Theorem. (βalgebraic versionβ; see [107]) Let π : π β π be a projective morphism of Noetherian schemes (seeβSchemesβ). Then there exists a scheme π, a finite morphism π : π β π and a projective morphism with connected fibres β : π β π such that
π = π β β .
Theorem. (βanalytic versionβ); (see [241]). Let π : π β π be a proper surjective mor-phism of reduced complex analytic spaces (seeβSpaces, analytic -β). Then there exists a reduced complex analytic space π, a surjective morphism π : π β π such that the fibres consist of a finite number of points and a surjective morphism with connected fibres β : π β π such that
π = π β β .
Subcanonical.
We say that a smooth subvariety π of a smooth algebraic variety π is subcanonical if there exists a line bundle πΏ on π such that πΏ|π = πΎπ, where πΎπis the canonical bundle of π (seeβCanonical bundle, canonical sheafβ).Surfaces, algebraic -.
([18], [22], [25], [65], [93], [107], [253]). In the sequel the word surface will denote a smooth projective algebraic variety of dimension 2 over β.We will denote the projective space βπβby βπ.
We start by stating some basic theorems and notation.
Let π be a surface; then π»4(π, β€) and π»4(π, β€) are isomorphic to β€, since π is oriented (as it is a complex manifold), compact and of real dimension 4 (seeβSingular homol-ogy and cohomolhomol-ogyβ). We denote by [π] the fundamental class of π, that is, the gener-ator of π»4(π, β€) giving the orientation of π. If πΏ and πΏσΈ are two holomorphic line bundles on π, we denote by πΏ β πΏσΈ the number obtained by evaluating π1(πΏ) βͺ π1(πΏσΈ ) β π»4(π, β€) in [π] (where βͺ is the cup product; see againβSingular homology and cohomologyβ).
For any divisors π· and π·σΈ on π, let π· β π·σΈ their intersection number (seeβIntersection of cyclesβ).
Observe that, for any holomorphic line bundle πΏ on π, we have that π»π(O(πΏ)) = 0 for π β₯ 3; to prove this, we can apply the Abstract de Rhamβs theorem (seeβSheavesβ) to the exact sequence
0 σ³¨βO(πΏ) σ³¨β πΆβ(πΏ)σ³¨β π΄π 0,1(πΏ)σ³¨β π΄π 0,2(πΏ) σ³¨β 0,
where πΆβ(πΏ) is the sheaf of the πΆβsections of πΏ, π΄π,π(πΏ) is the sheaf of the πΆβ(π, π)-forms with values in πΏ and the mapO(πΏ) σ³¨β πΆβ(πΏ) is given by inclusion.
So, if we denote by π(O(πΏ)) the Euler characteristic of O(πΏ), i.e., the number
βπ=0,...,β(β1)πβπ(O(πΏ)), we have that, in this case (i.e., in the case of surfaces), π(O(πΏ)) = β0(O(πΏ)) β β1(O(πΏ)) + β2(O(πΏ)),
Surfaces, algebraic - | 181 and so π(Oπ) = β0(Oπ) β β1(Oπ) + β2(Oπ). Furthermore, we recall that π(π) denotes the Euler-PoincarΓ© characteristic of π, i.e., βπ=0,...,β(β1)πππ(π), where ππ(π) is the Betti number πππ π»π(π, β); in this case,
π(π) = π0(π) β π1(π) + π2(π) β π3(π) + π4(π) = 2π0(π) β 2π1(π) + π2(π) by PoincarΓ© duality (seeβSingular homology and cohomologyβ).
Index theorem. Let π be a surface. The intersection form is negative definite on a sub-space of codimension 1 in π»1,1(π, β), precisely is negative definite on the subspace of the primitive forms (seeβLefschetz decomposition and Hard Lefschetz theoremβfor the definition of primitive).
RiemannβRoch theorem for surfaces. Let πΏ be a holomorphic line bundle on a sur-face π. We have:
π(O(πΏ)) = π(Oπ) +πΏ β πΏ β πΏ β πΎπ
2 ,
where πΎπ is the canonical bundle of π (see βCanonical bundle, canonical sheafβ).
Noetherβs theorem. Let π be a surface. We denote by ππ(π) the π-th Chern class of the holomorphic tangent bundle π1,0π and we denote by π1(π)2the cup product π1(π)βͺπ1(π).
