ConceptComplete

Birational Geometry of Surfaces

Birational geometry studies surfaces "up to birational equivalence"---identifying two surfaces when they share a common dense open subset. For surfaces, the minimal model program has a complete and beautiful answer: every surface is obtained from a unique minimal model (for ΞΊβ‰₯0\kappa \geq 0) by a sequence of blowups, and the Enriques--Kodaira classification organizes all minimal surfaces by Kodaira dimension.


Birational maps and equivalence

Definition5.13Birational map

A rational map Ο•:Xβ‡’Y\phi: X \dashrightarrow Y between smooth projective surfaces is a morphism defined on a dense open subset UβŠ‚XU \subset X. It is birational if there exists a rational map ψ:Yβ‡’X\psi: Y \dashrightarrow X such that Οˆβˆ˜Ο•=idX\psi \circ \phi = \mathrm{id}_X and Ο•βˆ˜Οˆ=idY\phi \circ \psi = \mathrm{id}_Y on appropriate dense open subsets.

Equivalently, XX and YY are birationally equivalent if their function fields are isomorphic: k(X)β‰…k(Y)k(X) \cong k(Y).

A surface XX is rational if it is birational to P2\mathbb{P}^2, i.e., k(X)β‰…k(x,y)k(X) \cong k(x, y).

ExampleThe standard Cremona transformation

The standard quadratic transformation Οƒ:P2β‡’P2\sigma: \mathbb{P}^2 \dashrightarrow \mathbb{P}^2 is defined by:

Οƒ([x0:x1:x2])=[x1x2:x0x2:x0x1].\sigma([x_0 : x_1 : x_2]) = [x_1 x_2 : x_0 x_2 : x_0 x_1].

This is a birational involution (Οƒ2=id\sigma^2 = \mathrm{id}). It is undefined at the three coordinate points P0=[1:0:0]P_0 = [1:0:0], P1=[0:1:0]P_1 = [0:1:0], P2=[0:0:1]P_2 = [0:0:1], and it contracts the three coordinate lines xi=0x_i = 0 to the opposite coordinate points.

Resolving: blow up P0,P1,P2P_0, P_1, P_2 to get P~2\tilde{\mathbb{P}}^2. On P~2\tilde{\mathbb{P}}^2, Οƒ\sigma becomes a morphism that contracts the proper transforms of the three coordinate lines (each now a (βˆ’1)(-1)-curve) and blows down to P2\mathbb{P}^2.

ExampleBirationally equivalent surfaces
  • P2\mathbb{P}^2 and P1Γ—P1\mathbb{P}^1 \times \mathbb{P}^1 are birational: the map P2β‡’P1Γ—P1\mathbb{P}^2 \dashrightarrow \mathbb{P}^1 \times \mathbb{P}^1 given by projection from a point realizes this. Concretely, Bl⁑P(P2)β‰…F1\operatorname{Bl}_P(\mathbb{P}^2) \cong \mathbb{F}_1 admits a morphism to P1Γ—P1\mathbb{P}^1 \times \mathbb{P}^1 by blowing down a different (βˆ’1)(-1)-curve.

  • The Hirzebruch surfaces Fn\mathbb{F}_n for all nβ‰₯0n \geq 0 are rational: each is birational to P1Γ—P1β‰…F0\mathbb{P}^1 \times \mathbb{P}^1 \cong \mathbb{F}_0.

  • A smooth cubic surface SβŠ‚P3S \subset \mathbb{P}^3 is rational. Projecting from a line β„“βŠ‚S\ell \subset S gives a birational map Sβ‡’P2S \dashrightarrow \mathbb{P}^2. Since SS contains exactly 2727 lines, there are many such projections.

  • A smooth quartic surface S4βŠ‚P3S_4 \subset \mathbb{P}^3 is a K3 surface and is not rational (it has ΞΊ=0\kappa = 0, while rational surfaces have ΞΊ=βˆ’βˆž\kappa = -\infty).


Castelnuovo's contraction theorem

Definition5.14Exceptional curve ((-1)-curve)

An exceptional curve of the first kind (or (βˆ’1)(-1)-curve) on a smooth projective surface XX is an irreducible curve EβŠ‚XE \subset X with:

Eβ‰…P1,E2=βˆ’1.E \cong \mathbb{P}^1, \quad E^2 = -1.

By the adjunction formula, this is equivalent to E2=βˆ’1E^2 = -1 and KXβ‹…E=βˆ’1K_X \cdot E = -1 (since 2g(E)βˆ’2=E2+KXβ‹…E2g(E) - 2 = E^2 + K_X \cdot E gives βˆ’2=βˆ’1+KXβ‹…E-2 = -1 + K_X \cdot E).

