Intersection Numbers of Varieties

April 11, 2025

We adopt an alternative definition of Cartier divisors than seen for example in Vakil's FOAG. Let XX be a scheme, and define the sheaf of total quotient rings KX\fank_X as the sheaf whose UU-points KX(U)\fank_X(U) is the localization of OX(U)\ox(U) by the multiplicative set of non zero divisors. We can define a Cartier divisor as a global section of the sheaf KX/OX\fank_X^*/\ox^*, where KX(U)\fank_X^*(U) is the group of invertible elements in KX(U)\fank_X(U).

Perhaps more concretely, a Cartier divisor is a collection of tuples (Ui,fi)(U_i, f_i) where UiU_i is an open cover of XX and fif_i is a unit in KX(Ui)\fank_X(U_i), such that for any other tuple (Uj,fj)(U_j, f_j) we have fi/fjOX(UiUj)f_i/f_j \in \calo_X^*(U_i \cap U_j).

Note that if XX is reduced then we may take a cover by integral (=reduced + irreducible) UiU_i such that KX(Ui)=K(Ui)\fank_X(U_i) = K(U_i) is just the field of fractions of OX(Ui)\ox(U_i). Then fif_i is merely a nonzero rational function on UiU_i such that fi/fjf_i/f_j is a nonvanishing regular function on UiUjU_i \cap U_j.

We say a Cartier divisor is principal if it is the image of the natural map

H0(X,KX)H0(X,KX/OX).H^0(X, \fank^*_X) \to H^0(X, \fank^*_X/\ox^*).

That is, it can be defined by a global nonzero rational function on XX. Two Cartier divisors D,DD,D' are linearly equivalent if DDD-D' is principal, in which case we write DDD \equiv D'. We say DD is effective if for all tuples (Ui,fi)(U_i, f_i), the fif_i are not zero divisors in OX(Ui)\ox(U_i); the fif_i cut out a codimension 11 subscheme of XX whose reduction is the support of DD.

This means we have a natural bijection between effective Cartier divisors DD and subschemes of XX which are locally cut out by a single non zero divisor. For this reason, effective Cartier divisor may refer to either the algebraic or geometric definition depending on context.

Hilbert Functions

Fix a field kk. For this section we will assume all schemes are finite type over k.k.

Let XX be a proper scheme and F\fanf a coherent sheaf on X.X. Recall the Euler characteristic is defined as the alternating sum

χ(X,F):=i=0(1)ihi(X,Fi).\chi(X, \fanf) := \sum_{i=0}^{\infty} (-1)^i h^i(X, \fanf^{i}).

Let XX be a subscheme of PkN\bbp^N_k of dimension nn (which is projective hence proper), and recall we have the Serre twisting sheaf OX(m)\ox(m) restricted to X.X. For a more general coherent sheaf F\fanf, we defined F(m):=FOX(m).\fanf(m) := \fanf \otimes \ox(m). Then the function

hF(m):=χ(X,F(m))h_{\fanf}(m) := \chi(X, \fanf(m))

is a polynomial of degree dimSuppF\dim \supp \fanf, called the Hilbert function of F.\fanf. By Serre vanishing, this implies

pF(m):=h0(X,F(m))p_{\fanf}(m) := h^0(X, \fanf(m))

is a polynomial for m0m \gg 0 called the Hilbert polynomial of F.\fanf. In particular, if F=OX\fanf = \ox then we define pX(m)=pOX(m)p_X(m) = p_{\ox}(m) as the Hilbert polynomial of XX, which has degree dimX\dim X.

We make a somewhat unintuitive definition of the degree of XX, which is n!n! times the leading coefficient of pX(m)p_X(m). That is, pX(m)=(deg(X)/n!)mn+p_X(m) = (\deg(X)/n!) m^n + \ldots We can give some justification as for why this is a good notion of degree.

  1. The degree of a hypersurface HH in PkN\bbp^N_k is the degree of the single nonzero homogeneous polynomial that cuts out H.H. Show that this definition coincides with the above.
  2. Suppose a regular curve CC is embedded in PkN\bbp^N_k via an invertible sheaf (line bundle) of degree dd. Show that the degree of the embedded curve with the above definition is also dd.

Now we return to the world of Cartier divisors and attempt to define the intersection product. Let D1,,DrD_1, \ldots, D_r be Cartier divisors on a proper scheme X,X, and let F\fanf be a coherent sheaf on X.X. Consider the generalized Hilbert function

hF(m1,,mr):=χ(X,F(m1D1++mrDr)).h_{\fanf}(m_1,\ldots,m_r) := \chi(X, \fanf(m_1 D_1 + \ldots + m_r D_r)).