Hopf theorem (seeβGaussβBonnetβHopf theoremβ)π2(π1,0π) = π(π).
CastelnuovoβEnriques criterion. Let π be a surface and πΆ a smooth rational curve in π with πΆ β πΆ = β1. Then there exists a surface πσΈ and a map π : π β πσΈ such that π is the blow-up of πσΈ in a point π and πβ1(π) = πΆ.
Structure of birational maps on surfaces. Let π1and π2be two surfaces and π : π1β π2be a birational map. Then there exists another surface Μπ and two morphisms, π1 and π2, both given by a sequence of blowing-up maps, such that the following diagram commutes:
182 | Surfaces, algebraic
-In other words, a birational map between surfaces is given by a sequence of blowing-ups followed by a sequence of blowing-downs.
Definition. We say that a surface π is minimal if there does not exist another surface
Μπ and a blowing-up π β Μπ. A minimal model for a surface π is a minimal surface bira-tional to π.
We recall that if π is a compact complex manifold of dimension π, we define β π(π) = β1,0(π) (irregularity);
β ππ(π) = β0(O(πΎππ )) for any π β β (plurigenera);
β ππ(π) = βπ,0(π) = π1(π) (geometric genus).
One can prove that they are birational invariants. With the notation above, we have π(Oπ) = β0(Oπ) β β1(Oπ) + β2(Oπ)
= β0,0(π) β β0,1(π) + β0,2(π) = 1 β π(π) + ππ(π) by Dolbeaultβs theorem and the Hodge theorem (seeβHodge theoryβ).
The most important tool to classify surfaces is Kodaira dimension πΎ (seeβKodaira dimension (or Kodaira number)β). For surfaces, it can be obviously only ββ, 0, 1, 2.
First, we will deal with rational and ruled surfaces.
Definition. We say that a surface is rational if it is birational to β2.
Definition. We say that a surface is ruled over a compact Riemann surface πΆ if it is birational to πΆ Γ β1.
Remark. Let π be a surface. Then π is ruled on β1β π is rational (in fact it is birational to β1Γ β1if and only if it is birational to β2).
Definition. We say that a surface π is geometrically ruled over a compact Riemann surface πΆ if there exists a smooth morphism π β πΆ such that the fibres are isomorphic to β1.
(Please note that in some works the term βruledβ means βgeometrically ruledβ.) NoetherβEnriques theorem. A geometrically ruled surface is equal to the projectivized β(πΈ) of a vector bundle πΈ of rank 2.
Proposition 1. Let π be a ruled surface on a compact Riemann surface πΆ. Let π(πΆ) be the genus of πΆ. Then
π(π) = π(πΆ), ππ(π) = 0, ππ(π) = 0 βπ β₯ 2.
Moreover, if π is geometrically ruled, then
πΎπ2= 8(βπ(πΆ) + 1), π2(π) = 2 .
Surfaces, algebraic - | 183
Definition. Let π β β. The π-th Hirzebruch surface is ππ= β(Oβ1(π) βOβ1),
whereOβ1is the trivial bundle on β1andOβ1(π) is the π-th power of the hyperplane bundle on β1; seeβHyperplane bundles, twisting sheavesβ(we are making a slight abuse of the notation:Oβ1(π) generally denotes the sheaf of the holomorphic sections of the π-th power of the hyperplane bundle, but sometimes, as here, it denotes just the π-th power of the hyperplane bundle).
Remark. The Hirzebruch surfaces are exactly the surfaces that are geometrically ruled on β1. In fact, by the GrothendieckβSegre theorem (see βGrothendieckβSegre theo-remβ) and by the NoetherβEnriques theorem, we have that a surface that is geomet-rically ruled on β1is equal to β(πΏ1β πΏ2) = β((πΏ1β πΏβ12 ) βOβ1) for some πΏ1and πΏ2 holomorphic line bundles on β1and β((πΏ1β πΏβ12 ) βOβ1) = β(Oβ1(π) βOβ1) for some π. Obviously we can suppose π β β.
We can prove that for all π ΜΈ= 0 the surface ππis the unique β1-bundle on β1with an irreducible curve πΈ with πΈ β πΈ = βπ and that the blow-up of β2in a point is π1. We can also prove that ππβ1can be obtained from ππby a up followed by a blowing-down.
