Patrick Lei

All notes

Varieties

Lectures by Jenia Tevelev, UMass, Spring 2019.

Abstract

This will be the first course in algebraic geometry - the study of geometric spaces locally defined by polynomial equations. It is a central subject in mathematics with strong connections to differential geometry, number theory, and representation theory. We will pursue an algebraic approach to the subject, when local data is studied via the commutative algebra of quotients of polynomial rings in several variables. The emphasis will be on basic constructions and examples. Topics will include projective varieties, resolution of singularities, divisors and differential forms. Examples will include algebraic curves of low genus and surfaces in projective 3-space. In addition to theoretical approach, we will also learn how to use computer algebra software, specifically the Macaulay 2 package, to help with basic calculations in commutative algebra and algebraic geometry.

Forms of evaluation: biweekly homeworks (25%), take-home midterms (50%) and computer algebra project (25%).

1 Organizational/Philosophical

Warning: All jokes and conversations are reproduced as best as I can remember and my transcription is not necessarily faithful. In addition, footnotes and some definitions are based on my understanding of algebraic geometry and related topics. Also, I am a terrible person and do not include diagrams. Form your own geometric intuition.

The syllabus is on Moodle. He wanted to project the syllabus on the board, but was unable to get the projector to work. Jenia promised to email everyone the syllabus today.

Jenia considers Shafarevich (hereto referred to as Shaf) to be the best way to learn algebraic geometry.1 It combines the Russian algebraic geometry approach (by Shafarevich, others, and Jenia himself) with the translation by Miles Reid (British, more colloquial approach). He has used the book to teach both undergrads and graduate students. Knowing projective algebraic varieties is very helpful in the future whether you use homological/categorical, complex (holomorphic), or scheme theory methods. The plan is to cover all of Shafarevich in this semester.

There are people who attempt to read Mumford instead (it’s fantastic), but the book is much more difficult and he attempts to develop both varieties and schemes at the same time.

Jenia notes that people who do algebraic geometry use computers, and he is going to be teaching us some Macaulay 2 (especially because computer packages can do commmutative algebra). Note that this is not a course in computational algebraic geometry (which Jenia, Paul, and Eyal don’t know very much about). There is David Cox at Amherst, who is retiring in June, who wrote a book titled Ideals, Varieties, and Algorithms. There will be a computer algebra project.

There will be two kinds of homework problems. Some are normal homework assignments while others will be posted on Moodle. Two homeworks will be designated as take-home midterms.

Jenia struggled to turn off the projector but managed to do so.

2 Affine Plane Curves

2.1 Lecture 1 (Jan 22)

Historically, algebraic geometry came from two directions: projective geometry and abelian integrals. These are the two big sources of algebraic geometry, and much early progress was about the two subjects.

2.1.1 Enumerative Geometry

Theorem 1 (Butterfly Theorem).

Consider an ellipse with a chord A⁢C and let B be the midpoint of A⁢C. Draw two chords through B and form a “butterfly.” We have two new points P,Q. Then P⁢B=B⁢Q.

Proof.

Observe we have three plane curves passing through four points: A conic e given by f and two unions of two lines c1,c2 given by f1,f2. We have a linear system of conics passing through four points.

The space of all conics forms an ℝ6. Every point imposes a linear condition on the coefficients, so conics passing through out four points, so the space we care about is at least two-dimensional. We will later prove that this is exactly two.

We see that f2=α⁢f+β⁢f1. Now we restrict our polynomials to the line passing through A,C and assume it is the x−a⁢x⁢i⁢s and that B is the origin. We see that f∣y=0=(x−c)⁢(x+c), f1∣y=0=x2, and f2∣y=0=(x−p)⁢(x−q). Then we must have that p+q=0. □

Jenia learned this theorem in high school.2 He told a story about Sheldon Katz and one of his interviews.

Lemma 2.

The dimension of the vector space is exactly 2.

Proof.

Assume dim(V)≥3. Let X be another point of the ellipse and consider the subspace V0⊂V of conics passing through X. Then we see that dim(V0)≥2. Then we have two linearly independent conics passing C,D passing through five points. This contradicts Bezout’s Theorem which states that if C has degree n and D has degree n, then |C∩D|≤n⁢m unless the curves have a shared component. □

Jenia says this is nice argument because it connects to enumerative geometry, the connection between algebraic and geometric objects, and curves on surfaces, which we will talk about later when we talk about divisors.

Another famous example of a similar argument is Pascal’s Theorem:

Theorem 3 (Pascal).

Given a conic with an inscribed hexagon, then opposite sides intersect at three collinear points.

Jenia attempted to draw the diagram on the board, but diagrams are not to scale. The argument is very similar to the butterfly proof and can be found in Shaf.

2.1.2 Basic Notions

We will give some definitions to formalize affine plane curves.

Definition 4 (Affine Plane Curve).

Let k be algebraically closed and denote by 𝔸2 the affine plane. Let f∈k⁡[x,y] be nonconstant of degree d>0. Then the vanishing locus C of f is an affine plane curve of degree d. Examples are lines, conics, cubics, quartics, and quintics (which are only for experts).

Remark 5.

Observe that k⁡[x,y] is a UFD (more generally, if R is a UFD, then R⁡[x] is a UFD). This is an application of Gauss’s Lemma.

Definition 6 (Irreducible Components).

We factor f=f1⁢⋯⁢fr into irreducibles. Then the vanishing locus ci of each fi is an irreducible component of C.

Why do we consider algebraically closed fields? One reason is to have more points. We can see that if our field is algebraically closed, our curve has points.

Lemma 7.

Let k=k¯. Then:

  1. 1.

    C contains infinitely many points.

  2. 2.

    Suppose f∈k⁡[x,y] is irreducible and g∈k⁡[x,y]. Then the set {f=g=0} is finite unless f divides g. In particular, an irreducible polynomial is uniquely determined up to a scalar by its curve.

Proof.

We use the fact that k is algebraically closed.

  1. 1.

    Every algebraically closed field is infinite, so we can write f=∑gi⁢(x)⁢yi. Then take any x0 and solve for y. Note each gi is a polynomial and has finitely many roots, so there are only finitely many x0 such that ∑gi⁢(x0)⁢yi is a nonzero constant.

  2. 2.

    We use Gauss’s Lemma in a different form: Let R be a UFD with field of fractions K. Then let f∈R⁡[y]. Then if f is irreducible over R, it must be irreducible over K. We see that f,g∈k⁡[x,y]=k⁡[x]⁢[y]⊂k⁡(x)⁢[y]. By Gauss’s Lemma, we know that f does not divide g in k⁡(x)⁢[y] and that f is irreducible. Note that k⁡(x)⁢[y] is a PID, so f,g are coprime and using Bezout’s Lemma, we have 1=α⁢f+β⁢g. Let p be an LCM of the denominators of the coefficients of α,β. Then we have p⁡(x)=α0⁢f+β0⁢g. Suppose f⁡(x0,y0)=g⁡(x0,y0)=0. Then p⁡(x0)=0, which has only finitely many solutions for x0. By the same argument, there are only finitely many possible values for y0.

□

During the proof of the previous lemma, Jenia said, “Every time we run into a problem, we have to use a little bit more commutative algebra. That’s how algebraic geometry works.” In addition, Jenia’s phone rang and made frog noises.

We now have enough vocabulary in plane curves to state Bezout’s Theorem, but Jenia now wants to talk about Abelian integrals.

2.1.3 Abelian Integrals

We want to be able to calculate

where u is a rational function in two complex variables. In Calc 2, we use trig substitutions, but Jenia learned something called Euler’s substitutions. We can rewrite the integral over the unit circle, which has a rational parameterization

We can now express the integral in terms of t and compute the integral using partial fractions.

This method is harder than using trig substitution, but is more general. If your curve is rational, then integrating over it is amenable to this method.

Definition 8 (Rational Curve).

An irreducible curve C=(f=0) is rational if there exist φ⁡(t),ψ⁡(t)∈k⁡(t) nonconstant such that f⁡(φ⁡(t),ψ⁡(t))=0.3

Corollary 9.

If C is rational, then the integral over the curve ∫u⁡(x,y)⁢𝑑x, where u is rational, can be computed using partial fractions.

Remark 10.

Later in the semester, we will talk about differential forms, which are a way to talk about integration is multiple dimensions.

2.2 Lecture 2 (Jan 24)

2.2.1 Non-Mathematics

Jenia was actually able to make the projector work today. He told us to install Macaulay 2 which you can run with the command M2. Jenia showed us some basic commands in Macaulay 2. Next, he showed us the Moodle and the forum with problems for us to solve.

2.2.2 Rational Curves Continued

We continue with our preliminary discussion of plane algebraic curves to discuss some history and motivation. Last time we discussed Abelian integrals focusing on the example of the circle. From the rational parameterization (2⁢t1+t2,t2−1t2+1), we can find all possible Pythagorean triples by choosing rational values for t. We saw last time that we can reduce an integral of the form ∫u⁡(x,y)⁢dx+∫v⁡(x,y)⁢dy to an integral of the form ∫φ⁡(t)⁢dt.

We introduce a new notion, the field of rational functions k⁡(C) which is the field of all rational functions on the curve. Observe, however, that rational functions are not everywhere defined. There are finitely many points on the curve where the denominator vanishes. We can say that u1=u2 if they agree outside their bad points. Algebraically, we can say that p1q1∼p2q2⇔f∣p1⁢q2−p2⁢q1.

Alternately, we can define the coordinate algebra k⁡[C]=k⁡[x,y]/(f) and then define k⁡(C) to be its field of fractions.4

Definition 11 (Regular Function).

u∈k⁡(C) is regular at P∈C if there exist p,q∈k⁡[x,y] such that u=pq such that q⁡(P)≠0.

Example 12.

Let C={x2+y2=1} and u=1−xy. Then we can write u=y1+x, so it is regular at the point (1,0).

We can see that k⁡(C) is finitely generated by x,y. We also see that k⁡(C) has transcendence degree5 1 because f⁡(x,y)=0 is an algebraic dependence.

Lemma 13.

Every finitely generated field of transcendence degree 1 is isomorphic to k⁡(C) for some irreducible curve C.

Remark 14.

Different curves can have the same function field.

Definition 15.

We say that curves C,D are birational if k⁡(C)≃k⁡(D).6

Example 16.

Let C be the circle and observe that k⁡(C)≃k⁡(t)≃k⁡(𝔸1)=k⁡(x), so C is birational to 𝔸1.

Suppose C is a rational curve {f=0}. Then there exist φ,ψ∈k⁡(t) nonconstant such that f⁡(φ⁡(t),ψ⁡(t))=0.

Lemma 17.

If C is a rational curve, then k⁡(C)↪k⁡(t).

Proof.

Take the obvious map. Then we see that this is defined because if q⁡(φ⁡(t),ψ⁡(t))=0, then f∣q. □

So we see that rational curves have function field a subfield of k⁡(t). However, we invoke a theorem from algebra.

Theorem 18 (Lüroth).

Every subfield of k⁡(t) that is not a subfield of k⁡(t) is isomorphic to k⁡(t).

Corollary 19.

A curve C is rational if and only if C is birational to a line.

Now we discuss nonrational curves and try to find a simple example.

2.2.3 Elliptic Curves

In 1655, Wallis was interested in computing the arc length of an ellipse x2a2+y2b2=1. Then we see that y=ba⁢a2−x2 and y′=−b⁢xa⁢a2−x2. We substitute into the formula for arc length ∫1+(y′)2⁢dx, change variables, and obtain an integral of the form

Then set C={y2=(1−e2⁢x2)⁢(1−x2)}. We will see that this curve is an elliptic curve.

Let’s perform an invertible change of variables by moving 1 to ∞. Then we set t=11−x. We now obtain the equation

Now multiply by t4 and obtain the equation

before setting s=y⁢t2 to obtain an equation of the form s2=f3⁢(t), where f3⁢(t) is a cubic polynomial. Then, after further simplification,7 we obtain an elliptic curve D given by s2=w3+a⁢w+b, which is birational to our original curve C.

Theorem 20.

Elliptic curves are not rational.8 Proof of this fact will give us motivation to understand other definitions

Sketch.

First, we projectivize our curves, so we have an elliptic curve C↪ℙ2 and a line L↪ℙ2. The next step is to show that the two curves are non-singular. Third, we show that if C,L are birational and non-singular, then they are isomorphic. Then, we see that C has a simple involution φ given by s↦−s. Then we see that s has four fixed points. Now we use this involution to create an involution f∘φ∘f−1 with 4>2 fixed points. However, every automorphism9 of L is given by the images of three points,10 so this is impossible. □

2.2.4 Non-singularity of plane curves

Conider a curve C={f=0} and let (a,b)∈C. Then we have a Taylor expansion f⁡(x,y)=f⁡(a,b)+fx⁢(a,b)⁢(x−a)+fy⁢(a,b)⁢(y−b)+⋯.

Definition 21 (Non-singularity).

A plane curve C is nonsingular at (a,b) if at least one of fx⁢(a,b),fy⁢(a,b)≠0. In this case, f⁡(x,y)=f⁡(a,b)+fx⁢(a,b)⁢(x−a)+fy⁢(a,b)⁢(y−b) is called the tangent line.

Definition 22 (Multiplicity).

We say that (a,b)∈C has multiplicity m is the smallest positive integer such that an order m partial derivative does not vanish at m.

We see that a point is singular if it has multiplicity at least 2.

Example 23 (Singular Conics).

Take (0,0)∈C and suppose the origin is a singularity. Then f=α⁢x2+β⁢y2+γ⁢x⁢y=(a⁢x+b⁢y)⁢(c⁢x+𝑑y). Thus C must be a union of two lines.

3 Affine Zariski-Closed Sets

3.1 Lecture 3 (Jan 29)

We will put plane curves away for a while and then return to them. In this section, we will review several facts from commutative algebra.

Theorem 24 (Noether Normalization Lemma).

Let A be a finitely generated k-algebra.11 Then A contains x1,…,xn algebraically independent over k and such that A is integral over k⁡[x1,…,xn].

Proof.

We will assume that k is infinite. Choose some generators y1,…,yr. We argue by induction on r. If r=0, then A=k, so there is nothing to prove. If y1,…,yr are algebraically independent, then A=k⁡[y1,…,yr].

Suppose that f⁡(y1,…,yr)=0 for some f∈k⁡[Y1,…,Yr]. We can write

If hd=1, then yr is integral over the subring B generated by y1,…,yr−1. By induction, B contains algebraically independent x1,…,xn such that B is integral over k⁡[x1,…,xn]. Therefore, A must be integral over k⁡[x1,…,xn] by transitivity of integrality.

Noether’s trick is to show that we can always reduce to the simple case by a linear change of variables. Introduce y1′=y1−λ1⁢yr,…,yr−1′=λr−1⁢yr,yr′=yr. We see that y1′,…,yr′ generate A. Then we see that f⁡(y1,…,yr)=f⁡(y1′+λ1⁢yr′,…,yr−1′+λr−1⁢yr′,yr′)=0. Write f=F+l.o.t. and suppose F is of degree d. Then we see that F⁡(y1′+λ1⁢yr′,…,yr−1′+λr−1⁢yr′,yr′)=F⁡(λ1,…,λr−1,1)⁢yr′d. We simply need F⁡(λ1,…,λr−1,1)≠0, which is always possible if k is infinite. □

Recall the fundamental theorem of algebra, which states that ℂ is algebraically closed. This gives a bijection between ℂ and maximal ideals m⊂ℂ⁡[x] by a↔(x−a). Indeed, over any field, maximal ideals of k⁡[x] are principal ideals generated by irreducible monic polynomials. If k=k¯, then all monic polynomials are linear.

Theorem 25 (Weak Nullstellensatz).

Let k=k¯. Then there is a bijection between 𝔸n=kn and maximal ideals in k⁡[x1,…,xn] given by (a1,…,an)↔(x1−a1,…,xn−an).

Proof.

Given a point (a1,…,an)∈𝔸n, consider a homomorphism k⁡[x1,…,xn]→k given by f↦f⁡(a1,…,an). Then the kernel of this morphism is a maximal ideal, but it must be equal to (a1,…,an)

Now take m⊂k⁡[x1,…,xn] maximal and suppose k⁡[x1,…,xn]/m→≃𝜑K⊃k is a field. If K=k, define ai=φ⁡(xi). Then because φ acts on the variables just like the evaluation morphism, we must have m=ker⁢φ=(x1−a1,…,xn−an).

We need to show that K=k. Note that K must be a finitely generated k-algebra. By Noether, K is integral over a polynomial subring k⁡[x1,…,xn]. Recall the fact that if A is a field and A is integral over B, then B is a field. To show this, we see that b−1∈A, so b−1=(b−1)r+a1⁢(b−1)r−1+⋯+ar=0 where a1,…,ar∈B. Multiplying through by br−1, we see that b−1∈B. Therefore k⁡[x1,…,xn] is a field and n=0 and K is integral over k. Because k is algebraically closed, K=k. □

Definition 26.

A subset X⊂𝔸n is called a closed affine set12 if X={𝐚∈𝔸n|f1⁢(𝐚)=⋯=fr⁢(𝐚)=0} for some polynomials f1,…,fr∈k⁡[x1,…,xn].

Lemma 27.

Closed sets have the form V⁡(I)={(a1,…,an)∈𝔸n|f⁡(a1,…,an)=0⁢ for all f∈I} for some ideal I⊂k⁡[x1,…,xn].

