Patrick Lei

All notes

Intersection theory

Columbia, Spring 2021.

Intersection Theory Learning Seminar
Spring 2021
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

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

Acknowledgements

I would like to thank Nicolás Vilches for pointing out that I was missing the accent in his name.

1 Caleb (Jan 22): Rational Equivalence

The idea of intersection theory is to define a well-defined intersection product on a scheme or variety that behaves like the intersection form in ordinary homology of a manifold.

Example 1.0.1.

Let f,g∈𝔸k2 be two plane curves. Then we can define the intersection multiplicity by

which is the dimension of the intersection scheme. However, this does not generalize, so we will need to use Serre’s formula, which involves the Tor functor.

1.1 Chow Groups

We want to define zeroes and poles for a singular variety. If V is codimension 1 in X for a nonsingular variety X, then consider r∈K⁢(X)×, where K⁡(X)=Frac⁡(𝒪V,X), where this local ring is a discrete valuation ring. Then we can write (r)=∑ordV⁢(r)⁢[V], which is a Weil divisor. For a curve, we can compute ordV⁢(r)=dimk(𝒪V,X/(r)). In general, we can write ordV⁢(r)=lenA⁢(A/(r)), where A=𝒪V,X.

Example 1.1.1.

If X=𝔸3 and r=y2⁢(x−2)⁢(x⁢y+1)z⁢y+x, we see that along y=0, we have A=k⁢[x,y,z](y) and thus lemA⁢(A/r)=2.

Proposition 1.1.2.

The map ordV:R⁢(X)∗→ℤ is a homomorphism.

Proposition 1.1.3.

For any r, there are finitely many V such that ordV⁢(r)≠0.

Examples 1.1.4.
  1. 1.

    Let f be irreducible and f,g be plane curves. Then i⁡(P,F⋅G)=ordP⁢(g¯).

  2. 2.

    Let X~→X,V~→V be normalizations. Then ordV⁢(r)=∑ordV~⁢(r)⁢[K⁡(V~):K⁡(V)].

Now we will define rational equivalence. Let W be a (k+1)-dimensional subvariety of X and r∈R⁢(W)×. Then we may obtain cycles of the form [div⁡(r)], and such cycles form the group Ratk⁡X.

Definition 1.1.5.

The Chow groups of X are defined by Ak⁢(X)≔Zk⁢(X)/Ratk⁡X.

Example 1.1.6 (Not Mayer-Vietoris).

Let X1,X2 be closed subschemes of X. Then the sequence

is exact.

1.2 Pushforward of Cycles

Let f:X→Y be a morphism. We want to define a pushforward Ak⁢(X)→Ak⁢(Y). This would imply that Ratk⁢(X)→Ratk⁢(Y). This is true when f is proper. Then V→f⁡(V)=W is a map of cycles because proper implies universally closed. We obtain a field K⁡(W)⊆K⁡(V), which is finite if V,W are of equal dimension. Then we define

Definition 1.2.1.

The pushforward of a cycle V is f∗⁢[V]=deg⁡(V/W)⁢[W].

Theorem 1.2.2.

Let f is proper. If α∼0, then f∗⁢α∼0.

Proposition 1.2.3.

Let f be proper and surjective and let r∈K⁢(X)×. Then

  1. 1.

    f∗⁢[div⁡(r)]=0 if dimY<dimX;

  2. 2.

    f∗⁢[div⁡(r)]=div⁡(N⁡(r)) if dimY=dimX.

Proof.

Consider the map ℙ1→Spec⁡k. Because the order map is a homomorphism, let r∈k⁡[t]⊂k⁡(t) be an irreducible polynomial, where p has degree n. Then div⁡(r)=[P]−n⁡[∞] and then we have

Now suppose f is finite. Given W⊂Y of codimension 1, write (A,𝔪)=(𝒪W,Y,𝔪W,Y). Construct a domain B such that B/A is finite and B⊗AK⁡(Y)=K⁡(X) such that {Ui}→W are obtained by B𝔪i=𝒪Ui,X. Here, we need to show that

Now if r∈B, we want φ:B→𝑟B, and then the LHS becomes ℓA⁢(coker⁡φ)=ordW⁢(det⁡φK). In general, we need to apply results from EGA III to show that B exists.

Now in the different-dimensional case, we replace Y with Spec⁡K⁡(Y) and reduce to the case of a curve over Spec⁡K⁡(Y). □

Corollary 1.2.4 (Bezout’s Theorem).

Let k=k¯ and suppose f,g are plane curves of degree m,n with no common component. Then ∑i⁡(P,F⋅G)=m⁢n.

Proof.

Assume f is irreducible and replace G with G′=Ln. Then G/G′=r∈K⁡(CF) and therefore

so we may as well assume F,G are lines and then the result is obvious. □

Definition 1.2.5.

We can define the degree deg⁡(α) to be deg⁡(α)=∫Xα=∑nP⁢[K⁡(P):K].

1.3 Alternative Definition of Rational Equivalence

Let X be a scheme with irreducible components X1,…,Xt. Then let mi≔ℓ𝒪X,x⁢(𝒪X,x). We define the fundamental class [X]=∑i=1tmi⁢[Xi].

Proposition 1.3.1.

A cycle α∈Zk⁢(X) is rationally equivalent to 0 if and only if there exist (k+1)-dimensional subvarieties V1,…,Vt of X×ℙ1 such that fi:Vi→ℙ1 are dominant and α=∑[Vi⁢(0)]−[Vi⁢(∞)].

Sketch of Proof.

Suppose α=[div⁡(r)] for some r∈K⁢(W)×. Now r defines W→ℙ1, so we will define V=Γ⁡(r)¯ to be the closure of the graph. This gives a dominant rational map f:V⇢ℙ1 and thus [div⁡(r)]=p∗⁢(div⁡(f))=[V⁡(0)]−[V⁡(∞)]. The other direction is easy. □

Remark 1.3.2.

This discussion and more machinery allows us to show that if f:X→Y is flat of relative dimension n, then we can define pullbacks of cycles. Here, if α∼0 is a k-cycle on Y, then f∗⁢α∼0 in Zk+n⁢(X), so we have

2 Avi (Jan 29): Intersecting with divisors and the first Chern class

Note: These are the speaker’s notes. Minor edits to the source were made.

2.1 Cartier and Weil divisors

Let X be a variety of dimension n over a field k. We want to introduce two notions of divisors, one familiar from the last chapter.

Definition 2.1.1.

A Weil divisor of X is an n−1-cycle on X, i.e. a finite formal linear combination of codimension 1 subvarieties of X. Thus the Weil divisors form a group Zn−1⁢X.

Definition 2.1.2.

A Cartier divisor consists of the following data:

  • •

    an open cover {Uα} of X;

  • •

    for each α a nonzero rational function fα on Uα, defined up to multiplication by a unit , i.e. a function without zeros or poles, such that for any α,β we have fα/fβ a unit on Uα∩Uβ.

Like the Weil divisors, the Cartier divisors form an abelian group: ({Uα,fα})+({Uα,gα})=({Uα,fα⁢gα}) (we can assume that the open covers are the same, since if not they refine to {Uα∩Vβ}). We call this abelian group Div⁡X.

Given a Cartier divisor D=({Uα,fα}) and a codimension 1 subvariety V of X, we define

for α such that Uα∩V is nonempty; since each fα is defined up to a unit, this order is well-defined. We define the associated Weil divsior

This defines a homomorphism

For any rational function f on X, we get a principal Cartier divisor div⁡(f) by choosing any cover {Uα} and defining fα=f|Uα. It is immediate that the image [div⁡(f)] of this divisor under the map to Zn−1⁢X is the Weil principal divisor. Say that two Cartier divisors D and D′ are linearly equivalent if D−D′=div⁡(f) for some f; then we define Pic⁡X to be the group of Cartier divisors modulo linear equivalence, and the above then shows that the map Div⁡X→Zn−1⁢X descends to a map Pic⁡X→An−1⁢X. This map is in general neither injective nor surjective.

Notice that the definition of a Cartier divisor yields that of a line bundle on X: given a divisor D=({Uα,fα}), define a line bundle L=𝒪⁡(D) to be trivialized on each Uα with transition functions fα/fβ. Two Cartier divisors D and D′ are linearly equivalent if and only if 𝒪⁡(D)=𝒪⁡(D′), and so we get the alternate description of Pic⁡X as the abelian group of line bundles on X with group operation given by the tensor product. Conversely, given a line bundle L, this determines a Cartier divisor D⁡(L) up to some additional data: a nonzero rational section s of L. Therefore we can also think of Cartier divisors as the data of a line bundle together with a nonzero rational section.

We define the support supp⁡D or |D| of a Cartier divisor D to be the union of codimension 1 subvarieties V of X such that fα is not a unit for Uα nontrivially intersecting V, i.e. ordV⁡D is nonzero.

We say that a Cartier divisor D=({Uα,fα}) if all of the fα are regular, i.e. have no poles.

2.2 Pseudo-divisors

In general, Cartier divisors are not well-behaved under pullbacks (although line bundles are). In particular, given the data of a line bundle L and a nonzero rational section s and a morphism f:Y→X, there is no guarantee that the pullback f∗⁢s is nonzero. Therefore we enlarge the notion to make it behave better: let L be a line bundle on X, Z⊂X be a closed subset, and s be a nowhere vanishing section of L restricted to X−Z, or equivalently a trivialization of L|X−Z. A pseudo-divisor on X consists of the data of such a triple (L,Z,s), up to the following equivalence: two triples (L,Z,s) and (L′,Z′,s′) define the same pseudo-divisor if Z=Z′ and there exists an isomorphism σ:L→L′ such that restricted to X−Z we have σ∘s=s′. Note that this is well-behaved under pullback.

Example 2.2.1.

Let D=({Uα,fα}) be a Cartier divisor, with support |D|. Then each fα away from |D gives a local section of the associated line bundle 𝒪⁡(D), and so these glue to a section sD of 𝒪⁡(D) on X−|D|; this makes (𝒪⁡(D),|D|,sD) a pseudo-divisor.

We say that a Cartier divisor D represents a pseudo-divisor (L,Z,s) when |D|⊆Z and there exists an isomorphism σ:𝒪⁡(D)→L such that restricted to X−Z we have σ∘sD=s, with notation as above.

Lemma 2.2.2.

If X is a variety, then every pseudo-divisor (L,Z,s) on X is represented by a Cartier divisor D. If Z⊊X, then D is unique; if Z=X, then D is unique up to linear equivalence.

Proof.

If Z=X, then s is a section on X−X={}, and so a pseudo-divisor is just a line bundle; and we saw in the previous section that the group of Cartier divisors up to linear equivalence is isomorphic to the group of line bundles, so L corresponds to a unique linear equivalence class of Cartier divisors.

If Z≠X, let U=X−Z. As above, choose a Cartier divisor D=({Uα,fα}) with 𝒪⁡(D)≃L. The section s consists of a collection of functions sα on U∩Uα such that sα=fα/fβ⋅sβ on U∩Uα∩Uβ; thus sα/fα=sβ/fβ on each intersection, i.e. there exists some rational function r such that sα/fα=r on each U∩Uα. Then D′≔D+div⁡(r) is the Cartier divisor ({Uα,fα⁢r}) and by definition fα⁢r=sα on each U∩Uα; therefore using the definition above sD′=s. Since D′ is linearly equivalent to D, it corresponds to the same line bundle, and since r is regular on each Uα the support of div⁡(r) is contained in Z; therefore D′ represents (L,Z,s).

