factorization, then the converse is still true by Krull’s principal ideal theorem.

To prove the statement for a Noetherian UFD, then there are finitely many irreducible components Z1,…,Zs and if Zi=V⁡(fi), then Z=V⁢(∏fi). Therefore we can assume Z is irreducible, but then every height 1 prime ideal is principal.

In the other direction, we may assume that f is irreducible. If there exists a prime ideal 0⊆𝔭⊆(f), then we can assume that ht⁡(𝔭)=1 by the Noetherian assumption. But then we know p is principal, so 𝔭=(f′), and this implies that f∣f′ and thus (f)=(f′). Thus ht⁡((f))=1.

Remark 1.7.20.

A Noetherian domain A is a UFD if and only if every prime ideal of height 1 is principal.

Remark 1.7.21.

Let Q=V⁡(x⁢y=z2)⊂𝔸k3. Then V⁡(z) has two irreducible components, and each line cannot be cut out by a principal ideal.

Theorem 1.7.22 (Krull principal ideal theorem).

Let A be Noetherian and f∈A be nonzero. Let 𝔭∋f be a minimal prime ideal containing f. Then ht⁡(𝔭)≤1. In fact, if f is not a zero divisor, then ht⁡(𝔭)=1.

Remark 1.7.23.

If X is locally of finite type, we can prove Krull by Noether normalization and the UFD property.

Theorem 1.7.24.

Let X be locally Noetherian and f∈Γ⁡(X,𝒪X). Then every irreducible component of V⁡(f) has codimension 0 or 1.

Corollary 1.7.25.

Let X be locally Noetherian and f1,…,fr∈Γ⁡(X,𝒪X). Then every irreducible component of V⁡(f1,…,fr) has codimsion at most r.

Proof.

We may assume that X=Spec⁡A is affine with A Noetherian. We will induct on r. When r=1, then this is just Krull’s principal ideal theorem. Now consider V⁡(f1,…,fr−1)⊇V⁡(f1,…,fr) and now let Z⊆V⁡(f1,…,fr) be an irreducible component. Then let W⊆V⁡(f1,…,fr−1) be an irreducible component containing Z. By induction, we know W∩V⁡(fr)⊇Z. Then every irrreducible component of W∩V⁡(fr) has codimension 0 or 1 in W, so it has codimension at most r in X. Therefore, codimX⁡(Z)≤r. □

1.8 Separated Morphisms

Definition 1.8.1.

A topological space X is Hausdorff if Δ⊆X×X is closed.

Remark 1.8.2.

A scheme X is almost never Hausdorff.

Theorem 1.8.3.

In the topological definition, we are taking the product topology on X×X and set-theoretically we have the Cartesian product.

For schemes, we can consider the fiber product X×SX endowed with the Zariski topology as a scheme.

Definition 1.8.4.

Let π:X→S be a morphism. Then π is separated if the morphism δπ:X×SX determined by the identity on each copy of X is a closed immersion.

A scheme X is separated if X→Spec⁡ℤ is separated, and π:X→S is called quasi-separated if δπ is quasi-compact.

Definition 1.8.5.

A morphism f:X→Y is a locally closed immersion if f is a closed immersion into some open subset U⊆Y. Equivalently, f is a homeomorphism onto a locally closed subset of Y and 𝒪Y,f⁡(x)→𝒪X,x is surjective.

Example 1.8.6.

Any morphism π:X=Spec⁡A→Spec⁡B=Y is separated. Here, we note that X×YX=Spec⁡A⊗BA, and the natural morphism A⊗BA→A is surjective.

Proposition 1.8.7.

Let π:X→S be a morphism of schemes. Then δπ:X→X×SX is always a locally closed immersion.

As a consequence, π:X→S is separated if and only if δπ⁢(X)⊆X×SX is closed.

Remark 1.8.8.

Note that if π is separated, then it is quasi-separated.

Proof of Proposition.

Let S=⋃Vβ be a cover by open affines and X=⋃Uα⁢β be a cover by open affines such that f⁡(Uα⁢β)⊆Vβ. Write Vβ=Spec⁡Bβ and Uα⁢β=Spec⁡Aα⁢β. Now we set

Note that X→𝒰 is closed because Spec⁡Aα⁢β→Spec⁡Aα⁢β⊗BβAα⁢β is a closed immersion. □

Proposition 1.8.9.

Affine morphisms are separated because morphisms of affine schemes are separated.

Example 1.8.10.

Being separated is local on the target.

Corollary 1.8.11.

Closed immersions are separated.

Example 1.8.12.

Open immersions are separated.

Lemma 1.8.13.

If π:X→S is separated, then for all open affines U,V⊆X that map to a common affine open subset of S, then U∩V is affine and 𝒪X⁢(U)⊗ℤ𝒪X⁢(V)↠𝒪X⁢(U∩V).

Example 1.8.14.

We have U∩V=U×SV∩δπ⁢(X)=δπ−1⁢(U×SV).

Proof.

We can assume that S=Spec⁡R is affine. Write U=Spec⁡A,V=Spec⁡B. Then we know that U∩V→U×SV=Spec⁡A⊗RB is a closed immersion. Therefore, we have a surjection A⊗ℤB↠A⊗RB↠𝒪X⁢(U∩V). □

Proposition 1.8.15.

Let S=Spec⁡R. Then π:X→S is separated if and only if for all U,V⊆X open affines, U∩V is affine and 𝒪X⁢(U)⊗R𝒪X⁢(V)→𝒪X⁢(U∩V) is surjective. Equivalently, there exists a covering of X by open affines such that the conditions hold.

Corollary 1.8.16.

Let S=Spec⁡R. Then ℙSn→S is separated.

Recall that closed immersions are preserved under base change. This follows from stability of affine morphisms and the fact that the tensor product is right exact.

Proposition 1.8.17.

Being separated and quasi-separated are preserved under base change.

Proof.

Let f:Y′→Y be separated. Let X→Y be a morphism and X′→X be the base change of f. Now consider the Cartesian diagram

Because the bottom arrow is a closed immersion, so is the top arrow. The proof for the quasi-separated case is similar. □

Proposition 1.8.18.

Being separated (or quasi-separated) is closed under composition. If f:X→Y,g:Y→Z are separated, then so is h=g∘f.

Proof.

Consider Xi→fiY→Z for i=1,2. Then we want to show that the diagram

is Cartesian. Note that ε is the morphism X1×YX2⇉Xi→Y and γ is constructed using the universal property for X1×ZX2⇉Xi→Z. Now consider a scheme T with diagram

Now a map φ:T→X1×ZX2 is given by φ=(φ1,φ2), and we know that (f1∘φ,f2∘φ2)=(ψ,ψ). But now the universal property of the fiber product gives us the desired result. □

Exercise 1.8.19.

Open immersions are separated.

Corollary 1.8.20.

Quasi-projective schemes are separated.

Proof of Proposition.

The square in the following diagram is Cartesian

and thus all horizontal arrows are closed immersions. □

Proposition 1.8.21.

Let f:X→Y be quasi-compact and quasi-separated and ℱ be a quasicoherent sheaf on X. Then f∗⁢ℱ is quasicoherent.

Consider a morphism f:X→Y of S-schemes. Then we define the graph morphism Γf:X→X×FY.

Proposition 1.8.22.

Let f:X→Y be a morphism of S-schemes.

  1. 1.

    Γf:X→X×SY is locally closed.

  2. 2.

    If Y→S is separated, then Γf:X×SY is closed.

Proof.

Consider the Cartesian diagram

Because Δ is a (locally) closed immersion, so is Γf. □

Now let f,g⇉Y be two morphisms of S-schemes. We define the equalizer to be the limit of this diagram if it exists. We need a scheme structure on the equalizer, and in the diagram

we see that the equalizer is simply Y×(Y×SY)X⊆X. This is a locally closed subscheme. Proving that this is actually the equalizer is simply and we simply note that (f∘ρ,g∘ρ) factors through Y→ΔY×SY and therefore factors through the fiber product Y×(Y×SY)X.

Example 1.8.23.

Consider the morphisms f,g:𝔸1⇉𝔸1 given by x↦0,x↦x2. Then the diagrams of (schemes, rings) are

and it is easy to see that Eq=Spec⁡k⁡[t]/t2.

Corollary 1.8.24.

Let f,g:X⇉Y be a morphism and X be reduced and Y separated. Suppose there exists a dense open U⊆X such that f|U=g|U. Then f=g everywhere on X.

Proof.

We know U⊆Eq⁡(f,g) is closed because Y is separated, and thus Eq⁡(f,g)=X. □

Example 1.8.25.

The affine line with two origins is not separated! Note that the inclusions of the two copies of 𝔸1 coincide on 𝔸1∖{0}, but globally are different morphisms.

Example 1.8.26.

Consider the maps Spec⁡k⁡[x]/x2⇉Spec⁡k⁡[x]/x2 given by the identity and killing the maximal ideal. These agree set-theoretically but are different as morphisms of schemes.

We return to rational functions on integral schemes of finite type over an algebraically closed field k. More generally, if X is a reduced locally Noetherian scheme, we say that

Definition 1.8.27.

A rational function on X is an equivalence class of (U,f), where U⊆X is dense and open and f∈𝒪X⁢(U). We declare (U,f)∼(U′,f′) if f,f′ agree on U∩U′.

Definition 1.8.28.

We say f is regular at x∈X if there exists a representative (U,f) such that x∈U.

Lemma 1.8.29.

If {(U,f)} is a rational function, there is a maximal open subset of regular points. This is called the domain of definition.

Proof of this is simply the sheaf axioms. Now the set of rational functions on X form a ring. If X=Spec⁡A, this is the total ring of fractions of A. If X is integral, this is k⁡(ηX).

Example 1.8.30.

On X=Spec⁡k⁡[x,y]/x⁢y, we see that 1(x−1)⁢(y−3) is a rational function but 1x⁡(y−3) is not.

Now let X be reduced and Y be a scheme.

Definition 1.8.31.

A rational map f:X⇢Y is an equivalence class of pairs (U,f) where U⊆X is open and dense and f:U→Y is a morphism. We declare (U,f)∼(U′,f′) when there exists V⊆U∩U′ open and dense such that f,f′ agree on V.

In particular, if Y is separated, then we can take V=U∩U′.

