18.

721: Introduction to Algebraic Geometry

Lecturer: Professor Mike Artin
Notes by: Andrew Lin

Spring 2020

1 February 3, 2020
Algebraic geometry is a beautiful subject, and it’s usually taught as a mid-level graduate course, so we’ll need to
discuss things in this class without a lot of background. In particular, we won’t assume commutative algebra (18.705),
though that might be useful. (18.702 is essential, though.)
Throughout this class, we’ll work with the scalars C – algebraic geometry can be done with any field of scalars,
but we’re making this choice to give a bit of intuition into the geometry. (Real numbers don’t work, by the way – we
need algebraic closure.)
The central objects of study here are systems of polynomial equations in multiple variables x = (x1 , x2 , · · · , xn )
of the form {f1 (x) = 0, · · · , fk (x) = 0}. Geometrically, this means we’ll look at the locus of solutions X in affine
space An = Cn , and algebraically, it means we’ll look at the polynomial ring C[x] (that is, polynomials in our variables
xi ), modded out by our polynomials, which yields an algebra of the form

A = C[x]/(f1 , · · · , fk ).

Note that any ring that can be generated by finitely many elements will be of this form – being finitely generated
means that there’s a surjective map C[x] → A, and the Hilbert Basis Theorem tells us that the kernel can always be
generated by finitely many elements.
Remember that the ideal of C[x] generated by f1 , · · · , fk is the set of linear combinations g1 f1 + · · · + gk fk , where
gi ∈ C[x]. The point is that the geometry and algebra of this situation will complement each other!
We’ll start with some simple examples:

Example 1
Consider the two-variable polynomial f (x, y ) = y 2 − x 3 − x 2 .

Let X be the locus (of zeros), described via y 2 = x 3 + x 2 . X lives in 4-space (because x and y can be complex),
but we can at least graph it in real space:

1
 The geometry is reflected in the algebra here, because we can actually parameterize this curve using polynomials:
if we draw a line of slope t from the origin (which is a double zero), then y = tx, so

f (x, tx) = t 2 x 2 − x 3 − x 2 = x 2 (t 2 − x − 1),

so x = t 2 − 1, which tells us that y = t 3 − t. In algebraic terms, this means that we can construct a map A1t → X
(the one-dimensional affine space maps to our locus X). In addition, we can take the algebra A = C[x, y ]/(f ) and
map it to C[t] (though these are not isomorphic algebras). Really, what’s going on is that we take the one-dimensional
complex t and identifying two points together, and that gives us our zero locus.

Example 2
On the other hand, consider the curve f (x, y ) = y 2 − x 3 + x.

Now we want y 2 = x 3 − x, but notice that the (real) curve that comes out of this is no longer just one piece –
things are more complicated here. We’ll come back to this in a second.

Affine space is interesting, but it’s often important to study projective space:

Definition 3
Points in projective space Pn are classes of nonzero vectors (x0 ; x1 , · · · , xn ) with the equivalence relation that

(x0 ; x1 , · · · , xn ) = (λx0 ; · · · , λxn )

for any λ ∈ C, λ 6= 0.

One reason this is important is that Pn is compact, while An is not! Let’s verify this:

Proof. Let V be Cn+1 , so then V − {0} maps to Pn . Then equivalence classes of Pn map to 1-dimensional subspaces
of V , so we can look at the unit sphere S in V . (Since V is a 2n + 2-dimensional real space, S has dimension 2n + 1.) S
∑
is the set of points x such that x i xi = 1; since we’re modding out the equivalence relation, every point in projective

2
space is represented by points in S (not uniquely). So there is a surjective map from S to Pn . Clearly S is closed and
bounded, so it is compact, and therefore Pn is compact as well.

So how can we find loci in projective space Pn ? We want polynomials f (x0 , · · · , xn ) so that if f (x) = 0, then
f (λx) = 0 as well for all complex λ. If we write out

f (x) = f0 + f1 + f2 + · · · + fd

(splitting up our polynomial by degree), then

f (λx) = f0 + λf1 + λ2 f2 + · · · + λd fd .

For any given x where f (x) = 0, we can substitute in the values of f0 (x), f1 (x), and so on into our equation. We now
have a polynomial in λ that needs to be satisfied for all complex values of λ: this is only possible if the polynomial is
the zero polynomial! So that means fi (x) = 0 for all i .
In other words, in Pn , we should only be studying the zeros of homogeneous polynomial equations in x =
x0 , · · · , xn – this tells us the same information as if we try work with polynomials in general.
So how can we study a polynomial like f (x, y ) = y 2 − x 3 + x in projective space if it’s not homogeneous? The
solution is to use the extra variable to homogenize into a cubic

f (x, y , z ) = y 2 z − x 3 + xz 2 .

Let’s discuss the locus of this polynomial in P2x,y ,z . First, let’s go back to a more generic description: any point
x0 , as long as x0 6= 0. This gives us a subset U 0 ⊂ P2 of
xi
(x0 , x1 , x2 ) can be identified with (1, u1 , u2 ), where ui =
points, corresponding to points (u1 , u2 ) in the affine plane A2u1 ,u2 . Meanwhile, the points (x0 , x1 , x2 ) where x0 = 0 are
on a line L (defined by {x0 = 0}). So P = U ∪ L : U is often called the points at finite distance, and L0 is often
0 2 0 0 0

called the line at infinity.

With this, we can go back to our equation y 2 z − x 3 + xz 2 = 0: relabeling, we can say that

x22 x0 − x13 + x1 x02 = 0.

If we consider x0 = 1, we get the affine space curve x22 − x13 + x1 (which is what we started with). But if x0 = 0, we
see what happens at infinity: that just gives us x13 , which means that there’s only one point at infinity, (0, 0, 1), on
the curve. (This has to do with being able to compactify the locus in one point.)
So let’s compute the Euler characteristic of this locus X (the number of vertices plus the number of faces, minus
the number of edges). We have four branch points (the three zeros on the x1 -axis, plus the point at infinity), but
at every other point there are two different values of x2 . This means that X covers most of P1 twice, so we should
triangulate the two copies of P1 instead. This gives two edges, faces, and vertices over every edge, face, and vertex,
except at the four branch points (vertices), which are double-counted:

e(X) = 2e(P1 ) − 4 = 2 · 2 − 4 = 0.

So the genus of the surface here satisfies

e = 2 − 2g =⇒ g(X) = 1.

This means that if we drew the locus in complex space, we see a torus (except missing the point at infinity).

3
2 February 5, 2020
We’ll start by taking another look at our picture of the (real) projective plane, which can be described with points
in three dimensions. Let’s take the plane containing the points (1, 0, 0), (0, 1, 0), and (0, 0, 1), which has equation
x0 + x1 + x2 = 1. Every one-dimensional subspace of R3 goes through this plane exactly once, except the subspaces
with x0 + x1 + x2 = 0.
Yesterday, we drew the curve x0 x22 = x13 − x02 x1 : there are three points where x2 = 0, which are at x1 = 0, ±x0 . In
addition, the curve intersects the line at infinity x0 = 0 with multiplicity 3 at the point (0, 1, 0), so that yields a “flex
point” at the intersection:

triple zero

x0 = 0
x1 = 0

x2 = 0

Remark 4. Note that a point like (−1, −1, 1) and a point like (1, 1, −1) are the same in projective space, so the
regions “wrap back” around.

Another way to draw this real projective plane is to take a 2-sphere and identify opposite points with each other,
but we won’t use that picture here.
With that, let’s move on to projective plane curves. We want to look at the locus of zeros for polynomials f ,
but remember here that we want to make sure f is an irreducible homogeneous polynomial.
The simplest case is a line: it takes the form

f = s0 x0 + s1 x1 + s2 x2 = 0.

Another description is that we can take two points p = (p0 , p1 , p2 ) and q = (q0 , q1 , q2 ) and join them together to
get a line Lpq = {ap + bq}, which corresponds to a one-dimensional subspace (a, b) ∈ P1 . So a line looks like a
one-dimensional projective space.
A conic is next: it’s a combination of the degree-two monomials, which are x02 , x12 , x22 , x0 x1 , x0 x2 , x1 x2 . First, let’s
see how to do linear changes of coordinates in P2 : if x = (x0 , x1 , x2 ), we just let x ′ = P x for some 3 × 3 complex
invertible matrix.

Proposition 5
For any conic C, we can choose coordinates so that the equation of C in new coordinates is x0 x1 + x0 x2 + x1 x2 = 0.

So there’s only one conic up to change in coordinates in projective space!

Proof. Choose three points on C, not on a line. (Being able to do this is the first problem on our homework, due next
week.) Adjust coordinates so that these are (1, 0, 0), (0, 1, 0), and (0, 0, 1).
Then the modified function f satisfies f (1, 0, 0) = 0, so the coefficient of x02 is 0. The same holds for x12 and x22 ,
so now f is of the form ax0 x1 + bx0 x2 + cx1 x2 . And now scale the variables so that a, b, c become 1.

Next, let’s talk a bit about the tangent space: let C be a curve corresponding to the locus of zeros for f , and let
p ∈ C be a smooth point (this means the partial derivatives ∂f
∂xi (p) are not all zero – the other case is called a singular

4
point). For example, x03 + x13 + x23 = 0 is smooth (except at the origin, which is not a point of projective space), but
x0 x22 − x13 − x0 x12 has partial derivatives (x22 − x12 , −3x12 − 2x0 x1 , 2x0 x2 ), which is zero for example at (1, 0, 0).
There’s a nice formula we can use:

Theorem 6 (Euler)
If f is homogeneous with degree d, then

x0 f0 + x1 f1 + x2 f2 = d · f .

where f0 refers to the partial derivative of f with respect to x0 .

We can just check this for a monomial: for example, if f = x02 x1 , then

x0 f0 + x1 f1 + x2 f2 = 2x02 x1 + x02 x1 = 3f .

(xi fi contributes the degree of xi towards the count of f .) So let’s write down the tangent line of a curve C at a point
p. To do that, pick another point q distinct from p, and define the line (this is equivalent to {ap + bq} but better
suited for our purposes)
Lpq = {p + qt} ∪ {q}.

Let’s now expand f (p + qt) in a Taylor series:

 
(∑ ) 1 ∑
f (p + qt) = f (p) + fi (p)qi t+  qi fij qj  t 2 + O(t 3 ).
2
i,j

Looking term by term now lets us know when f (p + qt) is a tangent line. p is on the curve, so f (p) is always zero.
But the first derivative tells us about the slope, so we want to control the derivatives. We can write down the gradient

∇f = (f0 , f1 , f2 )

and the Hessian matrix

∂2f
H = (fij ), fij = .
∂xi ∂xj
   
p0 q
   0
If we write our two points p = p1  and q = q1 
  
 as column vectors, we can now rewrite
p2 q2

1
Lpq = f (p) + ∇p qt + (q t Hp q)t 2 + O(t 3 ).
2
Define a bilinear form on column vectors in C3 via

hu, v i = u t Hp v ,

and now we can rewrite all of our coefficients for our line: Euler’s formula applied to the gradient tells us that

∇p · p = d · f (p),

5
and Euler’s formula applied to the Hessian tells us that
 
[ ] f00 f01 f02  [ ]
p t Hp = p0 p1 p2  f f
 10 11 12  f  = (d − 1) f0 f1 f2 = (d − 1)∇p .
f20 f21 f22

Therefore,
p t Hp p = (d − 1)∇p p = d(d − 1)f (p),

and now Lpq can be written as

1 1 1
Lpq = hp, pi + hp, qit + hq, qit 2 + O(t 3 ).
d(d − 1) d −1 2

p is on our curve C if hp, pi = 0, the line is tangent if hp, qi = 0 as well, and we get a flex if hq, qi = 0.
This line of reasoning only works if p is a smooth point on the curve, but we won’t be talking about the special
cases with singular points much in this class.

3 February 7, 2020
We’ll talk today about the first subtle fact of this subject, the dual curve. Let P = P2 be the projective plane, and
consider a line L in Px (that means points are labeled as (x0 , x1 , x2 )) with the equation s0 x0 + s1 x1 + s2 x2 = 0.
We can think of points (x0 , x1 , x2 ) on this line, but we can also think of this equation as a point (s0 , s1 , s2 ) – both
of these can be thought of as points in the projective plane, because the coefficients are only determined up to scaling.
So that means that a line L in P gives a point (s0 , s1 , s2 ), which we can call L∗ in the dual plane P∗ = P2s , and we can
also take a point p ∈ P and turn it into a line p ∗ in P∗ (by taking (a, b, c) and turning it into ax0 + bx1 + cx2 = 0).
One way we can then write down this interchange between points and lines is that

p ∈ L ⇐⇒ L∗ ∈ p ∗

(the roles of points and lines switch).

So now let C be a plane curve of the form f = 0 (which is irreducible and homogeneous). Recall that a point
p ∈ C is singular if the partial derivatives f0 (p) = f1 (p) = f2 (p) = 0. All other points are smooth – it’s easy to show
that the number of singular points for an irreducible f is finite. Thus, let U be the set of smooth points; we can map
U to the dual space
t : U → P∗ .

Explicitly, how is this map defined? If we take a point p ∈ U, let L be the tangent line to C at p (that’s why we leave
out the singular points, because the tangent line isn’t defined). Then we can let t(p) = L∗ .

Theorem 7
Let U ∗ = t(U). Then the closure of U ∗ is a curve C ∗ in the dual space P∗ , and C ∗∗ = C.

Basically, there’s some special cases to think about for our curve C: maybe we have a bitangent, which means
that a single tangent line is shared between two points, so the two points will have the same image in the dual space.
(So the curve will cross itself, and this gives us a node in the dual space.) Also, maybe we have a flex point where
the derivative goes to 0 for a moment – this turns out to give a cusp in the dual space. That’s why we don’t take

6
C to be smooth – C ∗ might still end up being singular even when C is smooth. Plus, we can use the degree of C to
count the number of nodes and cusps, too, which can be interesting.

Example 8
A smooth cubic has 9 flex points (no bitangents because that would mean slightly perturbing the bitangent would
make it intersect a line four times), and a generic quartic has 28 bitangents. (This means that the curve C ∗
coming from a generic quartic will have 28 nodes.)

How do we find the degree of the dual curve? If C is cubic and smooth, then C ∗ has degree 6, and it has 9 cusps
(corresponding to the 9 flex points on the cubic). A famous picture for the quartic: intersect two ellipses, and take
the product of the two equations to get an equation of the form

(ax 2 + by 2 − c)(a′ x 2 + b′ y 2 − c ′ ) = f (x, y ).

And now if we add or subtract a small number, we can think about what happens to the zero locus. For example, the
locus of zeros where f (x, y ) + ε = 0 will look like “four beans” (in the four areas inside one ellipse but not the other),
and that actually gives us 28 bitangents – four between each pair of beans, and one within each bean.
We’re going to use something called the transcendence degree, which we should read on our own (it’s algebra,
not algebraic geometry). We’ll review the main ideas here.

Definition 9
Let K/F be a field extension. α1 , · · · , αk ∈ K are algebraically dependent over F if there exist coefficients
f (x1 , · · · , xk ) ∈ F [x1 , · · · , xk ] such that
f (α1 , · · · , αk ) = 0.

(Otherwise, they are called algebraically independent.) The transcendence degree of K/F is the order of a
maximal set of algebraically independent elements of K.

(It can be checked that the transcendence degree is independent of the choice of set that we use.)

Example 10
The field of rational polynomials K = F (x1 , · · · , xk ) has transcendence degree k, so any set of k + 1 elements is
dependent.

With this, we can prove Theorem 7:

Proof. We’re working in K = C(x0 , x1 , x2 ) here, and we’re looking for a polynomial ϕ(s) which is zero on U ∗ = t(U).
(Let’s not worry about whether it’s homogeneous or irreducible first.)
The transcendence degree of K over C is 3, so any four polynomials will be algebraically dependent: let’s take f (x)
(which is the polynomial that defines C), as well as f0 (x), f1 (x), and f2 (x) (the partial derivatives). Because these four
polynomials are dependent, there exists an Ψ(s0 , s1 , s2 , t) such that

Ψ(f0 (x), f1 (x), f2 (x), f (x)) = 0

is identically zero. We can assume Ψ does not have any factors of t, and now if we define ϕ(s0 , s1 , s2 ) = Ψ(s0 , s1 , s2 , 0),
then ϕ is now not the zero polynomial. On the other hand, if we let x = (x 0 , x 1 , x 2 ) be a point of C (so f (x) = 0

7
because we’re on the zero locus), then

ϕ(f0 (x), f1 (x), f2 (x)) = Ψ(f0 (x), f1 (x), f2 (x), f (x)) = 0

(because Ψ is the zero polynomial here). So for all points x on U (that is, all smooth points of C of our curve), we
have ϕ(f0 (x), f1 (x), f2 (x)) = 0.
Now, what is the tangent line at this point x? Remember the Taylor series expansion

f (p + tq) = f (p) + (∇p q)t + O(t 2 ),

where ∇p = (f0 (x), f1 (x), f2 (x)). So if (s0 , s1 , s2 ) = t(x) (where t is the map from U to P∗ ), then (s0 , s1 , s2 ) =
(f0 (x), f1 (x), f2 (x)). That means that ϕ(s) = 0 is zero on U ∗ , and now we have a polynomial that vanishes on U ∗ .
Since λx is the same point as x, we can make ϕ homogeneous (by just looking at one of its homogeneous parts).
Now letting g(x) = ϕ(f0 , f1 , f2 ), we know that g vanishes on U and is homogeneous: factor g, which corresponds to a
factorization of ϕ. Now f will still divide one of the factors of g, and the corresponding irreducible factor ϕ will vanish
on U (and therefore C): thus we’ve indeed shown that the closure of U ∗ is a curve in the dual space.
Let’s spend some more time on why C ∗∗ = C. Say we have a point p0 with tangent line L0 on the curve C – this
means that L∗0 will lie on p0∗ on the curve C ∗ . It seems pretty plausible that p0∗ will be the tangent line, and we’ll try
to show that here.
To do that, consider another point p1 on C with tangent line L1 , and let L0 and L1 intersect at q. Since q lies on
L0 and L1 , the line q ∗ goes through points L∗0 and L∗1 . But now have p1 approach p0 – L∗1 will approach L∗0 , so the
line q ∗ will approach the tangent line. And now we just need to show that q ∗ is equal to p0∗ .
We’ll work with affine coordinates for this (specifically, work in P2 where z = 1) – pick them such that p0 is the
origin (0, 0, 1) and the tangent line at p0 is the line y = 0. Then C is some curve f (x, y ) = 0 – we can solve for some
function y = y (x) such that f (x, y (x)) = 0. Letting p1 = (x1 , y1 ), the equation of L1 (the tangent to p1 ) is of the
form
y − y1 = y1′ (x − x1 ),

where y1 = y (x1 ) and y1′ = y ′ (x1 ). Then q = (xq , 0, 1) lies on this tangent line, so we can solve to find
y1
xq = x1 − .
y1′

As (x1 , y1 ) goes to (x0 , y0 ), xq goes to 0, because the derivative y ′ has a zero of order one less than that of y . So the
limit as p1 goes to p0 of q is indeed p0 , which is what we want – this tells us that we do have C ∗∗ = C.

One quick application of this is that C being smooth means C ∗ has no bitangents and no flex points (otherwise
∗
C ’s dual would have a node or cusp). Soon, we’ll talk about the Plücker formulas, which tell us a bit more.

4 February 10, 2020

The first quiz has been moved from Friday to Wednesday due to popular vote.
Today, we’re going to talk about resultants. Say we have two polynomials

f (x) = a0 x 3 + a1 x 2 + a2 x + a3 , g(x) = b0 x 2 + b1 x + b2

(they could also be homogeneous if we just put y s to bring the degrees up). Then the resultant Res(f, g) is a
polynomial in the coefficients {ai , bj }, with the important property that Res(f , g) = 0 if and only if f and g have a

8
common root.
At first glance, it might be confusing why we care about such a polynomial – it turns out this actually computes
a kind of “projection.” Say that f = f (t, x) and g = g(t, x) are polynomials, so the coefficients ai and bj above are
now polynomials in t. Then the resultant Resx (f , g) with respect to x is a polynomial in t as well – it has zeros at the
“images” of the intersection points when we project down onto the t-axis.
We can do this in higher dimensions as well: say f and g are polynomials of x, y , z . Then Resz (f , g) is a polynomial
in x and y , and it’s zero at the points where we take the intersections of the zero loci for f and g and project that
down onto the xy -plane.
So what polynomial in the coefficients {ai , bj } are we trying to take here? Say f , g have a common root x = x,
and define h = x − x. Then h divides f and g, so we can write f = hp and g = hq for polynomials p, q, so
fg
= pg = f q
h
are two different ways of writing this polynomial. If f and g have, for example, degree 3 and 2, then p and q have
degrees 2 and 1, respectively. We can think of pg = f q as a relation among two polynomials – specifically, we can
consider what monomials can appear in p. pg is a combination of x 2 g, xg, and 1g if p is a quadratic, and f q is a
combination of xf and 1f . This gives us 5 dependent polynomials, and we’ll put these in a matrix. Because the
relation pg = f q has degree 4, we’ll write things in a 5 × 5 table.

x4 x3 x2 x 1
xf a0 a1 a2 a3 0
1f 0 a0 a1 a2 a3
2
x g b0 b1 b2 0 0
xg 0 b0 b1 b2 0
1g 0 0 b0 b1 b2
We now have a 5 by 5 matrix R, and if the common root exists, then the determinant of this matrix should be
0 because the polynomials are dependent on each other.

Definition 11
The resultant of two polynomials f and g is the determinant of the matrix R that comes out of this process
above.

Proposition 12
Suppose that f and g are monic, and suppose f has roots α1 , α2 , α3 and g has roots β1 , β2 . Then the resultant
is the product
∏ ∏
Res(f , g) = (αi − βj ) = g(αi )
i,j i

(because f = (x − α1 )(x − α2 )(x − α3 ) and g = (x − β1 )(x − β2 )).

This is not immediately obvious, but we’ll show the proof. The idea is that the coefficients ai and bj are elementary
symmetric polynomials in the roots αi , βj , so there must be a way to write this resultant in terms of ai and bj .
∏ ∏
Proof. Let αi , βj be our roots, and let f = (x − αi ) and g = (x − βj ). The resultant Res(f , g) is a polynomial in
the roots, and we divide this polynomial by (αi − βj ), which is a monic polynomial in the αi . Then

Res(f , g) = (αi − βj )q + r,

9
where q, r are polynomials in αi , βj and r has degree 0 in αi . Now if we set αi = βj , Res(f , g) = 0 (because we have
a common root), so r = 0 as well. But r can’t depend on αi , so r = 0, and therefore αi − βj divides the resultant for
all i, j.
To finish, we just need to check that the degrees are the same – we claim that the resultant as a function of αi
and βj is homogeneous with degree deg(f ) · deg(g). (This would show that we’re within a constant factor, and we
can just use f (x) = x m and g(x) = x n − 1 to check that constant factor. Then computing the determinant is not too
bad – it’s going to be ±1.) We’ll come back to that next time.

For the rest of today, we’ll talk about the discriminant of a polynomial.

Definition 13
The discriminant of a polynomial f is the resultant of f with its derivative f ′ .

For example, if f (x) = ax 2 + bx + c, then f ′ (x) = 2ax + b, and we have

 
a b c
 
discr(f ) = det  
2a b 0 = ab + 4a c − 2ab = 4a c − ab = −a(b − 4ac).
2 2 2 2 2 2

0 2a b

Often, people get rid of the a by working with a monic polynomial, and there’s the extra negative sign, but the whole
point is that this looks a lot like the b2 − 4ac we’re familiar with.
What does the discriminant tell us? If f has a double root, then f ′ will share a root with f , and the discriminant
will be zero. But also, if f is a function of t and x, we can plot it in the tx-plane – then the discriminant of f with
respect to x is zero for values of t where there is a “vertical” tangent or singular point (because there’s a double root
in x at that fixed value of t).

Proposition 14
Let f be monic with roots α1 , · · · , αn . Then
∏ ∏
discr(f ) = (αi − αj ) = f ′ (αi ).
i̸=j i

(Both (i , j) and (j, i ) appear in the product.)

The logic here is the same as before – if the discriminant vanishes with αi = αj , then the discriminant as a
polynomial in the roots must include (αi − αj ) as a factor for all i , j. For example, in the case where f is a quadratic,

f = (x − α1 )(x − α2 ) =⇒ discr(f ) = −(α1 − α2 )2 .

As an application, we’ll compute the genus of a smooth curve in the plane – let C be a smooth curve in P2 of
degree d using coordinates (x, y , z ), and we choose a generic point q ∈ P2 . We’ll project from P2 down onto P1 :
changing coordinates so that q = (0, 0, 1), we draw lines from (0, 0, 1) to points on the curve and see where they
intersect P1 (x, y ). Then the discriminant with respect to the variable z is 0 at the points p where the lines connecting
q and p are tangent to the curve.
We’ll assume for now that the points of tangency are double roots – this is a tricky point, and we’ll talk about it
next time. But the degree of the discriminant is d 2 − d in the roots (take any pair of roots, and they’re included twice

10
in the product), so the projection has d 2 − d branch points. Then the Euler characteristic of the curve is

e(C) = d · e(P1 ) − (d 2 − d) = 2d − (d 2 − d) = 3d − d 2 ,

because the projective line P1 is just a sphere. Because the Euler characteristic e(C) = 2 − 2g, this tells us that the
genus of a smooth curve of degree d is
1
g= (d − 1)(d − 2).
2
Let’s make a quick table:

d 1 2 3 4 5 6
g 0 0 1 3 6 10

This is interesting, because the genus of a curve can’t be 2 or 4 or 5 if it’s smooth. Also, let’s compute the degree
of the dual curve – the interesting points are where the lines from q is a tangent line to C. If we go over to the dual
space P∗ and look at our curve C ∗ , we have a line q ∗ which intersects it – the degree of C ∗ is just the number of
intersections with the generic line q ∗ .
Well, every intersection L∗ between C ∗ and q ∗ is some tangent line L to the original curve C, so the degree of the
dual curve is the same as the degree of the discriminant! Thus, the degree of C ∗ , the dual curve, is equal to d 2 − d.
(And this verifies that the degree of the dual curve for a generic cubic is 6.)

5 February 12, 2020

The due date for problem set 1 was pushed back to Friday, because we’ll cover a few more ideas today that will be
interesting.
Say that we have an affine curve C corresponding to f (x, y ) = 0, and we separate out the different degrees via

f = f0 + f1 + · · · + fd .

Let p = (0, 0) – then p ∈ C if f0 = 0, and p is a smooth point of C if f0 = 0 and f1 6= 0. Meanwhile, if f0 = f1 = 0

but f2 6= 0, then p is called a double point – in general, a point has multiplicity k if f0 , · · · , fk−1 are all zero, but
fk (p) 6= 0.
We’ll look at specifically at double points: then we can write

f = (ax 2 + bxy + cy 2 ) + (dx 3 + · · · ).

y
We’ll analyze by “blowing up” the xy -plane: let t = x, which means that we substitute in y = tx. Then we have a
map π from the xt-plane to the xy -plane given by (x, y ) = (x, tx): the t-axis gets sent to the origin under π. So this
is a blowup because the origin gets sent to a line, and the y -axis doesn’t get hit except at the origin, but everything
else goes bijectively. (And horizontal lines in the xt-plane go to lines through the origin.)
So if we substitute y = tx into our polynomial f , we get a polynomial (we can get rid of the x 2 factor throughout)

f (x, tx)
g(x, t) = = a + bt + ct 2 + dx + · · · ,
x2
where everything else after the dx term is divisible by x. The constant term (if we look at g as a function of x) is
q(t) = a + bt + ct 2 . If q(t) has distinct roots, and we normalize our function so that c = 1 (when it’s 0, just change
the coordinates slightly), we have
g(x, t) = (t − α)(t − β) + dx + · · · .

11
Then the partial derivative with respect to t is

gt = 2t − α − β + x(· · · ).

If we look at the points in the (x, t)-plane where g is zero, we have (0, α) and (0, β), so the curve goes through those
two points. But gt (0, α) = α − β 6= 0, and we can solve g(x, t) = 0 to find an analytic function t = u(x) for v (0) = α
(for small x). Similarly, we can find an analytic function t = v (x) with v (0) = β – the point is that the projection π
to the xy -plane gives us two lines through the origin. This means the curve intersects itself, which is called a node.
What about the case where q(t) has a repeated root? Then g looks like

g(x, t) = (t − α)2 + x(d + · · · ).

Now gt (0, α) = 0, but the partial with respect to x, gx (0, α) is nonzero if d 6= 0. So we can again solve x = w (t)
for small t: it’ll have a vertical tangent at α = 0 in the xt-plane. Projecting this back down to the xy -plane gives us
a cusp. And the last case is where we have a double root, and we also have d = 0: then the locus g = 0 is singular,
so the conclusion is that the blowup curve g = 0 is smooth above a point p if and only if the singularity is a node or a
cusp.
Let’s look a bit more at the geometry of the cusps – the standard cusp is the point (0, 0) for the curve y 2 −x 3 = 0.
All cusps are analytically equivalent to this one, so we’ll look a bit at the geometry here. We parameterize via x = t 2
and y = t 3 : the question we’re asking is “what does the cusp look like if we slice across a cross-section?”
To figure this out, let t = e iθ . Then x = e 2iθ and y = e 3iθ , so as θ ranges from 0 to 2π, x goes around the unit
circle twice, while y goes around three times. The product of the two unit circles is a torus, so this gives us a picture
on the surface of the torus traced out by the cusp as seen below:

If we look at this more closely, it’s actually a trefoil knot! So that’s what a cusp looks like on the cross-section
|x| = |y | = 1. (But in four real dimensions, it doesn’t really make sense to call it a “knot” anymore.) We could also
think about the unit sphere xx + y y = 1 and intersect it with a cusp, and do a stereographic progression from the
3-sphere into 3-space – this also gives a kind of knot.
We’ll now move on: Hensel’s lemma will be our next topic. Let f (x), g(x) be polynomials; we can write out the
polynomial p(x) = f (x)g(x):

f (x)g(x) = (a0 x 3 + a1 x 2 + a2 x + a3 )(b0 x 2 + b1 x + b2 ) = c0 x 5 + c1 x 4 + c2 x 3 + c3 x 2 + c4 x + c5 = p(x).

The coefficients ck can be computed in terms of the ai and bj :

12
 c0 = a0 b0
c1 = a0 b1 +a1 b0
c2 = a0 b2 +a1 b1 +a2 b0
c3 = a1 b2 +a2 b1 +a3 b0
c4 = a2 b2 +a3 b1
c5 = a3 b2 .

These are the “product equations,” and it’s an interesting question to ask whether we can solve for the ai and bj .
One useful idea is the Jacobian criterion: say that f and g are monic, so that a0 = b0 = 1. Then let the Jacobian
matrix be
∂(c1 , c2 , c3 , c4 , c5 )
J= .
∂(b1 , b2 , a1 , a2 , a3 )
We can calculate the entries of this matrix explicitly:
 
a0 0 b0 0 0
 
a1 a0 b1 b0 0
 
 
J = a2 a1 b2 b1 b0  .
 
a a 0 b2 b1 
 3 2 
0 a3 0 0 b2

This is the transpose of the resultant matrix R. Thus, the Jacobian matrix is singular (has determinant 0) if and only
if f and g have a common root, which gives us the following result:

Corollary 15 (Hensel’s lemma)

Suppose that the coefficients ai , bj are analytic functions of t, and let ai = ai (0), bj = bj (0). Here f = f (t, x)
and g = g(t, x) are polynomials in x with coefficients that are polynomials in t, and p = f g: let f = f (0, x) and
define g, p similarly. Then if f , g have no common roots, and f is monic, then there exist f , g such that

p(t, x) = f (t, x)g(t, x)

and p = f g for small t.

Proof. Since f is monic, we have a0 = 0, so the leading coefficient b0 for the other factor is also nonzero. Now
the Jacobian matrix is nonsingular, so the Implicit Function Theorem tells us that there is a unique solution for the
remaining coefficients for small t, as desired. (If we don’t know what the Implicit Function Theorem is, we should
read this on our own. Here, we’re treating different powers of x as different “functions.”)

Basically, if we have some curve p(t, x) = 0 and we’re interested in how it looks near t = 0, suppose that p(0, x)
factors into two polynomials with distinct roots. Then Hensel’s lemma says that we can factor p(t, x) near 0.
The simplest example is if we have two polynomials
?
c0 (t)x 2 + c1 (t)x + c2 (t) = (x + a1 )(b0 x + b1 ),

where a1 , b0 , b1 are functions of t. The product equations are

c0 = b0 , c1 = a1 b0 + b1 , c2 = a1 b1 .