For uniqueness, suppose that two Cartier divisors D1=({Uα,fα}) and D2=({Vβ,gβ}) both represent (L,Z,s). Then similarly there must exist some rational function r such that r⁢fα=r⁢gβ on each Uα∩Vβ. But since sD1=sD2=s, if Z≠X, i.e. U is nonempty, then sD1 and sD2 must agree on every U∩Uα∩Vβ, and so r restricted to U must be 1; since f is rational it follows that f=1 and D1=D2. □

For any pseudo-divisor D=(L,Z,s), as for Weil divisors we will write 𝒪⁡(D)=L, |D|=Z, and sD=s.

If D=(L,Z,s) and D′=(L′,Z′,s′) are two pseudo-divisors, we can define their sum

This agrees with the sum on Cartier divisors, except that the supports may be larger in this case. Similarly defining

makes the set of pseudo-divisors into an abelian group.

Given a pseudo-divisor D on a variety X of dimension X, we can define the Weil class divisor [D] by taking D~ to be the Cartier divisor which represents D and setting [D]≔[D~], the associated Weil divisor from the previous section. The above lemma shows that this yields a well-defined element of An−1⁢X; this gives a homomorphism from the group of pseudo-divisors to An−1⁢X.

2.3 Intersecting with divisors

Let X be a variety of dimension n, D be a pseudo-divisor on X, and V be a subvariety of dimension k. Let j:V↪X be the inclusion of V into X; then the pullback j∗⁢D is a pseudo-divisor on V with support V∩|D|. We define the class D⋅[V] in Ak−1⁢(V∩|D|) given by the Weil class divisor of j∗⁢D:

For any closed subscheme Y⊂X containing V∩|D|, we can also view this as an element of Ak−1⁢Y; we will also denote this by D⋅[V].

Let α=∑VnV⋅V be a k-cycle on X, with support |α the union of the subvarieties V such that nV is nonzero. For a pseudo-divisor D on X, we define the intersection class D⋅α in Ak−1⁢(V∩|D|) by

As above, we can also view this as an element of Ak−1⁢Y for any Y containing |α|∩|D|.

We will apply this in two main cases. First: f |D|=X, then the data of D=(L,X,s) is just that of a line bundle as above; in this case the action of D on a k-cycle α is called that of the first Chern class, written D⋅α=c1⁢(L)∩α.

Second: if i:|D|↪X is the inclusion of |D| into X, then D⋅α is called the Gysin pullback i∗⁢α.

Theorem 2.3.1.

Let X be a scheme, D be a pseudo-divisor on X, and α be a k-cycle on X.

  1. (a)

    Let α′ be a k-cycle on X. Then

    in Ak−1⁢((|α|∪|α′|)∩|D|).

  2. (b)

    Let D′ be a pseudo-divisor on X. Then

    in Ak−1⁢(|α|∩(|D|∪|D′|)).

  3. (c)

    Let f:Y→X be a proper morphism, β be a k-cycle on Y, and g:|β|∩f−1⁢(|D|)→f⁡(|β|)∩|D| be the restriction of f to |β|∩f−1⁢(|D|). Then

    in Ak−1⁢(f⁡(|β|)∩|D|).

  4. (d)

    Let f:Y→X be a flat morphism of relative dimension n and g:f−1⁢(|α|∩|D|)→|α|∩|D| be the restriction of f to f−1⁢(|α|∩|D|). Then

    in An+k−1⁢(f−1⁢(|α|∩|D|)).

  5. (e)

    If the line bundle 𝒪⁡(D) is trivial, then

    in Ak−1⁢(|α|∩|D|).

Proof.

Part (a) is immediate from the definition. Using part (a), then, we can assume by linearity that α=[V] for some k-dimensional subvariety V⊂X. Restricting to V, (b) is just the statement that taking the Weil class divisor is compatible with sums.

For part (c), we can likewise assume that β=[W] for some k-dimensional subvariety W⊂Y; then f∗⁢D⋅β is the restriction of the Cartier divisor f∗⁢D~ representing f∗⁢D to W, and so we can assume that Y=W. Similarly on the right-hand side D⋅f∗⁢β=D⋅deg⁡(f⁡(W)/W)⁢[f⁡(W)] and so concerns only the restriction of D to f⁡(W), and so we can assume that f⁡(W)=X. In this case g=f on the support of D and so the statement is

since D⋅[X]=[D] and f∗⁢D⋅[Y]=f∗⁢[D]. If f is a map of degree d and D=div⁡(r) for some function r on some open subset of f⁡(W), then from last time we know that locally

where N is the determinant map from functions on subsets of W to functions on their images, since N⁡(f∗⁢r)=d⁢r since f has degree d. But locally we can always assume that [D] is principal, and so f∗⁢f∗⁢[D]=d⁡[D] as desired.

For (d), we can again assume that α=[V]=[X], so the statement similarly becomes

By linearity, we can assume D=[W] for some subvariety W of X=V, at which point the statement is f∗⁢[W]=[f−1⁢(W)], which is true whenever f is flat.

Finally for (e) we can again assume α=[V]=[X], so that the statement is [D]=0 in An−1⁢X whenever 𝒪⁡(D) is trivial, where n is the dimension of V=X. Letting D~ be the Cartier divisor representing D, we know from section 1 that 𝒪⁡(D) is trivial precisely when D~ is linearly equivalent to the trivial Cartier divisor 0=({Uα,1}) for which every local function is a unit; and we know that the associated Weil divisor map Div⁡X→Zn−1⁢X descends to a map Pic⁡X→An−1⁢X, i.e. [D]=[D~]=0 whenever 𝒪⁡(D) is trivial. □

2.4 Commutativity

Suppose that we have two Cartier divisors D,D′ on an n-dimensional variety X. Then they both determine associated Weil divisors [D],[D′]∈Zn−1⁢X (and thus in An−1⁢X), and so it is natural to consider the intersections

Theorem 2.4.1.

In An−2⁢(|D|∩|D′|), we have

Corollary 2.4.2.

Let D be a pseudo-divisor on a scheme X, and α be a k-cycle on X rationally equivalent to 0. Then

in Ak−1⁢(|D|).

Proof.

We can assume without loss of generality that α=[div⁡(f)] for some rational function f on a subvariety V of X. Then letting D~ be the Cartier divisor representing D we can replace D with D~ and X with V without changing the result; then we can apply Theorem 2.4.1 to get

But by part (e) of Theorem 2.3.1, we have div⁡(f)⋅[D~]=0. □

Given a closed subscheme Y⊂X and a k-cycle α on Y, we can construct its intersection D⋅α∈Ak−1⁢(Y∩|D|) for any pseudo-divisor D on X. This gives a map

The above corollary shows that in fact this map descends to a map

this is called intersecting with D.

Corollary 2.4.3.

For two pseudo-divisors D,D′ on a scheme X and a k-cycle α on X, we have

in Ak−2⁢(|α|∩|D|∩|D′|).

Proof.

We can assume without loss of generality that α=[V] for some subvariety V⊆X of dimension k. Then we can restrict D and D′ to V, so that D′⋅[V]=[id∗⁢D′]=[D′] and similarly D⋅[V]=[D]; and then applying Theorem 2.4.1 immediately gives the result. □

For pseudo-divisors D1,…,Dn on X and a k-cycle α on X, we can then define inductively

in Ak−n⁢(|α|∩(|D1|∪⋯∪|Dn|)). Theorem 2.4.1 implies that the order of the Di is unimportant, and parts (a) and (b) of Theorem 2.3.1 implies that the action is linear in each Di and in α. More generally if p⁡(t1,…,tn) is a homogeneous polynomial of degree d and Z is a closed subscheme of X containing |α|∩(|D1|∪⋯∪|Dn|), then we can define p⁡(D1,…,Dn)⋅α in Ak−d⁢(Z).

Definition 2.4.4.

We say that an algebraic variety Y is complete if for any variety Z the projection Y×Z→Y is a closed map.

For example, any projective variety is complete.

If n=k and Y=|α|∩(|D1|∪⋯∪|Dk|) is complete, then we can define the intersection number

Similarly if p is a homogeneous polynomial of degree k in k variables then we can define

For a subvariety V purely of dimension k, we will sometimes write simply V instead of [V]; similarly we will sometimes write D instead of [D].

Example 2.4.5.

Let X be the projective completion of the affine surface X′⊂𝔸3 defined by z2=x⁢y. Consider the Cartier divisor D on X defined everywhere by the equation x, corresponding to the subvariety cut out by x=0. Define the lines ℓ,ℓ′ by x=z=0 and y=z=0 respectively, and let P be the origin (0,0,0). Along the subvariety x=0, from the defining equation we also have z=0 (in affine space), and so [D]=ordℓ⁡D⋅[ℓ]; we have

where (in the affine variety) A=𝒪X,ℓ=K⁡[x,y,z]/(z2−x⁢y). Thus A/(x)=K⁡[x,y,z]/(z2−x⁢y,x)=K⁡[y,z]/(z2) which has length 2, with maximal proper subsequence of modules given by 0⊂K⁡[y]=K⁡[y,z]/(z)⊂K⁡[y,z]/(z2). Therefore [D]=2⁢[ℓ]. We can compute

where j is the inclusion of ℓ′ into X, since restricted to the line y=z=0 the equation x=0 specifies only the point P with multiplicity 1. Therefore there cannot exist any Cartier divisor D′ with [D′]=[ℓ′], since if there were we would have

in either Z1⁢X or A1⁢X, by Theorem 2.4.1 and the above calculation. This proves our above claim that the maps Div⁡X→ZdimX−1 and Pic⁡X→AdimX−1⁢X are not in general surjective.

2.5 The first Chern class

Let X be a scheme, V⊆X a subvariety of dimension k, and L a line bundle on X. The restriction of L to V is a line bundle on V and so is isomorphic to 𝒪⁡(C) for some Cartier divisor C on V, determined up to linear equivalence. This in turn defines a well-defined element [C] of Ak−1⁢X; we write c1⁢(L)∩[V]≔[C]. More generally, if α=∑VnV⋅[V] is a k-cycle on X then define CV for each V as above, and write

If L=𝒪⁡(D) for some pseudo-divisor D, then if j:V↪X is the inclusion then the Cartier divisor D~ on V representing j∗⁢D satisfies 𝒪⁡(D~)≃𝒪⁡(D) by construction; by definition, this means that [CV]=[j∗⁢D]=D⋅[V] and so

in Ak−1⁢X.

Theorem 2.5.1.

Let X be a scheme, L be a line bundle on X, and α be a k-cycle on X.

  1. (a)

    If α is rationally equivalent to 0, then c1⁢(L)∩α=0. Therefore there is an induced homomorphism c1⁢(L)∩−:Ak⁢X→Ak−1⁢X.

  2. (b)

    If L′ is a second line bundle on X, then

    in Ak−2⁢X.

  3. (c)

    If f:Y→X is a proper morphism and β is a k-cycle on Y, then

    in Ak−1⁢X.

  4. (d)

    If f:Y→X is a flat morphism of relative dimension n, then

    in An+k−1⁢Y.

