INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

1. 1/27/20 - Introduction
Announcements. The class will meet on MW in general, but not this Wednesday.
The references are:
• 3264 and all that by Joe Harris and David Eisenbud.
• Intersection Theory by William Fulton.

Let X be a set. A basic observation about subsets A, B ⊂ X is that

A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).
So ∩ distributes over ∪ just as × distributes over +. Can we convert this observation
into a ring structure? One way to make this work is by forming the Boolean algebra:
(F2 )X .
The product corresponds to ∩ and the sum corresponds to symmetric difference
(union excluding intersection). The multiplicative identity is X itself. Unfortu-
nately this ring is enormous, and it does not see any geometric structures on X.

Next assume that X is a smooth manifold, and that A, B ⊂ X are closed sub-
manifolds. If A and B intersect transversely (which can be arranged by perturbing
them slightly), then
codim(A ∩ B) = codim(A) + codim(B).
So if we restrict our attention to submanifolds of X, then we should be able to
define a graded ring structure, graded by codimension. To accommodate small per-
turbations, we need to consider equivalence classes instead of literal submanifolds.

This idea can be formalized using de Rham cohomology. Let X be an oriented

manifold of dimension n. To any closed oriented submanifold A ⊂ X of codimen-
sion a, we can define its Poincaré dual
[ωA ] ∈ H a (X, R),
where ωA is a closed a-form supported on a small tubular neighborhood of A with
the property that the integral of ωA over any transverse cross-section of the tube is 1.

Exercise. If [ωA ] and [ωB ] are the Poincaré duals of closed oriented submani-
folds A, B ⊂ X meeting transversely, then
ωA ∧ ωB = [ωA∩B ].
The de Rham cohomology ring gives us a model for transverse intersections:
Mn
H ∗ (X, R) = H a (X, R).
a=0
1
2 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

If we assume that X is compact, then the isomorphism X : H n (X, R) → R sends

R

the Poincaré dual of a point to 1. This allows us to count the (oriented) number of
intersection points between A and B when codim(A) + codim(B) = n.

Poincaré duality takes submanifolds to integral cohomology classes, and in fact the
wedge product can be defined directly on the integral cohomology groups, where it
is called the cup product.

Example. Let X be a closed Riemann surface of genus g. Then H 0 (X, Z) '

H 2 (X, Z) ' Z, and H 1 (X, Z) = Z2g , with a basis given by the duals of the stan-
dard loops α1 , . . . , αg , β1 , . . . , βg . Because 1 is odd, the wedge product pairing

H 1 (X, Z) ⊗ H 1 (X, Z) → H 2 (X, Z) ' Z

is skew-symmetric. The product counts intersections of two loops intersecting trans-
versely with orientations.

This can give funny answers. For example, in a genus 4 surface (with the holes
arranged like a cross), you can draw a pair of loops which meet twice, but their
oriented intersection number is 0. These loops cannot be deformed to disjoint loops.

One way to avoid the orientation issue is to study complex manifolds and their
holomorphic submanifolds. A complex vector space induces a natural orientation
on the underlying real vector space such that any two complementary complex sub-
spaces have positive intersection. Furthermore, the real codimension of a complex
submanifold is always even, so the wedge product is commutative.

Definition. A projective manifold is a closed complex submanifold of CPN .

Theorem. (Chow) Any projective manifold is cut out by polynomial equations.

As a consequence, if X is a projective variety, then any complex submanifold A ⊂ X

is a projective variety too. Perhaps there is a way to define an intersection ring
for algebraic subvarieties of X purely in terms of algebraic geometry, without using
smooth topology. This will be the main subject of the course.

Proposition. The cohomology ring of CPN is given by

(
Z ∗ even
H ∗ (CPN , Z) '
0 ∗ odd

Proof. Breaking CPN into AN ∪ AN −1 ∪ · · · ∪ A0 gives a cell decomposition which

yields a chain complex with a single generator in even degrees and 0 differential.
This complex computes cellular homology or cohomology. 

Each generator is represented by the Poincaré dual of a projective subspace Pk

for 0 ≤ k ≤ N . It is easy to see the ring structure in terms of these generators.
Since any two linear subspaces can be translated so that they intersect transversely,
[Pk ] · [Pl ] = [Pk+l−N ].
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 3

Let A ⊂ PN be a complex submanifold of (complex) codimension a. What is the

class [A] ∈ H 2a (PN , Z)? Well we know that [A] = d[PN −a ]. To find d, we just need
to intersect [A] with [Pa ]:
[A] · [Pa ] = d[P0 ].

Definition. The integer d > 0 is called the degree of A.

From the multiplication law on cohomology, we see that degree is multiplicative

under transverse intersections. If A ⊂ PN is a hypersurface, that is a submanifold
cut out by a single homogeneous polynomial, then d is the degree of the polynomial.

Theorem. (Bézout) Let C, D ⊂ P2 be two curves in P2 defined by irreducible

polynomials of degree d and e. As long as they are not the same curve, they will
intersect at de points, counted with multiplicity.

If C meets D transversely, then this theorem follows immediately from the dis-
cussion above. If they have non-transverse intersection points, then there is a
multiplicity involved.

Definition. Suppose that 0 ∈ C2 is an intersection point of C and D, defined

by polynomials f (x, y) = 0 and g(x, y) = 0. The intersection multiplicity at 0 is:

mult0 (C, D) = dimC C[x, y]/(f, g).

We will prove Bézout’s Theorem later on. To finish up, let us work out a sample
application of intersection theory.

Question. Fix 4 general lines in C3 . How many lines in C3 meet all 4?

Answer. 2! And the answer is the same for lines in R3 (easier to visualize).

To get this answer, we consider the moduli space of all lines in C3 . This space
is non-compact, so to use an intersection product, we compactify it by considering
the space of projective lines in CP3 . The result is a Grassmannian:

G(1, 3) = G(2, 4) = {2-dimensional subspaces U ⊂ C4 }

Now G(2, 4) is itself a projective manifold. In fact it is a hypersurface in P5 . To see

this, we define a map from G(2, 4) into P(∧2 C4 ) sending U = span(v, w) to [v ∧ w].
Choosing a different basis for U changes the image by a determinant, which is why
we must projectivize ∧2 C4 .

Exercise. The subset of P5 = P(∧2 C4 ) consisting of “pure wedges” is cut out

by the equation α ∧ α = 0, where α ∈ ∧2 C4 , so G(2, 4) is a quadric hypersurface.

Now that we have a nice moduli space, we need to impose the condition of meeting
the fixed lines `1 , `2 , `3 , `4 ⊂ P3 . Let’s define

Σ(`) := {lines P1 ⊂ P3 which meet `} ⊂ G(1, 3) = G(2, 4)

4 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

The count we are after is the size of Σ(`1 ) ∩ Σ(`2 ) ∩ Σ(`3 ) ∩ Σ(`4 ). A dimension
count reveals that each Σ(`i ) is a codimension 1 subvariety1 of G(1, 3). In fact, it is
cut out by the equation `i ∧ α = 0. The count we want is the intersection number
of a quadric hypersurface with 4 hyperplanes in P5 , which is 2 by the multiplicative
property of degree.

To generalize this example, we can study the cohomology ring of the Grassmannian:



 Z

Z


H 2∗ (G(1, 3), Z) ' Z2




 Z

Z.

The generator of H 2 (G(1, 3), Z) is called σ1 , and it is the class of Σ(`) for any line
` ⊂ P3 . The observations above imply that
Z
(σ1 )4 = 2.
G(1,3)

Question. What are the two generators of H 4 (G(1, 3), Z)? To be continued.

More generally, the Grassmannian G(1, n) has dimension 2n − 2. There is an

analogous class σ1 defined as the set of lines meeting a fixed codimension 2 linear
subspace Λ ⊂ Pn . Another natural question is then:
Z
(σ1 )2n−2 = ?
G(1,n)

Alternative approaches to the 4 lines problem:

1. The lines meeting `1 , `2 , `3 sweep out a quadric surface Q ⊂ P3 . The last

line `4 meets that surface in two points.

2. Assume that `1 and `2 intersect at p (so they are not in general position).
There are two ways for L to meet both `1 and `2 : either L lies in the plane H
spanned by `1 and `2 , or L contains p. In the first case, L must be the line between
`3 ∩ H and `4 ∩ H. In the second case, projecting from p sends `3 and `4 to a pair
of lines on a screen which meet at a single point q, so L must be pq.

1Σ(`) is a singular subvariety because it is the intersection of G(1, 3) with the hyperplane
tangent to G(1, 3) at [`].
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 5

2. 2/3/20 - Chow Groups

2.1. Basic Properties. Let X be a scheme (often a smooth quasi-projective va-
riety if intersections are involved) over a field k. We would like to define a theory
of algebraic cycles on X. Start with a free abelian group:
Z(X) = Z{algebraic subvarieties of X}
By variety,
P we mean a reduced, irreducible subscheme (separatedness is automatic).
A cycle ni Zi ∈ Z(X) is effective if all the ni > 0. If Y is a subscheme, its
associated cycle [Y ] is a sum over the irreducible components weighted by their
multiplicity. For an irreducible component, Yi ⊂ Yred , its multiplicity is defined as
the length of the local ring OY,Yi over itself. X is Noetherian, so any subscheme
has finite length over its local ring.

Let Rat(X) ⊂ Z(X) be the subgroup generated by

W ∩ ({0} × X) − W ∩ ({∞} × X)
for W ⊂ P1 × X a subvariety not contained in any fiber of P1 × X → P1 .

Definition. The Chow group A(X) is the quotient Z(X)/Rat(X).

The Chow group is graded by dimension:

M
A(X) = Ak (X).
k

Theorem. There exists a product structure on A(X) satisfying the condition that
if A, B ⊂ X are generically transverse subvarieties, then
[A] · [B] = [A ∩ B].
Generically here means in A ∩ B. The most intuitive definition of the intersection
product goes through the Moving Lemma.

Lemma. For every subvarieties A, B ⊂ X, there is a cycle α ∈ Z(X) ratio-

nally equivalent to [A] such that every component of α is generically transverse
to B. Furthermore, the class [α ∩ B] is independent of α. We will prove this on
Wednesday. NB: this can fail when X is singular.

Theorem. (Kleiman): Suppose that G is an algebraic group acting transitively on

X (characteristic 0), and let A, B ⊂ X be subvarieties. Then for generic g ∈ G, gA
is generically transverse to B.

Theorem. (Bertini): A general hyperplane section of a smooth projective va-

riety X is smooth. More generally, a general member of a linear system of divisors
on X is smooth away from its base locus.

How do we compute the Chow group of X? If dim(X) = n, then An (X) = Z[X] be-
cause P1 ×X is irreducible. If X is reducible but equidimensional, then An (X) = Zc .
Furthermore A(Xred ) = A(X).
6 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Affine space: A(An ) = Z. Indeed, let Y ⊂ An be any subvariety not contain-

ing the origin. Consider W ◦ ⊂ (A1 )× × An be the set of (t, ty) for y ∈ Y , and let
W be its closure in P1 × An . Precisely, it is defined by the polynomials f (y/t) for
all f ∈ I(Y ). The fiber over 1 ∈ P1 is Y , and the fiber over ∞ ∈ P1 is empty (there
is polynomial f with nonzero constant term, and it will not vanish at ∞).

Exercise. The fiber over 0 is the cone over Y ∩ H∞ .

Theorem. Let X be a scheme. For any subschemes X1 , X2 ⊂ X, we have an

exact sequence
A(X1 ∩ X2 ) → A(X1 ) ⊕ A(X2 ) → A(X1 ∪ X2 ) → 0.
For Y ⊂ X a closed subscheme, we have an exact sequence
A(Y ) → A(X) → A(U ) → 0
Proof. Both theorems are proved the same way.

0 / Z(Y × P1 ) / Z(X × P1 ) / Z(U × P1 ) /0

  
0 / Z(Y ) / Z(X) / Z(U ) /0

  
A(Y ) / A(X) / A(U )

  
0 0 0
the first vertical maps are ∂, defined to be zero on cycles contained in fibers, and
otherwise the difference between the fibers at {0} and {∞}. The theorem follows
from a diagram chase. The Mayer-Vietoris is similar. 

Corollary. If U ⊂ An is a nonempty open set, then A(U ) = An (U ) = Z. (The

first map is 0, and the second is an isomorphism.)

Remark. The right exact sequences above can be extended to the left using the
higher Chow groups of Bloch, which are a special case of motivic cohomology.

We say that X is stratified if it is a union of locally closed irreducible subschemes

Ui , and the closure of any Ui is a union of Uj . We say that a stratification is affine
if every Ui is isomorphic to an affine space. We say it is quasi-affine if every Ui is
isomorphic to an open subset of affine space.

Proposition. If X admits a quasi-affine stratification, then A(X) is generated

by the classes of closed strata Ui .

Proof. Induct on the number of strata. If U0 is a minimal stratum, it is closed.

Z[U0 ] = A(U0 ) → A(X) → A(X r U0 ) → 0.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 7

By induction, the last piece is generated by closed strata classes. 

Theorem. (Totaro 2014) The classes of closed strata in an affine stratification

form a basis of A(X).

Remark. In practice, we can avoid appealing to this theorem by checking that the
closed strata in complementary dimension have nonsingular intersection matrix.

2.2. Functoriality. Let f : X → Y be a proper map of schemes. Then for any

subvariety A ⊂ X, f (A) ⊂ Y will be a subvariety.

Define the pushforward on cycles f∗ : Z(X) → Z(Y ) sending [A] to 0 if dim(A) >
dim(f (A)), and [A] to d[f (A)] otherwise,
d = [R(A) : R(f (A))].
Proposition. The homomorphism f∗ descends to f∗ : A(X) → A(Y ).
Proof. Use the norm map on fields of rational functions, to be introduced later.


Exercise. Find a non-proper map such that f∗ is not well-defined (hint: try
an open immersion).

If X is proper over a point, then point classes are nonzero, which gives the no-
tion of degree for a 0-cycle.

Let f : X → Y be an arbitrary map of smooth quasi-projective varieties. We

would like to define a notion of pullback.

Definition. A subvariety A ⊂ Y is generically transverse to f if the pre-image

f −1 (A) is generically reduced and has the same codimension as A.

Theorem. There is a unique map of groups f ∗ : Ak (Y ) → Ak (X) such that

for any A ⊂ Y generically transverse to f ,
f ∗ [A] = [f −1 (A)].
Remark. This theorem actually gives the intersection product, by applying i∗ i∗
for i the inclusion of B. For this reason, f ∗ actually gives a homomorphism of
graded rings A∗ (Y ) → A∗ (X). If f is proper too, we have α · f∗ (β) = f∗ (f ∗ α · β).
This implies that f∗ : A∗ (X) → A∗ (Y ) is a homomorphism of A∗ (Y )-modules.

NB: In the case when f is flat, you can just define f ∗ : Z(Y ) → Z(X) and it
descends to Chow; no Moving Lemma is required. Fulton calls this the flat pull-
back, and the more general one the Gysin pullback denoted f ! .

2.3. Refinements. At this stage we have defined [A]·[B] ∈ A∗ (X) when A, B ∈ X

are generically transverse. Fulton defined a refined intersection [A]·[B] ∈ A∗ (A∩B).
A partial refinement is given by the (Gysin) pullback. More on this later.
8 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Suppose that A ∩ B is a union of components C of the correct codimension. Then

X
[A] · [B] = mC (A, B)[C] ∈ A∗ (X).
Here are some features of the intersection multiplicities mC (A, B):
• mC (A, B) ∈ Z+ .
• mC (A, B) = 1 iff A and B intersect generically transversely along C.
• If A and B are Cohen-Macaulay at a general point of C, then mC (A, B) is
the scheme-theoretic multiplicity of C.
• If not, then you need to use the Serre formula to find mC (A, B).
2.4. Grassmannians. Recall that G(1, 3) = G(2, 4) is a quadric hypersurface in
P5 = P(∧2 C4 ) given by the single equation α ∧ α = 0.

We will describe the Chow group A(G(1, 3)) by giving an affine stratification. Fix
a complete flag p ∈ L ⊂ H ⊂ P3 .
Σ0,0 = G(1, 3)
Σ1,0 = {Λ : Λ ∩ L 6= 0}
Σ2,0 = {Λ : Λ ∩ p 6= 0}
Σ1,1 = {Λ : Λ ⊂ H}
Σ2,1 = {Λ : p ∈ Λ ⊂ H}
Σ2,2 = {Λ : Λ = L}
1. The dense open stratum Σ◦0,0 can be thought of as the locus of V ⊂ C4 com-
plementary to a fixed L ⊂ C4 . Fix one such Ω (as the origin). All others will be
graphs of linear maps from Ω → L, so we get an A4 .

2. The next open Σ◦1,0 is the set {Λ : Λ ∩ L 6= 0, p ∈ / Λ, Λ 6⊂ H}. Let H 0 be a

plane containing p but not L. Any Λ above meets H away from H 0 ∩ H. We get
0

maps Σ◦1,0 → L r p and Σ◦1,0 → H 0 r (H 0 ∩ H), so A1 × A2 ' A3 .

3. The middle two can be thought of as P2 and P2∗ respectively, and you are
removing a P1 = Σ2,1 in both cases. 4. P1 = A1 ∪ {∞}.

We now know the Chow group completely. What is the ring structure? It’s easy
to see that
2 2
σ1,1 = 1, σ2,0 = 1, σ1,1 σ2,0 = 0, σ1 σ2,1 = 1.
Next up, we see that σ1 σ2 = σ2,1 and σ1 σ1,1 = σ2,1 also.

Lemma. σ12 = σ1,1 + σ2 .

Proof. If Λ meets L1 and L2 , and L1 meets L2 , then either it passes through

L1 ∩ L2 or it lies in the plane L1 L2 . Alternatively, you can just use the relations
we already know. σ12 = aσ2 + bσ1,1 .
a = σ12 σ2 = σ1 σ2,1 = 1
b = σ12 σ1,1 = σ1 σ2,1 = 1. 
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 9

3. 2/5/20 - Moving Cycles

3.1. Loose Ends. Today we will always assume that f : X → Y is a morphism of

smooth projective varieties (so automatically proper). Recall that f ∗ : A∗ (Y ) →
A∗ (X) is homomorphism of graded rings, and f∗ : A∗ (X) → A∗ (Y ) is a homomor-
phism of graded A∗ (Y )-modules.

Definition. Let L/K be a finite extension of fields. The norm NL/K : L → K is

defined as follows. For x ∈ L, the multiplication mx : L → L is K-linear.
NL/K (x) := det(mx ) ∈ K.
It can be also be computed in a normal closure of L as the product of conjugates:
Y
NL/K (x) = xσ .
σ
If f : X → Y is generically finite and dominant, then we have a finite extension
of rational function fields R(X)/R(Y ). Given a rational function ϕ, its associated
divisor is given by
div(ϕ) = ϕ−1 (0) − ϕ−1 (∞).
Lemma. If ϕ ∈ R(X), then:
f∗ (div(ϕ)) = div(NR(X)/R(Y ) (ϕ)).
A nice consequence of this is a more economical definition of rational equivalence
Ratk (X). Instead of taking all subvarieties W ⊂ X ×P1 , it suffices to consider those
which are graphs of rational functions on W ⊂ X of dimension k + 1. With this
definition, it is easy to see that the proper pushforward descends to Chow groups.

Last time, we saw that A0 (X) = Z[X]. The next case to consider is A1 (X).
Here, rational equivalence of divisors coincides with linear equivalence. By the
correspondence between divisors and line bundles, we have an isomorphism
c1 : Pic(X) → A1 (X).
Remark. If X is singular, then c1 is no longer surjective; in that case A1 (X) '
Cl(X), the Weil divisor class group.

The pullback f ∗ : A1 (Y ) → A1 (X) can be defined by pulling back line bundles.

To get a sense of the size of A1 (X), recall that for X a curve of genus g,
0 → Pic0 (X) → Pic(X) → Z → 0.
Over C, we have Pic0 (X) ' Cg /Z2g a complex torus (Abel’s Theorem). More
generally, for X smooth projective we have a cycle class map
cck : Ak (X) → H 2k (X, Z).
Its kernel can be quite large in general. In dimension 0, ccn : An (X) → Z is an
isomorphism if X is rationally connected (any two points can be linked by a ra-
tional curve). But there are actually varieties of general type such that An (X) ' Z.

Conjecture. (Bloch) If S is a surface with H 1,0 (S) = H 2,0 (S) = 0, then

A2 (X) ' Z.
10 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

3.2. Moving Lemma. Let A, B ⊂ X be subvarieties with dim(X) = n. There

exists α ∈ Z(X) such that [α] = [A] ∈ A∗ (X) but each component of α intersects
B generically transversely.