We’re given that p = p(0, x) factors as (x + a1 )(b0 x + b1 ), and those two linear factors don’t have a root in common.

13
This means that a1 6= b1
b0
, and Hensel’s lemma tells us that p factors via

p(t, x) = (x + a1 (t))(b0 (t)x + b1 (t)).

6 February 14, 2020

We’ll talk about the Plücker formulas today: they count things like bitangents, nodes, and cusps. The formulas are
completely unimportant, but the interesting thing is that there are formulas (and they don’t depend on anything except
for degrees).
Specifically, we’ll work with a degree d smooth curve C which is ordinary, which means the following things:

• Flex points are ordinary (we have a triple intersection but not higher).

• Bitangents are ordinary – neither point is a flex point, and there are no tritangents.

This means that our map t : C → C ∗ from our curve to the dual curve is defined everywhere. Pick a point not on
the curve, and choose coordinates so that q = (0, 0, 1). For every point p on the curve, draw a line from q to p, and
project this down to the xy -plane.
What does this look like in the dual plane? We have a special line q ∗ which will intersect our curve C ∗ , and it
intersects at points L∗ corresponding to tangent lines in our original plane. In other words,

deg C ∗ = #(C ∗ ∩ q ∗ ),

and this is the number of points p̃ which are images of tangent lines from q.
We can calculate this using the discriminant: at each tangent line from q, the discriminant should vanish, because
we have a repeated root. So the degree of C ∗ is the number of zeros of the discriminant f with respect to one of the
variables z. All of these zeros are simple zeros, and the discriminant of f with respect to z is the resultant
( )
∂f
Res f , .
∂z

f has degree d, and ∂f

∂z has degree d − 1, so the resultant has degree deg C ∗ = d(d − 1) – this is our first result.
We have some other numbers that we care about – the number of flex points f of C, as well as the number of
bitangents b. First, we can count f :

Lemma 16
A smooth point p of our curve
( 2 C ) is a flex point if and only if the determinant of the Hessian matrix Hp (at the
∂ f
point p) is 0, where Hp = ∂xi xj .
1≤i,j≤3

Proof. Remember that we defined the bilinear form

hu, v i = u t Hp v

where we’re looking at the points as column vectors. Then we know that p is a flex if and only if the first three terms
of the Taylor expansion
1 1 1
f (t + pq) = hp, pi + hp, qit + hq, qit 2 + O(t 3 )
d(d − 1) d −1 2

14
vanish. If f is a flex point, then the form must be degenerate, because it’s operating on a three-dimensional vector
space where p and q are independent vectors. On the other hand, if the determinant is zero, there is a null vector, so
there exists q which satisfies this equation above.

With this, how can we count the flex points? Let H be the Hessian curve {p : det Hp } = 0: we wish to compute
the degree of H. H is a 3 by 3 matrix, and all entries have degree d − 2. This means that the determinant det Hp has
degree 3(d − 2).
Here, we’re going to use a useful result:

Theorem 17 (Bézout)
Let X and Y be curves of degree m and n in projective space. Then the number of intersections is mn if
intersections are transversals (not double multiplicity).

(We’ll prove this using cohomology later on in the class – it’s not a very deep theorem, but it’s interesting.) Also,
we’ll use a result which is ugly to prove (mostly just computation):

Lemma 18
If p is an ordinary flex of our curve C, then C intersects the Hessian divisor curve H transversely at p.

So now, if the degree of H is 3(d − 2), and the degree of C is d, the number of flexes is just the number of
intersections:
f = 3d(d − 2) = 3d 2 − 6d

(as long as d ≥ 2). It’s a fact that flexes map to cusps in the dual curve, and bitangents map to nodes, but this isn’t
obvious. (We’ll assume it for now.) Define δ ∗ to be the number of nodes of C ∗ and κ∗ to be the number of cusps:
then δ ∗ = b, κ∗ = f .
Counting the number of bitangents is a bit harder, and there isn’t a very easy way to directly compute them on
our curve C. So instead, let’s count the number of nodes on C ∗ ! Remember that the bidual gives us back the original
curve: (C ∗ )∗ = C, as long as we take closures. Choose a generic point Q in the dual plane, and project the curve from
Q onto the xy -plane (still in the dual plane). Then there are three types of interesting lines: (1) tangent lines to our
dual curve, (2) lines through a node, and (3) lines through a cusp.
Remember that we have some equation ϕ(u, v , w ) = 0 for our dual curve, and the discriminant with respect to
one of the variables v has simple zeros for case (1), double zeros for case (2), and triple zeros for case (3). And this
gives us a formula: we know the degree of C ∗ is d ∗ = d(d − 1), so the degree of the discriminant is

deg discrv (ϕ) = d ∗ (d ∗ − 1).

How many tangents do we have (case (1))? Remember the tangent lines correspond to intersections in the dual with
our generic line q ∗ . So because we’re working with the dual curve here, we should have d = deg C total tangent lines.
So our equation is now
d ∗ (d ∗ − 1) = 1 · d + 2 · δ ∗ + 3 · κ∗ .

But we know d ∗ and κ∗ , so this allows us to compute δ ∗ = b: substituting in δ ∗ = b and κ∗ = f = 3d 2 − 6d, we have
that
(d 2 − d)(d 2 − d − 1) = d + 2b + 3(3d 2 − 6d).

15
Expanding this out tells us that
d 4 − 2d 3 − 9d 2 + 18d = 2b.

This is zero for d = 3 – cubics can’t have bitangents, so the first interesting case is d = 4:

256 − 128 − 144 + 72 = 2b =⇒ b = 28,

which is the result we’ve been talking about for a while.

So how do we show that there are double zeros for nodes and triple zeros for cusps? The main idea is to use
Hensel’s lemma. Remember that if we have a curve f (x, y ) = 0 in the plane and we project a node down onto the
x-plane, we get a double zero. Say that the node is at (0, 0): then we know that

f (y ) = f (0, y ) = y 2 h(y )

for some other polynomial h. (The other zeros of h come from the other intersections of x = 0 with our curve f = 0.)
Then Hensel’s lemma tells us that we can also factor the multivariable polynomial

f (x, y ) = g(x, y )h(x, y )

for small x, where g(0, y ) = y 2 and h(0, y ) = h. What do we know about the discriminant of f ? Treating x as a
parameter,
∏ ∏ ∏
discr(g(y )h(y )) = (αi − αj ) (βk − βℓ ) (αi − βk ).
i,j k,ℓ i,k

where αi are the roots of g and βk are the roots of h. But the first two terms are the discriminants of g and h, and
the last term is the resultant. So
discr(f ) = Res(g, h)discr(g)discr(h).

So if we avoid accidents, the discriminant of the polynomial h and the resultant of g and h will be nonzero at x = 0.
So the order of vanishing of g is the same as the order of vanishing of f , and then we compute things for a quadratic
using the usual formulas.
Starting next class, we’ll move on to the second chapter – basic algebraic geometry for affine varieties.

7 February 18, 2020

We’ll start by talking about the Zariski topology today on the affine space An = {x = (x1 , · · · , xn ) : xi ∈ C}. Denote
C[x] = C[x1 , · · · , xn ], and let f = (f1 , · · · , fk ) be a set of polynomials fi ∈ C[x]. Then we’ll define

V (f ) = {x ∈ An |f (x) = 0, i.e. fi (x) = 0 ∀i }.

These are known as the Zariski closed sets. There are a few things to understand about them, none of which are
particularly difficult to prove: for example, if I is the ideal generated by f , often denoted (f ) (remember this is vector
notation), then I consists of
I = {g1 f1 + · · · + gk fk , gi ∈ C[x]}.

Then it’s clear that if the fi s are zero at a point, then everything in the ideal is also zero at that point, so V (f ) = V (I).
(The Basis Theorem tells us that every ideal in C[x] will look like (f1 , · · · , fk ).) Also, we have a nesting property: if
I ⊃ J, then V (I) ⊂ V (J). (The inclusions are reversed because having more functions is more restrictive, so there are
less points in V (I).)

16
 Proposition 19
The Zariski closed sets are closed in the Zariski topology.

Proof. We need to check that ∅ and An are Zariski closed, and if Cj are closed, the (possibly infinite) intersection
∩
Cj is Zariski closed. We also need to show that C ∪ D is closed (which shows this for finite unions as well). All of
these are pretty obvious except the last one.
∅ is the zero locus V (1) (it’s where 1 = 0) and An is the zero locus V (0) (it’s where 0 = 0). To show that
intersections are closed, say that Cj = V (Ij ): then
∩ (∑ )
Cj = V Ij ,
j
∑
because the intersection of the zero loci are where all the polynomials in each of the Ij s are zero. Since Ij is also an
∩
ideal, Cj is also Zariski closed, as desired.
The last one – showing C ∪ D is closed – is slightly nontrivial: we’ll need to notice something about two ideals,
which is that
I ∩ J ⊃ IJ ⊃ (I ∩ J)2 .

(Here, IJ is the product ideal, which is the sums of products of elements of I and J.) From these inclusions, we can
see that if C = V (I) and D = V (J), then

V (I ∩ J) ⊂ V (IJ) ⊂ V ((I ∩ J)2 ),

but the zero locus of an ideal squared is the same as the zero locus of the ideal. Therefore, we actually have

V (I ∩ J) = V (IJ).

With this, we want to show that C ∪ D = V (IJ). We know that I ⊃ IJ, so V (IJ) ⊃ C, and similarly V (IJ) ⊃ D, so
this means V (IJ) ⊃ C ∪ D.
To show the other direction, we want to show that if x ∈ V (IJ), then x ∈ C or x ∈ D. We know that for all f ∈ I
and g ∈ J, if x ∈ V (IJ), then f (x)g(x) = 0. If f (x) = 0 for all f ∈ I, then x ∈ C. Otherwise, there exists an f such
that f (x) 6= 0. Using this f and looking over all g ∈ J, we see that we must have g(x) = 0 for all g, meaning x ∈ D.
This shows that V (IJ) ⊂ C ∪ D, and we’re done.

Remark 20. It’s important to keep in mind that closed sets in the Zariski topology look very different from what we
usually call “closed sets.”

One nice property of C[x] is that it is Noetherian: any increasing chain of ideals I1 < I2 < · · · must be finite.
What does this tell us about the Zariski closed sets? Let Cj be Zariski closed sets, and let Ij = {f |f = 0 on Cj }. This
is an ideal, and C1 > C2 if and only if I1 < I2 . This gives us an analogous condition to the one on ideals:

Proposition 21
The Zariski topology has the descending chain condition on closed sets: if C1 > C2 > · · · is a decreasing chain,
it must be finite.

Zariski open sets are the complement of Zariski closed sets, so we get an ascending chain condition for open sets
as well.

17
 Definition 22
A topological space is Noetherian if it has a descending chain condition on closed sets.

If X is a topological space and S is a subset of X, then S becomes a topological space with the induced topology:
T ⊂ S is closed if T = S ∩ C and C is a closed set in X. (And restricting to subsets of X will still give us the
descending chain condition.)

Definition 23
A topological space X is irreducible if it is not the union of two proper closed subsets.

In other words, if C and D are closed in an irreducible space X, and X = C ∪ D, then C = X or D = X.

Proposition 24
Let X be an irreducible topological space. If U ⊂ X is a (nonempty) open set, then the closure of U (smallest
closed set containing U) is X.

Proof. Say that U is the closure of U. Since X is irreducible, if we let C = X − U be the complement of U,

X = U ∪ C =⇒ X = U ∪ C.

And now C is not X, because that means U is nonempty, so U must be X.

Definition 25
A topological space X is connected if it is not the union of two proper disjoint closed sets.

It’s hard to satisfy irreducibility, because we don’t require our two closed sets C and D to be disjoint! For example,
two intersecting lines form a connected space but not an irreducible space.

Theorem 26
Let X be a Noetherian space. Then X is a finite union of irreducible spaces.

Proof. We prove the contrapositive. Suppose that X = C0 is not a finite union of irreducible spaces: then we can
write C0 = C1 ∪ D1 , where C1 , D1 are proper closed subsets of C0 . Then one of C1 and D1 is not a union of finitely
many irreducible spaces: say that C1 is not. Then we know that C0 > C1 , and now we repeat this argument with C1
instead of C0 . But this gives us an infinite descending chain of closed sets, so X is not Noetherian.

Definition 27
A Zariski closed subset X of An is a variety if it is irreducible.

We’ll ask a question (which we’ll answer next time): if I and J are ideals, when is V (I) = V (J)? It’s clear that I
and J don’t need to be equal: I can be generated by f and J can be generated by f 2 . But it turns out that this is
really the only reason that two ideals would have the same variety:

18
 Definition 28
The radical of an ideal I, denoted rad I, is the ideal

rad I = {f : f n ∈ I for some n > 0}.

rad(I) contains I, and the zero loci are the same: V (rad I) = V (I) (because f is zero if and only if f n = 0). This
radical is the key to answering our question, but it’ll take some work to prove:

Theorem 29
V (I) ⊃ V (J) if and only if rad I ⊂ rad J.

The backwards direction is clear – if rad I is contained in rad J, then V (rad I) contains V (rad J), so V (I) contains
V (J). But the other direction is something we’ll show tomorrow.
Another question we may want to ask: when is a Zariski closed set a variety?

Theorem 30
X = V (I) is a variety (in other words, it is irreducible) if and only if rad(I) is a prime ideal.

Recall that an ideal P is prime if

ab ∈ P =⇒ a ∈ P or b ∈ P.

Alternatively, if A, B are ideals,

AB ⊂ P =⇒ A ⊂ P or B ⊂ P.

Let’s do an example:

Example 31
Consider a map ϕ from A2x,y to A4t,u,v ,w of the form

ϕ(x, y ) = (x 3 , y 3 , x 2 y , xy 2 ).

There are some relations here: notice that tu = v w , v 3 = t 2 u, and w 3 = tu 2 . These three relations generate
some ideal I. Let’s show that the radical rad I is a prime ideal: we want to show that the locus of zeros X = V (I)
here is irreducible.
To do this, we’ll show that ϕ maps A2 surjectively to X. Given t, u, v , w which satisfy these relations, we just need
to find x and y : we can just set x = t 1/3 , y = u 1/3 . There are three choices for each of x and y , and we just need to
check that ϕ(x, y ) = (t, u, v , w ) – as long as our relations are satisfied, there is a choice of x and y that work here.
With this, it is easy to show that X is irreducible. Suppose otherwise: then its inverse image is the union of two
closed sets. But A2 is irreducible, which is a contradiction! So we have indeed shown that rad I is a prime ideal. And
in this particular case, rad I = I, but it requires more work to show this.

8 February 19, 2020

We’ll start with a bit of review: in the Zariski topology on the affine space An , the closed sets are of the form V (I)
for an ideal I in C[x1 , · · · , xn ]: they are the points p which vanish for all polynomials f ∈ I. (By the Basis Theorem,
all such ideals I are finitely generated.)

19
 We know that An is a Noetherian topological space – basically, it has the descending chain condition on closed
sets, which follows from the ascending chain condition on ideals. Thus, every closed set can be decomposed into
a finite union of irreducible closed sets. (Recall that X is irreducible if it cannot be broken up into proper subsets
C ∪ D.) Then an affine variety is an irreducible closed set under the Zariski topology.
At the end of last class, we defined the radical

rad I = {g : g n ∈ I for some n ≥ 1}.

We know that I ⊂ rad I, and also

V (I) = V (rad I)

because a polynomial f has the same zeros as f n for any positive integer n. We’ll be proving some things today that
yield the following results:

Proposition 32
We have
• V (I) ⊃ V (J) if and only if rad(I) ⊂ rad(J).

• Let P be a radical ideal (meaning rad P = P ). Then V (P ) is a variety if and only if P is a prime ideal.

Here’s the first main result:

Theorem 33 (Hilbert Nullstellensatz)

Let C[x] = C[x1 , · · · , xn ]. There is a bijective correspondence between the following three sets:
• Points p of affine space An ,

• Homomorphisms πp from C[x] to C,

• Maximal ideals mp of C[x].

Basically, πp evaluates a polynomial f ∈ C[x] at p, and mp is the kernel of πp .

This comes from 18.702, so we won’t go into it in much detail. For example, it’s clear that every homomorphism
is surjective, so the kernel is a maximal ideal. If we’re looking at a point p = (a1 , · · · , an ), then the maximal ideal mp
is generated as
mp = (x1 − a1 , · · · , xn − an ).

Definition 34
An algebra A is a ring that contains the complex numbers C. A finite-type algebra A can be generated as
an algebra by some finite set of elements {α1 , · · · , αn }, where αi ∈ A, so that all elements can be written as
polynomials in αi with coefficients in C.

Another way to define a finite-type algebra is to consider the map

τ : C[x1 , · · · , xn ] → A

which sends xi to αi . Then this map should be surjective if we can write every element of our algebra as a polynomial
in the αi s, and thus we can write
A∼
= C[x1 , · · · , xn ]/I

20
for some ideal I = ker τ by the First Isomorphism Theorem.

Fact 35
A finite module over a ring means that every element in the module is a linear combination of our generating set,
but a finite-type algebra is different – it lets us write general polynomials! So we should be careful not to confuse
the two.

Theorem 36 (Hilbert Nullstellensatz, version 2)

Let A be a finite-type algebra. Then there is a bijective correspondence between homomorphisms π : A → C and
maximal ideals mp of A.

Proof. This is basically the first version but applying the Correspondence Theorem. We can present

A = C[x1 , · · · , xn ]/I,

and then our homomorphisms π : A → C correspond bijectively to homomorphisms π : C[x] → C where ker π contains
the ideal I. (And the rest of the argument follows analogously.)

So if A = C[x]/I, the two sets in the above theorem (maximal ideals of A and homomorphisms to C) also correspond
bijectively to the locus of zeros of I in An . We’ll actually use this more: A has to be a finite-type algebra, but we
don’t have to express it explicitly in the form C[x1 , · · · , xn ]/I for the result to hold. The point is that we can skip the
presentation altogether:

Definition 37
Let A be a finite-type domain (no zero divisors). The spectrum of A, denoted Spec A, is a set of points where
we put p into Spec A for every maximal ideal m = mp of A.

By definition, then, the set of points of Spec A correspond to the maximal ideals mp , which correspond to the
homomorphisms π p . So if we have a presentation A = C[x1 , · · · , xn ]/I, then Spec A is just the zeros of the ideal V (I).

Theorem 38 (Strong Nullstellensatz)

Let f1 , · · · , fk , g ∈ C[x1 , · · · , xn ], and let V = V (f ) be the set of zeros of (all of) f1 , · · · , fk in An . Then if g is
identically zero on V , then some power of g is in the ideal I = (f1 , · · · , fk ):

g n = h1 f1 + h2 f2 + · · · + hk fk ,

where hi ∈ C[x].

Proof by Rainich. Add another variable y to the polynomial ring to get C[x1 , · · · , xn , y ], and let W be the locus of
zeros in An+1
x,y where f1 = f2 = · · · = fk = 0 and gy − 1 = 0. (In other words, g(x1 , · · · , xn )y = 1.) Consider some
(x, y ) ∈ W in the zero locus: then
f1 (x) = · · · = fk (x) = 0

implies that g(x) = 0 as well (because g is identically zero on V ). But then we can’t solve gy = 1, so the locus W
is empty. Now, what can we say about a set of polynomials (specifically (f1 , · · · , fk , gy − 1) here) whose zero locus
is empty? We’ll use a quick lemma here:

21
 Lemma 39
Any ideal I < R of a ring R is contained in a maximal ideal.

Using that fact, note that maximal ideals correspond to points in An , so if there are no points in the zero locus,
the ideal I generated by f1 , · · · , fk , gy − 1 must be the whole ring R (it’s not contained in a maximal ideal). Thus,
there exist p1 (x, y ), · · · , pk (x, y ), q(x, y ) ∈ C[x, y ] which satisfy

p1 f1 + · · · + pk fk + q(gy − 1) = 1.

Now let’s go to the ring R = C[x][y ]/(gy −1): this means that the residue of y is g −1 . (In other words, we’ve adjoined
an inverse of g(x) to C[x].) We can suppose g 6= 0 (or the statement is trivial), and thus C[x] ⊂ R.
But we also have our relation equation above, so in our ring R, gy − 1 = 0. This means that in R, we have

p1 f1 + · · · + pk fk = 1.

The pi s are polynomials in x and y , but we can get rid of y = g −1 by multiplying by a sufficiently high power of g. So
if we multiply by g N for some N and cancel the y ’s, we have a relation

h1 (x)f1 (x) + · · · + hk (x)fk (x) = g(x)N ,

which is exactly what we want – this means g N is in the ideal generated by f , as desired.

The first two results that we stated today, Proposition 32, can now be deduced from the Strong Nullstellensatz
pretty easily – this is an exercise!
The idea here is that we’re localizing our ring, which means to adjoin an inverse. What’s the meaning of this
name? In topology, something is true locally if it’s true in some open neighborhood. There are some open sets that
we understand well, and they are important – this proof illustrates some of the power of this startegy here!

9 February 21, 2020

We’ll start with a bit of review: recall that a finite domain A is isomorphic to some C[x]/P , where P is a prime
ideal. We define X = Spec A to contain the points p corresponding to maximal ideals mp , which correspond to
homomorphisms πp : A → C. Also, it’s good to remember that Spec A for a domain A = C[x]/I corresponds to the
variety V (I).
Elements α ∈ A determine functions on X = Spec A via the correspondence between points p in the zero locus
and maximal ideals π p :
α(p) = π p (α).

These elements of A are called the regular functions on X.

Lemma 40
The function πp (α) determines the element α. Equivalently (because πp is a homomorphism), if a function f is
zero, then α = 0.

Proof. If A = C[x]/P , then X = VAn (P ) in affine space An is the zeroset of the ideal P . (Note that the correspondence
theorem tells us that maximal ideals of A are the same as maximal ideals of C[x] that contain P .) So a function f

22
that is zero on X is in P , so it must be the zero element α in A.

Definition 41
In the Zariski topology on an affine variety X, a closed set is the set of zeros of some ideal I of A.

For the sake of notation, denote VX (I) to be the zeroset of I: p ∈ VX (I) means that all elements α ∈ I are zero
at p: this means that α is in the kernel of πp , and we can correspond this to the maximal ideals via

VX (I) = {p : I ⊂ mp }.

One important operation which we briefly talked about last time is the idea of localization. Let A be a finite
domain, and let X = Spec A. Pick a nonzero element s ∈ A – we can adjoin an inverse to A, which we denote as

As = A[s −1 ] = A[z]/(zs − 1).

As is called the localization of A, and Xs = Spec As is similarly called the localization of X.

Proposition 42
Xs corresponds to points of X at which s(p) 6= 0, and it’s an open subset of X because it’s the complement of
s = 0. Then the Zariski topology on Xs is the induced topology (a closed set of Xs is a closed set in X, intersected
with Xs ).

Proof. Let p ∈ X be a point, and we correspond this to the homomorphism πp : A → C. If s(p) 6= 0, then we can
extend πp to a homomorphism As → C in a unique way: we know where the elements of A go, and the inverse s −1
goes to the inverse of s(p), which determines the homomorphism. On the other hand, if s(p) = 0, we can’t define
where the inverse goes. So whenever we have a homomorphism πp , we have a point p ∈ Spec As .
For the topology, let D be a closed set in Xs : this means D is the set of zeros of some set of elements of As . But
an element of As looks like as −n , where a ∈ A, so we have some set

{a1 s −n , · · · , ak s −n }

of polynomials. (We can use the same exponent for all of the ai .) This has the same zeros as {a1 , · · · , ak } in Xs ,
because s doesn’t have any zeros. So if C is the zeroset of a1 , · · · , ak in X, we indeed have D = C ∩ Xs .

Localization is important for two main reasons: it’s easy to understand the relationship between A and As , and
the localizations form a basis for the topology on X. (Here, a family of open subsets is a basis for a topology if
every open set is a union of members of this family.) So any open set can be covered by localizations, but this doesn’t
necessarily mean all open sets are localizations!

Example 43
Let X = A2x,y , and let U = X − {0, 0}. This is an open set, but it is not a localization.

To show this, we need to find an element s such that we can get U by inverting s. But if s 6= 0, inverting it doesn’t
make the point disappear, so we need to invert a function that is zero at the origin – every such polynomial vanishes
on a curve.

23
 ∪
How do we show that any open set U can be covered by localizations? If we want to show that U = Xsi , we
should take complements: let C = X − U be a closed set. Then the complement of any localization, Zi = X − Xsi , is
the zeroset of si (because Xsi is the set of points where si is equal to zero). So we know that
∪ ∩
U= Xsi ⇐⇒ C = Zi ,

and by the definition of the Zariski topology, every closed set is the zeros of some elements in X, so we can indeed
write U as a union of localizations.
We’ll now move on to the concept of morphisms: let X = Spec A and Y = Spec B, and say that ϕ : A → B is
a homomorphism. Then define the map u : Y → X (associated to the homomorphism ϕ) as follows: for any point
q ∈ Y , we have a homomorphism πq : B → C, which we can compose with ϕ to give a map πq ϕ. By definition of
the spectrum of A, beause this is a homomorphism, it corresponds to some unique point p ∈ X = Spec A: define
u(q) = p.
What can we say about this to make it more clear what’s going on? Let α ∈ A and ϕ(α) = β ∈ B. We want to
evaluate α at a point p: by definition,

α(p) = πp (α) = πq ϕ(α) = πq (β) = β(q).

So we define u such that evaluating α at p is the same as evaluating β at q.

Definition 44
A morphism Y → X is a map determined by the algebra homomorphism ϕ : A → B as detailed above.

Let’s think about this when we’ve chosen a specific presentation

A = C[x]/(f ), x = (x1 , · · · , xn ), (f ) = (f1 , · · · , fk ).

Let τ be the canonical map from C[x] to A (modding out by the ideal (f )): composing this with ϕ gives us a map
Φ : C[x] → B. When can we tell that Φ comes from a map ϕ? Any homomorphism Φ comes from substituting in
some values bi for our xi s, and we need to make sure that everything in ker τ is also in ker Φ for the map ϕ to be
well-defined (this is both necessary and sufficient). So this means that Φ(fj ) = 0 for all j, meaning that fj (b) = 0.

Proposition 45
In other words, constructing a homomorphism from A = C[x]/(f ) to an arbitrary B, and thus constructing a
morphism from Spec B to Spec A, means we need to solve the equations fj (x) = 0 for x ∈ B.

Example 46
Let A = C[x, y ]/(y 2 − x 3 ), and let B = C[t]. Then we can solve the equation y 2 = x 3 in B: one solution is
y = t 3 , x = t 2 , and this defines a homomorphism A → B by doing exactly that (send (x, y ) to (t 2 , t 3 )). And we
know that Spec A is the cusp curve, and Spec B is the t-line, so we’ve found a morphism from the t-line into the
cusp curve.

Example 47
Let X = SL2 (C), and let A = C[a, b, c, d]/(ad − bc − 1). Then a map A → B means that we need to find a
B-valued matrix with determinant 1. And in each case, we’re finding a map into the special linear group.

24
 We can also look at the blowup of the plane (which we studied a few lectures ago): let A = C[x, y ], and let
B = C[x, w ]. We map A → B by sending (x, y ) → (x, xw ). Then we again have a morphism: Spec B maps to Spec A,
and the w -axis is mapped to the origin, while the rest of the y -axis is not in the image.

10 February 24, 2020

Today, we’ll talk about operations of a finite group on an affine variety, but let’s start with a bit of review. Recall that
if A and B are finite type domains, and X = Spec A, Y = Spec B, a homomorphism ϕ : A → B gives us a morphism
u : Y → X defined as follows: if q ∈ Y is a point, this corresponds to a homomorphism πq : B → C, which we can
compose with ϕ to get a map πp = πq ϕ. A way to think about this is that for any α ∈ A, we can define

α(p) = πp (α) = πq (ϕα) = [ϕα](q).

So now let G be a finite group of automorphisms of B. An element b ∈ B is invariant if σb = b for all σ ∈ G; then
the set of invariant elements, A = B G , is a subalgebra of B.
So for every automorphism σ : B → B, we get a morphism uσ : Y → Y (which we’ll just denote σ). Then G
operates on Y , but there’s a bit of a problem: suppose we compose two automorphisms together, so we get the maps
σ τ
B → B → B.

But the corresponding maps backwards look like

σ τ
Y ←Y ←Y

(going from algebras to varieties reverses arrows), which yields a different composition. So an easy fix is as follows:
let σ operate on the left on B, so we send b to σb, but let it operate on the right on Y , so q is sent to qσ. And this
fixes the problem: a map τ σ sends q ∈ Y to qτ σ.
So suppose we have an element β ∈ B: what can we say about

[σβ](q) ?

The definition of the regular function tells us that this is πq (σβ), and now this is a composition of functions (πq ◦σ)(β).
And now note that if p = qσ, then this is equal to πp (β) = β(p), so this does give us β(qσ) . (So this is a way to
move the σ back and forth.)
So if we map A = B G to B (for instance, via an inclusion), we have a map π from Y into X = Spec A, which can
be pretty interesting.

Example 48
Let B = C[y1 , · · · , yn ], which means Y = An , and let G = Sn operate on the indices. Then B G is generated by
the elementary symmetric functions, so

A = B G = C[s1 , s2 , · · · , sn ].

Then X = Ans is the affine space labeled by s.

If we have a point of X, which means we know the values of the symmetric functions si = ai , then the yi are the
roots of the polynomial x n − a1 x n−1 + · · · + (−1)n an . So the fibres of the map π from Y to X are the orbits of the

25
group G operating on Any : the standard notation for the set of orbits is to use X = Y /G.

Example 49
Let B = C[y1 , y2 ] and Y = Spec C[y1 , y2 ], and consider G = hσi to be the cyclic group generated by the map

σy1 = ζy1 , σy2 = ζ −1 y2 ,

where ζ = e 2πi/n .

The only invariants of σ are those where the difference of the exponents is 0 mod n, so

A = B G = C[u1 , u2 , w ]/(w n − u1 u2 ),

where u = y1n , u2 = y2n , w = y1 y2 . (It’s not hard to show that this generates all invariant polynomials.)
So suppose that we’re given u1 , u2 , w which satisfy the relation, and say u1 6= 0. Then there are n possible values
for y1 , and choosing this value fixes y2 (because we know the value of y1 y2 ). So the n values of y1 are the orbit of the
original value under G, and thus X = Spec A is again equal to the set of the G-orbits. (The edge case is the origin,
which is a fixed point under σ.)

Theorem 50
Let B be a finite-type domain, and let G be a finite group of automorphisms of B. Let A = B G be the invariant
ring, and let Y = Spec B, X = Spec A. Then we have an inclusion π : Y → X (because A is a subset of B). Then
• A is a finite-type domain, and B is a finite A-module.

• The fibres of π are the G-orbits of π: and we have a bijective map Y /G → X.

We’ll do proof by example:

“Proof”. Let Y = SL2 (C) be the set of 2×2 complex-valued matrices with determinant 1. Then B = C[a, b, c, d]/(ad−
bc = 1). Let G = hσi, where σ 2 = 1 and σ sends a point P ∈ Y to its inverse P −1 . In other words,
[ ] [ ]
a b d −b
7→ .
c d −c a

These two matrices are identified in the orbits of G. To show the first point in the theorem, we need to write
down some invariants: notice that t = a + d, u = b2 , v = c 2 , w = ad are all invariant under σ, and let R =
C[t, w , u, v ]/((w − 1)2 − uv ). (because ad − bc = 1 in B, we also have the relation (w − 1)2 = uv in R). Note that
we have the chain of inclusions R ⊂ A ⊂ B.
Now a, d are roots of x 2 − tx + w , and b, c are roots of x 2 − u, x 2 − v respectively. So let’s write B down as a
module over R: we have the elements 1, a, b, c, ab, ac, bc, abc (we don’t need d because we get it from t and a),
and any higher degree monomial in a, b, c will be divisible by at least one square, and we can use the equations to
reduce the degree (for example, a2 − ta + w = 0, so a2 = ta − w ). So everything in B can be written as a linear
combination of these eight boxed elements, with coefficients in R.
Since R is a finitely-generated algebra (it’s generated by t, w , u, v ), it’s Noetherian. Since A is an R-submodule of
B (which is a finite R-module), A is a finite R-module (this is property of Noetherian rings). To prove that A is a finite
type algebra, note that we have a finite number of generators of R as an algebra, and a finite number of generators
of A as an R-module. So we have a finite number of generators as an algebra, and we’ve proved our first point.