  5. (e)

    If L′ is a second line bundle on X, then

    and

    in Ak−1⁢X.

Proof.

A line bundle on X defines a pseudo-divisor with support X, and so the analogous properties from Theorem 2.3.1 and its corollaries immediately imply these. □

2.6 The Gysin map

Fix an effective Cartier divisor D on a scheme X, with the inclusion given by i:|D|↪X. Then we define the “Gysin homomorphism”

for k-cycles α on X.

Proposition 2.6.1.

With notation as above:

  1. (a)

    If α is rationally equivalent to 0, then i∗⁢α=0, and so there is an induced homomorphism i∗:Ak⁢X→Ak−1⁢(|D|).

  2. (b)

    We have

  3. (c)

    If β is a k-cycle on |D|, then

  4. (d)

    If X is purely n-dimensional, then

    in An−1⁢(|D|).

  5. (e)

    If L is a line bundle on X, then

    in Ak−2⁢(|D|).

All of these follow immediately from the definitions and the results above.

3 Alex (Feb 05): Chern classes and Segre classes of vector bundles

3.1 Chern and Segre Classes

Let L be a line bundle over a scheme X. If Vk⊆X is a subvariety, then there exists a Cartier divisor such that L|V=𝒪V⁢(C), which gives us a cycle [C]∈Ak−1.

Definition 3.1.1.

We will define the first Chern class of L by c1⁢(L)∩[V]=[C]. Therefore we obtain a map

Proposition 3.1.2.

Here are some properties of Chern classes:

  1. 1.

    The Chern class gives a well-defined map Ak→Ak−1.

  2. 2.

    We have c1⁢(L)∩c1⁢(L′)∩−=c1⁢(L′)∩c1⁢(L)∩−.

  3. 3.

    Let f:X→Y be proper and α∈Ak⁢X. For a line bundle L over Y, we have

  4. 4.

    If f:X→Y is flat of relative dimension n, L/Y, and α∈Ak⁢Y, we have

  5. 5.

    Let L,L′ be line bundles. Then c1⁢(L⊗L′)∩−=c1⁢(L)∩−+c1⁢(L′)∩−

In the smooth case, Segre classes agree with Chern classes, but they have a nice generalization to the singular case. Let E be a rank e+1 vector bundle over a scheme X and let P=ℙ⁡(E) be the associated projective bundle. Then we have a line bundle 𝒪E⁢(1), and so we define

If E is a line bundle, we see that s1⁢(E)∩α=−c1⁢(E)∩α.

Proposition 3.1.3.

Let L be a line bundle. Then

Now we will define Chern classes for an arbitrary vector bundle E of rank r=e+1. Set

and set

This tells us that cn⁢(E)=−s1⁢cn−1−s2⁢cn−2−⋯−sn. This gives us a map ci⁢(E):Ak→Ak−i. Here are some new properties of Chern classes:

  • •

    For all i>r, we have ci⁢(E)=0.

  • •

    If 0→E′→E→E′′→0 is an exact sequence of vector bundles, then ct⁢(E)=ct⁢(E′)⋅ct⁢(E′′).

Remark 3.1.4.

The projection formula, flat pullback, and c1 give us uniqueness for all of the Chern classes.

3.2 Splitting Principle

Let E be a vector bundle over X. Then there exists a flat f:X′→X such that

  1. 1.

    The pushforward f∗⁢Ak⁢X′→Ak⁢X is injective.

  2. 2.

    There exists a filtration E=Er⊇Er−1⊇⋯⊇E0=0 such that Ei/Ei−1=Li for some line bundle Li. In addition, we have

This gives us a splitting principle:

To prove a universal formula for Chern clases of some vector bundles with certain relations, it suffices to show that the formula is true for filtrations with line bundle quotients and the relations are preserved under flat pullback.

Here are some properties of Chern classes:

  1. 1.

    If E∨ is the dual of E, we have ci⁢(E∨)=(−1)i⁢ci⁢(E) and ct⁢(E∨)=∏(1−αi⁢t), where αi are the Chern roots of E.

  2. 2.

    Let E,F be vector bundles with Chern roots αi,βj. Then {αi+βj}i,j are Chern roots for E⊗F.

  3. 3.

    Let E be a rank r vector bundle. Then ⋀pE has Chern roots {αi1+⋯+αip}.

Example 3.2.1.

Recall the exact sequence

From the fact that ct⁢(𝒪ℙn)=1 and ct⁢(𝒪ℙn⁢(1))=(1+H⁢t), we can compute the Chern character of Tℙn.

Let E be a vector bundle on a nonsingular variety X and suppose p:ℙ⁡(E)→X is the projection. Then we have an exact sequence

This allows us to calculate

3.3 Rational equivalence of bundles

This generalizes the Gysin map from last week. Let E be a vector bundle of rank r=e+1 and let π:E→X be the projection. Then we obtain a commutative diagram

Theorem 3.3.1.
  1. 1.

    The map π∗:Ak−r→Ak⁢E is an isomorphism for all k.

  2. 2.

    A cycle β∈Ak⁢ℙ⁡(E) is uniquely expressible in the form

    where αi∈Ak+1−e⁢X.

Definition 3.3.2.

Let s∗:Ak⁢E→Ak−r⁢X be (π∗)−1. This is the Gysin map.

Proposition 3.3.3.

Let β∈Ak⁢E and β¯∈Ak⁢ℙ⁡(E⊕𝒪) that restricts to β. Then s∗⁢(β)=q∗⁢(cr⁢(ξ)∩β¯) where ξ is the universal rank r quotient bundle of q∗⁢(E⊕𝒪). Here, we have an exact sequence

4 Patrick (Feb 12): Cones: because not every coherent sheaf is locally free

Note: These are the speaker’s notes.

4.1 Cones and Segre Classes

Our goal is to define a Segre class s⁡(X,Y) of a subvariety X⊊Y and study its properties.

4.1.1 Cones

Definition 4.1.1.

Let S∙ be a sheaf of graded 𝒪X-algebras such that 𝒪X→S0 is surjective, S1 is coherent, and S∙ is generated by S1. Then any scheme of the form C=Spec𝒪X⁡(S∙) is called a cone.

If C is a cone, then ℙ⁡(C⊕1)=Proj⁡(S∙⁢[z]) is the projective completion with projection q:ℙ⁡(C⊕1→X). Let 𝒪⁡(1) be the canonical line bundle on ℙ⁡(C⊕1).

Definition 4.1.2.

The Segre class s⁡(C)∈A∗⁢X of C is defined as

Proposition 4.1.3.
  1. 1.

    If E is a vector bundle on X, then s⁡(E)=c⁢(E)−1∩[X], where c=1+c1+⋯ is the total Chern class.

  2. 2.

    Let ,1…,ct by the irreducible components of C with geometric multiplicity mi. Then

Example 4.1.4.

Let ℱ,ℱ′ be coherent sheaves and let ℰ be locally free. Then we may define C⁡(ℱ)=Spec⁡(Sym⁡ℱ). We may define s⁡(ℱ)=s⁡(C⁡(ℱ)). Now if

is exact, then s⁡(ℱ′)=c⁡(E)∩s⁡(ℱ).

4.1.2 Segre Class of a Subvariety

Let X be a closed subscheme of Y defined by the ideal sheaf ℐ and let

be the normal cone. Note that if X is regularly embedded in Y, then CX⁢Y is a vector bundle.

Definition 4.1.5.

The Segre class of X in Y is defined by

Lemma 4.1.6.

Let Y be a scheme of pure dimension m and let Y1,…,Yr be the irreducible components of Y with multiplicity mi. If X is a closed subscheme of Y and Xi=X∩Yi, then

Proposition 4.1.7.

Let f:Y′→Y be a morphism of pure-dimensional schemes, X⊆Y a closed subscheme, and g:X′=f−1⁢(X)→X be the induced morphism.

  1. 1.

    If f is proper, Y is irreducible, and f maps each irreducible component of Y′ onto Y, then

  2. 2.

    If f is flat, then g∗⁢(s⁡(X,Y))=s⁡(X′,Y′).

Remark 4.1.8.

If f is birational, then f∗⁢(s⁡(X′,Y′))=s⁡(X,Y). This says that Segre classes are unchanged by pushforward along birational modifications.

Corollary 4.1.9.

Let Y be a variety and X⊆Y be a proper closed subsecheme. Then let Y~=BlX⁡Y and X~=ℙ⁡(C) be the exceptional divisor with projection η:X~→X. Then

Example 4.1.10.

Let A,B,D be effective Cartier divisors on a surface Y. Then let A′=A+D,B′=B+D, and let X=A′∩B′. Suppose that A,B meet transversally at a single smooth point P∈Y. Then if Y~=BlP⁡Y and f:Y~→Y is the blowup with exceptional divisor E, we see that X~=f−1⁢(X)=f∗⁢D+E, so we have

If A,B both have multiplicity m at P and no common tangents at P, then

In general, the answer is more complicated.

4.1.3 Multiplicity

Let X⊆Y be an (irreducible) subvariety. Then the coefficient of [X] in the class s⁡(X,Y) is called the algebraic multiplicity of X on Y and is denoted eX⁢Y.

Suppose X has positive codimension n, p:ℙ⁡(CX⁢Y)→X and q:ℙ⁡(CX⁢Y⊕1)→X are the projections to X, and Y~=BlX⁡Y with exceptional divisor X~=ℙ⁡(C). Then we have

For example, if X is a point, then we have

Example 4.1.11.

Let C be a smooth curve of genus g and C(d) be the d-th symmetric power of C. Then let P0∈C, J=J⁡(C) be the Jacobian, and ud:C(d)→J be given by D↦D−d⁢P0. We know that the fibers of ud are the linear systems |D|≅ℙr; if d>2⁢g−2, then ud:C(d)→J is a projective bundle; and if 1≤d≤g, then μd is birational onto its image Wd. Now if deg⁡D=d and dim|D|=r, we have

where K=c1⁢(K|D|). When d is large, this follows from the second bullet, but if d is small, then we may embed

and then consider the normal bundle to this embedding restricted to |D|. Combined with Proposition 4.1.7, this gives us the Riemann-Kempf formula, which says that the multiplicity of Wd at ud⁢(D) is given by eμd⁢(D)⁢Wd=(g−d+rr).

Remark 4.1.12.

The previous example can be generalized to the Fano varieties of lines on a cubic threefold X. In particular if F is the Fano variety of lines on X, then there is a morphism of degree 6 from F×F to the theta divisor, and we can calculate (following Clemens-Griffiths) that

and then the theta divisor has a singular point of multiplicity 3.

4.1.4 Linear Systems

Let L be a line bundle on a variety X (of dimension n) and let V⊆|L| be a partial linear system of dimension r+1. Then let B be the base locus of V. Then if X~=BlB⁡X, we obtain a morphism f:X~→ℙr resolving the rational map X⇢ℙr. By definition, we have f∗⁢𝒪⁢(1)=π∗⁢(L)⊗𝒪⁡(−E). Define degf⁡X~ to be the degree of f∗⁢[X~]∈An⁢ℙr.

Proposition 4.1.13.

We have the identity

Example 4.1.14.

Let B⊂ℙn be the rational normal curve. Then let V⊂|𝒪⁡(2)| be the linear system of quadrics containing B. If P~n=BlB⁡ℙn, we see that

If n=4, then f⁡(ℙ~4)=Gr⁡(2,4)⊂ℙ5.