TheoremCastelnuovo's contraction theorem

Let EE be a (βˆ’1)(-1)-curve on a smooth projective surface XX. Then there exists a smooth projective surface YY and a morphism Ο€:Xβ†’Y\pi: X \to Y such that:

  • Ο€\pi is the blowup of YY at a point P∈YP \in Y, and E=Ο€βˆ’1(P)E = \pi^{-1}(P) is the exceptional divisor.
  • Ο€βˆ£Xβˆ–E:Xβˆ–Eβ†’βˆΌYβˆ–{P}\pi|_{X \setminus E}: X \setminus E \xrightarrow{\sim} Y \setminus \{P\} is an isomorphism.
  • KX=Ο€βˆ—KY+EK_X = \pi^* K_Y + E and ρ(X)=ρ(Y)+1\rho(X) = \rho(Y) + 1.

In other words, every (βˆ’1)(-1)-curve can be blown down (contracted) to a smooth point. This is the inverse operation to blowing up.

ExampleBlowing down on Bl_P(β„™Β²)

Let X=Bl⁑P(P2)X = \operatorname{Bl}_P(\mathbb{P}^2) with exceptional divisor EE and proper transform HH of a line through PP. Then Pic⁑(X)=ZHβŠ•ZE\operatorname{Pic}(X) = \mathbb{Z}H \oplus \mathbb{Z}E with H2=1H^2 = 1, E2=βˆ’1E^2 = -1, Hβ‹…E=0H \cdot E = 0.

The curve EE is a (βˆ’1)(-1)-curve, and blowing it down recovers P2\mathbb{P}^2. But Hβˆ’EH - E is also a (βˆ’1)(-1)-curve (it is the proper transform of a line not through PP... actually, HH itself satisfies H2=1H^2 = 1, so HH is not a (βˆ’1)(-1)-curve). The curve L=Hβˆ’EL = H - E satisfies L2=1βˆ’1=0L^2 = 1 - 1 = 0, so it is not a (βˆ’1)(-1)-curve either.

On Bl⁑P,Q(P2)\operatorname{Bl}_{P,Q}(\mathbb{P}^2) (blowup at two points), with basis H,E1,E2H, E_1, E_2: the proper transform of the line through PP and QQ is L=Hβˆ’E1βˆ’E2L = H - E_1 - E_2, with L2=1βˆ’1βˆ’1=βˆ’1L^2 = 1 - 1 - 1 = -1 and Lβ‰…P1L \cong \mathbb{P}^1. So LL is a (βˆ’1)(-1)-curve that can be blown down, yielding F1\mathbb{F}_1.

Example(-1)-curves on del Pezzo surfaces

A del Pezzo surface SdS_d of degree dd is a blowup of P2\mathbb{P}^2 at r=9βˆ’dr = 9 - d points in general position (for 1≀d≀91 \leq d \leq 9). The number of (βˆ’1)(-1)-curves is:

  • d=9d = 9 (r=0r = 0, i.e., P2\mathbb{P}^2): 00 exceptional curves.
  • d=8d = 8 (r=1r = 1): 11 curve (E1E_1).
  • d=7d = 7 (r=2r = 2): 33 curves (E1,E2,L12E_1, E_2, L_{12}).
  • d=6d = 6 (r=3r = 3): 66 curves.
  • d=5d = 5 (r=4r = 4): 1010 curves.
  • d=4d = 4 (r=5r = 5): 1616 curves.
  • d=3d = 3 (r=6r = 6): 2727 curves --- these are the famous 27 lines on a cubic surface.
  • d=2d = 2 (r=7r = 7): 5656 curves.
  • d=1d = 1 (r=8r = 8): 240240 curves.

The count for degree d≀6d \leq 6 is (r1)+(r2)+…\binom{r}{1} + \binom{r}{2} + \ldots (exceptional divisors, proper transforms of lines, conics, etc.), reflecting the ErE_r root system structure.


Minimal models

Definition5.15Minimal surface

A smooth projective surface XX is minimal if it contains no (βˆ’1)(-1)-curves. Equivalently, every birational morphism Xβ†’YX \to Y to another smooth projective surface is an isomorphism.

The process of reaching a minimal model: start with any surface XX, find a (βˆ’1)(-1)-curve, blow it down. Repeat. Since each blowdown decreases ρ(X)\rho(X) by 11 and ρβ‰₯1\rho \geq 1, this process terminates in finitely many steps at a minimal surface Xmin⁑X_{\min}.