26
Remark 51. Note that R isn’t the invariant ring here: there are other invariant polynomials, like bc, which are not in
R.

For the second point, we need to first (1) note that all points of Y in an orbit have the same image, then show
that (2) points in different orbits have distinct images. Then we’ll show that (3) every point in X is an image of a
point of Y /G – that’ll show that we have a bijection.
(3) is the most interesting: we will show that the map from Y to X is surjective. Let p ∈ X: then we have the
maximal ideal mp of A at p. Because A is contained in B, we can look at mp B, the “extended ideal,” whose elements
are combinations of the form
= {z1 b1 + · · · + zk bk |zi ∈ mp , bi ∈ B}.

This is indeed an ideal of B, and every ideal except the unit ideal is contained in a maximal ideal. Say mp B 6= B: then
mp B is inside some maximal ideal mq in B corresponding to a point q ∈ Y .
So now we know that mp B ⊂ mq : consider mq ∩ A, which is an ideal of A (and not the unit ideal, because it
doesn’t contain 1). On the other hand, mp is contained inside mq ∩ A, and mp is a maximal ideal. So mp = mq ∩ A,
which means the map from Y to X indeed sends q to p, and we have a surjective map as desired.
∑
So we just need to check whether mp B can be the unit ideal: can we write 1 = zi bi ? The answer is no –
suppose for the sake of contradiction, and sum over the orbit:
∑ ∑
σ(1) = σ(zi )σ(bi ).
σ∈G i,σ

σ(1) = 1, and the z’s are in the maximal ideal mp , so σ(zi )s are in A (the invariant ring). Therefore, this is
∑ ∑
= zi σ(bi ) = zi αi ,
i,σ i
∑
where α = σ(bi ). So equating both sides, we have
∑
|G| = zi αi ,
i

where αi ∈ A (because it’s fixed under σ), zi ∈ mp . So |G| ∈ mp , but |G| is invertible in C, which is a contradiction
(because mp will contain 1).

The central idea of this proof is to take advantage of certain invariants, which allow us to turn each generator of
B into a finite set of generators of B as an R-module.

11 February 26, 2020

We’ll talk for about three lectures about projective varieties. Recall that Pn is represented by the points (x0 , · · · , xn ) ∼
(λx0 , · · · , λxn ): then a Zariski closed set in Pn is the zeroset of a family of polynomials f1 , · · · , fk , where each fi is
homogeneous in C[x0 , · · · , xn ]. And now, just as in affine space, we have

V (f ) = V (I), I = (f1 , · · · , fk ).

27
 Lemma 52
An ideal I in C[x] is generated by homogeneous polynomials if and only if the homogeneous parts of f (of each
degree) are in I.

Just like with An , since the polynomial ring C[x] has the ascending chain condition on ideals, Pn has the descending
chain condition on closed sets, so it is a Noetherian space. (In particular, this means that every closed set is a finite
union of irreducible closed sets). Analogous to the affine case, an irreducible closed set is a projective variety.
There’s just one thing to be careful about: the maximal ideal

M = (x0 , · · · , xn )

is called the irrelevant ideal, because its zeroset is empty – (0, 0, · · · , 0) is not a point in projective space.

Proposition 53
If X ⊂ Pn is a projective variety, and P is the ideal of polynomials

P = {f |f = 0 on X},

then P is a prime ideal (different from M).

(The proof is analogous to the affine case.)

Example 54
A hypersurface is the zeroset of a single irreducible homogeneous polynomial f (x) – this is a projective variety.

Example 55
x × Py is a map s : Px × Py → Pw , where we index our coordinates via
The Segre embedding of Pm n m n N

w = {wij }, 0 ≤ i ≤ m, 0 ≤ j ≤ n.

Then the point (x, y ) maps to w via wij = xi yj .

The dimension of this image space is N = (m + 1)(n + 1) − 1 (the −1 comes from us being in projective space).

Proposition 56
The Segre map is injective, and the image is the zero locus of the Segre equations

wij wkℓ = wiℓ wkj

for all indices i , j, k, ℓ.

Proof. First of all, these above equations obviously hold if wij = xi yj . Let w be a point in PN which satisfies the Segre
equations: then some wij 6= 0. We’ll say it’s w00 without loss of generality, and we’ll set w00 = 1 by scaling.
This means x0 , y0 are nonzero, and we’ll set x0 = 1, y0 = 1. Then

wij w00 = wi0 w0j

28
tells us that we should set wij = xi yj , and indeed this is the unique solution.

Example 57
Let m = n = 1. Then we have a map P1 × P1 → P3 sending (x0 , x1 ) × (y0 , y1 ) to (w00 , w01 , w10 , w11 ): then the
image is the zero locus of w00 w11 = w10 w01 , which means P1 × P1 is actually a quadric.

Example 58
The Veronese embedding of Pn is a map v : Pnx → PN
v , where we index coordinates via

v = {vij }, 0 ≤ i ≤ j ≤ n.

Then (x) maps to vij = xi xj .

Proposition 59
The Veronese map is injective, and its image is the zero locus of the equations

vij vkℓ = viℓ vij , 0 ≤ i ≤ k ≤ ℓ ≤ j ≤ n.

Example 60
We can send P1 → P2 under the Veronese embedding: (x0 , x1 ) maps to the zero locus (x00 , x01 , x11 ) where
2
v00 v11 = v01 . (And this means that P1 is a conic.)

The Veronese embedding uses a quadratic basis, but we could take cubics or higher-degree monomials:

Example 61
Take the cubic Veronese embedding, which maps P1 = (x0 , x1 ) to P3 = (z000 , z001 , z011 , z111 ) = (x03 , x02 x1 , x0 x12 , x13 ).
We’ll relabel the points in the image as (a, b, c, d): this forms a twisted cubic with the equations b2 = ac, ad =
bc, c 2 = bd.

[ ]
a b c
It turns out that these are the 2 × 2 minors of the matrix : the points of the twisted cubic are often
b c d
written as (1, t, t 2 , t 3 ) plus the point (0, 0, 0, 1). And this is equivalent to (u 3 , u 2 , u, 1) plus the point (1, 0, 0, 0).
where u = t −1 .
With this, let’s go back to the Segre embedding: we know that Pm × Pn gets put into the zero locus of some
homogeneous polynomial equations. Is Pm × Pn a variety?
We don’t really have a definition for this yet, but we can look at Pm × Pn ’s image in the Segre embedding and see
whether it’s a variety in PN . But this is not obvious, and we need to ask the question “what’s the Zariski topology in
PN ?”

Fact 62
If X and Y are varieties, then the Zariski topology on X × Y is not the product topology (unless one of X and Y
is a point).

29
 To show an example of this, consider P1 × P1 . The product topology is the coarsest topology (with fewest open
sets) which makes the projections continuous. The closed sets in P1 are just finite sets, so in the product topology,
the inverse image of a closed set in P1 × P1 should be closed. This means the product topology just requires finite
unions of “vertical/horizontal” lines and points to be closed, which is definitely not the Zariski topology.

Proposition 63
Let X and Y be irreducible topological spaces. Suppose there exists a topology on Π = X × Y such that
• The projections Π → X, Π → Y are continuous. (This means that the topology on Π is at least as fine as
the product topology.)

• For any y ∈ Y , the map from Πy = X × y → X is a homeomorphism, and similar for ΠX → Y . (This means
horizontal/vertical lines X × y and Y × x are “the same” as the original spaces X and Y .)
Then Π is an irreducible topological space.

(And this shows the result we want, because Pm and Pn are irreducible.)

Proof. Let C1 , C2 be closed in Π such that Π = C1 ∪ C2 . We work with the complements (open sets) W1 = Π − C1
and W2 = Π − C2 : it suffices to show that W1 ∩ W2 = ∅.
Let U1 be the image of W1 in Y , and let U2 be the image of W2 . Our first step is to show that if W is open in Π,
then the image of U in Y is open. Intersect W1 with a vertical fibre Πx = x × Y : then W ∩ Πx is open in Πx , and Πx
maps bijectively to Y . This means the image of Ux in Y is open, and now note that
∪ ∪
W = Wx =⇒ U = Ux

must also be an open set. So now we know that U1 , U2 are open – suppose that W1 and W2 are not both empty. Since
Y is irreducible, U1 ∩ U2 is not empty (or else the union of their complements in Y would be Y ). Let y ∈ U1 ∩ U2 : U1 is
the image of W1 , so there is some point (x1 , y ) in W1y and (x2 , y ) in W2y . But W1y , W2y are open in Πy , which maps
bijectively to X. But that means Πy is irreducible, which means W1y ∩ W2y 6= ∅, and thus W1 ∩ W2 6= ∅, contradiction
with the fact that Π = C1 ∪ C2 .
Thus either W1 or W2 is empty, which means C1 = Π or C2 = Π, as desired.

12 February 28, 2020

Today, we’ll be talking about morphisms for projective varieties. We know that algebra homomorphisms give us affine
varieties, but it’s harder to make a definition for projective varieties (because points are equivalence classes in Pn ).

Example 64
Let’s project a conic C onto a line X = P1 . We choose our coordinates so that we project through the point
(1, 0, 0) onto the line x0 = 0.

If we have a point p = (a, b, c) on the conic, then πp will just be (b, c), because the line containing p and (1, 0, 0)
is cx1 = bx2 . But what happens to the point q itself? We want to take a tangent line T at the point q, and see where
that intersects X. To do more, we’ll write out the explicit equation of the curve:

f = x0 x1 + x0 x 2 + x 1 x2 .

30
Then the tangent line is of the form
T : (∇f )q · x = 0,

where f0 = x1 + x2 , f = x0 + x2 , f2 = x0 + x1 . The gradient evaluated at q = (1, 0, 0) is (0, 1, 1); thus, a point x is on

the tangent line T if x1 + x2 = 0 and x0 = 0, which gives the point (0, 1, −1).
The central idea, though, is that there’s no way to write down a formula in polynomials that works for every point
on the conic! And there is no algebraic map from P2 to P1 except for constant maps, so there’s something tricky to
deal with here.
In general, when we look at projective varieties, we’ll call any nonempty open subset of a projective variety
a variety as well. (This is generally called a quasiprojective variety.) There are varieties that can’t be put into a
projective space, but we won’t talk about them here.

Fact 65
An affine variety is quasiprojective, because we have the standard affine open sets U j = {xj 6= 0} ⊂ Pnx . And then
∼ An (because we need to normalize one of the coordinates to 1); another way to say that is that (defining
Uj =
some new variables) [ ]
x0 x1 xn
U = Spec C [u0j , · · · , unj ] = Spec C
j
, ,··· , .
xj xj xj
Then an affine variety is a closed set in U j , which gives us what we want.

Next, let X be a projective variety in Pn , and let X j = X ∩ U j be a (closed) affine subset of U j , ignoring the indices
where X ∩ Uj = ∅. (This won’t happen if we pick generic coordinates.) Then Xi ∩ Xj ⊂ Xj is the subset of Xj where
xi 6= 0; that is, uij = xi
xj 6= 0.
We know that X j is the Spec of some algebra Aj , which is a quotient ring of C[u0j , · · · , unj ]. We’re doing some
localizing here: note that
X i = X j = Spec Aj [uij−1 ] = Spec Ai [uji−1 ].

Thus, the field of fractions of Aj is the same for all j:

Definition 66
The function field K of X is the fraction field of any algebra Aj where Xj 6= 0. A rational function α is a nonzero
element of K.

And open sets of this will still give the same function field: all open subvarieties of a (projective) variety have
the same function field.

Definition 67
If α is a rational function on X, and p ∈ X, then α is regular at p if there is some index j and some s ∈ Aj such
that α ∈ (Aj )s (adjoining s −1 to Aj ).

We’ll use these rational functions to define morphisms from a projective variety X to some projective space PN .

Definition 68
A point with values in K is the equivalence class of vectors α = (α0 , · · · , αN ), with αi ∈ K not all zero, where
scaling the vector by any λ ∈ K gives the same point.

31
 If p ∈ X is sent to some point with values in K, there are different cases that we should consider. If the αi are all
regular at p, and αi (p) are not all zero, then define our map α : X → PN via

α = (α0 (p), · · · , αn (p)).

On the other hand, if the αi are not all regular or are all zero at p, then we look for a λ ∈ K such that setting α′j = λaj
allows us to evaluate α′ at p.

Example 69
Let’s go back to our example from the beginning of class: what is the function field for the conic C?

Substituting our variables in, we find that (normalizing u0 = 1)

u1
u1 u2 + u2 + u1 = 0 =⇒ u2 = − .
u1 + 1
This means that our function field is just
K = C(u), u = u1 ,

because we’ve already adjoined u2 (a rational function of u1 ) once we do u1 . And this gives us the map (just multiplying
by various λs) ( )
u
π(1, u1 , u2 ) = (u1 , u2 ) = u, − = (1 + u, −1)
1+u
which is a good function as long as our functions are regular (u 6= −1). Then writing v = u −1 , our map π sends a
point (1, u1 , u2 ) to the point
(1 + v −1 , −1) ∼ (v + 1, −v )

which is okay even at v = 0 =⇒ u = ∞.

Definition 70
A point α with values in K is a good point if for all p ∈ X, there exists a λ ∈ K such that α′i = λai such that α′i
is regular at p for all i (and α′i (p) 6= 0 at at least one point). Then a morphism is a map defined by a good point.

This tells us how to define an isomorphism between two projective varieties X and Y : it is a morphism whose
inverse is a morphism. (And remember that X and Y can be open subsets of projective varieties as well.)
Here’s an example where this is useful:

Definition 71
An affine open subset of a variety X is an open subset U such that U is isomorphic to an affine variety Spec A.

It’s hard to know whether a subset is an affine open subset or not, especially in higher dimensions, but we have
a nice result: let U, V be affine open subvarieties of a variety X (which sits inside some projective space and is not
necessarily closed). Then U ∩ V is an affine open subvariety as well, and we’ll prove this next time.

13 March 2, 2020
Before looking some more at affine open sets, we’ll do a bit of review: a variety X is an open subset of a projective
i
variety X. The function field K of X is defined as follows: let X = X ∩ U i , where U i is the standard open subset.

32
 i i
Then if X is not empty, then it is an affine variety: we can write X = Spec Ai , and then we define K to be the
fraction field of Ai . But we don’t really know what the affine open sets look like, so this is a bit clumsy.

Remark 72. All open subvarieties of X have the same fraction field – note that we’ve been dealing with the closure
this whole time.

Nonzero elements of the function field are called rational functions: such a function f is regular at a point p if we
i i
can write p ∈ X for some i , meaning X = Spec Ai , and then the fraction f has a denominator that does not vanish
at our point p – we then evaluate at our point.
A useful notion here is that of a point of projective space with values in K: this is just a collection α =
(α0 , · · · , αr ) with αi ∈ K (not all zero) and the equivalence relation α ∼ λα for any λ ∈ K. Then we can say that
α is a point of X is it satisfies the defining equations of X – this helps us define a morphism α from a variety X to
projective space Pn . Specifically, if we have a point p ∈ X we look for a λ ∈ K in the functional field such that
• α′i = λαi for all i ,
• α′i are not all zero, and they are all regular at p.
If this is possible for every point p ∈ X, α is called a good point, and then we just define the map

α(p) = (α′0 (p), · · · , α′n (p)).

α gives us a morphism of X to Pn : if the image is contained in subvariety Y of Pn , then α is a morphism to Y .

Lemma 73
Let {X i } be an open covering of X. Then if αi are morphisms from X i → Pn , and αi , αj are equal on X i ∩ X j ,
then we can consistently define a morphism α from X to Pn .

Proof. The X i have the same function field, and if αi , αj are equal, then αj = λαi for some λ ∈ K. So we adjust
αj such that αi = αj for all j (using different λs for each j, of course), and we can just let this function that we’ve
defined be α.

Definition 74
Let X ⊂ Pr , Y ⊂ Pn , and suppose we have a morphism α from X to Y . Then α is an isomorphism if it is bijective
and its inverse is also a morphism.

We can use the above lemma to help us now:

Lemma 75
Suppose {Yi } is an open covering of Y . Let X i be the inverse image of Y i : then α restricts to maps αi from X i
to Y i . Then if αi is an isomorphism for all i , then α is an isomorphism.

Proof. Let β i be the inverse morphism Y i → X i . Then β i = β j on Y i ∩ Y j , because αi and αj are equal. This means
β is a morphism as well by our previous lemma, meaning that α is an isomorphism.

So if we want to check for isomorphisms, this can be just done locally (on open subsets). Recall from last time
than an open subset U of a variety X is an affine open subset if U is isomorphic to Spec A for some finite-type domain
A.

33
 Theorem 76
Let U, V be affine open subsets of a variety X. Then U ∩ V is an affine open subset.

Proof. Identify U with Spec A and V with Spec B. To prove U ∩V is an affine variety, we claim the coordinate algebra is
R = [A, B], the finite-type domain generated by A and B. (In other words, if A = C[a1 , · · · , ar ] and B = C[b1 , · · · , bs ],
A and B both have fraction fields equal to K, the function field of X, and we have C[a1 , · · · , ar ; b1 , · · · , bs ].) We will
show that if W = Spec R, then W ∼ =U ∩V.
Since X is a variety, it sits inside some Pn , and U, V sit inside X. A maps to R by inclusion, so we have a map
W → U (and similarly we have a map W → V ).
The inclusion map X → Pn is some morphism α, where α is a point of Pn with values in K (it defines the
embedding). But then α also gives us a morphism W → Pn : note that W can either map to U or V first, which means
the image of W under α is contained in U ∩ V .
We want to show the map α is an isomorphism. Choose an affine open set Z ⊂ U ∩ V . We want Z to be a
localization of U and of V : specifically, we wish to show that

Z = Us = Vt ,

where Us = Spec As = Spec A[s −1 ] and Vt = Spec B[t −1 ].

Lemma 77
Let p ∈ X. Then there exists an affine open subset Z contaning p in U ∩ V that is both a localization of U and
V : Z = Us = Vt .

Proof. We know that if U = Spec A, then the localizations of U form a basis for the topology on U. So if p ∈ U ∩ V ,
we can choose a nonzero s ∈ A such that p ∈ Us ⊂ U ∩ V . Similarly, choose t ∈ B so that p ∈ Vt ⊂ Us (we can do
this because Us is open). Finally, choose r ∈ A so that p ∈ Ur ⊂ Vt ⊂ Us ⊂ U ∩ V .
Now r ∈ Vt , so we can write r = bt −k for some b ∈ B. We can now write the localization Vtb = (Vt )b = (Vt )r ,
and this last expression is just Ur (r is invertible in Vt ). Thus, we’ve found the desired localization.

Wwe can do this for every point p ∈ X, so we can cover U ∩ V with sets Z which are localizations of both U and
V . Remember that R = [A, B]: then if Us = Vt , we know that As = Bt , so

Rs = [A, B]s = [As , B] = [Bt , B] = Bt = As .

This means our morphism α yields an isomorphism from Ws to Z for any s, and now we can cover U ∩ V with these
Zs. Therefore, W → U ∩ V is an isomorphism (by consistency of the morphisms).

We’re a day behind schedule – we’ll spend a day on the Grassmannian G(2, 4), but we should read about the
exterior algebra ∧V on our own, where V is a complex vector space.

14 March 4, 2020
We’ll discuss the Grassmannian today:

34
 Definition 78
The Grassmannian G(k, n) parameterizes the subspaces U of dimension k in an n-dimensional vector space
V = Cn .

We should think of G(k, n) as a set of points – one point for every subspace. For example, G(1, n) is one-
dimensional subspaces of Cn , which is the same thing as Pn−1 . Similarly, G(n − 1, n) is the dual space (Pn−1 )∗ , which
has the same dimension.
This means that the first interesting case not covered by the above examples is G(2, 4): these actually correspond
to the lines in P3 .

Remark 79. Professor Mattuck at MIT did a lot of research on Grassmannians, even if he’s known for a lot of teaching
too.

Let’s try to describe this in another way. Suppose U is a 2-dimensional subspace of V = C4 , and say that U has a
basis u1 , u2 . This gives us a 2 × 4 matrix, and we can usually change the basis via row reduction to
[ ]
1 0 ∗ ∗
,
0 1 ∗ ∗
[ ]
u1
where the ∗ yields a 4-dimensional subspace. This is true whenever we can row reduce the matrix in this way,
u2
which yields an open set W12 isomorphic to A4 . If we can’t row reduce with the first two columns, we can always do
so with some other two columns (because the two rows are linearly independent). For example,
{[ ]}
∗ 1 ∗ 0
W24 = .
∗ 0 ∗ 1

There are six such open sets, and therefore we can write the Grassmannian G(2, 4) as the union of 6 affine spaces A4 .
To say a bit more, say that (v1 , v2 , v3 , v4 ) is a basis of V = C4 . Then the exterior algebra

∧V = ∧0 V ⊕ ∧1 V ⊕ · · ·

is the (noncommutative but associative) algebra generated by V with the relations

v w = −w v ∀v , w ∈ V.

(It’s equivalent to just assume that v v = 0 for all v ∈ V : expand (v + w )(v + w ) = 0.) Here, ∧k V is the set of
elements in ∧V where we multiply k elements of V together:
• ∧0 V is just C, so it has a basis of 1.
• ∧1 V has dimension 4, and it’s generated by v1 , v2 , v3 , v4 .
• ∧2 V has dimension 6, and it’s generated by v1 v2 , v1 v3 , v1 v4 , v2 v3 , v2 v4 , v3 v4 .
• ∧3 V has dimension 4, and it’s generated by v1 v2 v3 , v1 v2 v4 , v1 v3 v4 , v2 v3 v4 .
• ∧4 V has dimension 1, and it’s generated by v1 v2 v3 v4 .
In higher dimensions, there is nothing: if we try to multiply 5 things together, each monomial in v1 , v2 , v3 , v4 will
have at least two copies of one of the basis elements, so it just becomes 0. And it’s easy to generalize this to an
exterior algebra ∧V where V is an n-dimensional vector space.

35
 Now note that
(v1 v4 )(v2 v3 ) = −v1 v2 v4 v3 = v1 v2 v3 v4 ,

and similarly
(v2 v3 )(v1 v4 ) = −v2 v1 v3 v4 = v1 v2 v3 v4 .

So ∧2 V is actually commutative – notice that this actually carries the structure of G(2, 4) inside it, up to scalars.

Definition 80
A decomposable element in ∧2 V is an element w such that w = uv for some u, v ∈ V .

Proposition 81
∑
If we can write w = i<j aij vi vj , then w is decomposable if and only if

a12 a34 − a13 a24 + a14 a23 = 0.

Proof. We’ll do the more interesting direction: if w satisfies the relation, then we can try

uv = (a12 v2 + a13 v3 + a14 v4 )(−a12 v1 + a23 v3 + a24 v4 ).

(There’s a negative sign because it’s the negative of a21 .) And now we can just expand this to find

2
w = a12 v1 v2 + a12 a23 v2 v3 + a12 a24 v2 v4 + a12 a13 v1 v3 + 0 + a13 a24 v3 v4 + a12 a14 v1 v4 − a14 a23 v3 v4 + 0

(remembering that v3 v3 = v4 v4 = 0). And then combining the two v3 v4 terms with our relation, and canceling the a12
everywhere indeed gives us w . (And if a12 = 0, use a different collection of the indices.)

Proposition 82
If a nonzero element w ∈ ∧2 V can be written as u1 u2 , then (u1 , u2 ) form a basis of a 2-dimensional subspace of
V . (In other words, they are independent.)

(Otherwise, u2 is a multiple of u1 , and u1 u1 = 0.) So now we have a way to characterize our subspaces: a
decomposable element of ∧2 V is a subspace, which is a point in the Grassmannian. We just need to show that
different w ’s give different subspaces:

Proposition 83
Let w = u1 u2 , w ′ = u1′ u2′ , and define U, U ′ to be the span of (u1 , u2 ), (u1′ , u2′ ) respectively. Then if U = U ′ , w ′ is
a scalar times w .

Proof. Since U = U ′ , we can write u1′ = au1 + bu2 and u2′ = cu1 + du2 . Expanding,

u1′ u2′ = (ad − bc)u1 u2 =⇒ w ′ = (ad − bc)w .

(the other terms are zeros because they are squares)

In particular, this means that the Grassmannian is a collection of points that satisfy a12 a34 − a13 a24 + a14 a23 = 0,
up to scaling. So G(2, 4) corresponds bijectively to a quadric in P5 with that above equation! And in particular, that

36
identifies G(2, 4) as a variety. (Note that this means G(2, 4) has dimension 4, but that’s something we already found
out earlier.)
This helps us answer some other questions:

Problem 84
Let X be a surface in P3 . (We’ll use coordinates (x1 , x2 , x3 , x4 ) here.) Does it contain a line ℓ?

Let’s look at the case where the line is

ℓ0 : {(x1 , x2 , 0, 0)},

or equivalently x3 = x4 = 0. To find out if ℓ0 ⊂ X, say that X is the zero locus f (x1 , x2 , x3 , x4 ) = 0, where f is an
irreducible (homogeneous, because we’re in projective space) polynomial of degree d. Then when we plug in x3 , x4
into f , the polynomial must be identically zero (because x1 , x2 can be anything, except (0, 0)).
Well, plugging in x3 = x4 = 0 will in general give us the monomials

c0 x1d + c1 x1d−1 + · · · + cd x2d ,

where ci are some of the coefficients of f . Therefore, for surfaces of a given degree, containing a line is equivalent
to setting (d + 1) coefficients equal to zero.
(d+3)
What’s another way to say this? The number of monomials in x1 , x2 , x3 , x4 of degree d is N = 3 , so the set of
surfaces S in P of degree d is parameterized by projective space of dimension N − 1. (We want f to be irreducible, but
3

this is an open subset of that projective space.) The lines that contain ℓ0 form a linear subspace of S of codimension
d + 1, which means we have dimension N − 1 − (d + 1) = N − d − 2.
But we care about containing some line, not just ℓ0 . However, the surfaces containing any line ℓ form a linear
subspace of codimension d + 1. Consider
Γ ⊂ G(2, 4) × S

where Γ corresponds to ordered pairs ([ℓ], [X]): ℓ is a line corresponding to a point in G(2, 4), and X is a surface,
where ℓ ⊂ X. The dimension of S is what we found earlier, and the dimension of G(2, 4) is 4, so we can add those
dimensions together:
dim Γ = 4 + (N − d − 2) = N − d + 2 = dim S − d + 3 .
• When d = 1, we have dim Γ = dim S + 2, so the fibres of the map Γ → S have dimension 2. Indeed, degree 1
hypersurfaces are planes, and they contain a 2-dimensional subspace of lines.
• When d = 2, we have dim Γ = dim S + 1, so the fibres have dimension 1. This is correct: quadrics are like
hyperboloids, which have a one-dimensional family of rulings.
• When d = 3, we have dim Γ = dim S, so the fibres have dimension 0 most of the time. In other words, we just
have finite sets – most (generic) cubics have a finite set of lines, and the right number is 27. We’ll try to be
able to prove this! Remember that a generic quartic in P2 has 28 bitangents, and this is not an accident.

We don’t always need to have a finite set of lines for all cubics – we can always take a cubic in three variables and
draw all of the lines from that cubic to another point outside that plane, and then we’ll have infinitely many lines.

15 March 6, 2020
We’ll start talking about dimension today: there are two equivalent definitions, and we’ll choose one of them.

37
 Definition 85
Let X be a variety with function field K. The dimension of X, denoted dim X, is the transcendence degree of K
(that is, the maximum number of algebraically independent elements of K).

Another, more useful, definition is that dim X is the longest possible length k of a chain of closed subvarieties
(irreducible closed subsets) in X of the form (notice there are k + 1 subvarieties here)

X = C0 > C1 > C2 > Ck = point.

Let’s show that this definition is equivalent:

Proposition 86
All chains of closed subvarieties have length at most dim X, and equality holds if we have a maximal chain.

We’ll start by reviewing a concept: α1 , · · · , αk ∈ K are algebraically independent if for all polynomials f (x1 , · · · , xk ),
f (α) 6= 0 – that is, we can’t find a polynomial relation in the αi s. The transcendence basis is a set of elements of a
maximal algebraically independent set (if finite). (The maximum number can be infinite, though it isn’t when K is a
function field).
All transcendence bases have the same order – this is fairly simple to prove by induction.

Lemma 87
Suppose α1 , · · · , αk , β are elements of our function field K, and {α1 , · · · , αk } are algebraically independent. Then
adding β will keep our set algebraically independent if and only if β is not algebraic over C(α1 , · · · , αk ).

Proposition 88
Let A be a domain (which is also an algebra), and let K be the fraction field of A. If the transcendence degree of
K is finite, then there exists a basis of elements of A.

Proof. We proceed by induction – suppose α1 , · · · , αk are elements of A that are algebraically independent but do
not form a transcendence basis. Then there is some β ∈ K which is not algebraic over the field C(α1 , · · · , αk ). We
can add β to our set (which we don’t want to keep, because we want our basis to contain elements of A), which is of
the form b
a for a, b ∈ A. But a, b can’t both be algebraic, or their quotient would be algebraic, so we can add either
a or b to our transcendence basis. Therefore, we can always find another element in A to add, and we’ve created an
algebraically independent set (α1 , · · · , αk+1 ).

It’s helpful to understand the dimensions along our chain of subvarieties C0 > C1 > · · · > Ck :

Theorem 89 (Krull)
Let X be an affine variety of dimension n with X = Spec A, and let α ∈ A be a nonzero element. Let V be the
set of zeros of α in X – this forms a closed subset of X, which means it’s a finite union of irreducible closed sets.
Then every irreducible component of V has dimension n − 1.

The proof is a bit harder than we might initially think:

38
 Fact 90
This proof was presented during class, but it was simplified a bit too much and became incorrect. I’ve included
the correct proof later in the notes.

Proof. Choose α1 , · · · , αn ∈ A to be a transcendence basis for K. Specifically, we can assume α is not a scalar
(otherwise V is empty, so the statement is vacuously true), so we can assume α = αn .
Let W be an irreducible component of V . V is the union of finitely many components: let Z be the union of the
others. “We want Z to be empty,” so we localize: each of W and Z are defined by some equations, and in particular
some elements of A are zero on Z. Pick an s ∈ A such that s is identically zero on Z but not identically zero on
W , and we’ll look at Xs = Spec A[s −1 ].
Xs has the same dimension as X, because it has the same field of fractions, and

Xs = X − (zeros of s).

In particular, Zs = Z ∩ Xs is the empty set, but Ws = W ∩ Xs is not empty (because s is identically zero on Z,
but not necessarily on W ). Since Xs , Ws have the same dimensions as X, W respectively, we can replace them in our
statement and use Xs , Ws instead.
Now since W (secretly W ∩ Xs ) is a closed subvariety of X, we can write W = Spec B, where B = A/P for some
prime ideal P . In particular, the zeros of P in W are the zeros of αn , which is exactly W . Now P and αA have the
same zeros, so the Strong Nullstellensatz tells us that

rad(αA) = rad P = P.

Remember that rad I is the set of elements a such that some power of a is in I. If β ∈ A and β vanishes on W
(meaning it’s in the prime ideal P ), then β k is in αA for some k.
We want to show that dim W = dim A − 1: let αi be the residue of αi in B (mod our prime ideal P ), which gives
us (α1 , · · · , αn−1 ). (Remember that αn = 0 because α goes to 0 when we mod out by P .) We claim that these form
a transcendence basis for K, the fraction field of B – this would show that dim W = n − 1.
There are two parts of this – we need to show they’re algebraically independent, and also every other element is
algebraic over C[α1 , · · · , αn−1 ].
First, we’ll show that αi s are algebraically independent. Say f (x1 , · · · , xn−1 ) is a polynomial with f (α) = 0: then
f (α1 , · · · , αn−1 ) ∈ P , where P is the radical of αA. (? We’ll postpone the rest of this proof for next time.)
On the other hand, let’s show that α1 , · · · , αn−1 , β are always algebraically dependent. Suppose otherwise: then
α1 , · · · , αn−1 , β form a transcendence basis in K if β is not algebraic over C(α1 , · · · , αn−1 ). We know that β must
be algebraic over C(α1 , · · · , αn ), so we can write
∑
ci (α)β i = 0,

where ci (α) are rational functions of α. Clearing denominators, we can assume they are polynomials, so we have some
polynomial f (x1 , · · · , xn , y ) such that f (α1 , · · · , αn , β) = 0. This tells us that (modding out by P )

f (α1 , · · · , αn−1 , 0, β) = 0,

We can assume that xn does not divide f , so that when we plug in 0, f is not identically zero. But now we have a
polynomial relation between β, α1 , · · · , αn−1 , which is a contradiction.