4.2 Deformation to the Normal Cone

Let X⊆Y be a closed subscheme and C=CX⁢Y be the normal cone. We will construct a scheme M=MX⁢Y and a closed embedding X×ℙ1⊆M such that

comutes and such that

  1. 1.

    Away from ∞, we have q−1⁢(𝔸1)=Y×𝔸1 and the embedding is the trivial embedding X×𝔸1⊆Y×𝔸1.

  2. 2.

    Over ∞, M∞=ℙ⁡(C⊕1)+Y~ is a sum of two Cartier divisors, where Y~=BlX⁡Y. The embedding of X is given by X↪C↪ℙ⁡(C⊕1). We also have ℙ⁡(C⊕1)∩Y~=ℙ⁡(C), which is embedded as the hyperplane at ∞ in ℙ⁡(C⊕1) and as the exceptional divisor in Y~.

We will now construct this deformation. Let M=BlX×∞⁡Y×ℙ1. Clearly we have CX×∞⁢Y×ℙ1=C⊕1. But now we can embed X×ℙ1⊆M. The first property is obvious by the blowup construction, so now we need to show the second property.

We may assume Y=Spec⁡A is affine and X is defined by the ideal I. Identify ℙ1∖0=𝔸1=Spec⁡k⁡[t]. Then if we write Sn=I,Tn, then we see that BlX×0⁡Y×𝔸1=Proj⁡S∙. But now this is covered by affines

Now for a∈I, we see that ℙ⁡(C⊕1)⊆Spec⁡S(a)∙ is defined by the equation a/1, while Y~ is defined by T/a, and now we see that

as desired.

Now this allows us to define a specialization morphism

Proposition 4.2.1.

Specialization preserves rational equivalence. Therefore we have a specialization morphism

Remark 4.2.2.

Supposing that X,Y are smooth, then the embedding of X⊂ℙ⁡(N⊕1) is nicer than X⊂Y in several ways:

  1. 1.

    There is a retraction ℙ⁡(N⊕1)→X.

  2. 2.

    There is a vector bundle ξ on ℙ⁡(N⊕1) or rank codimY⁡X and a section s∈Γ⁡(ξ) such that V⁡(s)=X. Therefore X is represented by the top Chern class of ξ.

5 Caleb (Feb 19): Chern classes and intersection products

5.1 Chern Classes

We will recall the notion of Chern classes in topology. Recall that characteristic classes are natural transformations from ℂ−𝖵𝖾𝖼𝗍n⁢(−)→H∗⁢(−,ℤ). Because complex vector bundles of rank n are reprsented by Gr⁡(n,∞), the Yoneda lemma tells us that characteristic classes are given by H∗⁢(Gr⁡(n,∞),ℤ). Now Chern classes satisfy the following properties:

  1. 1.

    c⁡(ℂ)=1, where ℂ denotes the trivial bundle.

  2. 2.

    If E has rank n, then c⁡(E) is nonzero only in degrees 0,…,2⁢n.

  3. 3.

    c⁡(E⊕E′)=c⁡(E)⁢c⁡(E′).

  4. 4.

    c⁡(𝒪ℂ⁢ℙ∞⁢(−1)) generates H2⁢(ℂ⁢ℙ∞).

Remark 5.1.1.

Chern classes can be computed using the exponential sequence

where

under the isomorphism H1⁢(𝒪X×)≅Pic⁡X.

We now return to algebraic geometry. Let L be a line bundle and σ be a rational section. Then c1⁢(L)=[div⁡σ]∈An−1⁢(X).

Example 5.1.2.

If X is a smooth variety over ℂ, then we can write ωX=⋀nΩX, and KX=c1⁢(ωX). For example, if X=ℙn, then we can choose Θ=dx1∧⋯∧dxn, and KX=(n+1)⁢H.

Here are some properties of Chern classes of vector bundles. Let E be a vector bundle on a smooth quasiprojective variety X. Then we define

  1. 1.

    If L is a line bundle, then c⁡(L)=1+c1⁢(L).

  2. 2.

    If τ0,…,τn−i are global sections with degeneracy locus having codimension i, then ci⁢(E)=[D]∈An−i⁢(X)

  3. 3.

    If 0→E→F→G→0 is an exact sequence, then c⁡(F)=c⁡(E)⁢c⁡(G).

  4. 4.

    If φ:Y→X is a morphism, then φ∗⁢c⁢(E)=c⁡(φ∗⁢E).

This axiomatic construction of Chern classes was done first by Grothendieck in 1958.

Now we discuss degeneracy loci. We know that c1⁢(L)=0 if and only if L≃𝒪X, whic happens if and only if L has a nonvanishing section. Now if E has rank n and we have general global sections τ0,…,τn−i, the degeneracy locus is where

Now recall that we defined the map c1⁢(L)∩−:Ak→Ak−1 and the map si⁢(E)∩−:Ak→Ak−i. Finally, we defined c⁡(E)=s⁢(E)−1. We will now return to splitting. If E→X is a vector bundle of rank r, then there exists f:Y→X such that f∗ is surjective and f∗ is injective on A∗ and f∗⁢E has a filtration by line bundles. The proof is by induction and is thus omitted. This allows us to write

where αi=c1⁢(Li) are the Chern roots.

Example 5.1.3.

If E has roots αi, then Sk⁢E has roots {αi1+⋯+αik}i1≤i2≤⋯≤ik.

Example 5.1.4.

We will now compute the 27 lines on a cubic surface. Note that there is a 20-dimensional locus of cubic hypersurfaces in ℙ3. On the other hand, there is a 4-dimensional space of cubic forms on a line L. Therefore, L lies on a cubic surface f if and only if f|L=0.

As L varies, the cubic forms give a vector bundle S3⁢E∨. Now each f gives us a section, so the locus where L is contained in f is precisely the zero-locus of f. Now we have a vector bundle S3⁢E∨ on the Grassmannian G⁡(2,4) of lines in ℙ3. Now we may assume that

(the number of lines on the cubic surface) is finite, so we want to calculate it. We know E has roots α1,α2. Then c1⁢(E)=α1+α2,c2⁢(E)=α1⁢α2, so

Therefore we have c4=9⁢c22+18⁢c12⁢c2. But then ∫c2⁢(E∨)2=1 is the number of lines contained in two planes. On the other hand, ∫{c1⁢(E∨)}2⁢c2⁢(E∨)=1 is the number of lines intersecting two lines in a given plane, so we see that c4=27, as desired.

5.2 Intersection Products

Let i:X→Y be a regular closed embedding of codimension d with normal bundle NX⁢Y. Let V be purely k-dimensional with a map f:C→Y. Our goal is to construct X⋅V∈Ak−d⁢(X). If we want this to look like cohomology, use formal Poincare duality. We will use deformation to the normal cone for our construction. Consider the pullback

Now define N=g∗⁢NX⁢Y be a rank d vector bundle on W and π:N→W be the projection. Then if I is the ideal sheaf of i:X→Y, this generates the ideal sheaf J of p:W→V. This means that

is surjective. This gives a closed embedding of the normal cone C=CW⁢V↪N. By a result in the appendix, C is purely k-dimensional. Now consider te cycle [C] on N and let s be the zero-section of N. Now we will define

where s∗:Ak⁢(N)→An−d⁢(W) is the Gysin map.

Here are some properties of the intersection product:

  1. 1.

    The intersection product is compactible with proper pushforwards.

  2. 2.

    Intersection products are compatible with flat pullbacks.

  3. 3.

    The intersection product is commutative.

  4. 4.

    Functoriality.

6 Nicolás (Mar 12): families of algebraic cycles

Note: these notes were adapted from the source for the beamer provided by the speaker. Very light edits were made to the source.

6.1 Introduction

“A typical problem in enumerative geometry is to find the number of geometric figures in a given family which satisfy certain conditions”. One of the classical examples is that given five points in general position in ℙ2, there exists a unique smooth conic passing through them.

Figure 6.1: Five points in general position determine a unique conic.

The idea is that “conics are parametrized by ℙ5, and passing through a point is a degree 1 equation in ℙ5”. But this might be dangerous:

  • •

    ℙ5 parametrizes conics, not smooth conics.

  • •

    We need transversal intersections.

For instance, the argument does not work for smooth conics tangent to five lines. Each tangency is a degree 2 equation on ℙ5, but they are not transversal (the conics of the form {L2=0} are “tangent” to all lines).

The correct number is 1, which may be seen by taking the dual conic (the set of tangent lines, as a subset of (ℙ2)∗).

6.1.1 Conservation of number

The classical principle is called conservation of number: if the problem has a finite numerical answer, this number is constant (or jumps to infinity).

Sadly, this does not work. Given four lines and a point in general position, there exists 14⋅2=2 smooth conics tangent to the lines and passing through the point (as one can show by taking the dual problem). But if the point lies in the diagonals of the quadrilateral given by the lines, then the number of smooth solutions decreases to 1 or 0.

Today we will discuss strong foundations for this principle, and some applications in enumerative geometry.

6.2 Families of cycle classes

During this section, T will denote an irreducible variety of dimension m>0. We take t∈T a regular closed point, and we denote

for the point and the inclusion.

We will use script letters (e.g. 𝒳,𝒴) for schemes over T, and the corresponding latin letters (e.g. Xt,Yt) for the corresponding fibers over t (as schemes over {t}). If f:𝒳→𝒴 is a morphism, we denote ft:Xt→Yt the map on the fibers.

6.2.1 Specialization

Let p:𝒴→T,α∈Ak+m⁢𝒴. We define αt∈Ak⁢Yt by

where t! is the refined Gysin homomorphism induced by

For instance, if α=[𝒱] and 𝒱⊆Yt, then [𝒱]t=0.

Here are some basic properties.

Proposition 6.2.1.
  1. 1.

    If f:𝒳→𝒴 is proper, α∈Ak+m⁢𝒳, then

  2. 2.

    If f:𝒳→𝒴 is flat of relative dimension n, α∈Ak+m⁢𝒴

  3. 3.

    If i:𝒳→𝒴 is a regular embedding of codimension d, such that it:Xt→Yt is also a regular embedding of codimension d, f:𝒱→𝒴 a morphism, α∈Ak+m⁢𝒱, then

  1. 4.

    If E is a vector bundle over 𝒴, α∈Ak+m⁢𝒴, then

The proof follows directly from similar statements for the refined Gysin homomorphism (see § 6.2–6.4).

We would now like to relate different fibers. Given a family 𝒳→T and α∈Ak+m⁢𝒳, it is natural to compare αt∈Ak⁢(Xt) for different values of t. It is not obvious that such relation exists, even if 𝒳=Y×T is the trivial family.

Example 6.2.2.

Let Y=T be a projective curve of genus g≥2, and Δ⊆Y×T the diagonal. If α=[Δ]∈A1⁢(Y×T), then αt=[t]∈A0⁢Y. But for t1≠t2, we have that αt1 and αt2 are not rationally equivalent.

This can be solved if we assume that 𝒳=Y×T, and if for every t1,t2∈T, they can be connected by a chain of rational curves in T (see Example 10.1.7).

Here is a useful corollary.

Corollary 6.2.3.

