ConceptComplete

Intersection Theory on Surfaces

Intersection theory on surfaces assigns an integer Cβ‹…DC \cdot D to any two divisors on a smooth projective surface, generalizing the naive count of intersection points. This bilinear pairing on Pic⁑(X)\operatorname{Pic}(X) is the fundamental tool for studying the geometry of surfaces.


The intersection pairing

Definition5.1Intersection number

Let XX be a smooth projective surface over an algebraically closed field kk. For divisors C,DC, D on XX, the intersection number is:

Cβ‹…D=Ο‡(OX)βˆ’Ο‡(OX(βˆ’C))βˆ’Ο‡(OX(βˆ’D))+Ο‡(OX(βˆ’Cβˆ’D)).C \cdot D = \chi(\mathcal{O}_X) - \chi(\mathcal{O}_X(-C)) - \chi(\mathcal{O}_X(-D)) + \chi(\mathcal{O}_X(-C-D)).

This depends only on the linear equivalence classes of CC and DD, so it defines a symmetric bilinear pairing:

Pic⁑(X)Γ—Pic⁑(X)β†’Z,(C,D)↦Cβ‹…D.\operatorname{Pic}(X) \times \operatorname{Pic}(X) \to \mathbb{Z}, \quad (C, D) \mapsto C \cdot D.

When CC and DD are effective and meet transversally, Cβ‹…D=∣C∩D∣C \cdot D = |C \cap D| (the naive count).

ExampleTransversal intersections in β„™Β²

In P2\mathbb{P}^2, let C=V(F)C = V(F) and D=V(G)D = V(G) be curves of degrees dd and ee. If they meet transversally:

Cβ‹…D=de.C \cdot D = de.

This is Bezout's theorem. Concretely:

  • A line and a conic (d=1,e=2d = 1, e = 2): 1β‹…2=21 \cdot 2 = 2 intersection points.
  • Two conics (d=e=2d = e = 2): 2β‹…2=42 \cdot 2 = 4 intersection points.
  • A line and a cubic (d=1,e=3d = 1, e = 3): 1β‹…3=31 \cdot 3 = 3 intersection points.
  • Two cubics (d=e=3d = e = 3): 3β‹…3=93 \cdot 3 = 9 points β€” this gives the Cayley–Bacharach theorem.
ExampleNon-transversal intersections

When curves meet with higher multiplicity, the intersection number accounts for it via local intersection multiplicities:

Cβ‹…D=βˆ‘P∈C∩D(Cβ‹…D)P,(Cβ‹…D)P=dim⁑kOX,P/(f,g),C \cdot D = \sum_{P \in C \cap D} (C \cdot D)_P, \quad (C \cdot D)_P = \dim_k \mathcal{O}_{X,P}/(f, g),

where f,gf, g are local equations for C,DC, D at PP.

  • Two lines meeting at a point: (Cβ‹…D)P=1(C \cdot D)_P = 1.
  • Line y=0y = 0 tangent to parabola y=x2y = x^2 at origin: local equations yy and yβˆ’x2y - x^2. The ring k[x,y](x,y)/(y,yβˆ’x2)β‰…k[x]/(x2)k[x,y]_{(x,y)}/(y, y - x^2) \cong k[x]/(x^2) has dimension 22. So (Cβ‹…D)O=2(C \cdot D)_O = 2.
  • Cusp y2=x3y^2 = x^3 and line y=0y = 0 at origin: k[x,y](x,y)/(y,y2βˆ’x3)β‰…k[x]/(x3)k[x,y]_{(x,y)}/(y, y^2 - x^3) \cong k[x]/(x^3), dimension 33.

Self-intersection

Definition5.2Self-intersection number

The self-intersection C2=Cβ‹…CC^2 = C \cdot C measures the "twisting" of the normal bundle NC/XN_{C/X}:

C2=deg⁑(NC/X)=deg⁑(OX(C)∣C).C^2 = \deg(N_{C/X}) = \deg(\mathcal{O}_X(C)|_C).

Unlike intersection with a different curve, C2C^2 can be negative! A curve with C2<0C^2 < 0 is called a negative curve.

ExampleSelf-intersection in β„™Β²

In P2\mathbb{P}^2, a curve CC of degree dd has C2=d2C^2 = d^2:

  • A line LL: L2=1L^2 = 1. Geometrically: perturbing LL to another line Lβ€²L', they meet in 11 point.
  • A conic: C2=4C^2 = 4.
  • A cubic: C2=9C^2 = 9.