39
16 March 9, 2020
We’ll have a quiz next Wednesday during class – we’ll make the next problem set due after spring break, as long as
MIT isn’t canceled by then.
By the way, the proof of Krull’s theorem last time was incorrect, because we simplified the proof too much and it
became incorrect. So we’ll leave it for now and come back to it later on.
Today, we’ll start with the Nakayama Lemma. Let A be a ring and P = (pij ) be an n × n matrix with entries in A.
Then there exists another A-matrix C such that CP = (det P ) · I. Here, C is the cofactor matrix: the entry Cij is
(−1)i+j times the determinant of P when we remove [ the ]jth row and the i th column.
[ What’s
] important is that this
a b d −b
formula works in any ring. As an example, if P = , then we have C = (we should remember to
c d −c a
take the transpose after looking at the matrix minors).
So now let M be an A-module, and let v = (v1 , · · · , vn ) be a vector with entries in M. Let P be an n × n A-matrix:
then v is an eigenvector with value λ if

P v = λv =⇒ (λI − P )v = 0.

Now if we let C be the cofactor matrix for (λI − P ), we know that C(λI − P )v = det(λI − P )v = 0. In other words,
we know that p(λ), the characteristic polynomial det(λI − P ), kills the vector v .

Proposition 91 (Nakayama Lemma)

Let M be a finite A-module, and let J be an ideal of A. If JM = M, then there exists z ∈ J such that (1−z)M = 0
– in other words, zm = m for all m ∈ M.

Proof. Let v1 , · 

· · , vn generate the module M, so every element of M is a combination of the vi s with coefficients in
v1
 
 v2 
 
A. Let v =  . . Since JM = M, each generator vi can be written as a sum of terms jm, where j ∈ J and m ∈ M,
 .. 
 
vn
and now each m is an A-linear combination of the vj s. Moving the coefficients from the vj s to the j’s (which is allowed
by closure of the ideal J), we find that
∑
vi = pij vj ,
j

where the coefficients pij are all in J. Thus v is an eigenvector of the matrix P = (pij ) with eigenvalue 1.
This means that if we have our characteristic polynomial p(t) = det(tI − P ), then p(1)v = 0. But every element
in M is generated by the vi s, so p(1)m = 0 for all m ∈ M. Now we just need to find an element z: set

det(I − P ) = 1 − z,

and notice that the left hand side is some polynomial z in the pij s (which are all in J), so z ∈ J, as desired.
For example, [ ]
1−a b
= (1 − a)(1 − d) − bc,
c 1−d
so this gives the element z = a + d + bc − ad.

40
 Corollary 92
Let A be a Noetherian domain, and let I, J be ideals of A. If IJ = I, then either I is the zero ideal or J is the unit
ideal.

Proof. Since A is Noetherian, I is a finitely generated ideal, which means I is a finite A-module. Therefore, the
Nakayama Lemma tells us that there is a z ∈ J such that (1 − z)I = 0. Therefore, either I = 0 or there is an element
in I. In the latter case, we must have z = 1 (because there are no zero divisors in a domain), meaning J is the unit
ideal because z = 1 ∈ J.

Corollary 93
Let A be a Noetherian domain, and let x ∈ A be not a unit. Then
∩
x n A = (0).

In other words, if y 6= 0 is in A, then x n does not divide y for sufficiently large n.

∩
Proof. If x = 0, the statement is clearly true. Otherwise, let I = x n A, and let J = xA. Then JI = I (we just
multiply an extra x in every term), and J is not the unit ideal, so I is the zero ideal by the previous corollary.

This turns out to be true when our ideal J is not just a principal ideal, but it’s hard to show in that case. Also,
note that the assumption that A is a domain is important: a counterexample otherwise is to take A = C × C and
x = (1, 0).
We’ll look at something a bit different now:

Definition 94
Let A ⊂ B be domains. Then an element β ∈ B is integral over A if it is the root of a monic polynomial
f (x) = x n + a1 x n−1 + · · · + an , where ai ∈ A.

The set of elements in B that are integral over A form a subring of B, which we can check by verifying all of the
closure assumptions.

Lemma 95
Let A be a Noetherian ring. Then β is integral if and only if A[β] is a finite A-module.

Proof. If β is integral, we have the relation β n =(polynomial in β), so (1, β, β 2 , · · · , β n−1 ) generate A[β].
For the other direction, if A[β] is a finite A-module, then we keep picking powers of β until they generate A[β].
Then we have some finite list of powers of β, so we can write β N in terms of them, giving us a monic polynomial.

Lemma 96
Let A ⊂ B be finite type domains. Then B be a finite A-module if and only if all elements of B are integral over
A.

41
Proof. Any submodule of B is a finite A-module, so all the elements are integral (because A[β] is always a submodule
of B). For the other direction, we know that B is generated by finitely many elements as an algebra. If each one is
integral, then B is a finite module because we adjoin one of them and it’s a finite module, then add the next one, and
so on.

Example 97
Let A = C[x], and let B = A[y ]/(f (y )). Here, f is of the form

f (y ) = a0 y n + a1 y n−1 + · · · + an ,

where ai ∈ A are functions of x.

If a0 = 1 (or any other nonzero complex number), then the residue of y in B is integral. Thus, every element of
B is integral over A. But if a0 is not a constant, it will vanish at some points where a0 (x) = 0. The fibre over any x0
is a finite set of points, but at each of the points where a0 vanishes, the degree of f goes down. So that corresponds
to a vertical asymptote, where y goes to ±∞ as x → x0 . We don’t like these vertical asymptotes, which explains why
we care about the idea of β being integral!
For example, when f (x, y ) = xy − 1, then a0 (x) = x, and indeed as x → 0, y → ∞. More generally, if we write

f = a0 y n + · · · + an ,

we have
f a1 an
= g = y n + y n−1 + · · · + .
a0 a0 a0
6 0, then this just looks like f , but when a0 = 0, at least one of the coefficients goes to ∞. Since the coefficients
If a0 =
are symmetric functions in the roots, this means at least one root must go to infinity.

Theorem 98 (Noether Normalization Theorem)

Let A be a finite-type domain. Then there exist y1 , · · · , yr in A that are algebraically independent, such that A
is a finite-module over a polynomial subring C[y1 , · · · , yn ] = P (so all the elements of A are integral over the
polynomials).

Proof. Choose generators x1 , · · · , xn for our algebra A. If they are algebraically independent, then we’re done – P is
equal to the polynomial ring. Otherwise, let f be the polynomial that evaluates to 0 for our variables xi . We’ll use the
following result:

Lemma 99
There exists a linear change of variables so that f (x1 , · · · , xn ) becomes a monic polynomial in xn of deg d = deg f .

Proof of lemma. Let xi = xi′ + ci xn′ , where the ci s are to be determined. (This is indeed invertible.) Take the
highest-degree homogeneous part of f : then

f (x1 , · · · , xn ) = f (x1′ + c1 xn′ , · · · , xn′ + cn xn′ ).

Our goal is to show that f can be monic in xn′ . The coefficient of degree d comes from putting in x1′ = x2′ =
′
· · · + xn−1 = 0 and xn′ = 1: we’ll get
f (c1 , · · · , cn−1 , 1 + cn ),

42
which can be nonzero if we choose the ci s appropriately, and then we can normalize to finish.

Then the Normalization Theorem becomes very simple: make that change of variables, and let R be the algebra
C[x1 , · · · , xn−1 ] ⊂ A. But we chose a polynomial f earlier that evaluates to zero at our point (x1 , · · · , xn ), and x is
monic, so xn is integral over R. Therefore, A is a finite R-module.
And now we induct: we may assume R is a finite module over the polynomial subring P . Then A must be a finite
module over P , as desired.

This turns out to be extremely useful, and we’ll see applications shortly.

17 March 11, 2020

There won’t be a quiz next week. Professor Artin will post a problem set for us to work on for the next few weeks.
Lectures after this week might be done by Zoom, but things are a bit unclear for now. (We’ll still have lecture on
Friday.)
We’ll jump ahead a bit for today’s lecture, because it’ll be hard to get things organized. First, a bit of review: if
A ⊂ B, we say that B is integral over A if every element b ∈ B satisfies a monic polynomial relation with coefficients
in A. In particular, if A and B are finite-type domains, then B is integral over A if and only if B is a finite A-module.
It’s also possible to do this when B is an A-module, but then we’re not necessarily working with an inclusion
anymore:

Definition 100
Let Y = Spec B and X = Spec A. If there is a homomorphism ϕ : A → B, then we call the morphism u : Y → X
a finite morphism. If A ⊂ B and B is a finite A-module, then Y → X is an integral morphism.

Example 101
Let Y = Spec B be a closed subvariety of X = Spec A. Then the inclusion map Y → X is a finite morphism,
because the corresponding homomorphism ϕ : A → B is a quotient map, and B is clearly a finite A-module
because it is generated by {1}. (Here, it’s helpful to note that B is an A-module in the sense that it is closed
under multiplication by ϕ(A).)

We can also extend our definitions of finite and integral morphisms beyond the affine case as well. Any morphism
u : Y → X is a finite (resp: integral) morphism if we can cover X by affine open sets, and for every such affine open
set X ′ = Spec A ∈ X, the inverse image Y ′ = Spec B is affine, and u : Y ′ → X ′ is finite (resp: integral). In general,
this means we only need to check on the affine open sets! To show that this definition is consistent with the previous
one, we need to look at the localizations (which form a basis for the topology), though the proof is a bit long because
we need to check a lot of different conditions.
For now, we’ll move on:

Theorem 102 (Chevalley’s finiteness theorem, version 1)

Let Y be a closed subvariety of Pn × X. If the projection π : Y → X has finite fibres, then π is a finite morphism.

43
 Theorem 103 (Chevalley’s finiteness theorem, version 2)
Let X, Y be projective varieties, and say that u : Y → X is a morphism. Then if the fibres of u are finite, then u
is a finite morphism.

Let’s first show that version 1 implies version 2. Let Γ be the graph of u, which consists of points of the form
(y , u(y )). Then we have Y ⊂ Pn , and Γ ⊂ Y × X ⊂ Pn × X. Since Γ is a closed subvariety of Pn × X and Γ is
isomorphic to Y , we can compose the isomorphism with the morphism from Γ to X.

Proof of Chevalley (version 1). We’ll prove this in the case of P1 – since Pn just has more indices, we’ll avoid those.
We know that Y sits inside P × X, and π is the projection of a point down to X. The graph of Γ looks like some curve
in the X-axis and P1 -axis: pick coordinates (y0 , y1 ) in P1 . Let X ∞ be the points at infinity – that is, where y0 = 0.
In case 1 (this is the main case), we have that Y ∩ X ∞ = ∅. Because we don’t hit the line at infinity, we can
assume X is affine, so X = Spec A for some A.
Consider the two standard open sets U 0 × X (where y0 6= 0) and U 1 × X (where y1 6= 0). Then we can say that

U 0 × X = Spec A[u], U 1 × X = Spec A[v ],

where u = y1
y0 , v = y0
y1 = u −1 . In this case, because Y ∩ X ∞ = ∅, we know that

Y 0 = (U 0 × X) ∩ Y = Y.

The secret is to look at U 1 × X first: this can be written as Spec R1 , where R1 = A[v ]. Then Y 1 = (U 1 × X) ∩ Y
is a closed subset of U 1 × X, and it is irreducible (because it is an open subset of Y , which is irreducible), so Y 1 is
the locus of some prime ideal Q. (Remember that when we work in U 1 × X, we’re leaving out the parts of the space
where y1 = 0.)
Now X ∞ is the locus where v = 0, and Y 1 ∩ X ∞ is empty because Y ∩ X ∞ is empty. But we have two closed
subsets that don’t intersect, which must generate the unit ideal. Since X ∞ is generated by v , we have that

Q + v R1 = (1) = R1 .

Then there exist f (v ) ∈ Q, g(v ) ∈ R1 in R1 = A[v ] (we’re writing everything as polynomials in v , with coefficients in
A) such that
f (v ) + v g(v ) = 1.

Write out f (v ) = a0 v n + · · · + an ; we know that an = 1 by matching constant coefficients. But now we can write
1
everything in terms of v = u instead: multiplying this through by a high power of u to clear all denominators, and
we’ll find that the left hand side is a monic polynomial in u, because the original constant term was 1.
To understand where this polynomial f vanishes, note that if we leave out both 0 and ∞ in the projective coordinate,
we can go back and forth between u and v . This monic polynomial vanishes on all of Y − X ∞ , and thus it vanishes
on Y by closure. That means that u is integral over A, but we can also write B as a quotient of A[u] because Y is a
closed affine subset of U 0 × X. Thus B is a finite A-module, and we’ve proved the desired result.
In case 2, Y does intersect X ∞ . Then the plan is to cover X by localizations so that we can get away from our
problem: for any p ∈ X, p × P1 intersects Y at only finitely many points. Since the localizations form a basis, we can
cover X by X i = Xsi so that after a change of coordinates, we localize the points p ∈ X where Y intersects X ∞ (this
is a closed set). Then we have a finite number of localizations (because of the finite fibre assumption)

Y i = (Pi × X i ) ∩ Y,

44
and this doesn’t meet X ∞ . Then X i = Spec Ai , and Y i is the inverse image of X i under π, so Y i = Spec Bi , where
Bi is a finite Ai -module from case 1.
And the localizations cover X, so s1 , · · · , sr generate the unit ideal of A.

Lemma 104
With all of the notation above, we have that Y = Spec B and B is a finite A-module.

This kind of argument often comes up in algebraic geometry, so we should pay attention to it here.
∩
Proof of lemma. Let R = Bi – all of the Bi s have the same fraction field, so we can take the intersection there.
We’ll show that Y = Spec R and R is a finite A-module.
Localize R, so (∩ )
Rsj = (Bi ) .
sj

If we invert sj in all of the Bi s separately, we get something at least as big, so

∩
Rsj ⊂ ((Bi )sj ).

Thus, for all β ∈ B in the right-hand set, we can write

β = bi sj−k , bi ∈ Bi ,

where we use the same k for all i for simplicity. Thus βsjk = bi for all i , and therefore βsjk is in the intersection of the
∩
Bi s, which is R. Therefore, every element β ∈ ((Bi )sj ) is also in Rsj , so we have
∩
Rsj = ((Bi )sj ).

Bi is the coordinate algebra of Y i , so

Y i ∩ Y j = Spec B i ∩ Spec B j = Spec(Bi )sj ,

because we need a function that vanishes on the complement of Y j . And notice that (Bj )sj = Bj (sj is already invertible
in Bj ), which means that
Y i ∩ Y j = Spec(Bi )sj = Spec(Bj )si .

Thus (Bi )sj = (Bj )si , and both of these are at least as big as Bj . But when i = j, we get equality, so indeed we have
∩
(Bi )sj = Bj .

So R is a nice ring, because localizing gives us the Bj s. Let Ỹ = Spec R: we wish to show that Ỹ = Y . When we
localize Ỹsj , we end up with Spec Rsj = Spec Bj = Y j . Thus Ỹsj is isomorphic to Y j , and the former is a covering of Ỹ
while the latter is a covering of Y . Thus we’ve shown that Y and Ỹ are indeed isomorphic.
To finish, note that Bi = Rsi is a finite Ai -module by construction for each i , so we can pick finitely many generators
for each Rsi . Then we can multiply by si (which is invertible in Ai ), so that our generators are all elements of R, and
we do this for all (finitely many) localizations i . Putting all of these together yields a set of generators for R.

And now B is again a finite A-module, completing the proof.

45
18 March 30, 2020

Fact 105
Classes from this day onward were conducted over Zoom.

We’re going to talk about double planes today, specifically projective double planes (we can read about affine
double planes on our own). Throughout this class, we’ll let X = P2 with coordinates x0 , x1 , x2 , and we’ll let y be
another variable of weight d: in other words, we’ll let f (x) be a homogeneous polynomial of degree 2d, and we say
that y 2 = f (x). Then we’ll look at the zero locus of y 2 − f (x) in the weighted projective space, which is defined by
the relation
(x0 , x1 , x2 , y ) = (λx0 , λx1 , λx2 , λd y ).

Indeed, y 2 − f (x) can be thought as a homogeneous degree 2d polynomial if y has weight d:

Definition 106
A double plane is a zero locus Y for y 2 − f (x) in the weighted projective space.

Since X is the ordinary projective plane, we have the projection map π : Y → X sending (x0 , x1 , x2 , y ) to (x0 , x1 , x2 ).
The fibres of this map generally have two points (x, y ) and (x, −y ), but these are the same points when f (x) = 0. So
the branch locus ∆ corresponds to the points in P2 where the fibres of π are just a single point (x, 0). And there’s an
automorphism σ : Y → Y of the double plane given by

σ(x, y ) = (x, −y );

this is an order 2 automorphism.

Since our map π is two-to-one, the image of Y will have dimension 1 as a subvariety of X. Let C be a curve in X
which is the zero locus h(x) = 0 of an irreducible homogeneous polynomial h. There are three cases here:
• C ramifies, which means that C is contained in the branch locus. This means that h divides f – this only occurs
for a finite number of curves. In this case, π −1 C will map bijectively to C, because every point in C corresponds
to one point in π −1 (C). (If we were talking about schemes, we’d want to say that the fibre is “two times the
curve,” because y 2 = 0. We don’t have to worry about this, though.)
• C splits, which means that π −1 C is made up of two curves D1 ∪ D2 in Y . Applying σ to Y will switch the curves
D1 and D2 with each other, and Di → C are generically bijective for both i = 1 and i = 2.
• C doesn’t split, which means that π −1 C is an irreducible curve. This happens when the ideal

(y 2 − f , h)

is a prime ideal in C[x, y ].

All of this is analogous to what happens when we adjoin a quadratic element δ to Z (this is material from 18.702):

Example 107
Consider Z ⊂ Z[δ], where δ 2 = −5.

There are a few different possibilities here, which correspond to the above cases:

46
 • Notice that the equation y 2 = −5 becomes y 2 = 0 when we’re looking mod 5, so that explains why the prime
p = 5 ramifies in Z[δ].

• On the other hand, p = 3 splits as

(3) = (3, 1 + δ), (3, 1 − δ).

Since (3) does not split as a principal ideal, it makes it harder to analyze what the ideals actually look like in
Z[δ]. So it’s generally not very easy to say what the ideals look like in the case above where C splits. Meanwhile,
p = 41 does split into principal ideals:
(41) = (6 + δ)(6 − δ).

• Finally, the prime p = 7 remains prime, which corresponds to the third case above.

So if we go back to the double plane y 2 = f (x), remember that f (x) has even degree. Let’s take the particular
example where d = 2 and deg f = 4: we want to ask when a line C = L splits.
We can choose our coordinates so that the line is {x0 = 0}. Since L splits, we can say that the ideal (y 2 − f , x0 )
is not prime. Taking this ideal mod x0 , we can define f (x1 , x2 ) = f (0, x1 , x2 ), and then we can rephrase our statement
as saying that (y 2 − f ) is not a prime ideal in C[x1 , x2 , y ]. Since C[x1 , x2 , y ] is a UFD, this is true if and only if y 2 − f
is not irreducible.
Since y 2 − f is quadratic in y , a factorization will look like

y 2 − f = (y + α)(y − α) =⇒ f = α2

(each factor has to be linear in y , and the polynomial is monic in y ). So the line L splits if and only if f is a square.
This means that all of the roots of f have to have order 2, so ∆ has to be tangent to the line L. So in this special
case that we’re talking about with d = 2, lines that split correspond to bitangents of the branch locus ∆, so that
both of the two zeros have order 2. (It could also be a single fourfold tangent, but that’s not generic, so we’ll ignore
it.) Note similarly that a line L that splits in a double plane with d = 3 is a tritangent for the same reason. But
generic curves don’t have tritangents, because there are finitely many bitangents!
With this, we can explain the following famous property:

Proposition 108
Every smooth cubic surface contains 27 lines.

Proof. Suppose we have a cubic surface S ⊂ P3 , and we label P3 with coordinates (x0 , x1 , x2 , z ) (it’s not weighted,
but we do want to distinguish one coordinate). Choose coordinates so that q = (0, 0, 0, 1) is a point on the curve
S: then the defining equation of S must be of the form

az 2 + bz + c = 0,

where a, b, c are homogeneous polynomials of degree 1, 2, 3 in the x-variables. (Because (0, 0, 0, 1) is on the curve,
the coefficient of z 3 is zero.)
Now the discriminant of f = b2 − 4ac is a homogeneous quartic, and thus we can describe the zero locus similarly
to the locus of the quartic double plane
Y : y2 = f ,

47
where the quadratic formula tells us that
√
−b + f −b + z
z= = =⇒ y = 2az + b.
2a 2a
(The zero loci here are birational.) So the correspondence between S and Y makes sense whenever a 6= 0 – let’s look
at the a = 0 special case first. Then we can find a line L0 in X which splits in Y , because the defining equation of S
looks like
bz + c = 0

modulo a. The locus of a = 0 then defines a plane H0 in P3 with coordinates (x0 , x1 , x2 , z ), and this plane contains
the point q. Meanwhile, the locus of bz + c = 0 is a (non-generic) cubic curve in H0 – this equation is singular at the
point q, because there is no z 3 coefficient. (To verify this, we can just suppose that a is the coordinate x0 by a change
of coordinates, and we can note that it’s not possible for that cubic curve to be the union of a line and a quadratic if
S is generic.
But now let’s look at the case where a 6= 0, we can take L to be another bitangent to the curve f = b2 − 4ac = 0.
L now intersects our first bitangent L0 in one point, and the map between S and Y is defined except at a single point
of L. Since L splits in Y (it’s a bitangent), we know that it is the union of a line and a quadratic. Since there are 27
bitangents distinct from L0 , this means that there are 27 lines on a cubic, as desired.

19 April 1, 2020
There’s a lot on the syllabus for today – we’ll talk about valuations and then discuss a little about normalization.

Definition 109
Let K be a field. A (discrete rank 1) valuation on K is a surjective homomorphism from K × → Z+ , such that
v (αβ) = v (α) + v (β) for all α, β ∈ K, and v (α + β) ≥ min(v (α), v (β)).

Example 110
Let K = C(t), and let t = a be a point in A1t .

We can define the valuation vp with the following process: if f (t) is a polynomial, and t − a divides f k times, then
vp (t) = k. And if we have a rational function α = gf , we’ll just define vp (x) = vp (f ) − vp (g).
It turns out that all valuations of K = C(t) are either vp for some p or the valuation at the point at ∞ (which we
1
get by working with t instead).

Definition 111
The valuation ring R of a valuation v is the set of α ∈ K such that v (α) ≥ 0.

This ring has the maximal ideal

M = {α : v (α) > 0}.

Indeed, if we take any other element in the ring, v (α) = 0. Then v (α−1 ) = 0 as well, so α is a unit of R. In other
words, all elements not in M are units, so M must be a maximal ideal.
M is also a principal ideal of the form xR, where x is any element where v (x) = 1. Indeed, if y ∈ R and v (y ) = n,
then v (x −n y ) = 0. Thus u = x −n y is a unit, so y = x n u, meaning that y ∈ x n R = M n . Thus, the ideals of R are

48
either the zero ideal or of the form Rn , and by the same logic, any element y of a proper R-submodule of K is of the
form x k R, and thus N = x k R (for the minimal k ∈ Z) or K (if no such k exists).
This also tells us that there is no ring R′ such that R < R′ < K. If there were such an R′ , it would be an
R-submodule of K, and it must be of the form x k R for some k ≥ 0 (or else x k squared wouldn’t be in the ring). But
R is contained in R′ , so k = 0.

Definition 112
A local ring R is a Noetherian ring with exactly one maximal ideal M, where M is not the zero ideal.

In particular, a valuation ring is a local ring.

Proposition 113
Let R be a local domain with maximal ideal M. Then M is a principal ideal if and only if R is a valuation ring.

Proof. We’ve already shown the backwards direction above. For the forward direction, first assume that the maximal
ideal in our local ring M = xR is a principal ideal. Let y be a nonzero element of R: by the Nakayama Lemma, we
know that the powers of x that divide y are bounded, so we can write y = x k u, where x does not divide u. But u is
not divisible by x, so it must be a unit in the ring: we can then define v (y ) = k. So we’ve now defined a valuation,
and it just remains to check the other conditions in the definition (which are not too hard).

With this, we can move on to the concept of a local ring at a point:

Definition 114
Let X be an affine variety, and let p be a point of X. Then the local ring of X at p is

Ap = {α ∈ K : α regular at p}.

In other words, there exists an affine neighborhood Spec A of the point p in X, such that α ∈ A.

To explain why this is indeed a local ring, we need to show that there is only one maximal ideal. The maximal ideal
at a point p of X is
mp = {α : α regular at p, α(p) = 0}.

It suffices to show that if s ∈ Ap and α(s) 6= 0, then α is a unit, so α−1 ∈ Ap as well. We know that α−1 ∈ A[α−1 ],
and Spec A[α−1 ] is Spec A with the zeros of α removed. Thus, p ∈ Spec A[x −1 ], so α−1 is regular at p, as desired.
We’ll now move on to our next topic – normalization.

Definition 115
Let A be a Noetherian domain with a fraction field K of characteristic zero. Then the normalization of A, denoted
A# , is the set of α ∈ K that are integral over A. A is normal if it’s equal to its normalization A# .

It is clear that A ⊂ A# ⊂ K, and we have some more structure here:

Theorem 116
A# is a finite A-module. In other words, the map X # → X is an integral morphism.

Proof sketch. The proof is involved, and one important object that will come up later in the class is the trace.

49
 Definition 117
Let L be a finite field extension of K, and let β ∈ K. Multiplication by β is a linear operator on L – the trace of
β is the trace of this linear operator.

We can use this trace to define a bilinear form on L, sending α, β to tr(αβ). This is an element of A (this is easy
to verify because the trace of an operator is basis-independent). From here, the main ideas of the proof are to assume
A is normal, use a basis for L such that all elements are integral, and then consider the map of L into K n via the
operator
T (β) = (hv1 , βi, · · · , hvn , βi).

This is injective, and it maps the integral closure B of A (in L) to An , which means that B is isomorphic to a submodule
of An , which is finitely generated because A is Noetherian.

Let’s show an example of this (the most important case is that of curves):

Example 118
Let A = C[x, y ]/(y 2 − x 3 ), and let z = yx .

y2 x3
z is integral over A, since z 2 = x2 = x2 = x (where we’re using x and y for the residues in A as well). In fact, we
also have y = z 3 , so the normalization A #
is just polynomials C[z]. In other words, we’ve “straightened out the cusp”
to create the affine line! We do need the following result to show that we’ve found our normalization, though:

Lemma 119
Any unique factorization domain is normal – in particular, polynomial rings over a field, such as C[z], are normal.

r
We can show this by showing that any element β = s is integral and writing out its defining equation.

Example 120
Consider the ring A = Z[δ], where δ 2 = −3.

As we might know from 18.702, this isn’t an integrally closed ring: we get the nice properties by using A# = Z[η],
1+δ
where η = 2 .

Definition 121
A local ring A has dimension 1 if the only prime ideals of A are 0 and the maximal ideal M.

For example, the valuation vp at a point p is a one-dimensional local ring.

Theorem 122
Let A be a local ring. Then A is one-dimensional and normal if and only if it is a valuation ring.

Proof. The backwards direction is easy, and the forward direction is trickier. Our plan will be to show that the maximal
ideal M of A is a principal ideal, and then we can use Proposition 113. Let x ∈ M: the radical ideal rad(xR) is the

50
intersection of some prime ideals, but there are only two prime ideals in our local ring, and one of them is the zero
ideal. Thus rad(xR) = M, which means that M r ⊂ xR for some r > 0.
Let r be the minimal such possible exponent: then M r −1 is not contained in xR, but M r is, and we can pick an
element of y that is in M r −1 but not xR. So now y M ⊂ M r ⊂ xR, and now we can consider the fraction z = yx . This
is not in R (because y is not in xR) but zM is in R (because y M is in xR).
But now notice that zM is an ideal of R, because it’s closed under multiplication by R and addition. Since there
is only one maximal ideal, either zM is contained in M, or zM is the whole ring. Let’s write out those two cases:
(1) If zM is the whole ring R, then we can let t = z1 , and we know that M = tR (by multiplying through by t).
Thus t is actually in R (because M contains t · 1), and now we’ve shown that M is a principal ideal, so we’re done.
(2) if zM ⊂ M, we can apply Nakayama: say that v1 , · · · , vk generate the ideal M. Then we can write
∑
zvi = pij vj ,
j

where the pij are in R. Therefore, (v1 , · · · , vk ) is an eigenvector of P with eigenvalue z, which means that p(z) = 0.
But this means that z is integral, which is a contradiction (since z is not in R).

Fact 123
We will also include the proof of Krull’s theorem here, now that we have enough tools. Recall that we want to
show that any closed subset VX (α), corresponding to the zero set of α in an affine vareity X of dimension d, has
irreducible components of dimension (d − 1).

Proof of Krull’s Theorem. We know that the dimension of any component C is less than d, because C is a proper
closed subset of X. Assume for the sake of contradiction that dim C < d − 1.
Let A be the coordinate algebra of X, and let B be its normalization. Then Y = Spec B has dimension d (because
A and B have the same fraction field, thus the same transcendence degree), and the morphism Y → X is an integral
morphism by definition. This map is also surjective, because every point in X is a maximal ideal m in A, and the
extension mB in B is not the unit ideal (this is a lemma from earlier in the class), so it corresponds to a point in Y .
Since we have an integral morphism, closed sets are mapped to closed sets, and there is some irreducible component
of VY (α) which maps to our irreducible component C of VX (α). Now the morphism D → C is an integral morphism
as well (it’s defined the same way as the one from Y to X), so D and C have the same dimension, meaning that it
suffices to prove Krull’s theorem for the case where A is normal (because we can replace X with Y and have the same
dimensionality).
Let the zero locus of α be written as C ∪ V . Similarly, as our earlier attempt, we want V to be empty, so we’ll
localize by finding an element s that is identically zero on V but not on C. Then Vs is empty (because s is zero
everywhere), but Cs is not empty. We’ve assumed here that A (and therefore X) is normal, so the localization Xs
is normal as well. Since Xs and Cs have the same dimensions as X and C, respectively, we can use the localizations
instead, and this means that we can assume α’s zero locus is irreducible.
Now we can say that C is both the zero locus of P and of α. This means that P = rad P = rad(αA), which means
that P n ⊆ αA for sufficiently large n. We’re assuming that C has codimension at least 2 (in other words, dimension
at most d − 2). By definition, this means that C ⊂ Z for some subvariety Z of X with codimension 1 – say that this
corresponds to the prime ideal Q. Now P ⊃ Q, but α is not in Q (since α does not vanish on all of X), so αA is not
contained in Q.

51
 Lemma 124
With the notation above, there is an element γ ∈ A such that γ is not in αA but P γ is contained in αA.

Proof of lemma. Pick some β ∈ Q. We have a Noetherian domain here, so by corollaries of the Nakayama Lemma,
we know that there is some k such that αk is the largest power of α dividing β. Now γ = β
αk vanishes on Z \ C, which
is a dense subset of Z, which means that γ vanishes on all of Z. So now we have a preliminary candidate: γ ∈ Q and
α does not divide γ.
Now to show that we can find a γ such that P γ is contained in αA, we know from above that P n ⊆ αA for some n
and thus P n Q ⊆ αA – let r be the smallest integer such that P r Q 6⊆ αA. (This exists because we found a preliminary
γ above) Then there is some element of this ring P r Q that is not in αA, and that γ satisfies the conditions that we
want.

Taking this γ, we can consider the element δ = γ

α. We know that P δ ⊂ P , because α is in P , but γ 6∈ αA =⇒ δ 6∈
A by construction. P δ then vanishes on Z \ C, which is a dense subset of Z, which means that P δ vanishes on all of Z.
Therefore, P δ ⊂ P , meaning δ is integral over A. By normality, this means that δ ∈ A, which is a contradiction.

20 April 3, 2020

Definition 125
A curve X is a variety of dimension 1.

The proper closed subsets of a curve are the finite subsets (by Krull’s theorem).

Definition 126
Writing a curve X as Spec A, if A is normal, then X is a smooth affine curve. In general, X is smooth if it can
be covered by smooth affine curves.

In general, if X is not smooth, we can take the normalization of A, and then X # will be a smooth curve. (We
need to normalize each of the affine open sets of a covering, and we need to check that this is consistent.) Then the
morphism X # → X is integral (because every element of A# is integral over A), so we have a surjective morphism.