Let πΉ be a fibre of β(Oβ1(π) βOP1) and π be the image of the section given by the zero section ofOβ1(π) and by the section 1 ofOP1. Let ππ,πbe the image of ππby the map π(π+ππΉ)associated to the linear system |π + ππΉ| (seeβBundles fibre -β). We can prove that it is a surface in βπ+2π+1of degree π+2π, which is the minimal achievable degree for a surface in βπ+2π+1(seeβMinimal degreeβ). The surfaces ππ,πare the rational normal scrolls of dimension 2 (seeβScrolls, rational normal -β).
Proposition 2. Every nondegenerate surface of minimal degree in βπ, i.e., every non-degenerate surface of degree π β 1 in βπ, is either a rational normal scroll or the Veronese surface in β5, i.e., the image of β2embedded in β5byOβ2(2) (seeβVeronese embeddingβ).
Another example of rational surfaces are Del Pezzo surfaces. Let π β€ 6. Let π1, . . . , ππ be π distinct points in β2in general position (i.e., no 3 of them are collinear, and no 6 of them lie on a conic). Let π : π β β2be the blowing-up of β2in π1, . . . , ππ. Then
βπΎπdefines an embedding π : π σ³¨ β β9βπ, whose image has degree 9 β π and is called a Del Pezzo surface.
If π = 6, π(π) is a smooth cubic in β3; if π = 5, π(π) is a complete intersection of two quadrics in β4.
One can show that π(π) contains only a finite number of lines, precisely the images under π of
(a) the exceptional curves;
(b) the strict transforms of the lines ππππfor π ΜΈ= π;
184 | Surfaces, algebraic
-(c) the strict transforms of the conics through 5 of the ππ. In the case π = 6 we have exactly 27 lines.
Theorem.
(i) The minimal ruled surfaces on β1are isomorphic either to the Hirzebruch surfaces or to β2.
(ii) The minimal ruled surfaces on a Riemann surface of genus β₯ 1 are the geometri-cally ruled ones and the minimal model is not unique.
Theorem. The nonruled surfaces have a unique minimal model (up to isomor-phisms).
Remark. If π is a rational surface, then, for every π β₯ 1, we have ππ(π) = π(π) = ππ(π) = 0.
The following theorem tells us that also the converse is true.
CastelnuovoβEnriques theorem. Let π be a surface such that π(π) = π2(π) = 0. Then π is rational.
Theorem. Any unirational surface (see βUnirational, LΓΌroth problemβ) is ratio-nal.
Theorem (CastelnuovoβDe Franchis).
β Let π be a minimal surface with π(π) < 0; then π is irrational ruled.
β Let π be a minimal surface with π1(π)2< 0; then π is irrational ruled.
β Let π be a minimal surface with π((Oπ)) < 0; then π is irrational ruled.
(Observe that the third statement follows trivially from the first two and Noetherβs the-orem.)
Enriquesβ theorem. A surface π is ruled if and only if π12(π) = 0.
As we have already said, the most important tool to classify surfaces is Kodaira dimen-sion.
Enriquesβ theorem and Proposition 1 tell us that πΎ(π) = ββ if and only if π is ruled.
Definition. We say that a surface π is hyperelliptic or bielliptic if it is equal to (πΈ Γ πΉ)/πΊ,
where πΈ and πΉ are elliptic curves and πΊ is a finite group of translations of πΈ acting on πΉ in such a way that πΉ/πΊ = β1.
Definition. Let πΆ be a smooth curve. An elliptic surface π with base πΆ is a surface such that there exists a surjective morphism π β πΆ such that the generic fibre is an elliptic (irreducible) curve.
Surfaces, algebraic - | 185 Classification theorem (EnriquesβKodaira).
β A minimal surface π with πΎ(π) = 0 is one of the following:
(a) a surface with π = 0 and ππ= 1 (this implies πΎπ=O); such surfaces are called K3 surfaces;
(b) a surface with π = 0 and ππ = 0 (this implies πΎπβ2 = O); such surfaces are called Enriques surfaces;
(c) a hyperelliptic surface if π = 1;
(d) an Abelian variety if π = 2 (seeβTori, complex - and Abelian varietiesβ).
β A surface π with πΎ(π) = 1 is elliptic.