Assume T is non-singular, t∈T rational over the ground field, 𝒴 smooth over T with relative dimension n. If α∈Ak+m⁢(𝒴),β∈Al+m⁢(𝒴), then

This gives us a strategy to show that a⋅b=c in a non-singular variety Y. We construct a family 𝒴→T with Yt=Y for some t, and such that a,b,c can be lifted to α,β,γ. Then, it suffices to show that α⋅β=γ, which we can try to prove generically.

6.2.2 Sample Application

Let C be a non-singular curve, C(n) its nth symmetric product (which points are effective divisors of degree n over C). If A is an effective divisor on C of degree <n, define

One can show that if A and B have disjoint support, then XA and XB intersect transversally, and so

This is true even if A and B intersect, by using Corollary 6.2.3 and by “moving” A.

6.3 Conservation of number

We have seen that for α∈Ak⁢𝒴, it is not clear that {αt}t∈T are related, even if 𝒴=Y×T is the trivial family. We have the following substitute.

Proposition 6.3.1 (Conservation of number).

Let p:𝒴→T be a proper morphism, dimT=m as before. Let α be an m-cycle on 𝒴. Then αt∈A0⁢(Yt) all have the same degree (which is obtained by pt⁣∗⁢(αt)=deg⁡αt⋅[{t}]).

The idea of the proof is write p∗⁢(α)=N⁡[T]∈Am⁢(T), for some N∈ℤ. Then, by Proposition 6.2.1 we get

This proposition can be improved to compute the degree of intersections with Chern classes or some divisors (see § 10.2 for precise statements). We will need the following result.

Corollary 6.3.2.

Let Y be a scheme, ℋi⊆Y×T effective Cartier divisors which are flat over T,i=1,…,d. Let a be a d-cycle on Y. Assume that

is proper over T. Then

is independent of t.

6.4 An enumerative problem

Our main application of these techniques will be to solve the following problem.

Given an r-dimensional family of plane curves, and r curves in general position in the plane, how many curves in the family are tangent to the r given curves?

The answer will require to compute the characteristics μk⁢νr−k of the family, which are the number of curves in the family passing through k general points and tangent to r−k general lines.

For instance, if we consider the family of smooth conics, then

  1. 1.

    We will study the incidence correspondence

    This can be seen as a ℙ1-bundle over ℙ2. In fact, if E is the kernel of

    then I=ℙ⁡(E).

    This allows us to compute A∙⁢(I) (see Example 8.3.4), with a basis

    where λ⁢ζ=λ2+ζ2,λ3=ζ3=0, and λ,ζ the pullbacks of c1⁢(𝒪ℙ2⁢(1)),c1⁢(𝒪ℙ2⁣∗⁢(1)).

    Now, if M is a line and Q a point, consider

    One can show that

  2. 2.

    Let D⊆ℙ2 be a curve without multiple components. Define D′⊆I as the closure of

    We claim that

    where n is the degree and m the class of D (the number of tangents from a general point to D). The idea is to compute

    which has generically #⁢D′∩M′′=n points.

    The equivalence [D′]=m⁡[M′]+n⁡[Q′] can be computed explicitely. Take P0 a general point, M a general line, and let Q1,…,Qm the intersections of M with the tangents from P0.

    P0Q1Q2
    Figure 6.2: Projection from P0 to the line M.

    The projection from P0 to M gives a family 𝒟→𝔸1 with 𝒟1=[D′],𝒟0=n⁡[M′]+∑[Qi′]. (There is a explicit computation in § 10.4.)

  3. 3.

    Let 𝒳⊆ℙ2×S be a flat family of plane curves, dimS=r, S non-singular. Assume Xs has no multiple compontents for general s, and let S0⊆S an open set with Xs reduced for s∈S.

    Let 𝒳⁡(r)⊆Ir×S0 given by (P1,L1),…,(Pr,Lr),s such that Pi is a simple point of Xs, and Li is tangent in Pi. Note that dim𝒳⁡(r)=2⁢r.

    Take D1,…,Dr⊆ℙ2 reduced curves, and consider

    We can move D1,…,Dr, so that the interseccion between 𝒳⁡(r) and D1′×⋯×Dr′ is transversal (by taking a general element in PGL⁡(2)r). This way, W has N (reduced) points.

    Now, compactify 𝒳¯⊆ℙ2×S0¯, and 𝒳⁡(r)¯⊆Ir×S0¯. If Z is a closed subsed of dimension less than 2⁢r, which contains all 𝒳⁡(r)¯−𝒳⁡(r), then the number N does not change after we remove Z.

  4. 4.

    We now degenerate each Di to a multiple line (as we did for D). This gives a diagram

    The space 𝒳⁡(r)¯ is complete, so 𝒲 is proper over 𝔸r. This way, we may take an open neighborhood T of (1,…,1) and (0,…,0), so that 𝒲 is proper over T and disjoint from Z.

    Now, Corollary 6.2.3 applies, and so

    where Di′,Ei′ are the fibers over 1 and 0.

    The right hand side is just

    where each curve Di has degree ni and class mi.

    The left hand side is the number of points N, provided that we take a convenient Z (which avoids technical difficulties such as bitangents).

6.4.1 A famous example

The most famous example is the Steiner’s conic problem, which tries to determine the number of conics tangent to five smooth conics in general position.

The natural family here is the family of smooth conics (as a subset of ℙ5), which has characteristics

(in characteristic zero!)

This way, the number of conics tangent to five non-singular curves of degree n in general position is

which for n=2 gives the famous number 3264.

7 Patrick (Mar 19): Doing Italian-style algebraic geometry rigorously

Note: these are the speaker’s notes.

7.1 Intersection multiplicities

Consider a Cartesian square

where i is a regular embedding of codimension d and V has pure dimension k. Let C=CW⁢V have components C1,…,Cr with multiplicity mi. Let Zi be the support of Ci. We call the Zi the distinguished varieties of the intersection.

Lemma 7.1.1.
  1. (a)

    Every irreducible component of W is distinguished.

  2. (b)

    For any distinguished variety Z, we have k−d≤dimZ≤k.

Definition 7.1.2.

An irreducible component Z of W=f−1⁢(X) is a proper component of intersection of V by X if dimZ=k−d. The intersection multiplicity of Z in X⋅V, denoted i⁡(Z,X⋅V;Y)=i⁡(Z,X⋅V)=i⁡(Z) is the coefficient of Z in the class X⋅V∈Ak−d⁢(W).

If NZ is the pullback of NX⁢Y to Z, then i⁡(Z,X⋅V;Y) is the coefficient of NZ in [C]. Now let A=𝒪Z,V and J⊂A be the ideal of W. Then A/J has finite length when Z is an irreducible component of W.

Proposition 7.1.3.

Assume Z is a proper component of W.

  1. (a)

    If ℓ⁡(A/J) is the length of A/J, then 1≤i⁡(Z,X⋅V;Y)≤ℓ⁡(A/J).

  2. (b)

    If J is generated by a regular sequence of length d, then i⁡(Z,X⋅V;Y)=ℓ⁡(A/J).

If A is Cohen-Macaulay, then local equations for X in Y give a regular sequence generating J and equality in (b) holds.

Now suppose Z is a proper component of the intersection of V by X on Y. Let A=𝒪V,Z and J be the ideal in A generated by the ideal (sheaf) of X in Y, and let 𝔪 be the maximal ideal of A.

Proposition 7.1.4.

Suppose V is a variety. Then i⁡(Z,X⋅V;Y)=1 if and only if A is regular and J=𝔪.

7.2 Nonsingular varieties

Let Y be a smooth variety of dimension n. Then Δ⊂Y×Y is regularly embedded with codimension n. The global intersection product is the map

where δ∗ is the Gysin homomorphism.

More generally, let X be a scheme and f:X→Y a morphism to a smooth variety. Then Γf is regularly embedded in Y, so we can define a cap product

If X is smooth, then we write f∗⁢y=f∗⁢y∩[X].

Remark 7.2.1.

We may also replace the Gysin homomorphisms Δ∗,Γf∗ with the refined Gysin homomorphisms Δ!,Γf!.

Definition 7.2.2.

Let f:X→Y be a morphism with Y smooth of dimension n. Let pX:X′→X,pY:Y′→Y be morphisms of schemes. Then form the square

Now define the refined intersection product by

for x∈Ak⁢(X′),Y∈Aℓ⁢(Y′). When X′=X,Y′=Y, this is the global product.

Proposition 7.2.3.

The refined products satisfy the following formal properties:

  1. (a)

    (Associativity) If X→𝑓Y→𝑔Z with Y,Z smooth, then

  2. (b)

    (Commutativity) If fi:X→Yi with Yi smooth, then

  3. (c)

    (Projection) Let X→𝑓Y→𝑔Z with Z smooth. Let f′:X′→Y′ be proper with pY⁢f′=f⁢pX and f′′=f′×ZidZ. Then

  4. (d)

    (Compatibility) Let f:X→Y with Y smooth and g:V′→Y′ be a regular embedding. Then

Corollary 7.2.4.

Let Y be smooth and j:V→Y be a regular embedding. If x is a cycle on Y, then x⋅[V]=j!(x)∈A∗(|x|∩V).

Corollary 7.2.5.

Let f:X→Y with X,Y smooth. Then x⋅fy=(x×y)⋅[Γf]∈A∗⁢(|x|∩f−1⁢(|y|)).

Corollary 7.2.6.

Let f:X→Y with Y smooth and x a cycle on X. Then x⋅f[Y]=x.

Definition 7.2.7.

Let f:X→Y be a morphism with X purely m-dimensional and Y a smooth n-dimensional variety Y. For any morphism g:Y′→Y, define a refined Gysin homomorphism

Proposition 7.2.8.
  1. (a)

    If f is flat, then f!(y)=f′∗(y), where f′:X×YY′→Y′ is the base change.

  2. (b)

    If f is a local complete intersection morphism, then f! agrees with the morphism constructed in Section 6.6 of Fulton.

Now let Y be a smooth variety of dimension n. Let V,W be closed subschemes of Y of pure dimension k,ℓ. Now a component Z⊆V∩W is a proper component if dimZ=k+ℓ−n. If Z is proper, then the coefficient of Z in V⋅W∈Ak+ℓ−n⁢(V∩W) is called the intersection multiplicity i⁡(Z,V⋅W;Y)=i⁡(Z,ΔY⋅(V×W);Y×Y). If every component of V∩W is proper, then the intersection class is

Proposition 7.2.9.

Assume Z is a proper component of V∩W. Then

  1. (a)

    1≤i⁡(Z,V⋅W;Y)≤𝒪V∩W,Z.

  2. (b)

    If the local ring 𝒪V∩W,Z is Cohen-Macaulay, then i⁡(Z,V⋅W;Y)=ℓ⁡(𝒪V∩W,Z).

  3. (c)

    If V,W are varieties, then i⁡(Z,V⋅W;Y)=1 if and only if the maximal ideal of 𝒪Y,Z is the sum of the prime ideals of V and W. In fact, 𝒪V,Z,𝒪W,Z are regular.

Now let Y be a smooth variety of dimension n. Set Ap⁢(Y)=An−p⁢(Y). Now this indexing, the intersection product is now Ap⁢(Y)⊗Aq⁢(Y)→Ap+q⁢(Y). In addition, if f:X→Y is a morphism, the cap product is now Ap⁢(Y)⊗Aq⁢(X)→∩Aq−p⁢(X). If X is also smooth, the pullback now reads f∗:Ap⁢(Y)→Ap⁢(X).