Example 127
A good way to think about this is that X # removes singular points from X: for example, a cusp will be straightened
out.

Proposition 128
Local rings of smooth curves are valuation rings.

By definition, smooth curves are dimension 1. Remember that the local ring at each point is the set of rational
functions that are regular, so we just need to know that (assuming that X is affine here) the local ring is also normal.
But we won’t focus too much on the details here! We should just think of the local ring at each point as corresponding
to the valuation vp which evaluates the multiplicity of the zero or pole.
We’ll move along to talking about constructible sets: throughout this discussion, X is a variety.

52
 Definition 129
A locally closed set L is an intersection of the form Y ∩ U, where Y is closed and U is open.

Example 130
Let X = A2 , and let L be a line with one point p deleted. Then this is the intersection of L with the complement
of p, so this is a locally closed set.

Clearly, open and closed sets are locally closed.

Proposition 131
L is locally closed if and only if it is a closed subset of an open set, which is also equivalent to it being an open
subset of a closed set.

Proof. We can replace Y ∩ U with Y ∩ (U ∩ Y ) to get an open subset of a closed set, and we can replace it with
U ∩ (U ∩ Y ) to get a closed subset of an open set. In general, we can find the minimal open and minimal closed sets
containing L (which exist by looking at chains of closed sets).

Definition 132
A constructible set S is a finite union of locally closed sets.

Example 133
Take X = A2 , and let S be the union of the complement of the y -axis with the origin. Then S is constructible.

Lemma 134
Suppose that S is constructible. Then we can write S = L1 ∪ · · · ∪ Lk , such that Li = Yi ∩ Ui and the Yi are
irreducible and distinct.

Proof. We decompose each closed component Yi into its irreducible components, and then we “merge” parts that
share the same closed set – this preserves the property that all Li are locally closed.

Theorem 135
Let S be the smallest family of subsets such that
• open sets are in S,

• finite unions and intersections are in S (S1 ∪ S2 and S1 ∩ S2 are in S for S1 , S2 ∈ S).

• complements S C = X − S are in S for any S ∈ S.

Then S is the family of constructible sets.

All constructible sets can be formed from open sets. To prove the remainder of the result, note that the union and
intersection of two constructible sets is constructible – if S1 = L1 ∪ · · · ∪ Lk and S2 = M1 ∪ · · · ∪ Mn , we can write
∪
S1 ∩ S2 = i,j Li ∩ Mj . This is because the intersection of locally closed sets is locally closed. We can similarly show
that complements are also constructible, so we can show closure.

53
 Theorem 136
Suppose f : Y → X is a morphism, and S is a constructible subset of Y . Then f (S) is a constructible subset of
X.

Conversely, the inverse image f −1 (T ) of a constructible set T ∈ X is constructible (but this is easier; it follows
from continuity of morphisms).

Proof. (This basically consists of reducing the problem until there is nothing left to prove.) Since we’re working with
Noetherian rings, we can prove this by Noetherian induction – by the descending chain condition, we just need to
show that if a statement is true for all closed proper subsets of Y , it is true for Y as well (and the same is true for
X, which is also closed). In other words, we can assume this statement is already true if S is contained in a proper
closed subvariety of Y or f (S) is contained in a proper closed subvariety of X.
We can now proceed with the proof. If Y1 is a proper closed subvariety of Y , we can write the constructible set S
as (S ∩ Y1 ) ∪ (S ∩ (Y \ Y1 ))), and we can check that both of these pieces are constructible subsets. But then we’ve
shown the condition for S ∩ Y1 by Noetherian induction already, so we can replace Y with Y \ Y1 finitely many times,
which lets us assume that Y is a nonempty open subvariety. By basically the same logic, we can replace X with a
nonempty open subvariety.
It suffices to show that a locally closed set satisfies this condition (because S is a finite union of such sets), and
we can write this set as C ∩ U, where C is closed and irreducible and U is open. Now Y = C ∩ (Y \ C), but Y \ C does
not intersect C, so we can assume that Y is just C. In other words, S = C ∩ U = Y ∩ U, and now we can assume Y
is just U – we’ve now reduced the problem to the case where S = Y .
Remember that we showed above that we can replace X and Y with open subvarieties, so we can write Y = Spec B
and X = Spec A. We now have an associated homomorphism ϕ : A → B – this homomorphism is injective because the
kernel corresponds to a proper closed subset (which we’ve already removed from the picture in an earlier simplification).
To finish, Noether Normalization tells us that we can pick s ∈ A such that Bs is a finite module over As [y1 , · · · , yk ].
We have surjective maps from Ys to Spec As , as well as from Spec As to Xs (because we have a finite module), which
means that there is a surjective map from Ys to Xs . Once again replacing X and Y with the open subsets Xs and Ys
gives us a surjective map, which indeed means that our image is constructible, as desired.

We’ll move on again to a nice topic: analyzing with smooth curves. Suppose that C is a smooth affine curve, q
is a point of C, and C ′ = C − {q} is its complement. We should think of q as a “limit point.”

Proposition 137
C is the closure of C ′ in the Zariski and in the classical topology.

Proof. This is true in the Zariski topology because C is irreducible (so we can’t break it up further, meaning C ′ is not
closed). In the classical topology, C ′ ’s closure must be either C ′ or C because C is closed – if it were C ′ , then q would
be open, but it is also closed. Thus, q would be an isolated point, and no such points exist.

Theorem 138
Let X be a variety, and let S be a constructible set in X. Then S is closed if and only if for all morphisms
f : C → X, we have that f (C ′ ) ⊂ S =⇒ f (C) ⊂ S.

54
 In other words, we can test if S is constructible by testing curves. Think of q as a limit point of C ′ : this says that
if S contains all of its limit points, it’s closed.

Proof sketch. The idea is that we need to “find enough curves:” we need to be able to test any point p in the closure
S to see if it is in S. To do this, we map in a curve C intersecting S at this point p, and use that as the removed
point. Here’s the main lemma that we need:

Lemma 139
Let S be a constructible subset of X, and let p ∈ S be a point. Then there exists a curve C and a point q such
that there exists a morphism f : C → X with f (C ′ ) ⊂ S and f (q) = p.

Proof of lemma. Let S = L1 ∪ · · · ∪ Lk , where Li = Yi ∩ Ui ; we can assume that the Yi are irreducible by Lemma 134.
Each Li is then some dense open subset of Yi , meaning the closure of each Li is Yi . Thus, the closure of S is the
union of Y1 through Yk , meaning any point p is in Yi for some i .
Therefore, there exists an irreducible closed set Y that contains p, as well as a nonempty open subset V ∈ Y such
that V ⊂ S (this is basically the corresponding locally closed set Li ). We can assume that Y is affine with Y = Spec B
(by taking X to be affine – this is okay because we have a Noetherian space). If dim Y = 0, then Y is just a single
point, so p ∈ S. Otherwise, we can reduce to dim Y = 1 by Krull’s theorem: this is because we can take a zero locus
from an element β ∈ B such that β(p) = 0, but β is not identically zero on any component of Z except p itself, and
consider β ∩ Y . By construction, this zero locus is closed and nonempty, and because it contains p, at least one of
these components of the zero locus contains p, and we can replace Y by one of those components – this will decrease
the dimension by at least 1.
So we can now say that dim Y = 1. Let C = Y # be a smooth curve: we know that π : Y # → Y is surjective
(because we have an integral morphism), so there exists a q ∈ C such that π(q) = p. We delete the (finite set of)
other points in the fibre over p by picking an affine open subset, and this shows that we indeed have the morphism
that we desire.

Using this curve to “test” closure gives us the result – if such a limit point p were not in S (meaning that S is not
closed), then we would not have that f (C ′ ) ⊂ S =⇒ f (C) ⊂ S.

Corollary 140
If S is a constructible subset of X, then it is closed in the Zariski topology if and only if it is closed in the classical
topology.

The classical topology is finer, so anything closed in the Zariski topology is closed in the classical topology. The
converse comes from the “testing curve” idea above.
We’ll finish today with two more results about projective space being proper:

Theorem 141
Let Z be a variety, and let Y be a closed subset of Z × Pn . Let π : Z × Pn → Z be the projection map. Then
π(Y ) is closed in Z.

Proof. This is true in the classical topology because Pn is compact, so projection takes closed sets to closed sets.
More explicitly, take a sequence of points pi in π(Y ) converging to a limit point – for each one, we can pick a point

55
(zi , xi ) in Y ⊂ Z × Pn that maps to it under π, and then we can pick a subsequence of {xi } with a limit point x in Pn ,
because of compactness. The zi also converge to some z (the same limit point that the pi converge to). Now (z, x)
is in Y (because Y is closed and therefore contains all of its limit points), and therefore its image p, which is the limit
point of the pi s, is also in π(Y ).
But now images of constructible subsets are constructible by Theorem 136, and Y is constructible (because it is
closed). Because π(Y ) is closed in the classical topology, it is closed in the Zariski topology.

Corollary 142
Let f : X → Z be a morphism of projective varieties, and Y be closed in X. Then f (Y ) is closed in Z.

Proof. Use the graph Γf , which consists of the points (x, f (x)) ⊂ X×Z. This is isomorphic to X, and so Y is isomorphic
to a closed subset of Γf ⊂ X ×Z. Now getting the values for f (Y ) is a projection map, and X ⊂ Pn =⇒ Γf ⊂ Pn ×Z,
meaning that we can use the above result.

21 April 6, 2020
We’ll be spending the next week or so on modules – the first topic here is the structure sheaf of a variety. Let X be
a variety (everything here is defined with respect to X) – we can define the category (opens), where the objects of
the category are open subsets of X and the morphisms are inclusions. (There is a morphism from V → U if V ⊂ U,
and no morphism if V 6⊂ U.)

Definition 143
The structure sheaf O is the functor from (opens) to (algebras). Specifically, if U is open, then O(U) is the
algebra of rational functions α that are regular on U.

Recall that if α is in a function field K, then α is regular at a point p ∈ X if there exists an affine open set such
that U = Spec A, p ∈ U, and α ∈ A. Now if we have an inclusion V → U of open sets, then α being regular on U
implies that it is regular on V – this means that V ⊂ U =⇒ O(U) ⊂ O(V ) , and that’s what makes the structure
sheaf into a functor. (Notice that the arrows are reversed here, so this is contravariant.)
Suppose we want to check that a rational function is regular on some open set or on X – we don’t want to check
every affine open set.

Lemma 144
Let U = Spec A be an affine open set in X that is covered by other affine sets {U i = Spec Ai } (we can assume
∩
this set is finite). Then A = Ai .

In other words, the rational functions on Spec A are just the elements of A, and checking whether functions are
regular only requires a specific open covering, not all of them.

Proof. Note that being covered here means that the U i are all subsets of U, so having maps U i → U means we have
∩
maps A → Ai – this means that every element of A is in Ai .
We know that the localizations of U form a basis for the topology. Thus, we can always assume our open coverings
are localizations, and write U i = Spec A[si−1 ], where si ∈ A are nonzero for all 1 ≤ i ≤ k. The U i s cover U, so there

56
is no point where the si s are all zero – that means that s1 , · · · , sk generate the unit ideal A. Let α be an element of
∩
the intersection Ai : since α ∈ A[si−1 ], we know that sin α ∈ A for some n. There are finitely many si s, so we can
use the same n for everything. Since {si } generate the unit ideal, so do the powers {sin }, and thus we can write
∑ ∑
1= ai sin =⇒ α = ai sin α ∈ A,
i
∩
and thus every element of Ai is in A.

Lemma 145
∩
Let U be any open set in X, and let {U i = Spec Ai } be an affine open cover of U. Then O(U) = O(U i ).

The lemma tells us that the affine open sets are the important ones here. This is basically the same as the argument
we made above – we’ll omit the proof.

Proposition 146
∩
O has the sheaf property: O(U) = O(U i ) if {U i } is an affine cover of U.

The structure sheaf is a functor, and we want it because the variety X is not just a topological space – all
one-dimensional curves are homeomorphic, but we do care about what the functions are.
We’ll now move on to the definition of O-modules: let (affines) be the subcategory of (opens), where the objects
are the affine open subsets of X, and the morphisms are localizations. This second part is not so obvious: what it
says is that a morphism V → U in (opens) – that is, an inclusion – will be a morphism in affines if (1) U = Spec A is
affine and (2) V is a localization of U, meaning that V = Us = Spec A[s −1 ] for some s 6= 0 in A.
The reason we like to restrict to (affines) is that we know what localizations look like, and we don’t have very
much to say about inclusions of open sets that aren’t localizations.

Definition 147
An O-module is a functor M from (affines) to (modules).

So in an O-module, we assign a module to each affine open set of X, which should be compatible with the
localization. Specifically, if U = Spec A is an affine open set, the M(U) is an O(U)-module, where O(U) = A. Since
our maps look like localizations, we can explain this more explicitly: if Us is a localization of U, then M(Us ) is the
localization of M(U), which we denote as M(U)s . In particular, if M(U) is an A-module M, then M(U)s is just the
localization Ms .

Definition 148
Elements of M(U) are sections of M on U.If V → U is a map in (affines), so V = Us , then the image of a
section m ∈ M(U) is the restriction of m to V .

By custom, we denote the restriction to V by the same letter m.

Example 149
In the free O-module On , the sections are vectors of sections of O.

57
Example 150
If X = Spec A is affine, then O-modules are A-modules and vice versa.

Example 151
Let p ∈ X be a point. The residue field module κp is defined as follows: if U is an element of (affines) and
U = Spec A, such that p ∈ U, then A has a residue field κ(p), and we define κp (u) = κ(p). If p 6∈ U, then we
define κp (U) = 0.

Definition 152
An ideal I of O is an O-submodule of O. A maximal ideal at a point p ∈ X is mp (U) if U = Spec A and p ∈ U
and O(U) if p 6∈ U.

Definition 153
Let M and N be O-modules. A homomorphism ϕ : M → N is defined as follows: for every U ∈ Spec A, we know
that M(U) and N (U) are O(U)-modules, and we have the homomorphism ϕ : M(U) → N (U) of O(U)-modules,
such that whenever Us is the localization of U, ϕ(Us ) is the localization of ϕ(U).

It seems annoying to define homomorphisms on every affine open set, but all of the baggage takes care of itself:

Example 154
Let ϕ : M → N be a homomorphism. Then we have a kernel K, a cokernel C, and an image – we know that the
kernel K should be an O-module, which means we need to understand K(U) for any affine U. But we can just
say that K(U) is the kernel of the homomorphism ϕ(U) : M(U) → N (U).

We just need to make sure that localization is compatible:

Lemma 155
If ϕ : M → N is a homomorphism of A-modules for a domain A, and s ∈ A is a nonzero element, then we have a
homomorphism ϕs : Ms → Ns by localization. Then ker ϕs = (ker ϕ)s .

Definition 156
f g
An sequence of O-modules · · · M → N → P · · · is exact if for any two adjacent maps f , g as above, we have
Im f = ker g.

Example 157
We have an exact sequence
0 → mp → O → κp → 0,

which means that the kernel of the homomorphism O → κp is mp and that we have an injective map mp → O.

58
22 April 8, 2020
We’ll continue to discuss modules today. Recall that O is the structure sheaf on a variety X, where O(U) is a regular
function on U. Then an O-module M is a (contravariant) functor associates a module to every affine open set: if
U = Spec A is an affine open set, then M(U) will be an O(U)-module, which is just an A-module.
We defined the category (affines) last time; the morphisms in this category are the localizations Us → U for some
nonzero s ∈ A. In order for this to be valid, we just need M(Us ) to be the localization of M(U). And again, if we
have an affine open set Spec A, then M(Us ) = Ms is an As -module.
Our goal is to extend M to all opens (instead of just affines) – our question is to ask for global sections of M,
which are the elements of M(X). They should be given by sections on an open covering {U i = Spec Ai } which agree
on the overlap, so any element M(U) should be defined with a vector (m1 , · · · , mk ), where mi ∈ M(U i ), such that
the restrictions of mi , mj to the smaller open set U ij are equal.
But we know that U ij = U i ∩ U j is the intersection of affine open sets, so it is affine open as well: it’s Spec[Ai , Aj ].
This does not always give us a map from M(U i ) to M(U ij ) unless U ij is actually a localization, but it’s not such an
important point.
In order for this to work out, we want the following sequence to be exact:
∏ ∏
0 → M(U) → M(U i ) → M(U ij ) ,
i

∏
where the first arrow tells us that the map in the second arrow from M(U) → i M(U i ) is injective (and describes
∏
the vector (m1 , · · · , mk )). Meanwhile, we should think of the last space M(U ij ) as the space of k × k matrices:
then we want our map β to send (m1 , · · · , mk ) → (zij ), where

zij = mj − mi

when restricted to U ij . When these conditions are satisfied, this yields the sheaf property.

Theorem 158
Let M be an O-module. Then M extends uniquely to a functor M from (opens) to (modules) – such that M
has the sheaf property. In addition, any homomorphism of O-modules f : M → N extends uniquely.

This proof is long – we need to check that our map β above makes sense, which amounts to showing that when
we ave two affines with one contained in the other, there exists a map. We should read about this on our own.
∏
Note that in our exact sequence, we can just take the product i<j M(U ij ), because U ij and U ji are the same and
U ii = U i .

Example 159
Suppose X = U 0 ∪ U 1 is covered by those two affine open subsets. Then the sheaf axiom is equivalent to the
following sequence being exact:

0 → M(U) → M(U ′ ) × M(U 1 ) → M(U 01 ).

If we say that X = P1 is the projective line, U 0 = U 1 are the standard affine open sets in X, and U 0 = Spec A0
and U1 = Spec A1 . Then A0 = C[u], where u = x1
x0 , and A1 = C[v ] = x0
x1 = u −1 . Thus

U 01 = Spec C[u, u −1 ]

59
which is the Laurent polynomials in u. So if we have an O-module, we can get an A0 -module M0 and an A1 -module
M1 . Now the Mi are the sections of M(U i ), and we must have

M(U 01 ) = M0 [u −1 ] = M1 [v −1 ].

On the other hand, if we have an A0 -module M0 and an A1 -module M1 , and suppose there’s an isomorphism
θ : M0 [u −1 ] → M1 [v −1 ] An O-module, we should have compatibility from localization, and thus we get an O-module
M.

Example 160
Suppose M0 is a free A0 -module with basis B0 , and M1 is a free B1 -module with basis B1 .

When we do our localizations, M0 [u −1 ] and M1 [v −1 ] are free A01 -modules with bases B0 , B1 , and they are isomor-
phic. So θ is some invertible matrix P with entries in A01 – note that we can chagne B0 by an invertible A0 -matrix Q0
and and B1 by an invertible A1 -matrix to Q1 , so here P determines M.

Theorem 161 (Birkhoff-Grothendieck)

Given an invertible A01 -matrix P (this is the Laurent polynomial ring), there exist an invertible A0 -matrix Q0 and
an invertible A1 -matrix Q1 such that Q−1
0 P Q1 is diagonal.

We know that the units of A0 are cu k for some k ∈ Z, This means that Q−1
0 P Q1 looks like a diagonal matrix with
entries of the form uiei , where all ei are integers. We’ll prove this later on with cohomology.

Corollary 162
Suppose that X = Spec A is affine. Then an O-module is equivalent to an A-module.

Proof. If we’re given an O-module M, then M = M(X) is an A-module. The localizations form a basis, so it’s enough
to just define M(Xs ) = Ms .

So the localizations are enough in this case where x is just an (affine)!

Proposition 163 (Coherence property)

If Y is any open set, and s ∈ Oy is a regular function, then Ys is Y \ {(zeros of s}. Then M(Ys ) is the localization
of Y .

Remember that the definition of a module M means it has to be given on affines – thus we know that this is true by
definition for an affine open set Y . This property extends that fact to opens in general! We then call an O-module M
that has been extended to all opens a quasicoherent sheaf. Note that working with the extended module is harder –
we like working with (affines) and localizations, because many operations on modules are compatible with localizations,
while very few operations work on all open sets. The only notable exception to this is the kernel – for example, the
cokernel doesn’t work:

Example 164
Suppose f : M → N is a homomorphism of O-modules. Then c = coker f is the module such that C(U) is the
cokernel of the map M(U) → N (U) for an affine open set U = Spec A, but not otherwise.

60
23 April 10, 2020
Recall that for an O-module M for a variety X extends uniquely to a functor on all opens with the sheaf property.

Lemma 165
The sections of M on the empty set are {0}.

This is more semantic than anything else – we’re bringing it up because we’ve been avoiding the empty set.

Example 166
A tensor product is an example of an O-module.

Recall that that if we have two A-modules M, N, we can construct an A-module M ⊗ N generated by tensors of
the form m ⊗A n (with m ∈ M, n ∈ N), with the bilinear relations

(m1 + m2 ) ⊗ n = m1 ⊗ n + m2 ⊗ n, m ⊗ (n1 + n2 ) = m ⊗ n1 + m ⊗ n2 , am ⊗ n = m ⊗ an.

One special case, where M and N are free modules with bases (m1 , · · · , mr ) and (n1 , · · · , ns ), yields a tensor product
M ⊗ N which is a free module with basis {mi ⊗ nj }. We can think of M = Ar as forming the rows and N = As as
forming the columns of an r × s matrix.

Definition 167
Let M and N be O-modules. Then the tensor product is defined by its sections on U:

M ⊗O N(U) = M(U) ⊗O(U) N(U).

(In other words, the scalars of the tensor product are the regular functions on U.) This is a valid definition, because
taking any nonzero element s ∈ A yields
(M ⊗A N)s ∼
= M s ⊗A s N s .

Definition 168
Let f : Y → X be a morphism, and let N be an OY -module. The direct image f∗ N is an OX -module defined as
follows: suppose U = Spec A is an affine open set in X, meaning that f −1 U is some open set in Y (not necessarily
affine). Then the sections of f∗ N on U are the sections of N on f −1 U.

We can write this with less words as

(f∗ N )(U) = N (f −1 U).

Lemma 169
The direct image f∗ N is an OX -module.

We prove this using the coherence property.

61
 Example 170
i
Suppose Y → X is an inclusion of a closed subvariety Y into X. Then i −1 (U) = U ∩ Y .

Then we should have

f∗ N (U) = N (U ∩ Y ).

We’re mentioning this special case because it’s called the extension of N by zero. This is because any open set
U ∈ X that doesn’t intersect Y yields
(f∗ N )(U) = N (∅) = 0.

Example 171
j
Suppose Y → X is an inclusion of an open set.

Inverse images are intersections again, so we still have

(f∗ N )(U) = N (U ∩ Y ).

Meanwhile, we can also restrict an OX module M to Y , because V being (affine) open in Y implies that it is
(affine) open in X. This means that M(V ) is defined. For example, restricting OX to some open Y will give us the
structure sheaf OY on Y – this is different from the direct image. If M is an OX module, we’ll say that the restriction
of M = MX to an open Y is denoted MY .

Example 172
What happens when we combine a restriction and a direct image: what is j∗ OY ?

We know that the sections on U are the same as the sections of OY on U ∩ Y (which is a smaller open set in U),
and this is the same as the regular functions on U ∩ Y – thus the sections can have poles in U but outside U ∩ Y , and
we have a map OX → j∗ OY .
Our next topic is that of twisting modules on projective space Pr . Recall that a homogeneous fraction f = g
h
is a fraction where g and h are each homogeneous: f has degree deg f = (deg g − deg h), and if g, h are relatively
prime, then f is regular on an open set U if h does not vanish on U.

Definition 173
The sections of the twisting module O(n) on an open set U ∈ Pr are defined to be the homogeneous functions
of degree n that are regular on U.

These are particularly important – we’ll see why soon. Notably, if f is a homogeneous polynomial of degree d, then
multiplying by f defines a homomorphism from O(n) to O(n + d), because f is regular everywhere.
Suppose that we restrict O(n) to the standard open set U 0 (that is, the points where x0 6= 0). Then the restriction
is a free OU 0 -module of rank 1 with basis {x0n }. This is because the homogeneous polynomials that are regular on
all of U 0 are (scalar multiples of) powers of x0 (everything else vanishes somewhere). Thus, the regular homogeneous
g
fractions look like f = x0k
for some k – we can show the rest as an exercise.
Here’s another definition of the twisting module: let H be the hyperplane where x0 = 0. Then we can define the
sections of O(nH) on U to be the rational functions f such that f x0n is regular on U. So now f can have poles of

62
order at most n on U ∩ H. Now we have an isomorphism
xn
O(nH) →
0
O(n).

We know that O ⊂ O(H) ⊂ O(2H) · · · , and we can consider

lim O(nH).
n→∞

This is the set of rational functions such that f x0n is regular for large enough n. Note that if f x0n is regular on an open
set V , then f is a section of O on V ∩ U 0 , so f is an element of the direct image j∗ OU 0 . In other words,

lim O(nH) = j∗ OU 0 ,

where j is the inclusion of U 0 into Pr .

We’ll finish by discussing how to generate an O-module M. Suppose we have global sections m = (m1 , · · · , mk )
m
of M – they are elements of M(Pr ), and we get homomorphisms Ok → M, sending
∑
(α1 , · · · , αk ) → α i mi .

When this map is surjective, we say that m generates the OPr module M. Then M is a finite O-module for every
affine open set U = Spec A if M = M(U) is a finite A = O(U)-module. (Note that this is only true because U is
affine!)
We’ll introduce the notation
M(n) = M ⊗O O(n).

Theorem 174
Say that M is a finite O-module on Pr . Then the twisting module M(n) can be (finitely) generated by global
sections for sufficiently large n.

M itself is not generated by global sections – this is only true on the affines, and M does not need to have any
sections at all. We’ll review this a bit at the beginning of next lecture and then move on to cohomology.

24 April 13, 2020

Our goal for today is to show the result from the end of last time: given a finite O-module M on Pn , we can generate
the twisting module M(n) by global sections. Recall that the twisting module O(n) is defined such that the sections
g
on an open set U are the homogeneous fractions f = h are regular on all of U. For the hyperplane H at infinity, we can
define O(nH) similarly: its sections on U are the homogeneous rational functions of degree 0 such that x n α is regular of
degree n. Then multiplying by x n gives us an isomorphism from O(nH) to O(n). Recall that O ⊂ O(H) ⊂ O(2H) · · ·
and so on, and taking the limit as n → ∞ yields a limit lim O(nH) such that its sections on an open set U are the
rational functions α with x0n α regular for sufficiently large n.
We discussed that this limit lim O(nH) is the direct image j∗ OU 0 , where j is the inclusion of U 0 into X = Pr .
Recall that the sections of the direct image are defined as

[j∗ OU0 ](V ) = OU 0 (V ) = OX (V ),

which is the set of regular functions on V . Let’s show this explicitly:

63
 g
Proof. A homogeneous fraction f = h (for relatively prime g, h) is regular on an open set U 0 if the denominator
doesn’t vanish on U 0 , meaning it must be a power of x0 (times a scalar). To show when f is regular on an open set
V ∩ U 0 , we factor the denominator h = h1 · · · hk , where the hi are irreducible polynomials. Then we care about having
h 6= 0 everywhere on V ∩ U 0 , and this happens whenever all hi 6= 0 on V ∩ U 0 . Letting Yi be the set of zeros of hi ,
we have an irreducible closed set in X = Pr . But V ∩ U 0 is open, so Yi ∩ (V ∩ U 0 ) is open in Yi , meaning this set is
dense or empty.
And now if hi 6= 0, we’re in the case where we want Yi ∩ (V ∩ U 0 ) must be empty. We can write this as
(Yi ∩ V ) ∩ (Yi ∩ U 0 ), and both terms here are either dense in Yi or empty – thus we need at least one of them to be
empty. In the latter case, we have that hi = cx0 (analogous to the simple case of being regular on U 0 ), which we’ve
analyzed before. But in the former case, we have hi 6= 0 on all of V . Thus, f = g
h is regular on V ∩ U 0 if h = h′ x0k ,
where h′ is a regular function on V . (Basically, we have a power of x0 , and the rest can’t vanish on all of V .) Therefore
we need that f x0k is regular on V if k is sufficiently large.
But now [j∗ OU 0 ](V ) = OU 0 (V ∩ U 0 ) is the set of rational functions of degree 0 satisfying the condition that x0k α
is regular on V for sufficiently large k: thus we do have j∗ OU 0 as the limit O(nH).

Remember that we can twist a module: because O ⊂ O(H) ⊂ · · · , we can define

M(n) = M ⊗O O(n), M(nH) = MM ⊗ O(nH).

(These maps may not be injective if M has torsion.) Then the limit here is

lim M(nH) = M ⊗O lim O(nH) = M ⊗O j∗ OU 0 = j∗ MU 0 .

n→∞

This is important, and we’ll discuss it a bit more: if we’re looking at a standard affine open of Pr , then O(U i ) are
the rational functions with denominator xik . Then the sections of O(n) on U i are homogeneous fractions f = g
x0k
of
degree n, so g is of degree n + k. Then [O(nH)](U ) is the set of fractions of degree 0 such that multiplying by x0n
i

yields a regular function on U i , so x0n f = g

xik
=⇒ f = g
x0m xik
, where k is aribtrary, m ≤ n, and g is a function of degree
(n + k).
But now O maps isomorphically to O(n) on U i by the multiply-by-xin map, and we have an isomorphism on X from
O(n) to O(nH) by dividing by x0n . So then an O-module M goes to M(n) via the isomorphism map 1 ⊗ xin , while
M(n) goes to M(nH) via the isomorphism map 1 ⊗ x0−n .
So now let’s go back to the result we’re trying to show: recall that a O-module M is a finite O-module if for
every affine open set U = Spec A, M(U) is a finite O(U)-module (that is, a finite A-module). We discussed trying to
generate an O-module M by global sections m1 , · · · , mk ∈ M(X) by considering the homomorphism
∑
(α1 , · · · , αk → αi m i ,

and the homomorphism being surjective means that we generate M(X).

Proof of theorem. Let M = M(U 0 ) and A = O(U 0 ). M is a finite A-module, and it is also [M(U)](U 0 ).
It’s enough to show that these global sections of M(nH) restrict to generators of A mod M. This is
because we know that whenever U = Spec A is affine, OU -modules M correspond to A-modules M. And we know
how to localize this, so we can extend to the localizations of U, which is enough. Then if the global sections of
M(nH) generate M, then they generate MU 0 . We have a map OUk 0 → MU 0 – let C be the cokernel of the supposed
generators. We know that the cokernel CU 0 is zero here, and we can similarly show that CU i is zero (possibly with a
larger value of n), and then we use the sheaf axiom: remember that we have an injective map which can be represented

64
 ∏
as 0 → C(V ) → i C(V i ).
But lim M(nH) = j∗ MU 0 , where j is again the inclusion map, and we can pick generators for M = M(U 0 ) – this
is supposed to be a finite A-module. We can represent by global sections of M(nH) – call them m1′ , · · · , mk′ – since
these map to m1 , · · · , mk , these generate M.

25 April 15, 2020

We’ll begin talking about cohomology of O-modules today. There isn’t really any easy way to uniquely describe what’s
going on here, though there are lots of different constructions.
Let X be a variety, and consider the short exact sequence
f g
0 → L → M → N = 0.

This means that f is injective, ker g = Im f , and g is surjective. Then if U = Spec A is an affine open set of X, we
can map the sections via
f (U) g(U)
0 → L(U) → M(U) → N (U) → 0 :

this is an exact sequence of A-modules, where A = O(U). However, the sequence may not be exact if U is open but
not completely affine:

Lemma 175
f (U) g(U)
If we have the short exact sequence 0 → L → M → N → 0 above, then 0 → L(U) → M(U) → N (U) is
exact for any open set U. (We do not need the surjectivity of g for this.)

Proof. Remember that an O-module is defined on the affine opens, and it extends to all opens by the sheaf property
of L, M, and N . If {U i } is an affine cover of U, then we have the exact sequence
∏ ∏
0 → M(U) → M(U i ) → M(U i ∩ U j )
i i,j

by the definition of M(U). We can then draw a commutative diagram:

0 0 0

0 L(U) M(U) N (U)

∏ ∏ ∏
0 i L(U i ) i M(U i ) i N (U i )

∏ ∏ ∏
0 i,j L(U ij ) i,j M(U ij ) i,j N (U ij )

The columns are exact by the sheaf property, and because the U i s and U ij s are affine, the last two rows are exact
by definition (of 0 → L → M → N → 0 being exact). Our goal is to prove that the first row is also exact (in other
words, that the additional blue 0 and arrows are valid).

65
 ∏
Take an element of L(U) that goes to zero in M(U), and take its image in L(U i ). L(U i ) is injective into
∏
i M(U i ), so it goes to 0, but then the L column is zero, so L(U) must be zero. We can use a similar Snake Lemma
idea to show the middle part works out too.