A particular case of K3 surfaces are the Kummer surfaces. A Kummer surface π is a surface obtained in the following way: let π΄ be an Abelian surface and let Μπ΄ be the surface obtained from π΄ by blowing up the 16 points of order 2; define
π = Μπ΄/β¨πΌ, πβ©,
where π is the map induced on Μπ΄ by the map π σ³¨β βπ on π΄.
Proposition. The quotient of a K3 surface by a fixed-point free involution is an Enriques surface.
Conversely, let π be an Enriques surface. As we have already said, πΎπβ2is the trivial bundle. Let Μπ = {π’ β πΎπ| π(π’ β π’) = 1} where π is the isomorphism πΎπβ2 β π Γ β. The surface Μπ is a (nonramified) double covering space of π, where the covering map is induced by the projection of πΎπto π, and it is a K3 surface.
Definition. We say that a surface π is of general type if πΎ(π) = 2.
BogomolovβMiyaokaβYau inequality. For a surface π of general type, it holds that π12(π) β€ 3 π2(π),
where ππ(π) = ππ(π1,0π).
Noether inequality. For a minimal surface of general type, π, it holds that ππ(π) β€ 12π12(π) + 2,
where ππ(π) = ππ(π1,0π).
Bombieri and Mumford extended the classification of surfaces to arbitrary alge-braically closed fields (see [25]). The classification in the case where the characteristic is different from 2, 3 is analogous to the one over the complex numbers.
186 | Surfaces, algebraic
-Symmetric polynomials
([76], [164], [178], [181], [236], [238]). Let π β β. Let π β β€[π₯1, . . . , π₯π]; we say that π is symmetric if it is invariant for the action of the symmetric group π΄π(in other words, if interchanging any of the variables does not modify the polynomial).We will denote by β€[π₯1, . . . , π₯π]π΄π the set of the symmetric polynomials in π₯1, . . . , π₯π with coefficients in β€.
Let πΈπ(π₯1, . . . , π₯π) be the sum of the squarefree monomials of degree π in π₯1, . . . , π₯π (squarefree means not divisible by the square of a variable). The polynomials πΈπare symmetric and they are called elementary symmetric polynomials.
For example,
β πΈ0(π₯1, . . . , π₯π) = 1;
β πΈ1(π₯1, . . . , π₯π) = π₯1+ β β β + π₯π; β πΈ2(π₯1, . . . , π₯π) = β1β€π<πβ€ππ₯ππ₯π.
Observe that πΈπ(π₯1, . . . , π₯π) = 0 for any π β₯ π + 1.
The elementary symmetric polynomials can be defined also by the following formula:
π±π=1,...,π(1 + π₯ππ‘) = β
πββπΈπ(π₯1, . . . , π₯π)π‘π.
Gaussβ theorem. If π β β€[π₯1, . . . , π₯π]π΄π, then there exists a polynomial πΉ β β€[π₯1, . . . , π₯π] such that π = πΉ(πΈ1, . . . , πΈπ).
Let πΆπ(π₯1, . . . , π₯π) be the sum of the monomials of degree π in π₯1, . . . , π₯π. The πΆπ are called complete symmetric polynomials.
For example:
β πΆ0(π₯1, . . . , π₯π) = 1;
β πΆ1(π₯1, . . . , π₯π) = π₯1+ β β β + π₯π; β πΆ2(π₯1, . . . , π₯π) = β1β€πβ€πβ€ππ₯ππ₯π.
The complete symmetric polynomials can be defined also by the following formula:
π±π=1,...,π 1 1 β π₯ππ‘ = β
πββπΆπ(π₯1, . . . , π₯π)π‘π. Remark. The following relations hold:
β
πββ
π₯πβππ (β1)ππΈπ(π₯1, . . . , π₯π) = 0 βπ β {1, . . . , π},
β
πββ(β1)ππΈππΆπβπ= 0 βπ β₯ 1.
Surfaces, algebraic - | 187 The first follows from the second definition of πΈπtaking π‘ = β1/π₯π. The second follows from the second definitions of πΈπand πΆπ:
1 = π±π=1,...,π(1 β π₯ππ‘) π±π=1,...,π 1 1 β π₯ππ‘
= ( β
πββ(β1)ππΈπ(π₯1, . . . , π₯π)π‘π)( β
πββπΆπ(π₯1, . . . , π₯π)π‘π).