Proposition 7.2.10.
  1. (a)

    Suppose Y is a smooth variety. Then the intersection product makes A∗⁢(Y) into a commutative graded ring with unit A0∋1=[Y]∈An. Then Y↦A∗⁢(Y) is a contravariant functor from smooth varieties to rings.

  2. (b)

    If f:X→Y is a morphism from a scheme X to a smooth variety Y, then A∗⁢X is a A∗⁢Y-module with action

  3. (c)

    If f:X→Y is a proper morphism of smooth varieties, then

    for all classes x on X and y on Y.

7.3 Bézout’s Theorem

We will now use this theory to discuss something classical. It is easy to see that Ak⁢(ℙn)≅ℤ and is generated by [Lk] for a linear subspace Lk⊂ℙn. If α is a k-cycle on ℙn, we define the degree deg⁡(α) to be the integer satisfying α=deg⁡(α)⋅[Lk]. Equivalently, we may define

Theorem 7.3.1 (Bézout).

Let αi∈Adi⁢(ℙn) for i=1,…,r. If d1+⋯+dr≤n, then

Proof.

We have an isomorphism A∗⁢(ℙn)=ℤ⁡[h]/(hn+1), where h=[Ln−1]. Thus [Ln−k]=hk, and the desired result follows. □

Now if subschemes V1,…,Vr⊆ℙn representing α1,…,αr meet properly, then

where Zj are the components of ⋂Vi. Then Bézout’s theorem gives us the identity

If H1,…,Hn are hypersurfaces intersecting properly (so the intersection is a finite number of points), then consider the local ring 𝒪⋂Hi,P. Then complete intersections are Cohen-Macaulay, so

Now dimk𝒪⋂Hi,P=deg⁡P⁢ℓ⁡(𝒪⋂Hi,P), and thus we obtain

This recovers the very classical Bézout’s theorem.

Example 7.3.2.

Let s be the hyperplane class on ℙn and t be the hyperplane class on ℙm. Then

  1. 1.

    A∗⁢(ℙn×ℙm)=ℤ⁡[s,t]/(sn+1,tm+1).

  2. 2.

    If H1,…,Hn+m are hypersurfaces in ℙn×ℙm with bidegree (ai,bi), then

  3. 3.

    If Δ is the diagonal in ℙn×ℙn, then [Δ]=∑i=0nsi⁢tn−i∈An⁢(ℙn×ℙn). This formula follows from intersecting Δ with [L1×L2], where L1,L2 are linear subspaces of complementary dimension.

  4. 4.

    Let s:ℙn×ℙm→ℙn⁢m+n+m be the Segre embedding. It h is the hyperplane class on ℙn⁢m+n+m, then s∗⁢u=s+t. Also, the degree of s⁡(ℙn×ℙm) is (n+mn).

  5. 5.

    If vm:ℙn→ℙ(n+mn)−1 is the Veronese embedding and s,h are the hyperplane classes on the source and target, then vm∗⁢u=m⋅s. If V is a k-dimensional subvarity of ℙn of degree d, then deg⁡(vm⁢(V))=d⋅mk.

8 Morena (Mar 26): Grothendieck-Riemann-Roch

We will begin by reviewing some notions from algebraic topology.

  1. 1.

    Let E be a vector bundle over X with Chern roots α1,…,αr. Then we may define the Chern character

    where sn is the n-th Newton polynomial. Recall that if

    is exact, then ch⁡(E)=ch⁡(E′)+ch⁡(E′′) and ch⁡(E⊗F)=ch⁡(E)⁢ch⁡(F).

  2. 2.

    We may define the Todd class

    Similarly to Chern classes, if

    is exact, then td⁡(E)=td⁡(E′)⁢td⁡(E′′).

  3. 3.

    Now define the ring

    This is well-behaved under pullback, but pushforward is not well-defined.

In order to fix the problems with pushforwards of vector bundles, we will attempt to define the Grothendieck group K0⁢(X) of coherent sheaves. Unfortunately, pullbacks of coherent sheaves are only right exacts. Also, pushforwards do exist for proper morphisms, but this is only left exact. Therefore we need to consider the higher derived functors, and now we define

This is a finite sum because we assume all of our schemes are Noetherian.

For a smooth variety, there is always a map K0⁢(X)→K0⁢(X). In fact, there is an inverse, because we can resolve any coherent sheaf ℱ by vector bundles. In this case, we will write K0⁢(X)=K0⁢(X)=K⁡(X). If ℱ is resolved by

we write [ℱ]=∑i=0n[ℒi] and ch⁡(ℱ)=∑i=0n(−1)i⁢ch⁡(ℒi).

8.1 Statement of the theorem

Recall the Chern character ch:K⁡(X)→A⁡(X)⊗ℚ. In topology, we need the Todd class to make the diagram

commute. Now we are able to state the Grothendieck-Riemann-Roch theorem:

Theorem 8.1.1 (Grothendieck-Riemann-Roch).

Let f:X→Y be proper where X,Y are smooth quasiprojective varieties over k. Then the diagram

commutes. In other words, f∗⁢(ch⁡(−)⋅td⁡(TX))=ch⁡(f∗⁢(−))⋅td⁡(TY).

First, we will see why this implies the classical (Hirzebruch)-Riemann-Roch theorem. First, let X be a curve, Y=pt, and ℒ be a line bundle. Then the right hand side is

On the other side, we have

This recovers the classical Riemann-Roch theorem.

Now let X be a smooth surface. Then our formula becomes

This recovers the Hirzebruch-Riemann-Roch theorem.

8.2 Proof for ℙm→pt

First, we have the following theorem:

Theorem 8.2.1.

K⁡(ℙm) is generated by [𝒪ℙm⁢(n)] for 0≤n≤m.

To prove this, we need to find a resolution of a coherent sheaf ℱ by line bundles, so we have

Because 𝒪⁡(1) is ample, ℱ⊗𝒪⁡(N) is generated by global sections, so we have the exact sequence

and we can tensor this by 𝒪⁡(−N). Now if we set V=⨁m+1𝒪⁡(−1), we can take the Koszul complex

Now we have the exact sequence

In particular, this means that we can generate K⁡(ℙm) by {𝒪,𝒪⁡(1),…,𝒪⁡(m)}.

Now it remains to prove Grothendieck-Riemann-Roch just for 𝒪⁡(n),0≤n≤m. Here, because Y is a point, we have ch⁡(f∗⁢(−))⁢td⁡(T∗)=χ⁡(X,−). In our case, we know that for 0<i<m, Hi⁢(ℙm,𝒪⁡(n))=0 and Hm⁢(ℙm,𝒪⁡(n))=H0⁢(ℙn,𝒪⁡(−n−m−1))∨=0, so

by basic combinatorics.1

On the other hand, we want to compute f∗⁢(ch⁡(𝒪⁡(n))⁢td⁡(Tℙm)). Recall the Euler exact sequence

Therefore we have td⁡(Tℙm)=td⁡(𝒪ℙn)m+1. Also, ch⁡(𝒪⁡(n))=ch⁡(𝒪⁡(1))n=ex⁢n. Looking for the xm-coefficient in ex⁢n⁢xm+1(1−e−x)m+1, we can multiply by xm+1 and compute the residue to obtain (m+nn).

Remark 8.2.2.

Grothendieck wrote the following about this result:

Um dieser Aussage über f:X→Y einen approximativen Sunn zu geben, musste ich nahezu zwei Stunden lang die Geduld der Zuhörer missbrauchen. Schwartz auf weiss (in Springer’s Lecture Notes) nimmt’s wohl an die 400, 500 Seiten. Ein packendes Beispiel dafür, wie unser Wissens - und Entdeckungsdrang sich immer mehr in einem lebensentrücken logischen Delirium auslabt, während das Leben selbst auf tausendfache Art zum Teufel geht – und mit endgültiger Vernichtung bedroht ist. Höchste Zeit, unsern Kurs zu ändern!

The English translation of this,2 keeping sentence structure intact as much as possible, is

To give this statement about f:X→Y an approximate meaning I’d have to abuse the patience of the listeners for nearly two hours. Black on white (in Springer’s Lecture Notes) it should take about 400, 500 pages. A gripping example of how our thirst for knowledge and discovery indulges more and more in a life-divorced logical delirium, while life itself is going to hell in a thousand different ways – and is threatened with absolute extinction. High time to change our course!

9 Patrick (Apr 02): Fine moduli memes for 1-categorical teens

Note: these are the speaker’s notes. Throughout this lecture, we work over ℂ.

9.1 The moduli space

The space M¯0,n parameterizes curves that look like this:

Figure 9.1: A stable curve

More precisely, these are reduced connected curves that are a tree of ℙ1s such that each ℙ1 has at least three marked points or nodes. In addition, at most two components meet at each node and we want H1⁢(C,𝒪C)=0. If all of these conditions are satisfied, we call our curve stable. More precisely, we want to represent the functor

Theorem 9.1.1 (Knudsen).

There exists a smooth complete variety M¯0,n and universal curve U0,n→M¯0,n with universal sections s1,…,sn that is a fine moduli space for this functor. M¯0,n also contains the space M0,n=(ℙ1∖{0,1,∞})n−3∖Δ as a dense open subset.

In fact, Knudsen also shows that U0,n=M¯0,n+1 and U0,n+1 is a blowup of M¯0,n+1×M¯0,nM¯0,n+1 along some subscheme of the diagonal. In order to prove this, Knudsen introduces two operations that we can perform, called contraction and stabilization. Contraction happens when we delete a marked point and stabilization happens when we add a marked point.

Theorem 9.1.2 (Knudsen).

Contraction and stabilization are functorial! Moreover, they commute with base change.

Here are some pictorial depictions of our operations:

45123delete 51234
Figure 9.2: Contraction
1234add 5 at node12345
Figure 9.3: Stabilization (1)
45123add 5 at 41234
Figure 9.4: Stabilization (2)

9.2 Keel construction of M¯0,n

First, we will describe the boundary divisors of M¯0,n. Let T⊆{1,…,n}≕[n] satisfy |T|,|TC|≥2. Then define

It is easy to see that DT=D(TC). Knudsen proves that DT is a smooth divisor and that DT≅M¯|T|+1×M¯|TC|+1.

Now consider the map π:M¯0,n+1→M¯0,n coming from the identification M¯0,n+1=U0,n. Now Keel proves that we can factor π as

where π1,2,3,n+1:M¯0,n+1→M¯0,n forgets all sections besides 1,2,3,n+1. Next, Keel shows that π1 is a composition of blowups along smooth codimension 2 subvarieties using an inductive construction.

Set B1=M¯0,n×M¯0,4. Then the universal sections s1,…,sn:M¯0,n→M¯0,n+1 induce sections p∘s1,…,p∘sn. In fact, DT≅p∘si⁢(DT) and this is independent of i. We also have

Lemma 9.2.1 (Keel).

The divisors DT⊂M¯0,n+1 with T⊂[n] are the exceptional divisors of π1.

Now we set B2 to be the blowup of B1 at ⋃|TC|=2DT. Inductively, we set Bk+1 to be the blowup of Bk at ⋃|TC|=k+1DT. Now we summarize the main results as follows:

Theorem 9.2.2 (Keel).

The map π1 factors through Bk and M¯0,n+1=Bn−2.