Example 1.8.32.

Consider ℙn⇢ℙn−1 given by projection from a point. This sends [x0,…,xn]↦[x1,…,xn] and is defined everywhere except [1,0,…,0].

Example 1.8.33.

Consider the map ℙ2⇢ℙ2 given by [x,y,z]↦[1x,1y,1z]. This is called the Cremona transformation and is defined everywhere besides [1,0,0],[0,1,0],[0,0,1].

Example 1.8.34.

The graph of the rational map ℙ2⇢ℙ1 given by projection from a point is simply the blowup of ℙ2 in a point.

Lemma 1.8.35.

The set of regular points of a rational map is open. If Y is separated, then there exists a morphism f:Ureg→Y representing the rational map.

Definition 1.8.36.

Let f:X⇢Y be a rational map over S with Y separated. Then let U⊆X be the set of regular points. Then the graph of f is the closed subscheme Γf⊆X×SY given by the closure of f|U. In fact, the graph is independent of the dense open subset U chosen.

Definition 1.8.37.

A rational map f:X⇢Y over S is called dominant if there exists a representative (U,f) such that f:U→Y is dominant.

Example 1.8.38.

The map ℙn→𝔸n is dominant (choose a distinguished open subset, then take the identity to 𝔸1).

Definition 1.8.39.

Let X,Y be reduced. A rational map f:X⇢Y is called birational if it is dominant and there exists a rational dominant map g:Y⇢X that is inverse to f as rational maps.

Proposition 1.8.40.

Let X,Y be reduced and f:X⇢Y over S be birational. Then there exist dense open subsets U⊆X,V⊆Y such that f|U:U→V is an isomorphism.

Proof.

Let g:Y⇢X be the inverse. Then we may assume that U,V are affine. Next, if we write Z=Y∖V, then we can replace U by an open affine in U∖f−1⁢(Z). Thus we can assume X,Y are affine. Call U′≔f−1⁢(V), Then U′→𝑓V→𝑔X. This means that g∘f is the inclusion of U′ in V. Now clearly we may replace V with g−1⁢(U′)≕V′, and now f,g are inverse on U′,V′. □

Example 1.8.41.

Let f:X⇢Y be a rational map. Then the projection π:Γf→X is birational, and the maximal open subset where ρ=π−1 is defined the domain of definition of f.

Example 1.8.42.

Consider 𝔸2⇢ℙ1 given by (x,y)↦[x,y]. Then the graph is the blowup Bl0⁡𝔸2 of 𝔸2 at the origin.

1.9 Proper Morphisms

Definition 1.9.1.

A morphism f:X→Y is called proper if f is of finite type, separated, and universally closed (closed and being closed is preserved under base change).

Example 1.9.2.

Closed immersions are proper. Clearly they are finite type (because they are affine of the form A→A/I), separated, and clearly universally closed because closed immersions are closed and stable under base change.

Example 1.9.3.

The map 𝔸1→Spec⁡k is closed, but not universally closed. For example, the map 𝔸2=𝔸1×k𝔸1→𝔸1 is not closed (take the closed subset defined by x⁢y=1).

Example 1.9.4.

Let f:X→Y be an integral morphism. Then f is affine and thus separated. Also, integral morphisms are closed and stable under base change, so f is universally closed. Then f is of finite type if and only if it is finite, so finite morphisms are proper.

Proposition 1.9.5.

Being proper is stable under base change, stable under composition, and local on the target.

Remark 1.9.6.

Let 𝒫 be a property of a morphism of schemes that is stable under base change and composition. Suppose closed immersions satisfy 𝒫. Then for all morphisms f:X→Y, if X→X has 𝒫 and Y→X is separated, then f has 𝒫.

Proof.

Note that f:X→ΓfX×SY→Y. Then X×SY→Y has 𝒫, and Γf is a closed immersion (because Y is separated) and thus has 𝒫, so f has 𝒫. □

Remark 1.9.7.

The property 𝒫 can be taken to be proper, separated, closed immersion, etc.

Proposition 1.9.8 (Image of proper to separated is proper).

Let f:X→Y be a surjective morphism and suppose X is proper and Y is separated of finite type. Then Y is proper.

Proof.

We only need to check that Y→S is universally closed. We show that Y→S is closed. Because f is surjective, then for Z⊆Y, we know Z=f⁡(f−1⁢(Z)), so the image of Z in S is closed. To show that f is universally closed, we simply base change the entire diagram. □

Remark 1.9.9.

We can eliminate the condition that f is surjective by replacing Y with the scheme-theoretic image of f.

Proposition 1.9.10.

Let X be a reduced scheme and f:X→Y be a morphism. Then the scheme theoretic image coincides with the closure of the set-theoretic image.

Proposition 1.9.11.

Let X be a proper connected reduced scheme over a field k. Then k⊆Γ⁡(X,𝒪X) is integral. If k is algebraically closed, then Γ⁡(X,𝒪X)=k.

Proof.

Let f∈Γ⁡(X,𝒪X). Then we can view f:X→𝔸1. This is proper, so f⁡(X) is a closed connected reduced subscheme of 𝔸1. Thus it suffices to show that f⁡(X)≠𝔸1, but this is simply because 𝔸1 is not proper (despite being separated of finite type). Thus f⁡(X) must be a closed point, so the map k⁡[t]→Γ⁡(X,𝒪X) factors through k⁡(f⁡(X)). But then f⁡(X)=V⁡(p⁡(t)) for some irreducible polynomial p, so f also satisfies p. □

Theorem 1.9.12.

Let R be a ring. Then ℙRn→Spec⁡R is proper.

Proof.

We already know ℙn is finite type and separated, so we need to show it is universally closed. Therefore it suffices to prove that ℙℤn is universally closed (because being proper is preserved by base change). Now we need to show that ℙn×ℤX→X is closed for all schemes X. In fact, we can check this locally, so we only need to show that ℙAn→Spec⁡A is closed. □

1.10 Proj construction

We want to think of ℙn as Proj⁡(k⁡[x0,…,xn]) under some definition of Proj. This construction should work for any graded ring R=⨁i≥0Ri. Then there is an irrelevant ideal R+=⨁i>0Ri.

We will define Proj⁡R as a set as the set of all graded prime ideals 𝔭⊆R such that 𝔭⊉R+. Now we want to consider Proj⁡R has a topological space, and we may consider the Zariski topology as in the case of Spec⁡R.

Lemma 1.10.1.

Let 𝔞⊆R be a homogeneous ideal. Then 𝔞 is the intersection of all homogeneous prime ideals containing 𝔞 and V⁡(𝔞)=∅ if and only if 𝔞=R+.

Finally we are ready to define Proj⁡R as a locally ringed space. For any f∈R, define Uf≔(Proj⁡R)f=Proj⁡R∖V⁡(f).

Lemma 1.10.2.

The sets {Uf}f∈Rk,k≥1 form a basis for the topology of Proj⁡R.

Remark 1.10.3.

We have an isomorphism Uf=(Proj⁡R)f≃Spec⁡R(f), where R(f) is the degree zero localization of R at f. Here, the maps are given by

and

Finally we are able to describe the structure sheaf on each open subset. Here, we simply define 𝒪Proj⁡R⁢((Proj⁡R)f)=R(f), and so we need to check the gluing axioms.

  1. 1.

    Let f,gi∈R+ be homogeneous and suppose (Proj⁡R)f=⋃(Proj⁡R)gi. This is the same as f∈∑gi⁢R, so there exists n such that fn=∑ai⁢gi for some ai∈R.

  2. 2.

    If (Proj⁡R)f⊆(Proj⁡R)g, then fn=a⋅g and thus there exists a canonical map R(g)→R(f).

  3. 3.

    We have ⋃(Proj⁡R)gi=Proj⁡R if and only if for all homogeneous primes 𝔭 not containing the irrelevant ideal, there exists gi∉𝔭. This is equivalent to V⁢(∑gi⁢R)=∅, which is equivalent to ∑gi⁢R=R+.

Now we not only obtain the structure sheaf but also a sheaf M~ associated to any graded R-module M=⨁k≥0Mk. In particular, Proj⁡R is a scheme. If R=R0⊕R+, then there is of course a natural continuous map Proj⁡R→Spec⁡R0. Note that we have a commutative diagram

Remark 1.10.4.

If R is finitely generated as an R0-algebra, then π:Proj⁡R→Spec⁡R0 is of finite type.

Proposition 1.10.5.

The map π:Proj⁡R→Spec⁡R0 is separated.

Proof.

Apply the criterion that if U,V⊆X is open affine, then U∩V is affine and 𝒪X⁢(U)⊗R𝒪X⁢(V)→𝒪X⁢(U∩V) is surjective implies that f is separated. Now apply this to distinguished open subsets. □

We would like to define functoriality for this construction. Let R,S be graded rings and φ:S→R be a morphism of graded rings. Can we define a map Proj⁡R→Proj⁡S?

Example 1.10.6.

Let S=k⁡[x1,…,xn]⁢↪𝜑⁢R=k⁡[x0,…,xn]. Then Proj⁡R=ℙn and Proj⁡S=ℙn−1. This is not defined globally because (x1,…,xn)∩S=S+ is the irrelevant ideal. Instead, we consider V⁡(φ⁡(S+))=Z, and then we obtain a morphism ψ:Proj⁡R∖Z→Proj⁡S. In fact, here we obtain projection from a point.

Remark 1.10.7.

ψ is an affine morphism.

Example 1.10.8.

If φ⁡(S+)=R+, then Z≠∅, so we have a morphism Proj⁡R→Proj⁡S.

Example 1.10.9.

For example, if φ is surjective, then ker⁡φ is a homogeneous ideal and in fact we have Proj⁡R=Proj⁡S/ker⁡φ⊆Proj⁡S, and the last inclusion is a closed immersion.

Example 1.10.10.

The morphism k⁡[x0,…,xn]→𝜑k⁡[x,t]=R given by xi↦sn−i⁢ti defines the Veronese embedding.

Example 1.10.11.

Suppose R is finitely generated as an R0-algebra by finitely many elements in degree 1. Then we have a surjection R0⁢[x1,…,xn]↠R, and thus Proj⁡R is a closed subscheme of ℙR0n.