Proof.

Take X={f1=⋯=fr=0}. Then let I=(f1,…,fr)⊂k⁡[x1,…,xn]. Then V⁡(I) must equal X. On the other hand, take V⁡(I). Because k⁡[x1,…,xn] is Noetherian, let f1,…,fr to be generators of I. Then V⁡(f1,…,fr)=V⁡(I)=X. □

Now let X⊂𝔸n be a closed set. Define I⁡(X)={f∈k⁡[x1,…,xn]|f⁡(𝐚)=0⁢ for all 𝐚∈X}. We now have two maps {Closed Sets}↔I⁡(X)V⁡(I){Ideals}.

Theorem 28 (Strong Nullstellensatz).

V⁡(I⁡(X)) and I⁡(V⁡(J))=J.

Proof.

Take g=I⁢(V⁡(I)). Suppose I=(f1,…,fr). We need to show that if g vanishes at every point (a1,…,an) such that fi⁢(a1,…,an)=0, then gℓ=h1⁢f1+⋯+hr⁢fr for some ℓ. We use a trick of Rabinowitch: Let B=(f1,…,fr,1−g⁢xn+1)⊂k⁡[x1,…,xn+1]. We show that B=k⁡[x1,…,xn+1]. If not, it is contained in some maximal ideal B⊂m=(x1−a1,…,xn+1−an+1). However, fi⁢(a1,…,an)=0 and 1−g⁡(a1,…,an)⁢an+1=0. But remember g∈I⁡(X), so we obtain that 1=0.

Thus we can write 1=∑hi⁢fi+hn+1⁢(1−g⁢xn+1). Then we write xn+1=1g⁡(x1,…,xn). Then we obtain an expression of the form

Then we can clear denominators and obtain that g∈I.

Finally we show that V⁡(I⁡(X))=X. Then there exists an ideal J such that X=V⁡(J). Then V⁡(I⁡(X))=V⁡(I⁡(V⁡(J)))=V⁡(J)=V⁡(J)=X. □

Remark 29.

I⁡(X) is always radical. In fact V⁡(I)=V⁡(I).

Corollary 30.

The operations I and V give a bijection between closed sets in 𝔸n and radical ideals in k⁡[x1,…,xn].

3.2 Lecture 4 (Jan 31)

Recall that the weak Nullstellensatz gives a bijection between points in 𝔸n and maximal ideals in k⁡[x1,…,xn]. Also, the strong Nullstellensatz gives a bijection between affine closed sets and radical ideals.

3.2.1 Zariski Topology on 𝔸n

Proposition 31.

The affine closed sets V⁡(I) form the set of closed sets for the Zariski topology on 𝔸n.

Proof.
  1. 1.

    𝔸n=V⁡(0) and ∅=V⁡((1));

  2. 2.

    ∩i∈IYi=V⁡(∑Ii);

  3. 3.

    ∪i=1rYi=V⁡(I1⁢⋯⁢Ir)=V⁡(I1∩⋯∩Ir).

□

Definition 32 (Irreducible Closed Sets).

A closed set is called irreducible if it is not the union of two proper closed subsets.

Example 33.

In ℝn with the Euclidean topology only points are irreducible.

Theorem 34.

Under the correspondence between V⁡(I),I⁡(X), we will see that irreducible subsets Y⊂𝔸n correspond to prime ideals 𝔭⊂k⁡[x1,…,xn], which we denote by Spec⁢k⁢[x1,…,xn].

Proof.

Suppose Y⊂𝔸n is reducible with Y=Y1∪Y2. Then I⁡(Y)⊊I⁡(Yi) for i=1,2 by the Nullstellensatz. Then choose f1∈I⁡(Y1)∖I⁡(Y) and f2∈I⁡(Y2)∖I⁡(Y). Then f1⁢f2 vanishes on Y and therefore I⁡(Y) is not prime.

Suppose I⁡(Y) is not prime. Then there exist f,g∉I⁡(Y)=I with f⁢g∈I⁡(Y). Then write Y1=V⁡(f)∩Y and Y2=V⁡(g)∩Y, so we see that Y=Y1∪Y2. □

Now we consider this correspondence in the case of the plane.

Irreducible subsets Prime ideals
Points (a,b) Maximal ideals (x−a,y−b)
Irreducible affine plane curves (f)
𝔸2 0
Figure 1: Irreducible subsets of 𝔸2 and prime ideals in k⁡[x,y].

Why is there nothing else? Suppose we have a prime ideal 0≠𝔭⊂k⁡[x,y]. Then write 𝔭=(f1,…,fr)=f⁡(g1,…,gr) where f=gcd⁡(f1,…,fr). Then either f∈𝔭 or gi∈𝔭 for all i. If f∈𝔭, we have that 𝔭=(f). Otherwise, 𝔭=(g1,…,gr). However, the gi are coprime so V⁡(𝔭) is a finite union of points, so it must be a single point. Thus 𝔭 is maximal.

“That’s what most arguments in algebraic geometry look like: some input from algebra and some input from geometry.”

Definition 35 (Irreducible Component).

Let Y⊂𝔸n be a Zariski closed set. Then an irreducible component of Y is a maximal irreducible subset of Y.

Lemma 36.

There exist only finitely many irreducible components Y1,…,Ys and Y=Y1∪⋯∪Ys.

Proof.

We show that Y can be written as a finite union z1∪⋯∪zr of irreducible subsets. Given that, we can also assume that zi⊊zj for all i≠j. Take some irreducible subset W⊂Y=Z1∪⋯∪Zr. But in fact W=∪(W∩Zi), so W⊂Zi for some I. Then in particular, if W is an irreducible component, then W=Zi. Therefore this decomposition is a decomposition into irreducible components.

To prove the claim, if Y is irreducible, then clearly this is true. Then write Y=Y1∪Y2. If Y1,Y2 are irreducible, the we are done. Suppose Y1 is reducible. Then write it as a union and continue to form a binary tree. But this tree must be finite because k⁡[x1,…,xn] is Noetherian. □

Remark 37.

The algebraic counterpart to that is I=𝔭1∩⋯∩𝔭s.

Proof.

Take f∈𝔭1∩⋯∩𝔭s. Then f vanishes on every yi, so it vanishes on y. Thus f∈I. □

Theorem 38.

Every radical ideal in l⁡[x1,…,xn] is the intersection of minimal prime ideals which contain I.13

We now shift to a more general perspective. So far we have considered 𝔸n and k⁡[x1,…,xn]. Now we will consider X,k⁡[X] where X is an irreducible Zariski-closed subset of 𝔸n and k⁡[X]=k⁡[x1,…,xn]/I⁡(X) the coordinate ring of X. We see that X is irreducible if and only if k⁡[X] is an integral domain.

Theorem 39 (Generalized Nullstellensatz).

There is a bijection between points of X and maximal ideals of k⁡[X], between closed subsets of X and radical ideals of k⁡[X], and between irreducible closed subsets of X and prime ideals of k⁡[X].

Proof.

Suppose Y⊂X is a closed subset. Then I⁡(Y)⊃I⁡(X) as radical ideals. Then I⁡(Y)/I⁡(X)⊂k⁡[X] is a radical ideal. □

3.3 Lecture 5 (Feb 5)

3.3.1 Macaulay 2 Interlude

Recall that we have algebraic sets X↪𝔸n and the coordinate algebra k⁡[X]=k⁡[x1,…,xn]/I⁡(X). Then for any Y⊂X, I⁡(Y)⊂k⁡[X] is a radical ideal and Y irreducible implies I⁡(Y) is prime. Then we can decompose Y as a union of irreducible components and I⁡(Y) as an intersection of minimal primes.

There are algorithms that can do these computations, but they generally use Gröbner bases, which are difficult to compute by hand. Fortunately, we can use computers, such as Macaulay2, to perform these calculations.

E = QQ[x] -* polynomial ring *-
ideal(x^2+1) -* ideal of i *-
isPrime oo -* true *-
clearAll
E = QQ[x,I]/ideal(I^2+1) -* i is in the ring *-
A = ideal(x^2+1)
isPrime A -* false *-
L = decompose A -* List of two ideals *-
length L -* 2 *-
L_0 -* x-i *-
L_1 -* x+i *-
clearAll

We now consider the Clebsch cubic surface x3+y3+z3+1=(x+y+z+1)3. There is a famous theorem that any smooth cubic surfaces contain exactly 27 lines. We will calculate these 27 lines.

R = ZZ/101[x,y,z,x0,y0,z0,a,b,c,t] -* initialize all parameters *-
cubic = x^3+y^3+z^3+1-(x+y+z+1)^3 -* cubic *-
cubt = sub(cubic, {x=>x0+a*t, y=>y0+b*t, z=>z0+c*t}) -* substitute parametric equation of line *-
cubic0 = sub(cubt, t=>0) -* constant term *-
cubic1 = sub(diff(t,cubt), t=>0) -* linear term *-
cubic2 = sub(diff(t,diff(t,cubt)), t=>0) -* quadratic term *-
cubic3 = sub(diff(t,diff(t,diff(t,cubt))), t=>0) -* cubic term *-
Lines = ideal(cubic0,cubic1,cubic2,cubic3) -* set identically zero *-
L=decompose Lines -* find all ideals *-
#L -* should be 25 *-
f = x^3+y^3+z^3-(x+y+z)^3 -* ideal at infinity *-
factor f -* three lines *-

We found an interesting phenomenon where the computation took longer and was the computer was unable to find the expected 25 affine lines when we worked over ℤ/103⁢ℤ because 5 is not a quadratic residue mod 103.14

3.3.2 Morphisms of Affine Closed Sets

Let f:𝔸n→𝔸m be given by yi=fi⁢(x1,…,xn), where fi∈k⁡[x1,…,xn] for i=1,…,m. Algebraically, this corresponds to a morphism k⁡[y1,…,ym]→𝜑k⁡[x1,…,xn], which is given by yi↦fi⁢(x1,…,xn).

We know that the kernel of this morphism is a prime ideal because the quotient by it is a subring of k⁡[x1,…,xn], so is an integral domain. Then V⁡(Ker⁢φ)=f⁡(𝔸n)¯. To see this, suppose that (b1,…,bm)=f⁡(a1,…,an). Then let g∈Ker⁢φ. We see that

In fact, Im⁢f¯=V⁡(Ker⁢φ).

Let X↪𝔸n. Now we define a morphism to be a restriction of a polynomial morphism from 𝔸n→𝔸m. We pull back k⁡[y1,…,ym]→k⁡[X] in the same way.

Finally, suppose X⊂𝔸n,Y⊂𝔸m. Then let f:X→𝔸m and suppose f⁡(X)⊂Y. Then we have a morphism X→Y and a pullback morphism k⁡[Y]→k⁡[X].

Theorem 40.

There is a bijection Hom⁡(X,Y)≃Hom⁡(k⁡[Y],k⁡[X]).15

3.4 Lecture 6 (Feb 7)

3.4.1 Morphisms Continued

Recall the definition of a morphism of affine algebraic sets. Also recall that any morphism f:X→Y induces a pullback homomorphism f∗:k⁡[Y]→k⁡[X].

Also recall Theorem 40. In particular, X≃Y if and only if k⁡[X]≃k⁡[Y].

Proof of Theorem 40.

Recover f from f∗ by bi=yi⁢(b)=yi⁢(f⁡(a))=(f∗⁢yi)⁢(a). Then given a morphism α:k⁡[Y]→k⁡[X], consider the following diagram.

k⁡[Y]k⁡[X]k⁡[y1,…,ym]k⁡[x1,…,xn]α

Then there exist fi such that the image of fi under the map k⁡[x1,…,xn]→k⁡[X] is α⁡(yi). Define f=(f1,…,fm). Clearly α=f∗.

We check that f⁡(X)⊂Y. Note g∈O⁡(Y) given by g⁡(f⁡(a1,…,an))=0 if and only if g⁡(f1,…,fm)∈I⁡(X) if and only if α⁡(g)∈I⁡(X) if and only if g restricted to Y is zero. □

Recall that X↪𝔸n corresponds algebraically to k⁡[x1,…,xn]↠k⁡[X], which is choice of generators of k⁡[X].

Let G be a finite group acting on X. Then G must also act on k⁡[X].

Theorem 41.

If the char⁢k does not divide |G|, then k⁢[x]G is finitely generated.

Choose generators g1,…,gn of k⁢[X]G given by ψ:k⁡[y1,…,yn]↠k⁢[X]G. Let I=ker⁡ψ.

Definition 42 (Quotient by a finite group).

Define Y=V⁡(I)⊂𝔸n. Then k⁡[Y]≃k⁢[X]G. Then Y=X/G. Note that k⁢[X]G↪k⁡[X], so there is a quotient morphism π:X→X/G.

Example 43.

Consider the action of ℤ/2⁢ℤ on 𝔸2 given by (x,y)↦(−x,−y). This gives an action on k⁡[x,y], so k⁢[x,y]ℤ2=k⁡[x2,y2,x⁢y]=k⁡[u,v,w]/(u⁢v−w2). Then we see 𝔸2/(ℤ/2⁢ℤ)=V⁡(u⁢v−w2)↪𝔸3 and the quotient morphism is given by (x,y)→(x2,y2,x⁢y).

Now we can translate some geometry into algebra and some algebra into geometry.

Definition 44 (Dominance).

A morphism f:X→Y is dominant if f⁡(X) is (Zariski-)dense in Y.

Proposition 45.

f is dominant if and only if f∗ is injective.

Proof.

Suppose f is dominant. Let φ∈ker⁡φ. We compute φ⁡(f⁡(a))=(f∗⁢φ)⁢(a)=0, so φ|f⁡(X)=0. However, f⁡(X)⊂Y is dense, so φ|Y=0.

Now suppose f∗ is injective but f is not dominant. Therefore f⁡(X) is not dense in Y, which is possible if and only if there exists φ∈K⁡[Y] nonzero such that φ|f⁡(X)=0. Therefore φ⁡(f⁡(a))=0 for all a∈X and thus f∗⁢φ=0, so f∗ is not injective. □

Definition 46 (Finite Morphism).

A dominant morphism f:X→Y is called finite if k⁡[X] is integral over k⁡[Y]. More generally, a morphism is finite if k⁡[X] is integral over the image of k⁡[Y].

Example 47 (Noether Normalization).

An integral extension k⁡[X]↩k⁡[y1,…,ym] corresponds to a finite morphism X→𝔸m.

Example 48 (Normalization of Cusp).

Note the map 𝔸1→V⁡(y2−x3)⊂𝔸2 given by t↦(t2,t3). Algebraically, k⁡[x,y]/(y2−x3)→k⁡[t] given by x↦t3,y↦t2. Because the morphism is dominant, we have R≃k⁡[t2,t3]↪k⁡[t] is an integral extension (t is a root of T2−t2).

Example 49.

Consider k⁢[X]G⊂k⁡[X]. This is always integral, so X↠X/G is finite. Consider α∈K⁡[X]. It is a root of ∏g∈G(T−g⁢α). Each coefficient is invariant because they are elementary symmetric functions on the g⁢α.

Recall the Going-up theorem:

Theorem 50 (Going-up).

Let 𝔭⊂A be a prime ideal and let A↪B be an integral extension. Then there exists a prime ideal 𝔮⊂B such that 𝔮∩A=𝔭.

Theorem 51.

Any dominant finite morphism is surjective and has finite fibers.

Proof.

Let α be a finite dominant morphism. Take y∈Y, which corresponds to a maximal ideal m⊂K⁡[Y]. We find a point x∈X (a maximal ideal n⊂k⁡[X]) such that α⁡(x)=y (n∩k⁡[Y]=m). By going-up, there exists a prime ideal q⊂k⁡[X] such that q∩k⁡[Y]=m. Take any maximal ideal n⊃q. Then n∩k⁡[Y]⊃m, so by maximality of m, n∩k⁡[Y]=m.

Now we prove that the fibers are finite. Let y∈Y correspond to a maximal ideal m⊂k⁡[Y]. Then we show that there are finitely many maximal ideals in k⁡[X]/(k⁡[x]⋅m). Then k⁡[X] is integral over k⁡[Y], so A=k⁡[X]/(k⁡[x]⋅m) is integral over k⁡[Y]/m=k. Therefore A is a finitely-generated k-module (a finite-dimensional k-vector space).16 □

3.5 Lecture 7 (Feb 14 ♡)

We begin with a remark: Let α:X→Y be a morphism of affine algebraic sets X→Y. Then there is a pullback α∗:k⁡[Y]→k⁡[X]. We note that if Z⊂Y is closed, then α−1⁢(Z)⊂X is also closed. Then if I⁡(Z)=(f1,…,fr), the ideal of α−1⁢(Z) is given by (α∗⁢f1,…,α∗⁢fr).

This becomes trickier when α is dominant (which is equivalent to α∗ being injective). Then we recall the computation of the fiber over a point to prove that dominant morphisms are surjective with finite fibers.17

We prove that Artinian rings have finitely many maximal ideals:

Lemma 52.

Artinian rings have finitely many maximal ideals.

Proof.

Consider all finite intersections of maximal ideals. This set contains a minimal element m1∩⋯∩mrx. Therefore m1,∩⋯∩mr⊂m for al maximal ideals m. Then we see m=mi for some i. (Otherwise, there exists xi∈mi∖m, and then x1⁢⋯⁢xr∈m, which contradicts maximality.) □

We recall our discussion of group actions and quotients. Suppose we have a finite group G acting on X. Then we have a finitely generated algebra k⁢[X]G↪k⁡[X]. Then we define X/G to be an affine algebraic set such that k⁡[Y]≃k⁢[X]G. Then there exists a quotient morphism π:X→X/G.