Proof. As a warm-up, let us prove the case where A is a divisor. Let L be

an ample line bundle on X. For m  0, we have that L⊗m and L⊗m ⊗ OX (D) are
very ample. General sections D1 and D2 of them meet B generically transversely
by Bertini’s Theorem. In that case, we can take α = (s2 ) − (s1 ).
To prove the lemma in general, embed X into some projective space PN , and let
Λ ⊂ PN be a general linear subspace of dimension N − n − 1, so Λ ∩ X = ∅. Let
πΛ : X → Pn be the linear projection, a finite map in this case since Λ is general.
Let Ae = π −1 πΛ (A) = X ∩ Λ, A. The bar denotes linear join: the union of all lines
Λ
e = A + A0 , where A0 is generically reduced
linking Λ with A. Since πΛ is finite, A
with the same dimension as A (this follows from Bertini’s Theorem). From here,
the strategy consists of three steps:
(1) Show that A e can be made transverse to B.
(2) Show that no component C ⊂ A0 ∩ B is also a component of A ∩ B.
(3) Show that A is generically transverse to B ∗ = B r (A ∩ B).
The Moving Lemma will follow by induction: write A = A e − A0 , and then apply
0
the same construction to A .

(1) Recall that Ae is equal to X ∩ Λ, A. By Kleiman’s Theorem, a general P GLN +1 -

translate of Λ, A in PN will be generically transverse to X and to B.

(2) Every component C ⊂ A ∩ B contains points p such that Λ does not meet
Tp X. Indeed, this is a Zariski open condition, and we can choose Λ so that it’s
true for one point on each component. The projection πΛ : X → Pn is nonsingular
at such points p. Hence p ∈/ A ∩ A0 because A
e = A ∪ A0 is singular along A ∩ A0 .

(3) Here we punt a little and just show that A is dimensionally transverse to B ∗ .
With another page of work, we could show that they are generically transverse, but
with the theory of intersection multiplicities the weaker statement suffices. Consider
Ψ = {(Λ, p, q) ∈ G(N − n − 1, N ) × A × B ∗ : Λ ∩ pq 6= ∅}
The fiber over a point (a, b) ∈ A × B ∗ is the set Λ meeting a fixed line, which is
codimension n inside G. Thus Ψ is irreducible, and
dim Ψ = dim A + dim B + dim G − n.
e ∩ B ∗ = A0 ∩ B ∗ . That fiber has
Now, a general fiber of Ψ → G surjects onto A
dimension dim A + dim B − n, so
e ∩ B ∗ ) ≤ dim A + dim B − n.
dim(A
On the other hand, dim(A e ∩ B ∗ ) ≥ dim A + dim B − n because codimension is
subadditive, so we are done. 

To get an idea of what is happening, consider the example where X ⊂ P3 is a

smooth quadric, so isomorphic to P1 × P1 , and A = B = P1 is one of the rulings.
e = P1 ∪ P1 . Now A
In this case, Λ is a point and A e can be moved to a general linear
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 11

section of X (a smooth conic C), which meets B at a single point. On the other
hand, A0 already meets B in a single point. We see that A · B = 0 using
A ∼ C − A0 .
3.3. Applications. Now we return to the Grassmannian G(1, 3) to see what the
intersection product can do for us.

Example. Let C ⊂ P3 be a smooth, nondegenerate curve. Consider the locus

Σ(C) = {lines Λ ⊂ P3 meeting C}.
It is not hard to see that Σ(C) ⊂ G(1, 3) is codimension 1. Since A1 (G(1, 3)) = Zσ1 ,
we know that [Σ(C)] = dσ1 . To find d, we intersect with the complementary σ2,1 :
d = [Σ(C)] · σ2,1
Recall that Σ2,1 a pencil of lines through a point p lying in a plane H ⊂ P3 . Since
H intersects C in deg(C) points, we find that d = deg(C).

Example. Let C ⊂ P3 be as before. Consider the locus

Sec(C) = {lines Λ ⊂ P3 meeting C twice}.
It is not hard to see that Sec(C) ⊂ G(1, 3) is codimension 2, so we know that
[Sec(C)] = aσ1,1 + bσ2
To find a and b, we intersect with the complementary cycles σ1,1 and σ2 :
a = [Sec(C)] · σ1,1
b = [Sec(C)] · σ2 .
Recall that Σ1,1 is the locus of lines contained in H. Since H intersects C in
d = deg(C) points, we find that a = d2 .

Recall that Σ2 is the locus of lines through p. The linear projection πp : P3 99K P2
sends C to a degree d plane curve with b nodes. A smooth plane curve always
has genus d−1


2 by the Riemann-Hurwitz formula. The number of nodes is the
difference between the arithmetic and geometric genera of πp (C) ⊂ P2 :
 
d−1
b= − g(C).
2
The genus g(C) must be part of the problem; it can take any value from 0 to the
Castelnuovo bound, which is quadratic in d.
12 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

4. 2/10/20 - Schubert Calculus

4.1. Grassmannians. Today we will discuss the Chow ring of a general G(k, n).
As in the case G(2, 4) from last week, G(k, n) admits an affine stratification where
the strata are indexed by sequences a of integers of length k such that:
n − k ≥ a1 ≥ a2 ≥ · · · ≥ ak ≥ 0.
This can be viewed as a Young tableau contained in a box of height n−k and width
k, with the bars decreasing in height from left to right. We will suppress trailing
zeroes from the notation.

Remark. There are nk such tableaux, by counting possibilities for the jagged

boundary of the tableau.

To define the stratification, fix a complete flag in Cn :

0 = V0 ⊂ V1 ⊂ · · · ⊂ Vn−1 ⊂ Vn = Cn .
Given a sequence a as above, we set
Σa := {Λ ∈ G(k, n) : dim(Λ ∩ Vn−k+i−ai ) ≥ i}.
To parse this, note that for Λ general, dim Λ ∩ Vn−k+i = i, so the non-generic strata
correspond to earlier intersections than expected. If a ≤ a0 , then Σa ⊃ Σa0 . The
following are special cases:

Example. The locus of Λ contained inside some Vr corresponds to a = (n − r)k .

Example. The locus of Λ containing some Vr corresponds to a = (n − k)r .

Example. The locus of Λ meeting some Vl non-trivially corresponds to

a = n − k − (l − 1).
For simplicity, from now on we assume that the flag is the standard flag in Cn .

Proposition. The open stratum Σ◦a is an affine space of codimension |a| =

P
ai .

Proof. We can choose a basis for Λ0 inductively which consists of vectors in

Vn−k+i−ai as soon as they become available. If Λ0 ∈ G(k, n) is generic then this
basis looks like:    
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
   
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
   
∗ ∗ ∗ ∗ → 1 0 0 0
n   
0 ∗ ∗ ∗ 0 1 0 0
   
0 0 ∗ ∗ 0 0 1 0
0 0 0 ∗ 0 0 0 1
The highest column of 0’s has height k − 1. If Λ0 is inside Σa , then we can make
the ith column of 0’s higher by ai , after performing the column operations, we can
read off the dimension of the affine stratum. 
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 13

Given a partition a, there is a complementary partition a∗ given by

a∗i = n − k − ak+1−i
Proposition. Schubert classes in complementary dimension have product 0, except
for the case:
σa · σa∗ = 1.
Proof. The action of GLn simply changes the flag, so by Kleiman’s Theorem
any intersection product can be computed using two generic flags, V• and W• . To
intersect Σa (V• ) with Σb (W• ) consider the conditions in pairs (i, k + 1 − i):
dim(Λ ∩ Vn−k+i−ai ) ≥ i
dim(Λ ∩ Wn+1−i−bk+1−i ) ≥ k + 1 − i.
If you view both of these intersections inside Λ, their dimensions sum to k + 1, so
they must have non-trivial intersection. Since V• and W• are generic, their dimen-
sions in Cn must also add up to ≥ n + 1. This forces b = a∗ . There is a unique
Λ by taking the span of the k vectors you get by intersection each complementary
pair Vn−k+i−ai ∩ Wn+1−i−a∗k+1−i =k+1−i+ai . 

With this in hand, we can encode the full ring structure in terms of structure
constants:
c
γa,b = σa · σb · σc∗ .
c
Remark. The γab are always positive because they are actually the same as
Littlewood-Richardson coefficients. Irreducible representations of GLk are classified
by partitions of length k (Schur functors applied to the standard representation).
If you tensor together two irreps, you can decompose the result into irreps. This
gives a ring called the representation ring, Rep(GLk ) which can also be expressed
in terms of characters as polynomial functions in the entries of a matrix, invariant
under conjugation. We have a sequence of isomorphisms
Rep(GLk ) → Sym∗ (gl∨
k)
GLk
→ H 2∗ (BGLk )
(closure followed by Chern-Weil). The latter has a natural map to → H 2∗ (G(k, n))
via the tautological bundle on G(k, n). A particular Schur irrep maps to the corre-
sponding Schubert class via the composition.

Consider the example G(1, n) = G(2, n + 1) of lines in Pn , which has dimension

2n − 2. A natural questions is what is its Plücker degree, that is σ12n−2 ?
Σa,b = {Λ ∈ G(1, n) : Λ ∩ P2−a 6= ∅, Λ ⊂ P3−b }.
Proposition. σ1 · σa,b = σa+1,b + σa,b+1 , where a summand may be zero if that
sequence is not allowed.

Proof. Similar to our proof of the fact that σ12 = σ1,1 + σ2 . Move the Pn−2
defining Σ1 so that it intersects P2−a in codimension 1, and P3−b in codimension
2. There are two ways for Λ to lie in Σ1 ∩ Σa,b : either it meets Pn−2 ∩ P2−a , or it
lies inside the span of P2−a and Pn−2 ∩ P3−b , which is a P3−b−1 . 

Corollary. In G(1, n), the top intersection σ12n−2 has degree Catalan(n+1).
14 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Proof. Stacking unit boxes inside the (n + 1) × 2 box such that each step is a
Young tableau is the same as placing n+1 pairs of parentheses. 

Theorem. (Pieri) In G(k, n), σc · σa is equal to the sum over all tableaux ob-
tained by adding c blocks to a, at most one block in each row.

These σc are called special Schubert classes. Actually, any class can be expressed
as a polynomial in these special classes:

Theorem. (Giambelli) In G(k, n), any σa can be expressed as the determinant

of a matrix with special Schubert classes as the entries:
 
 σa1 σa1 +1 . . . σa1 +l−1 

 σa2 −1 σa2 . . . σa2 +l−2 
σa1 ,a2 ,...,al =  . .

 .. ..
 . 

σa −l+1 σa −l+2 . . . σ al

l l

Proof. Use the Pieri formula with induction by expanding along the last column.
For example:
 
 σ σa1 +1 
σa1 ,a2 =  a1 = σa1 σa2 − σa1 +1 σa2 −1 . 
σa2 −1 σa2 
Pieri and Giambelli together give an algorithm to compute any intersection prod-
uct. Vakil gave a more efficient algorithm for computing LR coefficients in terms
of checkers puzzles. Positivity is clear from Vakil’s approach.

Lemma. (1 − σ1 + σ1,1 − · · · ± σ1k )(1 + σ1 + σ2 + · · · + σn−k ) = 1.

Proof. For d > 0, using Pieri we have

d
X
= σd − (σd + σd−1,1 ) + (σd−1,1 + σd−2,1,1 ) − · · · ± (σ2,1d−2 + σ1d ) ∓ σ1d
i=0

4.2. Chern classes. The cohomology of BGLk = G(k, ∞) is a free polynomial

ring on special Schubert classes called Chern classes ci (1 ≤ i ≤ k):
ci = (−1)i σ1i .
These give a set of characteristic classes. They are uniquely determined by c1 (OPn (1)) =
H and the Whitney sum formula: for any short exact sequence
0 → E → F → G → 0,
P
the total Chern classes c = ci satisfy c(F ) = c(E)c(G). Below we will give a
geometric characterization of ci for globally generated bundles:

Proposition. Let E be a globally generated vector bundle on X, and let σ0 , σ1 , σr−i

be r − i + 1 general global sections. Then Z(σ0 ∧ σ1 ∧ . . . σr−i ) is generically reduced
of codimension i, and it represents ci (E).
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 15

Proof. Suppose that H 0 (E) has dimension m. Let φ : X → G(m − r, m) be

the morphism sending p ∈ X to the kernel of the evaluation map.
0 → K → O⊕m → E → 0.
If U ⊂ H 0 (E) is the span of the chosen sections, then Z(σ0 ∧ σ1 ∧ . . . σr−i ) is
φ−1 (Σk (U )), the locus where Kp meets U non-trivially. 

On any Grassmannian G(k, n), we have a tautological sequence

0 → S → O⊕n → Q → 0.
The proposition above implies that ci (Q) = σi . The Whitney sum formula then
implies that ci (S) = (−1)i σ1i . The tangent bundle to the Grassmannian is
Hom(S, Q) = S ∗ ⊗ Q.
In the case where k = 1, we recover the Euler sequence for Pn :
0 → O(−1) → On+1 → Q → 0
0 → O → O(1)n+1 → TPn → 0
Theorem. (Poincaré-Hopf) Let X be a compact complex manifold.
ctop (X) = χtop (X)
Example. We can now compute the Euler characteristic of a smooth hypersurface
of degree d in Pn .
0 → TX → TPn |X → NX/Pn → 0
Now as a sheaf, NX/Pn is OX (X), so we have
(1 + H)n+1
c(TX ) =
(1 + dH)
   
n+1
= 1 + (n + 1)H + H 2 + . . . (1 − dH + d2 H 2 − . . . )
2
n−1 
X n+1 
n−1
cn−1 (TX ) = (−d)n−1−i HX
i=0
i
n−1  
n+1+i n + 1
X
= (−1) dn−i
i=0
i
For example, when n = 3, d = 4 you get 43 (1) − 42 (4) + 4(6) = 24 (K3 surface).
16 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

5. 2/12/20 - Chern Classes in Geometry

Theorem. (Splitting Principle) Any identity involving the Chern classes of vector
bundles on X holds if and only if it is true for vector bundles which are direct sums
of line bundles on X.

Proof. Suppose that a vector bundle E on X admits a filtration

0 ⊂ F1 ⊂ F2 ⊂ . . . Fr = E,
such that the successive quotients Fi+1 /Fi are line bundles. By the Whitney sum
formula, the Chern classes of E can be computed from a direct sum of line bundles.
Not every E admits such a filtration. To prove the theorem in general, we will
produce a projective morphism π : Y → X such that (1) π ∗ E admits a filtration
as above, and (2) the pull-back map π ∗ : A∗ (X) → A∗ (Y ) is injective. For (1) we
induct on the rank r of E. Consider the projectivization of E ∨ :
π1 : PE ∨ = ProjX (Sym∗ E) → X.
The pull-back π1∗ E ∨ fits into the relative tautological sequence
0 → S → π1∗ E ∨ → Q → 0
0 → Q∨ → π1∗ E → S ∨ → 0
By induction, Q∨ admits a filtration to we are done. For (2), if ζ = c1 (OPE ∨ (1)) is
the relative hyperplane class, then for any α ∈ A∗ (X),
α = π1∗ π1∗ (α) · ζ r−1 . 

As a first application of this splitting principle, we can count lines on hypersurfaces.

Let Xd ⊂ Pn be a hypersurface of degree d.
F (X) = {lines on X} ⊂ G(1, n).
Before we find the class of F (X), we compute its dimension using an incidence
correspondence:
n+d
Ψ = {(X, L) : L ⊂ X} ⊂ PN =( d )−1 × G(1, n).

The second projection pr2 has fiber a linear subspace of PN of codimension d + 1,

that is the kernel of the restriction
H 0 (Pn , O(d)) → H 0 (P1 , O(d)).
Hence, Ψ is smooth of dimension 2n − 2 + N − (d + 1). If the first projection is
dominant, then by properness it will be surjective. Then for a general hypersurface
X its Fano variety of lines will have dimension
dim F (X) = 2n − d − 3.
Harris and Eisenbud show dominance of pr1 (when this number is positive) by pro-
ducing an explicit hypersurface X containing a line L whose normal bundle NL/X
has 2n − d − 3 sections.

Conjecture. (Debarre-de Jong) If d ≤ n, then for every smooth hypersurface,

dim F (X) = 2n − d − 3.
In other words, the word general encompasses all smooth hypersurfaces for small d.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 17

Remark. The term “Fano variety” has two meanings. In this context it means
the variety of lines contained in X. More broadly, a Fano variety is a variety
whose anticanonical line bundle is ample. The meanings do not match; F (X) is
often not a Fano variety in the second sense. To add to the confusion, the hyper-
surfaces appearing in the Debarre-de Jong Conjecture are Fano in the second sense.

We can compute the class of F (X) ∈ Ad+1 (G(1, n)) as the top Chern class of
a vector bundle E on G(1, n) = G(2, n + 1). There is a tautological sub-bundle
S ⊂ O⊕n+1 , and we set E = Symd (S ∨ ). The fiber of E at a line L ' P1 ⊂ Pn is
canonically identified with H 0 (L, O(d)). The equation defining X ⊂ Pn is a degree
d polynomial, an element of H 0 (Pn , O(d)). For any line L, we can restrict the
polynomial to H 0 (L, O(d)), so there is a section σX ∈ H 0 (E). The zero locus of
σX consists of lines contained in X, which is precisely the locus F (X).

To be precise, dualizing and taking the symmetric power of the tautological in-
clusion S ,→ O⊕n+1 , we have
∨⊕n+1
OG(2,n+1) ⊗ H 0 (Pn , O(d)) ' Symd (OG(2,n+1) ) → Symd (S ∨ ) → 0
Taking global sections, we have a map H 0 (Pn , O(d)) → H 0 (E) sending the equation
defining X to σX .

To compute cd+1 (E), we use the Splitting Principle. If S ∨ were isomorphic to

a direct sum of line bundles L ⊕ M , then
Symd (S ∨ ) ' (L⊗d ) ⊕ (L⊗d−1 ⊗ M ) ⊕ · · · ⊕ (Ld−1 ⊗ M ) ⊕ (M ⊗d ).
Using the Whitney sum formula, we know that if c1 (L) = α and c1 (M ) = β,
c(L ⊕ M ) = (1 + α)(1 + β) = 1 + (α + β) + (αβ).
We can rewrite the symmetric expression
d
Y
c(Symd (S ∨ )) = (1 + iα + (d − i)β)
i=0
as a polynomial in (α + β) and αβ, and the result gives an identity relating the
Chern classes of Symd (S ∨ ) with those of S ∨ .

As a first example, we carry this for X ⊂ P3 a cubic surface. We want to compute

c4 (Sym3 (S ∨ )) ∈ A0 (G(1, 3)).
The degree 4 term of the product formula above (for the pretend split case) reads
(3α)(2α + β)(α + 2β)(3β) = 9(αβ)(2α2 + 2β 2 + 5αβ)
= 9(αβ)(2(α + β)2 + αβ)
= 9c2 (S ∨ )(2c1 (S ∨ )2 + c2 (S ∨ ))
= 9σ11 (2σ12 + σ11 )
= 9σ11 (2σ2 + 3σ11 ) = 27σ22 .
This proves that there are 27 lines on a generic cubic surface X3 ⊂ P3 . In fact,
all smooth cubic surfaces contain 27 lines. A similar calculation allows us to count
2875 lines on a generic quintic threefold X5 ⊂ P4 . The reader can readily count
18 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

lines on any hypersurface with 2n − d − 3 = 0. What happens if 2n − d − 3 = 1?

There, F (X) will be a curve (smooth for generic choice of X), and the Chern class
calculation tells us its degree as a multiple of the generator σn−1,n−2 ∈ A1 (G(1, n)).

Question. What is the genus of F (X) when 2n − d − 3 = 1?