Now we would like to consider what happens under base change. Let R=R0⊕R+ and consider a morphism S0→R0. Then we can consider the scheme Proj⁡R×R0Spec⁡S0. Alternatively, we may consider the graded ring R′=R⊗R0S0=⨁Ri⊗R0S0. Of course we have a morphism Proj⁡R′→Proj⁡R×R0Spec⁡S0, and in fact on open subsets, this defines an isomorphism.

As a corollary, let X be a scheme and ℛ=⨁ℛi be a sheaf of graded algebras with ℛi suasicoherent for all i≥0. Then there exists a scheme Proj𝒪X⁡(ℛ)→X such that over an affine open Spec⁡A where ℛi=Ri, we have ProjA⁡⨁Ri→Spec⁡A.

Example 1.10.12.

If ℱ is a finitely generated quasicoherent sheaf, then we may consider ℛ=⨁Symn⁢ℱ. Then we write ℙX⁡(ℱ)≔Proj𝒪𝒳⁡(ℛ). In particular, if ℱ=𝒪X⊗V, we obtain ℙX⁡(V∨).

Now if R is a graded ring, we can consider the ring R⁡(d)=⨁k≥0Rk⁢d, where Rk⁢d now has degree k. Then we have an inclusion Proj⁡R→Proj⁡R⁡(d). But then if R+⊄𝔭, we know that if 𝔭k⁢d⊇Rk⁢d, we know that Rk⊆𝔭 for all k>0. In particular, we obtain an honest morphism ψ:Proj⁡R→Proj⁡R⁡(d). We can check on principal open subsets that ψ is an isomorphism.

Corollary 1.10.13.

Let R,R′ be rings such that R0=R0′ and there exists N such that

then Proj⁡R≃Proj⁡R′.

Proof.

Replace R,R′ with R⁡(d),R′⁢(d) for sufficiently large d. □

Now let R,S be graded rings with R0=S0. Now of course we have the fiber product

Then if ℝ′=⨁Ri⊗R0Si, we can show that Proj⁡R′=Proj⁡R×R0Proj⁡S.

Theorem 1.10.14.

For any graded ring R, the morphism Proj⁡R→Spec⁡R0 is closed.

This follows from the fact that ℙn→Spec⁡R0 is universally closed and Proj⁡R⊆ℙn is a closed immersion.

Exercise 1.10.15.

Up to taking R⁡(d) for an appropriate d, we can assume that R is finitely generated in degree 1.

Proof.

We want to show that π:ℙAn→Spec⁡A is closed. Let Z=V⁡(I). If 𝔭 is such that ℙk⁡(𝔭)n∩V⁡(I)=∅, then there exists an open subset U∋𝔭 of Spec⁡A such that π−1⁢(U)∩V⁡(I)=∅. If I=(f1,…,fk), then f¯1,…,f¯k have no nontrivial solutions in k⁡(𝔭)⁢[x0,…,xn]. The idea here is to use Nakayama. □

We now want to consider sheaves on X≔Proj⁡R. We want a theory such that 𝒪X=R~. More generally, for any graded R-module M, define the sheaf M~ by

This gives an exact functor, which is not faithful.

Remark 1.10.16.

Compare this to the affine case where we had an equivalence of categories.

Remark 1.10.17.

The grading is important for this construction. We can construct the sheaves (M⁡(d))k=Md+k, and in particular, we have the sheaf R⁡(d)~≕𝒪X⁢(d).

Proposition 1.10.18.

If R is finitely generated in degree 1, the sheaves 𝒪X⁢(d) are locally free of rank 1.

Proof.

We know that X=⋃f∈R1Xf. For f∈R1, we know R⁡(1)~(Xf)=R(1)(f), which consists of elements of the form hfn, where deg⁡h=1+n. This implies that multiplication by f defines an isomorphism R(f)→R⁢(1)(f). □

Remark 1.10.19.

Note that multiplication gives a map R→R⁡(1), which after sheafification gives us a map 𝒪X→𝒪X⁢(1). Combining these, we obtain a map R1→Γ⁡(X,𝒪X⁢(1)). However, we cannot say much about this map in general. All of this can be generalized to Rd→Γ⁡(X,𝒪X⁢(d)).

Remark 1.10.20.

Consider R=A⁡[x0,…,xn] and write Proj⁡R=ℙAn=⋃Ui. We can write 𝒪X⁢(1) in terms of Cech cocycles. On each Ui, we have maps 𝒪X|Ui→xi𝒪X⁢(1)|Ui. On overlaps, the transitions are given by xj/xi, so we obtain 𝒪X⁢(1) by gluing copies of 𝒪X with these gluing functions.

Proposition 1.10.21.
  1. 1.

    For all graded modules M and d∈ℤ, M⁡(d)~≅M~⊗𝒪X𝒪X⁢(d).

  2. 2.

    For all m,n, 𝒪X⁢(n)⊗𝒪X𝒪X⁢(m)≃𝒪X⁢(n+m). In particualr, when m=−n, we have 𝒪X⁢(n)∨=𝒪X⁢(−n).

This is proved by considering the maps M⊗RR⁡(1)→M⁡(1). Similarly to above, we also have maps Md→Γ⁡(X,M⁡(d)~) for all d∈ℤ.

Definition 1.10.22.

For any sheaf ℱ of 𝒪X-modules, define the graded module

which is a graded Γ∗⁢(𝒪X)-module. For a graded module M, we obtain a module Γ∗⁢(M~).

Proposition 1.10.23.

If R=A⁡[x0,…,xn], then R=Γ∗⁢(R~). Therefore, for all d≥0, we have Rd=A⁢[x0,…,xn]d≃Γ⁡(ℙAn,𝒪⁡(d)).

Proposition 1.10.24.

Let ℱ be a quasicoherent sheaf on X=Proj⁡R. Then there exists a natural isomorphism ℱ=Γ∗⁢(ℱ)~.

Corollary 1.10.25.

Any closed subscheme of X=Proj⁡R is defined by some graded ideal I⊆R.

Remark 1.10.26.

If we replace R=R′=⨁i≥0Ri⁢d, we know X=Proj⁡R→𝜑∼Proj⁡R′=Y, and φ∗⁢𝒪⁢(1)=𝒪⁡(d).

Remark 1.10.27.

Let R→S be a morphism of graded algebras of degree 0. Then we have a map Proj⁡S⊇V→𝜑Proj⁡S, and φ∗⁢(𝒪⁡(1))=𝒪⁡(1).

We may also define global versions of this on an arbitrary scheme S. Let ℛ=⨁ℛi be a graded 𝒪S-algebra. We assume that ℛ is generated as an ℛ0-algebra by ℛ1. This gives us a morphism ProjS⁡(ℛ)≕X→𝜋S. Also, we have a natural surjection S∗⁢ℛ1↠ℛ, so we have an embedding X↪ℙ⁢ℛ1. Now if ℳ is a quasicoherent graded ℛ-module, we obtain a module ℳ~ on X. If ℳ=ℛ⁡(d), we have ℛ⁡(d)~≕𝒪π⁢(d).

Proposition 1.10.28.

If ℰ is locally free over S, then ℙ⁢ℰ=ProjS⁡(S∗⁢ℰ)→𝜋, and π∗⁢𝒪ℙ⁢ℰ⁢(d)=Sd⁢ℰ.

Remark 1.10.29.

If ℒ is a line bundle, then ℙ⁢ℰ=ℙ⁡(ℰ⊗ℒ).

Now recall that Hom𝖲𝖼𝗁⁡(X,ℙℤn) is in bijection with the set of surjections 𝒪Xn+1↠ℒ for line bundles ℒ. Now if we consider Proj⁡R, where R is finitely generated in degree 1, then we obtain a map 𝒪ℙn⊗R1↠𝒪ℙn⁢(1), so we obtain a map R1→Γ⁡(X,ℒ). Now we want to consider HomS=Spec⁡R0⁡(X,Proj⁡R) for arbitrary R.

Theorem 1.10.30.

The set HomS⁡(X,Proj⁡S∗⁢R1) is in bijection with the set of invertible sheaves ℒ on X equipped with a map R1→𝜑Γ⁡(X,ℒ) which globally generate ℒ. The bijection is given by ℒ=f∗⁢(𝒪⁡(1)) and R1=Γ⁡(𝒪⁡(1))→Γ⁡(X,f∗⁢(𝒪⁡(1))).

Proof.

Given a line bundle ℒ on X and φ:R1→Γ⁡(X,ℒ) which globally generates ℒ, write R1∋f↦sf∈Γ⁡(X,ℒ). Then we know X=⋃Xsf. Locally, we have the section 𝒪X→𝑠ℒ, so ℒ|Xs=𝒪X|Xs. To define f:X→Proj⁡R≕ℙ, we define it locally. In fact, we give morphism Xsf→ℙf=Spec⁡R(f). This is equivalent to giving a morphism R(f)→Γ⁡(Xsf,𝒪X). But this is simply given on R1 by f↦1, and so the map S∗⁢(R1)→Γ⁡(Xsf,𝒪X) factors through (S∗⁢R1)f, and thus we have a map (S∗⁢R1)(f)→Γ⁡(Xsf,𝒪Xsf). □

Remark 1.10.31.

The map f is uniquely determined by ℒ and φ, and thus gluing is given by this uniqueness.

Remark 1.10.32.

Given a surjection S∗⁢R1↠R which induces Proj⁡R↪Proj⁡S∗⁢R1, a map X→Proj⁡(S∗⁢R1) factors through Proj⁡R if and only if the map R1→Γ⁡(X,ℒ) satisfies the property that S∗⁢R1→Γ∗⁢ℒ factors through R.

Theorem 1.10.33.

Let f:X→S be a morphism of schemes and ℛ=S∗⁢ℛ1 be a quasicoherent graded 𝒪S-algebra. Then HomS⁡(X,ProjS⁡ℛ) is in bijection with the set of line bundles ℒ on X equipped with a surjection φ:f∗⁢ℛ1↠ℒ.

Corollary 1.10.34.

Let ℰ→X be locally free. Then sections of ℙ⁢ℰ→X are in bijection with surjections ℰ→ℒ, where ℒ is a line bundle on X.

Example 1.10.35.