9.3 Intersection theory of M¯0,n

There are several major results about the intersection theory of M¯0,n. In fact, once we state these results, we will only be seven pages through Keel’s paper, and the rest of the paper is dedicated to proving these results.

Theorem 9.3.1.

We have an isomorphism A∗⁢(M¯0,n)→H∗⁢(M¯0,n). In particular, M¯0,n has no odd homology and A∗⁢(M¯0,n+1) is a finitely generated free abelian group. In fact, if a scheme Y satisfies A∗⁢(Y)=H∗⁢(Y), then so does Y×M¯0,n.

Theorem 9.3.2.

For any scheme S, there is an isomorphism A∗⁢(M¯0,n×S)=A∗⁢(M¯0,n)⊗A∗⁢(S).

Theorem 9.3.3.

For all k, we have an isomorphism

which is induced by the maps

where g,j are as in the diagram

Theorem 9.3.4.

The Chow groups Ak⁢(M¯0,n) are free abelian and the ranks ak⁢(n)=rk⁡(Ak⁢(M¯0,n)) are given by the recursive formula

In particular, we have the Picard rank a1⁢(n)=2n−1−(n2)−1.

Theorem 9.3.5.

The Chow ring A∗⁢(M¯0,n) is the quotient of ℤ⁡[DT∣T⊂[n],|T|,|TC|≥2] by the relations

  1. 1.

    DT=D(TC);

  2. 2.

    For any distinct i,j,k,ℓ∈[n], we have the equality

  3. 3.

    For T1,T2⊂[n], DT1⁢DT2=0 unless one of T1⊂T2,T2⊂T1,T1⊂T2C,T2⊂T1C holds.

Remark 9.3.6.

All of the relations encode geometric content:

  1. 1.

    As divisors, we already know that DT=D(TC).

  2. 2.

    If we consider the map πi,j,k,ℓ:M¯0,n→M¯0,4, then the three sums are the pullbacks of the three boundary divisors Di,j,Di,k,Di,ℓ⊂M¯0,4=ℙ1.

  3. 3.

    The final relation encodes the fact that DT1∩DT2=∅ unless one of the four inclusions holds. Pictorially, this is encoded in the diagram below:

    Figure 9.5: Degeneration to a common stable curve

9.4 Intersection theory of regular blowups

Let i:X⊂Y be a regularly embedded subvariety, π:Y~→Y be the blowup along X, and X~ be the exceptional divisor. Let g:X~→X and j:X~→Y~.

Theorem 9.4.1.

Suppose i∗ is surjective. Then

where P⁡(T) has constant term [X] and i∗⁢P⁢(T)=Td+Td−1⁢c1⁢(NX⁢Y)+⋯+cd⁢(NX⁢Y), where d is the codimension of X in Y. This is induced by −T=[X~].

Theorem 9.4.2.

A scheme X is HI if A∗⁢(X)=H∗⁢(X). If X,Y are both HI, then so is Y~.

Theorem 9.4.3.

The map

is an isomorphism.

10 Patrick (Apr 09): Do we even need derived categories?

Note: these are the speaker’s notes.

10.1 Serre formula

Recall the following from my March 19 lecture (this is Proposition 7.2.9 in my notes):

Proposition.

Let X be a smooth variety, V,W⊂X be closed subschemes that intersect properly, and Z be an irreducible component of V∩W. Then

where 1≤i⁡(Z,V⋅W;X)≤ℓ⁡(𝒪V∩W,Z) is the intersection number and i⁡(Z,V⋅W;X)=ℓ⁡(𝒪V∩W,Z) if and only if the local ring is Cohen-Macaulay.

However, most rings are not Cohen-Macaulay, so we would like a formula to compute the intersection multiplicities in all cases. Serre gives a formula in terms of higher Tor functors (because 𝒪V∩W,Z=𝒪V,Z⊗𝒪W,Z). Before we state the formula, first we will state some results about the higher Tor functors.

First, if X is a locally ringed space and ℱ,𝒢 are modules on X, then

This follows from the construction of the derived tensor product ⊗L in Stacks, which exposits derived categories much better than I ever could.

Lemma 10.1.1.

Let X be a locally Noetherian scheme. If ℱ,𝒢 are coherent, so is Tori𝒪X⁡(ℱ,𝒢). Also, if L,K∈Dcoh−⁢(𝒪X) (this means bounded above complexes of quasicoherent sheaves with coherent homology), then so is L⊗LK.

Proof of this fact is pure homological algebra, and again can be found in the Stacks project.

Lemma 10.1.2.

Let X be a smooth variety and ℱ,𝒢 be coherent sheaves. Then Tori𝒪X⁡(ℱ,𝒢) is supported on Supp⁡ℱ∩Supp⁡𝒢, and is nonzero only when 0≤i≤dimX.

Proof.

The support condition is clear by looking at the stalks, so we need to consider when the stalks are nonzero. Here, we note that because X is smooth, the local rings 𝒪X,x are regular local rings. By a result of Serre (Theorem 4.4.1 in my commutative algebra notes), 𝒪X,x has finite global dimension dim𝒪X,x≤dimX. Here, the global dimension and the dimension are the same by Theorem 4.3.12 of my commutative algebra notes.1 □

Now we can compute the intersection multiplicities as (Stacks gives this as the definition of intersection multiplicity)

This formula is due to Serre, and Stacks writes the total intersection as

Remark 10.1.3.

Stacks writes the intersection multiplicity as e⁡(X,V⋅W,Z). I am using the notation in Fulton’s book.

Lemma 10.1.4.

Assume that ℓ⁡(𝒪V∩W,Z)=1. Then i⁡(Z,V⋅W;X)=1 and V,W are smooth in a general point of Z.

Proof.

Write A=𝒪X,Z. Then dimA=dimX−dimZ. Let I,J be the ideals of V,W. By Proposition 7.2.9 of the notes,2 I+J=𝔪. Thus there exists f1,…,fr∈I,g1,…,fs∈J Forming a basis for 𝔪/𝔪2. But this is a regular sequence and a system of parameters, so A/(f1,…,fr) is a regular local ring of dimension dimX−dimV, so I=(f1,…,fr). Similarly, J=(g1,…,gs). Now by Corollary 4.4.3 of commutative algebra, the Koszul complex K⁡(f1,…,fr,A) resolves A/I, so we obtain

By Theorem 4.4.2 from commutative algebra, we only have H0=k. □

Example 10.1.5.

Suppose V,W⊂X are closed subvarieties, dimX=4, 𝒪̂X,p=ℂ⁡[[x,y,z,w]] and V=(x⁢z,x⁢w,y⁢z,y⁢w),W=(x−z,y−w). Then ℓ⁡(ℂ⁡[[x,y,z,w]]/(x⁢z,x⁢w,y⁢z,y⁢w,x−z,x−w))=3, but the intersection multiplicity is 2 because V is locally a union (x=y=0)∪(z=w=0).

10.2 Some algebra

Let (A,𝔪,k) be a Noetherian local ring. If M is a module and I is an ideal of definition, recall the Hilbert-Samuel polynomial φI,M⁢(n)=ℓ⁡(In⁢M/In+1⁢M). Similarly recall the function

Recall that d⁡(M)≔deg⁡χ is independent of I and equals the dimension of the support of M (from the proof of Theorem 3.2.9 in my commutative algebta notes). Now write χI,M⁢(n)=eI⁢(M,d)⁢ndd!+O⁡(nd−1).

Definition 10.2.1.

For d=d⁡(M) we write eI⁢(M,d) as above, and for d>d⁡(M), we set eI⁢(M,d)=0.

Lemma 10.2.2.

For all I,M, we have

Lemma 10.2.3.

Let P be a polynomial of degree r with leading coefficient a. Then

for any t.

Proof.

Write Δ for the operator taking a polynomial P to P⁡(t)−P⁡(t−1). Then

The desired claim follows from Pascal’s identity. □

Theorem 10.2.4.

Let A be a Noetherian local ring and I=(f1,…,fr) be an ideal of definition. Then

There is a very long proof of this statement in Stacks using spectral sequences.

10.3 Computing intersection multiplicities without derived categories

We give some cases where intersection multiplicities can be computed without using derived categories.

Lemma 10.3.1.

Suppose 𝒪V,Z and 𝒪W,Z are Cohen-Macaulay. Then i⁡(Z,V⋅W;X)=ℓ⁡(𝒪V∩W,Z).

Proof.

Write A=𝒪X,Z,B=𝒪V,Z,C=𝒪W,Z. Then by Auslander-Buchsbaum (exercise 4d of the final CA homework), we have a resolution F∙→B of length depth⁡A−depth⁡B=dimA−dimB=dimC. Then F∙⊗C represents B⊗LC and is supported in {𝔪A}, so by Lemma 10.108.2 in Stacks, it has nonzero cohomology only in degree 0. □

Lemma 10.3.2.

Let A be a Noetherian local ring and I=(f1,…,fr) is generated by a regular sequence. If M is a finite A-module with dimSupp⁡M/I⁢M=0, then

In what follows, we will assume V is cut out in 𝒪X,Z by a regular sequence (f1,…,fc).

Lemma 10.3.3.

In this case, we have i⁡(Z,V⋅W;X)=c!. This is the leading coefficient of the “Hilbert polynomial” n↦ℓ⁡(𝒪W,Z/(f1,…,fc)t).

Proof.

By the previous lemma, e⁡(Z,V⋅W;X)=e(f1,…,fc)⁢(𝒪W,Z⁢(c)). Now we need to show that dim𝒪W,Z=c. But now if dimV=r,dimW=s,dimX=n, dimZ=r+s−n, so k⁡(Z) has transcendence degree r+s−n. Because f1,…,fc is a regular sequence, r+c=n, so dim𝒪W,Z=s−(r+s−n)=s−(n−c+s−n)=c. □

Lemma 10.3.4.

Assume c=1 (for example, V is an effeective Cartier divisor). Then i⁡(Z,V⋅W;X)=ℓ⁡(𝒪W,Z/(f1)).

Proof.

Note that 𝒪W,Z is a Noetherian local domain of dimension 1. Then it is clear that ℓ⁡(𝒪W,Z/(f1t))=t⁢ℓ⁢(𝒪W,Z/(f1)) for all t≥1. □

Lemma 10.3.5.

Asssume 𝒪W,Z is Cohen-Macaulay. Then i⁡(Z,V⋅W;X)=ℓ⁡(𝒪W,Z/(f1,…,fc)).

Proof.

Because f1,…,fc is a regular sequence, it is also quasi-regular by Proposition 3.5.6 of my commutative algebra notes. Then

Now take the leading coefficient. □

11 Caleb (Apr 16): Motivic cohomology

Note: these are the speaker’s notes. No changes were made except to bring notational and stylistic conventions in line with the rest of the notes and to adapt pictures that were not available to me. Also, the quote of Grothendieck was changed to the original French.

These are my notes on motivic cohomology. Essentially everything here is based off of Voevodsky’s lectures, now turned into a book by Mazza and Weibel.

11.1 Bloch’s higher Chow groups

11.1.1 Some topological motivation

Recall the following exact sequence.

Proposition 11.1.1 (Fulton, Prop. 1.8).

Let Y be a closed subscheme of a scheme X, and let U=X−Y. Let i:Y→X,j:U→X be the inclusions. Then the sequence

is exact for all k.