To answer this, we need a general version of the adjunction formula. Recall that if
Y is a smooth projective variety, and D ⊂ Y is a divisor, then
KD = (KY + D)|D .
This can be proved by taking c1 of the terms in the normal bundle short exact
sequence:
0 → TD → TY |D → ND/Y → 0,
together with the observation that ND/Y ' O(D)|D . The latter is true because
∨ 2
ND/Y ' ID /ID ' ID ⊗ OY /ID ' OY (−D) ⊗ OD = O(−D)|D .
If instead D ⊂ Y is subvariety of codimension r cut out by a section of a rank
r vector bundle E, then we have ND/Y ' E|D . To prove this, take the Koszul
complex for OY → OD , and restrict it to D. Putting all this together, we deduce
the following formula for the genus g of F (X) in the case where 2n − d − 3 = 1:
 
2g − 2 = deg KF (Y ) = deg KG(1,n) + c1 (Symd (S ∨ )) 

F (X)
 
d ∨ d ∨
= deg KG(1,n) + c1 (Sym (S )) · ctop (Sym (S )).
The splitting principle can be used to prove several identities, which we leave as
exercises. For vector bundles E and F or ranks e and f , respectively:
ck (E ∨ ) = (−1)k ck (E);
c1 (E ⊗ F ) = f c1 (E) + ec1 (F );
k  
X r−k+i
ck (E ⊗ L) = c1 (L)i ck−i (E);
i=0
i
cef (E ⊗ F ) = Resultant(ct (E), ct (F )).
Given any sequence of symmetric analytic functions
f1 (x1 ), f2 (x1 , x2 ), f3 (x1 , x2 , x3 ), . . .
one can define a characteristic class of algebraic vector bundles. If E is a bundle
of rank r, rewrite fr as a power series in the elementary symmetric polynomials
pk , and then replace each pk with ck . For particular choices of fr the resulting
characteristic class may have nice geometric properties. For example:
r
Y
fr = (1 + xi ) (Total Chern class)
i=1
r
X
fr = exi (Chern character)
i=1
r
Y xi
fr = (Todd class).
i=1
1 − e−xi
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 19

6. 2/19/20 - The Riemann-Roch Theorem

The Chern character (defined last time) has the advantage that it is additive for
short exact squences:
0→E→F →G→0

ch(F ) = ch(E) + ch(G)

and also multiplicative for tensor products:

ch(E ⊗ F ) = ch(E)ch(F ).

In other words, we have a ring homomorphism

ch : K0 (X) → A∗ (X) ⊗ Q.

Here K0 (X) is the Grothendieck group of X, generated by vector bundles E on

X modulo the relation that [F ] = [E] + [G] is there is a short exact sequence of
vector bundles 0 → E → F → G → 0. The product structure on K0 (X) is given by
the tensor product. In is natural to ask for a functoriality property for morphisms
f : X → Y of smooth quasi-projective varieties. From basic properties of Chern
classes, the pullback functors fit into a commutative square:

f∗
K0 (Y ) / K0 (X)

ch ch
 ∗ 
A∗ (Y )Q
f
/ A∗ (X)Q .

For covariant (pushfoward) functoriality, we must define a candidate for the proper
pushforward of a vector bundle E. The pushforward of the locally sheaf E is not
necessary locally free. For example, the pushforward OZ via a closed immersion
Z ,→ X will be a torsion sheaf. The correct setting to define the proper pushfor-
ward is on the (larger) abelian category of coherent sheaves on X.

Theorem. For X smooth projective, K0 (X) ' K0 (Coh(X)).

Proof. It suffices to show that every coherent sheaf admits a resolution by lo-
cally free sheaves. This follows from the Hilbert Syzygy Theorem. 

For f∗ to be well-defined on K0 (Coh(X)), we need to use the derived pushforward.

X
f∗ [E] = [Ri f∗ (E)].
i≥0

The higher pushforwards Ri f∗ (E) are explicitly given by sheafifying the presheaf

(U ⊂ Y ) 7→ H i (f −1 (U ), E|f −1 (U ) ).

The fact that f∗ is well-defined on K0 follows from the long exact sequence. The fact
that it is a covariant functor follows from the spectral sequence for compositions.
20 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Now we may hope for a commutative square for all proper f : X → Y :

K0 (X)
f∗
/ K0 (Y )

ch ch
 
A∗ (X)Q
f∗
/ A∗ (Y )Q .

But alas, this is still not so. Consider the case when X is a curve and Y is a point.
h0 (L) − h1 (L) 6= deg(L).
The Riemann-Roch Theorem (for curves) says that
h0 (L) − h1 (L) = deg(L) + 1 − g = deg(L).
The Riemann-Roch Theorem (for surfaces) says that
c1 (L)2 + c1 (L)c1 (TX ) c1 (TX )2 + c2 (TX )
h0 (L) − h1 (L) + h2 (L) = + .
2 12
Notice that this is the codimension 2 part of
c1 (L)2 c1 (TS ) c1 (TX )2 + c2 (TX )
  
1 + c1 (L) + 1+ +
2 2 12
The right hand factor is the 2-truncated part of the Todd class. To see a bit more:
c1 c2 + c2 c1 c2 −c41 + 4c21 c2 + c1 c3 + 3c22 − c4
1+ + 1 + + + ...
2 12 24 720
Lemma. If E has rank r, then
Xr
(−1)i ch(∧i E ∨ ) = cr (E) · Td(E)−1 .
i=0
Proof. This is a straightforward application of the splitting principle:
r r r
X
i i ∨
Y
−αi
Y 1 − e−αi
(−1) ch(∧ E ) = (1 − e ) = (α1 α2 . . . αr ) . 
i=0 i=1 i=1
αi
Theorem. (Grothendieck) Let f : X → Y be a morphism of smooth projective
varieties and E a coherent sheaf on X. Then in A∗ (Y ) we have the equality:
ch(f∗ E) · Td(Y ) = f∗ (ch(E) · Td(X)) .

K0 (X)
f∗
/ K0 (Y )

ch(−)·Td(X) ch(−)·Td(Y )
 
A∗ (X)Q
f∗
/ A∗ (Y )Q .

Proof. It is more convenient to write ch(f∗ E) = f∗ ch(E) · Td(X) · f ∗ Td(Y )−1 .

Step 1: Check that if the identity holds for f : X → Y and g : Y → Z separately,

then it holds for the composite g ◦ f : X → Z.
ch((g ◦ f )∗ E) = ch(g∗ f∗ E)
= g∗ (ch(f∗ E) · Td(Y /Z))
= g∗ (f∗ (ch(E) · Td(X/Y )) · Td(Y /Z))
= g∗ f∗ (ch(E) · Td(X/Z))
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 21

Step 2: Check that the identity holds for X = Pm × Y → Y . This can be reduced
to checking it for Y = pt: the box product K0 (Pm ) ⊗ K0 (Y ) → K0 (Pm × Y ) is
surjective, by induction on m with the localization sequence for K0 . If the outer
and upper squares commute, then lower square commutes:
K0 (Pm ) ⊗ K0 (Y ) / K0 (pt) ⊗ K0 (Y )

 ∼
 
K0 (Pm × Y )
f∗
/ K0 (Y )

ch(−)·Td(Pm )·Td(Y ) ch(−)·Td(Y )

 
A∗ (Pm × Y )
f∗
/ A∗ (Y ).

On the generating set [O(n)], 0 ≤ n ≤ m, for K0 (Pm ), we have:

 
m 0 m m+n
χ(P , O(n)) = H (P , O(n)) = .
n
On the other hand,
" m+1 #
enx

nH H
e · = Resx=0
1 − e−H (1 − e−x )m+1
m
(1 − y)−n
= Resy=0
y m+1
 
m+n
= .
n
Step 3: Check that the identity holds for a closed immersion f : X ,→ Y . This can
be reduced to checking it for Y = P(N ⊕ O) an arbitrary completed bundle on X,
with f the zero section x 7→ [0 : 1], and p : Y → X the structure map. Start by
computing ch(f∗ OX ) using a Koszul resolution. The image of X in P(N ⊕ O) is
cut out by a section of Q, the tautological quotient bundle on Y :
OY → p∗ N ⊕ OY → Q.
This gives us a resolution of f∗ OX by vector bundles:
0 → ∧r Q∨ → · · · → ∧2 Q∨ → Q∨ → OY → f∗ OX → 0.
By the additivity of the Chern character and the earlier Lemma,
r
X
ch(f∗ OX ) = (−1)i ch(∧i Q∨ ) = cr (Q) · Td(Q)−1 .
i=0
Then using the fact that f ∗ Q = NX/Y = N we obtain
cr (Q) · Td(Q)−1 = f∗ f ∗ Td(Q)−1 = f∗ (Td(N )−1 ) = f∗ (Td(X/Y )).

The proof for ch(f∗ E) is similar; the Koszul complex above tensored with p∗ E gives
a resolution of f∗ E and the rest follows.

Putting everything together, we use the fact that any morphism f : X → Y can be
factored X → Γf ⊂ Pm × Y as a closed immersion followed by a simple projection
with fiber Pm . The deformation to the normal cone trick (next lecture) allows us
to reduce the statement for closed embeddings to the setting of Step 3. 
22 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Corollary. The Chern character gives an isomorphism of rings:

ch : K0 (X) ⊗ Q → A∗ (X) ⊗ Q.
Proof. There is a natural filtration F i on K0 (X) given by the codimension of
the support of a sheaf being ≥ i, and A∗ (X) has a similar filtration (it is already
graded). By the GRR formula, for a k-codimensional subvariety Z ⊂ X we have
ch[OZ ] = iZ∗ (1 + α) ∈ [Z] + F k+1 A∗ (X).
Hence, the assignment [Z] 7→ [OZ ] defines an isomorphism
A∗ (X) → gr∗ K0 (X).
Since it is an isomorphism at the level of graded Q-algebras, it is an isomorphism
at the level of filtered Q-algebras too. 

Example. When X = Pn the Chern character isomorphism sends x = [OPn (1)] to

eh ∈ A∗ (Pn ):
K0 (Pn ) ⊗ Q ' Q[x]/(x − 1)n+1 → Q[h]/(hn+1 ) ' A∗ (Pn ) ⊗ Q
Notice that x is invertible in the quotient because the geometric series is finite.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 23

7. 2/24/20 - Deformation to the Normal Cone

Today we will take a step back and go through some foundational material, culmi-
nating in two key statements (starred) that were used in the proof of GRR last time.

Definition. Given a vector bundle E → X, its projectivization is given by

p
PE = Proj (Sym∗ E ∨ ) → X.
As such, it has a line bundle OPE (1) which restricts to O(1) on each fiber. The
relative tautological sequence for PE → X reads
0 → S → p∗ E → Q → 0,
and S ' OPE (−1), which can be checked locally by trivializing the bundle.

Universal Property. Given a scheme Y with a morphism π : Y → X, there

is bijection between commutative triangles

Y
f
/ PE

π p
}
X
and sub-line bundles L ⊂ π ∗ E over Y . To go from a commutative triangle to a line
bundle, simply pull back the tautological line bundle S via f :
f ∗ S ⊂ f ∗ p∗ E = π ∗ E
To go the other direction, cover X by open sets trivializing both E and L, and
define f |U using the universal property for projective space.

Proposition. If E (resp. L) is a vector (resp. line) bundle on X, then setting

E 0 = E ⊗ L, we have an isomorphism
PE ' PE 0 .
Proof. Define a morphism f : PE 0 → PE via the universal property: tensoring
OPE 0 (−1) ⊂ p0∗ E 0 ' p0∗ E ⊗ p0∗ L with p0∗ L−1 , we get
OPE 0 (−1) ⊗ p0∗ L−1 ⊂ p0∗ E.
We can define a similar morphism g : PE → PE 0 . Their composition is the identity,
as can be seen by pulling back the line bundles through the composition. 

Remark for experts. As a consequence of the construction above,

f ∗ OPE (1) ' OPE 0 (1) ⊗ p0∗ L.
So if L was ample, then OPE 0 (1) is less ample than OPE (1) via the isomorphism f .
This may seem backwards, and indeed it reflects the different convention between
Fulton and Grothendieck in the definition of projectivization. To make things con-
sistent, an ample vector bundle is one for which OPE ∨ (1) is ample.

In what follows, ζ = c1 (OPE (1)) is the relative hyperplane class, which depends
on the choice of E, not just on the projective bundle PE. We will assume that E
has rank r + 1 for now so that PE has fiber Pr .
24 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Theorem. The Chow group of PE is given by

r
M
A(PE) ' p∗ A(X) · ζ i .
i=0
∗
NB: We will sometimes suppress p in the notation. This is not so egregious be-
cause since p is flat, p∗ [A] = [p−1 A] always.

Proof. Define ψ : A(PE) → A(X)⊕r+1 by

M
ψ(β) = p∗ (ζ r−i · β).
i
⊕r+1
Define ϕ : A(X) → A(PE) by
X
ϕ((αi )ri=0 ) = ζ i · p∗ αi .
i

Now using the fact that (for dimensional reasons)

(
i ∗ i α if i = r
p∗ (ζ p α) = p∗ (ζ )α = ,
0 if i < r
we see that ψ ◦ ϕ is upper triangular with 1’s along the diagonal, so ψ is injective.
To show surjectivity, we use induction to write a general cycle as a sum of cycles
of the form ζ i · p∗ α:

Lemma. Given a subvariety Z ⊂ PE of dimension k, let W = p(Z) ⊂ X have

dimension l ≤ k, so the general fiber of p|Z : Z → W has dimension k − l. We can
always write
X
Z ∼ Z0 + Bj ,
where [Z 0 ] = ζ r−k+l [W ] and dim(p(Bj )) < l.

Proof. We start with the case of projective space Pr (so X = pt) and dim(Z) = n.
It’s possible to find coordinates xi on Pr such that (x0 = x1 = · · · = xn = 0) is
disjoint from Z. Define
 
I 0
gt = n+1 ∈ P GL(r + 1).
0 tIr−n
The flat limit of gt (Z) as t = 0 will be the nonreduced cycle deg(Z) · Pn , so
Z ∼ ζ r−n deg(Z).
The case of PE is similar. If L is ample on X then for N  0, E ∨ ⊗ L⊗N is globally
generated. Replacing E with E 0 = E ⊗ L−⊗N does not affect the projectivization.
Fix a point x ∈ W ⊂ X. For a general choice of global sections τ0 , τ1 , . . . , τr ,
(1) (τ0 )x , (τ1 )x , . . . , (τn )x forms a basis for the fiber Ex∨ .
(2) (τ0 = τ1 = · · · = τk−l = 0) is disjoint from Zx .
In fact, both conditions are Zariski open in X, so they are true on some U ⊂ X.
Using condition (1),
PE|U = U × Pr ,
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 25

so we can perform the same construction as in the point case over U .

 
Ik−l+1 0
gt = ∈ P GL(r + 1).
0 tIr−k+l
Set Zt = gt (ZU ). The flat limit Z0 will have class d · [Pk−l × U ] over U , plus some
components Bi over W r W ∩ U . 

Theorem. As a ring,
A∗ (PE) ' A∗ (X)[ζ]/(ζ r+1 + c1 (E)ζ r + · · · + cr+1 (E))
Proof. It suffices to prove that the monic polynomial relation above is satisfied.
There can be no other relations by our description of the group structure. From
0 → S → p∗ E → Q → 0,
we have c(S)c(Q) = c(p∗ E), so c(Q) = c(p∗ E)(1 + ζ + ζ 2 + . . . ). Since rk(Q) = r,
0 = cr+1 (Q) = ζ r+1 + c1 (E)ζ r + · · · + cr+1 (E). 

Remark. Fulton uses this result as the definition of (algebraic) Chern classes.
⊕r+1
Remark. Applying the theorem to the trivial bundle OX , we obtain
A∗ (X × Pr ) ' A∗ (X) ⊗ A∗ (Pr ).
This is called a Chow-Künneth formula. No such formula holds for general products.

For the rest of the lecture, we will assume that E has rank r because the main
player will be P(E ⊕ OX ), which has fiber Pr .

Proposition. If F ⊂ E is a sub-bundle of rank s, then PF ⊂ PE. The class

of PF in CH r−s (PE) is given by
[PF ] = ζ r−s + γ1 ζ r−s−1 + · · · + γr−s ,
where γi = ci (E/F ).

Proof. Consider the composition S → p∗ E → p∗ (E/F ). The composition vanishes

identically over points of PF . This means that we can compute [PF ] as the top
Chern class of the bundle
Hom(S, p∗ (E/F )) ' OPE (1) ⊗ p∗ (E/F ).
The formula now follows easily from the Splitting Principle. 

We will refer to P(E ⊕ OX ) as the completion of E. It has a zero section

j : X → P(E ⊕ OX )
whose image is P(0 ⊕ OX ), and it contains a hyperplane at ∞, which is P(E ⊕ 0).
These two cycles are clearly disjoint, and by the previous proposition their classes
in A∗ (P(E ⊕ OX )) are given by
[POX ] = ζ r + c1 (E)ζ r−1 + · · · + cr (E)
[PE] = ζ
26 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Proposition. The zero section POX ⊂ P(E ⊕ OX ) has self-intersection j∗ cr (E).

Proof. To compute, [POX ] · [POX ] we write one of them in terms of ζ and leave
the other alone. Since zero and ∞ are disjoint, [POX ] · ζ = 0.
[POX ]2 = [POX ] · (ζ r + c1 (E)ζ r−1 + · · · + cr (E)
= [POX ] · p∗ cr (E)
= j∗ (cr (E)). 
Proposition. (?) More generally, for any cycle α ∈ A∗ (X), j ∗ j∗ (α) = α · cr (E).

Proof. For any β ∈ A∗ (P(E ⊕ OX )), by the Moving Lemma j ∗ β = p∗ (β · [POX ]).
j ∗ j∗ (α) = p∗ (j∗ (α) · [POX ])
= p∗ (p∗ α · [POX ] · [POX ])
= p∗ (p∗ α · j∗ cr (E))
= α · p∗ j∗ cr (E)
= α · cr (E). 
Definition. Let Z ⊂ X a closed subscheme defined by an ideal sheaf I. The
blow-up of X along Z is a new scheme given by Proj of the Rees algebra:


BlZ X := Proj OX ⊕ I ⊕ I 2 ⊕ . . . → X.
If we restrict  to Z ⊂ X, the fibered product is obtained by tensoring the Rees
algebra with OX /I:
EZ X := Proj OX /I ⊕ I/I 2 ⊕ I 2 /I 3 ⊕ . . .

If Z is smooth (or more generally l.c.i.), then I k /I k+1 ' Symk NZ/X
∨
. In that case,
 
EZ X ' Proj Sym∗ NZ/X ∨
= PNZ/X .
By a dimension count, we find that EZ X ⊂ BlZ X is a divisor. In fact, the blow up
is the minimal way to replace Z with a Cartier divisor.

Lemma. If Y ⊂ X is a subvariety, its proper transform in BlZ X is:

−1 (Y r (Y ∩ Z)) ' BlY ∩Z Y.
The proof is technical, so we omit it. Instead, we will finish by introducing the
“deformation to the normal cone” (?). Consider X × P1 , viewed as a family over
P1 , and blow up Z × {0} ⊂ X × P1 to get M → P1 . The exceptional divisor E is
P(NZ×{0}/X×P1 ) ' P(NZ/X ⊕ OZ ),
the completed normal bundle. The proper transform of X × {0} is BlZ X, and it
meets E along EZ X, which coincides with the hyperplane at ∞ in P(NZ/X ⊕ OZ ).
The proper transform of Z × P1 is simply Z × P1 because blowing up a Cartier
divisor does nothing, but it meets E along the zero section POZ .

We have a flat family M → P1 . Over t 6= 0, Z ⊂ X is embedded somehow.

Over t = 0, Z ⊂ P(NZ/X ⊕ OZ ) is embedded as the zero section in its completed
normal bundle. Over t = 0, there is a Zariski neighborhood of Z isomorphic NZ/X .
This is analogous to the tubular neighborhood theorem in differential geometry.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 27

8. 2/26/20 - Gysin Pullback and Excess Intersection

The deformation to the normal cone construction produces a family of varieties
M = BlZ×{0} X × P1 → P1
with general fiber X, and with special fiber a reducible, normal crossing variety
M0 = BlZ X ∪ P(NZ/X ⊕ OZ )
EZ X

Here EZ X ' PNZ/X is the hyperplane at ∞ in P(NZ/X ⊕ OZ ). Recall that

A∗ (P(N ⊕ O)) ' A∗ (Z)[ζ]/(ζ r+1 + c1 (N )ζ + · · · + cr (N )ζ).
The relation factors as ζ(ζ r + ζ r−1 c1 (N ) + · · · + cr (N )) = 0 which corresponds to
[PO] · [PN ] = 0.
The localization sequence for Chow groups gives
A∗ (PN ) → A∗ (P(N ⊕ O)) → A∗ (N ) → 0,
where the first map sends α 7→ ζ · α. This implies that the ideal (ζ) vanishes in
A∗ (N ), so we have
A∗ (N ) ' A∗ (Z).
To be more explicit, the pull-back π ∗ : A∗ (Z) → A∗ (N ) is an isomorphism. We can
give the inverse map on a cycle β ∈ A∗ (N ) by taking its closure β ∈ A∗ (P(N ⊕

O)),
and then pushing forward the intersection with the zero section: p∗ β · [PO] .

Fulton and MacPherson had the idea to specialize cycles in X to cycles in NZ/X in
M ◦ := M r BlZ X → P1 .
At the level of sets, a subvariety B ⊂ X specializes to the normal cone
CB∩Z/B ⊂ NZ/X ,
and this descends to a map on Chow groups:

Theorem. The assignment [B] 7→ [CB∩Z/B ] gives a well defined map

σ : A∗ (X) → A∗ (NZ/X )
Proof. Consider the localization sequence again:

A∗+1 (NZ/X )
i∗
/ A∗+1 (M ◦ ) / A∗+1 (X × A1 ) /0
O
i∗ ∼
 w
A∗ (NZ/X ) A∗ (X).
The dotted arrow exists because for i : NZ/X ,→ M ◦ the inclusion, i∗ i∗ = 0. 