Proposition 53.

The fibers of π are G-orbits.

Proof.

First we show that y=g⁢x implies π⁡(X)=π⁡(y). Suppose not. Then there exists f∈k⁡[X/G] such that f⁡(π⁡(x))=0 and f⁡(π⁡(y))=1. We have a pullback π∗:k⁢[X]g↪k⁡[X]. Then there exists a G-invariant function f such that f⁡(x)=0 and f⁡(y)=1, which is impossible.

Next we show that the fiber is precisely the orbit. Suppose that G-orbits of x,y are disjoint. Then there exists f∈k⁡[X] such that f|G⋅x=0 and f|G⋅y=1. Then we define f#⁢(x)=1#⁢G⁢∑g∈Gf⁡(g⋅x). This new polynomial is now G-invariant and is always 0 on the orbit of x and 1 on the orbit of y. Then we see that f#⁢(π⁡(x))=(π∗⁢f#)⁢(x)=f#⁢(x)=0 and f#⁢(π⁡(y))=(π∗⁢f#)⁢(y)=f#⁢(y)=1. Thus π⁡(x)≠p⁢i⁢(y). □

3.5.1 Rational Maps

Now let X be an affine variety. Then k⁡[X] is an integral domain.

Definition 54 (Function Field).

The function field k⁡(X) of an affine variety X is the field of fractions of k⁡[X].

Definition 55 (Regular function).

Let f∈k⁡(X). Then f is regular at x∈X if f can be written as a fraction f=pq where p,q∈k⁡[X] and q⁡(x)≠0.

Definition 56 (Domain of Definition).

The domain of definition of f is the set of all x∈X such that f is regular at X.

Proposition 57.

The domain of definition is open.

Proof.

D=⋃f=p/q(X∖V⁡(q)). □

Proposition 58.

f⁡(x) is regular at every x∈X if and only if f∈k⁡[X].

Proof.

For all x∈X, there exists an expression f=px/qx where qx⁢(x)≠0. Then take I=(qx). The vanishing set V⁡(I)=∅, so I=(1), which implies that 1∈I. We can write 1=a1⁢qx1+a2⁢qx2+⋯+ar⁢qxr. Then

□

Definition 59 (Local ring of a point.).

Let x∈X. Then the ring 𝒪x is the set of all functions regular at x. Note that this is the same as k⁢[X]mx. This is a local ring.

Definition 60 (Rational Maps).

Let Y↪𝔸n. Then a rational map is a map f=(f1,…,fn), where fi∈k⁡(X) such that f⁡(x)∈Y whenever f1,…,fn are regular at x. Given this, we have a pullback homomorphism f∗:k⁡[Y]→k⁡(X).

Proposition 61.

We will see that f∗ is injective if and only if f⁡(X) is dense in Y. If so, we induce a field extension k⁡(Y)↪k⁡(X). Additionally, the other way around, given α:k⁡(Y)↪k⁡(X), we can construct a dominant rational map f:X⇢Y such that α=f∗.

Proof.

Take f=(f1,…,fn)=(α⁡(y1),…,α⁡(yn)). □

Definition 62 (Birational Equivalence).

X and Y are birationally equivalent if k⁡(X)≃k⁡(Y), or equivalently, there exist dominant rational maps f,g:X⇠⇢Y such that f∘g,g∘f are the identity wherever they are defined. In particular, X is rational if it is birational to an affine space.

We will skip projective and quasiprojective varieties and return to them next week.

3.5.2 Dimension

Let X⊂𝔸N. What is the dimension of X? Here are some ideas:

  • •

    “Maximal number of independent parameters.” To make this rigorous, we define the dimension of X to be the transcendence degree of k⁡(X);

  • •

    “Maximum possible dimension of a subspace +1.” Convert this into algebra and we define the dimension of X to be the Krull dimension of k⁡[X].

Theorem 63.

If X is an affine variety, then the two definitions of dimension agree.

Proof.

By Noether’s Normalization Lemma, k⁡[X] is integral over its polynomial subalgebra k⁡[y1,…,yn]=k⁡[𝔸n]. Then dim⁢X=trdeg⁢k⁢(X)=trdeg⁢k⁢(y1,…,yn)=n=dim⁢𝔸n.

First we show that the Krull dimension of k⁡[y1,…,yn]=n, and then we show that Krull dimension is preserved by integral extensions.

To see the second claim, let p1⊊⋯⊊pr⊊k⁡[X] be a chain of prime ideals. Then p1∩R⊂⋯⊂pr∩R⊂R is a chain of prime ideals. As an exercise, show that pi∩R≠pi+1∩R. Now let q1⊂⋯⁢qr⊂R be a chain of prime ideals in R. Then we use a stronger going-up theorem to lift the whole chain to k⁡[X]. If the qi are different, then the lifted ideals are different. Thus the Krull dimensions are equal. □

4 Projective Space

4.1 Lecture 8 (Feb 21, given by Luca Schaffler)

We will work over an algebraically closed field.

Definition 64 (Projective Space).

Let n∈ℤ be positive. Then the projective space ℙn is defined as ℙn=(kn+1∖{0})/Gm, which is the set of lines in kn+1 through the origin.18

In ℙn, we denote the equivalence class of (z0,…,zn) by (z0:⋯:zn) in homogeneous coordinates. Observe that at least one of z0,…,zn is nonzero.

Definition 65.

We say that a polynomial f∈k⁡[z0,…,zn] vanishes at a point ξ∈ℙn if f⁡(x0,…,xn) for every choice of homogeneous coordinates ξ=(x0:⋯:xn). In this case, we simply write f⁡(ξ)=0.

Definition 66 (Homogeneous Polynomial).

A polynomial f∈k⁡[z0,…,zn] is homogeneous if every monomial term in f has the same degree.

Example 67.

z03+z0⁢z12+z12⁢z2 is homogeneous with degree 3.

Remark 68.

Every polynomial f∈k⁡[z0,…,zn] can be decomposed as the sum of its homogeneous components, i.e. k⁡[x1,…,xn] is a graded algebra.

Proposition 69.

If f∈k⁡[z0,…,zn] vanishes at p∈ℙn, then all of its homogeneous components vanish at p.

Proof.

Let f=∑d≥0fd and p=(x0:⋯:xn). Then for all λ∈k∗ we have

Because k is infinite, then fd⁢(x0,…,xn)=0 for all d≥0. Therefore all fd vanish at p. □

Remark 70.

With the same assumptions as the previous proposition, the above proof also implies that f0=0.

Definition 71 (Closed Set).

Let X⊂ℙn is called closed if X is the vanishing set of some set of polynomials f1,…,fr∈k⁡[x0,…,zn].

By the previous proposition, it is not restrictive to assume that f1,…,fr are homogeneous.

Definition 72 (Homogeneous Ideal).

An ideal I⊂k⁡[z0,…,zn] is called homogeneous if it is generated by homogeneous polynomials. Equivalently, I is closed under taking homogeneous parts.

Definition 73 (Vanishing Set of Ideal).

Let U⊂k⁡[z0,…,zn] be an ideal. Define V⁡(I)={p∈ℙn|f⁡(p)=0⁢ for all f∈I }.

Remark 74.

If I⊂k⁡[z0,…,zn] is a homogeneous ideal, then by the Hilbert Basis Theorem, I=(f1,…,fr), where each fi can be assumed to be homogeneous.

Definition 75 (Zariski Topology on ℙn).

The closed subsets of ℙn are V⁡(I), where I⊂k⁡[z0,…,zn] is a homogeneous ideal. This forms a topology. If X⊂ℙn is closed, then the Zariski topology on X is the one induced by ℙn.

4.1.1 Affine Constructions

We will now consider a useful construction.

Definition 76 (Affine Cone).

Let I⊂k⁡[z0,…,zn] be a homogeneous ideal. We can conside the vanishing set V⁡(I)⊂ℙn or the affine cone Va⁢(I)⊂𝔸n+1.

Example 77.

Consider V⁡(z0)⊂ℙn, which is a point. Then Va⁢(z0) is a line.

Remark 78.

V⁡(I)=(Va⁢(I)∖{0})/Gm.

Remark 79.

I⁢(V⁡(I))=I⁢(Va⁢(I)).

Remark 80.

In the affine cone construction, then something happens: If I=k⁡[z0,…,zn], then V⁡(I)=∅ and Va⁢(I)=∅. Let I=(z0,…,zn). Then V⁡(I)=∅ but Va⁢(I)={0}. The map Va⁢(I)↦V⁡(I) fails to be injective exactly in this case.

Proposition 81.

Let I⊂k⁡[z0,…,zn] be a homogeneous ideal. Then the following are equivalent:

  1. 1.

    V⁡(I)=∅;

  2. 2.

    I⊃(z0,…,zn);

  3. 3.

    There exists s∈ℤ>0 such that I⊃Is and V⁡(Is)=∅ but Va⁢(I)={0}.

Proof.

First we prove (1)⇔(2). This is because V⁡(I)=∅ if and only if Va⁢(I)⊂{0} if and only if I=I⁢(Va⁢(I))⊃I⁡(0)=(z0,…,zn). To show that (3)⇒(2), note that Is⊂I, so (z0,…,zn)=Is⊂I.

To see that (2)⇒(3), then for all i=0,…,n, there exists pi∈ℤ>0 such that zipi∈I. Let p=max⁡{pi}. Then if s>(n+1)⁢(p−1), Is⊂I. □

Definition 82.

In ℙn consider the open subset 𝔸in={(z0:⋯:zn)∈ℙn|zi≠0}. Note that there is an identification 𝔸in→𝔸n given by (z0:⋯:zn)→(z0zi,…,znzi). Moreover, for all X⊂ℙn closed, we can define Xi≔X∩𝔸in.

Explicitly, if X=V⁡(F1,…,Fr), then Xi=V⁡(f1(i),…,fr(i)), where fji⁢(x0,…,xn)=Fj⁢(x0,…,1,…,xn). this is known as de-homogeneization.

Conversely, every closed Y⊂𝔸n defines canonically Y¯⊂ℙn where we identify 𝔸n with 𝔸0n. The equations for Y¯ are given as follows. If Y=Va⁢(f1,…,fr), define Fj⁢(z0,…,zn)=z0deg⁢fj⁢fj⁢(z1z0,…,znz0), which is known as homogeneization.

Proposition 83.

Y¯=V⁡(F1,…,Fr).

Remark 84.

X is not nexessarily the closure of its affine patches. For example, consider a line at infinity. It is lost during the procedure.

Remark 85.

Let F be a homogeneous polynomial. Then F=zideg⁢F⁢f(i).

4.1.2 Irreducibility

Definition 86.

A closed set X⊂ℙn is irreducible if X cannot be written as a union X=X1∪X2, where X1,X2 are proper closed subsets.

Remark 87.

We have a decomposition into irreducible components like in the affine case. Also, X is irreducible if and only if I⁡(X) is prime.

4.1.3 Projective and Quasiprojective Varieties

Definition 88 (Projective Variety).

X⊂ℙn closed is called a projective variety.

Definition 89 (Quasiprojective Variety).

A quasiprojective variety is an open subset of a projective variety.

Example 90.

Affine and projective varieties are both quasiprojective.

4.2 Lecture 9 (Feb 26)

Let X⊂ℙn be a quasiprojective variety. We need to define a regular function on X. We define k⁡(ℙn)={f⁡(z0,…,zn)g⁡(z0,…,zn)|f,g∈k⁢[z0,…,zn]d⁢ for some d }, where z0,…,zn are homogeneous coordinates on ℙn.

Remark 91.

In an affine chart, 𝔸0n⊂ℙn given by z0≠0, we can write

We can reverse this process, so k⁡(ℙn)=k⁡(𝔸n).

If f∈k⁡(ℙn), f=pq, then f defines a function in some open neighborhood of x∈X.

Definition 92 (Regular at a Point).

We call this function f regular at x∈X.

Definition 93 (Regular Function).

A function f:X→k is regular on X if for all x∈X, f can be written as f=pq in some open neighborhood of x∈X with p,q∈k⁡[z0,…,zd] and q⁡(x)≠0.

Remark 94.

We write k⁡[X] for the set of regular functions on X.

Remark 95.

If X is affine, then k⁡[X]=k⁡[x1,…,xn]/I⁡(X).

Remark 96.

If X is projective, then k⁡[X]=k.

Remark 97.

There are examples of quasi-projective varieties X such that k⁡[X] is not finitely generated.19

Definition 98.

A map X→𝔸m is regular if f=(f1,…,fm) and fi∈k⁡[X].

Definition 99.

A map f:X→Y↪ℙm quasiprojective varieties is called regular if for all x∈X we can choose an affine chart 𝔸im⊂ℙm containing f⁡(x) and an open set U∋x such that f⁡(U)⊂𝔸im and the induced map f|U is regular.

Remark 100.

This definition does not depend on the choice of affine chart containing f⁡(x).

Definition 101.

Quasi-projective varieties X,Y are isomorphic if there exists regular maps f:X→Y and g:Y→X such that f=g−1.

Lemma 102.

Let X be a quasi-projective variety. Then every x∈X has an affine neighborhood.

Proof.

Let X⊂ℙn and x∈𝔸in. Then X∩𝔸in=X¯∩𝔸in−(X¯∖X)∩𝔸in, so x∈Y−Z⊂𝔸n, where Y,Z are both affine. Choose 0≠F∈I⁡(Z)⊂k⁡[Y] and set u=Y−V⁡((F))=:D⁡(F). We will show that D⁡(F) is affine, called a principal open set. Principal open sets form a basis of the Zariski topology.

To prove this, note that Y=(G1=⋯=Gs=0)⊂𝔸n. Define Z≔(G1=⋯⁢Gs=F⁢yn+1=0)⊂𝔸n+1. Then Z is affine and isomorphic to D⁡(F). Simply take the last coordinate to be 1/F, which is possible because F≠0 on Z. □

Definition 103.

A regular map f:X→Y of quasi-projective varieties is called finite if all y∈Y have an affine neighborhood such that f−1⁢(V)=U is also affine and the induced map f|U is a finite map of affine varieties.

Proposition 104.

If X,Y are both affine, then this agrees with the previous definition.

Proof.

Suppose f:X→Y is a map of affine varieties. Let k⁡[X]=B,k⁡[Y]=A and the pullback be f∗:A→B. For all y∈Y, there exists an affine y∈V⊂Y such that U=f−1⁢(V)⊂X is affine and f|U is finite.

Choose a principal open set y∈D⁡(F)⊂V⊂Y for some f∈A. Then f−1⁢(D⁡(F))⊂U⊂X. We see that f−1⁢(D⁡(F))=D⁡(f∗⁢F), so it is affine.

Now we show that f|D⁡(f∗⁢F) is finite. To prove this claim, note that k⁡[U] is a finite k⁡[V]-module via f∗, so we show that k⁡[U]⁢[1f∗⁢F] is a finite k⁡[V]⁢[1F]-module. Simply take the basis to be the basis of k⁡[U] over k⁡[V].

We need to show that B is a finite A-module via f∗. We know that Y is covered by principal open sets D⁡(Fα) such that k⁡[D⁡(f∗⁢F)] is a finite k⁡[D⁡(F)]-module via f∗. We see that k⁡[D⁡(F)]=A⁡[1F], so k⁡[D⁡(f∗⁢F)]=B⁡[1f∗⁢F]. Also, Y is covered by D⁡(Fα) if and only if (Fα)=A. Thus we can write 1=∑hα⁢Fα. In particular, we need only finitely many D⁡(Fα).

We know that B⁡[1f∗⁢Fα] is a finite A⁡[1Fα]-module. Choose a basis of the form ωα,i∈B. We will show that the ωα,i form a basis of B over A. Choose b∈B. Then for all α we can write b=∑ωα,i⁢ai,αFαnα. We can still write 1=∑Hα⁢Fαnα.

We write b=b⁢∑Hα⁢Fαnα=∑Hα⁢(∑iωα,i⁢ai,α). □

4.3 Lecture 10 (Feb 28)

Let f:X→Y be a morphism of quasiprojective varieties. Then f is locally given by a polynomial map f=(f1,…,fm) where fi=pi⁢(x0,…,xn)qi⁢(x0,…,xn) and fi is of degree 0.

Homogeneizing, we obtain that f is given by f⁡(x)=[F0:⋯:Fm], where Fi⁢(x)∈k⁢[x0,…,xn]d for some d. Also, for every x∈X, there exists a presentation [f0:⋯:Fm] such that at least one Fi⁢(x)≠0.

Proposition 105.

The above two notions are equivalent.

Proof.

Suppose f regular at x and f⁡(x)∈𝔸0m. Then f⁡(x)=[1:p1/q1:⋯:pm/qm], so we can clear denominators to get f=[q1⁢⋯⁢qm:⋯:⋯]. □

Example 106.

Consider the projection from ℙℓ⊂ℙn. On the affine cones, this is just a projection, so the projection is given by [x0:⋯:xn]↦[xℓ+1:⋯:xn]. In this case the Fi are just coordinates. This gives a regular map ℙn∖ℙℓ→Pn−ℓ−1.

Example 107.