From the second relation we have β€[πΆ1, . . . , πΆπ] = β€[πΈ1, . . . , πΈπ], which, by Gaussβ
theorem, is β€[π₯1, . . . , π₯π]π΄π. Thus both the elementary symmetric polynomials and the complete symmetric polynomials generate the algebra of the symmetric poly-nomials.
Now we will define another class of symmetric polynomials such that they generate β€[π₯1, . . . , π₯π]π΄πas β€-module: the Schur polynomials.
Let π = (π1, . . . , ππ) with ππβ β and π1β₯ β β β β₯ ππ(we call it a partition of π1+ β β β + ππ);
we can associate to π a diagram, called a Young diagram, with ππboxes in the π-th row for any π β {1, . . . , π} and the rows lined up on the left; see Figure19.
Fig. 19. Young diagram of (4, 3, 1).
Let π΄ be the matrix π Γ β:
(
1 π₯1 π₯21 . . . .
. . . .
. . . .
. . . .
1 π₯π π₯2π . . . . ) .
Number the columns of π΄ beginning from 0. For distinct π‘1, . . . , π‘π β β, define ππ‘1,...,π‘π(π₯1, . . . , π₯π) to be the determinant of the matrix obtained by taking the columns π‘1, . . . , π‘πof π΄.
For π = (π1, . . . , ππ) with ππβ β and π1β₯ β β β β₯ ππ, define ππ(π₯1, . . . , π₯π) = ππ+πΏ(π₯1, . . . , π₯π)
ππΏ(π₯1, . . . , π₯π) ,
where πΏ = (π β 1, π β 2, . . . , 0). The ππare symmetric polynomials and they are called Schur polynomials.
For example: let π = 3 and consider π = (1, 1, 0), which we write (1, 1) (in general the zeroes at the end of a partition are omitted); we have
π(1,1)= π₯23π₯22(π₯3β π₯2) β π₯21π₯23(π₯3β π₯1) + π₯22π₯21(π₯2β π₯1)
βπ>π(π₯πβ π₯π) = π₯1π₯2+ π₯1π₯3+ π₯2π₯3.
188 | Syzygies
Theorem. The Schur polynomials are a basis of β€[π₯1, . . . , π₯π]π΄πas β€-module.
The following formulas express the Schur polynomials in terms of elementary sym-metric polynomials and in terms of complete symsym-metric polynomials.
JacobiβTrudiβGiambelli formulas. For any π = (π1, . . . , ππ) with ππβ β and π1β₯ β β β β₯ column of the Young diagram of π (in other words, the Young diagram of πΎ is obtained from the Young diagram of π by interchanging rows and columns).
LittlewoodβRichardson rule. For any π = (π1, . . . , ππ) with ππ β β and π1 β₯ β β β β₯ ππ and for any π = (π1, . . . , ππ) with ππβ β and π1β₯ β β β β₯ ππ, we have
ππππ= β
π ππ,π,πππ,
where ππ,π,πis the number of the ways the Young diagram of π can be expanded to the Young diagram of π by a strict π-expansion, where
β a π = (π1, . . . , ππ)-expansion of the Young diagram of π is a Young diagram ob-tained from the Young diagram of π by adding π1boxes not two in the same col-umn, then π2boxes not two in the same column, and so on;
β a π-expansion is called strict if the following condition hold: put a 1 in each of the π1boxes, a 2 in each of the π2boxes, and so on; form a list reading the numbers in the boxes, reading from right to left and beginning from the top row; we must have that for every π with 1 β€ π β€ π1+ β β β + ππ, and for every π with 1 β€ π β€ π β 1, in the first π entries of the list, the number of the πβs is greater than or equal to the number of the (π + 1)βs.
Syzygies | 189 More generally, a syzygy among some π-uples of polynomials in πΎ[π₯1, . . . , π₯π],
π1= ( π11
...
π1π
) , . . . , ππ = ( ππ1
...
πππ ) ,
is a π-uple of polynomials (π1, . . . , ππ) with π1, . . . , ππβ πΎ[π₯1, . . . , π₯π] such that
β
π=1,...,πππππ= 0.