This allows us to give a definition for Gysin pullback without appealing to the
Moving Lemma. If j : Z → X is the inclusion, then
j ∗ := (π ∗ )−1 ◦ σ.
Remark. Fulton uses this construction to define the intersection product.

With a few formal observations, we can get a lot of mileage out of this. Recall
28 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

that the zero section PO ⊂ P(N ⊕ O) is cut out by a section of the tautological
quotient bundle Q. If β ∈ Ak (N ), then we have
(π ∗ )−1 β = p∗ β · [PO]




= p∗ β · cr (Q)


= p∗ β · c(Q) k−r
= p∗ β · c(p∗ (N ⊕ O)) · (1 + ζ + ζ 2 + . . . ) k−r

= c(N ) · p∗ β · (1 + ζ + ζ 2 + . . . ) k−r



The factor p∗ β · (1 + ζ + ζ 2 + . . . ) is an example of a total Segre class.

Definition. Let E → X be a vector bundle of rank r. The total Segre class

s(E) := p∗ (1 + ζ + ζ 2 + . . . ) ∈ A∗ (X),
where p : P(E ⊕ OX ) → X. The graded parts of s(E) are recovered by
si (E) = p∗ (ζ r+i )
for dimensional reasons. NB: we could have used P(E) instead of P(E ⊕ O).

Proposition. s(E) · c(E) = 1.

Proof. Apply p∗ to the identity c(Q) = c(p∗ E)(1 + ζ + ζ 2 + . . . ) and use the
push-pull formula. 

The total Segre class contains the same information as the total Chern class, but
it has the advantage of being definable using only Chow groups. For thisL reason,
the definition can be generalized to cones. Any graded OX -algebra S = i≥0 S i
such that OX  S 0 and S generated by S 1 gives rise to a cone scheme C =
Spec (S) → X. The Segre classes of C are defined in the analogous way, using
p : P(C ⊕ OX ) → X and ζ = c1 (OP(C⊕O) (1)).

s(C) = p∗ 1 + ζ + ζ 2 + . . . ∈ A∗ (X).

Recall that in the specialization construction above, β = [CB∩Z/B ] ∈ A∗ (NZ/X )

and β = [P(CB∩Z/B ⊕ OZ )] ∈ A∗ (P(NZ/X ⊕ OZ )). The hyperplane at ∞ (class ζ)
restricts well, so the formula above becomes
j ∗ [B] = (π ∗ )−1 [CB∩Z/B ] = c(NZ/X ) · g∗ s(CB∩Z/B ) k−r ∈ A∗ (Z),
 

where g : B ∩ Z → Z is the inclusion. By the push-pull formula, this can be refined:

[B ∩ Z] = g ∗ c(NZ/X ) · s(CB∩Z/B ) k−r ∈ A∗ (B ∩ Z).
 

This is referred to as the excess intersection formula. It allows us to find an inter-

section product without moving the cycles, and the answer is supported on their
set-theoretic intersection. If Z and B are generically transverse, then the (k − r)
piece will be 1, and their is no need to consider normal cones.

Remark. There is an asymmetry between Z and B. One of them must have

 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 29

a normal bundle (in our case Z) which is equivalent to it being regularly embedded,
that is cut out by a regular sequence of length r. In that case
Symk (IZ /IZ2 ) ' IZk /IZk+1 .
If Z, B, and B ∩ Z are all smooth, then using the normal bundle sequence
0 → NB∩Z/B → NB∩Z/X → NB/X |B∩Z → 0
we can rewrite the excess intersection formula more symmetrically:
c(NZ/X |B∩Z ) · c(NB/X |B∩Z )
 
[B ∩ Z] = .
c(NB∩Z/X ) k−r

On a smooth variety X, the diagonal ∆ : X → X × X is a regular embedding. This

is enough to define an intersection product for arbitrary cycles A, B; apply the
Gysin pullback ∆∗ to A × B ⊂ X × X. Note: X is not necessarily quasi-projective!

The most general form of the excess intersection formula is stated for Z → X
a regular embedding and B → X an arbitrary morphism. The fibered product
W = B ×X Z fits into a Cartesian square

W
i /B
g f
 
Z
j
/ X.
Theorem. For any class α ∈ Ak (Z), we have
f ∗ j∗ (α) = i∗ g ∗ (α · c(NZ/X )) · s(CW/B ) k+dim(B)−dim(X) .

Corollary. Suppose that φ : X → Y is generically finite. If y ∈ Y is a special

point for which F = φ−1 (y) is not finite, then
deg(ϕ) = [s(NF/X )]0 .
2 2
For example, for Bl0 P → P the normal bundle to the exceptional fiber is OP1 (−1).
As a more exotic example, Donagi-Smith computed the degree of the Prym map
R6 → A5 by considering a special fiber, and the answer was 27!

Lastly, let us return to the reduction step in the proof of the Grothendieck-Riemann-
Roch theorem. To relate the case of an arbitrary closed embedding to the zero
section of the completed normal bundle, we cited the deformation to the normal
cone trick. The key fact is that M → P1 is flat, by the following “Miracle Flatness:”

Theorem. If q : M → N is a surjective morphism of schemes, with M Cohen-

Macaulay and N regular, such that for all n ∈ N , dim(M ) = dim(N )+dim(q −1 (n)),
then q is flat.

Let E be a vector bundle on Z, and pull it back to Ẽ on Z × P1 . Via the proper

transform, we have an embedding ι : Z × P1 → M . Take a resolution G• of ι∗ M .
Since M → P1 is flat, the Gi are flat over P1 , so the restrictions of G• to M0 and
to M∞ remain exact. This observation combined with repeated application of the
projection formula completes the desired reduction.
30 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

9. 3/2/20 - The Five Conics Problem

Question. How many conics are tangent to five general conics?

Let C1 , . . . , C5 be general conics in P2 . Let Zi ⊂ P5 be the locus of conics tangent

to Ci . Consider the incidence correspondence
Ωi = {(C, p) : C tangent to Ci at p} ⊂ P5 × Ci .
The fiber of pr2 at p ∈ Ci is a codimension 2 linear subspace of P5 - defined by
the vanishing of the first two coefficients in the Taylor expansion of the degree 4
polynomial C|Ci at p. Since pr1 is finite, we see that Zi is a hypersurface. To find
its degree, intersect it with a line in P5 , that is a pencil of conics. Restricting the
pencil to Ci ' P1 , we get a basepoint free linear series of degree 4 divisors on P1 .
That is the same a degree 4 map P1 → P1 , which by Riemann-Hurwitz must have
6 branch points.

Since each Zi is a sextic hypersurface, and we intersecting them inside P5 , it is

tempting to conclude that the answer is
deg[Zi ]5 = 65 = 7776.
But that is incorrect. To see why, observe that any double line is “tangent” to Ci ,
so each of the Zi contains the Veronese surface
S = v2 (P2 ) ⊂ P5 .
This means that set-theoretically
5
\
Zi = S ∪ Γ,
i=1

where Γ is the finite set that we want to count. The correct answer is 65 minus the
excess contribution supported on S ' P2 . We will use h = c1 (OP2 (1)).
5
Y
excess = deg c(NZi /P5 |S ) · s(CT /P5 ).
i=1

Here T is the component of the scheme-theoretic intersection supported on S. First,

5
Y
c(NZi /P5 |S ) = (1 + 12h)5 = 1 + 60h + 1440h2 .
i=1

On the other hand,

c(TS ) 1 + 3h + 3h2
s(NS/P5 ) = = = 1 − 9h + 51h2
c(TP5 |S ) 1 + 12h + 30h2
The ideal of T is (IS )2 essentially because each Zi contains S with multiplicity 2.
This means you can just replace ζ with 2ζ in the definition of Segre class. This has
the effect of multiplying sk by 23+k in this case:
s(NT /P5 ) = 8 − 144h + 1632h2 .
Hence, the excess contribution is 1440 · 8 − 60 · 144 + 1632 = 4512, so we have
|Γ| = 7776 − 4512 = 3264.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 31

Question. How many conics are tangent to five general lines?

By the Riemann-Hurwitz argument, Zi is a quadric, so the computation becomes

25 − deg(1 + 4h)5 (1 − 9h + 51h2 ) = 25 − deg(1 + 20h + 160h2 )(1 − 9h + 51h2 )
= 32 − 31
=1
But this is not a surprise because the dual of a conic is a conic, and 5 general
points lie on a unique conic. In general, the dual of a smooth hypersurface X ⊂ Pn
has degree d(d−1)n−1 , by pulling back hyperplanes via the Gauss map G : X → Pn∗ .

A second approach to the five conics problem involves blowing up the Veronese
surface S to tease apart the scheme-theoretic behavior there. Luckily for us,
M ' BlS P5
has a nice moduli theoretic description. It happens to be the space of complete
conics. This is the closure in P5 × P5∗ of the locus U consisting of (C, C ∗ ) where C
is a smooth conic and C ∗ its dual. It also happens to be isomorphic to the coarse
moduli scheme of the Kontsevich stack of stable maps:
M0 (P2 , 2h) = {f : C → P2 | C is conn. nodal genus 0, f∗ [C] = 2h, |Aut(f )| < ∞}.
The fibers of the exceptional E → S are P2 ' Sym2 P1 , encoding the branch
points of the double cover C → P1 . From these models, it is clear that the proper
transforms Zei intersect only in a finite set away from the exceptional divisor:
5
\
Z
ei = Γ.
i=1

So it suffices to compute the intersection product inside the blow-up BlS P5 .

Proposition. Let  : BlZ X → X be the blow up of a regularly embedded subva-

riety Z. Then we have
A∗ (BlZ X) ' Ã∗ (EZ/X ) ⊕ A∗ (X),
where Ã∗ (E) ⊂ A∗ (E ' PN ) consists of cycles ζ i p∗ α where i < codim(Z/X) − 1.

Proof. The proper pushforward ∗ : A∗ (BlZ X) → A∗ (X) is surjective, and the

kernel is precisely Ã∗ (EZ/X ). The pullback ∗ provides a splitting. 

Proposition. The ring structure on A∗ (BlZ X) is given in terms of generators

in ∗ A∗ (X) and j∗ A∗ (E) by the following formulas:
∗ (α) · ∗ (α) = ∗ (α · α0 );
∗ (α) · j∗ (β) = j∗ (β · p∗ i∗Z (α));
j∗ (β) · j∗ (β 0 ) = −j∗ (β · β 0 · ζ).
Proof. The first follows from the fact that ∗ is a ring homomorphism. The second
follows from the push-pull formula. The third follows from the fact that
NE/BlZ X ' OPN (−1). 
32 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

The class of Zei in the blow up is 6H − 2E = 2(3H − E), because Zi vanishes with
multiplicity 2 along S. To prove this, note that a general pencil of conics containing
a double line meets Zi at only 4 other points, by the Riemann-Hurwitz argument.
(3H − E)5 = 243H 5 − 5 · 81H 4 E + 10 · 27H 3 E 2 − 10 · 9H 2 E 3 + 5 · 3HE 4 − E 5
= 243[pt] − 90 · j∗ (2h)2 ζ 2 + 15 · j∗ (2h)(−ζ 3 ) − j∗ (ζ 4 )

= 243[pt] − 360[pt] + 30j∗ h(ζ 2 9h + ζ30h2 ) + j∗ (ζ 2 9h + ζ30h2 )ζ

= (243 − 360 + 270 − 81 + 30)[pt]

= 102[pt].
Recall that in A∗ (PN ), we have ζ 2 h2 = [pt], ζ 3 = −ζ 2 c1 (N ) − ζc2 (N ), and we are
repeatedly using the fact that c(NS/P5 ) = 1 + 9h + 30h2 . To conclude,
e 5 = 25 · 102 = 3264.
deg[Z]
Two Generalizations. A quadric ϕ on PV ' Pn can be viewed as an element of
Sym2 (V ∨ ) or as a symmetric transformation V → V ∨ . One can associated to ϕ a
sequence of symmetric transformations
ϕi : ∧i V → ∧i V ∨ ' (∧i V )∨ .
The space M of complete quadrics on Pn is defined to be the closure of ϕ 7→
(ϕ1 , . . . , ϕn ) for ϕ non-degenerate.
Yn
P Sym2 (∧i V ∨ ) .


M⊂
i=1
M can alternatively be obtained by successively blowing up the strata of degenerate
quadrics, starting with the most degenerate one. The result is a Mori Dream Space.

On the other hand, a degree d rational curve in P2 can be viewed as a morphism

f : P1 → P2 , where two morphisms f and g are equivalent if they differ by a
reparametrization. We can define a pre-stack which associates to a scheme T , the
groupoid of diagrams: F / 2
C P


T
such that F∗ [Ct ] = dh. This pre-stack turns out to be a smooth algebraic stack. To
make it proper, we allow the curves Ct to be nodal, and require that any contracted
component have ≥ 3 nodes. This makes the isotropy groups finite. A coarse space
for a stack X is a morphism m : X → X to an algebraic space such that




m k :X k →X k
is a bijection, and any morphism to an algebraic space factors through m.

Theorem. (Keel-Mori) Any stack with finite inertia has a coarse moduli space.

The map m looks like [Tx X/G] → Tx X/G formally locally around a smooth point
x ∈ X with isotropy group G. By a theorem of Chevalley-Shephard-Todd, if G is
generated by pseudo-reflections, i.e. elements fixing a hyperplane, then the quotient
is smooth (and conversely). This is true when d = 2, but fails when d ≥ 3.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 33

10. 3/4/20 - Intersection Theory on Singular Spaces

Question. Can we do intersection theory on singular spaces? What about smooth
algebraic stacks?

On a singular quadric surface, Q ⊂ P3 , a line L through the vertex p fails the

Moving Lemma. Indeed, any curve rationally equivalent to L must pass through p
because curves which do not have even degree in P3 . Note that H ∩ C = 2 deg(π),
where π is the linear projection from p. To have a well-defined intersection product,
(2L)2 = 2 forces us to set
1
L2 = .
2
Mumford defined an intersection product on A1 (S) ⊗ Q for S a normal surface.
Take a resolution  : S̃ → S. For a curve A ⊂ S, we use à to denote its proper
transform in S̃.

Lemma. There is a unique Q-curve class A0 in S̃ supported on the exceptional

locus such that à · E + A0 · E = 0 for each component E of the exceptional locus.

Proof. Let E1 , . . . , Er be the components of the exceptional. We want to solve

Xr
λj Ej · Ei = −Ã · Ei
j=1

for all i. This follows from the Hodge Index Theorem, which says that the inter-
section matrix of the Ei is negative definite. 

Mumford’s intersection product is defined by

A · B = (Ã + A0 ) · (B̃ + B 0 ) ∈ Q.
In the case of the line on the singular quadric S, we get A0 = 12 E, since E 2 = −2:
1 1 1
L2 = (Ã + A0 )2 = Ã2 + ÃE + E 2 = 0 + 1 − = .
4 2 2
Things are a bit worse for the singular quadric threefold X. All lines in X are
rationally equivalent to a line through the vertex, but on the other hand we have
Li · Hj = δij pt. This implies that pt = 0 which is problematic.

Intersections with Q-Cartier divisors in A∗ (X) ⊗ Q can be done in general.

Definition. A divisor D is Q-Cartier if for some m ∈ Z, mD is Cartier.

For such D, we can define D · Z for any cycle [Z] ∈ Ak (X) by pushing forward
1
D|Z := c1 (O(mD)|Z ) .
m
This recovers Mumford’s construction, since a Cartier divisor D satisfies
∗ D · E = 0.
More generally, Chern classes of vector bundles can be defined and pulled back via
morphisms of schemes. The correct setting for a ring structure on the full Chow
group is that of smooth algebraic stacks.
34 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

A category fibered in groupoids (CFG) p : X → Sch/k is the same as a 2-functor

from (Sch/k)op → Grpd, sending T to p−1 (T ), the subcategory of X whose objects
lie over T and whose morphisms lie over idT . Any scheme T has an associated CFG
given by its functor of points, which is actually valued in Set, and we will abuse
notation slightly by referring to this CFG as T also.

Lemma. (2-Yoneda) For T ∈ Sch/k, there is an equivalence of groupoids between

Hom(T, X) ' p−1 (T )
For this reason, we often write X(T ) for p−1 (T ).

Definition. An algebraic (or Artin) stack X over k is a CFG p : X → Sch/k,

and a topos on Sch/k (usually fppf), such that
(1) For x, y ∈ X(U ), the presheaf Isom(x, y)(V ) := Isom(x|V , y|V ) is a sheaf.
(2) Every descent datum is effective, that is for every collection of objects
xi ∈ X(Ui ) with isomorphisms fij : xi |Uij → xj |Uij satisfying the cocycle
condition, there is an object x ∈ X(U ) and isomorphisms fi : x|Ui → xi .
(3) The diagonal ∆ : X → X×X is representable, separated, and quasi-compact.
(4) There exists a smooth epimorphism Y → X from a scheme (called an atlas).
The typical examples one should have in mind for Artin stacks are coherent sheaves
on a scheme - Coh(X), curves of arithmetic genus g - Mg , and BG = [pt/G] for G
an algebraic group. A motivating example is BGm which classifies line bundles. It
is easy to find two line bundles which are isomorphic (both trivial) on every subset
in a covering Ui , but non-isomorphic on U . This highlights the subtle necessity of
the descent axiom.

Axioms (1) and (2) assert that X is a 2-sheaf on the site Sch/k; isomorphisms
glue, and objects 2-glue. Axiom (3) implies that any morphism from a scheme to X
is representable, via the Magic Square (below), which is Cartesian in any category
with fibered products. This allows us to describe maps from schemes to X.
X ×Z Y / X ×Y

 
Z
∆ / Z × Z.
Representable morphisms between stacks can be modified by any adjective that
modifies morphisms of schemes, is local on both the domain and target, and is
preserved under base change (e.g. flat, smooth, finite, étale). Hence, Axiom (4)
only makes sense given Axiom (3).

Definition. An algebraic stack is Deligne-Mumford (DM) if it satisfies the sheaf

conditions (1) and (2) above with respect to the étale topos, and furthermore the
following equivalent conditions are satisfied:
(1) The diagonal ∆ : X → X × X is unramified (and quasi-finite).
(2) There exists an étale epimorphism Y → X from a scheme (called an atlas).
The typical examples one should have in mind for Deligne-Mumford stacks are
smooth curves - Mg , stable curves - Mg , and BG = [pt/G] for G a finite group.
Any scheme S is a stack (via its functor of points).
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 35

NB: The hypothesis on ∆ for a DM stack is slightly weaker than the hypothesis of
the Keel-Mori theorem (existence of coarse spaces) which was finiteness of inertia
stack IX → X, defined as IX = X × X, but this distinction rarely matters.
X×X
Objects of IX are pairs of isomorphisms from x → y. This category is equivalent
to one whose objects are pairs (x, φ) where φ is an automorphism of x.

Absolute properties of X like being reduced or finite type are defined from the
atlas scheme Y . Smoothness and properness can also be defined via lifting criteria.

Definition. A stack X is proper if for every discrete valuation ring R with fraction
field K, any morphism Spec K → X extends to Spec R → X.

Definition. A stack X is smooth if for every Artinian ring A with square-zero

ideal I, any morphism Spec A/I → X extends to Spec A → X.

Theorem. (Vistoli) Deligne-Mumford stacks have Chow groups and, if smooth,

an intersection product defined over Q. The product can be defined directly on the
coarse space. This gives a Q-intersection theory for all Q-varieties, i.e. varieties
which are étale locally the quotient of a smooth variety by a finite group.

Theorem. (Kresch) Artin stacks have Chow groups. If they are smooth and
(1) Deligne-Mumford, or (2) stratified by global quotients stacks, then there is an
intersection product defined over Z.

Vistoli uses the naive definition of Chow groups:

Z(X) := Z{integral closed substacks of X}
A closed substack is simply a representable morphism Y → X which is a closed
embedding. A rational function on an integral stack W is a morphism from an
open substack to A1k . This leads to the notion of rational equivalence:
M
Rat(X) := k ∗ (W);
W
A∗ (X) := Z(X)/∂Rat(X).
A dominant morphism of integral stacks has a well-defined degree valued in Q, de-
fined in terms of extensions of rational function fields on étale atlases. For example,
1
deg([pt/G] → pt) =
|G|
Representable morphisms have integral degree. This leads to a definition of proper
pushforward. The main technical challenge is to define the Gysin pullback without
using deformation to the normal cone. Vistoli defines pull-backs through finite type,
unramified maps (called local embeddings). The diagonal morphism X → X × X is
a local embedding.

