Horseshoe lemma. Let π be a ring and let
0 //π1 πΌ //π2 π½ //π3 //0 be an exact sequence of π -modules. Let
. . . //πΉπ‘ ππ‘// . . . π1 //πΉ0 π0 //π1 //0 and
. . . //πΊπ‘ ππ‘// . . . π1 //πΊ0 π0 //π3 //0
be two projective resolutions of π1and π3respectively (seeβInjective and projective resolutionsβ). Then there is a projective resolution of π2
. . . //πΉπ‘β πΊπ‘ ππ‘ // . . . π1 //πΉ0β πΊ0 π0 //π2 //0
is commutative with exact rows and columns (where, for any π‘, the map ππ‘is the obvi-ous injection and the map ππ‘is the obvious projection).
The lemma takes its name from the shape of the given part of the diagram (the short exact sequence, and the given projective resolutions).
An analogous statement holds in case we have an injective resolution of π1and an injective resolution of π3.
Hurwitzβs theorem.
SeeβRiemann surfaces (compact -) and algebraic curvesβ.Hypercohomology of a complex of sheaves.
([79], [93]). We follow strict-ly [93].Hyperplane bundles, twisting sheaves | 93 Let
β β β σ³¨βFπ πσ³¨βFπ+1σ³¨β β β β β β β
be a complex of sheaves of Abelian groups on a topological space π (seeβSheavesβ).
We denote it byFβ.
The cohomology sheafHπ(Fβ) is the sheaf on π associated to the presheaf
π σ³¨σ³¨β πΎππ (Fπ(π)σ³¨βπ Fπ+1(π)) πΌπ (Fπβ1(π)σ³¨βπ Fπ(π))
for any π open subset of π. LetU = {ππΌ}πΌβπ΄be an open covering of π. Consider the bigraded complex πΆπ(U,Fπ) (seeβSheavesβand in particular the cohomology of the sheaves for the notation) with the differentials
πΏ : πΆπ(U,Fπ) σ³¨β πΆπ+1(U,Fπ), π : πΆπ(U,Fπ) σ³¨β πΆπ(U,Fπ+1),
where π is induced by the differential ofFβ. Let πΆβ(U) be the associated single complex given by πΆπ(U) = βπ+π=ππΆπ(U,Fπ) with differential π· = π + πΏ.
The hypercohomology of the complex of sheavesFβis defined to be the limit onUof the cohomology of the complex πΆβ(U), where the set of the open coverings is partially ordered by refinement (seeβLimits, direct and inverse -β):
IHIπ(π,Fβ) = lim
βU
π»π(πΆβ(U)) for any π.
By considering the two filtrations of πΆβUand by passing to the limit forU, we get two spectral sequences (seeβSpectral sequencesβ) πΈπand πΈσΈ πwith πΈπ,π2 β π»π(π,Hπ(Fβ)) and πΈσΈ 2π,πβ π»ππ(π»π(π,Fβ)).
Proposition. If a map between complexes of sheaves induces an isomorphism on co-homology sheaves, then it also induces an isomorphism on hypercoco-homology.
Hyperelliptic Riemann surfaces.
SeeβRiemann surfaces (compact -) and al-gebraic curvesβ.Hyperelliptic or bielliptic surfaces.
SeeβSurfaces, algebraic -β.Hyperplane bundles, twisting sheaves.
([93], [107], [129]). Let πΎ be a field and π β β€. The line bundle π»π on βππΎis defined in the following way: let the total space be((πΎπ+1β {0}) Γ πΎ)/ βΌ, where βΌ is the following equivalence relation:
(π₯0, . . . , π₯π, π‘) βΌ (ππ₯0, . . . , ππ₯π, ππ π‘) βπ β πΎ β {0};
94 | Hyperplane bundles, twisting sheaves the projection
π : π»π β βππΎ is given by
π([(π₯0, . . . , π₯π, π‘)]) = [π₯0:β β β :π₯π].