Consider the d-th Veronese embedding ℙn→vdℙN given by x↦[Fi⁢(x)] where Fi runs through monomials of degree d and N+1=(n+dd). This is regular everywhere because one of the powers is nonzero. In fact, vd is an embedding.

An example of this is the rational normal curve.

Theorem 108.

Let X⊂ℙn be an irreducible projective variety. Then k⁡[X]=k.

Proof.

We will deduce this from another theorem, which is given below. Using the theorem, take f∈k⁡[X]. Then f is a morphism X→𝔸1↪ℙ1. If f⁡(X) is dense in 𝔸1, then it is dense in ℙ1. By the next theorem, f⁡(X)=ℙ1, which contradicts the fact that f⁡(X)⊂𝔸1.

If f⁡(X) is not dense in 𝔸1, then f⁡(X)={p1,…,pr}. However, X is irreducible, so f⁡(X)=p (consider the fiber above each point). □

Theorem 109.

Let X be a projective variety and f:X→Y. Then f⁡(X)⊂Y is closed.

Proof.

To prove this, we prove the main theorem of elimination theory. Now we consider the map f:X→Y. Then consider the graph Γf:X→X×Y given by x→(x,f⁡(x)). We show that Γ=Γf⁢(X) is closed for all X,Y. Using this, we reduce to the following:

Theorem 110.

Let X⊂X×Y be closed. Then π2⁢(Z)⊂Y is closed.

This motivates the following definition:

Definition 111.

A variety is called proper if for every variety Y and closed subvariety Z⊂X×Y, π2⁢(X)⊂Y is closed.

Next we cover Y by affine open sets Yi. Then Z=⋃z∩(X×Yi)=Zi and Zi is closed in X×Y. Then π2⁢(X)=∪π2⁢(Zi). It is enough to show that the projections of the zi are closed.

Now we have X⊂ℙn projective and Y⊂𝔸m affine. Then Z is defined by equations gi⁢(u,y) homogeneous in x and arbitrary in y. The second projection of Z is precisely the locus T from the main theorem of elimination theory, which is closed. □

Theorem 112 (Main Theorem of Elimination Theory).

Let gi⁢(u,y) be polynomials homogeneous in u=(u0,…,un) and arbitrary in y=(y1,…,ym). Let T={y0∈𝔸m|gi⁢(u,y0)=0⁢ has a nonzero solution }. Then T is closed.

Proof.

Recall gi⁢(u,y0)=0 has a nonzero solution if and only if (g1⁢(u,y0),…,gt⁢(u,y0))⊉Is for all s. Then T=∩n≥1Ts where Ts={y0∈𝔸m|(g1⁢(u,y0),…⁣gt⁢(u,y0))⊉Is}. It suffices to check that Ts is closed, or that the complement is open.

Indeed, we see that for all monomials Nα∈Is, then Mα=∑gi⁢(u,y0)⁢Fi,α⁢(u). Let Ni,β be all monomials of degree s−deg⁢gi. Thus Mα∈Is is in the linear span of gi⁢(u,y0)⋅Ni,β. Therefore, gi⁢(u,y0)⋅Ni,βspan the vector space of all degree s polynomials, which means that at least one of the maximal minors is nonzero. Therefore the complement of Ts is a union of open sets and is thus open. □

4.3.1 Products

We discuss products of projective spaces. To do this, we consider the Segre embedding ℙn×ℙm→ℙN, where N+1=(n+1)⁢(m+1), given by [x0:⋯:xn],[y0:⋯:ym]↦[xi⁢yj]i⁢j. Note that at least one xi and at least one yj are nonzero, so this is well-defined.

Alternately, we present this as [x],[y]↦[x⁢yT], which gives a linear map ψ:km+1→kn+1 with kernel y⟂ and image x. Therefore, the Segre embedding is injective. Matrices of this form are precisely the rank 1 matrices, so the image is defined by equations the 2×2 minors of the matrix. In particular, ℙn×ℙm is a projective variety. This endows ℙn×ℙm with the structure of a projective variety.

Remark 113.

In the charts 𝔸0n,𝔸0m, we see s⁡([1:x1:⋯:xn],[1:y1:⋯:ym])=[1:x1:⋯:xn:y1:⋯:ym:ci⁢yj]⊂𝔸0N. This image is isomorphic to 𝔸n+m. Therefore this agrees on the charts with a product of affine spaces.

If X⊂ℙn,Y⊂ℙm are projective varieties, then X×Y⊂ℙn×ℙm is a projective variety.

To define closed subvarieties in ℙn×ℙm (or ℙn×𝔸m) without referring to the ambient projective space ℙN, we do the following:

Theorem 114.

The closed subvarieties of ℙn×ℙm are given by multihomogeneous (homogeneous in both variables) polynomials gi⁢(u,v)=0. For ℙn×𝔸m, just remove the assumption that gi are homoegeneous in v.

Proof.

Consider the image under the Segre embedding. Then we see that Z=(Fα⁢(wi⁢j)=0) os the same as Fα⁢(xi,yj)=0 homogeneous in X and Y of the same degree. If the gi are not of the same degree in u,v (s>t), we can multiply by vjs−t to obtain equivalent equations. □

4.4 Lecture 11 (Mar 5)

We continue our discussion of dimension (from Valentine’s Day). Let X⊂ℙn be an irreducible quasi-projective variety. Then we define the local ring 𝒪X={f∈k⁡(ℙn)∣f⁢ is regular at x∈X }. Then we can define k⁡(X)=𝒪X/mX where mX is the maximal ideal of functions that vanish on X.

Then if U⊂X is open, k⁡(U)=k⁡(X). Thus we can assume X is affine. Then we see that X⊂𝔸n⊂ℙn is the vanishing X=V⁡(𝔭) and 𝒪X=k[x!,…,xn]𝔭. Then mX=𝔭⁢R𝔭, and this agrees with the old definition by a “qualifying exam problem.”

Then, if we define dim⁢X≔tr.deg⁢k⁢(X), it is easy to see that dim⁢X=dim⁢U for an open subset U⊂X.

Theorem 115.

If X is affine and irreducible, then dim⁢X=Kr.dim⁢k⁢[X].

Proof.

We use Noether normalization. Then there exists k⁡[x1,…,xn]⊂k⁡[X], which is an integral extension. First it is easy to see that dim⁢X=dim⁢𝔸n=n because the field extension is algebraic.

To see that the Krull dimensions are equal, any chain q0⊂q1⊊⋯⊊qr⊊k⁡[x1,…,xn] gives a chain p0⊊⋯⁢pr⊊k⁡[X] by going-up. However, all pi are distinct, so their intersections with k⁡[x1,…,xn] are distinct (quotient by pi, then take the field of fractions of the smaller ring).

Now we show that the Krull dimension of k⁡[x1,…,xn] is n. Clearly it is at least n because we have the following chain of prime ideals: (0)⊂(x1)⊂⋯⊂(x1,…,xn). Take a chain p0⊊⋯⊊ps. Then V⁡(ps)⊊⋯⊊V⁡(p0)⊊𝔸n. We prove a lemma which implies what we want. □

Lemma 116.

Let X⊂Y be irreducible affine varieties. Then dim⁢X≤dim⁢Y. If X≠Y, then the inequality is strict.

Proof.

Let X⊊Y⊊𝔸N and let n=dim⁢Y. Then given t1|Y,…,tN|, any n+1 are algebraically independent. However, t1|X,…,tN|X generate k⁡[X] and therefore k⁡(X), so dim⁢X≤n.

Suppose dim⁢X=dim⁢Y=n. Then some n coordinates t1|X,…,tn|X are algebraically independent in k⁡(X). Thus t1|Y,…,t1|Y are also algebraically independent. Choose 0≠u∈k⁡[Y] such that u|X=0 Then there is a relation a0⁢(t1,…,tn)⁢um+⋯+am⁢(t1,…,tn) which vanishes along Y. We can also assume that am≠0 along Y. However, it becomes 0 when restricted to X because u|X=0. □

We use the Krull Principal Ideal Theorem:

Theorem 117 (Krull Principal Ideal).

If X is affine, irreducible, and f∈k⁡[X], then all irreducible components of V⁡(f)⊂X have codimension 1.

Corollary 118.

In 𝔸n, irreducible hypersurfaces correspond exactly to irreducible polynomials f∈k⁡[x1,…,xn].

Proof.

Let f∈k⁡[x1,…,xn] be irreducible. Then because k⁡[x1,…,xn] is a UFD, (f) is prime, so V⁡(f) is irreducible and has the correct dimension by PIT.

In the other direction, let X⊂𝔸n be an irreducible hypersurface. Then choose f∈k⁡[x1,…,xn] such that f|X=0. Factor f=f1⁢⋯⁢fr into irreducibles. Then X⊂V⁡(fi) because V⁡(fi) are irreducible hypersurfaces. Because they have the same dimension, they are equal. □

Corollary 119.

Let X⊂ℙn be projective and F∈k⁡[x0,…,xn] be homogeneous of positive degree d. Then X⊂V⁡(F) is non-empty and all irreducible components have codimension 1.

Proof.

Work in the charts. Then use the PIT on the irreducible affine components. To show that the intersections in the affine charts are nonempty, pass to the affine cone. By the PIT, the intersection has codimension 1 and is non-empty because it contains the origin. Then the same is true for X. □

Remark 120.

This implies that if X,Y⊂ℙ2 are projective curves, X∩Y are nonempty. This implies that every irreducible curve of degree at least 3 has inflection points.

Theorem 121.

Let f:X→Y be a regular surjective map of irreducible varieties. Suppose dim⁢X=n,dim⁢Y=m. Then:

  1. 1.

    dim⁢F≥n−m for every component F of every fiber f−1⁢(y).

  2. 2.

    There exists U⊂Y open, nonempty such that dim⁢F=n−m for all y∈U.

  3. 3.

    Sets Yk={y∈Y∣dim⁢f−1⁢(y)≥k} are closed.

Proof.

To prove the first part, take y∈Y. Then we can take. We can assume X,Y are affine by taking affine charts. Choose f1≠0 such that f1⁢(y)=0. Then dim⁡(f1=0)=m−1 by the PIT. Choose f2|D≠0 (f2⁢(y)=0) for all irreducible components D of V⁡(f1). Then dim⁡(f1=f2=0)=m−2. Then y∈(f1=⋯=fm=0), which is a finite set of points. Now we pass from Y to D⁡(F)⊂Y where F⁡(y~)=0 for all y∈(f1=⋯=fm=0)∖{y}. With this new Y, y=(f1,=⋯=fm=0) where m=dim⁢Y, we see that f−1⁢(y)=V⁡(f∗⁢(f1),…,f∗⁢(fm)). By the PIT, the dimension is at least n−m. □

4.5 Lecture 12 (Mar 7)

Last time we discussed Krull’s Principal Ideal Theorem and began the proof of the theorem on dimension of fibers.

Corollary 122.

Let X be irreducible of dimension n and f1,…,fr∈k⁡[X]. Then every irreducible component of V⁡(f1,…,fr)⊂X has dimension at least n−r.

Sketch.

We induct on r. If r=1, we use the P.I.T. The rest is left as an exercise. □

Conclusion of proof of Theorem 121.

To prove the second part, we do something similar to the below theorem. We may assume X,Y are affine and f:X→Y dominant. Then we can decompose f∗:k⁡[Y]↪k⁡[Y]⁢[z1,…,zr]↪k⁡[X]. This corresponds to a map Y⁢←π1⁢Y×𝔸r⁢←𝑔⁢X. Then f−1⁢(y)=g−1⁢(Y×𝔸r).

We will show that there exists D⁡(F)⊂Y such that for all y∈D⁡(F), p1|f−1⁢(y),…,pℓ|f−1⁢(y) are algebraic over z1|f−1⁢(y),…,zr|f−1⁢(y), so every component of f−1⁢(y) has dimension at most r. By part 1, the dimension is at least r, so it must equal r.

To prove the claim, write the equations of algebraic dependence Fi⁢(pi,z1,…,zr,q1,…,qs)=0 with pi appearing in the polynomial. Then restrict to a specific y and show that not all coefficients become zero (which is because there is an open set where the product of all coefficients does not vanish).

Now we prove the third part, we induct on the dimension of Y. If Y is a point, there is nothing to prove. Also, by part (1), Yn−m=Y is closed. By part (2), there exists U⊂Y open where dim⁢f−1⁢(y)=n−m for all y∈U. Then Yj for j>n−m is contained in f−1⁢(Y∖U)=Y′. Now consider f restricted to f−1⁢(Y′). Because dim⁢Y′<dim⁢Y, Yj is closed. □

Theorem 123.

Let f:X→Y be a regular map of quasi-projective varieties. Then f⁡(X) contains an open subset U⊂f⁡(X)¯.

Proof.

We may assume that f is dominant and that Y is irreducible by considering the irreducible components of Y. Considering the irreducible components of X, we may assume that X is irreducible. Passing to an affine chart U⊂Y and its preimage, we may assume that Y is affine and irreducible. Take any affine chart in the preimage of U and we may assume that X is affine.

Thus we have X⁢→𝑋⁢Y and f∗:k⁡[Y]↪k⁡[X] a map of domains. Then we can decompose f∗:k⁡[Y]↪k⁡[Y]⁢[z1,…,zr]↪k⁡[X]. This corresponds to a map Y⁢←π1⁢Y×𝔸r⁢←𝑔⁢X. If α is integral, then g (and thus f) is surjective.

We use a trick that there exists a principal affine D⁡(F)⊂Y×𝔸r such that g|g−1⁢(D⁡(F))→D⁡(F) is finite. Let p1,…,pℓ∈k⁡[X] be generators of the algebra. Then they are algebraic over k⁡[Y]⁢[z1,…,zr] so we can write a0⁢pimi+a1⁢pimi−1+⋯+ami=0, where ai∈k⁡[Y]⁢[z1,…,zr]. If we invert F=a01⁢a02⁢⋯⁢a0ℓ, then pi are all integral over k⁡[Y]⁢[z1⁣…,zr]⁢[1/F]=k⁡[D⁡(F)].

Then the image of g contains D⁡(F) because finite maps are surjective. Then D⁡(F)↪Y×𝔸r→Y. We find U⊂Y such that U⊂π1⁢(D⁡(F)). Note that F=∑bi1⁢…⁢ir⁢z1i1⁢⋯⁢zrir. Define U=D⁡(bi1⁢…⁢ir) for some coefficient. If y∈U, then F⁡(y,z1,…,zr)≠0 for some choice of y,z1,…,zr. Thus (y,z1,…,zr)∈D⁡(F). □

Theorem 124.

Let f:X→Y be a regular map between projective varieties such that f⁡(X)=Y and suppose Y is irreducible. Suppose f−1⁢(y) is irreducible of the same dimension (dim⁢X−dim⁢Y). Then X is irreducible.

Proof.

Decompose X=∪Xi. Then f⁡(X1)=⋯=f⁡(Xs)=Y and f⁡(Xi)⊊Y for all i>s. Define fi=f|xi. Then there exists Ui⊂Y open such that f−1⁢(y) has smallest dimension ni. Set U=∩i=1sUi∖∪j>sf⁡(xj). Fix y0∈U. Then f−1⁢(y0)⊂∪fi−1⁢(y0). WLOG we see that f−1⁢(y0)=f1−1⁢(y0), which implies n=n1. Then for y∈Y, we see that f1−1⁢(y)⊂f1−1⁢(y). Because Y is irreducible, X=X1. □

Jenia apologizes that this is too abstract and wants to calculate some actual examples.

Theorem 125.

Every cubic surface contains a line.

Proof.

We postulate the existence of the Grassmanian. One property of Gr(k.n) is that the incidence variety ℒ={p∈ℙn−1,L∈Gr⁡(k,n)∣p∈L}⊂ℙn−1×Gr⁡(k,n) is closed. It is easy to see that ℒ is irreducible because p2:ℒ→Gr⁡(k,n) and all fibers are ℙk−1.

Define Cub=ℙ19=ℙ⁡(Sym3⁢(k4)) and W={p∈ℙ3,S∈Cub∣p∈S}. Clearly W is a projective variety. Define Z={L∈Gr⁡(2,4),S∈Cub∣L⊂S}⊂Gr⁡(2,4)×Cub. This is also a closed set (construct a map such that Z is the locus of points with 1-dimensional fibers). Therefore Z is projective.

Now consider π2⁢(Z)⊂Cub. Becuase the Grassmanian is projective, it is proper. Then π2⁢(Z) is closed. This is the set of cubic surfaces containing at least one line. If the image is not all cubic surfaces, we may use the theorem on the dimension of the fibers. We see that each cubic contains a positive-dimensional set of lines. But the Clebsch cubic surface contains only 27 lines, which is a contradiction. □

5 Local Properties

5.1 Lecture 13 (Mar 19)

We begin our discussion of local properties. Let p∈X be a point on a quasi-projective variety. We can assume X is affine and k⁡[X]=A. Then mp⊂A is a maximal ideal. The local ring is 𝒪p=Amp, which is independent of the affine chart. If X is irreducible, then 𝒪p⊂k⁡(X). If X is not irreducible, then A has zero-divisors and localization is harder to define.

5.1.1 Tangent space

Let X⊂𝔸n. Then X=V⁡(I),I=(F1,…,Fs),p=(x10,…,xn0).

Definition 126 (Tangent Line).

We say that a line L passing through p is tangent to X at p if the multiplicity of L∩X is at least 2.

Definition 127 (Tangent space, ambient version).

The tangent space Tp⁢X is the locus of all lines tangent to X at p.