Then this function takes the same values on Zr\bbz^r as a polynomial with rational coefficients of degree at most dimSuppF\dim \supp \fanf.

Throughout the proof we will use the fact that, for a short exact sequence

0GFH0,0 \to \fang \to \fanf \to \fanh \to 0,

if the result holds for H,G\fanh, \fang then it holds for F\fanf to reduce the problem to simpler statements.

Any coherent sheaf F\fanf on a Noetherian scheme XX admits a filtration

F=F0F1Fm=0\fanf = \fanf_0 \supset \fanf_1 \supset \ldots \supset \fanf_m = 0

such that Fi/Fi+1\fanf_i/\fanf_{i+1} is torsion free on an integral subscheme of X.X. We'll give a slightly nonrigorous justification. Let ξ\xi be an associated point of F\fanf and suppose Z={ξ}Z = \overline{\{\xi\}} is an integral closed subscheme of XX which is the closure of ξ.\xi. Since SuppF\supp \fanf is the closure of the (finitely many) associated points of F\fanf, ZZ must be contained in the support. The embedding ZXZ \to X yields a short exact sequence

0IX/ZFFFZ00 \to \fani_{X/Z} \otimes \fanf \to \fanf \to \fanf\vert_Z \to 0

of coherent sheaves. Let F1:=IX/ZF\fanf_1 := \fani_{X/Z} \otimes \fanf such that F/F1=FZ.\fanf/\fanf_1 = \fanf\vert_Z. Since F1\fanf_1 is the sheaf of sections of F\fanf that are annihilated by ξ,\xi, this quotient is torsion free along Z.Z. This also implies SuppF1\supp \fanf_1 is strictly contained in SuppF,\supp \fanf, so by the Noetherian condition this filtration must be finite.

Back to the Theorem at hand. The existence of the filtration implies it suffices to show the result for XX integral and F\fanf torsion-free. Then there exists a dense open UU of XX over which F\fanf is free of rank rr. The choice of isomorphism FOUOUr\fanf \otimes \calo_U \simeq \calo_U^{\oplus r} determines an embedding of the following form.

FKXr\fanf \hookrightarrow \fank_X^{\oplus r}

Let G=FOXr\fang = \fanf \cap \ox^{\oplus r} and note that in the sequences

0GFG100 \to \fang \to \fanf \to \fang_1 \to 0 0GOXrG200 \to \fang \to \ox^{\oplus r} \to \fang_2 \to 0

the sheaves G1,G2\fang_1, \fang_2 are not supported on UU, hence their supports are strictly contained in XX. By induction on the dimension of XX, we can reduce to the case F=OX\fanf = \calo_X and XX integral.

The rest of the proof can be found in Debarre's Higher-Dimensional Algebraic Geometry, but the gist is that this is a generalization of the fact that the Hilbert function in one variable is a polynomial. That is,

hF(m1,,mr)=P(m1,,mr)h_{\fanf}(m_1,\ldots,m_r) = P(m_1,\ldots,m_r)

is a polynomial where PQ[T1,,Tr]P \in \bbq[T_1,\ldots,T_r]. Let dd be the total degree of PP, and let n1,,nrn_1,\ldots,n_r be integers such that

Q(T)=P(n1T,,nrT)Q(T) = P(n_1 T, \ldots, n_r T)

still has degree dd. Then since Q(m)=χ(X,m(n1D1++nrdr))Q(m) = \chi(X,m(n_1D_1 + \ldots + n_r d_r)) it follows that ddimXd \leq \dim X.

Intersection Numbers

The intersection number

D1DrD_1 \cdots D_r

of a collection of rdimXr \geq \dim X Cartier divisors D1,,DrD_1,\ldots,D_r on a proper scheme XX is defined as the coefficient of m1mrm_1 \cdots m_r in the polynomial

χ(X,m1D1++mrDr)\chi(X, m_1D_1 + \ldots + m_r D_r)

In particular, we can show that this number is an integer and that it vanishes for r>dimXr > \dim X. If the DiD_i are effective and meet properly in a finite number of points, it has a nice geometric interpretation as the number of points contained in the intersection D1DrD_1 \cap \ldots \cap D_r counted with multiplicity.

For YY a subscheme of XX, we can define

D1DrY=D1YDrYD_1 \cdots D_r \cdot Y = D_1 \vert_Y \cdots D_r \vert_Y

and similarly for F\fanf a coherent sheaf on XX with dimSuppFr\dim \supp \fanf \leq r, we can define

D1DrFD_1 \cdots D_r \cdot \fanf

as the coefficient of m1mrm_1 \cdots m_r in

χ(X,F(m1D1++mrDr))\chi(X, \fanf(m_1 D_1 + \ldots + m_rD_r))