We can take as trivializing subsets for π»π the sets
ππ= {[π₯0:β β β :π₯π] β βππΎ| π₯π ΜΈ= 0}
for π = 0, . . . , π, and the trivializing function ππ: ππΓ πΎ β π»π |ππis ππ([π₯0:β β β :π₯π], π‘) = [(π₯0, . . . , π₯π, π₯π ππ‘)].
Thus the transition functions are
ππ,π([π₯0:β β β :π₯π]) = π₯βπ π π₯π π.
The bundle π»1is called a hyperplane bundle (since it is associated to the divisor given by a hyperplane) and is the dual of the universal bundle (seeβTautological (or univer-sal) bundleβ). Obviously π»π = π»βπ . The sheafO(π ) (or, more precisely,OβππΎ(π )) is de-fined to be the sheaf of the sections of π»π . SometimesO(π ) denotes also the line bundle itself. The sheavesO(π ) are called Serreβs twisting sheaves.
For any sheafEon βππΎ, we denoteEβO(π ) byE(π ).
A more general way to define the hyperplane bundles and the twisting sheaves is the following one.
Let π be a graded commutative ring with unity. For any graded π -module π, letFπ
be the following sheaf on the scheme ππππ(π ) (seeβSchemesβ): for every π β ππππ(π ), let π(π)be the group of the elements of degree 0 in the localization of π with respect to the multiplicative system of the homogeneous elements in π β π (seeβLocalization, quotient ring, quotient fieldβ); for every open subset π of ππππ(π ), letFπ(π) be the set of the functions
π : π β βπβππ(π)
such that π(π) β π(π)for every π β π and π is locally a fraction, i.e., for all π β π, there exists a neighborhood π of π in π and homogeneous elements π β π, π β π of the same degree such that, for all πσΈ β π, we have π ΜΈβ πσΈ and π(πσΈ ) = ππ.
For any graded π -module π and π β β€, let π(π ) be the graded module whose part of degree π is ππ+π (the part of degree π + π of π). For any π β β€, we define the twisting sheafOπ(π ) on π := ππππ(π ) in the following way:
Oπ(π ) =Fπ (π ),
i.e.,Oπ(π ) is defined to be the sheaf associated to the π -module π (π ). We can prove that Oπ(π ) is locally free (seeβSheavesβ) of rank 1 and that, for any π , π‘ β β€,
Oπ(π ) βOπ(π‘) =Oπ(π‘ + π ).
Injective and projective modules | 95 If we take π = ππππ(π ) where π = πΎ[π₯0, . . . , π₯π] for some algebraically closed field πΎ (thus π is the scheme corresponding to the projective space of dimension π over πΎ), we get the sheaves we have described before.
Theorem.
β Let π΄ be a Noetherian graded commutative ring with unity and let π = π΄[π₯0, . . . , π₯π]. Let π = ππππ(π ). DenoteOπ(π ) byO(π ) for the sake of brevity.
β By Serre duality (seeβSerre dualityβ)
β0(π,O(π )) = βπ(π,O(βπ β π β 1)) .
Theorem. Let πΎ be an algebraically closed field. Then any locally free sheaf of rank 1 on the scheme βππΎis isomorphic toO(π ) for some π β β€.
I
Injective and projective modules.
([41], [62], [116], [159], [208]). Let π be a ring. An module πΌ is said to be injective if for every injective morphism of π -modules π : π β π and every π -morphism π : π β πΌ there exists an π -morphismAn π -module π is said to be projective if for every surjective π -morphism of π -modules π : π β π and every π -morphism π : π β π there exists an π -morphism β : π β π
96 | Injective and projective resolutions
Proposition. Let π be a ring with unity and consider only unital modules.
Free π -modules (i.e., direct sums of copies of π ) are projective.
Any projective module is a direct summand of a free module.
Let πΎ be a field. Any finitely generated projective πΎ[π₯0, . . . , π₯π]-module is free (see [159, Chapter IV, Theorem 3.15]).
Injective and projective resolutions.
([41], [62], [116], [208]). Let π be a ring.An injective resolution of an π -module π is an exact sequence of π -modules 0 β π β πΌ0β β β β β πΌπβ β β β
with πΌπinjective π -modules.