This is great, but if we want homology to continue to the left. We certainly cannot put these sequences together. In fact, the indexing is a bit misleading in this way – instead, we should put the ks together for a grading of CH∗⁡(−). So, we really do need new groups if we want to continue this sequence. To construct these groups, we first recall the following definition of rational equivalence.

Definition 11.1.2 (Rat⁡(X)).

Let Z⁡(X) denote the cycles of a scheme X and let Φ be any subvariety of X×ℙ1. Then we define Rat⁡(X)⊂Z⁡(X) be the subgroup generated by differences of the form

Then rationally equivalent cycles are those which differ by something in Rat⁡(X). We see that it looks like there is a homotopy between rationally equivalent cycles.

11.1.2 Definition

Motivated by algebraic topology, we define the algebraic simplex

Let zi⁢(X,n) be the subgroup of Zi⁢(X×Δn) that meet all faces properly. This gives both a simplicial abelian group zi⁢(X,∙) and a chain complex zi⁢(X,∗).

Definition 11.1.3.

The higher Chow groups CHi⁡(X,m) are defined

11.1.3 Properties

  1. 1.

    Homotopy invariance:

    The projection X×𝔸1→X induces an isomorphism

  2. 2.

    Long exact sequence:

    There is a distinguished triangle

  3. 3.

    Isomorphism with rational K-theory:

We will see that Hp,q⁢(X;A)=CHq⁡(X,2⁢q−p;A). In particular, we have H2⁢q,q⁢(X,A)=CHq⁡(X)⊗A.

11.2 The category of correspondences

11.2.1 Correspondences

Let X,Y∈𝖲𝗆k be smooth separated schemes of finite type over k. Very informally, one can think of 𝖢𝗈𝗋⁡(X,Y) as a generalization of Hom⁡(X,Y) to multivalued morphisms.

Definition 11.2.1.

An elementary correspondence between a smooth connected scheme X/k to a separated scheme Y/k is an irreducible closed subset W⊂X×Y whose associated integral subscheme is finite and surjective over X.

If X is not connected, then an elementary correspondence refers to one that is one from a connected component of X to Y.

The group 𝖢𝗈𝗋⁡(X,Y) of finite correspondences is the free abelian group generated by the elementary correspondences.

Then given a closed subscheme Z⊂X×Y finite and surjective over X, we can associate the finite correspondence ∑ni⁢Wi where Wi are the irreducible components of the support of Z surjective over a component of X with generic points ξi and ni=ℓ⁡(OZ,ξi).

11.2.2 The category of correspondences

We compose correspondences V∈𝖢𝗈𝗋k⁢(X,Y) and W∈𝖢𝗈𝗋k⁢(Y,Z) as follows. Construct the cycle [T]=(V×Z)⋅(X×W) on X×Y×Z. Then take its pushforward along the projection p:X×Y×Z→X×Z.

It is not difficult to check that 𝖲𝗆k embeds into 𝖢𝗈𝗋k as a subcategory, where f:X→Y becomes the graph Γf⊂X×Y.

Furthermore, 𝖢𝗈𝗋k is a symmetric monoidal category. Indeed, the tensor product is simply X⊗Y=X×Y. Given V∈𝖢𝗈𝗋k⁢(X,X′) and W∈𝖢𝗈𝗋k⁢(Y,Y′), we get the desired cycle V×W∈𝖢𝗈𝗋k⁢(X⊗Y,X′⊗Y′).

11.2.3 Examples

  1. 1.

    𝖢𝗈𝗋k⁢(Spec⁡k,X) is generated by the 0-cycles of X.

  2. 2.

    𝖢𝗈𝗋k⁢(X,Spec⁡k) is generated by the irreducible components of X.

  3. 3.

    Take W∈𝖢𝗈𝗋k⁢(𝔸1,X) and two k-points s,t:Spec⁡k→𝔸1. Then the zero-cycles W∘Γs and W∘Γt are rationally equivalent.

11.3 Presheaves with transfers

11.3.1 Definition

Definition 11.3.1.

A presheaf with transfers is a contravariant additive functor F:𝖢𝗈𝗋k→𝖠𝖻.

Additivity gives a map

Thus there are extra “transfer maps" F⁡(Y)→F⁡(X) coming from 𝖢𝗈𝗋k⁢(X,Y).

Theorem 11.3.2.

𝖯𝖲𝖳⁡(k) is an abelian category with enough injectives and projectives.

11.3.2 Examples

Example 11.3.3.

The constant presheaf A on 𝖲𝗆k can be extended to a pst.

For W∈𝖢𝗈𝗋⁡(X,Y) with X,Y connected, the corresponding homomorphism A→A is multiplication by the degree of W over X.

Example 11.3.4.

𝒪∗ and 𝒪, at least for X normal. Use the norm and trace maps.

Example 11.3.5.

CHi⁡(−), the Chow groups.

Example 11.3.6.

Representable functors: hX⁢(−)

11.3.3 Representable functors of Cork⁢(X)

Take X∈Ob⁡(𝖢𝗈𝗋k⁢(X)). We denote

By Yoneda, ℤt⁢r⁢(X) is a projective object in 𝖯𝖲𝖳⁡(k).

Note that ℤt⁢r⁢(Spec⁡k) is just the constant sheaf ℤ on 𝖲𝗆k, with the transfer maps constructed in the example from the previous subsection. Let (X,x) be a pointed scheme. We define

The structure map X→Spec⁡k provides a splitting, so

Out of laziness we screenshot the following definitions from Voevodsky’s lectures.

Definition 11.3.7.

If (Xi,xi) are pointed schemes for i=1,…,n we define ℤt⁢r⁢((X1,x1)∧⋯∧(Xn,xn)), or ℤt⁢r⁢(X1∧⋯⁢Xn), to be:

Lemma 11.3.8.

The presheaf ℤt⁢r⁢((X1,x1)∧⋯∧(Xn,xn)) is a direct summand of ℤt⁢r⁢(X1×⋯×Xn). In particular, it is a projective object of 𝖯𝖲𝖳.

Moreover, the following sequence of presheaves with transfers is split-exact:

Consider the pointed scheme (𝔾m,1). We will be interested in the presheaf with transfers ℤt⁢r⁢(𝔾m∧q).

Before continuing, we recall our construction

Recall that a simplicial object of a category C is a functor F:Δo⁢p→C. Then if F is a presheaf of abelian groups on 𝖲𝗆k, then F⁡(U×Δ∙) is a simplicial abelian group. Then

is a simplicial presheaf with transfers. Similarly, C∗⁢F⁢(U) gives the complex of abelian groups

11.3.4 Homotopy invariant presheaves

Definition 11.3.9.

A presheaf F is homotopy invariant if for every X, the map p∗:F⁡(X)→F⁡(X×𝔸1) is an isomorphism.

Note that this is equivalent to p∗ being surjective. We can check that an equivalent condition is that for all X, we have

Furthermore, if F is any presheaf, we have that i0∗,i1∗:C∗⁢F⁢(X×𝔸1)→C∗⁢F⁢(X) are chain homotopic. From this we deduce that if F is a presheaf, then the homology presheaves

are homotopy invariant for all n.

Definition 11.3.10.

Two finite correspondences from X to Y are 𝔸1-homotopic if they are the restrictions along X×0 and X×1 of an element of 𝖢𝗈𝗋⁡(X×𝔸1,Y).

This is an equivalence relation on 𝖢𝗈𝗋⁡(X,Y). Note that it is not one if we just look at morphisms of schemes! With this definition though, we define f:X→Y to be an 𝔸1-homotopy equivalence in the expected way.

11.4 Motivic cohomology

11.4.1 The motivic complex

Definition 11.4.1.

For q∈ℤ≥0, the motivic complex ℤ⁡(q) is defined as the following complex of presheaves with transfers.

We can change coefficients to A∈𝖠𝖻 by setting A⁡(q)=ℤ⁡(q)⊗A.

These are actually complexes of sheaves with respect to the Zariski topology. In fact, they are also sheaves in the étale topology.

For example when q=0, applying this to a scheme Y we just get

which is quasi-isomorphic to just ℤ. When q=1, the complex looks like

11.4.2 Motivic cohomology groups

Definition 11.4.2.

The motivic cohomology groups Hp,q⁢(X,ℤ) are defined to be the hypercohomology of the motivic complexes ℤ⁡(q) with respect to the Zariski topology:

If A is any abelian group, we define

11.4.3 Weight 1

There is a quasi-isomorphism

Thus we have the following table.

ℤ⁡(X)𝒪∗⁢(X)Pic⁡(X)000000000000000H−2,2H−1,2H0,2H1,2H2,2H3,2qp
Figure 11.1: Weight q motivic cohomology

11.5 Relation to other fields

11.5.1 Algebraic K-theory

Atiyah-Hirzebruch:

In the algebraic setting, it is much more difficult. Indeed, both algebraic K-theory and motivic cohomology are significantly harder to define than their topological counterparts. In 2002, Suslin and Friedlander built upon previous work of Bloch and Lichtenbaum to show the following spectral sequence.

11.5.2 Motives

C’est pour parvenir à exprimer cette intuition de “parenté” entre théories cohomologiques différentes, que j’ai dégagé la notion du “motif” associé à une variété algébrique. Par ce terme, j’entends suggérer qu’il s’agit du “motif commun” (or de la “raison commune”) sous-jacent à cette multitude d’invariants cohomologiques différents associés à là variété, à l’aide de la multitude des toutes les théories cohomologiques possibles a priori.Alexander Grothendieck

Note: there was an English translation originally in the slideshow, but I don’t actually think the translation is very good. I provided my own translation at some point, but that was also bad so it was removed.

Grothendieck constructed Chow motives by replacing morphisms of schemes with correspondences (defined under rational equivalence, different from the correspondences discussed earlier), augmenting the category to look like an abelian category, and taking the opposite category. By construction, this works in that cohomology theories factor through it. However, to truly achieve what is desired from them, one must assume the standard conjectures on algebraic cycles (or some variants), which have been open for over 50 years!

Voevodsky used motivic cohomology to construct a triangulated category D⁢M⁢(k;R), which for all intents and purposes acts as the derived category of the desired category of motives. He studied mixed motives, which apply to all varieties (not just the smooth ones). These can be thought of as extensions of pure motives, and motivic cohomology studies these E⁢x⁢t groups.

11.5.3 Arithmetic geometry

There’s the Bloch-Kato conjecture and the Bloch-Kato conjectures, which are different!

The Bloch-Kato conjecture is now a theorem: the norm residue isomorphism theorem, proven by Voevodsky. Through proving it, Voevodsky developed motivic cohomology, motivic homotopy theory, motivic Steenrod algebra…

The norm residue isomorphism theorem (or Bloch-Kato conjecture) states that for a field k and an integer ℓ that is inverible in k, the norm residue map

from Milnor K-theory mod-ℓ to étale cohomology is an isomorphism. The case ℓ=2 is the Milnor conjecture, and the case n=2 is the Murkurjev-Suslin theorem.Wikipedia

The Bloch-Kato conjectures are on special values of L-functions.

Conjecture 11.5.1 (Soulé).

If X is regular and proper over Spec⁡ℤ, then for an integer n∈ℤ we have

Applied to elliptic curves, this implies (one of the two parts of) the Birch-Swinnerton Dyer conjecture!