In P2\mathbb{P}^2, all self-intersections are positive. This is a special feature of P2\mathbb{P}^2.

ExampleNegative self-intersection

On the blowup X~=Bl⁑P(P2)\tilde{X} = \operatorname{Bl}_P(\mathbb{P}^2), the exceptional divisor EE (the fiber over the blown-up point) satisfies:

E2=βˆ’1.E^2 = -1.

This is the prototype of a negative curve. More generally:

  • On a ruled surface Fn=P(OβŠ•O(βˆ’n))\mathbb{F}_n = \mathbb{P}(\mathcal{O} \oplus \mathcal{O}(-n)) over P1\mathbb{P}^1: the section C0C_0 with C02=βˆ’nC_0^2 = -n.
  • On a K3 surface: any smooth rational curve Cβ‰…P1C \cong \mathbb{P}^1 has C2=βˆ’2C^2 = -2 (by adjunction: βˆ’2=2g(C)βˆ’2=C2+Kβ‹…C=C2+0-2 = 2g(C) - 2 = C^2 + K \cdot C = C^2 + 0).

Intersection theory on product surfaces

Exampleβ„™ΒΉ Γ— β„™ΒΉ

On X=P1Γ—P1X = \mathbb{P}^1 \times \mathbb{P}^1, the Picard group is Pic⁑(X)β‰…Z2\operatorname{Pic}(X) \cong \mathbb{Z}^2, generated by:

  • F1=P1Γ—{pt}F_1 = \mathbb{P}^1 \times \{pt\} (a "horizontal" fiber),
  • F2={pt}Γ—P1F_2 = \{pt\} \times \mathbb{P}^1 (a "vertical" fiber).

The intersection pairing is:

  • F1β‹…F2=1F_1 \cdot F_2 = 1 (they meet in one point).
  • F12=0F_1^2 = 0 (parallel horizontal fibers don't meet).
  • F22=0F_2^2 = 0 (parallel vertical fibers don't meet).

A divisor of bidegree (a,b)(a, b) means D∼aF1+bF2D \sim aF_1 + bF_2. Then:

  • D2=2abD^2 = 2ab (since (aF1+bF2)2=2abβ‹…F1β‹…F2=2ab(aF_1 + bF_2)^2 = 2ab \cdot F_1 \cdot F_2 = 2ab).
  • Dβ‹…Dβ€²=abβ€²+aβ€²bD \cdot D' = ab' + a'b for Dβ€²βˆΌaβ€²F1+bβ€²F2D' \sim a'F_1 + b'F_2.
  • A curve of bidegree (2,2)(2, 2) is an elliptic curve (g=1g = 1 by adjunction).
ExampleIntersection on an abelian surface

Let AA be an abelian surface (a 22-dimensional abelian variety). If LL is an ample line bundle with L2=2dL^2 = 2d, then LL defines a polarization of type (1,d)(1, d).

  • Principally polarized (d=1d = 1): L2=2L^2 = 2. The theta divisor Θ\Theta satisfies Θ2=2\Theta^2 = 2.
  • (1,2)(1,2)-polarization: L2=4L^2 = 4.
  • By the Nakai–Moishezon criterion, LL is ample iff L2>0L^2 > 0 and Lβ‹…C>0L \cdot C > 0 for all curves CC.

The Neron–Severi group and numerical equivalence

Definition5.3Neron–Severi group

Two divisors C,DC, D are numerically equivalent (C≑DC \equiv D) if Cβ‹…E=Dβ‹…EC \cdot E = D \cdot E for all curves EE on XX.

The Neron–Severi group is NS⁑(X)=Pic⁑(X)/≑\operatorname{NS}(X) = \operatorname{Pic}(X)/\equiv. Its rank ρ(X)=rk⁑NS⁑(X)\rho(X) = \operatorname{rk} \operatorname{NS}(X) is the Picard number.

The intersection pairing descends to a non-degenerate bilinear form on NS⁑(X)βŠ—R\operatorname{NS}(X) \otimes \mathbb{R}, of signature (1,Οβˆ’1)(1, \rho - 1) by the Hodge Index Theorem.