A projective resolution of an π -module π is an exact sequence of π -modules
β β β β ππβ ππβ1β β β β β π0β π β 0 with ππprojective π -modules.
Integrally closed.
([12], [62], [164], [256]). Let π΅ be a commutative ring with unity.Let π΄ be a subring of π΅ (so the unity belongs to π΄). We say that π₯ β π΅ is integral over π΄ if there exists a monic polynomial π with coefficients in π΄ such that π(π₯) = 0.
The integral closure of π΄ in π΅ is defined to be the set πΆ := {π₯ β π΅ | π₯ integral over π΄}.
If π΄ = πΆ, then we say that π΄ is integrally closed in π΅. If π΅ = πΆ, then we say that π΅ is integral over π΄.
We say that a domain π΄ is integrally closed if it is integrally closed in its quotient field (seeβLocalization, quotient ring, quotient fieldβ).
Proposition. ring, quotient fieldβfor the notation πβ1πΆ).
Intersection of cycles | 97 (ii) Let π be a commutative ring with unity and let π σΈ , π σΈ σΈ be two subrings of π with π σΈ σΈ β π σΈ β π . If π is integral over π σΈ and π σΈ is integral over π σΈ σΈ , then π is integral over π σΈ σΈ .
(iii) Let π be a domain. Then π is integrally closed if and only if π πis integrally closed for every prime ideal π of π and this is true if and only if π πis integrally closed for every maximal ideal π of π (seeβLocalization, quotient ring, quotient fieldβfor the notation π πand π π).
(iv) A unique factorization domain is integrally closed, in particular, for any field πΎ, the ring πΎ[π₯1, . . . , π₯π] is integrally closed.
Intersection of cycles.
([72], [74], [93], [107], [214], [227], [228]).Intersection in topology. We follow mainly [93].
Let π be an oriented πΆβmanifold of (real) dimension π. Let π1and π2be two singular homology classes of cycles of complementary dimension, precisely,
π1β π»π(π, β€), π2β π»πβπ(π, β€)
(seeβSingular homology and cohomologyβ for the notation). Let πΎ1 and πΎ2 be two piecewise smooth cycles such that [πΎ1] = π1, [πΎ2] = π2, and πΎ1and πΎ2intersect trans-versely (i.e., every intersection point π is smooth in πΎ1and πΎ2and πππ = πππΎ1β πππΎ2, where ππdenotes the tangent space at π). We define
π1β π2= β
πβπΎ1β©πΎ2
ππ(πΎ1, πΎ2),
where ππ(πΎ1, πΎ2) is defined as follows: take a positively oriented basisB1of πππΎ1and a positively oriented basisB2of πππΎ2(observe that πΎ1and πΎ2have natural orientations given respectively by π1and π2); ifB1,B2is a positively oriented basis of πππ, then we define ππ(πΎ1, πΎ2) to be 1, otherwise we set ππ(πΎ1, πΎ2) = β1. One can show that we can find πΎ1and πΎ2as above and that π1β π2does not depend on the choice of πΎ1and πΎ2.
We can define also the intersection of two singular homology classes of cycles of non-complementary dimension. Precisely, let
π1β π»πβπ1(π, β€), π2β π»πβπ2(π, β€),
with π1+ π2 < π. Let πΎ1and πΎ2be two piecewise smooth cycles such that [πΎ1] = π1, [πΎ2] = π2 and πΎ1and πΎ2intersect transversely almost everywhere (we say that they intersect transversely in a point π if π is smooth in πΎ1and πΎ2and πππΎ1+ πππΎ2= πππ).
We define π1β π2to be the element of π»π βπ1βπ2(π, β€) given by πΎ1β© πΎ2with the following orientation: ifBis a positively oriented basis of ππ(πΎ1β©πΎ2), where π is a smooth point of πΎ1β©πΎ2, andB1,Bis a positively oriented basis of πππΎ1andB,B2is a positively oriented basis of πππΎ2, thenB1,B,B2is a positively oriented basis of πππ.