Basically, the sections are left exact but not necessarily right exact.
The next idea is to study global sections: by the above lemma, we know that we have the exact sequence

0 → L(X) → M(X) → N (X).

Then the cohomology is a sequence of functors H 0 , H 1 , · · · , from O-modules to vector spaces, which “substitutes for
the lack of exactness.”
This has three characteristic properties:

• H 0 (M) = M(X) are just the global sections.

• We have the cohomology sequence which describes the lack of exactness: if we have a short exact sequence
0 → L → M → N → 0, we get the long exact sequence
δ0
0 → H 0 (L) → H 0 (M) → H 0 (N ) →
δ1
→ H 1 (L) → H 1 (M) → H 1 (N ) → · · ·

δ 0 is called the coboundary map, so computing H 1 (L) will tells us to what extent exactness fails. We’ll see
later that all but finitely many of these are zero.

• Basically, if X is affine, then the global section functor is exact, so we don’t need higher cohomology. The actual
statement is a bit more complicated: if U = Spec A is an affine open subset of X, and N is an OU -module, then
H q (j∗ N ) = 0 for all positive q. (Meanwhile, H 0 (j∗ N ) is defined to be (j∗ N )(X) = N (U), so it is not zero.)

Corollary 176
If X is affine, then H q (M) = 0 for all q > 0 and all M.

Proof. Use the identity map X → X with the third charactersitic property. Then j∗ N is just N .

Theorem 177
There is a cohomology theory satisfying the three characteristic properties above, unique up to isomorphism.

Unfortunately, there is no natural construction of the cohomology – the best way to think about it is to just work
with the characteristic properties.

Proof of uniqueness of cohomology. Say we are given a cohomology. Choose an affine open cover U = {U ν } of X –
it will follow from the logic that the choice of affine cover doesn’t matter. Let j be the family of inclusions of U into
X. Let M be an O-module – we want to describe its cohomology. We can restrict to open subsets: let
∏ ∏
MU = MU ν , j∗ν MU ν .

66
 Lemma 178
Let R = j∗ Mν . There exists a canonical injective map from M → R, and H q (R) = 0 if q > 0.

∏
Proof. Let V be open in X. We know that V ν = V ∩ U ν cover V , so M(V ) is contained in the product M(V ν ),
∏
which we can write as a product MU ν (V ). By definition, this is R(V ), so M(V ) is contained in R(V ), meaning we
do have injectivity.
For the second part,
∏
H q (R) = H q (j∗µ MU ν ) = 0,

because each U ν is affine.

Now we can form a short exact sequence 0 → M → R → S → 0, where S is the cokernel of the map from M to
R. We know that the correspondending cohomology sequence has

0 → H 0 (M) → H 0 (R) → H 0 (S) →

H1 (M) → 0 → H 1 (S) . . . ,

but the key point here is that in the exact sequence

δ0
0 → H 0 (M) → H 0 (R) → H 0 (S) → H 1 (M)

with the cohomology given and a chosen covering of X, we know what the H 0 s look like – this gives us the first three
terms of the exact sequence, because H 0 (R) depends on the covering rather than the cohomology.
d
From here, we can look at 0 → H 1 (S) → H 2 (M) → 0: H 1 (S) is determined by M, not by the cohomology, so
H 1 (S) is unique, and therefore H 2 (M) is unique. Now we can repeat this argument inductively to show that everything
is unique: we do need to check uniquness in the δs, but the boxed sequence identifies H 1 (M) uniquely independent of
the cohomology – this fixes δ, and the idea is that knowing the coboundary maps is enough.

26 April 17, 2020

Recall the characteristic propertes of cohomology, which is a series of functors H 0 , H 1 , H 2 , · · · . Here, M is an OX -
module for some variety X.

• H 0 (X, M) = H 0 (M) = X is just the vector space of global sections. But there isn’t a very clear correspondence:
it’s not clear that when M is a finite module over the coordinate ring, H 0 (M) forms a vector space over the
scalars.

• We have the cohomology sequence: if we have a short exact sequence 0 → L → M → N → 0, this yields a
sequence of maps H q (L) → H q (M) → H q (M) (because H q is a functor, these maps exist by definition), and
then we have a coboundary map δ to H q+1 (L) such that the entire sequence is exact.

• If U is an affine open set in X, j is the inclusion map U → X, and N is an OU -module, then H q (X, j∗ N ) = 0)
for all q > 0. In other words, we don’t really have a cohomology for affine open sets – it just looks the same as
the exact sequence we already have.

We can talk about the third point here in more generality:

67
 Lemma 179
If Y is affine, f : Y → X is a morphism, and N is an OY -module, then H q (X, f∗ N ) = 0 for all q > 0.

This has consequences for what are called affine morphisms:

Definition 180
Let Y and X be varieties. A morphism f : Y → X is an affine morphism if for all affine open sets U in X,
V = f −1 (U) is affine in Y .

Example 181
The most important case is when f is an inclusion of a closed (or open) subvariety Y into X: this is indeed an
affine open set.

Theorem 182
Let f : Y → X be an affine morphism, and let N be an OY -module. Then the cohomology

H q (Y, N ) ∼
= H q (X, f∗ N ).

Proof. We want to show that the sequence of H q (X, f∗ N )s is the cohomology for N , so we just need to verify the
three characteristic properties. Define F q (N ) = H q (X, f∗ N ) for convenience.
For the first characteristic property, we wish to show that F 0 (N ) = N (Y ) forms the space of global sections. This
is H 0 (X, f∗ N ), which is the space of global sections of f∗ N on X. But f∗ is defined to have the sections of N on the
inverse image of X, which means this is indeed N (Y ) = H 0 (Y, N ).
For the second property, we need to look at a short exact sequence 0 → L → M → N → 0 of OY -modules.
Because f is an affine morphism, we claim that the sequence 0 → f∗ L → f∗ M → f∗ N → 0 is also exact. But the
definition of being exact depends on being exact on the affine open sets, and if U is open in X, we want to know
whether the sequence
0 → f∗ L(U) → f∗ M(U) → f∗ N (U) → 0

is also exact. But letting V ∈ Y be the inverse image f −1 (U), we konw that f∗ L(U) = L(V ), so that sequence is
actually just
0 → L(V ) → M(V ) → N (V ) → 0,

and this is exact because V is an affine open set of Y and 0 → L → M → N → 0 is exact.

For the third property, we take an affine open V ∈ Y and an OV -module N , and we let j : V → Y be the inclusion
map. We want to show that F q (j∗ N ) = 0 for all q > 0, and this is just

F q (j∗ N ) = H q (X, f∗ j∗ N ).

The composition f∗ ◦ j∗ is a morphism, and V is affine, so Lemma 179 tells us that this is zero, as desired.

Example 183
Our goal with this will be to compute a specific cohomology: that of the twisting module O(n) on projective
space Pd .

68
 First of all, H 0 (Pd , O(n) is the set of homogeneous fractions of degree n, but being a global section means that
our denominator must be a constant (so it doesn’t vanish anywhere), so this is the space of homogeneous polynomials
( )
d -degree vector space whenever n ≥ 0.
of degree n, which is a n+d
To proceed, let’s map OP (n) → OH (n) to the hyperplane at infinity where x0 = 0: we just send x0 to 0. The
kernel is the set of homogeneous polynomials of degree n − 1 times x0 , so we get the exact sequence
·x
0 → OP (n − 1) →0 OP (n) → OH (n) → 0.

This gives us a cohomology sequence: we’ll just write the dimensions of the spaces instead of the spaces themselves,
which yields ( ) ( ) ( )
(n − 1) + d n+d n + (d − 1)
0→ → → .
d d d −1
(Indeed, for this to be an exact sequence, the dimension of the middle space is supposed to be the sum of the
dimensions of the two outside spaces, and this is Pascal’s identity.)

Theorem 184
We have H q (O(n)) = 0 for all q > 0 and n ≥ 0.

Proof. We induct on d: we can assume this is true for the cohomology on H, which is Pd−1 . So then we can assume
the third columns of the dimensions of H q are all zero. But the first columns all have dimension zero by induction on
n, so the second column all has dimension zero.

Theorem 185
Suppose n = −r for some r > 0. Then the dimension hq (O(−r )) = 0 if q 6= d, and the dimension of the dth
cohomology ( )
r −1
hd (O(−r )) = .
d

Proof. First, we do r = 1. We look at our exact sequence (where i∗ is the inclusion of H into P)

0 → OP (−1) → OP → i∗ OH → 0 :

the dimensions on the seccond and third columsn are 1 on the first cohomology (corresponding to the constant
functions, the scalars), and zero everywhere else. So all of the cohomology on the first columnOP (−1) is 0 for O(−1).
And then an induction argument works just like for the positive case: consider

0 → OP (−r − 1) → OP (−r ) → OH (−r ) → 0,

inducting on r in the second column and d in the third.

Example 186
Let Y be a plane curve of degree r in P2 . Then we have (where i∗ is the inclusion Y → P2 )
·f
0 → OP (−r ) → OP → i∗ OY → 0.

(a regular function on P2 restricts to a regular function on Y , and the functions that are 0 on Y are those divisible
by f ).

69
 (r −1)
We’ve now computed the dimensions in the first two columns: column 1 (OP (−r ) has 0, 0, 2 , followed by a
bunch of zeros, and column 2 (OP ) has 1 followed by a bunch of 0s. So then using properties of exactness gives us
( )
that column 3 (OY ) has dimensions 1, r −1
2 , 0, · · · . In other words,

h1 (Y, OY ) = pa

is the arithmetic genus of Y , which is also equal to g, the geometric genus, when Y is smooth.
In both of the above cases, it was okay to just ignore the i∗ , because the cohomology transfers over:

H q (Y, OY ) ∼
= H q (P2 , i∗ OY ),

because i∗ is an affine morphism.

Theorem 187
Let M be a finite OX -module, where X = Pn . Then

H q (M(n)) = 0

for sufficiently large “twistings” n.

Proof. We know that M(r ) is generated by global sections for sufficiently large r : thus, there exists a map from
Om → M(r ) (where O is the structure sheaf – sufficiently large twists are finitely generated), and if K is the kernel,
we have the exact sequence
0 → K → Om → M(r ) → 0.

Twist this sequence, which yields

0 → K(r ) → O(n)m → M(n + r ) → 0.

We know that the middle column has dimension 0 for q > 0 and n ≥ 0 by Theorem 184. But then we have
δ
H q (O(n)m ) → H q (M(n + r )) → H q+1 (K(n)) → H q+1 (O(n)m ),

where the first and last spaces are 0.

Lemma 188
H q (M) = 0 for sufficiently large q.

Apply this to the coboundary map, which is an isomorphism: H q (M(n + 1)) → H q+1 (K(n)) for sufficiently large
q, so because H q+1 (K(n)) is zero for sufficiently large q, H q (M(n + r )) is also zero for sufficiently large q.

27 April 22, 2020

Due to popular vote, we’ll be replacing the quiz with a final assignment: it will be more cumulative but also more
routine than the usual homework questions.
We’ll begin by discussing support:

70
 Definition 189
Suppose that A is a finite type domain for simplicity and M is a finite A-module. The annihilator of M, denoted
ann M, is an ideal of A: it’s the set of α such that αM = 0. The support of M, denoted supp M, is the zero
locus V (ann M) of the annihilator in Spec A.

This definition carries over to O-modules, because both the support and annihilator are compatible with localization.

Proposition 190
Let M = M/(mp M) = M ⊗A k(p). Then

supp M = {p ∈ X : M 6= 0}.

In other words, the support of M is the “set of points of M that are not zero.”

Proof. Let I be the annihilator ann M. If I 6⊂ mp , then there exists α ∈ I such that α(p) 6= 0. But on M, scalar
multiplication by α is the same as multiplying by α(p). If this value is nonzero, then αM = α(p)M = M is not zero.
On the other hand, if M = 0, then mp M = M, and now Nakayama tells us that there exists an element z ∈ mp
such that (1 − z)M = 0. Thus an element of I, 1 − z ∈ ann M, so I is not contained in the maximal ideal.

The next result we’ll cover today is the vanishing of cohomology for large dimension. First of all, suppose that
X is a projective variety and we have an inclusion ι : X → P. Let M be an OX -module: we know that

H q (X, M) ∼
= H q (R, ι∗ M),

because ι is an affine morphism. So we can work with cohomology on projective space instead, which is convenient
because we know how twisting works there.
So now we can let M be an O-module on P, and let U 0 : x0 6= 0 be the standard affine open set. If we let
j : U 0 → P be the inclusion map, we have the sequence

M → M(H) → M(2H) → · · · ,

where M(H) = M ⊗O O(H), such that the limit M(nH) is the direct image j∗ MU 0 .

Theorem 191
Let M be a finite O-module on projective space, and suppose supp m has dimension k. Then H q (M) = 0 for all
q > k.

Proof. Consider the map from M(−1) to M, given by multiplying by x0 . Embed this in the exact sequence

0 → K → M(−1) → M → C → 0.

The twists have the same support as M: O(n) is locally isomorphic to O. x0 annihilates K and C, so we can assume
that the support of K (some closed set in projective space) is not contained in the hyperplane, meaning that the
dimensions of the support of K and C are both less than k. (The support of K is contained in the support of M, and
we can assume that x0 is not zero on any component of the support of M.)
Here, we have the exact sequences
0 → K → M−1 → M → 0

71
which corresponds to the cohomology sequence

H q (K) → H(M(−1)) → H q (M) → Hq+1 .

By induction, we know that the first and last of these are zero, so we have an isomorphism between H q (N ) and
H(M(−1)). Similarly,
0 → M(−1) → M → C → 0

means that H q (M(−1)) is isomorphic to Hq (M).

Now we can twist: replacing M by M(n),

H q (M(n − 1)) ∼
= H q (M(n))

for n > k. And then taking the limit, for all q > k,

lim H q M(n) = H q (lim M(n)) = H q (j∗ M0U ) = 0,

because U 0 is an affine open set. This means that H q must be zero for sufficiently large 0.

Corollary 192
If P = Pn , then H q (M) = 0 for all q > n.

Our next result basically says that “twisting kills the cohomology:”

Theorem 193
Let M be a finite O-module on P. Then H q (M(n)) = 0 for all q > 0 and n sufficiently large.

Proof. Recall that M(r ) is generated by global sections for large enough r , which means we have a surjective map
represented by the exact sequence Ok → M(r ) → 0. Letting N be the kernel, we have some exact sequence

0 → N → Ok → M(r ) → 0.

Twist this sequence to get

0 → N (d) → O(d)k → M(d + r ) → 0,

and we know this yield the cohomology map

H q O(d)k → H q M(d + r ) → H q+1 N (d) → H q+1 O(d)k ,

and the first and last thing here are zero, so the middle two are isomorphic. So by the previous result, we know that
H q (M(d + r )) = 0 for large enough q, and the same logic shows that H q N (d ′ ) = 0 for some large enough d ′ . Thus
H q−1 M(d ′ + r ) = 0, and we can keep repeating this down for smaller and smaller q (adding some r each time).

Theorem 194
Let M be a finite module on P. Then H q (M) is a finite-dimensional vector space for all q.

Proof for q > 0. As above, embed the multiply-by-x0 -map in an exact sequence

0 → K → M(−1) → M → C → 0.

72
We can induct on the dimension of the support: because K and C are of lower dimension, it suffices to assume
H q K, H q C are finite-dimensional vector spaces. We have the exact sequences

H q K → H q (M)(−1) → H q M → H q+1 K,

where the first and last term are both finite-dimensional vector spaces. Then the middle two must be both finite-
dimensional or both infinite-dimensional. The same argument works for the sequence

H q−1 C → H q−1 M → H q M → H q C.

so we know that H q M(−1) and H q (M) are either both finite-dimensional or both infinite-dimensional. Now we can
twist and use the previous theorem: H q (M(n)) will eventually have cohomology zero, and we can use descending
induction on n to show that all the twists have finite dimension.

The q = 0 case is harder to prove directly – it’s not obvious that the global sections form a finite-dimensional
vector space.

28 April 24, 2020

We’ll discuss the cohomology of projective smooth curves today – unfortunately, we won’t get time to talk about
surfaces.
First of all, a review: we define cohomology on projective space P for a finite O-module (finiteness isn’t officially
necessary, but it makes things easier for our discussion – for example, the support exists). There are three theorems
we shuld know about:
• If the dimension of the support supp M is k, then H q M = 0 for all q > k.

• The cohomology of the twist H q (MM(n)) is 0 for sufficiently large n for any q > 0. (This one will be less
important today.)

• H q (M) is a finite-dimensional vector space (this requires finite module).

These come from a paper by Serre from 1955 – this was also the first time the concept of an O-module was
introduced.
One other fact we should know: if i∗ : X → P is an inclusion of a projective variety, and M is an OX -module, then

H q (X, M) ∼
= H q (P, i∗ M).

This means that both the first and third result that we’ve stated above always hold for any projective variety.
We’ll now move to new materials. Recall that local rings of a smooth projective curve X are valuation rings: let
vp be the valuation at the point p ∈ X.

73
 Definition 195
A divisor on a smooth projective curve X is a finite combination of points
∑
D= ri pi , ri ∈ Z.
i

The divisor of a rational function f is

∑
divf = vp (f )p.
p∈X

Since the valuation is zero on all but finitely many points, the divisor of a rational function is indeed a finite sum:
it’s basically the sum of the zeros minus the sum of the poles, counting multiplicity. Here’s how we can make that
more precise:

Definition 196
If vp (f ) > 0, then f has a zero of order r = vp (f ). Meanwhile, when vp (f ) < 0, then f has a pole of order
r = −vp (f ).

Definition 197
A divisor D is effective if ri ≥ 0 for all i : then we say that D ≥ 0. Given an open set U, D is effective on U if
ri ≥ 0 for all pi ∈ U.

One classical problem of algebraic geometry is to determine the rational functions f whose poles are bounded by
a given effective divisor, and we’ll use cohomology to discuss this.

Definition 198
An O-module associated to a divisor D is defined such that its sections on O(D) on an open set U are the
rational functions f such that div f + D is effective, plus the zero function.

By definition, the cohomology H 0 (O(D) is exactly this set of rational functions f such that div f + D ≥ 0, as well
as zero. This is the set that we’re trying to understand in the classical problem.

Lemma 199
O(D) is an O-module, and it is locally free of rank 1 (so it is free on each open set, meaning it looks like O).

Example 200
Consider the divisor D = −p.

This is the set of points such that div f − p ≥ 0, meaning that f must have a zero at p (so that it cancels out with
the −p). Thus, on any open set U that contains p,

O(−p)(U) = functions with a zero at p = mp .

On the other hand, if U does not contain p, then the sections are just the regular functions.

74
 Example 201
Now consider the divisor D = p.

Then our functions are allowed to have a pole of order at most 1 at p, so on any open set U that contains p,

O(p)(U) = rational functions with a possible pole at p.

(And again, we just get the regular functions for any U not containing p.)
The reason we care about cohomology is that it tells us that dimension of these spaces. We’ll start by looking at
our first example with the maximal ideal: we have the exact sequence

0 → O(−p) → O → κp → 0,

where κp is the one-point module κ(p). We’ll tensor this with O(D): the tensor product is right exact normally, but
because O(D) is locally free, we’ll still have the whole exact sequence.

Lemma 202
The module associated to O(D + E) is isomorphic to O(D) ⊗O O(E).

Basically, we know that div f + div g = div f g: we’re doing a bit of handwaving here. But then we end up with
the exact sequence
0 → O(D − p) → O(D) → ε → 0,

where ε is a one-dimensional module concentrated at p (it’s also a residue field like κ(p), but not canonically). When
we write down the cohomology, we’re working with a curve, which means that the dimensions h2 and higher are zero
for all of O(D − p), O(D), and ε.
We know that ε is a one-dimensional module, so its global sections have dimension 1: h1 (ε) = 1 . And its support
has dimension 0 (a single point has dimension 0), so H 1 (ε) has dimension 0.
So now there are only two possibilities to account for the boxed 1: either we have

h0 (O(D)) = h0 O(D − p) + 1 =⇒ h1 (O(D)) = h1 O(D − p),

meaning the 1 is absorbed into the sequence at level h0 , or

h0 (O(D)) = h0 O(D − p) =⇒ h1 (O(D)) = h1 O(D − p) − 1

meaning the 1 is absorbed at level h1 .

Definition 203
Let M be a finite O-module. The Euler characteristic χ(M) is defined via
∑
χ(M) = (−1)q hq (M).
q

In the case of a curve, only the dimensions h0 and h1 can be nonzero, so

χ(M) = h0 (M) − h1 (M).

75
Note that this already requires the first and third results from above to be a consistent definition. But we can plug
this in to both of the two cases above:

Corollary 204
For a smooth curve, we always have
χ(O(D)) = χ(O(D − p)) + 1.

Our next result shows that we can actually compute the Euler characteristic:

Definition 205
∑
The degree of a divisor D i ri pi is the sum of its coefficients:
∑
deg D = ri .
i

Theorem 206 (Riemann-Roch, version 1)

We have
χ(O(D)) = deg D + χ(O).

So if there was no h1 , we would have already computed the dimension for our classical problem.

Proof. This theorem is clear for the zero divisor D. We can get to O(D) by adding or subtracting finitely many points.
But each time we go from D to D − p or vice versa, and the above corollary shows us that we gain the appropriate
1 in each case. So induction of our previous corollary yields the result.

This version of Riemann-Roch is good for h0 (looking at the global sections), but not for h1 .

Corollary 207
Let f be a rational function. Then deg(div(f )) = 0 (this means the number of zeros of f is equal to the number
of poles of f ), and this is also true if we replace “zeros of f ” with “zeros of f − c” for some c ∈ C.

f
Proof. Let D = div(f ), and consider the isomorphism O(D) → 0. Indeed, the sections of O(D) on U are the rational
functions α such that div(α) + D ≥ 0 on U, and

div(f α) = div(f ) + div(α) ≥ 0

on all of U. Thus, f α must be a section of O on U. Going backwards, we know by Riemann-Roch that χ(O(D)) =
deg D + χ(O), but we also know that χ(O(D)) = χ(O), so deg D = 0 because isomorphic modules have the same
characteristic. And f has the same poles as f − c, so we also get the same zeros.

Now notice that H 0 (O) = C: the number of zeros is equal to the number of poles, but global sections have no
poles, so there can’t be any zeros either. Let h1 (O) be the arithmetic genus pa of O: thus Riemann-Roch can be
rewritten as
χ(O(D)) = deg D + 1 − pa .

And thus notice that

lim dim H 0 (O(np)) = ∞,
n→∞

76
because deg(np) goes to infinity, meaning χ(O(D)) goes to infinity, and χ = h0 − h1 . In other words, there exists a
rational function f with a pole only at p.

Corollary 208
As a final example, we can show that any projective curve X is connected in the classical topology.

Proof. Suppose otherwise: say that p is an isolated point, and we can write X = X1 ∪ X2 for clopen X1 , X2 . Pick a
point p ∈ X1 , and choose a rational function f with a pole only at p (meaning there are no poles on X2 ). f is analytic
on X2 , which is a compact space, and a bounded analytic function on a compact manifold is some constant c (this is
the maximum principle).
But then f − c is zero on Xi , and it is a rational function on a curve, so it can only have finitely many zeros, which
is a contradiction.

29 April 27, 2020

Today, we’re going to apply the Riemann-Roch theorem to curves of genus 0, 1, and 2. We’ll start with the case
where we have a smooth curve Y with function field K.

Proposition 209
Any point (α0 , · · · , αn ) of Pn with values in K defines a morphism π : Y → Pn .

Proof. For the morphism to be defined on all of Y , we must be able to pick an index i for each q ∈ Y such that
αj αi
βj = αi are all regular at q (we don’t need to worry about being nonzero, since αi = 1). Then we define the morphism
via
π(q) = (β0 (q), · · · , βn (q)).

Because Y is a smooth curve, its local rings are valuation rings. Therefore, if we pick i so that
( ) the valuation at q, vq (αi ),
αj
is minimal (highest order pole), dividing must yield something that is regular (because vq αi = vq (αj ) − vq (αi )).

Proposition 210
Let Y be a projective curve, and let π : Y → X be a nonconstant morphism. Then π is finite (meaning that the
preimage Y ′ = π −1 X ′ of any affine open set X ′ = Spec A is affine, and writing it as Spec B makes B a finite
A-module).

Proof. The fibres of X are closed subsets of Y , but Y is a curve of dimension 1 and the whole curve doesn’t map to
a point. So our fibres are finite sets, and we can use Chevalley’s Finiteness Theorem.

Proposition 211
Let π : Y → X be a nonconstant finite morphism of curves, and let K and L be the function fields of X and Y ,
respectively. Then L is a finite field extension of K of some degree n, and the fibres of X have n elements for all
but a finite number of points in X.

77
Proof sketch. We can assume that X, Y are their affine open sets, X = Spec A, Y = Spec B, and B is a finite A-
module. Pick a primitive element of the field extension and look at the minimal polynomial of that element (which is
of degree n): then the discriminant is nonzero at all but finitely many points in X.

With that, we can move on to discussing Riemann-Roch.

• First of all, suppose Y has (arithmetic) genus 0, which means that the dimension pa = h1 OY is 0. This arithmetic
genus is also equal to the geometric genus g for projective smooth curves (which we’ll show later on).
Consider the exact sequence
0 → O → O(p) → ε → 0,

where ε is the one-dimensional module supported at p (isomorphic to the residue field module). Then the
cohomology sequence has dimensions as follows because h0 O is just the constant functions:

O O(p) ε
0
h 1 ? 1
h1 0 ? 0
2
h 0 0 0.

But we also have coboundary maps which help us figure out the rest of the table too:

O O(p) ε
0
h 1 2 1
1
h 0 1 0
2
h 0 0 0.

This means that H 0 O(p) contains 1 as well as some nonconstant function α: by looking at the divisors, α has
a pole of order 1 at p and no other poles. Consider the morphism
(1,α)
π : Y → P1 = X,

which evaluates at a point q to produce f (q) = (1, α(q)) when q 6= p and (α−1 (p), 1) = (0, 1) at q = p (where
we have a pole). Remember that Riemann-Roch shows that α has a simple pole and takes on every value exactly
once (because the divisor has degree 0, meaning that the number of zeros of f − c is the same as the number
of poles of f ), which means that the map π : Y → X is a bijective finite morphism. Since X and Y have equal
function fields, but both are normal, this means that all curves of genus 0 are isomorphic to P1 .
• In the case where we have genus 1, our table looks like the following:

O O(p) ε
h 0
1 ∗ 1
1
h 1 ? 0

There are two possibilities for ∗: either (h0 (O(p)), h1 (O(p))) = (1, 0) or (2, 1). But if h0 (O(p)) = 2, then there
is a nonconstant rational function α which means the curve is isomorphic to P1 , a contradiction. Thus we must
have the (1, 0) case.
Now we can consider the exact sequence

0 → O((k − 1)p) → O(kp) → ε → 0,

78
 where we’re working with a slightly different ε, and say that k ≥ 2. We can inductively compute the dimension:
putting in (k − 1, 0) in the first column for O((k − 1)p) means that we have a second column of (k, 0): in other
words, h0 O(kp) = k for all k ≥ 2. Then H 0 (O(2p)) contains 1, but it also contains a nonconstant function x
such that x has a pole of exactly order 2 at p (it can’t be order 1 by our earlier argument).
Next, we can consider h0 (O(3p)): this also contains 1 and x, and it also contains an element y with a pole of
order 3 (because it is in H 0 O(3p) but not H 0 O(2p))l, which is independent from x because we have different
order poles. We can consider the morphism
(1,x,y )
Y → P2

sending a point q to (1, x(q), y (q)) whenever q 6= p. Because y has a bigger order pole, this means that at
q = p, we send p to (y −1 , xy −1 , 1) = (0, 0, 1). This is a finite morphism, so the image is some curve X in P2
(its dimension has to be 1).
Now consider some line L : ax0 + bx1 + cx2 = 0 in P2 , where a, b, c are generic. Then the points of Y that map
to L are those q ∈ Y such that a + bx(q) + cy (q) = 0, a + bx + cy is a rational function on Y , and it has a pole
of order 3 at p, so it takes on every value 3 times. Thus, there are three points q with π(q) ∈ L ∩ X, meaning
that X is a cubic curve: let’s find its equation.
Notice that x 2 has a pole of order 4 at p, so it is in O(4p), and 1, x, y , x 2 are still independent. Thus, the
rational functions {1, x, y , x 2 } form a basis for H 0 O(4p), and similarly {1, x, y , x 2 , xy } for H 0 O(5p). Now y 2
and x 3 both have a pole of order 6 and we have a 6-dimensional space, so we get a linear dependence. This
gives us a plane cubic X, so every curve of genus 1 is isomorphic to an elliptic curve.

• Finally, we have the following table for curves of genus 2:

O O(p) ε
0
h 1 1 1
1
h 2 1 0

But now look at O → O(p) → O(2p) → ε:

O(p) O(2p) ε
h 0
1 ∗ 1
h 1
1 ∗ 0

The middle column either looks like (1, 0) or (2, 1), but it turns out that both are impossible.

We’ll move on soon to the second version of Riemann-Roch, which tells us more about what H 1 actually looks like:
that will allow us to say more about what’s going on in this last case!

30 April 29, 2020

As an interlude, we’ll discuss the Birkhoff-Grothendieck theorem today:

79
 Theorem 212 (Birkhoff-Grothendieck)
Let X = P1 , and let M be a locally free OX -module of rank r . Then
⊕
r
M∼
= O(ni ).
i=1

splits into a direct sum of (rank one) twisting modules.

This was proved by Grothendieck in this form, and it’s equivalent to a version with matrices proved by Birkhoff:
let A0 = C[u], A1 = C[v ], where u = x1
x0 , v = x0
x1 (these are the coordinate rings of the two standard affine open sets).
Then A01 ’s elements are the Laurent polynomials C[u, u −1 ]. Then we have the following:

Theorem 213
Suppose N is an invertible A01 -matrix. Then there exists an invertible A0 -matrix Q and an invertible A1 -matrix P
such that Q−1 NP is diagonal with diagonal entries that are integer powers of u.

When we look at a locally free module M, we’ll get a free module on U 0 and on U 1 . Then N basically tells us how
to glue those two together, and this idea above of changing basis tells us that we can do exactly that in a simple way.
We’re going to prove Grothendieck’s version of this here (we’ll verify the Birkhoff matrix version on our homework
for 2 × 2 matrices), but we’ll need to review some terminology first.

Definition 214
Let A be a domain and M be an A-module. An element m ∈ M is a torsion element if am = 0 for some a 6= 0 in
A (that is, the annihilator of m is nonzero). The set of torsion elements in M forms a module called the torsion
submodule. M is torsion-free if the submodule is zero, and it is a torsion module if all elements have torsion.

This works well with localization: if M is a torsion-free A-module and s 6= 0 is an element of A, then Ms is a
torsion-free As -module. Thus, the definition of “torsion-free” extends to O-modules as well.
We also want a few results about curves:

Lemma 215
Let X be a curve and M be an OX -module. If the torsion submodule is nonzero, then M has a global section:
H 0 M 6= 0.

Proof. Replace M by its torsion submodule, and assume it is finitely generated (take a finitely generated submodule).
This is okay, because we’re just trying to show that H 0 is not zero.
Each element in M is killed by something, so its annihilator is a nonzero ideal, meaning that the annihilator module
is nonzero. Thus, the support supp M, which is a closed subset, is a finite set S (this is where we use the fact that
X is a curve). So now we use the sheaf axiom: we can choose an affine open set U that contains S. Let V = X − S:
then X = U ∩ V is a union of two open sets, and M(V ) = 0, so M(X) = M(U). But U is affine, so M(U) is nonzero
as long as the module isn’t zero, and thus M(X) is nonzero.

Proposition 216
Let X be a smooth curve and M be a torsion-free finite O-module. Then M is locally free: that is, there exists
an affine covering {U i } for X such that M(U i ) is isomorphic to O(U i )r for some r .

80
Proof. Since this is a local statement, we can assume X = Spec A. Let p be a point in X: recall the local ring Ap is
the set of elements α ∈ K in the fraction field of A that are regular at p, and we can get Ap by adjoining all elements
a that don’t evaluate to zero at p.
Because X is a smooth curve, we know that this local ring is a valuation ring. If M is a torsion-free O-module,
then this corresponds to a finite A-module M (because O-modules and A-modules correspond). Then

Mp = M ⊗A Ap

is the module obtained by localizing M at p. Since Ap is a valuation ring, it is a principal ideal domain, so every torsion
free module is free: Mp is isomorphic to Apr by using a basis (m1 , · · · , mr ) of Mp . Since there are finitely many of
these, the mi are all in some localizations Ms for a nonzero s ∈ A, and then we get a map Ars → Ms , which we can
embed in the exact sequence
0 → K → Ars → Ms → C → 0.