Let X=ℙAn and f:𝒪X↪𝒪X⁢(1) for f∈A⁢[x0,…,xn]1. If we tensor with 𝒪X⁢(−1), we obtain a map f:𝒪X⁢(−1)↪𝒪X. This realizes 𝒪X⁢(−1) as an ideal sheaf of 𝒪X associated to f. Of course, the scheme associated to this ideal sheaf is V⁡(f).

Example 1.10.36.

Let X be locally factorial and Z⊆X be a codimension 1 subscheme. Then the ideal sheaf ℐZ is locally principal.

Now let X be a scheme, ℒ be a line bundle, and s0,…,sn∈Γ⁡(X,ℒ). This defines a rational map X⇢ℙn. We would like to put a scheme structure on X∖U≕Z.

Definition 1.10.37.

The base locus Bs⁡(s0,…,sn)⊆X is the closed subscheme defined by the following ideal sheaf. Write V≔⟨s0,…,sn⟩⊆Γ⁡(X,ℒ). Then we obtain a morphism V⊗𝒪S→evℒ. Tensoring with ℒ−1, we obtain a map V⊗ℒ−1→𝒪X. Now we define the ideal sheaf ℐV⊆𝒪X to be the image of this morphism.

We would like to now modify X such that we can define an actual morphism X→ℙn. Before we do the construction, we give some examples.

Example 1.10.38.

Suppose that the ideal sheaf I is invertible and locally principal. If ℒ=𝒪ℙn⁢(1) and s0=x1, then the base locus is V⁡(x1). Up to passing from ℒ to ℒ⊗ℐ, we may assume that the line bundle is globally generated by V, so our map extends.

We are now ready to consider blowups of closed subschemes Y⊆X. Here, X is a scheme and Y⊆X to be a closed subscheme with ideal sheaf ℐ. Write ℛ=⨁n≥0𝒥n

Definition 1.10.39.

The blowup of X along Y is the scheme ProjX⁡(ℛ)→𝜋X.

Note that π is proper. Now note that if f:X→Y, and ℐ⊆𝒪Y is an ideal sheaf, the map f∗⁢ℐ→f∗⁢𝒪Y=𝒪X is not injective, but we can consider the image, which is an ideal sheaf that we will call f−1⁢ℐ.

Example 1.10.40.

Consider the map 𝔸2→𝔸2 given by k⁡[s,t]→k⁡[x,y] given by (s,t)↦(x,x⁢y). Then

is not injective, because f−1⁢(s,t)=(x).

Remark 1.10.41.

If ℐ is quasicoherent, so it f−1⁢ℐ.

Remark 1.10.42.

If X→𝑓Y and ℐZ⊆𝒪Y is the ideal of a closed subscheme Z, then f−1⁢(Z) has ideal f−1⁢ℐZ.

Example 1.10.43.

We will consider the blowup of 0∈𝔸k2. Here, if A≔k⁡[x,y], then the blowup X⊆ℙ𝔸21 is defined by (x⁢T=y⁢S), where ℙ𝔸21=Proj⁡A⁡[S,T]. Now we want to check what f−1⁢(x,y) is, and we can do this locally. If we consider the chart US=Spec⁡k⁡[x,y,t], then X∩US is given by t⁢x=y. Thus f−1⁢(x,y)|US=(x). On the other hand, on the chart UT, we have the equation x=y⁢s, so f−1⁢(x,y)|UT=(y). In fact, X=Bl0⁡𝔸k2. To see this, we know that

Theorem 1.10.44.

Let X,Y,ℐ be as above. Then write BlY⁡X≕X~→𝜋X and set U=X∖Y.

  1. 1.

    π induces an isomorphism π:π−1⁢(U)→U.

  2. 2.

    The sheaf f−1⁢ℐ⊆𝒪X~ is invertible and corresponds to 𝒪X~⁢(1).

Proof.

On X∖Y=U, ℐ is trivial, so ℛ|U=⨁𝒪U=𝒪U⁢[T], so we are done. This proves the first part. For the second part, note 𝒥⋅ℛ=⨁n≥1𝒥n=ℛ⁡(1). □

Remark 1.10.45.

If ℐ is locally principal (thus locally trivial), then in fact BlY⁡X=X.

Remark 1.10.46.

When we have ℙAn=Proj⁡A⁡[x0,…,xn], 𝒪⁡(1) has sections A⁢[x0,…,xn]1, and here 𝒪⁡(−1)⊆𝒪 is an ideal sheaf. Here something was said about self intersections of exceptional divisors on surfaces (if you blow up a smooth point, you get a curve with self intersection −1).

We know that f−1⁢ℐ is a locally principal ideal sheaf on X~. If E⊆X~ is the subscheme defined by f−1⁢ℐ, then f−1⁢(Y)=E. Because ℐE is locally principal, then E|U=(fu) is a closed subscheme of codimension at most 1 for any open subset U⊆X~, and the codimension is 1 if fu is not a zero divisor. Later, we will see that E has pure codimension 1 and is a Cartier divisor, called the exceptional divisor.

We will check this affine locally on X. We may assume that X=Spec⁡A and R=⨁In. Then let (x1,…,xr)=I, so X~=⋃Uxi. Now we have a map

and we always have the relations xi⁢Tj=xj⁢Ti for all i,j. On Uxi=Spec⁡R(xi), we still have a morphism A⁢[T1,…,Tr](Ti)↠R(xi). Now we consider I⁢R(xi) and note that

Because xj=xi⁢φ⁢(TjTi), it follows that I⁢R(xi) is generated by xi and is thus principal. Now we need to show that xi is not a zero divisor, but this is clear by the localization process.

Proposition 1.10.47.

If X is integral, so is X~. Also, if X→S is separated or proper, so is X~→S. Finally, if X→S if X is Noetherian, then ℐ is coherent, and X~ is also Noetherian.

Returning to extending morphisms, let X be integral and ℒ be an invertible sheaf. Then choose s0,…,sn∈Γ⁡(X,ℒ) and write V=⟨s0,…,sn⟩⊆Γ⁡(X,ℒ). Then let Y be the base locus and ℐY=[V⊗ℒ−1↠ℐY⊆𝒪X] be the ideal sheaf. Then we would like to extend the morphism U≔X∖Y→ℙn. Last time, we saw that we could extend the morphism if ℐY was an invertible sheaf. Now let X~=BlY⁡X and π:X~→X. But now we have a surjection V⊗π∗⁢ℒ−1↠π−1⁢ℐY onto an invertible sheaf, so we obtain a surjection V⊗𝒪X~↠π−1⁢ℐY⊗π∗⁢ℒ, and so now we obtain a regular morphism X~→ℙSn. Note that we also write π−1⁢ℐY=𝒪X~⁢(−E).

Exercise 1.10.48.

We have an identification of X~ with the graph of X⇢ℙn.

Theorem 1.10.49 (Universal property of blowups).

Let X,Y,ℐ,X~→𝜋X be as before. Then for all f:Z→X such that f−1⁢ℐY⊆𝒪Z is an invertible sheaf, then there exists a unique g:Z→BlY⁡X making the diagram

commute.

Proof.

We use the characterization of morphisms to Proj⁡R. Here, a map X→Proj⁡S∗⁢ℛ1 was an invertible sheaf ℒ on X and a surjection 𝒽∗⁢ℛ1→ℒ. If ℛ is generated by ℛ1, then we need this surjection to factor through 𝒽∗⁢ℛ.

Now set ℒ=f−1⁢ℐY. By definition, there exists a surjective morphism f∗⁢ℐY→ℒ, so of course we obtain a morphism

This gives us the morphism Z→X~=Proj⁡ℛ→X. The proof of uniqueness is omitted. □

Corollary 1.10.50.

Consider π:BlY⁡X≕X~→X. Then for all f:Z→X, let Z~→Z be the blowup of Z along f−1⁢ℐY, then there exists f~:Z~→X~ lifting f.

Remark 1.10.51.

If f is a closed embedding, then so is f~. When this is the case, then Z~ is called the proper transform of Z in X~.

Corollary 1.10.52.

If Z is integral and Z⊆X is a closed embedding, then Z~=π−1⁢(X∖Y)∩Z¯⊆X~.

Exercise 1.10.53.

Consider proper transforms of nodal cubic curves in Bl0⁡𝔸2.

Now let E be the exceptional divisor of π−1⁢(Y) under the blowup π:X~→X. Then we know

For example, the exceptional divisor of Bl0⁡𝔸2→𝔸2 is

More generally, the blowup Bl0⁡𝔸n has exceptional divisor ℙn−1.

Now let A be a Noetherian ring and I⊆A be an ideal. Suppose I is generated by a regular sequence of length r.

Lemma 1.10.54.

If I is generated by a regular sequence of length r, then I/I2 is a free A/I-module of rank r, and

Definition 1.10.55.

Let X be a Noetherian scheme and Y⊆X be a closed subscheme. Then Y is called locally complete intersection of codimension r if ℐY is locally generated by a regular sequence of length r.

Example 1.10.56.

Let X=Spec⁡A where A is a unique factorization domain. Then Y⊆X of codimension 1 is a locally complete intersection.

Example 1.10.57.

If E⊆X~ is the exceptional divisor of a blowup, then E is a locally compelte intersection.

Example 1.10.58.

The twisted cubic is locally a complete intersection.

Proposition 1.10.59.

Let X be Noetherian and Y⊆X be locally complete intersection of codimension r. Then

  • •

    ℐ/ℐ2 is a locally free coherent sheaf of rank r;

  • •

    The map E=ℝ⁢ℐ/ℐ2→Y is a ℙr−1-bundle.

Remark 1.10.60.

Locally on X~⊆ℙXn this is defined by fi⁢Tj=fj⁢Ti, where I=(f1,…,fr).

1.11 Projective morphisms and (very) ample line bundles

Definition 1.11.1.

A morphism f:X→Y is called projective if X≃ProjY⁡ℛ for some ℛ=⨁ℛi, where ℛ is a quasicoherent graded algebra, ℛ0=𝒪Y, and ℛ is finitely generated in degree 1, so ℛ1 is locally of finite type.

This is equivalent to the existence of a factorization f:X↪ℙ⁢ℱ→Y, where ℱ is a quasicoherent sheaf locally of finite type. If either of these conditions are satisfied, X is said to be projective over Y.

Remark 1.11.2.