Definition 128 (Multiplicity).

Let L be defined parametrically. Then I|L⊂k⁡[t] is a principal ideal (f). Then f=tm⁢u, where u⁡(0)=0. The multiplicity of intersection is m.

Example 129.

Let X=(y=x2) the parabola. Then if L={x=x0+a⁢ty=y0+a⁢t we see that f⁡(t)=(b−2⁢a⁢x0)⁢t+O⁡(t2). Thus the line is tangent if b=2⁢a⁢x0, or b/a=2⁢x0.

Example 130.

Let X=(y2=x3) and p=x0. Then f⁡(t)=(b⁢t)2−(a⁢t)3=t2⁢(b2−a3⁢t), so every line is a tangent line through the origin, and the tangent space is 𝔸2.

Again let I=(F1,…,Fs) and p=(x10,…,xn0). Then

where ∑∂Fℓ∂xi⁢(p)⁢(xi−xi0) is the linearization of dp⁢Fℓ. Then Fℓ|L=∂Fℓ∂xi⁢(p)⁢ai⁢t+h.o.t. Thus L is a tangent line if and only if

for all ℓ=1,…,s. Thus the defining equations for Tp⁢X are

Note that Tp⁢X is an affine subspace of 𝔸n. One can also view the linear subspace parallel to Tp⁢X as the tangent space.

Theorem 131.

There is a canonical isomorphism Tp∗≃mp/mp2, where either we take mp⊂A=k⁡[X] or mp⊂𝒪p. This is known as the Zariski cotangent space.

Proof.

Let g∈mp. Then g=G+I for some G∈k⁡[x1,…,xn]. Then dp⁢G=∑∂G∂xi⁢(xi−xi0), but for all F∈I, dp⁢F|Tp⁢X=0. Therefore we can define dp⁢g=dp⁢G|Tp⁢X. This gives a map dp:mp→Tp∗. Then dp is clearly k-linear and surjective (linear function is its own differential). Then mp2⊂Ker⁢dp because mp2 is generated by (xi−xi0)⁢(xj−xj0), which vanish under the differential by Leibniz.

Thus dp induces a surjective map mp/mp2→Tp∗. We show that the kernel of dp is mp2. WLOG let p=(0,…,0). Suppose dp⁢g=0. Then g=G|X,G∈k⁡[x1,…,xn] and dp⁢G|Tp=0. Then Tp={dp⁢Fℓ=0∣ℓ=1,…,s}, so dp⁢G=λ1⁢dp⁢F1+⋯+λs⁢dp⁢Fs. Set G1=G−∑i=1sλi⁢Fi. Then g1|X=g and dp⁢G1=0, so G1∈(x1,…,xn)2. Therefore G1|X∈mp2. □

Corollary 132.

Tp and therefore its dimension is a local invariant of p∈X.

Remark 133.

Suppose X⊂ℙn is projective. Then X=(F1=⋯=Fs=0). Choose p∈X, and suppose WLOG that p∈𝔸0n, so p=(1:x10:⋯:xn0). Then Tp⁢X⊂𝔸0n is given by ∑i=1n∂Fℓ∂xi⁢(p)⁢(xi−xi0)=0 where ℓ=1,…,s.

We can take Tp⁢X¯⊂ℙn the projective tangent space. Recall Euler’s formula ∑i=0n∂Fℓ∂xi⁢(p)⁢xi0=(deg⁢Fℓ)⁢Fℓ⁢(p)=0. Therefore the projective tangent space is given by

Theorem 134.

Suppose X is irreducible. Then Xsm={p∈X∣dim⁢Tp⁢X=dim⁢X} is a non-empty open subset of “smooth” or “non-singular” points. The complement is known as the singular locus Xsing, and for all p∈Xsing, dim⁢Tp⁢X>dim⁢X.

Proof.

We assume X⊂𝔸N is affine. Then define T⁢X={(a,p)∈𝔸N×X∣a∈Tp⁢X}. This is called the tangent bundle (or fiber space). Its equations are F1,…,Fs the equations of X and the equations of tangency ∑∂Fℓ∂x⁢(x1,…,xn)⁢(yi−xi)=0. Thus T⁢X is algebraic.

Now the second projection π2:T⁢X→X has fiber the tangent space of X. By the theorem on the dimension of fibers, Xsm={p∈X∣dim⁢Tp⁢X=s}, where s is the minimum possible, is an open subset. It remains to be seen that s=n.

Note every irreducible algebraic variety is birational to a hypersurface. Also, we show that X is birational to Y, if and only if there exist U⊂X,V⊂Y open, nonempty such that U≃V.20

Given both claims, we can assume X⊂𝔸n+1 is a hypersurface. Then X=(F=0) for some irreducible F∈k⁡[x0,…,xn+1]. Then Tp⁢X is given by ∑i=0n∂F∂xi⁢(p)⁢(xi−xi0)=0. Then the dimension of the tangent space is n unless ∂F∂xi⁢(p)=0 for all i. This implies that ∂F∂xi∈(F) and thus the partial derivatives are zero. In characteristic 0, this implies F=0, which is a contradiction, and in positive characteristic, F is a pth power, so it is not irreducible. Thus we have a contradiction. □

Lemma 135.

Every irreducible variety is birational to a hypersurface.

Proof.

We work in characteristic zero. Choose a transcendence basis k⁡(x1,…,xn)⊂k⁡(X) a finite extension. This extension is separable, so we use the primitive element theorem. Then k⁡(x1,…,xn,y), where y has minimal polynomial F∈k⁡[x1,…,xn,y]. Thus k⁡(X)≃k⁡(H), where H=V⁡(F). □

If X is reducible, then p∈X is nonsingular if dim⁢Tp⁢X=max⁡(dim⁢Y) where Y∋p is an irreducible component of X. In fact, there is a theorem that if p lies on several irreducible components, then p is singular.

5.2 Lecture 14 (Mar 21)

Last time we considered local properties at a point. Today we will prove more theorems. Let p be a non-singular point of X.

Theorem 136.

𝒪p↪k⁡[[u1,…,un]].

Corollary 137.

𝒪p is an integral domain. Thus X has only one irreducible component passing through p.

Remark 138.

This is an analogue of the result from complex analysis that holomorphic functions are analytic and the map 𝒪U→ℂ⁡[[z]] is injective.

Definition 139.

u1,…,un∈mp are called local parameters if {ui} is a basis of mp/mp2. Equivalently, dp⁢u1,…,dp⁢un form a basis of Tp∨.

Fix u1,…,up∈mp a system of local parameters.

Lemma 140.

mp=(u1,…,un).

Proof.

Let U=(u1,…,un)⊂mp. Then mp⁢(mp/I)=mp/I, so by Nakayama’s lemma, mp/I=0. □

Definition 141.

A formal power series ϕ=∑k≥0ϕk⁢(u1,…,un) is called a Taylor series of f∈𝒪p if (f−∑k=0nϕk)∈mpn+1.

Lemma 142.

Every f∈𝒪p has a Taylor series.

Proof.

Note f=f⁡(p)+g1, where g1 vanishes at p, so g1∈mp. Then we can write g1=∑i=1nai⁢ui where ai∈𝒪p. Thus we can write ai=ai⁢(p)+ai¯, where ai¯∈mp. Then we can write g1=∑i=1nai⁢(p)⁢ui+∑ai¯⁢ui. Set the second sum to be g2∈mp2.

Now suppose gk∈mpk. Thus gk=∑ih1i⁢⋯⁢hki. Noting that hji=∑aj,ki⁢uk. Decompose this as above and then the claim follows by induction. □

Lemma 143.

A Taylor series is unique. Thus we have a morphism 𝒪p→k⁡[[u1,…,un]].

Lemma 144.

This morphism is injective.

Proof.

Take f∈𝒪p and suppose f has Taylor series 0. This happens if and only if f∈mpn for every n. We know that by Krull’s intersection theorem, if (R,m) is a Noetherian local ring and I⊂R is some ideal, then ∩n≥1In=0. Therefore f=0. □

Theorem 145 (Krull’s Intersection Theorem).

If (R,m) is a Noetherian local ring and I⊂R is some ideal, then ∩n≥1In=0.

Proof of Krull’s intersection Theorem.

Consider the Rees ring R~=R⊕I⊕I2⊕⋯ with graded multiplication. Then R~ is Noetherian. Consider N=∩n≥1In and N~=N⊕N⊕N⊕⋯⊂R~, which is an ideal. Thus N~ has a finite generating set. Thus all generators live in the first k components for some k. Then any n∈Nk+1 can be written n=∑ri~⁢ni where ri~∈I⊕I2⊕⋯. Then N=I⋅N. I is contained in a maximal ideal, so N=m⁢N, and therefore N=0 by Nakayama. □

Remark 146.

This leads to the Artin-Rees lemma.

Proof of Lemma 143.

Suppose f∈𝒪p has 2 different Taylor series. Equivalently, suppose 0 has a nontrivial Taylor series. Then ϕ0∈mp, so φ0=0. Then ϕ1=∑ai⁢ui∈mp2. However, u1,…,un are linearly independent mod mp2 by definition of local parameters. Thus a1=⋯=an=0. Thus ϕi=0.

Now ϕ2=∑ai⁢j⁢ui⁢uj∈mp3. We show that ϕk⁢(u1,…,un)∈mpk+1 if and only if ϕk is a trivial polynomial. If n=1, then mp=(u) and ϕk⁢(u)=α⁢uk∈(u)j+1. If α≠0, then (u)k=(u)k+1, which implies (u)=0 by Nakayama.

From the proof of Noether’s normalization, WLOG ϕk=u1k+other monomials We restrict to a hypersurface Y=V⁡(un)⊂X. Then all irreducible components of Y have codimension 1 by the PIT. We claim that Y is non-singular at p and u1|Y,…,un−1|Y are local parameters of Y at p.

Given the claim, ϕk|Y=ϕ⁡(u1|Y,…,un−1|Y,0) is a nontrivial polynomial. It contains (u1|Y)k. On the other hand, ϕk∈mpk+1, so ϕk|Y∈(mp,Y)k+1. By induction, this is a contradiction. □

Theorem 147.

Let u1,…,un∈mp⊂𝒪p be local parameters. Shrink X so that u1,…,un∈k⁡[X]. Then Xi=V⁡(ui) is nonsingular at p and {uj|xi}j≠i are local parameters of Xi. Also, ∩i=1nTp,Xi=0.

Proof.

Note ui∈I⁡(xi)⊂k⁡[X]. Then Tp⁢Xi={𝑑fp=0⁢ for all ⁢fp∈I⁡(xi)}⊂Li={dp⁢ui=0}⊂Tp⁢X. By the PIT, dimp⁢Xi=n−1. However, dim⁢Tp⁢Xi≤n−1. Thus xi is smooth and Tp⁢Xi=Li. Also, {dp⁢uj|Li}j≠i is a basis of Li∨. Therefore {uj|Xi}j≠i are local parameters on Xi. □

This completes the proof of Lemma 143.

5.2.1 Tangent Cone

Consider the cusp X={y2=x3},T0⁢X=𝔸2. Rescale x→t⁢x,y→t⁢y, so the equation becomes t2⁢y2=t3⁢x3, so y2=t⁢x3. Note that X≃Xt for t≠0. Then we set X0={y2=0}.

Let 0∈X⊂𝔸n. Then X=V⁡(I). For all f∈I we can write f=in⁡(f)+h.o.t.. Then define I0=in⁡(I)=(in⁡(f)∣f∈I). For example, in⁡(y2−x3)=(y2).

Definition 148.

The Tangent cone is defined to be V⁡(I0), at least if I0 is a radical ideal.

Theorem 149.

The tangent cone has the same dimension as the original variety. There is a family of varieties Xt≃X such that “limt→0Xt”=X0.

More precisely, there exists an affine variety 𝒳⊂𝔸n×𝔸t1 such that π2−1⁢(t)≃X for t≠0, π2⁢(1)=X, and π2−1⁢(0)=X0.

5.3 Lecture 15 (Mar 26)

We can consider lines (or curves on a variety X as maps ℙ1⇢X). Also, we can consider maps f:X⇢ℙ1. Then Y=f−1⁢(a)⊂X is a hypersurface.

Definition 150 (Locally Principal).

A hypersurface Y⊂X is called locally principal if it is locally given by 1 equation.

Definition 151 (Local Equations).

f1,…,fr∈𝒪p are called local equations of a subvariety Y⊂X if there exists and affine neighborhood U⊂X of p such that f1,…,fr∈k⁡[U] and I⁡(Y)=(f1,…,fr).

Lemma 152.

f1,…,fr are local equiations of Y at p if and only if ℒY,p⊂𝒪p is generated by f1,…,fr. Here ℒY,p={f∈𝒪p∣f|Y∩U=0⁢ for some neighborhood ⁢p∈U}. Equivalently, if X is already affine, I⁡(Y)⊂k⁡(X) and ℒY,p=I⁢(Y)mp.

Proof.

If f1,…,fr∈𝒪p are local equations, then after shrinking X to U, we have I⁡(Y)=(f1,…,fr)⊂k⁡[X]. Then ℒY,p=(f1,…,fr) because localization preserves generators.

Now suppose that ℒY,p=(f1,…,fr)⊂𝒪p. By shrinking, we can assume that f1,…,fr∈k⁡[X] and that X is affine. We know that I⁡(Y)=(g1,…,gs)⊂K⁡[X]. We can write gi=∑j=1rhi⁢j⁢fj in 𝒪p. We can shrink X: Choose a principal open set U=D⁡(w)⊂X such that hi⁢j∈k⁡[U]. We claim that in U, I⁡(Y∩U)=(g1,…,gs)⊂k⁡[U]=k⁡[X]⁢[1/w] is also generated by f1,…,fr. We know (g1,…,gs)⊂(f1,…,fr). Then fi|Y=0, so (f1,…,fr)⊂(g1,…,gs). □

Theorem 153 (Criterion for Smoothness).

Let X be a variety of dimension n. Then suppose p∈Y⊂X and Y is locally principal at p, as in ℒY,p=(g)⊂𝒪p. If Y is nonsingular at p, then so is X.

Proof.

Shrink X to be affine. Then I⁡(Y)=(g)⊂k⁡[X]. We see that I⁡(X)=(F1,…,Fs) and I⁡(Y)¯=(F1,…,Fs,G). Recall that Tp⁢X={dp⁢F1,⋯,fp⁢Fs=0} with dimension at least n. Then Tp⁢Y={dp⁢F1=⋯=dp⁢Fs=dp⁢G=0}. Then we see that Tp⁢Y has dimension n−1, so Tp⁢X must have dimension n, so X is nonsingular at p. □

Remark 154.

It is not enough to assume that Y=V⁡(g). For example, take X={z2=x⁢y} and Y=V⁡(x). Y is nonsingular at the origin but X is singular. In fact, this implies that Y⊂X is not locally principal, so I⁡(Y)=(x,z) needs at least 2 generators.

Theorem 155.

Suppose X is nonsingular at p and suppose p∈Y⊂X is an irreducible hypersurface. Then Y is locally principal at p.21

Theorem 156 (Equivalent Theorem).

Let p∈X be a nonsingular point. Then 𝒪p is a UFD.

Proof of Equivalence.

Suppose p∈Y⊂X where X is affine. Then consider f∈ℒY,p. This is a UFD, so we can factor f=f1⁢⋯⁢fr. We can assume that f∈k⁡[X] and f|Y=0. If Y is irreducible, then f1 vanishes on Y (up to renumbering). We will show that ℒY,p=(f1=g).

We know that Y⊂V⁡(g). We will sow that Y is the only irreducible component of V⁡(g) passing through p. Suppose Y′ is another component passing through p. Then there exists h|Y=h′|Y′=0, and h⁢h′|V⁡(g)=0. Thus g|(h⁢h′)k for some k. Thus g divides h or h′ in a UFD, which leads to a contradiction. This implies that Y=V⁡(g) is irreducible. Take s∈I⁡(Y). Then g|sk for some k. Because g is prime, it divides s. Thus s∈(g). In fact, I⁡(Y)=(g).

In the other direction, note that 𝒪p is Noetherian, which implies existence of prime factorization. We prove Euclid’s Lemma. Let g be irreducible. Let Y=V⁡(g). Then Y is irreducible at p. Take Y1⊂Y an irreducible component. By Krull, Y1 has codimension 1. Thus Y1 is locally principal, so Y1=V⁡(g1) after shrinking. We also know I⁡(Y1)=(g1). Then g|Y1=0, so g∈(g1). Thus g=g1⁢h1∈𝒪p, so h1 is a unit. Then (g)=(g1) and thus Y=Y1. Thus Y is irreducible and g is prime. □

Proof of Theorem 156.

The idea is to use the injection 𝒪p⊂k⁡[[T1,…,Tn]] where p∈X nonsingular and n=dim⁢X. We check that Euclid’s Lemma holds. We know that this holds in k⁡[[T1,…,Tn]]. Then if a,c∈𝒪p with a|c in 𝒪̂p. We also check that if a,b are coprime in 𝒪p then they are coprime in 𝒪̂p. We show that if I⊂𝒪p, then (I⁢𝒪̂p)∩𝒪p=I.