TheoremUniqueness of minimal models

Let XX be a smooth projective surface over an algebraically closed field kk.

Case ΞΊ(X)β‰₯0\kappa(X) \geq 0: The minimal model of XX is unique up to isomorphism. That is, if X1X_1 and X2X_2 are both minimal surfaces birational to XX, then X1β‰…X2X_1 \cong X_2.

Case ΞΊ(X)=βˆ’βˆž\kappa(X) = -\infty: The minimal model is not unique. Both P2\mathbb{P}^2 and Fn\mathbb{F}_n (for nβ‰ 1n \neq 1) are minimal and rational, but P2β‰…ΜΈFn\mathbb{P}^2 \not\cong \mathbb{F}_n. However, the possibilities are completely classified: a minimal rational surface is either P2\mathbb{P}^2 or a Hirzebruch surface Fn\mathbb{F}_n with nβ‰ 1n \neq 1.

ExampleThe minimal model process in action

Start with X=Bl⁑P1,…,P6(P2)X = \operatorname{Bl}_{P_1, \ldots, P_6}(\mathbb{P}^2), the blowup of P2\mathbb{P}^2 at 66 general points (a cubic surface S3βŠ‚P3S_3 \subset \mathbb{P}^3).

This surface has 2727 exceptional curves ((βˆ’1)(-1)-curves). Pick any one, say E1E_1, and blow it down to get Bl⁑P2,…,P6(P2)\operatorname{Bl}_{P_2, \ldots, P_6}(\mathbb{P}^2), a del Pezzo of degree 44 with 1616 exceptional curves. Continue:

  • Blow down E2E_2: get Bl⁑P3,…,P6(P2)\operatorname{Bl}_{P_3, \ldots, P_6}(\mathbb{P}^2) (1010 exceptional curves).
  • Blow down E3E_3: get Bl⁑P4,P5,P6(P2)\operatorname{Bl}_{P_4, P_5, P_6}(\mathbb{P}^2) (66 exceptional curves).
  • Blow down E4,E5,E6E_4, E_5, E_6: recover P2\mathbb{P}^2 (no exceptional curves, minimal).

Alternatively, at any stage we could blow down different (βˆ’1)(-1)-curves, potentially arriving at Fn\mathbb{F}_n instead of P2\mathbb{P}^2. This illustrates the non-uniqueness for ΞΊ=βˆ’βˆž\kappa = -\infty.

ExampleK3 surfaces are already minimal

A K3 surface XX has KX∼0K_X \sim 0 (trivial canonical class). If EE were a (βˆ’1)(-1)-curve, then KXβ‹…E=0K_X \cdot E = 0 but also KXβ‹…E=βˆ’1K_X \cdot E = -1 by adjunction, a contradiction. Therefore K3 surfaces contain no (βˆ’1)(-1)-curves and are automatically minimal.

The same argument applies to abelian surfaces (KX∼0K_X \sim 0) and Enriques surfaces (2KX∼02K_X \sim 0, so KXβ‹…E=0K_X \cdot E = 0 for any curve EE since 2KXβ‹…E=02K_X \cdot E = 0).


The minimal model program for surfaces

RemarkThe MMP algorithm for surfaces

The minimal model program (MMP) for surfaces proceeds as follows:

  1. Start with a smooth projective surface XX.
  2. If KXK_X is nef (i.e., KXβ‹…Cβ‰₯0K_X \cdot C \geq 0 for all curves CC), then XX is a minimal model. Stop.
  3. If KXK_X is not nef, there exists a curve CC with KXβ‹…C<0K_X \cdot C < 0. By the Cone Theorem, we can choose CC to be a (βˆ’1)(-1)-curve (i.e., Cβ‰…P1C \cong \mathbb{P}^1, C2=βˆ’1C^2 = -1).
  4. Contract CC via Castelnuovo's theorem to get a smooth surface Xβ€²X'. Replace XX with Xβ€²X' and go to step 2.

This terminates because ρ(X)\rho(X) decreases at each step. The result is either a minimal model (if ΞΊβ‰₯0\kappa \geq 0) or a surface with a P1\mathbb{P}^1-fibration or isomorphic to P2\mathbb{P}^2 (if ΞΊ=βˆ’βˆž\kappa = -\infty).