Kresch uses an idea of Graham-Eddidin (next lecture), who defined equivariant

Chow groups AG ∗ (X) = A∗ [X/G] for any G-variety X by considering (X × V )/G
for sufficiently large G-representations V . The G-representations are replaced with
arbitrary vector bundles on the Artin stack X.
36 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Definition. A quasi-coherent sheaf on a stack X (with structure functor p : X →

Sch/k) consists of the following data. For every object x ∈ X, a sheaf ξ(x) on
the scheme p(x), and for every morphism f : x → y over F : p(x) → p(y), an
isomorphism F ∗ ξ(x) → ξ(y), satisfying the obvious compatibility for compositions.

Example. A vector bundle on BG is the same as a G-representation. Given a

vector bundle V on BG, consider its value on the trivial G-bundle over a point.
This is a vector space V , with an action of G coming from automorphisms of the
trivial G-bundle. Conversely, given a G-representation V , we can construct a vector
bundle on B for every principal G-bundle P → B by forming
(V × P )/G.
Let BX denote the category of vector bundles E on X with the partially ordering
E ≤ F if there exists F  E. Kresch defines:
Âk (X) := lim Ak+rk(E) (E).
→ BX

The natural morphism Ak (X) → Âk (X) is an isomorphism when X is a scheme.

Next, we take a further enlargement to obtain AKr
k (X), which satisfies all the axioms.
The issue is that while X may have no non-trivial vector bundles, it may contain a
global quotient stack, which has many. Let UX be the set of projective morphisms
f : Y → X, partially ordered by inclusions of components.
AKr
k (X) = lim Âk (Y)/B̂k (Y),
→ UX

where B̂k (Y) is a union over all stacks T with pairs of projective morphisms p1 , p2 :
T → Y such that f ◦ p1 ' f ◦ p2 of the following abelian group:
{p1∗ β1 − p2∗ β2 | (β1 , β2 ) ∈ Âpk1 (T) ⊕ Âpk2 (T) satisfies ιp1 (β1 ) = ιp2 (β2 )}.
To make sense of this, we need to define
Âpk (T) := lim Ak+rk(E) (p∗ E); ιp : Âpk (T) → Âk (T).
→ BY

There is a natural sequence of morphisms

Ak (X) → Âk (X) → AKr
k (X).
The last one is an isomorphism when X a global quotient. The composite is an
isomorphism over Q when X is smooth and Deligne-Mumford.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 37

11. 3/9/20 - Quotient Stacks

Let G be an algebraic group. Recall that a coherent sheaf on BG = [pt/G] is the
same as a G-representation. We can define K0 (BG) as the Grothendieck group
of the abelian category of G-representations, with multiplication given by tensor
product. This ring is sometimes called the representation ring, Rep(G).

If H ⊂ G is a subgroup, there is a morphism of stacks BH → BG. The pull-

back on K0 is the restriction of representations, and the pushforward on K0 is the
induction of representations.

More generally, if X is a smooth G-variety, we have the stack quotient [X/G].

A coherent sheaf on [X/G] is the same as a G-equivariant sheaf on X. We define
K0 [X/G] to be the Grothendieck group of G-equivariant vector bundles on X. Our
goal is to formulate an equivariant version of Grothendieck-Riemann-Roch.

Theorem-Definition. (Totaro) Let V be a G-rep such that the action is free

outside a closed subset S of codimension s > i. Define the Chow groups of BG as:
Ai (BG) := Ai ((V − S)/G).
The definition is independent of the choice of S because if S 0 ⊃ S is a larger subset,
then we have the localization sequence
A∗ ((S 0 − S)/G) → A∗ ((V − S)/G) → A∗ ((V − S 0 )/G) → 0.
For ∗ > dim(S 0 ) − dim(G) = n − s − dim(G), the first group vanishes, so the second
two groups are isomorphic. Matching the dimension notation with codimension
notation in the statement above,
∗ = n − dim(G) − i > n − s − dim(G).
This is equivalent to i < s. The definition is independent of the representation
V as well; suppose we have chosen (V, SV ) and (W, SW ) which both satisfy the
conditions. Then we have two vector bundles.
((V − SV ) × W )/G → (V − SV )/G, (V × (W − SW ))/G → (W − SW )/G
By S-independence applied to V ⊕ W with SV × W versus V × SW , the total spaces
of the bundles have isomorphic Chow groups Ai , and this implies that the base
spaces also have isomorphic Chow groups too. To show that G-representations V
with S of large codimension always exist, take any faithful representation V0 of G
of dimension m, and set
V = Hom(k m+N , V0 ),
for N  0, and S the locus of non-surjective linear maps. Note that in this case,
(V − S)/G is actually quasi-projective! 

Definition. (Edidin-Graham) For a smooth G-variety X, choose V and S as

before, and define the G-equivariant Chow groups as:
Ai [X/G] := Ai ((X × (V − S))/G).
Theorem. (Edidin-Graham) The intersection product on A∗ (X/G) is well-defined.
38 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

If E → X is a G-equivariant vector bundle, we can form

(E × (V − S))/G → (X × (V − S))/G
which is an ordinary vector bundle, so we have G-equivariant chern classes
cG i
i (E) ∈ A [X/G].
We now have all the necessary pieces to formulate the equivariant GRR theorem.
The wrinkle is that we must take the completions with respect to augmentation
ideals for the theorem to work. This is essentially because Ai (BG) can be nonzero
for i arbitrarily large.

Theorem. (Edidin-Graham) The map τ (V ) = ch(V ) · Td(TX − g), where g is

the trivial bundle with the adjoint action of G, defines a ring homomorphism
τ : K0 [X/G] → Â∗ [X/G] ⊗ Q
which factors through an isomorphism
τ̂ : K̂0 [X/G] ⊗ Q → Â∗ [X/G] ⊗ Q.
The completion of K0 [X/G] is taken with respect to the kernel of rk : K0 [X/G] → Z.
The completion of A∗ [X/G] is taken with respect to the ideal of positive degree
elements, so Y
Â∗ [X/G] = Ai [X/G].
i≥0
Theorem. (Equivariant GRR) The map τ is covariant for proper, representable
morphisms of stacks which are global quotients.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 39

12. 3/11/20 - Applications of Equivariant GRR

Example. Consider the case of BC× . The K0 ring is
Rep(C× ) ' Z[t, t−1 ]
The Chow ring is Z[h] because BC× is approximated by projective spaces Pl , l  0.
The tangent bundle is trivial, and g is a trivial representation so τ : Z[t, t−1 ] → QJhK
is simply the Chern character: τ (t) = exp(h). This does NOT descend to a ring
isomorphism Q((t)) → QJhK. If instead we complete Q[t, t−1 ] at the ideal (t − 1),
then we do get an isomorphism:
τ̂ : QJuK → QJhK
h2 h3
u 7→ exp(h) − 1 = h + + + ...
2 3!
Example. Consider the case of [P /C ], where the action of C× has weights
n ×

(a0 , . . . , an ). The approximations to Chow are given by (Pn ×(Cl −0))/C× → Pl−1 ,
which is isomorphic to:
P(O(a0 ) ⊕ · · · ⊕ O(an )) → Pl−1
The Chow ring of this variety is given by:
Z[h, ζ]/(hl , ζ n+1 + σ1 ζ n + · · · + σn+1 ),
where σi is the ith symmetric polynomial in the variables a0 h, a1 h, . . . , an h. As we
send l to infinity, we are left with
Z[h, ζ]/(ζ n+1 + σ1 (a)hζ n + · · · + hn+1 σn+1 (a)).
In particular, if C× acts on P1 with weights ±1, we get Z[h, ζ]/(ζ 2 − h2 ).

Theorem. (Weyl) Let G be a semisimple Lie group, g its Lie algebra, and h ⊂ g
the Cartan subalgebra. The irreducible representations Π of G are classified by
their highest weight, λ. For such an irrep, the character of eH ∈ T is given by
(w) ew(λ+ρ)(H)
P
chΠ (e ) = Q w∈W α(H)/2
H
.
α∈∆+ e − e−α(H)/2
The character formula is enough to determine chΠ on all of G because any semi-
simple element can be conjugated into T , and semisimple elements are Zariski dense.
Setting H = 0, we recover the dimension formula for irreps.

Proof. In the case of SL2 , the torus is C× . The highest weight is a nonneg-
ative integer n, and the representation can be realized as H 0 (P1 , O(n)). Weyl’s
formula in this case reads
ei(n+1)θ − e−i(n+1)θ
 iθ 
e 0 sin((n + 1)θ)
tr πn −iθ = =
0 e eiθ − e−iθ sin(θ)
This will be a consequence of EGRR for the action of C× on P1 with weights ±1.
×
χC
K0 [P1 /C× ] / K0 [pt/C× ]

τ τ
 
Â∗Q [P1 /C× ]
π∗
/ Â∗ [pt/C× ]
Q
40 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

The equivariant Euler characteristic of OP1 (n) is an element of K0 (BC× ) = Z[t, t−1 ].
Since OP1 (n) has no higher cohomology, we get the representation Πn = H 0 (P1 , O(n)).
 
1 nζ 2ζ
τ (Πn ) = π∗ (ch(OP (n)) · Td(P )) = π∗ e
1
1 − e−2ζ
Lemma. For p(ζ) ∈ A∗Q [P1 /C× ] = Q[h, ζ]/(ζ 2 − h2 ),
p(h) − p(−h)
π∗ (p(ζ)) =
2h
and the same is true after completion.

Proof. From the P1 -bundle model, any polynomial in h only gets killed by π∗ .
Since ζ 2 = h2 , all even degree terms in p(ζ) go to 0, whereas an odd degree term
a2k+1 ζ 2k+1 = a2k+1 ζh2k goes to a2k+1 h2k . 

Now we compute using the lemma:

e(n+1)ζ e(n+1)h − e−(n+1)h
 
τ (Πn ) = π∗ 2ζ ζ −ζ
= .
e −e eh − e−h
Note that τ is the sum of exponentials of the weights, under the diagonalization
isomorphism (which holds for any torus T ):
Rep(T ) ⊗ C ' C[T ].
The Weyl character formula for general G can be proven in a similar way, using an
action of a maximal torus T ⊂ B ⊂ G on the flag variety G/B (exercise). 

The morphism [X/G] → pt is not representable, so GRR does not apply directly.
However, there is a generalization of Hirzebruch-Riemann-Roch in the case when
[X/G] is proper and Deligne-Mumford. We would like a formula for the Euler
characteristic of a bundle V on [X/G] in terms of intersection numbers (degrees of
0-cycles) coming from Chern characters and Todd classes. A 0-dimensional substack
of [X/G] is a gerbe for a finite group Gx , and we set:
deg(BGx ) = 1/|Gx |.
Occasionally we write deg as an integral over [X/G], in analogy with cohomology.

Definition. Let V be a vector bundle on [X/G] a proper DM stack. Define

X
χ([X/G], V ) := (−1)i dim H i (X, V )G .
i
This is a finite sum because if m : [X/G] → M is the coarse space map, then
H i (X, V )G = H i (M, p∗ V ).
Remark. This definition of χ is the correct definition of pushforward to pt because
applying K0 to pt → BG → pt, we get
Z → Rep(G) → Z
1 7→ C[G] → 1.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 41

13. 3/30/20 - Localization Theorem in Equivariant K-theory

Our goal is to prove a version of Hirzebruch-Riemann-Roch for smooth proper DM
stacks which are global quotients [X/G]. This will allow us e.g. to compute e.g.
the dimensions of spaces of classical modular forms. Along the way, we will gain a
better understanding of equivariant K-theory in general.

Recall that K0 [X/G] is a ring, but it also a module over K0 [pt/G] via the pull-
back through [X/G] → [pt/G]. Geometrically, K0 [X/G]C is a coherent sheaf over
Spec K0 [pt/G]C . If G is a diagonalizable group (subgroup of a torus), then
Spec K0 [pt/G]C = Spec Rep(G)C ' G.
For a more general linear algebraic group,
Spec K0 [pt/G]C = Spec Rep(G)C ' G ad G,
whose points are conjugacy classes of semisimple elements of G. Both of these
isomorphisms send a representation to its character. The augmentation ideal in
Rep(G)C corresponds to the identity element 1 ∈ G, since the trace 1 recovers the
dimension of a representation.

Lemma. (Thomason) If h ∈ G does not fix any points in X, then the localization
(K0 [X/G]Q )h = 0.
If [X/G] is Deligne-Mumford, then only finite order elements h ∈ H can have fixed
points in X. This implies that K0 [X/G]C is supported over finitely many points hi :
k
Y
K0 [X/G]Q ' (K0 [X/G]Q )hi .
i=1

Theorem. (Edidin-Graham) If [X/G] is proper DM and α ∈ K0 [X/G]Q , then

deg ch(α)Td(TX − g) = χ([X/G], α1 ),
where α1 is the part of α over the augmentation ideal (called 1).

Proof. EGRR says that τ is an isomorphism after taking the completion K̂0
at 1, but τ it is only covariant for representable morphisms. There exists a smooth
G-variety X 0 with a finite G-equivariant surjection f : X 0 → X such that [X 0 /G] is
represented by a smooth variety. Now use the covariance of τ and the surjectivity
of f∗ on Chow groups, which implies surjectivity of f∗ on K̂0 :

K0 (X 0 /G)Q
f∗
/ K̂0 [X/G]Q / K0 (pt)Q = Q

∼ τ ∼ ∼
  
A∗ (X 0 /G)Q
f∗
/ Â∗ [X/G]Q / A∗ (pt)Q = Q.

The left square commutes by EGRR, and large square commutes by HRR. 

To compute the full χ([X/G], α), we must add up contributions from the remaining
sectors, closed points in the support of K0 [X/G]Q over Rep(G)Q . Each contribution
will be the degree of a Hirzebruch-Riemann-Roch type expression. Our main tools
will be localization in K-theory, followed by twist operators.
42 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

Theorem. (Thomason) Let G be a diagonalizable group acting on a scheme X.

(K0 [X/G])h ' K0 [X h /G] h

as modules over K0 [pt/G] ' Rep(G). This is referred to as the localization theorem.

Proof. We start by proving it in the case of C× acting on Pn with weights

(a0 , a1 , . . . , an ). The general case can be reduced to this one. The equivariant
K-theory of projective space is given in terms of x = [O(1)] by
n
Y
K0 [Pn /C× ] = Z[t, t−1 ][x]/ (x − t−ai ).
i=0
To see this, a version of the Hilbert syzygy theorem for multi-graded modules
implies that Z[t, t−1 , x, x−1 ] generate K0 . To check the relation, consider the equi-
variant tautological sequence
0 → OPn (−1) → OPn+1
n →Q→0
which implies that x · tai = 1 + qx, where q = [Q]. Now apply the λ operation
P
on K-theory (which sends sums to products):
rkE
X
λ(E) := (−1)i [∧i E].
i=0
The equation above becomes
n
Y
λ(tai x) = 0 · λ(qx)
i=0
n
Y
(1 − tai x) = 0
i=0
as desired (up to units). To check that 1, x, x2 , . . . , xn are independent, simply
apply χ (the ordinary Euler characteristic) to a linear relation, multiplied by x−m
to make all but one of the exponents negative.

Exercise. Check that the equivariant Chern character

n
Y Yn
Z[t, t−1 ][x]/ (x − t−ai ) → QJh, ζK/ (ζ + ai h)
i=0 i=0

is well-defined (send t 7→ eh and x 7→ eζ ).

Now, the localization theorem follows from geometry! If the weights are distinct
and h is general, then there are n + 1 fixed points, so
n
Y Yn
K0 [(Pn )h /C× ] = K0 (BC× ) = Z[t, t−1 ]
i=0 i=0
m−1
If m > 1 weights coincide, then there is a P in the fixed locus which contributes
×
K0 (BC ) ⊗ Z[x]/(x − 1) m
= Z[t, t−1 ][x]/(x − 1)m .
When you localize at h, the two Artinian rings match geometrically (graph x = t−a ),
and this correspondence can be refined for h a root of unity.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 43

To go from the case of projective space to a general smooth projective X, first

embed X ,→ Pn equivariantly (by choosing a G-linearized very ample line bundle).
Pulling back give ring homomorphisms
K0 [(Pn r (Pn )h )/G] → K0 [(X r X h )/G]
K0 [(Pn r (Pn )h )/G]h → K0 [(X r X h )/G]h .
Since K0 [(Pn r (Pn )h )/G]h = 0 by localization for Pn , K0 [(X r X h )/G]h = 0 as
well - this is Thomason’s Lemma above for Pn . Next we use the excision sequence
for K-theory, and the fact that localization is an exact functor:
· · · → K1 [(X r X h )/G] → K0 [X h /G] → K0 [X/G] → K0 [(X r X h )/G] → 0
· · · → K1 [(X r X h )/G]h → K0 [X h /G]h → K0 [X/G]h → K0 [(X r X h )/G]h → 0.
We have not defined K1 yet, but it is a module over K0 . Since the latter vanishes
in this case, the former also vanishes, and we get the full localization theorem. 

Explicitly, the localization theorem isomorphism is induced by the pullback i∗ for

i : [X h /G] → [X/G]
the inclusion. Using the flat deformation to the normal cone, we know that
i∗ i∗ (β) = λ(N ∨ ) · β,
where N = NX h /X is the normal bundle, so the inverse of the isomorphism in the
theorem is given by  
β
β 7→ i∗ .
λ(N ∨ )
Remark. The class λ(N ∨ ) is invertible in K0 [X h /G] h because it is a product

over weights complementary to the fixed locus weights.

Using the inverse formula, along with functoriality of [X h /G] → [X/G] → pt,
we find that for αh ∈ (K0 [X/G])h ,
i∗ αh
 
h
χ([X/G], αh ) = χ [X /G], .
λ(N ∨ )
44 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

14. 4/1/20 - Twisted Sectors

Let G be a diagonalizable group. Consider a G-variety Y such that h ∈ G acts
trivially, and assume that h has finite order, so we have H = hhi ' Z/m. Given a
G-equivariant vector bundle V on Y , it has a direct sum decomposition
M
V ' Vξ
ξ∈H ∨

over characters of H. In other words, K0 [Y /H] ' Rep(H) ⊗ K0 (Y ).

Definition. The twist operator twh : K0 [Y /G] → K0 [Y /G] is defined by

X
[V ] 7→ ξ(h)[Vξ ].
ξ∈H ∨

Proposition. The twist operator twh sends

(K0 [Y /G])h → (K0 [Y /G])1 .
Proof. The pull-back morphisms K0 [pt/G] → K0 [Y /G] → K0 [Y /H] are both twh -
equivariant. Geometrically, the twist can be P viewed as the multiplication by h−1
from H → H. Indeed, if f ∈ C[H], then f = ai ξi , then we have
X X X
twh f (g) = ci twh ξi (g) = ci ξi (h)ξi (g) = ci ξi (hg) = f (hg).

If f vanishes at a point g, then twh f vanishes at h−1 g. 

The other key property of twh is that it preserves G-invariants, so it preserves

the Euler characteristic that we seek. Let Y be the fixed locus X h from last time.
The operator twh moves the sector at h to the sector at 1 so that we can apply
the Hirzebruch-Riemann-Roch theorem. Recall
that α ∈ K0 [X/G] decomposes into
components αh ∈ (K0 [X/G])h ' K0 [X h /G] h and χ is additive, so we have:

Theorem. If [X/G] is proper DM and α ∈ K0 [X/G]Q , then

!!
i∗ αh
X Z
χ([X/G], α) = ch twh ∨ Td(TXh − g)
[X h /G] λ(NX h /X )
h∈supp K0 [X/G]

There is a slicker way to write this formula as a single integral over the inertia stack
I[X/G] if you care to unwind the statement. Let
I(X, G) = {(x, h) : hx = x} ⊂ X × G,
with the G-action given by g · (x, h) = (gx, ghg −1 ). The inertia stack is a global
quotient in this case: I[X/G] = [I(X, G)/G]. Note that I(X, G) admits a finite G-
equivariant decomposition into connected components indexed by h ∈ supp K0 [X/G],
so one can define a global twist operator tw.

Theorem. (Edidin) If X = [X/G] is proper DM, α ∈ K0 [X/G]Q , and f : IX → X

is the natural morphism from the inertia stack, then
!!
f ∗α
Z
χ(X, α) = ch tw Td(IX ).
IX λ(Nf∨ )
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 45

Example. Consider the stack P(4, 6) = (C2 r {0})/C× , acting with weights (4, 6).
To compute K0 , we use
K0 [{0}/C× ] → K0 [C2 /C× ] → K0 (P(4, 6)) → 0.
The first two terms are both isomorphic to Z[t, t−1 ], since all vector bundles on C2
are trivial. The pushforward is multiplication by
λ(T0∨ ) = (1 − t−4 )(1 − t−6 ).
Hence, K0 (P(4, 6)) ' Z[t, t−1 ]/(t4 − 1)(t6 − 1). As a sheaf over Spec Z[t, t−1 ] it is
supported at ±1, ±i, η = e±πi/3 , ω = e±2πi/3 . Using the formula above,
1 (−1)k ik + (−i)k ω k + ω −k + η k + η −k
χ(P(4, 6), tk ) = (k + 5) + (k + 5) + + .
24 24 8 12
This recovers the dimension formula for spaces of classical modular forms because
P(4, 6) is isomorphic to the moduli space of elliptic curves, M 1,1 . In this case,
χ(tk ) = h0 (tk ) because the higher cohomologies vanish.