If f:X→Y is projective, then on X we have a line bundle ℒ=𝒪⁡(1)=ℛ⁡(1)~. Conversely, if there exists ℒ and a quasicoherent sheaf ℛ1 of finite type with f∗⁢ℛ1↠ℒ such that f factors as X↪ProjY⁡S∗⁢ℛ1→Y, then f is projective.

Example 1.11.3.

Blowups are projective if X is Noetherian and ℐ is coherent.

Example 1.11.4.

Clearly closed embeddings are projective.

Remark 1.11.5.

Projective morphisms are proper and stable under base change.

Warning 1.11.6.

Hartshorne defines projective morphisms as being X↪ℙYn→Y. This is strictly stronger than our definition.

Definition 1.11.7.

If Y is quasicompact, then f:X→Y is called quasiprojective if it can be factored as a quasicompact embedding followed by a projective morphism.

Remark 1.11.8.

Projective morphisms are separated and proper. Also, there exists some invertible sheaf ℒ=i∗⁢𝒪ℙ⁢ℱ⁢(1).

Now we want consitions for an invertible sheaf ℒ on X to define an immersion X→ℙ⁢ℱ.

Definition 1.11.9.

Let f:X→Y be a morphism of finite type with Y Noetherian. Then ℒ is called very ample over Y if there exists a coherent sheaf ℱ on Y such that f factors via an immersion as f=X↪ℙ⁢ℱ→Y such that ℒ=i∗⁢𝒪ℙ⁢ℱ⁢(1).

Remark 1.11.10.

If Z⊂X is a closed subscheme and ℒ is very ample, then so is i∗⁢ℒ.

Example 1.11.11.

Let f be as above. Then 𝒪ℙ⁢ℱ⁢(1) is very ample over Y.

Remark 1.11.12.

This is a relative notions. If X=Blx⁡ℙ2, then 𝒪X⁢(1) is very ample over ℙ2 but not ample over Spec⁡k (?).4

Remark 1.11.13.

If f is proper, then the morphism X→ℙ⁢ℱ is necessarily a closed immersion.

Remark 1.11.14.

Vakil defines the notion of being very ample only for proper morphisms. There is yet another different definition in Görtz-Wedhorn, but I cannot be bothered to copy it.

Remark 1.11.15.

By definition, if there exists a very ample ℒ, then f is separated.

Definition 1.11.16.

Let X be Noetherian and ℒ be an invertible sheaf on X. Then ℒ is ample if there exists n0 such that ℱ⊗ℒn is globally generated for all n≥n0 and all coherent ℱ.

Example 1.11.17.

If X=Spec⁡A is affine, then any invertible sheaf is ample.

Proposition 1.11.18.

Let X be a quasicompact and quasiseparated scheme, ℒ be a line bundle, ℱ quasicoherent, and f∈Γ⁡(X,ℒ).

  1. 1.

    Let s∈Γ⁡(X,ℱ) such that s|Xf=0. Then there exists n>0 such that fn⁢s=0 in Γ⁡(X,ℱ⊗ℒn).

  2. 2.

    Let t∈Γ⁡(Xf,ℱ). Then there exists n>0 such that fn⁢t lifts to a section of ℱ⊗ℒn.

Theorem 1.11.19 (Serre).

Let ℱ be quasicoherent on X=Proj⁡R, where R is finitely generated in degree 1. Then there exists n0>0 such that for all n≥n0, ℱ⁡(n) is generated by a finite number of global sections.

Proof.

By assumption, Proj⁡R is quasicompact and quasiseparated. Also. Now ℱ=M~i on Xfi, and these are all finitely generated. There are also finitely many fi, so we can choose generators and lift to ℱ⊗𝒪⁡(n). □

Remark 1.11.20.

The sheaf 𝒪Proj⁡R⁢(1) is ample.

Lemma 1.11.21.

The following are equivalent for a Noetherian scheme X:

  1. 1.

    ℒ is ample.

  2. 2.

    There exists n>0 such that ℒn is ample.

  3. 3.

    ℒn is ample for all n>0.

This means that ample line bundles form a cone. They are invariant under passing to positive powers. Now here is a useful fact. Let R be as above and assume R0 is Noetherian. Then for all ℱ quasicoherent on X, Γ∗⁢ℱ is finitely presented over R0.

Corollary 1.11.22.

Let f:X→Y be a projective morphism with Y Noetherian. then for all ℱ coherent on X, f∗⁢ℱ is coherent on Y.

In general the pushforward of a coherent sheaf is not coherent (consider open immersions).

Proposition 1.11.23 (Coherent extension).

Let X be Noetherian and ℱ be quasicoherent on X. Suppose U⊆X is open and 𝒢U⊂ℱ|U is a coherent subsheaf. Then there exists a coherent subsheaf 𝒢⊂ℱ such that 𝒢|U=GU.

Corollary 1.11.24.

Let X be Noetherian, U⊆X be open and ℒ be ample on X. Then ℒ|U is ample.

Proof.

For all ℱ coherent on U, j∗⁢ℱ is quasicoherent on X. Then use coherent extension to construct an extension ℱ¯ on X extending ℱ. If ℱ¯⊗ℒn is globally generated, so is ℱ⊗ℒn|U. □

Proof of proposition.

Consider the partially ordered set {(𝒢W,W)} of coherent extensions of W ordered in the obvious way. Then by Zorn, there exists a maximal element, and after reducing to the affine case, we can see that the maximal element is defined on all of X. □

Proposition 1.11.25.

Let X be Noetherian (or qcqs) and ℒ be an invertible sheaf. Then the following are equivalent:

  1. 1.

    ℒ is ample.

  2. 2.

    For all coherent ideals ℐ, there exists n0>0 such that ℐ⊗ℒn is globally generated for all n≥n0.

  3. 3.

    The open subsets of the form Xf,f∈Γ⁡(X,ℒn) for some n>0 form a basis for the topology of X.

  4. 4.

    There exists n0 and f1,…,fN∈Γ⁡(X,ℒn0) such that Xfi are affine and X=⋃Xfi.

Proof.
1 implies 2:

This is by definition.

2 implies 3:

Let U⊆X be open, x∈U be a closed point, and Y=X∖U have the reduced structure. Finally let ℐY⊆𝒪X be the ideal sheaf of Y. Then there exists n such that ℐY⊗ℒn is globally generated. Thus if f∈Γ⁡(X,ℐY⊗ℒn), so Xf⊆U.

3 implies 4:

By quasicompactness, there exists n such that f1,…,fn∈Γ⁡(X,ℒn) are such that X=⋃Xfi. Then 𝒳fi⊆Ui are affine and ℒ|Ui=𝒪Ui. Thus Xfi is a principal open subset of an affine, so it is affine.

4 implies 1:

This is the exact same proof as Serre’s theorem. □

Proposition 1.11.26.

Let X be quasicompact and quasiseparated and ℒ be ample. Then write R=⨁Γ⁡(X,ℒn). Then X→Spec⁡Γ⁡(X,𝒪X) factors through the open immersion X↪Proj⁡R.

Theorem 1.11.27.

Let f:X→Spec⁡A be of finite type with A Noetherian and ℒ be an invertible sheaf on X. Then ℒ is ample if and only if there exists n≥0 such that ℒn is very ample. Moreover, if this is the case, then the immersion X↪ℙ⁢ℱ can be taken to be X↪ℙAN.

Proof.

Assume ℒn is very ample. Then there exists ℱ coherent and also 𝒪XN+1↠ℱ, so we obtain an immersion j:X↪ℙAN. But now if ℱ is coherent on X, we can apply coherent extension to find a coherent subsheaf ℱ¯⊂j∗⁢ℱ on X¯. Finally i∗⁢ℱ¯ is coherent on ℙN, so ℒ is ample.

Conversely, if ℒ is ample, then there exist s1,…,sr∈Γ⁡(X,ℒn) such that X=⋃Xsi with Xsi=Spec⁡Bi affine. Also note that Bi is a finitely generated A-algebra. Now let bi⁢j be generators of Bi over A. Then there exists N such that ti⁢j=siN⁢bi⁢j lift to Γ⁡(X,ℒN). Now we have finitely many sections siN,ti⁢j∈Γ⁡(X,ℒn), and these define a regular morphism ψ:X→ℙm. On each Xsi→(ℙm)Ti, we see that the map A⁢[T1,…,Tr](Ti)→Bi is surjective. Thus ψ is a closed immersion into an open subset of ℙn. □

Remark 1.11.28.

If X is quasicompact and quasiseparated, i:Z→X is a quasicompact immersion, then the pullback of an ample line bundle is ample.

Theorem 1.11.29.

Let X be proper over an algebraically closed field k Now let s0,…,sn∈Γ⁡(X,ℒ) and set V=⟨s0,…,sn⟩. Then V defines a closed immersion if and only if

  1. 1.

    V separates closed points. This means for all x≠x′ there exists s∈V such that s⁡(x)=0,s⁡(x′)≠0.

  2. 2.

    V separates tangent directions. This means that for a closed point x and a tangent vector t∈Tx⁢X, there exists s∈V such that x∈Supp⁡s but t∉Tx⁢V⁢(s).5

Theorem 1.11.30.

Let f:X→Y be a quasiseparated morphism of finite type with X,Y Noetherian. The following are equivalent:

  1. 1.

    ℒn is very ample for some n>0.

  2. 2.

    There exists an affine open covering {Vi} of Y such that ℒ|XVi is very ample for all i.

  3. 3.

    For all V⊆Y affine open, ℒ|XV is ample.

Proof.
1 implies 3:

Note that if ℒn is very ample, then ℒn|XV is very ample, so ℒ|XV is ample.