To prove the claim, let I⊂𝒪p where I=(f1,…,fs) and x∈I⁡(w⁢i⁢d⁢e⁢h⁢a⁢t⁢𝒪p)∩𝒪p. Then x=∑fi⁢αi, αi∈𝒪̂p. Then for all n≥1 we see that αi=ai(n)+ξi(n) where ai∈𝒪p and ξi∈m̂pn. Thus x=∑ai(n)⁢fi+∑ξi(n)⁢fi. We know ξ=x−a∈m̂pn∩𝒪p, so ξ∈mpn. Therefore x∈I+mpn for all n, and thus x∈I by Krull’s Intersection Theorem.

Given the claim, we see that c∈(a)⁢𝒪̂p. However, c∈𝒪p, so c∈(a). Now we show that gcd is preserved by the embedding. Suppose a=γ⁢α,b=γ⁢β where (α,β)∈𝒪̂p. Choose n such that α,β∉m̂pn. Then α=xn+un,β=yn+vn where xn,yn∈𝒪p and un,vn∈m̂pn. Because a⁢β−b⁢α=0, we see that a⁢yn−b⁢xn=a⁡(β−vn)−b⁡(α−un)=−a⁢vn+b⁢un∈(a,b)⁢m̂pn. By the claim, a⁢yn−b⁢xn∈(a,b)⁢mpn, so a⁢yn−b⁢xn=an+b⁢sn where tn,sn∈mpn. Bt algebraic manipulation, α⁡(yn−tn)=β⁡(xn+sn). Because α,β are coprime, then α∣xn+sn, so xn+sn=α⁢λ. Thus λ=1+h.o.t. is a unit. Therefore xn+sn∣α, so xn+sn∣a in 𝒪p. Then a=(xn+sn)⁢h and therefore h⁡(yn−tn)=b, so b=h⁡(yn−tn). Therefore h is a unit. In the power series ring, α∣a and a∣α, so γ is a unit. □

5.4 Lecture 16 (Mar 28)

We prove that if p∈X is smooth, then 𝒪p is a UFD. We will finish the proof in last time’s section.

Corollary 157.

Let f:X⇢Y is a rational map with X nonsingular, Y projective. Then the indeterminancy locus has codimension at least 2. In particular, if X is nonsingular, f is regular.

Proof.

We work near p∈X and assume Y=ℙn. Then f=(f0:⋯:fn) where fi∈k⁡(X). We can assume that fi∈𝒪p with no common factor. We show that V⁡(f0,…,fn)⊂X has codimension 2 near p.

Suppose Y⊂V⁡(f0,…,fn) and Y has codimension 1. Then ℒY,p=(h) because 𝒪p is a UFD. Then fi|Y=gi⁢h. This contradicts the fact that the fi have no common factor. □

5.4.1 Blowups

Example 158.

Consider ξ=[1:0:⋯:0]∈ℙn and take the map π:ℙn⇢ℙn−1 given by [x0:⋯:xn]↦[x1:⋯:xn]. We know π is regular at ℙn∖ξ. We want to resolve π:

Blξ⁢ℙnℙnℙn−1σπ

Consider the map ℙn×ℙn−1⊃Blξ⁢ℙn=Γπ¯, where Γπ={(x,y)∣y=π⁡(x)}. By construction, σ=π1 and the right arrow is π2. We also know that Blξ⁢ℙn∖σ−1⁢(ξ)≃ℙn∖ξ. Also, the blowup is irreducible.

Fix [y1:⋯:yn]∈ℙn−1. We know π−1⁢(y)⊃{[1:ty1:⋯:t⁢yn]}. Then the closure contains ξ,[y1:⋯:yn]. Therefore σ−1⁢(ξ)=ℙn−1 and is called the exceptional divisor of the blowup.

Now we find equations for the blowup. We know that ℙn×ℙn−1 is given by xi⁢yj=xj⁢yi. Then if x1=⋯=xn=0 (ξ), we get ℙn−1. If x1≠0, then [x1:⋅:xn]∼[y1:⋯:yn] and we can use equations for Γπ.

5.5 Lecture 17 (Apr 02)

We finish the discussion of blowups from last time. Let f:X→Y be a birational regular surjective morphism. Then f restricts to an isomorphism U≃V of open subsets. Using the theorem on dimension of the fibers, there exists an open W⊂Y such that f has finite fibers over points in W. Away from W, every component of every fiber is positive dimensional.

Definition 159 (Exceptional Locus).

Exc⁡(f)=f−1⁢(Y∖W) is the exceptional locus of f.

Theorem 160.

Suppose f:X→Y is a surjective birational regular map. Suppose f⁡(x)=y and assume that Y is smooth at y (more generally, Y is locally factorial at y, meaning that 𝒪y is a UFD). Assume that g=f−1 is not defined at y. Then there exists a subvariety Z⊂X of codimension 1 passing through x such that Z⊂Exc⁡(f).

Corollary 161.

If f:X→Y is birational, surjective, regular, and codim⁡(Exc⁡(f))≥2, then Y is singular and in fact not even locally factorial.

Proof.

Pass to an affine neighborhood of x at X, and assume that X⊂𝔸n with g=f−1. Then denote gi=g∗⁢(fi)∈k⁡(Y). Then g is not defined at y∈Y, so one of the gi, say g1 is not regular at y∈Y. Thus g1∉𝒪y.

If 𝒪y is a UFD we can write g1=u/v where u,v are coprime and v⁡(y)=0. We know that t1 is a regular function on X, where t1=f∗⁢(g∗⁢(t1))=f∗⁢(g1)=f∗⁢(u)f∗⁢(v).

Define Z=V⁡(f∗⁢v)⊂X, which is a hypersurface. We know f∗⁢(u)|Z=(t1⁢f∗⁢v)|Z=0. Thus f⁡(Z)⊂V⁡(u)∩V⁡(v). We know V⁡(u,v)⊂Y has codimension 2 at y becuase the local ring at y is a UFD. □

We observe that the blowup is a local construction. Take 𝔸x1,…,xnn and let ξ=0. Then Blξ=Γ¯π⊂𝔸n×ℙn−1⊂Blξ⁢ℙn. To find equations, we dehomogeneize to find xi⁢yj=xj if we take yi=1. Also, other equations are xj⁢yk=xk⁢yj. Also Blξ⁢𝔸n=⋃i=1nUi where ui={yi≠0}. Thus Ui≃𝔸y1,…,xi,…,ynn and there is a map Ui→𝔸n given by (y1,…,xi,…⁣yn)↦(xi⁢y1,…,xi,xi⁢yn).

Corollary 162.

The blowup of 𝔸n is smooth, covered by 𝔸n’s.

Definition 163 (Proper Transform).

Take X⊂𝔸n. Then σ−1⁢(X)=E∪σ−1⁢(X∖ξ)¯. the components are E≃ℙn and X~, which is called a proper transform of X.

Example 164.

Consider X=(x2=t⁢x1)⊂𝔸2. Then consider U1≃𝔸x1,y22 and U2=𝔸y1,y22/ Them if α:U1→𝔸2 is the covering map, α−1⁢(x)=x1⁢y2−t⁢x1=0, so either x1=0 or y2=t. On U21, we see x1↦x2⁢y1,y2↦y2, so β−1⁢(x)=x2−t⁢x2⁢y1=0, so either x2=0 or 1−t⁢y1=0.

Example 165.

Consider X=(x22−x13=0) in U1. Then α−1⁢(X)=(x1⁢y2)2−x13=0, so either x1=0 (E) or x1=y22 (X~).

5.5.1 Local Blowup II

Let ξ∈X be a smooth point and the dimension of X be n. We find the blowup of X at ξ.

Definition 166.

Choose local parameters u1,…,un∈mξ. Near ξ, a rational map X⇢ℙn given by x↦[u1⁢(x):⋯:un⁢(x)] is not regular only at ξ. Then Blξ⁢X=Γ¯π⊂X×ℙ1 which is irreducible, smooth and the map σ:Blξ⁢(X)→X is an isomorphism away from ξ with equations as before.

Theorem 167.

Let ζ∈X⊂𝔸n with X smooth. Then X~⊂Blξ⁢𝔸n is isomorphic to Blξ⁢X.

Theorem 168.

Blξ⁢X is independent of the choice of local parameters. For two choices of local parameters, there exists a unique isomorphism ψ that commutes with the σ.

5.6 Lecture 18 (Apr 04)

Recall that local parameters are essentially a substitute for homolorphic local coordinates in complex geometry.

Lemma 169 (Weak Inverse Function Theorem).

Let p∈Y⊂X be a smooth point of both X and Y. Then there exist local parameters u1,…,un such that u1|Y,…,um|Y are local parameters on Y and I⁡(Y)=(um+1,…,un) locally in some affine neighborhood of p. Equivalently, a smooth subvariety of a smooth variety is a local complete intersection.

Proof.

Observe the following short exact sequences:

0000IY,X/(mp,X2∩IY,X)Tp,X∗Tp,Y∗00IY,Xmp,Xmp,Y00mp,X2∩IY,Xmp,X2mp,Y20000

Choose a basis u1,…,um of Tp,Y∗ and lift to X. Then choose um+1,…,un which generate the quotient. This gives local parameters u1,…,un in mp,X such that u1|Y,…,um|Y are local parameters on Y. Define Y′=V⁡(um+1,…,u+n). By previous results, Y′ is smooth at p and has dimension m. Then Y⊂Y′ is smooth with the same dimension, so they must be equal in some neighborhood of p. □

Remark 170.

Y is a local complete intersection at p if ℒY⊂𝒪p is generated by s=codimY⁢X equations.

Now let p∈X be smooth and u1,…,un be local parameters. Then X¯=Blp⁢X⊂X×ℙn−1. We check that this is smooth. We know that Blp⁢X∖E≃X∖{p}. Therefore we check for a point y0∈E. We show that my0⁢X¯ is generated by n elements. Take the chart Ui of X¯ given by Ui⊂X×𝔸y2,…,ynn−1. Then the equations are uj=u1⁢yj. Then

Thus X¯ is smooth at Y0.

Theorem 171.

Let 0∈X⊂𝔸n be a smooth point. The proper transform of X in Bl0⁢𝔸n is isomorphic to Bl0⁢X.

Proof.

Choose local parameters u1,…,un such that I⁡(X)=um+1,…,un locallat at 0. Then u1|X,…,um|X are local parameters at 0∈X. We know the blowup is independent of the choice of local parameters.

Therefore Bl0⁢𝔸n⊂𝔸n×ℙn−1 with equations ui⁢yj=uj⁢yi and Bl0⁢X⊂X×ℙm−1 is given by ui⁢yj=uj⁢yi (with appropriate indices). To get equations of Bl0⁢X from equations of Bl0⁢𝔸n, take equations um+1=⋯=un=ym+1=⋯=yn=0 to cut out X and 𝔸n.

To see that this is the proper transform X~, note that um+1=⋯=un along X~ because they cut X×ℙn−1 from 𝔸n×ℙn−1. Then choose one of the ui, say u1≠− at some part of X−0. Then u1⁢yj=uj⁢y1 for all j in Bl0⁢𝔸n. Since uj=0 for j=m+1,…,n, yj=0. □

5.6.1 Normal Varieties

Let p∈X be an irreducible variety. Then X is normal at p is 𝒪p⊂k⁡(X) is integrally closed. Then X is normal if it is normal at every point.

Lemma 172.

If p∈X is a smooth point, or factorial at p, then X is normal at p.

Proof.

𝒪p is a UFD and any UFD is integrally closed. To see this suppose f∈k⁡(X),f=a⁢b where a,b∈𝒪p and a,b are coprime, is integral over 𝒪p. Then fs+a1⁢fs−1+⋯+as=0. Clearing denominators, we see that as+a1⁢as−1⁢b+⋯+as⁢bs=0. Then b∣as, so b is a unit. Thus f∈𝒪p. □

Lemma 173.

Let X be affine. Then X is normal if and only if k⁡[X]⊂k⁡(X) is integrally closed.

Proof.

Let R⊂f.f.R be integrally closed. Then we know that R𝔭⊂f.f.R is integrally closed for every prime p⊂R.

In the other direction, note that Rm⊂K is integrally closed. Then note that R=∩Rm where the intersection is over all maximal ideals m. Then if x∈K is integral over R, it must be integral over Rm for all m, so it must be in Rm for all m. Thus x∈R.

To prove that R=∩mRm, define I={b∈R∣x⁢b∈R}. If I=R, then 1∈I and thus x∈R. Otherwise, I⊂m for some maximal ideal m. This is impossible because we can write x=ab where b∉m because x∈Rm. □

Remark 174.

Every irreducible surface in 𝔸3 with isolated singularities is normal.

Theorem 175 (Serre’s Criterion).

Let R be a Noetherian domain and K the fraction field of R. Then R⊂K is integrally closed if and only if

and for all minimal primes 𝔭⊂R, R𝔭 is a DVR.

Now let R=k⁡[X] for some affine X. Then we find when 𝔭⊂R is a minimal prime (of height 1) if dim⁢R/𝔭=dim⁢R−1, or Y=V⁡(𝔭)⊂X is a hypersurface.

Note that R𝔭=𝒪Y={f∈k(X)∣f is regular at some point of Y. Then note that ∩R𝔭={f∈k⁡(X)∣f⁢ is regular at some point of every hypersurface}. Therefore R=∩R𝔭 if and only if every rational function regular at some point of every hypersurface is regular everywhere.

Theorem 176 (Hartog’s Extension Principle).

Let Z⊂X of codimension at least 2. Then if f∈k⁡(x) is regular on X∖Z, f is regular at some point of every hypersurface, so f is regular.

5.7 Lecture 19 (Apr 09)

5.7.1 Differential of a regular map

Let φ:X→Y be regular. Then we want to define dx⁢φ:TX,x→TY,y. We may assume that X,Y are affine, so choose charts. Then recall that TX,x=(dx⁢f1=⋯⁢dx⁢fr=0) where I⁡(X)=(f1,…,fr), so we define dx⁢φ=(dx⁢φ1,…,dx⁢φm).

Proposition 177.

dx⁢φ⁢(TX,x)⊂TY,y.

Proof.

(dy⁢gi)⁢(dx⁢φ)=dx⁢(fi∘φ)=dx⁢0=0. □

More intrinsically, the dual map TY,y∨→TX,x∨ is induced by the pullback φ∗:𝒪Y,y→𝒪X,x.

Example 178.

Consider X=(x⁢y=t)⊂𝔸3 and φ:X→𝔸1 given by (x,y,t)↦t. Note that Tp⁢X={y0⁢𝑑x+x0⁢𝑑y−𝑑t=0}⊂𝔸3. Therefore X is smooth and Tp⁢X has coordinates 𝑑x,𝑑y (basis of Tp⁢X∨) and 𝑑φ=𝑑t=y0⁢𝑑x+x⁢d0⁢y, which is surjective unless x0=y0=0.

To see the geometry of this map, we see that the fiber above every nonzero point is smooth (a torus), but the fiber at 0 is the union of two coordinate axes.

Lemma 179.

Let φ:X→Y be a regular map of smooth varieties with φ⁡(x)=y. Suppose that dx⁢φ:TX,x→TY,y is surjective. Then v⁢a⁢r⁢p⁢h⁢i−1⁢(y) is nonsingular at x.

Proof.

Consider the differential dx⁢φ:TX,x→TY,y. We show that Tf−1⁢(y),x⊂Ker⁢dx⁢φ. Given the claim, we see that dim⁢Tf−1⁢(y),x≤dim⁢Ker⁢dx⁢φ=dim⁢TX,x−dim⁢TY,y=dim⁢X−dim⁢Y. On the other hand, dim⁢Tf−1⁢(y),x≥dimx⁢f−1⁢(y)≥dim⁢X−dim⁢Y. Thus all inequalities are equalities and f−1⁢(y) is smooth at x.

To proe the claim, consider the sequence of maps Tf−1⁢(y),x⁢→dx⁢ι⁢TX,x⁢→dx⁢φ⁢TY,y. Then Tf−1⁢(y),x→TY,y is a differential of the map φ−1⁢(y)→Y. This is constant, so the differential is zero. □

5.7.2 Normal Varieties Continued

Theorem 180.

A nonsingular variety is normal.

Proof of this theorem comes from Lemma 173.

Definition 181 (Discrete Valuation).

A discrete valuation on a field k is a function v:k∗→ℤ such that v⁡(f⁢g)=v⁡(f)+v⁡(g) and v⁡(x+y)≥min⁡{v⁡(x),v⁡(y)}.

Define R={f∈K∣v⁡(f)≥0} and m={f∈R∣v⁡(f)>0}.

Lemma 182.

R is a local ring with maximal ideal m (called a DVR) with field of fractions k.

Proof.

Clearly by definition of discrete valuation, R is a ring and m is an ideal. It is easy to see that everything with valuation 0 is a unit in R, so m is a maximal ideal. Then we see that for all x∈K, at least one of v⁡(x),v⁡(x−1) is nonnegative, so at least one of x,x−1 is in R. □

Lemma 183.

Every DVR is a PID and in fact, every ideal has form (tn) for some fixed t∈m∖m2.

Proof.

Choose t such that v⁡(t)=1. Then clearly t∈m. Take I⊂R an ideal. Then let n be the minimum valuation of any element of I, so v⁡(g/tn)≥0 and thus g∈(tn). Then choose a minimizing f, and we see that f/tn is a unit, so tn∈I. □

Theorem 184.

Let R be a ring. Then the following are equivalent:

  1. 1.

    R is a DVR.

  2. 2.

    R is a Noetherian local domain which is integrally closed and has Krull dimension 1.

Corollary 185.