Definition5.16Kodaira dimension

The Kodaira dimension ΞΊ(X)\kappa(X) of a smooth projective surface XX is defined as:

ΞΊ(X)=lim sup⁑mβ†’βˆžlog⁑h0(X,mKX)log⁑m,\kappa(X) = \limsup_{m \to \infty} \frac{\log h^0(X, mK_X)}{\log m},

with the convention ΞΊ=βˆ’βˆž\kappa = -\infty if h0(mKX)=0h^0(mK_X) = 0 for all mβ‰₯1m \geq 1. For surfaces, κ∈{βˆ’βˆž,0,1,2}\kappa \in \{-\infty, 0, 1, 2\}:

  • ΞΊ=βˆ’βˆž\kappa = -\infty: Pm=0P_m = 0 for all mβ‰₯1m \geq 1 (no pluricanonical sections).
  • ΞΊ=0\kappa = 0: Pm∈{0,1}P_m \in \{0, 1\} for all mm, and Pm=1P_m = 1 for some mm.
  • ΞΊ=1\kappa = 1: PmP_m grows linearly (Pm∼cmP_m \sim cm for large mm).
  • ΞΊ=2\kappa = 2: PmP_m grows quadratically (Pm∼cm2P_m \sim cm^2 for large mm). These are surfaces of general type.

The Enriques--Kodaira classification

Kodaira dimension ΞΊ=βˆ’βˆž\kappa = -\infty: Ruled and rational surfaces

Definition5.17Ruled surface

A smooth projective surface XX is ruled if it admits a surjective morphism π:X→C\pi: X \to C to a smooth curve CC whose fibers are isomorphic to P1\mathbb{P}^1. Equivalently, X≅P(E)X \cong \mathbb{P}(\mathcal{E}) for some rank-22 vector bundle E\mathcal{E} on CC.

XX is rational if Cβ‰…P1C \cong \mathbb{P}^1 (so XX is birational to P2\mathbb{P}^2). Otherwise, XX is an irrational ruled surface over a curve CC of genus g(C)β‰₯1g(C) \geq 1.

ExampleMinimal surfaces with κ = -∞

The minimal surfaces with ΞΊ=βˆ’βˆž\kappa = -\infty are completely classified:

Rational surfaces (q=0q = 0):

  • P2\mathbb{P}^2: the projective plane, KP2=βˆ’3HK_{\mathbb{P}^2} = -3H, K2=9K^2 = 9.
  • Fn=P(OP1βŠ•OP1(βˆ’n))\mathbb{F}_n = \mathbb{P}(\mathcal{O}_{\mathbb{P}^1} \oplus \mathcal{O}_{\mathbb{P}^1}(-n)) for nβ‰ 1n \neq 1: Hirzebruch surfaces. Note F0=P1Γ—P1\mathbb{F}_0 = \mathbb{P}^1 \times \mathbb{P}^1 and F1=Bl⁑P(P2)\mathbb{F}_1 = \operatorname{Bl}_P(\mathbb{P}^2) (not minimal since it contains a (βˆ’1)(-1)-curve).

Irrational ruled surfaces (q=g(C)β‰₯1q = g(C) \geq 1):

  • P(E)β†’C\mathbb{P}(\mathcal{E}) \to C for an indecomposable rank-22 bundle E\mathcal{E} on a curve CC of genus gβ‰₯1g \geq 1. These satisfy Pm=0P_m = 0 for all mm and q=g(C)β‰₯1q = g(C) \geq 1.

Castelnuovo's rationality criterion: A surface is rational if and only if q=h1(OX)=0q = h^1(\mathcal{O}_X) = 0 and P2=h0(2KX)=0P_2 = h^0(2K_X) = 0.

Kodaira dimension ΞΊ=0\kappa = 0

ExampleMinimal surfaces with ΞΊ = 0

There are exactly four types of minimal surfaces with ΞΊ=0\kappa = 0:

K3 surfaces: KX∼0K_X \sim 0, q=0q = 0, pg=1p_g = 1, Ο‡(OX)=2\chi(\mathcal{O}_X) = 2. Examples: smooth quartic in P3\mathbb{P}^3, complete intersection of a quadric and cubic in P4\mathbb{P}^4, double cover of P2\mathbb{P}^2 branched along a smooth sextic. The lattice H2(X,Z)H^2(X, \mathbb{Z}) is isomorphic to U3βŠ•E8(βˆ’1)2U^3 \oplus E_8(-1)^2 (rank 2222, signature (3,19)(3,19)).

Abelian surfaces: KX∼0K_X \sim 0, q=2q = 2, pg=1p_g = 1, Ο‡(OX)=0\chi(\mathcal{O}_X) = 0. These are 22-dimensional abelian varieties, e.g., E1Γ—E2E_1 \times E_2 for elliptic curves EiE_i, or the Jacobian Jac⁑(C)\operatorname{Jac}(C) of a genus-22 curve.