0
 k odd
dim Mk (SL2 (Z)) = bk/12c + 1 k 6≡ 2(12)

bk/12c k ≡ 2(12).

46 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

15. 4/6/20 - Bundles of Principal Parts

Let X be a smooth projective variety, and E locally free sheaf on X. For k ≥ 0, it
is natural to seek a vector bundle P k (E) whose fiber at a point x ∈ X is given by

P k (E)x = H 0 (E ⊗ OX,x /mk+1

x )
= {germs of sections of E at x}/{those vanishing to order k + 1 at x}.
This can be done globally by thickening the diagonal ∆X ⊂ X × X:
k+1
Definition. P k (E) = pr2∗ pr1∗ E ⊗ OX×X /I∆


.

Note that P 0 (E) = E itself. Consider the short exact sequence

k k+1 k+1 k
0 → I∆ /I∆ → OX×X /I∆ → OX×X /I∆ → 0,
tensor with pr1∗ E, and then apply pr2∗ to get

0 → E ⊗ Symk (ΩX ) → P k (E) → P k−1 (E) → 0.

We have used the fact that I∆ k k+1

/I∆ ' Symk (ΩX ). This allows us to compute the
total Chern class inductively, using the Whitney sum formula:
k
Y
c(P k (E)) = c E ⊗ Symi (ΩX ) .


i=0

Remark. The short exact sequence for k = 1,

0 → E ⊗ ΩX → P 1 (E) → E → 0,
gives us a naturally defined class a(E) ∈ Ext1 (E, E ⊗ ΩX ) = H 1 (End(E) ⊗ ΩX ).
This is called the Atiyah class of E, and its vanishing is equivalent to the existence
of a holomorphic connection on E - the correct setting is for X a compact Kähler
manifold. If E = L is a line bundle, then a(L) ∈ H 1 (ΩX ) ' H 1,1 (X, C), and
a(L) = −2πi · c1 (L). More generally, by the splitting principle

tr(∧k a(E)) = (−2πi)k ck (E) ∈ H k (ΩkX ).

If a(E) = 0, then all the rational Chern classes of E vanish, but the converse is false;
consider E = OP1 (1) ⊕ OP1 (−1). When dim(X) = 1, any holomorphic connection is
integrable (flat). The Atiyah class is also important in deformation theory: given
a first order deformation δ ∈ H 1 (TX ) of X and a coherent sheaf E on X, the ob-
struction to deforming E with X is δ · a(E) ∈ H 2 (End(E)). Illusie generalized this
to the case of singular varieties X, replacing ΩX with LX .

Our first application of principal parts will be counting singular elements in linear
series. A pencil of plane curves is given by a line P1 ' hF, Gi ⊂ PH 0 (OP2 (d)) ' PN .
A general pencil will contain curves with at worst ordinary double points, so we
consider P 1 (OP2 (d)), a bundle of rank 3. The global sections F, G of OP2 (d) induce
sections τF , τG of P 1 (OP1 (d)), and we want to count the points where the latter
two sections become linearly dependent. This is given by c2 :
deg c2 P 1 (OP2 (d) = 3(d − 1)2 .


 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 47

A similar computation with X = Pn and L = OPn (d) counts singular elements in a

general pencil of hypersurfaces:
deg cn P 1 (OPn (d) = (n + 1)(d − 1)n .

This formula can alternatively be obtained by looking at the universal singularity

Σ ⊂ Pn × PN , which is a complete intersection of n + 1 hypersurfaces of bidegree
(d − 1, 1), and pushing it forward to PN .

Proposition. The total Chern class for the bundle P k (OPn (d)) is given by
n+k
c P k (OPn (d)) = (1 − (d − k)h)( n )

Proof. The Euler sequence for Pn tensored with OPn (d) reads:
0 → ΩPn (d) → OPn (d − 1)⊕n+1 → OPn (d) → 0.
The middle term is not isomorphic to P 1 (OPn (d)), but they have the same Chern
classes! This handles the case k = 1. We leave the general case as an exercise. 

Proposition. Let L be a line bundle on X with dim(X) = n. For a pencil in

|L| such that each fiber has at most one singularity (a double point) lying outside
the base locus, the number of singular fibers is given by:
n
X
1
(i + 1)c1 (L)i cn−i (ΩX ).


deg cn P (L) = deg
i=0

Proof. Using the Whitney sum formula, c(P 1 (L)) = c(L)c(L ⊗ ΩX ). The degree
n part of the product becomes
cn P 1 (L) = cn (L ⊗ ΩX ) + c1 (L)cn−1 (L ⊗ ΩX )

n
X n−1
X
= c1 (L)i cn−i (ΩX ) + c1 (L) (i + 1)c1 (L)i cn−1−i (ΩX )
i=0 i=0
n
X
= (i + 1)c1 (L)i cn−i (ΩX ). 
i=0

Next, we can ask for the degree of the locus of plane curves with triple points,
which has codimension 4 in PN . Consider P 2 (OP2 (d)), a bundle of rank 6. Take a
general hF0 , F1 , F2 , F3 , F4 i ' P4 ⊂ PN , and consider the associated sections τFi of
P 2 (OP2 (d)). The locus where they become linearly dependent has size:
deg c2 P 2 (OP2 (d)) = 15(d2 − 4d + 4).

For example, the locus of asterisks in the space of plane cubics (P9 ) has degree 15.
Taking this argument to the extreme, the locus of cones inside the space PN of
degree d hypersurfaces in Pn has degree:
 n+d−1

d−1

n
deg cn P (OP (d)) =
n .
n

A second application of principal parts is counting lines meeting a hypersurface with

high multiplicity. For example, how many lines meet a quintic surface S5 ⊂ P3 at
a single point? Let U → G(1, 3) be the universal line, that is the projectivization
48 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

PS of the tautological sub-bundle on G(2, 4), and let L be the pull-back of OP3 (d)
via U → P3 . We form the bundle of relative principal parts using ∆U ⊂ U ×G U :
k
(L) = pr2∗ pr1∗ (L) ⊗ OU ×G U /I∆
k+1


PU/G .
Its fiber at a point (M, p) ∈ U is given by
k
PU/G (L)(M,p) = {germs of sections of L|M at p}/{those vanishing to order k + 1}.
These bundles fit into the same short exact sequences as before, only with the
cotangent sheaf ΩU replaced by its relative version, ΩU/G . The same relative con-
struction works for any smooth proper morphism X → Y .

Proposition. The number of lines meeting a general surface Sd ⊂ P3 (d ≥ 5)

at a point with multiplicity ≥ 5 is given by:
 
4
deg c5 PU/G (L) = 35d3 − 200d2 + 240d,
Proof. We compute using the Whitney sum formula applied to the 5 pieces
L, L ⊗ ΩPS/G , L ⊗ Sym2 (ΩPS/G ), L ⊗ Sym3 (ΩPS/G ), L ⊗ Sym4 (ΩPS/G ),
aided by the fact that the relative cotangent bundle ΩPS/G is a line bundle. Its first
Chern class is −2ζ + σ1 by the relative Euler sequence, so
  Y 4
4
c5 PU/G (L) = ((d − 2i)ζ + iσ1 ) ∈ A∗ (PS). 
i=0

Note: the d and d terms vanish because ζ 4 = ζ 5 = 0 (ζ is pulled back from P3 ).

4 5

The formula gives 575 lines for the quintic surface.

Question. What is the maximum number M (d) of lines that can appear on a
smooth surface Sd ⊂ P3 with d ≥ 4?

For reasons of Kodaira dimension, the number of lines must be finite, and indeed
for general Sd there are no lines at all. However, the Fermat surface
V (wd + xd + y d + z d = 0) ⊂ P3
has 3d2 lines on it (exercise). This gives a uniform lower bound on M (d), although
it can be improved for certain d. To produce an upper bound, look at the curve Γ
in U of lines meeting Sd at a point with multiplicity ≥ 4. It has class
  Y 3
3
c4 PU/G (L) = ((d − 2i)ζ + iσ1 ) ∈ A4 (PS).
i=0

The image of Γ in P3 has degree given by

 
3
ζ · c4 PU/G (L) = 11d2 − 24d.
If Sd contains a line, then that line must be a component of this curve, so 11d2 −24d
is an upper bound for M (d). The true value is not known past M (4) = 64.

Remark. The degree of the codimension d − 3 locus in PN consisting of sur-

faces which contain a line alsoprovides 
an upper bound for M (d), but it is higher.
Exercise: the answer is deg c4 Symd S ∨ on G(1, 3), which is O(d4 ).
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 49

16. 4/8/20 - Classical Moduli Spaces

Today we will survey some results on the Chow groups of some standard moduli
spaces. The most studied examples are:
• Mg - moduli of smooth curves of genus g,
• Ag - moduli of principally polarized abelian varieties of dimension g, and
• Kg - moduli of polarized K3 surfaces of degree 2g − 2.
Note that the g denotes something different for each row, but they are related.
The Jacobian of a genus g curve is an abelian variety of dimension g, and a gen-
eral element of the linear system on a polarized K3 surface is a curve of genus g.
The dimension of each moduli space is dim Mg = 3g − 3, dim Ag = g(g + 1)/2,
dim(Kg ) = 19. Smooth compactifications with divisorial boundary are best for
doing intersection theory (this excludes the Baily-Borel compactifications). The
following arrangement of the smooth examples highlights certain analogies.
Mg ⊂ Mct
g ⊂ Mg
vor
Ag ⊂ Ag
Σ
Kg ⊂ K
e g ⊂ Kg ?

Except for M0 and M1 , all of these moduli spaces are Deligne-Mumford stacks.
Recall that the coarse space morphism induces an isomorphism on Chow groups
A∗ (M)Q ' A∗ (M )Q . The integral Chow groups of a moduli stack are typically
hard to compute; here is a sample of what is known.

Theorem. (Mumford) A1 (M1,1 ) ' Z/12 and A1 (M1,1 ) ' Z.

Theorem. (Vistoli) A∗ (M2 ) ' Z[λ1 , λ2 ]/(10λ1 , 2λ21 − 24λ2 ).

Theorem. (Larson)
A∗ (M2 ) ' Z[λ1 , λ2 , δ1 ]/(24λ21 − 48λ2 , 20λ1 λ2 − 4δ1 λ2 , δ13 + δ12 λ1 , 2δ12 + 2δ1 λ1 ).
The first result that we will prove is Mumford’s relation in A1 (Mg )Q :

Theorem. Let π : C g → Mg be the universal family of curves with its rela-

tive dualizing sheaf ω = ωC g /Mg , and let E = π∗ (ω) be the Hodge bundle. For
λ = c1 (E), κ = π∗ (c1 (ω)2 ), and δ the boundary divisor class, we have:
κ+δ
λ= ∈ A1 (Mg )Q .
12
Proof. It suffices to check the relation on any family of curves π : S → B over a
base curve B. We apply the GRR formula to ω:
 
ch(E) − 1 = π∗ ec1 (ω) · Td(S/B)

The LHS is given by (g − 1)[B] + λ. The inside of the expression on the RHS has
codimension 2 part given by:
c1 (ω)2
(1) + c1 (ω) · Td1 (S/B) + Td2 (S/B)
2
50 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

We can compute the total Chern class

c(TS )
= (1 + c1 (TS ) + c2 (TS ))(1 − π ∗ c1 (TB ))
π ∗ c(TB )
= 1 − c1 (ω) + c2 (TS ) − c1 (TS ) · π ∗ c1 (TB )
The codimension 2 part has degree equal to δ via topological Euler characteristics:
χtop (S) = (2 − 2g)χtop (B) + δ;
c2 (TS ) = −c1 (ω) · π ∗ c1 (TB ) + δ
= c1 (TS ) · π ∗ c1 (TB ) + δ.
Using the expression for the Todd class,
c1 (ω) c1 (ω)2 + δ
(2) Td(S/B) = 1 − + .
2 12
Mumford’s relation now follows from substituting (2) into (1). 

Let us describe A1Q = PicQ for some moduli spaces. First, I claim that the Picard
group is discrete in each of the standard examples. By the localization sequence
for Chow groups, we have
A1 (∆) → A1 (M) → A1 (M) → 0,
so it suffices to show that Pic(M) ' H 1 (M, O× ) is discrete. By the exponential
long exact sequence
c
· · · → H 1 (M, O) → H 1 (M, O× ) −→
1
H 2 (M, Z) → . . . ,
it suffices to prove that the irregularity H 1 (M, O) = 0. By Hodge theory, it suffices
to show that H 1 (M, C) = 0. To see this, note that over C each of the examples
above is the quotient of a contractible space by a discrete group.
• M1,1 (C) ' BG where G is SL2 (Z).
• Mg (C) ' BG where G is the mapping class group MCGg .
• Ag (C) ' BG where G is the arithmetic group Sp2g (Z).
• Kg (C) ' BG where G is the arithmetic group O0 (Λ2g−2 ).
Each of these groups G has finite abelianization (MCGg is perfect for g ≥ 3, and
MCGab 2 = Z/10). Passing from M to M only makes π1 smaller, so H1 (M, C) = 0.

Theorem. (Harer) The cohomology groups of Mg stabilize: for 3k − 1 ≤ g,

H k (Mg ) ' H k (Mg+1 ).
Mumford conjectured that the stable cohomology ring H ∗ (M, Q) is a polynomial
ring in the κ classes κi = π∗ (c1 (ω)i+1 ).

Theorem. (Madsen-Weiss) Mumford’s Conjecture is true:

H ∗ (M∞ , Q) ' Q[κ1 , κ2 , . . . ].
In particular, this means that Pic(Mg )Q = Qκ = Qλ for all g ≥ 5. We will
see that the same is true for g ≥ 3, while Pic(M2 )Q = 0. Borel showed that
H 2 (BSp2g (Z))Q ' Q for g ≥ 4, so in fact Pic(Ag )Q = Qλ for g ≥ 4 as well.

In what follows, we compute PicQ (Mg ) for 2 ≤ g ≤ 5, using GIT models:

 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 51

To build a moduli space, we need to embed all the varieties X into the same
ambient space P , consider the appropriate Hilbert scheme, and then take the quo-
tient by the automorphism group of P . If every aut of X is induced by an aut of
the ambient space, then we have successfully constructed a moduli space. Canon-
ical embeddings are nice because any automorphism of X preserves the canonical
bundle. Since not every canonical map is an embedding, pluricanonical maps are
used to prove existence, but the canonical map is easier to study (see below).

g = 2. Every smooth curve C of genus 2 is hyperelliptic, and the canonical mor-

phism φK : C → P1 has 6 distinct branch points. To build M2 , we take the
GIT quotient of Sym6 (P1 ) ' P6 by the action of P GL(2). An orbit is GIT stable
(resp. semi-stable) iff < 3 (resp. ≤ 3) points collide, so all nodal curves are stable.
dim(M2 ) = 6 − 3 = 3. dim Pic(M2 )Q = 1 − 1 = 0.

g = 3. Let C be a smooth curve of genus 3. The canonical bundle gives a map

φK : C → P2 . There are two cases: either C is hyperelliptic and φK is the double
cover of a conic in P2 , or C is non-hyperelliptic and φK is an embedding (this is
the generic case). To build M3 , consider space P14 of plane quartics, and take the
GIT quotient by the action of P GL(3). Every smooth or nodal quartic is GIT sta-
ble. The double conics are strictly semi-stable. Blowing up the double conic locus
gives a model for M3 . A degenerating family of smooth quartics gives 8 points on
the conic, which are precisely the branch points of the hyperelliptic double cover.
dim(M3 ) = 14 − 8 = 6. dim Pic(M3 )Q = 1 + 1 − 1 = 1.

g = 4. Let C be a smooth curve of genus 4. The canonical bundle gives a map

φK : C → P3 . The hyperelliptic locus is no longer a divisor, so we can ignore it.
All remaining curves are complete intersections of a quadric surface and a cubic
surface in P3 . To build M4 , consider the P15 -bundle over P9 = PH 0 (P3 , O(2)), and
take the GIT quotient by the action of P GL(4).
0 → H 0 (P3 , O(1)) → H 0 (P3 , O(3)) → H 0 (OQ (3)) → 0
Every smooth or nodal curve is GIT stable. Those curves lying on a singular quadric
surface are called theta-null. dim(M4 ) = 24 − 15 = 9. dim Pic(M4 )Q = 2 − 1 = 1.

g = 5. Let C be a smooth curve of genus 5. The canonical bundle gives a map

φK : C → P4 , where the general curve is a complete intersection of three quadrics.
To build M4 , consider the Grassmannian G(3, 15) and take the GIT quotient by
the action of P GL(5). dim(M5 ) = 36 − 24 = 12. dim Pic(M5 )Q = 1 − 1 = 0?
There is actually a divisor of trigonal curves, for which the ideal IC (2) cuts out an
F1 ⊂ P4 containing C as a trisection, so dim Pic(M5 )Q = 1.

The hyperelliptic, theta-null, and trigonal loci can be used to produce an affine
stratication of Mg for 2 ≤ g ≤ 5. This gives a Q-basis for all the Chow groups.

To complete the description of Pic(Ag )Q for g ≤ 3, note that A1 ' M1,1 , A2 ' M2ct
and A3 ' M3ct , so they have Picard ranks 0, 1, and 2, respectively.

Theorem. dim Pic(Kg )Q → ∞ as g → ∞. (next time)

52 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

17. 4/13/20 - Test Curves on Mg and Kg

Using the affine stratifications described last time, the rational Chow ring is:

Theorem. (Loojienga-Fontanari) A∗ (Mg )Q = Q[λ]/λg−1 for g ≤ 5.

Looijenga proved (1995) that for all Mg , a polynomial in the kappa classes κi
of total degree g − 1 vanishes in A∗ (Mg ). This is consistent with S. Diaz’s theorem
(1984) that any complete subvariety of Mg has dimension ≤ g − 2.

Conjecture. (Faber) The tautological subring R∗ (Mg ) ⊂ A∗ (Mg )Q , which is

generated by the κi classes, is an even Poincaré duality algebra of dimension g − 2.

Is there a closed manifold with R∗ (Mg ) as its even cohomology? Wide open.

Next, we turn to Mg . The boundary ∆ = Mg r Mg is a union:

∆ = ∆0 ∪ ∆1 ∪ · · · ∪ ∆bg/2c
The general element of ∆0 is a smooth curve of genus g − 1 with two points glued
at a node. The general element of ∆i (i > 0) is a pair of smooth curves of genera i
and g − i, glued at a node. To compute Pic(Mg )Q , we use the localization sequence
Z1+bg/2c → Pic(Mg ) → Pic(Mg ) → 0.
Proposition. The components ∆i of the boundary of Mg have linearly indepen-
dent classes. In particular, dim Pic(Mg )Q = 2 + bg/2c for g ≥ 3.

Proof. The approach is to produce 1 + bg/2c complete curves Ci ⊂ Mg , such

that the matrix deg(Ci · ∆j ) of intersection numbers is nonsingular. The test curve
Ci will be inside ∆i for each i = 0, 1, . . . , bg/2c, constructed by sliding the point
of attachment. For C0 , fix a smooth curve Σ of geometric genus g − 1, and glue
marked points p and q. As the point p moves along Σ, we get a curve in ∆0 . For Ci
(i > 0), fix two marked curves (Σ1 , p) and (Σ2 , q) of genera i and g − i, respectively,
and glue the marked points. As the point p moves along Σ1 , we get a curve in ∆i .
The intersection numbers are given by:
 
2 − 2g 1 0 0 ...
 1 0 0 0 . . .
 
deg(Ci · ∆j ) = 
 0 0 −2 0 . . . .
 0 0 0 −4 . . . 
.. .. .. ..
 
..
. . . . .
To see this, we use the fact (from the deformation theory of nodes) that the fiber
of the normal bundle N∆i /Mg at a nodal curve Σ is naturally isomorphic to the
tensor product of the tangent spaces to the two branches of Σ at the node point.
By the excess intersection formula,
deg(Ci · ∆i ) = deg N∆i /Mg |Ci .
For i > 0, Ci is isomorphic to the component Σ1 , and N∆i /Mg |Ci ' TΣ1 , so
deg(Ci · ∆i ) = 2i − 2. For C0 , observe that the stable limit as p collides with q is
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 53

isomorphic to Σ glued to a rational nodal tail, a curve whose moduli point lies in
∆0 ∩ ∆1 . To construct the stable family explicitly, consider
S = Bl(q,q) (Σ × Σ) → Σ,
with the proper transforms of ∆Σ and Σ × {q} identified, to produce a family of
nodal curves over Σ ' C0 . The normal bundles of the two section curves have
degrees 2 − 2(g − 1) − 1 and −1, respectively, which gives deg C0 · ∆0 = 2 − 2g. 