But we can localize this sequence: when we localize at a point p, we have that Kp = Cp = 0 (because of the
isomorphism), so there exists some s ′ ∈ A such that Kss ′ = Css ′ = 0, and now we’ve shown the result we want
(isomorphism for a localization).

Definition 217
Let A be a Noetherian domain, and let M, N be A-modules. HomA (M, N) is an A-module consisting of the module
homomorphisms M → N: addition and scalar multiplication of these homomorphisms are defined in the obvious
way.

We’ll write this as A (M, N). There are some relevant functorial properties here: if we have a module homomorphism
N1 → N2 , we can map from A (M, N1 ) →A (M, N2 ). On the other hand, when we have a module homomorphism
ψ : M1 → M2 , we have a contravariant map A (M2 , N) →A (M1 , N).

Fact 218
If 0 → N1 → N2 → N3 is an exact sequence, then so is 0 →A (M1 , N1 ) →A (M2 , N2 ) →A (M3 , N3 ). Meanwhile,
if M1 → M2 → M3 → 0 is exact, then so is (again, contravariant) 0 →A (M3 , N) →A (M2 , N) →A (M1 , N).

Proposition 219 (Splitting)

If we have a set of maps N → M → X, then M is isomorphic to the direct sum M ∼
i s
= N ⊕ K, where K = ker s.

(We just need to check that N ∩ K is empty and that N + K = M.)

Corollary 220
Any functor F on O-modules carries direct sums to direct sums.

One final fact to know about Hom is the following:

Lemma 221
Let M, N be A-modules for some Noetherian domain A. Let s ∈ A be a nozero element: if M is a finite module,
then
As (Ms , Ns )
∼
= (A (M, N))s .

81
 We’ll finish the proof next time: we’ll want to refer to the dual module.

Definition 222
The dual module of an A-module M is
M ∗ =A (M, A).

This is compatible with localization: for instance, if M is an O-module and M∗ =O (M, O) (because Hom is
contravariant in the first variable), a map M1 → M2 will give us a map M∗2 → M∗1 . (And notice that we define the
dual module in this order because A (A, M) ∼
= M: the homomorphism is dependent on where 1 is sent.)

31 May 1, 2020
We’ll put the Birkhoff-Grothendieck theorem aside for now and focus on a new topic: the second version of Riemann-
Roch.
We’re going to need some new terminology, and our first topic will be that of branched covers of curves. Suppose
that π : Y → X is a nonconstant morphism of smooth projective curves: recall that Chevalley’s Finiteness Theorem
tells us that π is a finite morphism, so if U = Spec A is an affine open subset of X, then its preimage π −1 U is also
affine: writing it as Spec B, we’ll have an injective homomorphism A → B, and B will be a torsion-free finite A-module,
and therefore it will be locally free. This means that it makes sense to take the direct image π∗ OY , and this will be a
finite OX -module (because the modules on the affine open sets are all finite) that is locally free, say of rank n. We’ll
just make this our definition:

Definition 223
A morphism π : Y → X of smooth projective curves is called a branched cover.

One important concept here is that of ramification: let q ∈ Y have image p = π(q). We have a local ring at each
point of our smooth curve, so we can let x be the generator for our local ring of X at p (that is, it is a generator in
some affine open set). Because p is the image of q, and the local rings are valuation rings, we can also evaluate x at
q:
vq (x) = e

for some e ≥ 0, which we call the ramification index at q. Remember that our fibres are finite, so each point p ∈ X
has some points q1 , · · · , qk in the fibre over p.

Proposition 224
∑k
Let ci be the ramification index at qi . Then i=1 ei = n is the rank of the OX -module π∗ OY , and the extended
ideal can be written as
mp OY = meq11 · · · meqkk .

In the case where we have affines U = Spec A and π −1 U = Spec B, then mp is an ideal of A, and we have a map
A → B such that
mp B = meq11 · · · meqkk .

We’ll move on to another preliminary topic:

82
 Definition 225
Let Y be a smooth curve. An OY -module L is invertible if it is locally free of rank 1.

The reason for the name here is that

O(D) ⊗O O(E) ∼
= O(D + E) :

functions with poles D times functions with poles E are the functions with poles D + E. This means that

O(D) ⊗O O(−D) ∼
= O,

and since O can be thought of as a “unit” when doing an O-tensor product, O(D) and O(−D) are inverses.

Proposition 226
Every invertible module is isomorphic to OY (D) for some divisor D.

D isn’t necessarily unique here, so it’s interesting to ask when O(D) and O(E) are isomorphic. Remember that
the sections of this module on an open set U are the rational functions α on Y such that

div(α) + D

(where the divisor div counts the zeros minus the poles) is an effective divisor (≥ 0) on U. (And we also need to include
the element 0.) So suppose we have an isomorphism ϕ : O(D) → O(E): if we look on an open set not containing any
points in D, then we just want the sections of O, the structure sheaf.
The idea is to note that O(D) and O(E) are both subsets of the function field of Y , which we denote K. “Lifting”
the isomorphism to K → K, we know that any map from K to itself is a multiplication by a rational function f . Thus,
we need to find a function f such that

div(α) + D ≥ 0 ⇐⇒ div(f α) + E ≥ 0,

but because the divisor is multiplicative, this means div(f α) = div(f ) + div(α), and then we can equivalently say that

D ≥ 0 ⇐⇒ div(f ) + E ≥ 0,

meaning that div(f ) = D − E .

Definition 227
Two divisor D, E are linearly equivalent if D − E = div(f ) for some rational function f .

Corollary 228
O(D) ∼= O(E) if and only if D and E are linearly equivalent.

A final preliminary concept we’re going to need is that of a differential:

83
 Definition 229
Let A be an algebra (scalars C), and let M be an A-module. A derivation is a map δ : A → M such that
δ(a + b) = δ(a) + δ(b), δ(ab) = aδ(b) + bδ(a), and δc = 0 for all scalars c ∈ C.

(These are like the calculus rules for differentiation.)

Definition 230
A module of differentials ΩA is an A-module generated by elements da, for each a ∈ A, with the relations
d(a + b) = da + db, d(ab) = a db + b da, dc = 0 for c ∈ C.

These relations cut down the module to a finite A-module. Since the relations look the same in the two above
definitions, we can make the following observation:

Corollary 231
Derivations δ : A → M correspond bijectively to module homomorphism ϕ : ΩA → M, such that

δ =ϕ◦d

(where d maps A to ΩA).

Suppose, for example, that R = C[x1 , · · · , xn ]. Then ΩR is a free R-module with basis dx1 , · · · , dxn , because any
polynomial f has differential
∑ ∂f
df = dxi ,
∂xi
i

and we can inductively continue this reasoning.

Note that Ω is compatible with localization: if s 6= 0 is an element of a domain A, then ΩAs ∼
= (ΩA )s . That means
we can define the OX -module ΩX by letting its sections on an affine open set U = Spec A be ΩA .
One other fact: suppose that I is an ideal of a ring A, and A = A/I. Consider the submodule

N = dI + IΩA ,

which tells us that dα = 0 for any α ∈ I, as well as that αγ = 0 for any α ∈ I, γ ∈ ΩA . Then we have that
Ω ∼
A = ΩA /N.

Proposition 232
Let Y be a smooth curve. Then ΩY is an invertible module.

Therefore, we can write ΩY ∼

= O(K) for some divisor K, which we call the canonical divisor (up to linear
equivalence).
With all of this, we’re finally able to state our result:

Theorem 233 (Riemann-Roch, version 2)

Let Y be a smooth projective curve, let D be a divisor, and let K be a canonical divisor. Then

h0 O(D) = h1 O(K − D), h1 (O(D)) = h0 O(K − D).

84
 (These two equations are equivalent because we can plug in K − D for D in the first equation.) It’s not very clear
why differentials come up, and the proof won’t make that very clear either.
A corollary of this theorem will be the following result:

Theorem 234
For any smooth projective curve Y , we have pa = g (where pa = h1 OY and g is the topological genus). In addition,
the degree of the canonical divisor is 2g − 2 = 2pa − 2.

32 May 4, 2020
We’ll start with a bit of review: recall that for a variety X, we can construct an O-module of differentials ΩX . If we
write X = Spec A, then ΩA is generated by elements {da : a ∈ A}, such that d(a + b) = da + db, d(ab) = adb + bda,
and dc = 0 for a, b ∈ A and c ∈ C. When Y is a msooth curve, ΩY is locally free of rank 1 (it is invertible), so
we know by Birkhoff-Grothendieck that ΩY ∼
= O(K) for some canonical divisor K. (Note that we can have different
canonical divisors if they are linearly equivalent – that is, their difference is the divisor of a rational function.)
Then the second version of Riemann-Roch tells us that if Y is a smooth projective curve and D is a divisor, then

h0 (O(D)) = h1 (O(K − D))

(and therefore the same result holds when we swap the roles of D and K − D, since we can plug in D → K − D).
Note that when we apply this to the divisor D = 0, h0 O(K) = h1 O is the genus g of the variety, and h1 O(K) =
h0 O = 1. Thus, the Euler characteristic h0 − h1 here is χ(O(K)) = g − 1.

Corollary 235
The degree of the canonical divisor D is 2g − 2.

Proof. We know that

χ(O(K)) = deg K + (1 − g)

by Riemann-Roch version 1, and then the left hand side is g − 1 by the argument above.

Example 236
Let’s consider a curve of genus g = 2.

Then the degree of the canonical divisor, deg K, is 2g − 2 = 2, and the Euler characteristic χO(K) = 1. This
means that h0 O(K) > 0, so there exists some nonzero global section, which means that there exists a rational function
α such that
div(α) + K ≥ 0.

But now the left hand side is divK ′ for some canonical divisor K ′ (K and K ′ are linearly inequivalent), so we can
assume that K ′ is effective and of the form p1 + p2 , where it’s possible that p1 = p2 .
Riemnann-Roch now tells us that
h0 (O(D)) = h1 (O) = g = 2,

85
so H 0 has some basis (1, α) (we can always choose 1 to be a global section), such that α has poles at p1 and p2 . This
means α has order 2, so takes on every value 2 times (with multiplicity).
But now (1, α) is a point of P1 with values in the function field F of Y , and any such point defines a morphism
Y → P1 via π(q) = (1, α(q)) whenever q 6= p1 , p2 and ( α(q)
1
, 1) = (0, 1) when q = p1 , p2 . We already mentioned that
π is a double covering of P1 above.

Definition 237
A smooth projective curve Y is hyperelliptic if it can be represented as a double cover of P1 .

Every curve of genus 1 can be represented as such a covering, so this may explain the “elliptic” part of the definition.
One way to create such a curve is to consider y 2 = f (x0 , x1 ) in the weighted projective plane.

Corollary 238
Every smooth projective curve Y of genus 2 is hyperelliptic.

Example 239
Now let’s consider g = 3.

We can again assume that K is effective, so it is the sum of four points, and the global sections form a space of
dimension h0 O(K) = g = 3: suppose we have a basis (1, α, β). Then α, β must have poles in the set {p1 , p2 , p3 , p4 }.
We claim that all points must be represented. Suppose otherwise, so that both α and β are sections of O(p1 +
p2 + p3 ) = O(K − p), where p = p4 . Then

h0 O(p1 + p2 + p3 ) = h0 O(K − p) = h1 O(p)

by Riemann-Roh version 2. But now we can estimate the cohomology of Op via the exact sequence

0 → O → O(p) → ε → 0.

We know that O(p) does not have a nonconstant global section, so the table of cohomology dimensions looks like

O O(p) ε
0
h 1 1 1
1
h 3 2 0

But then h1 O(p) is not 3, which is a contradiction. So we can assume that all points are represented, and now
for any point q 6= p1 , p2 , p3 , p4 , we can construct the point with values in K (1, α, β). The only points where α and β
may not be regular are the pi s, so there are only the four points pi that are on the line X0 = 0. So the image of Y is
a plane curve, and we have a finite morphism Y toX such that the degree of X is the number of points in L ∩ X.
Since 4 points map to L, this means we either have X of degree 1 and a covering Y of degree 4, X of degree 2 and
a covering of degree 2, or X of degree 4 and a generically injective covering. But the first case can’t happen because
(1, α, β) are linearly independent.
The other two cases can be be described: if X is degree 2, then X is a conic isomorphic to P1 , and Y is a double
covering and therefore hyperelliptic. And finally, if X is degree 4, then Y is the normalization of X, since the genus
( )
of X has dimension 4−12 = 3.

86
 Corollary 240
A smooth projective curve of genus 3 is either hyperelliptic or isomorphic to a plane curve of degree 4.

We’ll finish by verifying Riemann-Roch in the case where Y = P1 :

Proof. If we apply Birkhoff-Grothendieck to invertible modules, we see that every invertible O-module is isomorphic
to O(n) for some n, which we wish to identify.
Identify ΩX with O(K), where X = U 0 ∪ U 1 . Identify U 0 = Spec C[u] and U 1 = Spec C[v ]. We know that ΩU 0
is free with basis du, and we know that du = dv −1 = −v −2 dv , so du has a pole of order 2 at the point at infinity,
meaning du is a global section of ΩX (2p∞ ). But this section is zero nowhere, so ΩX (2p∞ ) is free and isomorphic to
OX , meaning that
OX (−2p∞ ) ∼
= O(−2).

So now we can verify Riemann-Roch by noting that O(D) is isomorphic to O(n) and that O(K − D) is isomorphic to
O(−n − 2). We’re supposed to have h0 O(n) = h1 O(− − 2) and h1 O(n) = h0 O((n − 2).
When n ≥ 0: then h0 O(n) = n + 1 and h1 O(n) = 0 (because we know the cohomology of the twisting modules).
Meanwhile, when r > 0, h0 O(−r ) = 0 and h1 O(−r ) = −r − 1. Plugging in r = n + 2, we find that

h1 O(−n − 2) = n + 1, h0 O(−n − 2) = 0,

and indeed things work out.

33 May 6, 2020
Today, we’ll want to generalize Riemann-Roch from P1 to curves in general. Proving it for P1 is just the computation
that we did in class, and in general we will want to make Y into a branched covering of X = P1 . Then π : Y → X is
a finite morphism of some degree n, meaning that

H q (Y, OY (D)) = H q (X, π∗ OY (D)),

so we can get an analogous statement for OY (D) by taking its direct image and applying Riemann-Roch. But there
are two main problems with this: first of all, π∗ OY (D) is not invertible (unlike OY (D)), though it is locally free of rank
n. This isn’t too much of a problem, since we can use the Birkhoff-Grothendieck theorem to write the locally free
module as a direct sum of O(ri )s and then apply Riemann-Roch there. But we have a bigger issue in that we need to
relate π∗ OY (K − D) to π∗ OY (D), so we will need to rewrite some things here.
We’ll write OY (K) = ΩY as the module of differentials, and OY (D) = L as an invertible OY -module. Then the
dual module
L∗ =OY (L, OY ) ∼
= OY (−D).

87
 Lemma 241
Let M and N be OY -modules, and let D be a divisor. Let OY (M, N ) be the O-module with sections equal to
the homomorphisms M(U) → N (U). Then

OY (M, N ) =OY (M(D), N (D))

(where M(D) = M ⊗O O(D)), and O(K − D) is isomorphic to OY (L, ΩY ).

Proof. For the first part, we are given homomorphisms M → N , tensoring with O(D) gives us a homomorphism
M(D) → N (D). This is an invertible operation because we can tensor with O(−D) to get back to where we started.
For the second part, remember that O (O, M) ∼= M (because sections m of M correspond to the homomorphisms
that are multiplication by m).
OY (L, ΩY ) =OY (OY (D), OY (K)),

and tensoring both sides by O(−D) yields

∼
= OY (OY , OY (K − D)).

Definition 242
Let M be a locally free OY -module. Then the Serre dual of M, denoted M# , is the set of maps from M to
the differentials:
M# =OY (M, ΩY ).

For example,
OY (D)# =OY (O(D), O(K)) ∼
=OY (O, O(K − D)) ∼
= OY (K − D).

This tells us how to state Riemann-Roch version 2 in another way:

Theorem 243
Let M be a locally free OY -module. Then

h0 M = h1 M# , h1 M = h0 M# .

In order to apply the P1 -version of Riemann-Roch 2, we need to show that

π∗ (M# ) ∼
= (π∗ M)# ,

where M# is the Serre dual on Y , and the right side is the Serre dual on X. We’ll drop the π∗ symbol from here:
we’ll denote the direct image of an OY -module M on X by M as well. In other words, if U = Spec A is an affine open
in X, and V = π −1 U = Spec B is affine, then the B-module M = M(V ) can be viewed as an A-module because we
have restriction of scalars through the map A → B, and we’ll say that the sections of the direct image on U are also
denoted by M.
To make progress, we’ll need to talk about the trace of a differential. If we have a map OX → OY (which is the
direct image), then the trace is the map OY → OX such that the composition is multiplication by the degree of the
covering. We’ll let the trace of β be denoted as tr(β).

88
 To understand the trace for differentials, we’ll do a bit of computation: let p ∈ X be a point, and let x be a local
generator for the maximal ideal mp . Then we pick q ∈ Y such that π(q) = p, and let y be the local generator for mq ,
we can let the ramification index be e (meaning that x = uy e , and u is a local unit). Then

dx = y e du + ey e−1 udy ,

but dy is a local generator for the module of differentials ΩY , and u is a regular function on Y at q. Therefore, du
can also be written as w dy for some regular function w at q. Therefore,

dx = (w y e + ey e−1 u)dy = v y e−1 dy ,

where v = w y + eu is a local unit (invertible function locally at q), meaning that

1
dy = dx
v y e−1

(so we have a pole of order e − 1 when we write things in terms of dy ).

So now we can look on a fibre q1 , · · · , qk of p ∈ X. If β is a differential on Y that is regular at all qi , then we can
write
β = bdx

(using the above local computation), where b has poles of order at most ei − 1 at each point pi .

Definition 244
The trace is a map τ of modules of differentials ΩY → ΩX , such that

τ (β) = (tr b)dx.

Proposition 245 (Main Lemma)

If β is a differential on Y which is regular at q1 , · · · , qk (the fibre over p), then τ (β) is a regular differential on X
at p.

Proof. We know that β = bdx, where b has poles of order at most ei − 1 at qi . But we know that x has zeros of
order ei at qi , so xb is regular at all qi and evaluates to zero. Then

τ (xβ) = tr(xb)dx,

and the trace tr(xb) is regular at p: in fact, we have a general formula. If f is a regular function on Y , then
∑
[tr(f )](p) = ei f (qi ).
i

1
So if xb is zero at qi , then tr(xb) = 0 at p, meaning that x tr(xb) is regular at p as well. And because the trace is
OX -linear, this last expression is equal to the trace of b, meaning

τ (β) = tr(b)dx

is indeed regular at p, as desired.

89
 This trace defines a map ΩY → ΩX : note that

ΩY ∼
=OY (OY , ΩY ),

where the correspondence sends β to the map b → bβ. Then we can compose this map ΩY with τ to get to

OX (OY , ΩX )

(because the first map is OY -linear, but the second map is OX -linear).

Lemma 246
The map
τ◦
π∗ (OY (OY , ΩY ) → OX (π∗ OY , ΩX )

is bijective.

So differentials on Y are the same as maps from OY to ΩX , which is a strange fact. As an exercise, we can look
at the case where Y is defined by y 3 = x and X is the affine line. Unfortunately, the proof is a bit long, and we can
read it on our own.
So now if M is a locally-free OY -module, then
τ◦
OY (M, ΩY ) →OX (M, ΩX ),

because we can look at an open set and treat M as a direct sum of OY s, and then we can apply the above lemma.
So now the left side is the Serre dual M# computed on Y , while the right side is M# computed on X, and we’ve
shown Riemann-Roch version 2.

34 May 8, 2020
In these last two classes, we’ll focus on applications of the Riemann-Roch theorem.

Proposition 247
The arithmetic genus pa and topological genus g of a smooth curve are the same.

Proof. Recall that Riemann-Roch version 2 tells us that

h0 O(D) = h1 O(K − D), h1 O(D) = h0 O(K − D).

Applying this with D = 0 yields

1 = h0 O = h1 O(K)

and
pa = h1 O = h0 O(K).

But version 1 also tells us that

pa − 1 = χ(O(K)) = deg K + χ(O) = deg K + 1 − pa ,

so the degree of the canonical divisor must be 2pa − 2.

90
 We’ll also compute this in another way. Suppose we have a branched covering π : Y → X = P1 , and Y has branch
points qi with ramifications ei . Then looking at the differential dx on Y , if we let yi be a local generator for mqi , then

x = uyiei ,

meaning that dx has a zero of order ei − 1 at the point qi . We’ll asume that we’re looking at the points over ∞ on
P1 (this is just a change of coordinates): since dx has a pole of order 2 at the point ∞ on X, it must have a pole of
order 2 at all of the point p1 , . . . , pn in the fibre over ∞.
So now let K be the divisor of dx on Y : this will be the sum of the zeros minus the sum of the poles, which takes
the form
∑ ∑
n
(ei − 1)qi − 2 pi .
i i=1
∑
Thus, the degree of K is (ei − 1) − 2n, which is also 2pa − 2 by our above calculation. But we are trying to work
with the topological Euler characteristic of Y as well: we know that

e(Y ) = ne(X) − (number of times the different sheets come together),

and this can be written in terms of the ramification indices: since ei sheets come together at a point qi , we lose
(ei − 1). Thus,
∑ ∑
e(Y ) = n · 2 − (ei − 1) = 2n − (ei − 1) = 2 − 2pa
i i

by consulting the calculation above. But in a manifold of degree 2, the Euler characteristic is also 2 − 2g, so we must
have g = pa , as desired (and this tells us that deg K = 2g − 2).

We’ll now move on to a different topic:

Definition 248
Let D be a divisor, and let h0 O(D) > 0 (so we have some global sections). A point p ∈ Y is a base point of
O(D) if
h0 O(D − p) = h) O(D).

In other words, whenever we have a function f such that div(f ) + D ≥ 0, we also have div(f ) + D − p ≥ 0. We
can understand this using the usual exact sequence

0 → O(D − p) → O(D) → ε → 0,

which we use when proving Riemann-Roch version 1. When p is a base point, the h0 dimensions are the same, which
means that the h1 values for the two modules differ by 1. Otherwise, h0 will differ by 1.
In the case where D is effective, we can rephrase by saying that whenever p is not in the support supp D, if a
function f has poles at most D, then f has poles at most D − p.

Example 249
The module O(K) has no base points.

We can see this by computing h0 O(K − p): by Riemann-Roch,

h0 O(K − p) = h1 O(p), h1 O(K − p) = h0 O(p).

91
If Y is not the projective line, then h0 O(p) must be 1 – there is no function with just one simple pole. So

h0 O(p) = h0 O =⇒ h0 O(K − p) = h0 O(K) − 1,

which means that our arbitrary point p is not a base point.

Example 250
On the other hand, p is always a base point of O(K + p): every function with poles K + p doesn’t actually have
the extra pole at p.

This is because
h1 O(K + p) = h0 O(−p) = 0 = h0 O − 1,

and if the h0 dimensions differ, the h1 dimensions must be the same.

So now suppose that D > 0, and we use a basis of H 0 O(D) (of regular functions) to apply the morphism

π : Y → Pr

via (α0 , · · · , αr ), where h) O(D) = r + 1. Then remember that the degree of our morphism π is the number of points
in π −1 H, where H is some generic hyperplane of Pr .

Lemma 251
If D is effective and O(D) has no base points, then deg π = deg D.

Proof. Since H is generic, we’ll represent it in the form

∑
ci xi = 0.

Then the preimage of H is the set of points

∑
{q ∈ Y : ci αi (q) = 0}.

Let β be the regular function

∑
β= ci αi :

since the αi s form a basis, this is also a global section in H 0 O(D). The set of zeros of β is exactly the set π −1 H that
we want, but β will have all poles in the divisor D, since there are no base points in O(D). Thus, the number of poles
is deg D, and this is equal to the number of zeros of β as well.

In the next topic that we cover, we’ll let g ≥ 2. Then we can use a basis of H 0 O(K) (of size g) to map
π : Y → Pg−1 : this is known as the canonical map.

Theorem 252
The canonical map π is either an embedding of Y into Pg−1 , or X, the image of Y , is isomorphic to P1 and Y → X
is a double cover (meaning that Y is hyperelliptic).

Proof. Suppose the map π : Y → X is not injective, so we have two points q1 , q2 such that π(q1 ) = π(q2 ). Consider
O(K), O(K − q1 ), and O(K − q1 − q2 ). We know that K is only chosen up to linear equivalence, so we can assume
K is effective, and we can assume q1 , q2 are not in the support of K.

92
 Then O(K) has no base points, meaning

h0 O(K − q1 ) = h0 O(K) − 1.

In other words, sppose our basis for H 0 O(K) is (α0 , · · · , αg−1 ). Since this defines a morphism, these functions don’t
all vanish at q1 , but the dimension of functions that do vanish is only 1 less than the total dimension. So let’s assume
that α0 (q1 ) does not vanish, but all the other αi s do: then π(q1 ) = (1, 0, · · · , 0) = π(q2 ) (by the initial choice of
q1 , q2 ), which means α1 , · · · , αg−1 alsovanish at q2 (and α0 (q2 ) 6= 0 as well). So now

h0 O(K − q1 − q2 ) = h0 O(K − q1 ),

and also
h0 O(K − q1 − q2 ) = h0 O(K − q2 ) = g − 1,

because q2 is a base point of h0 O(K − q1 ). So now Riemann-Roch version 2 restates this as

h1 O(q1 + q2 ) = h1 O(q1 ) = g − 1,

which means that

χ(O(q1 + q2 )) = 2 + (1 − g) = 3 − g.

But h1 (O(q1 + q2 )) = g − 1, so h0 = 2, which means that there is a two-dimensional space of functions with the
poles q1 , q2 . Therefore, the map Y → P1 where we use a two-dimensional basis of h0 (O(q1 + q2 )) has degree 2, which
means Y is hyperelliptic.
Throughout all of this, we’ve assumed Y → X is not injective, so the other case is that Y → X is bijective. It’s
still possible that X is a cusp curve, but we’ll skip over that – the way to deal with this case is to consider 2q instead
of q1 + q2 and look at this reasoning.

Theorem 253
Suppose Y is a curve of genus g ≥ 2. Then there is at most one way to represent Y as a double cover of P1 .

We know that if Y is hyperelliptic, there is a way to do this, but this theorem tells us that the representation
is unique. We don’t have time to prove this, but a sketch of the main ideas is that when we have a double cover
π : Y → X = P1 , we can count the number of branch points by noting that

2 − 2g = 2e(X) − (branch points) = 4 − B,

meaning that B = 2g + 2. So if we look at the divisor of dx on Y , we have zeros of order 2 − 1 = 1 at each of the
branch points q1 , · · · , qB , and we have poles of order 2 at p1 , p2 (the points in the fibre over ∞). Let K be the divisor
of dx on Y , which can be written as
∑
B
qi − 2(p1 + p2 ).
i=1

35 May 11, 2020

We’ll discuss the canonical embedding of a curve today. Say that Y is a smooth projective curve: remember that Y
is hyperelliptic if there exists a degree 2 map Y → X = P1 , meaning that Y is a double cover of X. As we discussed

93
last time, there is at most one way to represent Y as a double cover of X when g ≥ 2.
It’s easy to construct hyperelliptic curves of any genus g: start with a homogeneous polynomial of degree 2d in
x0 , x1 , and form the double line
y 2 = f (x),

where y is a weight d variable. Then the map Y → X ramifies at the zeros of f , meaning that the Euler characteristic

e(Y ) = 2e(X) − 2d = 4 − 2d.

(because each of the 2d zeros is a branch point) But C(Y ) = 2 − 2g, so g = d − 1

Another relevant fact is that whenever Y has genus ≥ 2 and is not hyperelliptic, O(K) has no base points, so its
global sections define a map π : Y → Pg−1 . This map is then an embedding of Y as a closed subvariety of projective
space of degree g, and the degree of this embedding is deg K = 2g − 2.
When the genus is 2, Y is always hyperelliptic, because O(K) will have degree 2, meaning the map Y → P1 has
degree k = 2g − 2 = 2, which is a double cover. We will proceed by describing this embedding in a few cases:
g = 3, 4, 5.

Example 254
When g = 3, we are embedding Y into P2 , so Y is a plane curve of degree deg K = 2g − 2 = 4.

Indeed, this works out because the arithmetic and geometric genus are equal to
( )
4−1
g = pa = = 3.
2

But there’s another approach we can take here: take R = C[x0 , · · · , xn ] and impose a grading
⊕
R= Rd ,
(d+n) (d+2)
where Rd consists of the homogeneous polynomials of degree d. The dimension of Rd is n , so it is 2 in the
case where we have three variables.
So now let α = (α0 , α1 , α2 ) be a basis for the global sections H 0 O(K). Let A1 be the span of these elements, let
A2 be the span of the degree-2 terms in αi s, and so on: this means we have a grading on A as well of the form
⊕
A= Ad .

Now A1 = H 0 O(K), and A2 ⊂ H 0 O(2K) and so on (because adding K gives an effective divisor for each αi , so adding
2K will do so if we have αi αj ). Now Riemann-Roch tells us that whenever n ≥ 2,

h0 O(nK) = χO(nK) = deg(nK) + 1 − g = 4n + 1 − g,

(d+2)
because h1 O(nK) = 0 by Riemann-Roch version 2. The dimensions of Rd are 2 = 1, 3, 6, 10, 15, 21, · · · for
d = 0, 1, 2, 3, 4, 5, and the dimensions of h0 O(dK) are 1, 3, 6, 10, 14, 18 (by plugging in the above formula). Thus,
there exists some nonzero homogeneous polynomial f (x) of degree 4 such that f (α) = 0, and there exist three
homogeneous polynomials h1 , h2 , h3 of degree 5, but that’s not new information because they’re just x0 f , x1 f , x2 f .
But now Y is contained in the zero locus of this degree 4 polynomial f , so indeed we’ve seen again that every
non-hyperelliptic curve of genus 3 is a plane curve of degree 4.

94
 Example 255
When g = 4, we are embedding Y into P3 , and the degree of the embedding is 4 · 2 − 2 = 6.

(d+3)
We’ll count dimensions again: this time dim Rd = 3 , and dim Ad = deg(dK) + 1 − g = 6d + 1 − g for all d ≥ 2.
Then dim Rd = 1, 4, 10, 20 and dim Ad = 1, 4, 9, 15 for d = 0, 1, 2, 3, which means that there exists a homogeneous
quadratic f with f (α) = 0. Then there are five cubics including x0 f , x1 f , x2 f , x3 f : call the last one g. Then Y is
contained within the locus Z = {f = g = 0}, and now a version of Bezout’s theorem (in higher dimensions) says that
the degree of Z is 2 · 3 = 6. So every non-hyperelliptic curve of genus 4 is a complete intersection (it’s defined via
f = g = 0 in P3 ).

Example 256
When g = 5, we embed Y into P4 with deg K = 2g − 2 = 8.

(d+4)
Then dim Rd = 4 is 1, 5, 15 for d = 0, 1, 2, while h0 O(2K) = deg(2K) + (1 − g) = 2(2g − 2) + (1 − g) = 12.
Thus, there are linearly independent three quadrics Q1 , Q2 , Q3 that all contain Y .
Now Bezout’s theorem tells us that the degree of Q1 ∩ Q2 ∩ Q3 is 2 · 2 · 2 = 8, which will also be the degree of Y
if the dimension of the intersection Q1 ∩ Q2 ∩ Q3 is 1. So then Y will be the intersection of 3 quadrics.
There’s also another case above where the dimension of Q1 ∩ Q2 ∩ Q3 is not of dimension but of dimension 2, and
this case does happen. If we can represent Y as a triple cover of P1 , then Q1 ∩ Q2 ∩ Q3 will fall into his case, and Y
is called trigonal.

Remark 257. And similarly, for g = 6, deg Y = 10 doesn’t factor enough for Y to be a complete intersection.

95
 Math 137 - Problem Set 10
Due Friday, Apr 17
All rings are commutative, and k is an algebraically closed field.

1. (a) Let P1 , . . . , Pr ∈ P2 and d ≥ 1. Show that there is a plane curve of degree d that
doesn’t contain any of the Pi .
(b) Does (a) generalize to hypersurfaces of degree d in Pn ?