Enriques surfaces: KX≁0K_X \not\sim 0 but 2KX∼02K_X \sim 0, q=0q = 0, pg=0p_g = 0, Ο‡(OX)=1\chi(\mathcal{O}_X) = 1. An Enriques surface is the quotient of a K3 surface by a fixed-point-free involution. Over C\mathbb{C}, the fundamental group is Ο€1(X)β‰…Z/2\pi_1(X) \cong \mathbb{Z}/2.

Bielliptic (hyperelliptic) surfaces: KX≁0K_X \not\sim 0 but nKX∼0nK_X \sim 0 for some n∈{2,3,4,6}n \in \{2, 3, 4, 6\}, q=1q = 1, pg=0p_g = 0. These are quotients (E1Γ—E2)/G(E_1 \times E_2)/G where GG is a finite group acting by translations on E1E_1 and faithfully on E2E_2.

Kodaira dimension ΞΊ=1\kappa = 1

ExampleMinimal surfaces with ΞΊ = 1: elliptic surfaces

A minimal surface with κ=1\kappa = 1 admits a unique elliptic fibration π:X→C\pi: X \to C over a smooth curve CC, whose general fiber is an elliptic curve.

Invariants: KX2=0K_X^2 = 0 (since KXK_X is nef and ΞΊ=1\kappa = 1, Kodaira's formula gives KX∼QK_X \sim_\mathbb{Q} a sum along fibers), with the canonical bundle formula KXβˆΌΟ€βˆ—(KCβŠ—L)+βˆ‘iaiFiK_X \sim \pi^*(K_C \otimes L) + \sum_i a_i F_i involving the discriminant divisor and singular fibers.

Kodaira's classification of singular fibers: the singular fibers can be of type InI_n (cycle of nn rational curves), IIII (cuspidal rational curve), IIIIII (two tangent rational curves), IVIV (three concurrent rational curves), Inβˆ—I_n^*, IIβˆ—II^*, IIIβˆ—III^*, IVβˆ—IV^* (the starred types correspond to extended Dynkin diagrams D~n+4\tilde{D}_{n+4}, E~8\tilde{E}_8, E~7\tilde{E}_7, E~6\tilde{E}_6).

Concrete example: the Fermat pencil x3+y3+z3=tβ‹…xyzx^3 + y^3 + z^3 = t \cdot xyz in P2Γ—P1\mathbb{P}^2 \times \mathbb{P}^1. The general fiber is a smooth cubic (elliptic), with singular fibers at t3=27t^3 = 27 (nodal cubics, type I1I_1) and t=∞t = \infty (the triangle xyz=0xyz = 0, type I3I_3). After resolving base points, this gives an elliptic surface with ΞΊ=1\kappa = 1.

Kodaira dimension ΞΊ=2\kappa = 2: Surfaces of general type

ExampleMinimal surfaces with ΞΊ = 2: surfaces of general type

A surface of general type is a surface with ΞΊ=2\kappa = 2. For a minimal surface of general type, KXK_X is nef and big (KX2>0K_X^2 > 0).

Key inequalities for minimal surfaces of general type:

  • Noether's inequality: KX2β‰₯2pgβˆ’4K_X^2 \geq 2p_g - 4 (equivalently, Ο‡(OX)≀12KX2+2\chi(\mathcal{O}_X) \leq \frac{1}{2}K_X^2 + 2).
  • Bogomolov--Miyaoka--Yau inequality: KX2≀9Ο‡(OX)=9(1βˆ’q+pg)K_X^2 \leq 9\chi(\mathcal{O}_X) = 9(1 - q + p_g).
  • Equality KX2=9Ο‡(OX)K_X^2 = 9\chi(\mathcal{O}_X) holds iff XX is a ball quotient B2/Ξ“\mathbb{B}^2/\Gamma (Yau).

Examples:

  • Smooth quintic in P3\mathbb{P}^3: KX=OX(1)K_X = \mathcal{O}_X(1), KX2=5K_X^2 = 5, pg=4p_g = 4, q=0q = 0, Ο‡=5\chi = 5.
  • Product C1Γ—C2C_1 \times C_2 of curves with g(Ci)β‰₯2g(C_i) \geq 2: K2=8(g1βˆ’1)(g2βˆ’1)K^2 = 8(g_1 - 1)(g_2 - 1), Ο‡=(g1βˆ’1)(g2βˆ’1)\chi = (g_1 - 1)(g_2 - 1).
  • Godeaux surface: K2=1K^2 = 1, pg=0p_g = 0, the simplest surface of general type. Constructed as the quotient of the Fermat quintic x05+x15+x25+x35=0x_0^5 + x_1^5 + x_2^5 + x_3^5 = 0 by the Z/5\mathbb{Z}/5-action xi↦΢ixix_i \mapsto \zeta^i x_i (ΞΆ=e2Ο€i/5\zeta = e^{2\pi i/5}).
  • Barlow surface: a simply connected surface of general type with pg=0p_g = 0, K2=1K^2 = 1.
  • Beauville surface: a surface isogenous to a higher product C1Γ—C2/GC_1 \times C_2 / G.