ExamplePicard numbers of surfaces
  • P2\mathbb{P}^2: ρ=1\rho = 1. NS⁑(P2)=Zβ‹…H\operatorname{NS}(\mathbb{P}^2) = \mathbb{Z} \cdot H, H2=1H^2 = 1.
  • P1Γ—P1\mathbb{P}^1 \times \mathbb{P}^1: ρ=2\rho = 2. Intersection matrix (0110)\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, signature (1,1)(1, 1).
  • Blowup Bl⁑P(P2)\operatorname{Bl}_P(\mathbb{P}^2): ρ=2\rho = 2. Basis {H,E}\{H, E\} with H2=1H^2 = 1, E2=βˆ’1E^2 = -1, Hβ‹…E=0H \cdot E = 0. Matrix diag⁑(1,βˆ’1)\operatorname{diag}(1, -1), signature (1,1)(1,1).
  • K3 surface: ρ\rho ranges from 11 to 2020. A general K3 has ρ=1\rho = 1; the maximum ρ=20\rho = 20 is achieved by "singular K3 surfaces" (e.g., the Fermat quartic x4+y4+z4+w4=0x^4 + y^4 + z^4 + w^4 = 0 over Fpβ€Ύ\overline{\mathbb{F}_p} for some pp).
  • Abelian surface: ρ\rho ranges from 11 to 44. Over C\mathbb{C}: ρ=1\rho = 1 generically; ρ=4\rho = 4 for products E1Γ—E2E_1 \times E_2 with CM.

Ampleness criteria

TheoremNakai–Moishezon criterion

A divisor DD on a smooth projective surface XX is ample if and only if:

  • D2>0D^2 > 0, and
  • Dβ‹…C>0D \cdot C > 0 for every irreducible curve CβŠ‚XC \subset X.
ExampleTesting ampleness

On Bl⁑P(P2)\operatorname{Bl}_P(\mathbb{P}^2) with basis H,EH, E (H2=1,E2=βˆ’1,Hβ‹…E=0H^2 = 1, E^2 = -1, H \cdot E = 0):

Is D=2Hβˆ’ED = 2H - E ample?

  • D2=4βˆ’1=3>0D^2 = 4 - 1 = 3 > 0 βœ“
  • Dβ‹…H=2>0D \cdot H = 2 > 0 βœ“
  • Dβ‹…E=0βˆ’(βˆ’1)=1>0D \cdot E = 0 - (-1) = 1 > 0 βœ“
  • For any curve C∼aHβˆ’bEC \sim aH - bE with a>0a > 0: Dβ‹…C=2a+bD \cdot C = 2a + b. If CC is effective and irreducible, then Cβ‹…H=a>0C \cdot H = a > 0 and C2=a2βˆ’b2β‰₯βˆ’1C^2 = a^2 - b^2 \geq -1 (for smooth rational curves). One checks Dβ‹…C>0D \cdot C > 0 for all such CC.

So D=2Hβˆ’ED = 2H - E is ample. Geometrically, ∣2Hβˆ’E∣|2H - E| maps Bl⁑P(P2)\operatorname{Bl}_P(\mathbb{P}^2) to P2\mathbb{P}^2 via conics through PP, embedding the blowup as a cubic surface in P3\mathbb{P}^3... actually, ∣2Hβˆ’E∣|2H - E| is a g?2g^2_?; more precisely, 2Hβˆ’E2H - E embeds Bl⁑P(P2)\operatorname{Bl}_P(\mathbb{P}^2) as a degree-33 surface in P3\mathbb{P}^3 (a cubic surface is the blowup of P2\mathbb{P}^2 at 66 points, not 11 β€” let me correct: ∣3Hβˆ’E∣|3H - E| gives a cubic in P3\mathbb{P}^3).

TheoremKleiman's criterion

A divisor DD is ample if and only if DD is in the interior of the nef cone: Dβ‹…C>0D \cdot C > 0 for all nonzero effective 11-cycles CC. Equivalently, DD is nef (Dβ‹…Cβ‰₯0D \cdot C \geq 0 for all curves CC) and D2>0D^2 > 0.

The cone theorem (Mori): the closure of the effective cone of curves is a rational polyhedral cone in the half-space {KXβ‹…Cβ‰₯0}\{K_X \cdot C \geq 0\}, plus countably many extremal rays Rβ‰₯0[Ci]\mathbb{R}_{\geq 0}[C_i] with KXβ‹…Ci<0K_X \cdot C_i < 0.