2 implies 1:

We have a morphism ψi:XVi⁢ℙViNi such that ψi∗⁢𝒪⁢(1)=ℒni|XVi. Up to passing to Veronese embeddings, we may assume ni=n. Then we have a surjection f∗⁢𝒪ViNi+1↠ℒn|Vi. By the adjoint, we obtain a morphism 𝒪ViNi+1→f∗⁢ℒn|Vi. Now the image of this morphism is coherent, so let 𝒢i be a coherent extension of the ℱi to Y. Then we obtain a morphism ⨁𝒢i→f∗⁢ℒn, which gives us f∗⁢⨁𝒢i→ℒn. This is surjective because it is surjective on each XVi. Therefore we obtain a morphism ψ:X↪ℙ⁢⨁𝒢i, and we need to show this is an immersion. For each i, we have an immersion XVi→ℙY⁡(𝒢i). Now up to passing to an open subset, ψ factors through ℙ⁢⨁𝒢i, and so we need to show that if g∘f is an immersion, so is f. This is left as an exercise. □

1.12 Cartier and Weil divisors

Let ℒ be an invertible sheaf. Then there exists an open cover {Ui} of X such that ℒ|Ui→φi𝒪Ui, and on Ui∩Uj we have an isomorphism 𝒪Ui∩Uj→𝒪Ui∩Uj. Of course, this φi⁢j∈𝒪X⁢(Ui∩Uj)×. On Ui∩Uj∩Uk, we have φi⁢j⁢φj⁢k=φi⁢k. This is a Čech 1-cocycle. Then inside the set Z1⁢(𝒰,𝒪X×) we have a subgroup of boundaries B1⁢(𝒰,𝒪X×).

If ℒ=ℳ is an isomorphism of invertible sheaves with cocycles φi⁢j,ψi⁢j, there exist fi∈𝒪X⁢(Ui)× such that φi⁢j=ψi⁢j⁢fi⁢fj−1. Therefore, we can define the Picard group of a scheme to be the subgroup of isomorphism classes of line bundles. This is isomorphic to the cohomology H1⁢(X,𝒪X×).

Now consider sections s∈Γ⁡(X,ℒ). This is the same as a morphism s:𝒪X→ℒ, and if X is integral, this is always injective if s≠0. Dualizing, we have an injection ℒ∨⊆𝒪X, so we realize ℒ∨ as an ideal sheaf. Every section may determine a different embedding ℒ∨⊂𝒪X, and thus a different closed subscheme of X.

Example 1.12.1.

Consider 𝒪⁡(d) on X=ℙn. Then s∈Γ⁡(X,𝒪⁡(d))=k⁢[x0,…,xn]d, and we obtain the exact sequence

where H is a degree d hypersurface.

Remark 1.12.2.

Let X be proper over a field k and ℒ is very ample. Then if X↪ℙn is the embedding given by ℒ, then s∈Γ⁡(X,ℒ) corresponds to a section s∈Γ⁡(ℙn,𝒪⁡(1)), and V⁡(s)=X∩H, where H is the hyperplane cut out by s.

Now let X be a scheme. For an open U⊆X, consider the set S of elements of Γ⁡(U,𝒪X) that are not zero divisors and now write K⁡(U)=S−1⁢Γ⁢(U,𝒪X). Now we define a presheaf U↦K⁡(U), and after sheafifying we obtain sheaves 𝒦⁡(X),𝒦⁢(X)×.

Example 1.12.3.

When X is integral, this is a constant sheaf.

Of course, we have a injection 𝒪X×→𝒦X×, and now we have an exact sequence of sheaves

Example 1.12.4.

A Cartier divisor is an element D∈Γ⁡(X,𝒦X×/𝒪X×).

Concretely, we obtain a collection of compatible pairs (U,f), where U⊂X is open and f∈Γ⁡(U,𝒦X×) such that (U,f)∼(U′,f′) if f′⁢f−1∈𝒪X⁢(U∩U′)×. Now to a Cartier divisor D, we can consider Supp⁡(D)={x∣Dx≠1}={x∣fx∉𝒪X,x×}. Now if D={(Ui,fi)},{(Ui,gi)}, we denote D±E the Cartier divisor defined by {(Ui,fi⁢gi±1)}.

Definition 1.12.5.

A Cartier divisor is called principal if it is in the image of Γ⁡(X,ℒX×) given by D=(X,f). Here, we write D=(f). Two Cartier divisors are called linearly equivalent if D−E=(f) is principal.

Definition 1.12.6.

A Cartier divisor D is called effective if D={(Ui,fi)}, where fi∈𝒪X⁢(Ui)∩𝒦X×⁢(Ui). This is a regular function that is not a zero divisor, and so an effective Cartier divisor defines an ideal sheaf ℐD⊆𝒪X. Clearly this is locally free of rank 1.

Therefore, we see that effective Cartier divisors are the same as invertible ideal sheaves. Also, we will write D≥E if and only if D−E≥0. Now to a Cartier divisor D we will associate the sheaf 𝒪X⁢(±D) locally defined by fi∓1. Also, we see that 𝒪X⁢(−D)⊆𝒪X if and only if D≥X, so we have a group homomorphism from Cartier divisors to the Picard group Pic⁡X.

Proposition 1.12.7.

This assignment factors through the group of Cartier divisors modulo principal divisors and in fact, the kernel is precisely the set of principal divisors. If X is integral, the assignment is surjective.

Proof.

If D=(f), where f∈Γ⁡(X,𝒦X×), f determines an isomorphism 𝒪X→𝒪X⁢((f)). On the other hand, we need to show that if 𝒪X⁢(D)≃𝒪X, then D can be represented by (U,1). Finally, if X is integral, then the sheaf 𝒦⁡(X) is constant, so if ℒ corresponds to {φi⁢j}, then we fix j and think of fi≔φi⁢j∈Γ⁡(Ui,𝒦X×), and then everything will glue. □

Proposition 1.12.8.

If X is integral and ℒ is invertible, then effective Cartier divisors Cartier divisors D such that 𝒪X⁢(D)=ℒ correspond exactly to nonzero global sections of ℒ modulo invertible functions Γ⁡(X,𝒪X×). In particular, if X is proper over an algebraically closed field k, then dimkΓ⁡(X,ℒ)<∞.

Proof.

Suppose D≥0. Then 𝒪X⁢(−D)↪𝒪X, and tensoring with ℒ, we obtain a section. Conversely, given a section, we obtain an invertible ideal sheaf ℒ∨⊂𝒪X, and a Cartier divisor D. □

Now we will consider Weil divisors. Let X be an integral Noetherian scheme. Now a Weil divisor is a finite sum ∑ai⁢Di, where ai∈ℤ and Di is an integral subscheme of codimension 1. We need to assume that X is regular in codimension 1. Note that a local Noetherian ring of dimension 1 is regular if and only if it is a discrete valuation ring.

Definition 1.12.9.

Let f∈K×⁢(X) and W⊆X be a prime divisor. Then 𝒪X,W is a discrete valuation ring, and thus we obtain a valuation vW:K⁡(X)→ℤ. Then f is said to have a pole or zero along W if vW⁢(f) is negative (or positive). Now we define a function Γ⁡(X,𝒦X×)=K×⁢(X)→Z1⁢(X) to the group of Weil divisors by f↦∑vW⁢(f)≕(f).

Definition 1.12.10.

Given f, there exist finitely many prime divisors W such that vW⁢(f)≠0.

Definition 1.12.11.

A Weil divisor D=∑ai⁢Wi is called principal if there exists f∈K×⁢(X) such that D=(f).

In fact, we can define a map Γ⁡(X,𝒦X×/𝒪X×)→Z1⁢(X) sending principal Cartier divisors to principal Weil divisors. This induces an injection Pic⁡(X)↪Z1⁢(X)/K×⁢(X). If X is locally factorial, this is an isomorphism.

Example 1.12.12.

Consider ℙn. Then it is easy to see that Pic⁡(ℙn)=ℤ (if we think of everything as a Weil divisor ∑ni⁢Yi, then equivalence classes are classified by ∑ni).

2 Cohomology

2.1 Derived Functors

Let 𝒜,ℬ be abelian categories and F:𝒜→ℬ be a left exact functor. For example, if X is a topological space, we can have 𝒜 be the category of modules, ℬ=𝖠𝖻, and F=Γ⁡(X,−). This is left exact but not exact in general.

Example 2.1.1.

Let A be a ring and 𝒜=𝖬𝗈𝖽A. Then HomA⁡(M,−) is left exact.

Example 2.1.2.

Let X be a scheme and p≠q be closed points. Then we have the exact sequence

If X is proper over an algebraically closed field k, then Γ⁡(X,𝒪X)→k⊕k cannot be surjective.

Example 2.1.3.

If X=Spec⁡A is affine, then Γ⁡(X,−) is exact on 𝖰𝖼𝗈𝗁⁡(X).

Now we will construct a sequence of functors Ri⁢F for i≥0 such that R0⁢F=F. In some semse, this will measure the failure of exactness of F. To do this, we will replace 𝒜 with the derived category D⁡(𝒜) of complexes localized at quasi-isomorphisms. I have discussed chain homotopies and quasi-isomorphisms in my notes for several other courses,1 so I will omit the discussion here. In order to construct derived functors, we need to replace A∈𝒜 with a quasi-isomorphic complex I∙ of injective objects.

Definition 2.1.4.

An object I∈𝒜 is called injective if Hom𝒜⁡(−,I) is exact.

Example 2.1.5.

In 𝖵𝖾𝖼𝗍k, every object is injective.

Exercise 2.1.6.

If 0→A1→A2→A3→0 is a short exact sequence, then if A is injective, the sequence is split.

Definition 2.1.7.

An abelian category 𝒜 has enough injectives if for all A∈𝒜, there exists an injective object I and injection A↪I.

An injective resolution of A∈𝒜 a complex A→I∙ that is a long exact sequence. Clearly, 𝒜 has enough injectives if and only if injective resolutions always exist.

Lemma 2.1.8.

For A∈𝒜, any two injective resolutions are quasi-isomorphic.

If A is a ring, then 𝖬𝗈𝖽A has enough injectives.

Corollary 2.1.9.
  1. 1.

    Let (X,𝒪X) be a ringed space. Then 𝖬𝗈𝖽⁡(X) has enough injectives.

  2. 2.

    Let X be a topological space. Then 𝖠𝖻⁡(X) has enough injectives.

Proof.

For all x∈X, we have an injective module Ix with ℱx↪Ix. But then we obtain an injection

The target is injective, so we are done. In the second case, simply take the ringed space (X,ℤ). □

Now we may define the right derived functors of F.

Definition 2.1.10.

The right derived functors of F are the functors

Alternatively, we may consider the complex R⁢F⁢(A)=F⁡(I∙), which is well-defined up to quasi-isomorphism.

Remark 2.1.11.

Because F is left-exact, we H0⁢(F⁡(I))=R0⁢F⁢(A)=F⁡(A) as desired.