Let R be a Noetherian integrally closed domain and let 𝔭⊂R is a height 1 prime ideal. Then R𝔭 is a Noetherian, integrally closed, local domain of Krull dimension 1.

Proof of Theorem 184.

First let R be a DVR. Then R is a PID, so it is Noetherian and integrally closed. Also R is local with prime ideals (0) and m, so it has Krull dimension 1.

In the other direction, let m⊂R be the maximal ideal. Take x∈R nonzero. Then there exists a unique n≥0 such that x∈mn∖mn+1 by Krull intersection theorem.We need to show that m is principal. Given the claim, take x∈R∖0. Then x∈(tn)∖(tn+1) for a unique n. Thus x=u⁢tn for some unit u, and thus n defines a valuation k∗→ℤ.

To prove the claim, m≠m2 by Krull or by Nakayama. Take t∈m∖m2. Since the only prime ideals are 0,m, and every radical ideal is an intersection of primes, m=t.Thus mn⊂(t) for some n≥1. Choose n to be the smallest such n, so choose b∈mn−1∖(t). Take x=tb∈K. Then b∈(t), so x−1∉R. Because R is integrally closed, x−1 is not integral over K. This implies that x−1⁢m⊈m (otherwise, choose m=(e1,…,es) and write x−1⁢ei=∑ai⁢j⁢ej. Then ∑(x−1⁢δi⁢j−ai⁢j)⁢ej=0, so det⁢|x−1⁢δi⁢j−ai⁢j|=0. Thus x−1 is integral).

However, x−1⁢m⊂R because if z∈m then b⁢z∈mn⊂(t), so we can write b⁢z=r⁢t, so x−1⁢z=r∈R. We see that x−1⁢m⊂R, so it is an ideal R but not contained in a maximal ideal. Therefore x−1⁢m=R, so m=(x). □

Let X be a normal variety. Then we have local rings R𝔭⊂k⁡(X) and a DVR for every irreducible subvariety Y⊂X of codimension 1. Thus we have a collection of discrete valuations vY:k⁢(X)∗→ℤ. There are called orders of zeroes or poles of f∈k⁡(X) along Y.

5.8 Lecture 20 (Apr 11)

Recall that if X is a normal variety and Y⊂X is an irreducible subvariety, then 𝒪Y,X is a DVR. Last time we defined the order of zeroes or poles as the valuation.

Corollary 186.

Let X be normal and Y⊂X be irreducible. Then there exists an affine chart U⊂X with U∩Y≠0 such that I⁡(Y)⊂k⁡[U] is principal.

Proof.

Assume that X is affine. Then 𝔭=I⁡(Y)=(u1,…,um)⊂k⁡[X]=R. Then we note that 𝔭𝔭⊂R𝔭 is the maximal ideal in the DVR and it is principal. Then we can write t=ab where b∉𝔭. We can write ui=t⁢vi, where vi∈R𝔭, where vi=aibi where bi∉𝔭. Now we know t∈k⁡[U] and ui=t⁢vi, where vi∈k⁡[U]. Then I⁡(Y∩U)=(t)⊂k⁡[U]. □

Corollary 187.

The singular locus of a normal variety ha codimension at least 2.

Proof.

Suppose there exists an irreducible hypersurface Y in the singular locus. As a variety, Y has an open smooth locus. Thus we can shrink X to an affine chart U⊂X such that U∩Y is nonempty and smooth. By the previous corollary, we can also assume that X is affine. Then I⁡(Y)=(t). This implies X is smooth along Y by Theorem 153. This gives a contradiction. □

Corollary 188.

For curves, smooth is equivalent to normal.

5.8.1 Normalization

Trying to resolve singularities is not very well resolved, and has led to several Fields Medals. On the other hand, normalization is relatively easy to understand and always available.

Example 189.

Consider the curve X={y2=x3}. This is singular, so not normal. We see that t∈k⁡(X) is a root of a monic equation T2−x, but t∉k⁡[X]. We know that the rational parameterization from 𝔸1 is birational and finite, and the source is normal.

Definition 190.

A morphism ν:Xν→X is called a normalization if it is birational and finite and Xν is normal.

Theorem 191.

Every irreducible variety has a normalization, which is unique up to isomorphism.

Construction for affine X.

Let k⁡[X]⊂k⁡(X), so defien R to be the integral closure of k⁡[X] in k⁡(X). We show that R is a finitely generated k⁡[X]−m⁢o⁢d⁢u⁢l⁢e. Given the claim, R is also a finitely-generated k-algebra, so R=k⁡[Y] for some variety Y, which must be birational to X because k⁡[X]⊂k⁡[Y]⊂k⁡(X). The extension k⁡[Y]/k⁡[X] is integral, so Y→X is finite. Also, we know R is integrally closed, so Y is normal.

To prove the claim, use Noether’s normalization lemma. We have a diagram

k⁡[X]Rk⁡(X)k⁡[T1,…,Tn]k⁡(T1,…,Tn)

We know that A is a UFD, so it is integrally closed. Thus 𝔸n is normal, so R is integral over A. This implies that R is the integral closure of A in L. It remains to prove the following lemma. □

Lemma 192.

Let A be an integrally closed Noetherian domain in its field of fractions K. Then if L/K is a finite separable extension and B is the intgral closure of A in L, then B is a finitely generated A-module.

Proof.

Consider the map L→K given by the trace of the map given by multiplication by x. Because L/K is separable, then T⁢r⁢(x⁢y) is a nondegenerate quadratic form. It suffices to find a basis v1,…,vn of L over K such that B⊂A⁢v1+⋯+A⁢vn (this is because A is Noetherian).

To find the basis, first choose some basis u1,…,un. After rescaling by elements of A, we can assume that u1,…,un∈B. Then the ui are algebraic over K, so it is a root of some polynomial a0⁢uidi+⋯+adi=0. Then we see that a0⁢ui∈B. Finally, let v1,…,vn be a dual basis via the trace form. Take x∈B and write it as x1⁢v1+⋯+xn⁢vn. First note that T⁢r⁢(x⁢uj)=xj. On the other hand, x∈B, ui∈B implies that x⁢ui∈B, so T⁢r⁢(x⁢ui)∈A. Therefore xi∈A for all i. □

Theorem 193 (Universal Property of Normalization).
  1. 1.

    Suppose g:Y→X is a finite birational regular map. Then there exists a unique morphism f:Xν→Y such that ν=g∘f.

  2. 2.

    If Y is a normal variety with a dominant regular map g:Y→X, then there exists a unique morphism f:Y→Xν such that g=ν∘f.

Proof.

Assume X is affine. In 1, we can also asusme Y is affine. Then note that k⁡(Y)=k⁡(X), so we must conclude that k⁡[Y]⊂k⁡[Xν], and this inclusion is unique.

In 2, we can also assume Y is affine. Then we again know that k⁡[X]⊂k⁡[Y] and k⁡[X]⊂k⁡[Xν] and that k⁡(X)⊂k⁡(Y). We know k⁡[Y] is integrally closed, so k⁡[Y]⊃k⁡[Xν] and this containment is unique. □

5.9 Lecture 21 (Apr 16)

Homework is now due next Tuesday as opposed to this Thursday. There will be one more homework and then the second take-home midterm.

6 Divisors

Let X be an irreducible variety.

Definition 194 (Prime Divisor).

A prime divisor D⊂X is an irreducible subvariety of codimension 1 (hypersurface).

Definition 195 (Divisor Group).

The divisor group Div⁡X is a free abelian group generated by symbols [D] for each prime divisor D⊂X.

Definition 196.

A divisor D=∑ni⁢Hi is effective, or D≥0, if ni≥i for all i.

Suppose X is normal (or at least non-singular in codimension 1) and let H⊂X a prime divisor. Then 𝒪H,X⊂k⁡(X) is a DVR with valuation vH:k⁢(X)∗→ℤ. If vH⁢(f)>0 then we say that f has a zero of order vH⁢(f) along H. If vH⁢(f)<0, then f has a pole. To every f∈k⁢(X)∗ we associate a divisor (f)=∑H⊂XvH⁢(f)⁢[H] called the divisor of zeros and poles.

Lemma 197.

vH⁢(f)=0 for all but finitely many prime divisors.

Proof.

We can shrink X as much as we want because if U⊂X is open, then X∖U contains only finitely many hypersurfaces. We can also assume that X is affine and that f∈k⁡[X]. We can also remove V⁡(f) and assume that f is invertible. Then it is easy to see that (f)=0.Indeed, f is invertible on every hypersurface, so its valuation must be 0. □

Remark 198.

Divisors of the form (f) are called principal divisors. Principal divisors form a subgrop in Div⁡X because the valuation is like a logarithm (takes products to sums).

Example 199.

If X is normal and projective and f∈k⁡(X), (f)≥0 if and only if f is constant.

Definition 200.

The divisor class group Cl⁡X≔Div⁡X/{(f)} is the quotient of the divisor group by the subgroup of principal divisors.

Example 201.

Take X=𝔸n. Then a prime divisor H⊂X is an irreducible hypersurface given by an irreducible polynomial F∈k⁡[x1,…,xn]. We claim that d⁢i⁢v⁢(F)=H. In particular, all divisors are principal, so Cl⁡(𝔸n)=0. To see the claim, we compute vH⁢(F)=1 and then for any other hypersurface, vH′⁢(f)=0.

Remark 202.

Suppose X is normal, f∈k⁡(X) and suppose (f)≥0. Then f is regular outside of that divisor and at general points of irreducible components of (f), so f is not regular only in codimension 2. Therefore f is regular everywhere.

Proposition 203 (Homework).

Let X be an affine variety. Then the class group of X is trivial if and only if k⁡[X] is a UFD.

Example 204.

Let X=ℙn. Then H⊂X irreducible hypersurface is given by an irreducible homogeneous polynomial F∈k⁡[T0,…,Tn] of degree k. However, F∉k⁡(ℙn). However, we know I⁡(H∩𝔸in)=(FTik). Then we know vH~⁢(f) is 1 when H~=H, 0 when H~≠H and H~∩𝔸in≠∅, and unknown when H~ is the hyperplane at infinity.

To compute this, choose a different affine chart which gives −k. Now take any f∈k⁡(ℙn). We know f=∏Fini∏Gjmj, so we can write (f)=∑ni⁢[Fi=0]−∑mj⁢[Gj=0].

Corollary 205.

D∈Div⁡ℙn is principal if and only if deg⁡(D)=0, where def⁡(D)=∑ni⁢deg⁢(Fi), where Fi∈k⁡[T0,…,Tn] is a homogeneous equation of Hi. Therefore the class group of ℙn is ℤ.

Definition 206.

Divisors D,D′ on X are called linearly equivalent (D∼D′) if D−D′ is a principal divisor.

Remark 207.

Every divisor on ℙn is linearly equivalent to a unique multiple of a hyperplane.

The notion of a divisor we just introduced is called a Weil divisor.

Definition 208 (Cartier Divisor).

A Cartier divisor on an irreducible variety X is the following data:

  1. 1.

    A cover X=∪Ui by open sets;

  2. 2.

    Rational functions fi∈k⁢(X)∗ “defining a divisor on Ui” satisfying a compatibility condition: fi/fj is invertible on Ui∩Uj.

Two such data (Ui,fi) and (Vj,gj) are considered equivalent if fi/gj is an invertible regular function on Ui∩Vj.

Now suppose X is normal, or at least nonsingular in codimension 1. Then to each Cartier divisor (Ui,fi) we associate a Weil divisor F such that D∩Ui=(fi)∩Ui. Concretely, D=∑nj⁢Hj, where nj=vHj⁢(fi) as long as Hj∩Ui≠∅.

Example 209.

Take a prime divisor H⊂ℙn and take the standard cover ℙn=∪Ui. We see that I⁡(H∩Ui)=(fi) where fi∈k⁡[x0,…,xn]. Thus H is in fact given by a Cartier divisor (Ui,fi).

Why are the fi compatible? If H is given by homogeneous F of degree k, then fi=F/Tik. Then fi/fj=(Tj/Ti)k, which is regular and invertible on Ui∩Uj.

We can turn CDiv, the group of Cartier divisors into a group (Ui,fi)⁢⋯⁢(Vj,gj)=(Ui∩Vj,fi⁢gj). By the above construction, we have a homomorphism CDiv→Div. If f∈k⁢(X)∗, there is a Cartier divisor (X,f), so principal divisors lie in the Cartier divisors. If X is normal, this is an inclusion.

Definition 210 (Picard Group).

The Picard group Pic⁡X≔CDiv/{(f)} is the quotient of the group of Cartier divisors by principal divisors.

Note that the Picard group is contained in the class group if X is normal. We want to determine when every Weil divisor is Cartier. If X is nonsingular (more generally, 𝒪x,X is a UFD) for all x∈X, choose H⊂X a prime Weil divisor. We need to construct an open cover: first take U0=X∖H,f0=1. For all x∈X, there exists some affine neighborhood U⊃X, which depends on x∈X, such that H is principal in this neighborhood. This gives an infinite open cover, but we can choose a finite subcover. Thus H is a Cartier divisor.

Corollary 211.

If X is locally factorial, every divisor is Cartier.

6.1 Lecture 22 (Apr 18)

We will continue our discussion of divisors. Here we will assume that X is nonsingular and study the behavior of divisors under regular maps.

Let D=(Uα,fα) where Y=∪Uα, fα∈k⁢(Y)∗. Then we define φ∗⁢D=(φ−1⁢Uα,φ∗⁢fα). We need to see that the pullback of each fα is well defined and nonzero. This happens with φ⁡(X)¯ is not contained in the divisor of zeroes and poles of fα, so when φ⁡(X)¯⊈D. Fortunately, we can pass to a linearly equivalent divisor.

Lemma 212.

Let X be nonsingular. Then for any divisor D⊂X there exists D′∼D such that x1,…,xm∉D′.

Proof.

We argue by induction on m. We can assume that x1,…,xm−1∉D,xm∈D. We can also assume that D is prime and assume that X is affine. Thn D has a local equation at xm: (π)=ℒD⊂𝒪xm, where π=p/q for some p,q∈k⁡[X]. Near xm, we see that D=(p). Then D′=D−(p)∼D does not contain xm, but it may contain x1,…,xm−1.

Choose gi∈k⁡[X] such that gi⁢(xi)≠0. Then for all i=1,…,m−1, gi|D∪{x1,…,xm}=0. Then choose constants α1,…,αm−1 such that for f=p+∑αi⁢gi2, f⁡(xi)≠0. We claim that D′=D−(f) does nto contain x1,…,xm. We know that f⁡(xi)≠0, so x1,…,xm−1∉(f). Now we need to know that f is a local equation of D. Then because f=∑αi⁢gi2, we know that p|gi, so f=p⁢(1+∑αi⁢p⁢w2),so it is also a local equation of D. □

This defines a pullback on the level of Picard groups.

6.1.1 Divisors of rational maps to ℙn

Suppose φ:X⇢ℙn is a rational map defined by (φ0:⋯:φn). Suppose know (φ)=Di−D where D0,…,Dn,D are effective. Choose D to be the smallest possible.

We want to know where φ is regular. Choose x∈X. Then φi=pi/q, where pi is a local equation of Di and q is a local equation of D. Thus φ=(p0:⋯:pn). If pi⁢(x)≠0 for some x, φ is regular at p. The converse is also true because 𝒪x is a UFD.

Theorem 213.

The indeterminancy locus of φ is D0∩⋯∩Dn, where Di=(pi=0).

Example 214.

Consider the rational normal curve ℙ1→ℙn. Then φi=xi and (xi)=i⁡[0]−i⁡[∞]. In another chart, x=1y. Then D=n⁡[∞], so Di=i⁡[0]+(n−i)⁢[∞] and then D0∩⋯∩Dn=∅, so the map is indeed regular everywhere.

Suppose X is a nonsingular variety and D is a divisor. Then set ℒ⁡(D)={f∈k⁡(X)∣(f)+D≥0} the complete linear system.

Theorem 215 (Serre).

If X is projective, ℒ⁡(D) has dimension ℓ⁡(D)<∞.

We can define φD:X⇢ℙℓ⁡(D)−1. First choose a basis φ1,…,φℓ⁡(D) and then set φD=[φ1:⋯:ϖℓ⁡(D)].

Lemma 216.

If D∼D′ then ℒ⁡(D)≃ℒ⁡(D′) and φD=φD′, so we have distinguished maps to projective spaces parameterized by the Picard group.

Proof.

If D∼D′ then D=D′+(f). Therefore there are isomorphisms between the linear systems given by multiplying and dividing by f, so they are isomorphic. Then if φ1,…,φℓ⁡(D) is a basis of ℒ⁡(D), f⁢φ1,…,f⁢φℓ⁡(D) is a basis of ℒ⁡(D′). Then the two maps are the same. □

Remark 217.

If X is projective then if (f)+D=(g)+D, we must have (f)=(g), so (f/g)=0 and thus f/g is a constant.

If X is projective, then ℙe⁢l⁢l⁢(D)−1=ℙ⁡(ℒ⁡(D))=|D|, which is defined to be the set of effective divisors linearly equivalent to D.

Example 218.