The adjunction formula

TheoremAdjunction formula for curves on surfaces

If CC is a smooth curve on a smooth surface XX:

2g(C)βˆ’2=C2+KXβ‹…C,2g(C) - 2 = C^2 + K_X \cdot C,

or equivalently Ο‰Cβ‰…(Ο‰XβŠ—OX(C))∣C\omega_C \cong (\omega_X \otimes \mathcal{O}_X(C))|_C. This relates the genus of CC to its self-intersection and the canonical class of XX.

ExampleAdjunction in action
  • Degree-dd curve in P2\mathbb{P}^2: KP2=βˆ’3HK_{\mathbb{P}^2} = -3H, C∼dHC \sim dH, so 2gβˆ’2=d2βˆ’3d2g - 2 = d^2 - 3d, giving g=(dβˆ’1)(dβˆ’2)2g = \frac{(d-1)(d-2)}{2}. Recovers the genus-degree formula.

  • Curve of bidegree (a,b)(a,b) on P1Γ—P1\mathbb{P}^1 \times \mathbb{P}^1: K=βˆ’2F1βˆ’2F2K = -2F_1 - 2F_2, so 2gβˆ’2=2ab+(βˆ’2bβˆ’2a)=2abβˆ’2aβˆ’2b2g - 2 = 2ab + (-2b - 2a) = 2ab - 2a - 2b, giving g=(aβˆ’1)(bβˆ’1)g = (a-1)(b-1).

  • Smooth rational curve on a K3: KK3=0K_{K3} = 0, so 2gβˆ’2=C22g - 2 = C^2. For g=0g = 0: C2=βˆ’2C^2 = -2. Every (βˆ’2)(-2)-curve on a K3 is a smooth rational curve.

  • Exceptional curve on blowup: Eβ‰…P1E \cong \mathbb{P}^1, g=0g = 0, KX~β‹…E=βˆ’1K_{\tilde{X}} \cdot E = -1 (by KX~=Ο€βˆ—KX+EK_{\tilde{X}} = \pi^*K_X + E). So βˆ’2=E2+(βˆ’1)-2 = E^2 + (-1), giving E2=βˆ’1E^2 = -1 βœ“.

  • Canonical curve on a surface of general type: If C∈∣KX∣C \in |K_X| is smooth, then 2g(C)βˆ’2=KX2+KX2=2KX22g(C) - 2 = K_X^2 + K_X^2 = 2K_X^2, so g(C)=KX2+1g(C) = K_X^2 + 1.


Riemann–Roch and intersection numbers

RemarkRiemann–Roch for surfaces (preview)

The Riemann–Roch theorem for a divisor DD on a surface XX reads:

Ο‡(OX(D))=12Dβ‹…(Dβˆ’KX)+Ο‡(OX).\chi(\mathcal{O}_X(D)) = \frac{1}{2}D \cdot (D - K_X) + \chi(\mathcal{O}_X).

Every term involves intersection numbers:

  • Dβ‹…(Dβˆ’KX)=D2βˆ’Dβ‹…KXD \cdot (D - K_X) = D^2 - D \cdot K_X.
  • Ο‡(OX)=1+pgβˆ’q\chi(\mathcal{O}_X) = 1 + p_g - q where pg=h0(KX)p_g = h^0(K_X) and q=h1(OX)q = h^1(\mathcal{O}_X).

For D=0D = 0: Ο‡(OX)=Ο‡(OX)\chi(\mathcal{O}_X) = \chi(\mathcal{O}_X) βœ“. For D=KXD = K_X: Ο‡(KX)=12KX2βˆ’12KX2+Ο‡(OX)=Ο‡(OX)\chi(K_X) = \frac{1}{2}K_X^2 - \frac{1}{2}K_X^2 + \chi(\mathcal{O}_X) = \chi(\mathcal{O}_X). This recovers the Serre duality statement Ο‡(Ο‰X)=Ο‡(OX)\chi(\omega_X) = \chi(\mathcal{O}_X).


Summary

RemarkThe intersection form encodes geometry

The intersection pairing NS⁑(X)Γ—NS⁑(X)β†’Z\operatorname{NS}(X) \times \operatorname{NS}(X) \to \mathbb{Z} is the central invariant of a surface:

  • Its signature (1,Οβˆ’1)(1, \rho - 1) constrains the geometry (Hodge Index Theorem).
  • Ampleness is detected by positivity against all curves (Nakai–Moishezon).
  • Adjunction relates curve genus to intersection data.
  • Riemann–Roch expresses Euler characteristics in terms of intersection numbers.
  • The cone of curves and nef cone organize the birational geometry.