The Cremona group

Definition5.18Cremona group

The Cremona group Cr⁑2(k)=Bir⁑(Pk2)\operatorname{Cr}_2(k) = \operatorname{Bir}(\mathbb{P}^2_k) is the group of birational automorphisms of P2\mathbb{P}^2. Since P2\mathbb{P}^2 is rational, this equals Aut⁑k(k(x,y))\operatorname{Aut}_k(k(x, y)), the group of kk-automorphisms of the purely transcendental field extension k(x,y)/kk(x, y)/k.

The Cremona group contains:

  • PGL⁑3(k)=Aut⁑(P2)\operatorname{PGL}_3(k) = \operatorname{Aut}(\mathbb{P}^2) as a subgroup (the linear automorphisms).
  • The standard quadratic transformation Οƒ:[x:y:z]↦[yz:xz:xy]\sigma: [x:y:z] \mapsto [yz:xz:xy].
  • De Jonquieres transformations (birational maps preserving a pencil of lines through a point).
TheoremNoether–Castelnuovo theorem

Every birational automorphism of P2\mathbb{P}^2 (over an algebraically closed field) is a composition of linear automorphisms and the standard quadratic transformation. In other words:

Cr⁑2(k)=⟨PGL⁑3(k),ΟƒβŸ©.\operatorname{Cr}_2(k) = \langle \operatorname{PGL}_3(k), \sigma \rangle.

This means the Cremona group is generated by PGL⁑3(k)\operatorname{PGL}_3(k) and a single quadratic involution.

ExampleElements of the Cremona group
  • Linear maps: [x:y:z]↦[ax+by+cz:dx+ey+fz:gx+hy+iz][x:y:z] \mapsto [ax + by + cz : dx + ey + fz : gx + hy + iz], degree 11.

  • Standard Cremona Οƒ\sigma: degree 22. Resolves to 33 blowups followed by 33 blowdowns.

  • Degree-nn Cremona maps: for example, [x:y:z]↦[xn:yn:zn][x:y:z] \mapsto [x^n : y^n : z^n] is not birational (it has degree n2n^2). A degree-nn birational map is [x:y:z]↦[F0:F1:F2][x:y:z] \mapsto [F_0 : F_1 : F_2] with deg⁑Fi=n\deg F_i = n and a rational inverse.

  • De Jonquieres transformation of degree nn: preserves the pencil of lines through [1:0:0][1:0:0]. In affine coordinates (x,y)(x, y): (x,y)↦(x,a(x)y+b(x)c(x)y+d(x))(x, y) \mapsto (x, \frac{a(x)y + b(x)}{c(x)y + d(x)}) for polynomials a,b,c,da, b, c, d with adβˆ’bcβ‰ 0ad - bc \neq 0. These form a subgroup dJβ‘βŠ‚Cr⁑2\operatorname{dJ} \subset \operatorname{Cr}_2.

  • The Cremona group is not simple: Cantat and Lamy (2013) proved that Cr⁑2(C)\operatorname{Cr}_2(\mathbb{C}) is not simple, using the action on an infinite-dimensional hyperbolic space.


Factorization of birational maps

TheoremFactorization theorem for surfaces

Let ϕ:X⇒Y\phi: X \dashrightarrow Y be a birational map between smooth projective surfaces. Then ϕ\phi can be factored as a sequence of blowups followed by a sequence of blowdowns:

X=X0←X1←⋯←Xnβ†’β‹―β†’Xn+m=Y,X = X_0 \leftarrow X_1 \leftarrow \cdots \leftarrow X_n \rightarrow \cdots \rightarrow X_{n+m} = Y,

where each left arrow Xi+1β†’XiX_{i+1} \to X_i is a blowup at a point, and each right arrow Xjβ†’Xj+1X_j \to X_{j+1} is a blowdown of a (βˆ’1)(-1)-curve.

Equivalently, there exists a smooth surface ZZ and morphisms p:Zβ†’Xp: Z \to X, q:Zβ†’Yq: Z \to Y (both compositions of blowups) such that Ο•=q∘pβˆ’1\phi = q \circ p^{-1}.