Given a projective algebraic variety, one can study the syzygies among generators of the ideal of the variety and then the syzygies among these syzygies, and so on. In particular, one can study the degree of such syzygies. A definition which is often used is the following one; it is due to Green and Lazarsfeld (see [88], [89], [170]):
Let π be a smooth complex projective algebraic variety of dimension π and let πΏ be a holomorphic line bundle on π defining an embedding ππΏ : π β β, where β = β(π»0(π, πΏ)β¨) (seeβBundles, fibre -β). Set π = βπββππ¦πππ»0(O(πΏ)), the homoge-neous coordinate ring of the projective space β, and consider the graded π-module πΊ = βπββπ»0(π,O(πΏπ)). Let πΈβ
β β β β πΈ1β πΈ0β πΊ β 0
be a minimal graded free resolution of πΊ (seeβMinimal free resolutionsβ). For any π β β, we say that the line bundle πΏ satisfies Property ππif the two following condi-tions hold:
πΈ0= π,
πΈπ= βπ(βπ β 1) for 1 β€ π β€ π,
where the second condition means that πΈπis the direct sum of some copies of π(βπβ1).
Observe that the kernel of the map π β πΊ is the ideal of ππΏ(π) (take the cohomology of the exact sequence 0 βIππΏ(π)(π) βOβ(π) βOππΏ(π)(π) β 0, whereIππΏ(π)is the ideal sheaf ππΏ(π)).
Thus, πΏ satisfies Property π0if and only if πΏ is normally generated, πΏ satisfies Prop-erty π1if and only if it satisfies Property π0 and the ideal of ππΏ(π) is generated by quadrics, and πΏ satisfies Property π2if and only if it satisfies Property π1and the syzygies among these quadrics are linear, and so on.
See alsoβGroebner basesβ,βHilbert syzygy theoremβ.
190 | Tautological (or universal) bundle
T
Tautological (or universal) bundle.
([93], [188]).β Let π be a vector space of dimension π and let π < π. The tautological (or uni-versal bundle) on the Grassmannian of π-subspaces in π, πΊ(π, π) (see βGrass-manniansβ), is the bundle whose fibre on π β πΊ(π, π) is the π-subspace π. In particular the tautological bundle on the projective space β(π) is the line bundle whose fibre on π is the line π; it is the dual of the hyperplane bundle (see βHyper-plane bundles, twisting sheavesβ).
β Let πΈ be a vector bundle on a manifold (or an algebraic variety) π and let π : β(πΈ) β π be the projectivized bundle, i.e., the bundle whose fibre on π₯ β π is the projectivized of πΈπ₯. The tautological (or universal) bundle π on the projec-tivized bundle β(πΈ) is the following bundle: the subbundle of πβπΈ whose fibre on a point π of β(πΈ) is the line represented by π.
Its dual is the line bundleOβ(πΈ)(1), i.e., the line bundle whose restriction on β(πΈπ₯) isO(1) for all π₯ β π.
β More generally: let πΈ be a vector bundle on a manifold (or an algebraic variety) π, and let π : πΊ(π, πΈ) β π be the bundle whose fibre on π₯ β π is πΊ(π, πΈπ₯); the tau-tological (or universal) bundle π on πΊ(π, πΈ) is the following bundle on πΊ(π, πΈ):
the subbundle of πβπΈ whose fibre on a point π of πΊ(π, πΈ) is the π-subspace of πΈπ(π)= (πβπΈ)πgiven by π:
π β πβ(πΈ)
πΈ
πΊ(π, πΈ) π //π
Obviously taking π equal to a point, we get the notion of universal bundle on the Grassmannian and taking π = 1 we get the notion of universal bundle on the projectivized of a bundle.
Tor, TOR.
([41], [62], [93], [116]). Let π be a commutative ring with unity. Let π be an π -module. Letβ β β σ³¨β ππσ³¨β ππβ1σ³¨β β β β σ³¨β π0σ³¨β π σ³¨β 0
be a projective resolution of π (seeβInjective and projective resolutionsβ); we denote by πβthe complex
β β β σ³¨β ππσ³¨β ππβ1σ³¨β β β β . σ³¨β π0σ³¨β 0.
Let π be another π -module. We can consider the complex πββπ π:
β β β σ³¨β ππβπ π σ³¨β ππβ1βπ π σ³¨β β β β σ³¨β π0βπ π σ³¨β 0.