2. (a) Let Y be a set of 5 distinct points in P2 . Let V be the linear system of conics
that contain Y . Show that dim(V ) > 0 if and only if at least four of the points
are collinear.1
(b) Let Z be a set of 10 distinct points in P2 . Let W be the (possibly empty!) linear
system of cubics that contain Z. Show that dim(W ) > 0 if and only if at least 6
of the points are collinear or at least 9 of the points lie on a conic.

3. Let F be an irreducible plane curve of degree d. Assume the partial derivative Fx 6= 0.

(a) If P is a point on F , show that mP (Fx ) ≥ mP (F ) − 1.

(b) Using part (a) along with Bézout’s theorem, show that
X
mP (mP − 1) ≤ d(d − 1).
P ∈V (F )

(c) Conclude that F has at most 21 d(d − 1) multiple points.

(d) Give an example to show that the bound in (c) is not sharp. That is, show that
there is some d such that F cannot have 12 d(d − 1) multiple points.

4. Let n ≥ 1 and d ≥ 1. Define a map vd : Pn → PN by

vd ([x0 : x1 : . . . : xn ]) = [M0 : M1 : . . . : MN ],

where the Mi are all the degree d monomials in n + 1 variables. That is,

M0 = xd0 , M1 = x0d−1 x1 , . . . , MN = xdn .

This map is called the degree d Veronese embedding of Pn .

(a) What is N ? Your answer should be in terms of d and n.

(b) Show that vd is well-defined and injective.2
(c) Denote V := vd (Pn ), the image of P(n). You may assume V is Zariski closed.
Let H ⊂ PN be a hyperplane. Show that H ∩ V = vd (W ), where W ⊂ Pn is a
hypersurface of degree d.3
1
Hint: Remember that dim(V ) = 0 if and only if there is exactly one conic through those points.
2
In fact, it’s an isomorphism onto its image. You don’t need to prove this.
3
H ∩ V is called a hyperplane section of V .
 (d) Show that the converse of (c) holds. Specifically, let W ⊂ Pn be a hypersurface of
degree d. Show that there is some hyperplane H ⊂ PN such that H ∩ V = vd (W ).
That is, the hyperplane sections of V are exactly the degree d hypersurfaces in
Pn .
(e) Notice that the twisted cubic in P3 is the degree 3 Veronese embedding of P1 .4
Apply the last sentence in (d) to the case of the twisted cubic. What does it say,
concretely, in this case?

4
The Veronese embeddings of P1 are called rational normal curves.
 Math 137 - Problem Set 11
Due Wednesday, Apr 29
All rings are commutative, and k is an algebraically closed field.

1. Show that V = An − {0} is an affine variety if and only if n = 1. 1

2. Let C be the affine plane curve V (x3 + x2 − y 2 ) ⊂ A2 .

(a) Show that C is rational by explicitly describing a birational map f : A1 99K C

and checking that it induces an isomorphism of fields f ∗ : k(C) → k(A1 ) = k(t)
(b) Which DVR(s) in k(t) dominate the local ring f ∗ (O(0,0) (C))?

3. Let X and Y be varieties, and let f : X → Y be a function. Suppose X = ∪α Uα and

Y = ∪α Vα , where Uα ⊂ X and Vα ⊂ Y are open, and f (Uα ) ⊂ Vα for each α.

(a) Show that f is a morphism if and only if each restriction fα : Uα → Vα of f is a

morphism. 2
(b) If each Uα and Vα is an affine variety, show that f is a morphism if and only if
each f ∗ (Γ(Vα )) ⊂ Γ(Uα ).

4. Define a morphism f : P1 ×P1 → P3 by f ([x0 : x1 ], [y0 : y1 ]) = [x0 y0 : x0 y1 : x1 y0 : x1 y1 ].

This map is called the Segre embedding of P1 × P1 into P3 .

(a) Check that this map is well-defined and injective.

(b) (Extra credit) Show that f is a morphism.
(c) Let the homogeneous coordinates of P3 be t00 , t01 , t10 , and t11 . Define the ideal

I = (t01 t10 − t00 t11 ) ⊂ k[t00 , t01 , t10 , t11 ].

Show that the image of f is V (I).

(d) (Extra credit) Generalize the above to Pn × Pm . What is the ideal of the image
of the map in this case?

1
Hint: what is Γ(V )?
2
Your answer here should not be super technical and complicated. The crux of this problem is just parsing
definitions.
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #11 Due: 24 Nov 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let X = Pn and p = [1 : 0 : · · · : 0], and let M be the skyscraper sheaf supported

at p, i.e., (
C if p ∈ U,
M(U ) =
0 otherwise.

(a) Compute the Čech complex Č • (U, M) with respect to the standard affine open
cover U.
(b) Compute the Čech cohomology groups Ȟ q (U, M) for all q ≥ 0.

2. Let d ≥ 0 and consider the twisting module O(d) on X := P1 .

(a) Compute the Čech complex Č • (U, O(d)) with respect to the standard affine
open cover U.
(b) Compute the Čech cohomology groups Ȟ q (U, O(d)) for all q ≥ 0. Verify that
your answer agrees with the results on the cohomology groups of twisting
modules proved in class via the long exact sequence.

3. Let X be a variety. Recall that an invertible sheaf L is a locally free OX -module

of rank 1, so there exists an affine open cover U = {Ui }i of X and isomorphisms
∼
ϕi : L|Ui → OX |Ui for each i. Then
∼
ϕi (Uij ) ◦ ϕj (Uij )−1 : OX (Uij ) → OX (Uij )

is an isomorphism of OX (Uij )-modules for each i, j, where Uij = Ui ∩ Uj . Thus

×
ϕi (Uij )◦ϕj (Uij )−1 is given by multiplication by an invertible function gij ∈ OX (Uij ),
namely
gij = (ϕi (Uij ) ◦ ϕj (Uij )−1 )(1) ,
×
where OX is the sheaf of invertible regular functions.
× ×
(a) Show that g := (gij )ij ∈ Ž 1 (U, OX ) is a Čech 1-cocycle. (Note that OX (V ) is
a group under multiplication for each open V .)
 (b) Suppose we have another local trivialization of L over U, i.e., another family
∼
{ψi }i of isomorphisms ψi : L|Ui → OX |Ui . Let h = (hij )ij be the correspond-
ing 1-cocycle, as constructed in the previous part. Show that g and h differ
×
(multiplicatively) by a 1-coboundary, i.e., an element of B̌ 1 (U, OX ). (With a
∼ 1 ×
little more work, one can show that Pic(X) = Ȟ (U, OX ).)

4. Let A, B be 2 × 2 variable matrices, let P = C[aij , bij ] be the polynomial ring

in their entries, and let R be the quotient of P by the ideal that expresses the
condition that AB = BA. Show that R has a resolution as a P -module of the form

0 → P2 → P3 → P → R → 0,

i.e., one can construct an exact sequence of the form above. (Hint: Write the
equations in terms of a11 − a22 and b11 − b22 .)
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #8 Due: 5 Nov 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Prove that, when a variety X is covered by countably many constructible sets, a

finite number of those sets will cover X.

2. (a) Let X be a proper affine variety. Show that X is a point.

(b) Let f : X → Y be surjective morphism of varieties, and suppose X is a proper
variety. Prove that Y is proper.
(c) Let X and Y be varieties. Show that X × Y is proper if and only if both X
and Y are proper.

3. Consider the space Z defined in Example 5.5.7 of the 22 October version of the
lecture notes.

(a) Consider the subspace Zirred of irreducible polynomials. Prove that Zirred is
open in Z.
(b) Consider the projection map π : Σ → Z. Show that π −1 (z) has a component
of dimension greater than 0 if and only if the polynomial f corresponding to
z is not square-free.
(c) Show how the semicontinuity of the fibre dimension implies that the subspace
Zsq-free of squarefree polynomials is open in Z.

4. Finish the proof of Theorem 5.6.2. Hint: see also exercises 5.7.16 and 5.7.18 in
the lecture notes (but you are not obliged to use these).

5. Let u : Y → X be a surjective morphism of varieties and define ∆ : X → Z by

∆(x) = max δ(y).

y∈f −1 (x)

Suppose Y is proper, prove that ∆ is upper semicontinuous.

6. Let P = (0 : 0 : 1) ∈ P2 . Determine OP2 (P2 \ {P }).

 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #9 Due: 13 Nov 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let X = Pk .

(a) Prove that the multiplication morphism OX (m)⊗OX (n) → OX (m+n) is well-
defined and an isomorphism. Deduce that the global sections of the tensor
product does not have to be isomorphic to the tensor product of the global
sections.
(b) Prove that any non-zero morphism OX (m) → OX (n) is injective. Moreover,
if m > n, prove that no such morphism exists.

2. Let R = C[x, y, z], and let f = x2 − yz.

(a) Determine generators and defining relations for the ring R{f } of homogeneous
fractions of degree zero whose denominators are powers of f .
(b) Prove that the twisting module O(1) is not a free module on the open subset
of P2 at which f 6= 0.

3. Let X = P1 . Let M be a S locally free sheaf of modules of rank 1 on X, i.e. on for

each U in some covering U of X by affine opens, there exists an isomorphism
M|U → OU of OU modules. Let U0 and U1 be the standard affines of X. You may
assume that M(Ui ) is a finitely generated O(Ui )-module for i = 0, 1.

(a) Prove that M(Ui ) is isomorphic to O(Ui ) for i = 0, 1. Hint: use the structure
theorem for finitely generated modules over a principal ideal domain.
(b) Prove that M is isomorphic to one of the twisting sheafs O(n) for some n ∈ Z.

4. Prove the following coherence property of an O-module. Let Y be an open subset of

a variety X, let s be a nonzero regular function on Y , and let Ys be a localisation.
If M is an OX -module, then M(Ys ) is the localisation M(Y )s of M(Y ). In
particular, OX (Us ) is the localisation OX (U )s . (This is a requirement for an O-
module, when Y is affine.)

5. Let n ≥ 0 and consider the twisting module O(n) on P1 with homogeneous coor-
dinates x0 , x1 .
(a) Show that the global sections xn0 , xn1 generate O(n), i.e., that the map

O ⊕ O → O(n)
(s0 , s1 ) 7→ xn0 s0 + xn1 s1

is surjective.
(b) Show that the induced map on global sections is not surjective when n > 1.
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #10 Due: 19 Nov 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let 0 → V0 → · · · → Vn → 0 be a complex of finite-dimensional vector spaces.

Show that X X
(−1)i dim(Vi ) = (−1)q dim(H q (V • )) .
i q

2. On the projective line P1 , when are two direct sums O(m) ⊕ O(n) and O(r) ⊕ O(s)
isomorphic? (Hint: Use that twisting is an exact functor.)

3. (a) Let X = Pn and K be the function field of X. Let F be the constant function
field OX -module, that is F(U ) = K for any nonempty open U ⊆ X. Prove
that H q (X, F) = 0 for all q > 0.
(b) Let X = Pn and {U i }i be an open covering of X. Suppose that rational
functions fi are given such that fi − fj is a regular function on U ij := U i ∩ U j
for all i, j. Let Q = F/O and consider the exact sequence

0 → O → F → Q → 0.

Using this exact sequence, solve the Cousin Problem: is there a rational func-
tion f such that f − fi is a regular function on V i for each i?

4. (a) Let X = P2 \ {(0 : 0 : 1)}. Use the covering of X by the two standard affine
open sets U0 , U1 to compute the cohomology H q (X, OX ).
(b) Suppose Y is a variety covered by two affine open sets, and M is an OY -
module. Show that H q (Y, M) = 0 for all q > 1.
 Math 137 - Problem Set 8
Due Friday, Apr 3
All rings are commutative, and k is an algebraically closed field.

1. Let f ∈ k[x1 , . . . , xn ] and I = (f ). Let F ∈ k[x1 , . . . , xn+1 ] be the homogenization of

f with respect to xn+1 , and let J be the homogenization of I (i.e. the ideal generated
by the homogenization of elements of I). Show that J = (F ).
2. Let V ⊂ An be an affine algebraic set, and I = Ia (V ). Let J ⊂ k[x1 , . . . , xn+1 ] be the
homogenization of I. Define V̄ ⊂ Pn to be Vp (J).

(a) (Extra credit) Show that J is radical.

(b) Show that V̄ ∩Un+1 = V (where we’ve identified Un+1 and An in the natural way).
(c) Show that V̄ is equal to the Zariski closure of V in Pn . That is, V̄ is the smallest
projective algebraic set containing V .

3. The twisted cubic C ⊂ P3 is the image of the map φ : P1 → P3 defined

φ([s : t]) = [s3 : s2 t : st2 : t3 ].

(a) Show that C is the locus of points [x : y : z : w] in P3 such that the rank of the
matrix  
x y z
y z w
is 1. (Varieties described as the vanishing of minors of a matrix are called deter-
minantal varieties.)
(b) Show that no two of the minors in part (a) determine C. That is, all three are
necessary.

4. Let H = V ( ai xi ) be a hyperplane in Pn .
P

(a) Show that assigning [a1 : . . . : an+1 ] ∈ Pn to H gives a one-to-one correspondence

between hyperplanes in Pn and points in Pn . This is called the dual projective
space. If P ∈ Pn , let P ∗ be the corresponding hyperplane; if H is a hyperplane,
let H ∗ be the corresponding point.
(b) Show that P ∈ H if and only if H ∗ ∈ P ∗ .

5. (a) Let P1 , P2 , P3 ∈ P2 be three distinct points not lying on a line. Show that if
Q1 , Q2 , Q3 ∈ P2 are distinct points not lying on a line, there is a projective
change of coordinates T : P2 → P2 such that T (Pi ) = Qi . (Hint: Show there’s
a projective change of coordinates sending them to [0 : 0 : 1], [0 : 1 : 0], and
[1 : 0 : 0].)
(b) State and prove an analogous result for two sets of three lines in P2 . (What
conditions do you need to impose on the three lines?)
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #12 Due: 3 Dec 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. (a) Let D be a divisor of degree d on a smooth projective curve Y , and assume

that h0 (D) > 0. Prove that if p is a generic point of Y , then h0 (O(D − p)) =
h0 (D) − 1.
(b) Let Y be a smooth projective curve of genus g ≥ 2, and let D = g+1
P
i=1 qi be a
divisor that is the sum of g + 1 generic points of Y . Prove that h0 (O(D)) = 2.

2. Let Y be a smooth curve of genus 2 and suppose p is a point of Y such that

h1 (O(2p)) = 0. Show that there exists a basis of global sections of O(4p) of the
form (1, x, y), where x and y have poles of order 3 and 4 at p, respectively. Prove
that this basis defines a morphism Y → P2 whose image is a singular curve of
degree 4.

3. Let Y be a smooth curve of genus g.

(a) Show that if D is a divisor on Y with deg(D) < 0, then h0 (O(D)) = 0.

(b) Show that if D is a divisor on Y and deg(D) > 2g − 2, then h0 (O(D)) =
1 − g + deg(D).
(c) Suppose that g ≥ 2 and let p be a point of Y . Note that h0 (O(kp)) ≤
h0 (O((k + 1)p)) for all k ∈ Z. Show that for k in the range 0 ≤ k ≤ 2g − 2,
there are exactly g values of k where h0 (kp) = h0 (O((k + 1)p)). (These are
called Weierstrass gaps.)
 Math 137 - Problem Set 7
Due Friday, Mar 27
All rings are commutative, and k is an algebraically closed field.

1. Suppose P = (0, 0) is a point of multiplicity two on an affine plane curve f , and suppose
f has only one unique tangent L at P .

(a) Show that IP (f, L) ≥ 3. The curve f is said to have a cusp at P if equality
holds.
(b) Suppose L = y. Show that P is a cusp if and only if fxxx (P ) 6= 0, where fxxx is
the third partial derivative of f with respect to x.
(c) Show that if P is a cusp on f , then f has only one component passing through
P.

2. Let f ∈ k[x1 , . . . , xn+1 ]. Write f = i fi , where fi is a form of degree i. Let P ∈ Pn

P
and suppose f (a1 , . . . , an+1 ) = 0 for every choice of homogeneous coordinates [a1 : . . . :
an+1 ] for P . Show that for all i, fi (a1 , . . . , an+1 ) = 0 for every choice of homogeneous
coordinates for P .

You might want to wait until after class on Monday 3/23 to work on the remaining problems.

3. Let V ⊂ Pn be an algebraic set and I a homogeneous ideal. Verify the following claims
from class:

(a) If V 6= ∅, then Ia (C(V )) = Ip (V ).

(b) if Vp (I) 6= ∅, then C(Vp (I)) = Va (I).

4. Let I ⊂ k[x1 , . . . , xn+1 ] be a homogeneous ideal.

(a) Show that I is prime if and only if for any forms (i.e. homogeneous polynomials)
F, G ∈ k[x1 , . . . , xn+1 ], such that F G ∈ I, we have F ∈ I or G ∈ I. (One direction
should be obvious.)
√
(b) Show I is homogeneous.

5. Let S = k[x1 , . . . , xn+1 ], I ⊂ S a homogeneous ideal, and R = S/I.

(a) Let d ≥ 0 and Sd ⊂ S the set of forms of degree d. Show that Sd is a k-vector
space and calculate its dimension.
(b) Let Rd be the image of Sd in the quotient. Show Rd is also a k-vector space of
finite dimension.

6. Let S = k[x, y, z], and f ∈ S an irreducible form of degree n. Let C = V (f ) ⊂ P2 , and

R = S/(f ).
(a) Show that there is a short exact sequence
·f
0 → S → S → R → 0,

where the map from S to S is multiplication by f , and the second map is the
natural quotient. (i.e. this amounts to showing that the first map is injective,
and the second map is the quotient of the first.)
(b) For each d, calculate dimk (Rd ). Keep in mind that your answer may depend on
whether d > n, d = n, or d < n.
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #7 Due: 29 Oct 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let X be a variety and let C0 ) C1 ) . . . ) Ck be some chain of irreducible closed

subvarieties of X. Prove that this chain can be extended to a chain of length
dim(X), by putting new irreducible closed subvarieties between, before, or after
the existing ones.

2. Let f : X → Y be a finite morphism. Prove that for every y ∈ Y the fibre f −1 (y)
is finite.

3. For each of the following statement determine whether the statement is true. Give
a proof or disproof.

(a) Chevalley’s finiteness theorem (corollary 4.6.8 in the 15 Oct version of the
notes) also holds when Y is affine.
(b) If f : X → Y is a surjective morphism of varieties and dim(X) = dim(Y ),
then f is finite.
(c) Let f : X → Y be a surjective morphism of curves. Suppose that X is smooth.
Then Y is smooth.
(d) Let f : X → Y be a surjective morphism of curves. Suppose that Y is smooth.
Then X is smooth.

4. Prove that, if v is a (discrete) valuation on a field K that contains the complex

numbers, every nonzero complex number c has value zero.

5. Prove that every nonconstant morphism P2 → P2 is finite. Do this by showing the

fibres cannot have positive dimension.
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #7 Due: 29 Oct 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let X be a variety and let C0 ) C1 ) . . . ) Ck be some chain of irreducible closed

subvarieties of X. Prove that this chain can be extended to a chain of length
dim(X), by putting new irreducible closed subvarieties between, before, or after
the existing ones.

2. Let f : X → Y be a finite morphism. Prove that for every y ∈ Y the fibre f −1 (y)
is finite.

3. For each of the following statement determine whether the statement is true. Give
a proof or disproof.

(a) Chevalley’s finiteness theorem (corollary 4.6.8 in the 15 Oct version of the
notes) also holds when Y is affine.
(b) If f : X → Y is a surjective morphism of varieties and dim(X) = dim(Y ),
then f is finite.
(c) Let f : X → Y be a surjective morphism of curves. Suppose that X is smooth.
Then Y is smooth.
(d) Let f : X → Y be a surjective morphism of curves. Suppose that Y is smooth.
Then X is smooth.

4. Prove that, if v is a (discrete) valuation on a field K that contains the complex

numbers, every nonzero complex number c has value zero.

5. Prove that every nonconstant morphism P2 → P2 is finite. Do this by showing the

fibres cannot have positive dimension.
 Math 137 - Problem Set 9
Due Friday, Apr 10
All rings are commutative, and k is an algebraically closed field.

1. Let P be a nonsingular (simple) point on a plane curve F . Show that the tangent line
to F at P is
Fx (P )x + Fy (P )y + Fz (P )z = 0.
(Recall that the tangent line at a point is the projective closure of a tangent line in an
affine chart containing that point.)

2. For each of the following projective plane curves, find their multiple points and the
multiplicities and tangent lines at the multiple points.

(a) x2 y 3 + x2 z 3 + y 2 z 3
(b) y 2 z − x(x − z)(x − λz), λ ∈ k
(c) xn + y n + z n , n > 0

3. Let F be an irreducible projective plane curve. Show that F has only finitely many
singular points.

4. Let V = V (y − x3 ) and W = V (x) in A2 .

(a) Show that V and W are isomorphic as affine algebraic sets.

(b) Show that the projective closures of V and W in P2 are not isomorphic. (Do you
see what’s going on geometrically?)

5. Find two projective plane curves in P2 that are isomorphic but have the property that
when you dehomogenize with respect to z (i.e. intersect with U3 ), the resulting affine
plane curves are not isomorphic.

6. Let G be an irreducible cubic.

(a) Show that G has at most one singular point, and any such singular point must
have multiplicity 2.
(b) Show that if G has a cusp (i.e., in this case, has a single tangent line of multiplicity
2 at the singular point), then it is projectively equivalent to the curve y 2 z − x3 .
(Hint: Using a problem from the previous problem set, you can start with the
singular point at [0 : 0 : 1] and the tangent line y.)
(c) Show that if G has a node (i.e. has two distinct tangent lines at the singular
point), then it is projectively equivalent to the curve xyz = x3 + y 3 .

7. (Extra credit) Let H1 , H2 , . . . , Hr be hyperplanes in Pn . Show that the intersection

H1 ∩ . . . ∩ Hr is a linear space of dimension ≥ n − r.
Bokbeslut möte 4/2013 (131209)

Optimering
Global optimization : theory, algorithms, and applications / Locatelli, Marco
Paperback: 450 s.
SIAM, 2013
9781611972665

Linear programming with duals : a modern exposition / Tovey, Craig A.

Hardback:
Chapman & Hall, 2014
9781439887462

Data correcting approaches in combinatorial optimization / Goldengorin, Boris & Pardalos,

Panos M.
Paperback: 114 s.
Springer, 2012
9781461452850

Deterministic operations research : models and methods in linear optimization / Rader, David
J.
Hardback: 613 s.
Wiley-Blackwell, 2010
9780470484517

Metaheuristics : from design to implementation / Talbi, El-Ghazali

Hardback: 593 s.
Wiley-Blackwell, 2009
9780470278581

Allmänt

Current Developments in Mathematics, 2012

[various] (Author), David Jerison (Editor), Mark Kisin (Editor), Tomasz Mrowka (Editor),
Richard Stanley (Editor), Horng-Tzer Yau (Editor), Shing-Tung Yau (Editor)
· Paperback: 264 pages
· Publisher: International Press of Boston (October 30, 2013)
· ISBN-13: 978-1571462404

Current Developments in Mathematics, 2011

[various contributors] (Author), David Jerison (Editor), Barry Mazur (Editor), Tomasz
Mrowka (Editor), Wilfried Schmid (Editor), Richard P. Stanley (Editor), Shing-Tung Yau
(Editor)
· Paperback: 198 pages
· Publisher: International Press of Boston (November 4, 2012)
· ISBN-13: 978-1571462398
Current Developments in Mathematics, 2010
[various] (Author), David Jerison (Editor), Barry Mazur (Editor), Tomasz Mrowka (Editor),
Wilfried Schmid (Editor), Richard P. Stanley (Editor), Shing-Tung Yau (Editor)
· Paperback: 186 pages
· Publisher: International Press of Boston (November 4, 2011)
· ISBN-13: 978-1571462282

Current Developments in Mathematics, 2009

[various] (Author), David Jerison (M.I.T.) (Editor), Barry Mazur (Harvard University)
(Editor), Tomasz Mrowka (M.I.T.) (Editor), Wilfried Schmid (Harvard University) (Editor),
Richard Stanley (M.I.T) (Editor), Shing-Tung Yau (Harvard University) (Editor)
· Series: Current Developments in Mathematics
· Hardcover: 204 pages
· Publisher: International Press of Boston (November 12, 2010)
· ISBN-13: 978-1571461469

Current Developments in Mathematics (2007)

by various (Author) , David Jerison (MIT) (Editor) & 5 more
· Paperback: 245 pages
· Publisher: International Press of Boston (April 1, 2009)
· ISBN-13: 978-1571461346

Matematisk historia

Correspondance Entre Henri Carten Et Andre Weil 1928-1991 / Michele Audin (Editor)
· Series: Documents Mathematiques (Book 6)
· Hardcover: 720 pages
· Publisher: Amer Mathematical Society (July 15, 2011)
· Language: French
· ISBN-13: 978-2856293140

Contributions to the History of Number Theory in the 20th Century / Peter Roquette (Author)
· Series: Heritage of European Mathematics
· Hardcover: 289 pages
· Publisher: European Mathematical Society (December 15, 2012)
· ISBN-13: 978-3037191132

Algebraisk geometri

Lectures on the Theory of Pure Motives

Jacob P. Murre (Author), Jan Nagel (Author), Chris A. M. Peters (Author)
· Series: University Lecture Series
· Paperback: 149 pages
· Publisher: American Mathematical Society (December 3, 2012)
· ISBN-13: 978-082189434
Algebraic Geometry: A Problem Solving Approach
Thomas Garrity (Author), Richard Belshoff (Author), Lynette Boos (Author), Ryan Brown
(Author), Carl Lienert (Author)
· Series: Student Mathematical Library
· Paperback: 335 pages
· Publisher: American Mathematical Society in corporation with IAS/Park City Mathematics
Institute; Student edition (February 14, 2013)
· ISBN-13: 978-0821893968

Singularities in Geometry and Topology: Strasbourg 2009 / Vincent Blanloeil (Author,

Editor), Toru Ohmoto (Editor)
· Series: Irma Lectures in Mathematics and Theoretical Physics
· Paperback: 370 pages
· Publisher: European Mathematical Society (December 15, 2012)
· ISBN-13: 978-3037191187

Algebra

Matrix Theory
Xingzhi Zhan(Author)
· Series: Graduate Studies in Mathematics
· Hardcover: 264 pages
· Publisher: American Mathematical Society (June 28, 2013)
· ISBN-13: 978-0821894910

Frobenius Algebras I: Basic Representation Theory

Andrzej Skowronski (Author), Kunio Yamagata(Author)
· Series: Ems Textbooks in Mathematics
· Hardcover: 661 pages
· Publisher: European Mathematical Society/American Mathematical Society (December 15,
2011)
· ISBN-13: 978-3037191026

Statistik

Stochastic populated dynamics in ecology and conservation / Lande, Russell et al.

Hardback: 212 s.
Oxford university press, 2003
9780198525240

Measure-valued branching Markov processes / Li, Zenghu

Hardback: 350 s.
Springer, 2010
9783642150036

Bayesian filtering and smoothing / Särkkä, Simo

Hardback: 254 s.
Cambridge UP, 2013
9781107030657

Fundamentals of matrix-analytic methods / He, Qi-Ming

Hardback
Springer, 2013
9781461473299

Statistical intervals : a guide for practitioners / Hahn, Gerald J. 2. ed.

Hardback: 520 s.
Wiley-Blackwell, 2014
978047168177

Thinking, fast and slow / Kahneman, Daniel

Hardback: 512 s.
Farrar, Straus and Giroux, 2013
9780374533557

Analys, geometri

Oxford user’s guide to mathematics / Zeidler, Eberhard

Paperback
Oxford UP, 2013
9780199686926

A course in mathematical analysis / Garling, D. J. H.

Hardback
Cambridge UP, 2014
9781107032033

Decay of the Fourier transform : analytic and geometric aspects / Iosevich, Alex & Lifyand,
Elijah
Hardback
Birjhäuser/Springer, 2013
9783034806244

Global wellposedness of nonlinear evolutionary fluid equations / Qin, Yuming m.fl.

Paperback
Springer, 2013
9783034805933

Harmonic and geometric analysis / Citti, Giovanna m.fl.

Paperback
Birkhäuser/Springer, 2013
9783034804073

Paradoxes in mathematics / Farlow, Stanley

Paperback
Dover publications, 2014
9780486497167
Övrigt

Game-theoretical models in biology / Broom, Mark & Rychtar, Jan

Hardback: 494 s.
Chapman & Hall/CRC, 2013
9781439853214

Applied inverse problems / Beilina, Larisa

Hardback:
Springer, 2013
9781461478157

Homotopy type theory

Paperback
Inst. for advanced study, s.a.

Random heterogeneous materials : microstructure and macroscopic properties / Torquato,

Salvatore
Hardback: 701 s.
Springer, 2006
9780387951676

Ordinary differential equations with applications / Hsu, Sze-Bi

Hardback: 244 s.
World scientific, 2005
9789812563194

Riemannian geometry / Willmore, T. J.

Paperback: 330 s.
Oxford UP, 1996
9780198514923
 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #5 Due: 15 Oct 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Let X = SpecA be an affine variety. The following conditions defined close subsets
Y of X × Pn . Prove that they are equivalent.

(a) Y is the locus of zeros of a finite number of homogeneous polynomials in

A[x0 , . . . , xn ].
(b) For the standard affine opens Ui ⊂ Pn , the intersection Y ∩ (X × Ui ) is a
closed subset of the affine variety X × Ui .

2. Determine all the morphisms P2 → P1 .

3. Consider the Veronese embedding P2x,y,z → P5u by monomials of degree 2 defined

by (u0 , u1 , u2 , u3 , u4 , u5 ) = (z 2 , y 2 , x2 , yz, xz, xy). If we drop the coordinate u0 , we
obtain a map φ : P2 \ {q} −→ P4 that is defined at all points except q = (0, 0, 1)
by (x, y, z) 7→ (y 2 , x2 , yz, xz, xy). Prove that the inverse map φ−1 is a morphism,
and that the fibre over q is a projective line.

4. Let C be the projective plane curve x3 − y 2 z = 0.

(a) Show that the function field K(C) of C is the field C(t) of rational functions
in t = y/x.
(b) Show that the point (t2 − 1, t3 − 1) of P1 (K(C)) defines a morphism C → P1 .

5. Show that the conic C in P2 defined by the polynomial x2 + y 2 + z 2 = 0 and the

twisted cubic V = V (v0 v2 − v12 , v0 v3 − v1 v2 , v1 v3 − v22 ) in P3 are isomorphic, by
exhibiting a inverse morphisms between them.

6. Show that the morphism φ : A1 → A1 given by φ(t) = t2 is an integral morphism.

 18.721 Introduction to Algebraic Geometry
Fall 2021
Problem Set #6 Due: 22 Oct 2021

Description
Solutions are to be prepared in typeset form (typically via latex) and submitted electron-
ically as a pdf file via Gradescope by midnight on the due date. Late problem sets will
not be accepted – your lowest score is dropped, so you can afford to skip one problem
set without penalty.
Collaboration is permitted/encouraged, but you must write up your own solutions
and explicitly identify your collaborators, as well as any resources you consulted that are
not listed above. If there are none, please indicate this by writing Sources consulted:
none at the top of your submission.

1. Prove this alternate form of the Nullstellensatz: Let k be a field of characteristic

zero, and let B be a domain that is finitely generated k-algebra. If B is a field,
then [B : k] < ∞.

2. Let A be a finite type domain over C, R = C[t], X = Spec(A), and Y = SpecR.

Let φ∗ : A → R be a homomorphism whose image is not a subset of C, and let
φ : Y → X be the corresponding morphism. Show that R is a finite A-module,
and that the image φ is a closed subset of X.

3. Let α be an element of a domain A, and set β = α−1 . Prove that if β is integral

over A, the it is an element of A.

4. Let k be a field of characteristic zero. Let A ⊂ B be finite type domains over k with
fraction fields K ⊂ L of characteristic zero, and let Y → X be the corresponding
morphism of affine varieties. Prove the following:

(a) There is a nonzero element s in A such As is integrally closed.

(b) There is a nonzero element s in A such that Bs is a finite A-module over a
polynomial ring As [y1 , . . . , yd ].
(c) Supposed that L is a finite extension of K of degree d. There is a nonzero
element s ∈ A such that all fibres of the morphism Ys → Xs consist of d
points.

5. Identify the normalization of A = C[x, y, z]/(y 2 − xz 2 ).