ExampleFactoring the standard Cremona transformation

The standard Cremona Οƒ:P2β‡’P2\sigma: \mathbb{P}^2 \dashrightarrow \mathbb{P}^2 factors as:

  1. Three blowups: Blow up P0=[1:0:0]P_0 = [1:0:0], P1=[0:1:0]P_1 = [0:1:0], P2=[0:0:1]P_2 = [0:0:1] to obtain Z=Bl⁑P0,P1,P2(P2)Z = \operatorname{Bl}_{P_0, P_1, P_2}(\mathbb{P}^2). This introduces exceptional divisors E0,E1,E2E_0, E_1, E_2 with Ei2=βˆ’1E_i^2 = -1, and the proper transforms β„“01,β„“02,β„“12\ell_{01}, \ell_{02}, \ell_{12} of the coordinate lines now satisfy β„“ij2=βˆ’1\ell_{ij}^2 = -1.

  2. Three blowdowns: Contract β„“01,β„“02,β„“12\ell_{01}, \ell_{02}, \ell_{12} (the proper transforms) to get P2\mathbb{P}^2 again, where E0,E1,E2E_0, E_1, E_2 become the new coordinate lines.

The intermediate surface ZZ is the del Pezzo surface S6S_6 (degree 66, isomorphic to P2\mathbb{P}^2 blown up at 33 non-collinear points), and it admits two different contractions to P2\mathbb{P}^2.

ExampleBirational map between ruled surfaces

Consider F0=P1Γ—P1\mathbb{F}_0 = \mathbb{P}^1 \times \mathbb{P}^1 and F2\mathbb{F}_2. These are birational (both rational) but not isomorphic. A birational map F0β‡’F2\mathbb{F}_0 \dashrightarrow \mathbb{F}_2 factors through one blowup and one blowdown:

  1. Blow up a point P∈F0P \in \mathbb{F}_0: get Bl⁑P(F0)β‰…Bl⁑Q1,Q2(P2)\operatorname{Bl}_P(\mathbb{F}_0) \cong \operatorname{Bl}_{Q_1, Q_2}(\mathbb{P}^2) (blowup of P2\mathbb{P}^2 at two points), which is F1\mathbb{F}_1 with an extra blowup.
  2. The proper transform of the fiber through PP becomes a (βˆ’1)(-1)-curve. Blowing it down yields F2\mathbb{F}_2.

More generally, the elementary transformation elm⁑P:Fnβ‡’FnΒ±1\operatorname{elm}_P: \mathbb{F}_n \dashrightarrow \mathbb{F}_{n \pm 1} at a point PP is a blowup followed by blowdown of the proper transform of the fiber.


Comparison with higher-dimensional MMP

RemarkWhy surfaces are special

The minimal model program for surfaces is remarkably clean compared to higher dimensions:

What works for surfaces:

  • Every birational map factors into blowups and blowdowns of smooth points.
  • Contracting a (βˆ’1)(-1)-curve always produces a smooth surface.
  • Minimal models exist and are unique (for ΞΊβ‰₯0\kappa \geq 0).
  • The classification is complete (Enriques--Kodaira).

What changes in dimension β‰₯3\geq 3:

  • Flips are needed: contracting a KXK_X-negative curve in dimension 33 can create singularities worse than a smooth point. One must perform a flip (a surgery replacing a curve with another) to continue.
  • Terminal singularities: the MMP must work with mildly singular varieties (terminal or canonical singularities), not just smooth ones.
  • Flops: birational maps between minimal models (in ΞΊ=0\kappa = 0) can involve flops --- birational maps that are isomorphisms in codimension 11 but change the variety along codimension-22 loci.
  • Termination: proving that the sequence of flips terminates is a deep theorem in dimension 33 (Mori, 1988) and was established in full generality by Birkar--Cascini--Hacon--McKernan (BCHM, 2010) for varieties of general type.
  • Non-uniqueness of minimal models: in dimension β‰₯3\geq 3, minimal models with ΞΊ=0\kappa = 0 are related by flops but need not be isomorphic.
ExampleA flip in dimension 3

Consider a smooth 3-fold XX containing a curve Cβ‰…P1C \cong \mathbb{P}^1 with normal bundle NC/Xβ‰…O(βˆ’1)βŠ•O(βˆ’1)N_{C/X} \cong \mathcal{O}(-1) \oplus \mathcal{O}(-1). Then KXβ‹…C=βˆ’1K_X \cdot C = -1 (by adjunction on the 3-fold). The contraction Xβ†’YX \to Y that contracts CC to a point creates an ordinary double point (node) singularity.