If X=ℙr, then Pic⁡X=Cl⁡X=ℤ=ℤ⁡[H], where H is a hyperplane. Then |𝑑H| is the set of hypersurfaces of degree d, which is ℙ(r+dd)−1. We see that ℒ⁡(𝑑H)=ℒ⁡(𝑑H0)={f∈k⁡(ℙr)∣(f)+𝑑H0≥0}. Note that ℙr∖H0=𝔸x1,…,xrr. Then we have that all poles of f are at infinity. Thus f∈k⁡[x1,…,xr], so vH0⁢(f)=−deg⁡(f). After homogeneizing, this becomes the set of degree d hypersurfaces, so the map given by the divisor is the Veronese embedding.

6.2 Lecture 23 (Apr 23)

We continue with our discussion of linear systems of divisors. We will always assume that X is projective and nonsingular. Suppose φ:X⇢ℙr is a rational map.

Remark 219.

Given φ:X→ℙr, it can be degenerate, say φ⁡(X)⊂ℙr−1⊂ℙr. In this case, φ0,…,φr are linearly dependent. This does not happen for φD, so we can assome that φ:X⇢ℙr is not degenerate.

For a basis f0,…,fr of ℒ⁡(D),

  1. 1.

    We can assume that D is effective because φD=φD′ for all D∼D′.

  2. 2.

    If D is effective, (fi)=Di−D. However, it can happen that the Di have a common prime divisor.

Definition 220 (Fixed Part).

A fixed part of a linear system ℒ⁡(D) is the largest effective divisor F such that Di−F≥0. We can write Di=F+Mi, where Mi is the moving part of the linear system.

We can write a⁢b⁢s⁢D={D′∼D∣D′≥0}=ℙ⁡(ℒ⁡(D)). Another way to phrase this is to introduce the base locus.

Definition 221 (Base Locus).

The base locus B⁢LD of a divisor D is defined by B⁢LD=∩D′∈|D|D′⊃F.

We know that ℒ⁡(D)≃ℒ⁡(D−F) and that φD=φD−F. Thus we can always reduce to linear systems without fixed part.

Definition 222 (Base-point-free).

A divisor D is base-point-free if its base locus is empty. Note this implies φD is regular.

Definition 223 (Very Ample).

D is very ample if it is base-point-free and φD is an embedding.

Definition 224 (Ample).

D is called ample if k⁢D is very ample for some k>0.

When we start with φ:X⇢ℙr, φ0,…,φr∈ℒ⁡(D). Now we consider incomplete linear systems: a subspace V⊂ℒ⁡(D)⊂k⁡(X). Choose a basis f0,…,fs of V. Then φV=[f0:⋯:fs]:X⇢ℙs. We can complete f0,…,fs to a basis f0,…,fr of ℒ⁡(D) and recover φV as the composition of φD and the projection.

We can talk about the base locus, fixed part of V, define global generation, and very ampleness of V.

Example 225.

Consider ℙ2 with coordinates x,y,z and let H=(Z=0). Then ℒ⁡(2⁢H)={f∈k⁡(ℙ2)∣(f)+2⁢H≥0}={f∈k⁡[x,y]∣deg⁡f≤2}=k⁢[x,y,z]2. Thus |2⁢H| is the set of conics in ℙ2 and φ2⁢H=[z2:x⁢z:y⁢z:x2:x⁢y:y2], the Veronese embedding. Thus 2⁢H is very ample.

Now we consider an incomplete linear system. Chooise points p=[1:0:0],q=[0:1:0]. Consider V the set of conics passing through a,b. Then ℒ⁡(2⁢H)=k6 and dim⁡V≥4. In fact, V is spanned by 1,x,y,x⁢y. Then φV:ℙ2⇢ℙ3 is given by [x:y:1]↦[1:x:y:x⁢y]. In particular, φV is birational. The equation of the image is actually A⁢D−B⁢C=0, which makes φV a birational map ℙ2⇢Q to a quadric surface in ℙ3. We see that the base locus is p,q.

We consider the image of H. For any point [x:y:0], the image is [0:0:0:x⁢y]=[0:0:0:1]. Thus H collapses to the point [0:0:0:1]. In the hyperplane at infinity Q∩(A=0) is given by B⁢C=0, which is a union of two lines. We have two points where φV is not regular, so we resolve by blowing up at p,q. We blowup at p and consider two charts of the local blowup.

In the chart x=1, we see y=u,z=u⁢v, so v=z/y. Thus the chart u=0 is exceptional. In this chart, the map is [z2:z:y⁢z:y]=[u2⁢v2:u⁢v:u2⁢v:u]=[u⁢v2:v:u⁢v:1]. We see this is regular. Note v is a coordinate along ℙ1, so restricting to the exceptional divisor E, we get the map [0:v:0:1], which maps E onto a line A=C=0. The exceptional divisor over q will map to another line A=B=0.

Thus φV is given by blowing up p,q and then contracting the proper transform of H.

Example 226.

We will do similar calculations without being able to do it too explicitly. Choose 6 points p1,…,p6 on ℙ2 that are distinct, in general linear position, and not coconic. Consider the linear system ℒ⁡(3⁢H) and φ3⁢H. Note |3⁢H| is the space of cubics in ℙ2, so consider V to be the set of cubics through p1,…,p6. We check that the dimension of V is 4. If this is not the case, then every cubic that passes through p2,…,p6 also passes through p1. This is false because we can take a conic through p1,…,p6 and add a line not through p1. Thus we get a rational map φV:ℙ2⇢ℙ3.

First we find the base locus, which is {p1,…,p6}. For any other point q, we can take the conic through five of the points and a line containing p6 but not q. If q is on the conic, just change the five points and then note that q cannot be on the new conic because both conics are smooth and they already have four intersections points.

6.3 Lecture 24 (Apr 25)

Last time we were building up to a theorem of Clebsch:

Theorem 227 (Clebsch).

For a smooth projective surface X, the following are equivalent:

  1. 1.

    X≃B⁢l6⁢ℙ2 where the six points are in general position (no three are collinear and the six points are not coconic).

  2. 2.

    X is a smooth cubic surface in ℙ3.

  3. 3.

    X is a del Pezzo surface of degree 3 (Fano variety of dimension 2, which means the anticanonical divisor is ample).

We continue Example 226 from last time.

Example (Continuation of Example 226).

We show that for all p,q∈ℙ2∖{p1,…,p6}, φV⁢(p)≠φV⁢(q). To see this, take the conic through 1,2,3,4,5 and the line through 6,p and a similar combinatorial argument from the base locus calculation tells us that at least one choice works.

Next we show that the induced rational map B⁢l6⁢ℙ2→ℙ3 is regular. It suffices to show that it is regular at every point of E≃ℙ1 over p1. To do this, we find two cubics D1,D2∈V transversal at p1. To do this, we take the conic C through p2,…,p6 and then two distinct lines L1,L2 through p1. Then D1=C∪L1,D2=C∪L2. To see that this is enough, choose a basis f1,…,f4 of V such that D1=(f1=0) and D2=(f2=0). Assume p1=[0:0:1]. We know φV is given by [f1:f2:f3:f4]=[f1:z3:⋯:f4/z3]. We can dehomogeneize in the standard chart and write φV=[y+α:x+β:g3:g4], where α,β∈m(0,0)2 and g3⁢(0,0)=g4⁢(0,0)=0.

Next we blow up and consider the chart x=u,y=u⁢v. Then φV=[u⁢v+u2⁢g1:u+u2⁢g2:u⁢g3:u⁢g4]=[v+u⁢g1:1+u⁢g2:g3:g4]. The equation of E1 is u=0, so we substitute u=0 and in the second component we have 1, so φV is regular at every point of E1.

Remark 228.

We consider π∗⁢V where π:B⁢l6⁢ℙ2→ℙ2. Then for all D∈V, we see π∗⁢D=E1+⋯+E6+D′, so π∗⁢V=(E1+⋯+E6)+|D′|. Thus D′∼π∗⁢(3⁢H)−E1−⋯−E6. The D′ are the preimages of hyperplanes in ℙ3. Because φ⁡(Ei) are not points (it intersects a general hyperplane at one point), it must be a curve. In fact it must be a line because a general hyperplane will give a cubic in ℙ2 smooth at pi, so its proper transform intersects Ei in one point.

Next we will show that φ⁡(B⁢l6⁢ℙ2) is smooth. Indeed, if not, Tx has dimension 3 for some x∈X. (To be continued next lecture).

7 Regular Differential Forms

Now we will attempt to perform calculus on algebraic varieties. Let X be a variety and f∈k⁡[X]. Then dx⁢f is a linear function on Tx.

Definition 229.

Φ⁡[X]={ functions that assign every ⁢x∈X⁢ an element of ⁢Tx∨}. Note that Φ⁡[X] is a k⁡[X]-module. For f∈k⁡[X],α∈Φ⁡[X], (f⁢α)⁢(x)=f⁡(x)⁢α⁡(x).

Definition 230.

The regular differential forms Ω⁡[X]⊂Φ⁡[X] are α∈Φ⁡[X] such that for every x∈X, there exists a neighborhood U∋x such that α|U−∑ifi⁢𝑑gi. We have a map d:k⁡[X]→Ω⁡[X] and the Leibniz rule d⁡(f⁢g)=f⁡(𝑑g)+g⁡(𝑑f).

Lemma 231.

If X is affine, then Ω⁡[X] is generated by differentials.

Proof.

We use another algebraic partition of unity. For all x∈X, we can write α=∑fi,x⁢𝑑gi.x is some neighborhood of x. Clearing denominators, px⁢α=∑iri,x⁢𝑑gi,x where px,ri,x,gi,x∈k⁡[X] and px⁢(x)≠0. Then the ideal (px)=k⁡[X] by the Nullstellensatz, so we can choose finitely many of them and write ∑px⁢qx=1. Then multiply by qx and add, and we get α=∑i.xri,x⁢qx⁢𝑑gi,x. □

Corollary 232.

Let X be affine and g1,…,gs generate k⁡[X]. Then 𝑑g1,…,𝑑gs generate Ω⁡[X] as a k⁡[X]-module (use the Leibniz rule to prove this).

In the simplest case, Ω⁡[𝔸n] is a free k⁡[x1,…,xn]-module with generators 𝑑x1,…,𝑑xn (if ∑fi⁢𝑑xi=0 then ∃x∈𝔸n where one fi≠0, and then 𝑑x1,…,𝑑xn ar e linearly dependent in Tx⁢𝔸3, which is a contradiction).

Theorem 233.

Suppose x∈X is a nonsingular point with local parameters u1,…,un. Then there exists an affine neighborhood U of x such that Ω⁡[U] is a free k⁡[U]-module generated by 𝑑u1,…,𝑑un.

Proof.

Without loss of generality, x⊂𝔸N. Then I⁡(X)=(F1,…,Fm), so k⁡[X] is generated by fi=Ti|X, so Ω⁡[X] is generated by 𝑑t1,…,𝑑tN. We know Fi|X=0, so ∑j∂Fi∂Tj⁢𝑑tj=0 for all i. The Jacobian matrix has rank N−n at all nonsingular x∈X, so it has a nondegenerate minor with the last variables.

Thus t1,…,tn are local parameters of x. From We get that 𝑑tj=∑i=1nfji⁢𝑑ti for all j>n using Cramer’s rule, where fji∈𝒪X. Now we shrink X to an affine neighborhood U such that all fji∈k⁡[U]. After this shrinking, Ω⁡[U] is generated by 𝑑t1,…,𝑑tn and t1,…,tn are local parameters for all y∈U. □

7.1 Lecture 25 (Apr 30)

There will be extra office hours tomorrow from 2 to 4.

We will continue our study of multivariable calculus. Last time we discussed regular differential forms.

Example 234.

We calculate Ω1⁢[ℙ1]. We know Ω1⁢[𝔸1]=k⁡[x]⁢𝑑x. Take another chart with coordinate y=1/x, we can write ω=f⁡(x)⁢𝑑x=f⁡(1y)⁢d⁢1y=g⁡(y)yn⁢(−𝑑yy2)=g⁡(y)⁢𝑑yyn+2 where g⁡(y)≠0. This is not a polynomial multiple of 𝑑y, so Ω1⁢[ℙ1]=0.

Now we talk about differential r-forms. We introduce Φr⁢[X]={x∈X↦ω⁡(x)∈∧rTx∗} and Ωr⁢[X] as the subspace that can be written locally as ω=∑fi1,…,ir⁢𝑑gi1∧⋯∧𝑑gir.

Theorem 235.

Let X be nonsingular at x and U∋x an affine neighborhood with a system of local parameters u1,…,un. Then Ωr⁢[U] is a free k⁡[U]-module generated by (nr) forms of the form 𝑑ui1∧⋯∧𝑑uir.

Proof of this is the same as for Ω1. Now we discuss the case of Ωn⁢[X] the space of canonical differentials. Here we have Ωn⁢[u]=k⁡[U]⁢𝑑u1∧⋯∧𝑑un. If u1′,…,un′ is another system of local parameters, then 𝑑u1∧⋯∧𝑑un=𝒥⁡(u1,…,un,u1′,…,un′)⁢𝑑u1′∧⋯∧𝑑un′ where J is an invertible function.

We discuss rational canonical differentials Ωn⁢(X) where every pair is a form (U,ω) where U⊂x is open and ω∈Ωn⁢[U]. They are subject to equivalence by equality on intersections. We can assume from now on that X is nonsingular and ω∈Ωn⁢(X) is regular on some affine open with local parameters u1,…,un, so ω=f⁡(x)⁢𝑑u1∧⋯∧𝑑un where f∈k⁡(X) is regular on U.

Lemma 236.

Ωn⁢(X) is a one-dimensional vector space over k⁡(X) generated by 𝑑u1,…,𝑑un for all choices of local parameters somewhere.

Definition 237.

Choose ω∈Ωn⁢(X)∗. We will define (ω) as a Cartier divisor. Cover X=∪Uα such that for all alpha there exist u1α,…,unα are local parameters on Uα. Then ω|Uα=fα⁢𝑑u1α∧⋯∧𝑑unα for fα∈k⁡(X). This gives a Cartier divisor (Uα,fα).

To see this is Cartier, on the overlaps Uα∩Uβ we see that ω|Uα∩Uβ=fα⁢𝑑u1α∧⋯∧𝑑unα=fα′⁢𝑑u1α′∧⋯∧𝑑unα′=fα⁢𝒥⁢𝑑u1α′∧⋯∧𝑑unα′, so fα=𝒥⁢fα′, where the Jacobian is invertible.

Note that if ω′=f⁢ω then (ω′)=(f)+(ω), so they are equivalent. Thus the class of KX in the Picard group only depends on X.

Example 238.

We calculate the canonical divisor of ℙn. Let 𝔸n={z0≠0} and choose ω=𝑑x1∧⋯∧𝑑xn. This has no zeroes or poles on 𝔸1 and let H=ℙn∖𝔸n be the plane at infinity. We see that [1:x1:⋯:xn]=[1/x1:1:x2/x1:⋯:xn/x1], so writing y1=1/y1,x2=y2,…,xn=yn/y1, we have

Thus Kℙn=−(n+1)⁢H.

Definition 239.

X is Fano is −KX is ample.

Corollary 240.

ℙn is Fano. φ−KX=φ(n+1)⁢H is the (n+1)-th Veronese embedding of ℙn.

For example, −Kℙ2=3⁢H. and |−Kℙ2| is the set of cubic curves in ℙ2. We now determine the relationship between KS and KX if S is a blowup of X at a point p. Choose local parameters u,v at p and ω=𝑑u∧𝑑v on X. Then (ω) is disjoint from p, and we know k⁡(S)=k⁡(X), so ω∈Ωn⁢(S). We compute its divisor on S.

Working in a chart of the blowup with local parameters x,y where u=x,v=x⁢y, we have ω=𝑑u∧𝑑v=𝑑x∧(x⁢𝑑y+y⁢𝑑x)=x⁢𝑑x∧𝑑y. Near E, (ω) is precisely E. This proves the following:

Theorem 241.

Let S⁢→𝜋⁢X be the blowdown. Then KS=π∗⁢KX+E, where E is the exceptional divisor.

Corollary 242.

Let S be a blowup of ℙ2 at 6 points. Then KS=π∗⁢Kℙ2+E1+⋯+E6=−3⁢H+E1+⋯+E6 and thus −KS=3⁢H−E1−⋯−E6.

Last time, we constructed the map φ:S→ℙ3. We have the following:

Theorem 243.

φ is an isomorphism onto its image, which is a smooth cubic surface.

Proof.

We already know that φ is birational and bijective away from E1,…,E6. Also, φ|E+i is a line in ℙ3. Therefore φ has finite fibers. We claim that φ is finite. Instead of a proof of the claim, we state the Stein Factorization Theorem (after the end of this proof).

Given Stein, φ is finite and birational, so we need to check that Y is normal. Note X is smooth, so it is normal. We use Serre’s criterion. If X is a hypersurface in ℙn then X is Cohen-Macaulay and is thus S⁢1 by Hartog’s principle. Now we need to show that X=φ⁡(S)⊂ℙ3 has isolated singularities. If not, then X is singular along C⊂X, so every plane section X∩H is singular at C∩H. Thus X∩H is a singular cubic curve, which means it is rational. However, on S, X∩H is a member of |3⁢H|−p1−⋯−p6, which for some H is a smooth cubic curve, which is not rational. □

Corollary 244.

−KS is very ample, so S is a Fano surface, or a del Pezzo surface.

Theorem 245 (Stein Factorization Theorem).

Let f:X→Y be a proper regular map. Then there exists a commutative diagram

XYZ

where X→Z has conncted fibers and Z→Y is finite. In particular, if f is proper and has finite fibers, then f is finite.