Theorem. (O’Grady) For any N , there exists g such that dim Pic(Kg )Q > N .

Proof. Recall that points of Kg are pairs (S, L) where S is a K3 surface, meaning
that H 1 (OS ) = 0 and KS ' OS , and L ∈ Pic(S) is primitive and ample, with
c1 (L)2 = 2g − 2. To construct examples of polarized K3 surfaces, take the double
cover of a surface T branched along a smooth curve B ∈ | − 2KT |. If T = P1 × P1 ,
then ϕ : S → T is branched along curve of bidegree (4, 4). For any integers a, b > 0,
L = ϕ∗ (ah1 + bh2 )
is ample with c1 (L)2 = 4ab, and it is primitive as long as (a, b) = 1. Hence, for
each pair a, b of coprime positive integer we have a divisor Da,b ⊂ K1+2ab . Indeed,
counting dimensions, h0 (T, 4h1 + 4h2 ) = 25 so we have dim P24 − dim Aut(T ) = 18.

For each Da,b , we will construct a test curve Ca,b ⊂ Da,b such that:
• deg Ca,b · Da,b < 0,
• deg Ca,b · Da,b = 0 if min{c, d} > 2 min{a, b} or min{c, d} < 12 min{a, b}.
The second property comes from the fact that Ca,b and Dc,d are disjoint under the
given inequalities. For g = 1 + 2p1 p2 p3 . . . p2N (with pi distinct primes), we can
produce a diagonal intersection matrix (of size N ) with nonzero determinant.

Let P1 → P24 be a general pencil of (4, 4) curves in P1 × P1 . There are 68 nodal

members of the pencil; the corresponding K3 surface S → T will have a node. Let
C → P1 be the double cover branched at those 68 points. The family of curves
Y ⊂C ×T →C
has tridegree (2, 4, 4), and the total space Y has 68 nodes. Let X → C × T be the
double cover branched along Y , with 68 ordinary double points. A small resolution
X̂ → X → C gives a family of smooth K3 surfaces. If we pullback the primitive
ample line bundle L = ah1 + bh2 from T , we get a family of polarized K3 surfaces,
which corresponds to a moduli map Ca,b = C → K2ab+1 .

By Noether-Lefschetz theory (next time), the normal bundle NDa,b /Kg is isomor-
phic to the restriction of the dual Hodge line bundle to Da,b . The Hodge bundle
is ample. For disjointness, assume WLOG that a = min{a, b}, and suppose that
we have found an automorphism f of a K3 surface S such that f ∗ ϕ∗ (ch1 + dh2 ) '
ϕ∗ (ah1 + bh2 ). Replacing ϕ∗ with ∼ for readability, we have
f∗ (h̃2 ) · (ch̃1 + dh̃2 ) = h̃2 (ah̃1 + bh̃2 ) = 2a
On the other hand, using the fact that h2 is nef, f∗ (h̃2 )·(ch̃1 +dh̃2 ) ≥ min{c, d} > 2a,
contradiction. Swapping the roles of {a, b} with {c, d} gives a contradiction for the
second inequality too. 
54 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

18. 4/15/20 - Borcherds Modularity

To get more detailed information about the moduli spaces Kg of polarized K3
surfaces, we need to introduce some Hodge theory. Given (S, L), the primitive
cohomology lattice hc1 (L)i⊥ ⊂ H 2 (S, Z) is abstractly isomorphic to
Λg = Z(2 − 2g) ⊕ U ⊕2 ⊕ E8⊕2 (−1).
If we choose an isomorphism H 2 (S, Z) ' Λg , then the Hodge decomposition gives
Λg ⊗ C = H 0,2 ⊕ H 1,1 ⊕ H 2,0 ' C ' C ⊕ C19 ⊕ C.
The moduli space of Hodge structures on Λg is a Hermitian symmetric domain D
of Type IV, meaning that it has a complex structure and an action of the real
orthogonal group O(Λg ⊗ R) ' O(2, 19) such that
D ' O(2, 19)/O(2) × O(19).
To see the complex structure, we may view D as an open subset of the quadric
hypersurface Q ⊂ P(Λg ⊗ C) defined by the lattice pairing (v, v) = 0, with the open
condition on Cv ∈ P(Λg ⊗ C) given by (v, v) > 0. For a dual lattice vector x ∈ Λ∨g,
x⊥ is a hyperplane section of the quadric Q, and it is non-empty iff (x, x) < 0.

The moduli space of polarized K3 surfaces of degree 2g − 2 has a period map:

ρ : Kg → D/Γ,
where Γ = O0 (Λg ) ⊂ O(Λg ) is the subgroup of lattice automorphisms which act
trivially on the discriminant Λ∨
g /Λg , or equivalently extend to an automorphism of
the whole cohomology lattice ΛK3 fixing c1 (L).

Theorem. (Piatetskii-Shapiro, Shafarevich) The period map ρ is an open em-

bedding, and the complement of the image consists of a union of hyperplanes r⊥ ,
for r ∈ Λg with (r, r) = −2. Since Γ acts transitively on the set of such r, we get
an irreducible hypersurface in the quotient D/Γ.

Remark. If we relax the ampleness condition on L and only require that it be

nef, then we get an isomorphism K e g ' D/Γ, the moduli space of quasi-polarized
K3 surfaces. Geometrically this corresponds to adding mildly singular (ADE) K3
surfaces to the moduli space, using simultaneous resolution.

There is a natural Hodge line bundle λ on D, whose fiber is H 2,0 , and it descends
to the quotient D/Γ. Baily-Borel proved that some power of λ has enough sections
to embed D/Γ into projective space (λ is ample). The closure of D/Γ in projective
space is a singular projective variety, often denoted (D/Γ)BB .

For any x ∈ Λ∨g , the hyperplane x

⊥
and its Γ-translates descend to a hypersur-
face in D/Γ. These divisors are called Noether-Lefschetz divisors because they
parametrize K3 surfaces with Picard number > 1. If Pic(S) contains a sublattice
 
2g − 2 k
hc1 (L), βi = ,
k 2m
then the orthogonal projection of β is an element x ∈ Λ∨
g . If we index these
hypersurfaces by −n = (x, x) ∈ Q>0 , we can make a formal power series with
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 55

classes in A1 (K
e g ) as coefficients:
X X
f (q) = −λ + [x⊥ ]q n .
n∈Q>0 x∈Λ∨
g /Γ
(x,x)=−2n

Note: n can have a denominator of at most 2g − 2.

Theorem. (Borcherds) f (q) is the Fourier series in q = e2πiτ of a modular form of

weight 21/2 and level Γ(2g − 2), with values in A1 (K
e g ).

e g )Q is spanned by the classes [x⊥ ].

Theorem. (Bergeron-Li-Millson-Moeglin) A1 (K

Although there are infinitely many divisors [x⊥ ], their span is finite-dimensional
in A1 (K e g )∨ , we have
e g )Q . Indeed, for each element of µ ∈ A1 (K
Q

µ · f (q) ∈ Mod(21/2, 2g − 2),

so we have an injective morphism A1 (K e g )∨ → Mod(21/2, 2g − 2). Bruinier was
Q
able to describe the image of this morphism with a simple vanishing criterion,
which leads to a dimension formula for Pic(Kg )Q using the Selberg trace formula2
31
for spaces of modular forms dim Pic(Kg )Q ∼ 24 g.

Remark. The growth of dim Pic(Mg )Q came from the boundary divisors, which
were related to lower genus curves. The growth of dim Pic(Kg )Q comes from the
NL-divisors, which are themselves locally symmetric spaces for the group O(2, 18).

Theorem. (Maulik) For any infinite sequence of elements xi ∈ Λ∨

g , a complete
curve C → Kg must intersect one of the NL-divisors x⊥
i .

Corollary. (BKPS) Any complete family with constant Picard number is isotrivial.

Proof. The crucial fact is ampleness of the Hodge line bundle λ (Baily-Borel).
It suffices to express aλ as a combination of the NL-divisors x⊥ i . Let θ(q) be the
Siegel theta function of weight 1/2, and let E10 (q) be the Eisenstein series of weight
10. If f (q) ∈ Mod(21/2, 2g − 2) with constant term c0 , then
f (q) − c0 θ(q)E10 (q) ∈ Mod(21/2, 2g − 2)
is a cusp form. The coefficients of a cusp form grow like O(nwt/2+ ) = O(n11/4+ ).
On the other hand, the coefficients of θ(q)E10 (q) are bounded from below by O(n9 ).
These estimates imply that the coefficients of f (q) are eventually nonzero. Next
consider the composition:
α
M
Qei → Mod(21/2, 2g − 2)∨ → Pic(Kg )Q ,
i

which sends ei to [x⊥ If an element f (q) ∈ Mod(21/2, 2g − 2)∨ vanishes on the

i ].
span of α(ei ), then it vanishes on e0 as well. Hence α(e0 ) ∈ spani (α(ei )), so
λ ∈ spani [x⊥
i ]. 

2The Borcherds result is slightly more refined; you get a vector-valued modular form for the
Weil representation on C[Λ∨
g /Λg ]. This is why we use the trace formula to compute the dimension.
56 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

19. 4/20/20 - Hirzebruch-Mumford Proportionality

Hirzebruch proved a statement relating certain top intersection products on X =
Γ\D (for D = G/K a symmetric space and Γ ⊂ Aut(D) torsion-free, cocompact)
with those on Y , the flag variety containing D as an open subset. The relation is a
universal constant of proportionality c(Γ) between the two.

Remark. Every arithmetic group has a torsion-free subgroup of finite index.

Torsion-free discrete groups Γ act freely on symmetric spaces3.

To set this up, let G be a semi-simple R-group, and let K be a maximal compact
subgroup such that D = G/K has a complex structure. The standard examples
are SL(2, R)/SO(2), more generally Sp(2n, R)/U (n), O(2, n)/SO(2) × SO(n), and
U (1, n)/U (1) × U (n). The compact dual of D is a flag variety Y = GC /P , where
P = KC ·P+ is a parabolic subgroup with unipotent radical P+ such that P ∩G = K.

To produce (complex) vector bundles on D take a representation ρ : K → GLn (C),

and form the bundle
Vρ = G ×K Cn → G/K = D,
which descends to Γ\D. These are called automorphic bundles. There is a nat-
ural extension of V to a vector bundle Ṽ on the compact dual Y , by extending
the representation linearly from K to KC and trivially to P = KC · P+ , and then
performing the same quotient construction.

Theorem. (Hirzebruch) Any top intersection cα (Vα ) on X = Γ\D is propor-

tional to the analogous
P top intersection cα (Ṽα ) on Y , where α = (α1 , α2 , . . . , αk ) is
a multi-index with αi = dim(X) = dim(Y ). The constant depends only on Γ,
and it is equal to (−1)dim(X) · vol(Γ\D), with respect to Haar measure.
Σ
Mumford gave a prescription for constructing toroidal compactifications Γ\D ,
which depend on a choice of fan Σ, and how to extend automorphic bundles. By
studying singular metrics on these bundles, he proved a proportionality statement
for when Γ\D is non-compact.

Theorem. (Mumford) The same statement as above, replacing X = Γ\D with

Σ
Γ\D a toroidal compactification that is smooth as a stack, so Γ can have torsion.

Our main application of the principle will be to Ag , the moduli space of ppav’s,
and its toroidal compactifications. Here D = Hg is the Siegel upper-half space, and
the compact dual is Yg = Sp(2g, C)/P is the Lagrangian Grassmannian of (C2g , ω).
To compute the cohomology of Yg , fix a flag of isotropic subspaces:
0 ⊂ Z1 ⊂ Z2 ⊂ · · · ⊂ Zg ⊂ C2g ,
⊥
and extend it by taking Zg+i = Zg−i . Given a partition µ satisfying µi − 1 ≤
µi+1 ≤ µi fitting inside the g × g box (there are 2g such partitions), we define
Σµ (Z• ) = {L ∈ Yg : dim(L ∩ Zg+i−µi ) ≥ i} ⊂ Yg .

3Note that discreteness of Γ implies that for C ⊂ D compact, {g ∈ Γ : gC ∩ C 6= ∅} is finite.

 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 57

These are the only Schubert cycles in G(g, 2g) which lie in Yg because of the fact
that dim(L ∩ Zi⊥ ) = dim(L ∩ Zi ) + g − i.

Proposition. The Chow ring of Yg the following (graded) presentation:

A∗ (Yg ) ' Z[u1 , u2 , . . . , ug ]/(1 + u1 + u2 + · · · + ug )(1 − u1 + u2 − · · · ± ug ) − 1.
Proof. We have the tautological sequence on Yg given by
0 → S → OYg ⊗ C2n → Q → 0,
and using the symplectic form ω, we have Q ' S ∨ . The Whitney sum formula
then gives the desired relation, and one checks that there are no further relations
by comparing the Betti numbers with the numbers of strict partitions. 

Remark. A more concise way of writing the relation is ch2k (S) = 0 for k ≥ 1.

The restriction of the bundle S from Yg to the open subset Hg coincides with the
Hodge bundle E = π∗ (Ω1Xg /Ag ) pulled back from Ag . It is the automorphic bundle
associated to the standard representation of U (n). We define the tautological ring
R∗ (Ag ) ⊂ A∗ (Ag )Q to be generated by the Chern classes of the Hodge bundle.
Σ
Since E extends over any toroidal compactification Ag , we define the tautological
ring of smooth compactifications in the same way. The Proportionality Principle
implies (sending ui 7→ λi ) that
 Σ
R∗ Ag ' A∗ (Yg ),
but the degrees of 0-cycles differ by the constant multiple:
g
1 Y
c(Γ) = (−1)g(g+1)/2 g ζ(1 − 2k).
2
k=1
This is the reciprocal of a large integer (1/24 for g = 1 is familiar).

Theorem. R∗ (Ag ) ' A∗ (Yg−1 ).

Proof. By the localization sequence, we know that R∗ (Ag ) is a quotient of A∗ (Yg ).

It turns out that the kernel is (ug ). To see this, apply GRR4 to OXg on the universal
family π : Xg → Ag . On an abelian variety X, we have H i (OX ) ' ∧i H 1 (OX ), so
by relative Serre duality the LHS reads:
ch(π∗ [OXg ]) = ch(1 − E∨ + ∧2 E∨ − · · · ± ∧g E∨ ) = cg (E) · Td(E)−1 .
The other hand, the RHS is π∗ Td(Xg /Ag ) = 0, since TXg /Ag is pulled back from
the base. To show that there are no further relations, we use the algebraic fact
g(g−1)/2
that A∗ (Yg−1 ) is a Gorenstein ring with socle u1 , so it suffices to show that
g(g−1)/2
λ1 6= 0. To do this, use ampleness of λ1 (Baily-Borel) combined with the
existence of a complete subvariety of dimension g(g − 1)/2 over fields of charac-
teristic p, namely the locus of abelian varieties with p-torsion rank 0. The p-rank
is generically g, and drop in rank occurs with codimension 1. Since semi-abelian
varieties with non-trivial Gm part have positive p-rank, this locus is complete. 
4Applying GRR to the relative theta divisor instead, one can reprove the relation above on
Ag , but we avoid that computation by appealing to Hirzebruch-Mumford Proportionality.
58 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

20. 4/22/20 - Interactions between Kg and Ag

Both Kg and Ag are examples of Shimura varieties Γ\G/K. As such, they have
a rich collection of Shimura subvarieties corresponding to subgroups G0 ⊂ G. If
K 0 = K ∩ G0 and Γ0 = Γ ∩ G0 , then we have
Γ0 \G0 /K 0 ,→ Γ\G/K.
For this to be a subvariety, we need K 0 maximal and Γ0 arithmetic (use Baily-Borel).

In the case of K3 surfaces, let V ⊂ Λg ⊗ Q be a Q-subspace such that the pair-

ing is non-degenerate there. This implies that Λg ⊗ Q ' V ⊕ V ⊥ . Recall that
D = O(2, 19)/O(2) × O(19) is the space of positive definite 2-planes in (Λg ) ⊗ R.
Define a subdomain of codimension equal to dim(V ):
DV = {Z ∈ D : Z = Z ∩ V + Z ∩ V ⊥ },
GV ⊂ G the stabilizer of V and ΓV = Γ ∩ GV . In the case where V is a negative
definite line, then ΓV \DV is a Noether-Lefschetz divisor and GV = O(2, 18). In
the case where V is a negative definite subspace of dimension m, then ΓV \DV is
a higher NL-locus, parametrizing polarized K3 surfaces with Picard rank m, and
GV = O(2, 19 − m). These are classified (modulo Γ) by their Gram matrices M
with respect to a basis in Λ∨ 5
g , and they satisfy a modularity theorem as well:

Theorem. (Kudla-Millson, Zhang) For m ≥ 1, let Sm (t) be the set of positive

semi-definite m × m matrices of rank t with Q-coefficients. Then the power series
X X
F (q) = (−λ)m + (−λ)m−1 [ΓM \DM ]q M + · · · + [ΓM \DM ]q M .
M ∈Sm (1) M ∈Sm (m)

is a Siegel modular form (weight 21/2, level 2g − 2) with coefficients in Am (K

e g ).

Definition. A Siegel modular form is a section of λ⊗k on Ag (g = m, k = 21/2

in the theorem) which satisfies a polynomial growth condition at the cusps. More
explicitly, they are holomorphic functions on the space of symmetric g × g matrices
τ with positive definite imaginary part. The notation q M denotes eπi tr(M τ ) .

The proof uses a technique called theta lifting, which is a way of passing between
automorphic forms for symplectic and orthogonal groups. The key fact is that if
W is a symplectic vector space and V is an orthogonal vector space, then W ⊗ V is
naturally symplectic, and we have Sp(W ) × O(V ) ⊂ Sp(W ⊗ V ). The Siegel theta
function on Sp(W ⊗ V ) restricts to a function Θ, which serves as a correspondence
between automorphic forms for the two groups:
Sp(W ) ← Sp(W ) × O(V ) → O(V ).
A second interaction between Kg and Ag stems from the exceptional isomorphism
SO(2, 3)+ ' Sp(4, R).

5We have only given the statement for K with coefficients in Chow groups, but Kudla-Millson
g
prove a statement for all O(p, q) symmetric spaces with coefficients in cohomology.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 59

This implies that higher NL-loci of dimension 3 are Shimura varieties for Sp(4, R).
Using the structure theorem for R∗ (A2 ) from last time, we have
λ21 = 2λ2 = 0 ∈ R2 (A2 ).
Theorem. (v.d.Geer-Katsura) If Γ\D is a Shimura variety for O(2, n), then
λn−1 = 0 ∈ A∗ (Γ\D)Q
In particular, any complete subvariety of Kg has dimension ≤ 17.

Proof. We argue by induction on the dimension n, with the base case n = 3

coming
P from the exceptional isomorphism above. Write λ as a linear combination
ci Hi of NL-divisors Hi which are themselves Shimura varieties.
X X
λn−1 = ci Hi λn−2 = ci ι∗ (λn−2
Hi ) = 0.

Here ι denotes the inclusion the relevant Hi . 

A third interaction is the Kuga-Satake construction, which produces an enormous

abelian variety from a K3 surface. More precisely, it produces a weight one PHS
of dimension 2n−2 from a K3-type weight two PHS H of dimension n. The even
Clifford algebra Cl+ (HR ) is a real vector space of dimension 2n−1 , and one uses the
K3 decomposition to produce a complex structure on it by hand, and then a PHS
of weight one. This gives an injection
Kg ,→ A219 ,δ
where δ is a non-principal polarization type depending g. The construction is useful
for importing arithmetic and cycle theoretic results about abelian varieties to K3
surfaces. Examples include the Tate Conjecture, Shafarevich Conjecture, versions
of the Hodge conjecture, etc.
60 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

21. 4/27/20 - Quillen K-Theory

The last topic for the course will be higher Chow groups and their cousins, higher
algebraic K-groups. The ultimate goal is to extend the localization sequence toward
the left. We would like to define groups A∗ (X, n) such that for any nested pair of
schemes (X, Z) with U = X r Z open, we have a long exact sequence:
· · · → A∗ (Z, 1) → A∗ (X, 1) → A∗ (U, 1) → A∗ (Z, 0) → A∗ (X, 0) → A∗ (U, 0) → 0,
where A∗ (X) = A∗ (X, 0) recovers the usual Chow groups. Bloch defined the higher
Chow groups in 1986. He was primarily motivated by the parallel story in alge-
braic K-theory due to Quillen around 1972. We have already encountered the
Grothendieck group K0 (X) = K0 (Coh(X)), which is (rationally) isomorphic to the
Chow group A∗ (X, 0) via the natural transformation τX of Grothendieck-Riemann-
Roch. A similar natural isomorphism τX (n) exists between Kn (X)Q and A∗ (X, n)Q .