The flip Xβ‡’X+X \dashrightarrow X^+ replaces Cβ‰…P1C \cong \mathbb{P}^1 with a different P1\mathbb{P}^1 having normal bundle O(βˆ’1)βŠ•O(βˆ’1)\mathcal{O}(-1) \oplus \mathcal{O}(-1) but with KX+β‹…C+=+1K_{X^+} \cdot C^+ = +1. This phenomenon has no analogue for surfaces.

The classical example: the Atiyah flop on the conifold xy=zwxy = zw in C4\mathbb{C}^4. The two small resolutions (blowing up the ideal (x,z)(x, z) vs. (x,w)(x, w)) are related by a flop.


Birational invariants

Definition5.19Birational invariants of surfaces

The following are birational invariants of smooth projective surfaces (unchanged under birational equivalence):

  • The Kodaira dimension ΞΊ(X)\kappa(X).
  • The plurigenera Pm=h0(X,mKX)P_m = h^0(X, mK_X) for all mβ‰₯1m \geq 1.
  • The irregularity q=h1(X,OX)=h0(X,Ξ©X1)q = h^1(X, \mathcal{O}_X) = h^0(X, \Omega^1_X).
  • The fundamental group Ο€1(X)\pi_1(X) (in characteristic 00).

The following are not birational invariants:

  • The Picard number ρ(X)\rho(X) (changes under blowup: ρ(Bl⁑PX)=ρ(X)+1\rho(\operatorname{Bl}_P X) = \rho(X) + 1).
  • The self-intersection KX2K_X^2 (changes: KBl⁑PX2=KX2βˆ’1K_{\operatorname{Bl}_P X}^2 = K_X^2 - 1).
  • The Euler characteristic e(X)e(X) (changes: e(Bl⁑PX)=e(X)+1e(\operatorname{Bl}_P X) = e(X) + 1).
  • But KX2+ρ(X)K_X^2 + \rho(X) is a birational invariant (since both change by Β±1\pm 1).
ExampleComputing birational invariants

Rational surfaces (P2\mathbb{P}^2, Fn\mathbb{F}_n, del Pezzo surfaces): ΞΊ=βˆ’βˆž\kappa = -\infty, q=0q = 0, Pm=0P_m = 0 for all mm, Ο€1=0\pi_1 = 0. But K2K^2 varies: KP22=9K^2_{\mathbb{P}^2} = 9, KFn2=8K^2_{\mathbb{F}_n} = 8, KSd2=dK^2_{S_d} = d for del Pezzo of degree dd.

K3 surfaces: ΞΊ=0\kappa = 0, q=0q = 0, P1=1P_1 = 1, Pm=1P_m = 1 for all mβ‰₯1m \geq 1. All K3 surfaces are diffeomorphic (over C\mathbb{C}) and have Ο€1=0\pi_1 = 0.

Abelian surfaces: ΞΊ=0\kappa = 0, q=2q = 2, P1=1P_1 = 1, Pm=1P_m = 1 for all mm. The fundamental group Ο€1β‰…Z4\pi_1 \cong \mathbb{Z}^4 distinguishes them from K3 surfaces.

Quintic surface S5βŠ‚P3S_5 \subset \mathbb{P}^3 (general type): ΞΊ=2\kappa = 2, q=0q = 0, K2=5K^2 = 5, pg=P1=4p_g = P_1 = 4, Ο‡(O)=5\chi(\mathcal{O}) = 5, Ο€1=0\pi_1 = 0.


Summary

RemarkThe structure of birational geometry for surfaces

The birational geometry of surfaces is governed by three pillars:

  • Castelnuovo contraction: every (βˆ’1)(-1)-curve can be blown down, reducing the surface toward a minimal model. For ΞΊβ‰₯0\kappa \geq 0, this minimal model is unique.
  • Enriques--Kodaira classification: minimal surfaces fall into four classes by Kodaira dimension: ruled (ΞΊ=βˆ’βˆž\kappa = -\infty), K3/abelian/Enriques/bielliptic (ΞΊ=0\kappa = 0), elliptic fibrations (ΞΊ=1\kappa = 1), and general type (ΞΊ=2\kappa = 2).
  • Factorization: every birational map between surfaces is a composition of blowups and blowdowns, and the Cremona group Cr⁑2\operatorname{Cr}_2 is generated by PGL⁑3\operatorname{PGL}_3 and a single quadratic involution.

This complete picture --- achieved through the work of the Italian school, Kodaira, Bombieri, Mumford, and many others --- serves as both the template and the motivation for the higher-dimensional minimal model program.