The crucial observation is that this definition does not depend on the choice of injective resolution.

Example 2.1.12.

Let X be a topological space and F=Γ⁡(X,−). Then Ri⁢Γ⁢(X,−)≕Hi⁢(X,−) are called the cohomology functors of X.

Theorem 2.1.13.

Let F:𝒜→ℬ be an exact functor as above. Then

  1. 1.

    The derived functors Ri⁢F are well-defined and additive.

  2. 2.

    If 0→A′→A→A′′→0 is a short exact sequence in 𝒜, we have a long exact sequence

  3. 3.

    Given two short exact sequences and morphism f:A∙→B∙ in 𝒜, the diagram

    commutes.

Example 2.1.14.

If I is injective, then Ri⁢F⁢(I)=0 for all i>0.

Injective resolutions are hard to compute, so we will try construct a resolution that is easier to compute.

Definition 2.1.15.

A∈𝒜 is called F-acyclic if Ri⁢F⁢(A)=0 for all i>0.

Example 2.1.16.

Injective objects are acyclic for all left-exact functors.

Proposition 2.1.17.

If A→J∙ is ann F-acyclic resolution, then Ri⁢F⁢(A)=Hi⁢(F⁡(J)).

Proof.

Consider the diagram of exact sequences

Becuase the Ij are injective, the map J∙→I∙ is a quasi-isomorphism. Next, K∙ is exact. Because Ji,Ii are acyclic, so are the Ki. But this means that the the map F⁡(J∙)→F⁡(I∙) is a quasi-isomorphism. To complete this, it is an exercise that if C∙ is an exact complex of acyclic objects, then F⁡(C∙) is exact. □

Definition 2.1.18.

A sheaf ℱ∈𝖠𝖻⁡(X) is called flasque if for all V⊆U the restriction ℱ⁡(U)→ℱ⁡(V) is surjective.

Exercise 2.1.19.

Let 0→ℱ′→ℱ→ℱ′′→0 be a short exact sequence. If ℱ′ is flasque, then Γ⁡(X,ℱ)↠Γ⁡(X,ℱ′′) is surjective.

Example 2.1.20.

If (X,𝒪X) is a ringed space, then injective sheaves are flasque.

Proposition 2.1.21.

Let X be a topological space and ℱ∈𝖠𝖻⁡(X) be flasque. Then ℱ is Γ⁡(X,−)-acyclic.

Proof.

Let 0→ℱ→I→𝒢→0 be an exact sequence with I injective. Then because ℱ is flasque, we obtain an exact sequence

Because Hi⁢(X,I)=0 for all i>0, we know H1⁢(X,ℱ)=0 and Hi⁢(X,𝒢)≃Hi+1⁢(X,ℱ). By induction, we see that ℱ is acyclic. □

Remark 2.1.22.

If (X,𝒪X) is a ringed space, cohomology in 𝖬𝗈𝖽⁡(X) is the same as cohomology in 𝖠𝖻⁡(X).

Theorem 2.1.23.

Let X be a Noetherian topological space of dimension n. Then Hi⁢(X,ℱ)=0 for i>n.

Lemma 2.1.24.

Let i:Y↪X be a closed immersion. Then Hi⁢(Y,ℱ)=Hi⁢(X,i∗⁢ℱ).

2.2 Cohomology of Noetherian Schemes

Theorem 2.2.1.

Let X=Spec⁡A be a Noetherian scheme and ℱ be quasicoherent. Then ℱ is Γ⁡(X,−)-acyclic. In other words, Hi⁢(X,ℱ)=0 for all i>0.

This result is implied by the following proposition:

Proposition 2.2.2.

Let I be a injective A=module. Then I~ is flasque.

To prove the theoremm from this proposition, write ℱ=M~ and let M→I∙ be an injective resolution in 𝖬𝗈𝖽⁡(A). Then M~→I~∙ is a flasque resolution, so it computes the cohomology.

Corollary 2.2.3.

Let X be a Noetherian scheme and ℱ be quasicoherent. Then for ℱ coherent, there exists 𝒢 quasicoherent and flasque with ℱ↪𝒢.

Proof.

Cover X be affines Ui. Then we have an injection

Theorem 2.2.4 (Serre).

Let X be Noetherian. The following are equivalent:

  1. 1.

    X is affine.

  2. 2.

    Hi⁢(X,ℱ)=0 for all quasicoherent ℱ and i>0.

  3. 3.

    H1⁢(X,ℐ)=0 for all coherent ideal sheaves ℐ⊆𝒪X.

Proof.

Clearly 1 implies 2 implies 3, so now choose p∈X a closed point and U∋p be an open neighborhood. Then if Y=X∖U and Z=Y∪p, we have an exact sequence

From the vanishing of H1 for coherent ideal sheaves, there exists f∈Γ⁡(X,𝒪X) such that f⁡(p)≠0. Then Xf⊆U is affine, so now we need to show that ⟨f1,…,fk⟩=Γ⁡(X,𝒪X). □

2.3 Čech cohomology

Let X be a topological space and 𝒰={Ui}i∈I and fix an ordering of I. Then let ℱ∈𝖠𝖻⁡(X). For all p≥0, define the sheaf

and the group

Now we may define a complex by

Example 2.3.1.

The kernel of d0 is precisely the global sections.

Exercise 2.3.2.

The complex ℱ→𝒞0⁢(𝒰,ℱ)→𝒞1⁢(𝒰,ℱ)→⋯ is a resolution of ℱ.

Definition 2.3.3.

Let X,𝒰,ℱ be as above. Then the Čech cohomology of the covering is defined as

then the Čech cohomology of X is defined as Hˇi⁢(X,ℱ)=lim→𝒰⁡Hˇi⁢(𝒰,ℱ).

Now we will compare Čech cohomology and derived functor cohomology.

  1. 1.

    Flasque sheaves have no higher Čech cohomology. In particular, if ℱ is flasque then Hˇi⁢(ℱ,ℱ)=0 for all 𝒰 and all i>0.

    To see this, note that ℱ|U is flasque, so ℱ→𝒞∙⁢(𝒰,ℱ) is a flasque resolution. Therefore the Čech cohomology is the same as the usual cohomology, so by flasqueness, they must vanish.

  2. 2.

    Let X be a topological space and 𝒰 be an open covering. Then there exists a functorial map Hˇi⁢(𝒰,ℱ)→Hi⁢(X,ℱ).

  3. 3.

    If X is Noetherian and separated, then for every ℱ quasicoherent and open affine cover 𝒰, the map Hˇi⁢(𝒰,ℱ)→Hi⁢(X,ℱ) is an isomorphism.

    Because X is separated, Ui∩Uj is affine, so if we denote Uα=Ui0∩⋯∩Uip for any multi-index α=i0⁢…⁢ip, then we can consider an exact sequence

    Because 𝒢 is quasicoherent and flasque, it has no derived functor cohomology and no Čech cohomology. Because the Uα are affine, then

    is exact, so we obtain an exact sequence

    of complexes. Now we obtain a long exact sequence in the Hi⁢(X,−), so now Hi⁢(X,ℰ)=Hi+1⁢(X,ℱ). Using the snake lemma, we haave the same result for the Čech cohomology. Because the desired result holds for i=0, we use induction to obtain it for all i.

2.4 Cohomology of projective schemes

Theorem 2.4.1.

Let A be a Noetherian ring and X=ℙAr.

  1. 1.

    As graded rings, ⨁H0⁢(X,𝒪X⁢(n))≃A⁡[x0,…,xr].

  2. 2.

    For all n∈ℤ and 0<i<r, Hi⁢(X,𝒪X⁢(n))=0.

  3. 3.

    Hr⁢(X,𝒪X⁢(−r−1))=A.

  4. 4.

    The map H0⁢(X,𝒪X⁢(n))×Hr⁢(X,𝒪X⁢(−r−1−n))→Hr⁢(X,𝒪X⁢(−r−1)) is a perfect pairing.

Remark 2.4.2.

The bundle 𝒪X⁢(−r−1)≃ωX is the canonical bundle det⁡ΩℙAn1=⋀rΩℙAn1. Compare the third result to hr⁢(ℙℂr,Ωℂr)=hr,r⁢(ℙℂr)=1.

Remark 2.4.3.

These are particular instances of Serre duality, which says that when A=k, then

  1. 1.

    Hr⁢(ωX)=k;

  2. 2.

    The map Hom⁡(ℱ,ωX)×Hr⁢(X,ℱ)→Hr⁢(X,ωX)=k is a perfect pairing;

  3. 3.

    Exti⁡(ℱ,ωX)≃Hr−i⁢(X,ℱ)∨,

which holds when X is a projective scheme over k and ωX is the dualizing sheaf. In nice cases, for example when X is smooth, ωX is just the canonical bundle.

The proof of the theorem is a direct computation using Čech cohomology. Now we will state important finite results for cohomology of coherent sheaves on Noetherian projective schemes.

Theorem 2.4.4 (Serre).

Let X→Spec⁡A be a projective scheme of finite type over A with A Noetherian. Let ℱ be a coherent sheaf on X and 𝒪X⁢(1) be very ample. Then

  1. 1.

    Hi⁢(X,ℱ) is a finitely generated A-module for all i;

  2. 2.

    There exists n0 such that for all i>0 and n≥n0, Hi⁢(X,ℱ⁡(n))=0.

Proof.

We can reduce this to the case of X=ℙAr. For i>r, we know Hi⁢(X,ℱ)=0 because ℙAr is covered by r+1 affine open subsets and thus 𝒞p⁢(𝒰,ℱ)=0 for p>r+1. For the second part, we use descending induction on i, using the fact that there exists a surjection 𝒪XN⁢(n)↠ℱ for n≥n0 and what we already know about Hi⁢(𝒪X⁢(n)). □

Remark 2.4.5.

The n0 depends very much on ℱ. A crucial point in the construction of Hilbert schemes is to find an n0 that works for any sheaf of ideals ℐZ as long as we fix the Hilbert polynomial of Z.

Definition 2.4.6.

Let X be projective over k with 𝒪X⁢(i) ample and ℱ be coherent. Then define the Euler characteristic

This is additive on short exact sequences of coherent sheaves.

Definition 2.4.7.