Definition. Let C be a small category. The nerve of C is defined to be the

simplicial set N (C) whose p-simplices are compositions
X0 → X1 → . . . Xp .
The classifying space BC is the geometric realization of the nerve.

Example. If G is a group, we can form the category G with a single object.

Then the classifying space of the category matches usual notion: BG = BG.

Definition. If X is an object of C, we can define the homotopy groups:

πn (C, X) := πn (BC, X).
This notion of classifying space has a number of nice properties:
• BC = BCop .
• B(C × C0 ) = B(C) × B(C0 ).
• A functor F : C → C0 induces a cellular map BF : BC → BC0 .
• A natural transformation F ⇒ G induces a homotopy from BF to BG.
• If F has a right or left adjoint, then BF is a homotopy equivalence.
• If C has an initial or final object, then BC is contractible.
Definition. Let C be an abelian category. Quillen’s Q-construction gives a category
QC whose objects are the same as those of C. A morphism in QC from M → N is
defined by a pair: M 0 ,→ M , M 0  N . The composition of morphisms is given by:
MO

MO 0 /N
O

M 0 ×N N 0 / N0 / L,

where the vertical arrows are monos and the horizontal arrows are epis.

Note: 0 is the final object of C, but not of QC because you lose uniqueness.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 61

Theorem. The fundamental group π1 (QC) is isomorphic to K0 (C).

Proof. Given M ∈ K0 (C), we have two natural morphisms in M → 0 in QC,

M ⊃ M → 0 and M ⊃ 0 → 0, which give a loop γM in BQC. Given a short exact
sequence 0 → M 0 → M → M 00 → 0, the composition of the loops γM 0 and γM 00
is homotopic to γM (draw a picture of a 2-sphere with three disks cut out). Con-
versely, send a loop (based at 0) consisting of arrows with alternating orientation
in QC to the alternating sum of the vertices as elements of K0 (C). 

Definition. The algebraic K-groups of an abelian category C are defined by:

Kn (C) = πn+1 (QC, 0).
Definition. If R is a regular ring, define Kn (R) = Kn (Modf (R)). If R is not
regular then we instead use the subcategory generated by projective R-modules.

Theorem. (Dévissage) Let B ⊂ C be a full abelian subcategory such that any

object M ∈ C admits a finite filtration whose successive quotients are in B. Then
the induced map BQB → BQC is a homotopy equivalence, so Kn (B) ' Kn (C).

Proof. By the long exact sequence of homotopy groups, it suffices to show that
the homotopy fiber of BQB → BQC is contractible. Recall that the homotopy
fiber of a continuous map of pointed spaces f : B → C is the space of pairs
(b ∈ B, γ : f (b) → 0). To avoid alternating arrow orientations, Quillen proves that
a certain homotopy fiber category is contractible for any choice of basepoint. 

Example. Let G be a reductive group. Taking C the category of G-modules

and B the collection of irreps α for G, we obtain
M
Kn (C) = Kn (End(α)),
α
and by Schur’s Lemma each End(α) is a division ring.

Example. Let I ⊂ R be a nilpotent ideal. Then Kn (R/I) ' Kn (R).

Theorem. (Localization) Let S ⊂ C be a Serre subcategory, that is a full abelian

subcategory that is closed under extensions. Then there is a fibration sequence
BQS → BQC → BQ(C/S) which induces a long exact sequence on homotopy:
· · · → Kn+1 (C/S) → Kn (S) → Kn (C) → Kn (C/S) → . . .
Recall: The localized category C/S is obtained by formally inverting all morphisms
whose kernel and cokernel lie in S.

Corollary. Let R be a Dedekind domain with fraction field F . Taking C the

category of R-modules and S to be the category of torsion R-modules, we obtain:
M
· · · → Kn+1 (F ) → Kn (R/m) → Kn (R) → Kn (F ) → . . .
m
We have used the Dévissage Theorem to describe Kn (S) in terms the residue fields.
62 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

22. 4/29/20 - Bloch’s Formula

Last time, we showed that π1 (BQC, 0) was the Grothendieck group. Is there a
similarly concrete description of K1 (C) = π2 (BQC, 0)?

Theorem. For a commutative ring R, K1 (R) ' GL(R)ab .

Proof. Construct a map GL(R) → π2 (BQC, 0) by sending an element T ∈ GLn (R)

to the sphere in BQModf (R) built by gluing together two cones over the loop
A : R⊕n → R⊕n , obtained by composing T with the two canonical Q-morphisms
R⊕n ⊃ 0 → 0 and R⊕n ⊃ R⊕n → 0. Since π2 is abelian, this map factors through
GL(R)ab . We leave the isomorphism fact as an exercise. 

Theorem. (Whitehead) For R a local ring or Euclidean domain, GL(R)ab ' R× .

Theorem. (Milnor) For R = OF with F a number field, GL(R)ab ' R× .

For a scheme X, let Mr (X) ⊂ Coh(X) be the Serre subcategory of coherent sheaves
whose support has codimension ≥ r. Then the localization Mr (X)/Mr+1 (X) has
a dévissage over all points x ∈ X of codimension r:
M M
··· → Kn+1 (F (x)) → Kn (Mr+1 (X)) → Kn (Mr (X)) → Kn (F (x)) → . . .
x∈X (r) x∈X (r)

Here we have fixed the codimension r. If instead we use the full filtration M• (X)
of Coh(X), the long exact sequences above combine to produce an exact couple
whose spectral sequence6 relates the K-theory of functions fields with the K-theory
of Coh(X), filtered by K-theory of Mr (X).
M
E1pq = K−p−q (F (x)) ⇒ K−q (X).
x∈X (p)

Lemma. (Quillen) The map Kn (Mr+1 (X)) → Kn (Mr (X)) vanishes if X is a

regular local scheme over a field (this vanishing is called Gersten’s Conjecture).

Corollary. For X as above, the following Gersten sequence is exact:

M M M
0 → Kn (X) → Kn (F (x)) → Kn−1 (F (x)) → · · · → K0 (F (x)) → 0.
x∈X (0) x∈X (1) x∈X (n)

Proof. The vanishing of the differential implies that the E2 = E∞ in the spectral
sequence, and furthermore the coniveau filtration on Kn (Coh(X)) is the stupid one.
This means that the only nonvanishing terms in E2 are E20q ' Kq (X). This is the
0-th cohomology of the E1 -page complex which is otherwise exact. 

Theorem. (Bloch’s Formula) Given a smooth variety X, consider the sheaf Kn (X)
of abelian groups associated to the presheaf U 7→ Kn (O(U )). The classical Chow
groups are computed by sheaf cohomology:
An (X, 0) = H n (X, Kn (X)).

6The construction of the spectral sequence is very analogous to the Serre spectral sequence.
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 63

Proof. By sheafifying the Gersten sequence, we have a resolution

M M M
0 → Kn (X) → ix∗ Kn (F (x)) → ix∗ Kn−1 (F (x)) → · · · → ix∗ K0 (F (x)) → 0.
x∈X (0) x∈X (1) x∈X (n)
Each sheaf in the resolution is a sum of skyscraper sheaves, so this is a flasque
resolution. As a result,
 
M M
H n (X, Kn (X)) ' coker  K1 (F (x)) → K0 (F (x))
x∈X (n−1) x∈X (n)

The algebraic K-theory of fields is far simpler than the general case: K0 (F ) ' Z
and K1 (F ) ' F × . So the map in question goes from
M
F (x)× → Z n (X),
x∈X (n−1)
and is actually the map div from Lecture 3. Quillen reduces to the case of a DVR,
where K1 (F ) → K0 (R/m) is the ord function. 

Remark. Bloch’s Formula is essentially a definition for n = 0, 1: H 0 (X, Z) ' Z,

H 1 (X, Gm ) = Pic(X) ' A1 (X), and it was first proved by Bloch for n = 2 using
dilogarithms. Quillen proved the statement for all n.

The Steinberg group is the universal central extension of E(R) ⊂ GL(R), and its
center is K2 (R) = H2 (E(R), Z): 0 → K2 (R) → St(R) → E(R) → 1.
a
More explicitly, St(R) is the free group on Xij , where i, j ∈ N and a ∈ R, modulo
the relations satisfied by elementary matrices:
a b a+b
Xij Xij = Xij ;
a b ab
[Xij , Xjk ] = Xik , i 6= k;
a b
[Xij , Xkl ] = 1, i 6= l, j 6= k.
The kernel of St(R) → E(R) is a quotient of R∗ ⊗ R∗ , by the Steinberg relations:
{a, b} = {b, a}−1 ; {a, b}{a0 , b} = {aa0 , b}; {a, 1 − a} = 1.
For X/k a regular scheme with function field F , there is a well-defined pairing
d log ∧ d log : F × ⊗ F × → Ω2F/k .
It satisfies the Steinberg relations, so it descends to K2 (F ). When restricted to
K2 (X) via the Gersten resolution, it lifts to a map K2 (X) → Ω2X/k . Bloch proves
that the induced map
H 2 (X, K2 (X)) → H 2 (X, Ω2X/k )
is the cycle class map via his isomorphism H 2 (X, K2 (X)) ' A2 (X).

Remark. There are several equivalent definitions of higher algebraic K-theory:

the +-construction, the Q-construction, and then S∗ -construction. They each give
an infinite loop space (or spectrum), and the algebraic K-groups are equal to its
homotopy groups. The Whitehead bracket πn ⊗ πn → πn+m−1 gives a product on
the algebraic K-groups:
Kn ⊗ Km → Kn+m .
64 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

23. 5/4/20 - Higher Chow Groups

We start by defining an algebraic version of an n-simplex (without the positivity
assumption on coordinates). For each n ≥ 0, consider the affine space
 X 
∆n := Spec k[t0 , t1 , . . . , tn ]/ 1 − ti ' Ank .
For each ordered map ρ : {0, 1, . . . , m} → {0, 1, . . . , n}, define p̃ : ∆m → ∆n by
X
ρ̃∗ (ti ) = tj .
ρ(j)=i

If ρ is injective, ρ̃ is called a face map (facet if m = n − 1). The higher cycle groups
Z(X, n) ⊂ Z(X × ∆n ) are the subgroups generated by cycles which intersect each
face X ×∆m dimensionally transversely. If ∂i denotes the pullback to the i-th facet,
we can define an algebraic boundary map
Xn
∂= (−1)i ∂i : Z(X, n) → Z(X, n − 1).
i=0

Definition. The higher Chow groups A(X, n) are defined by taking the homology:
Z(X, n) = Hn (Z(X, ∗))
Bloch verifies the functoriality properties (proper pushforward, flat pullback) and
the property that A∗ (X, n) ' A∗ (X × A1 , n) directly from the definitions.

Theorem. (Localization) For a quasi-projective scheme X over k and a closed

subscheme Y ⊂ X with complement U , there is a long exact sequence
· · · → A∗ (Y, n) → A∗ (X, n) → A∗ (U, n) → A∗ (Y, n − 1) → . . .
Proof. Since higher Chow groups are the homology of a complex, we expect the
long exact sequence to come from a short exact sequence of cycle groups. In fact
the sequence
0 → Z(Y, n) → Z(X, n) → Z(U, n)
is only left exact for degrees n > 0. Nonetheless, the injective map of complexes
Z(X, ∗)/Z(Y, ∗) → Z(U, ∗)
is a quasi-isomorphism. The proof of this is quite technical and uses higher moving
lemmas, relative to the facets. 

Corollary. For any vector bundle E → X, the flat pullback A∗ (X, n) → A∗ (E, n)
is an isomorphism.

Theorem. (Gysin pullback) The functor A∗ (−, n) is contravariant for morphisms

of smooth, quasi-projective varieties.

Proof. Let f : X → Y be a morphism of smooth, quasi-projective varieties.

Define the subcomplex
Z∗f (Y ) → Z∗ (Y )
to be generated by cycle classes dimensionally transverse to f . Using higher moving
lemmas, Bloch proves that this is a quasi-isomorphism. Define the Gysin pullback
on transverse cycles in the obvious way, using f −1 with multiplicities. 
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 65

Using triangulations of product simplices, Bloch defines an external product

A∗ (X, n) ⊗ A∗ (Y, m) → A∗ (X × Y, n + m).
Composing with the Gysin pullback via the diagonal, we get an internal product
A∗ (X, n) ⊗ A∗ (X, m) → A∗ (X, n + m),
which generalizes the intersection product.

Theorem. For any projective bundle PE → X, if ζ ∈ A1 (PE, 0) is the relative

hyperplane class, then we have an isomorphism (for r + 1 = rk E).
Mr
A∗ (PE, n) ' ζ i · A∗ (X, n) ' A∗ (PE, 0) ⊗ A∗ (X, n),
A∗ (X,0)
i=0
Proof. Using the localization sequence with induction on dim(X), it suffices to
prove the statement for trivial bundles. This is done using the localization sequence
again and induction on r. 

Theorem. In the codimension 1 case, we have

A1 (X, n) = 0, n ≥ 2;
×
A1 (X, 1) = H 0 (X, OX );
A1 (X, 0) = Pic(X).
Proof. Define the algebraic n-sphere to be
S n = ∆n ∪n ∆n ,
∂∆
where ∂∆n is the union of the facets. Bloch proves that
A1 (X, n) ' Pic(X × S n )/Pic(X).
The theorem follows easily from there (note that S 0 is two points). 

Exercise. If X is an algebraic surface, then A2 (X, 1) is generated by formal sums

X
(Ci , fi ),
i
×
P
where Ci ⊂ X is a curve, fi ∈ k(Ci ) , and i div(fi ) = 0 as an element of Z0 (X).

Theorem. (Totaro) A2 (Spec k, 2) ' K2 (k). More generally,

An (Spec k, n) ' KnM (k),
the Milnor K-groups, which form the simplest part of the higher K-theory of fields.

There is a higher Chern character map ch(n) : Kn (X) → A∗ (X, n) due to Gillet.
It is completely formal, and uses the Quillen +-construction of algebraic K-theory,
adapted to the world of simplicial schemes. Multiplying by the Todd class of X,
τ (n) : Kn (X) → A∗ (X, n) satisfies a higher Grothendieck-Riemann-Roch for proper
morphisms X → Y , and gives an isomorphism over Q.
66 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

24. 5/6/20 - Motives and L-Functions

A famous conjecture of Birch and Swinnerton-Dyer says that for an elliptic curve
E defined over Z, its Hasse-Weil L-function LE (s) has a zero at s = 1 of order
determined by the rank of the Mordell-Weil group E(Q):
ords=1 LE (s) = rk E(Q).
The L-function is an infinite Euler product over primes p of local factors related to
the number of points of E modulo p. The L-function is closely related to the zeta
function, which encodes the full motive of E instead of only the weight 1 part:
ZE (s) = ζ(s)ζ(s − 1)LE (s)−1 .
How can we generalize this to varieties which are not group schemes? Recall that
A0 (E) ' Pic(E) fits into the short exact sequence
deg
0 → Pic0 (E) → Pic(E) −→ Z → 0,
and Pic0 (E) ' E(Q) after choosing an origin. Every variety has Chow groups.

Conjecture. (Lichtenbaum-Soulé) If X is a smooth variety defined over Z, then

for d = dim(X) and ZX (s) its zeta function, we have
X
−ords=n ZX (s) = (−1)i rk An (X, i).
i≥0

In general we do not even know whether this sum is finite!

Remark. It is important to use cycles over Q in the definition for the Chow
groups; over C they are often infinite rank.

The general philosophy is that motivic cohomology groups (defined by Voevodsky)

are supposed to go here, but higher Chow groups are a good approximation.
p
HM (X, Q(q)) ' Aq (X, 2q − p) ⊗ Q.
Beilinson’s Conjecture predicts the order of vanishing of an L-function at certain
integer points s, and the Bloch-Kato Conjecture predicts the leading coefficient at
those points. Both statements are formulated in terms of a regulator:
p p
R(p, q) : HM (X, Q(q)) ⊗ C → HD (X, Q(q)) ⊗ C.

Conjecture. (Beilinson) If X be a smooth projective variety defined over Z, then

for d = dim(X) and LX (s) its L-function in weight d, we have
ords=d+1−q LX (s) = dim ker(R(d + 1, q)).
for d + 1 < 2q. The critical case d + 1 = 2q (ordinary Chow group) is slightly larger:
ords=q LX (s) = rk Aq (X)hom .
In that case, the Beilinson regulator fits into the middle of the diagram
0 / Aq (X)hom / H 2q (X, Z(q) / Hdg(q) /0
M

AJ R(2q,q) =
  
0 / Jac(X) / H 2q (X, Z(q)) / Hdg(q) / 0.
D
 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY 67

Definition. The Deligne cohomology is defined by a hypercohomology group:

p q−1
HD (X, Z(q)) ' Hp (X, 0 → ZX (q) → OX → Ω1X → . . . ΩX → 0).
Its key property is the long exact sequence:
p p−1
· · · → HD (X, Z(q)) → H p (X, (2πi)q Z) → H p (X, C)/F q → HD (X, Z(q)) → . . .
which produces the short exact sequence above, assuming the Hodge Conjecture.

The Bloch-Kato conjecture gives the leading coefficient of LX (s) at the special
points. It is too complicated to state here, but in the case where X = OF for a
number field F it recovers the class number formula:

2r1 · (2π)r2 · RF · hF
lim (s − 1)ζF (s) = √ ,
s→1 ωF DF
as well as the Dirichlet unit theorem:
rk OF× = r1 + r2 − 1.
68 INTERSECTION THEORY IN ALGEBRAIC GEOMETRY

References
[1] S. Bloch. K2 and algebraic cycles. Ann. Math., 99:349–379, 1974.
[2] S. Bloch. Algebraic cycles and higher K-theory. Adv. Math., 61:267–304, 1986.
[3] D. Edidin. Riemann-Roch for Deligne-Mumford stacks. In B. Hassett, J. McKernan, J. Starr,
and R. Vakil, editors, A Celebration of Algebraic Geometry, volume 18, pages 241–266. Clay
Mathematics Proceedings, 2013.
[4] D. Edidin and W. Graham. Equivariant intersection theory. Invent. Math., 131:595–634, 1998.
[5] D. Edidin and W. Graham. Riemann-Roch for equivariant Chow groups. Duke Math. J.,
102(3):567–594, 2000.
[6] D. Eisenbud and J. Harris. 3264 and All That: A Second Course in Algebraic Geometry.
Cambridge University Press, 2016.
[7] C. Fontanari and E. Looijenga. A perfect stratification of Mg for g ≤ 5. Geom. Dedicata,
136:133–143, 2008.
[8] W. Fulton. Intersection theory. Springer-Verlag, 1984.
[9] H. Gillet. Riemann-Roch theorems for higher K-theory. Adv. Math., 40:203–289, 1981.
[10] J. Harris and I. Morrison. Moduli of curves. Springer-Verlag, 1998.
[11] A. Kresch. Cycle groups for Artin stacks. Invent. Math., 138:495–536, 1999.
[12] S. Kudla and J. Millson. Intersection numbers of cycles on locally symmetric spaces and
Fourier coefficients of holomorphic modular forms in several complex variables. Publ. Math.
IHÉS, 71:121–172, 1990.
[13] D. Maulik. Supersingular K3 surfaces for large primes. Duke Math. J., 163(13):2357–2425,
2014.
[14] D. Mumford. Hirzebruch’s proportionality theorem in the non-compact case. Invent. Math.,
42:239–272, 1977.
[15] H. Nielsen. Diagonalizably linearized coherent sheaves. Bull. Soc. Math. Fr., 102:85–97, 1974.
[16] K. O’Grady. On the Picard group of the moduli space for K3 surfaces. Duke Math. J.,
53(1):117–124, 1986.
[17] D. Quillen. Higher algebraic K-theory I. In H. Bass, editor, Higher K-theories, pages 85–147.
Springer, 1972.
[18] R.W. Thomason. Une formula de Lefschetz en K-théorie equivariante algébrique. Duke Math.
J., 68:447–462, 1992.
[19] G. van der Geer. The cohomology of the moduli space of abelian varieties. In The Handbook
of Moduli, Advanced Lectures in Mathematics, pages 415–458. Int. Press, 2011.
[20] G. van der Geer and T. Katsura. Note on tautological classes of moduli of K3 surfaces. Mosc.
Math. J., 5(4):775–779, 2005.
[21] A. Vistoli. Intersection theory on algebraic stacks and on their moduli spaces. Invent. Math.,
97:613–670, 1989.