The function n↦χ⁡(X,ℱ⁡(n)) is a polynomial with rational coefficients, called the Hilbert polynomial pℱ⁢(n).

Example 2.4.8.

If X=ℙr, then p𝒪X⁢(n)=pX⁢(n)=(n+rr).

Exercise 2.4.9.

Compute the Hilbert polynomial of a degree d hypersurface Y⊆ℙr.

Remark 2.4.10.

Given ℱ, the coefficients of the Hilbert polynomial pℱ⁢(n) are important invariants of ℱ.

Remark 2.4.11.

We will see later that the Hilbert polynomial is constant in flat families.

2.5 Higher direct images

Let f:X→Y be a continuous map of topological spaces. Then f∗:𝔄⁢𝔟⁢(X)→𝖠𝖻⁡(Y) is left exact. We also know that 𝖠𝖻⁡(X) has enough injectives, so we may consider the right derived functors.

Definition 2.5.1.

The functors Ri⁢f∗:𝖠𝖻⁡(X)→𝖠𝖻⁡(Y) are the higher direct image functors.

Proposition 2.5.2.

The higher direct image Ri⁢f∗⁢ℱ is the sheaf associated to the presheaf

In particular, Ri⁢f∗⁢ℱ|V=Ri⁢(f|f−1⁢(V))∗⁢(𝔉|f−1⁢(V)) and if ℱ is flasque, then Ri⁢f∗⁢ℱ=0 for all i>0. This means that flasque sheaves are f∗-acyclic, so they may be used to compute higher direct images. Also computing higher direct images is the same in 𝖠𝖻⁡(X) and in 𝖬𝗈𝖽⁡(X).

Proposition 2.5.3.

Let f:X→Y=Spec⁡A with X Noetherian. Let ℱ be quasicoherent. Then Ri⁢f∗⁢ℱ=Hi⁢(X,ℱ)~. Therefore Ri⁢f∗⁢ℱ is quasicoherent, and if f is projective with A Noetherian, then Ri⁢f∗ preserves coherent sheaves.

Proposition 2.5.4.

Let f:X→Y be a morphism of separated Noetherian schemes, ℱ be quasicoherent on X, and 𝒰={Ui} be an open affine cover. Let 𝒞∙⁢(𝒰,ℱ) be the Čech resolution of ℱ. Then Ri⁢f∗⁢ℱ≃Hi⁢(f∗⁢𝒞∙⁢(𝒰,ℱ)).

Proof.

Because Y is separated, for all V⊆Y affine, f−1⁢(V)∩Ui is affine. Therefore we may assume that Y=Spec⁡A is affine. But now

□

Remark 2.5.5.

It is also often useful to use the long exact sequence for right derived functors.

Theorem 2.5.6.

Let f:X→Y be a projective morphism of finite type with X,Y Noetherian. Suppose 𝒪X⁢(1) is very ample over Y and let ℱ be coherent on X.

  1. 1.

    There exists n0 such that for n≥n0 the map f∗⁢f∗⁢ℱ⁢(n)→ℱ⁡(n) is surjective;

  2. 2.

    Ri⁢f∗⁢ℱ is coherent;

  3. 3.

    There exists n0 such that for all n≥n0 and i>0, Ri⁢f∗⁢ℱ⁢(n)=0 for i>0.

Proof.

Because Y is quasicompact, we may reduce to the affine case. □

Now we will consider base change of Ri⁢f∗⁢ℱ along general morphisms X′→Y.

Proposition 2.5.7.

Let X,Y be Noetherian and separated schemes, f:X→Y be of finite type, and Y′ be Noetherian. Suppose ℱ is quasicoherent and let

be Cartesian. Then there exists a base change morphism

which is an isomorphism if v is flat.

Proof.

We may assume Y=Spec⁡A,Y′=Spec⁡A′ are affine. Then Ri⁢f∗⁢ℱ=Hi⁢(X,ℱ)~, so

On the other hand, if we cover X′ by Ui×ASpec⁡A′≕Ui′, then 𝒞∙⁢(𝒰′,u∗⁢ℱ)=u∗⁢𝒞∙⁢(𝒰,ℱ) and C∙⁢(𝒰′,u∗⁢ℱ)=C∗⁢(𝒰,ℱ)⊗AA′. Therefore we obtain a map

which is an isomorphism if A→A′ is flat. □

Remark 2.5.8.

If y∈Y is a point, then y→Y is in general not flat, so it is not easy to compare (Ri⁢f∗⁢ℱ)y and Hi⁢(Xy,ℱy).

Example 2.5.9.

Let C be a smooth curve over k that is irrational. Then consider π=p2:C×C→C and let ℒ=𝒪C×C⁢(Σ−Δ) where Σ=p0×C and Δ is the diagonal. Then π∗⁢ℒ is torsion free on C and thus locally free. However,

Therefore the rank of H0⁢(C,(π∗⁢ℒ)p) changes.

Remark 2.5.10.

When f:X→Y is flat, we will see some criteria to understand what happens.

2.6 Flatness and base change

Definition 2.6.1.

Let f:X→Y be a morphism of schemees. then ℱ is flat over Y if for all x∈X the stalk ℱx is a flat 𝒪Y,f⁡(x)-module. If 𝒪X is flat over Y, then f is said to be a flat morphism.

Here are some important results about flat morphisms:

  1. 1.

    Flat morphisms of locally Noetherian schemes are equidimensional. This means that for all x∈X, dimx(Xy)+dimyY=dimxX. To prove this, use going down.

  2. 2.

    If X is integral of dimension 1 and Y is regular, then f is flat if and only if it is dominant. In fact, without assuming that X is integral, f is flat if and only if every associaated point of X dominates Y.

  3. 3.

    If Y is regular of dimension 1, p is a closed point, and U≔Y∖p, then for all XU⊆ℙUn flat over U, there exists a flat limit X¯U:X→Y sending Xp to p.

Now for any morphism f:X→Y, we know that the Čech resolution 𝒞∙⁢(𝒰,ℱ) computes Ri⁢f∗⁢ℱ, and this is compatible with base change to an open subset of Y or with flat base change. For an arbitrary base change, the Čech resolution does not work, but if ℱ is flat over Y, we can cook up a complex that computes cohomology compatibly with base change.

Theorem 2.6.2.

Let f:X→Y be a projective morphism of Noetherian schemes with Y=Spec⁡A. Let ℱ be coherent on X. Then there exists a finite complex of finitely generated projective A-modules

such that for all A→A′, there exists a natural isomorphism Hi⁢(X′,u∗⁢ℱ)=Hi⁢(K∙⊗AA′), where

Corollary 2.6.3.

There exists a complex K~∙ of locally free coherent sheaves such that

Remark 2.6.4.

The proof of this only uses coherence of Ri⁢f∗⁢ℱ which holds more generally for proper morphisms.

Corollary 2.6.5.

The map y↦dimk⁡(Y)Hi⁢(Xy,ℱy) is upper semicontinuous. In addition, y↦χ⁡(Xy,ℱy) is locally constant.

Proof.

Note that

and the last two terms are lower semicontinuous because the Kj are locally free. For the second part, note that

which is locally constant on Y. □

Corollary 2.6.6.

The Hilbert polynomial is constant in flat families.

Corollary 2.6.7.

Assume that Y is reduced. The following are equivalent:

  1. 1.

    The map y↦dimHi⁢(Xy,ℱy) is constant.

  2. 2.

    The sheaf Ri⁢f∗⁢ℱ is locally free and Ri⁢f∗⁢ℱ⊗k⁡(y)→Hi⁢(Xy,ℱy) is an isomorphism.

Moreover, if these conditions are satisfied, then

is also an isomorphism.

Proof.

Use the following two facts:

  1. 1.

    If ℱ is coherent on Y, then dimk⁡(y)(ℱ⊗k⁡(y))≡r if and only if ℱ is locally free of rank r.

  2. 2.

    If ℱ,𝒢 are locally free on Y and φ:ℱ→𝒢 is a morphism such that rk⁡(φy)=r for all y, then locally on Y, there exists a splitting ℱ=ℱ1⊕ℱ2,ℱ=𝒢1⊕𝒢2 with all ℱi,𝒢i locally free such that

    with ψ:ℱ2→𝒢1 an isomorphism. □

Corollary 2.6.8.

Let Y be reduced. If Hi⁢(Xy,ℱy)=0 for all y∈Y, then Ri⁢f∗⁢ℱ=0.

Corollary 2.6.9.

Let Y be reduced. If Ri⁢f∗⁢ℱ=0 for all i≥i0, then Hi⁢(Xy,ℱy)=0 for all i≥i0 and y∈Y.

In fact, we can prove a stronger result.

Theorem 2.6.10.

Let f:X→Y be a projective morphism of fintie type between Noetherian schemes. Let ℱ be flat over Y. Then

  1. 1.

    If φYi:Ri⁢f∗⁢ℱ⊗k⁡(y)→Hi⁢(Xy,ℱy) is surjective at y, then it is an isomorphism at y and the same is true in a neighborhood of y∈Y.

  2. 2.

    If φyi is surjective, then the following are equivalent:

    1. (a)

      φyi−1 is also surjective;

    2. (b)

      Ri⁢f∗⁢ℱ is locally free in a neighborhood of y.

Corollary 2.6.11.

If Hi⁢(Xy,ℱy)=0 for all y∈Y, then φyi is surjective for all y, and thus φyi is an isomorphism, so (Ri⁢f∗⁢ℱ)y=0. This implies that Ri⁢f∗⁢ℱ=0 around y, and thus φyi−1 is also surjective.

Exercise 2.6.12.

Let X,Y be Noetherian and f:X→Y be flat and proper. Suppose that for all y∈Y, Xy≃ℙk⁡(y)n. If ℒ is invertible such that ℒ|Xy=𝒪Xy for all y∈Y, then there exists an invertible sheaf ℳ such that ℒ=f∗⁢ℳ.

Hint: Set ℳ=f∗⁢ℒ and prove that f∗⁢f∗⁢ℒ→ℒ is an isomorphism.

Exercise 2.6.13.

Let X be Noetherian and connected. Show that Pic⁡(X×ℙℤn)≃Pic⁡X×ℤ.

Hint: Show that the map Pic⁡X×Pic⁡ℙn→Pic⁡(X×ℙn) is an isomorphism.