98 | Inverse image sheaf
One can prove that the intersection of cycles corresponds through the PoincarΓ© duality to the cup product; seeβSingular homology and cohomologyβ.
Intersection in algebraic geometry. Let π be a smooth algebraic variety of dimension π over an algebraic closed field πΎ and let π1and π2be two subvarieties of π of co-dimension π1and π2, respectively. Suppose π1+ π2β€ π. We say that π1and π2intersect properly if the codimension of every irreducible component of π1β©π2is equal to π1+π2. Suppose that π1and π2intersect properly. We define the intersection cycle of π1and π2, which we denote by π1β π2or simply by π1π2, to be the algebraic cycle (seeβCyclesβ)
β
πππ(π1, π2) π,
where the sum runs over all the irreducible components of π1β© π2and the number ππ(π1, π2) is the so-called βintersection multiplicityβ of π1and π2along π. There are several definitions of intersection multiplicity; we report Serreβs definition (see [227]):
if π is an irreducible component of π1β© π2, ππ(π1, π2) := β
π (β1)ππππππ‘β (ππππ΄π(π΄/πΌ1, π΄/πΌ2)) ,
where π΄ is the local ringOπ,π₯of π at a generic point π₯ β π and πΌπis the ideal of ππin π΄ for π = 1, 2 (seeβLength of a moduleβ,βTor,TORβfor the definitions of πππππ‘β and πππ). Serre showed that the numbers ππ(π1, π2) are nonnegative.
By linearity we can define the intersection cycle of any two cycles π1 and π2when they intersect properly, i.e., when the subvarieties of which π1is a linear combination intersect properly the subvarieties of which π2is a linear combination.
Chowβs moving lemma states that if π is a smooth quasi-projective algebraic variety and π1and π2are cycles of π of codimension π1and π2, respectively, we can find a cycle π1σΈ rationally equivalent to π1 and such that π1σΈ and π2intersect properly (see
βEquivalence, algebraic, rational, linear -, Chow, NeronβSeveri and Picard groupsβ
for the definition of rational equivalence). Moreover, one can show that if π1and π1σΈ are rationally equivalent and intersect properly π2, then π1β π2and π1σΈ β π2are rationally equivalent. This allows us to define the intersection of algebraic cycles (also intersect-ing not properly) up to rational equivalence.
Inverse image sheaf.
SeeβDirect and inverse image sheavesβ.Irreducible topological space.
We say that a topological space is irreducible if it is not the union of two proper closed subsets.Irregularity.
([93], [107]). Let π be a complex manifold or an algebraic variety over a field πΎ. The irregularity of π isβ1(π,Oπ),
whereOπis the sheaf of the holomorphic functions, respectively of the regular func-tions, on π. The irregularity of π is generally denoted by π(π).
Jacobians of compact Riemann surfaces | 99 If π is a compact KΓ€hler manifold (seeβHermitian and KΓ€hlerian metricsβ), then, by Dolbeaultβs theorem and by Hodgeβs theorem (seeβDolbeaultβs theoremβ,βHodge the-oryβ), we have that β1(π,Oπ) = β0,1(π) = β1,0(π) = β0(π, πΊ1π) (and somewhere the irregularity is defined to be β1,0(π)).
If π is a smooth projective algebraic surface over an algebraic closed field, then the irregularity is equal to ππ(π)βππ(π) (where ππ(π) and ππ(π) are respectively the arith-metic and the geometric genus of π; seeβGenus, arithmetic, geometric, real, virtual -β
andβDualizing sheafβ).
J
Jacobians of compact Riemann surfaces.
([93], [101], [102], [163], [165], [195]). To every compact Riemann surface we can associate a principally polarized Abelian variety, called its Jacobian. Jacobians were the first Abelian varieties to be studied.Let π be a compact Riemann surface of genus π (seeβRiemann surfaces (com-pact -) and algebraic curvesβ). By Riemannβs theorem, the complex vector space π»0(π,O(πΎπ)), where πΎπis the canonical bundle of π (seeβCanonical bundle, canon-ical sheafβ), has dimension π. Consider the map
π : π»1(π, β€) σ³¨β π»0(π,O(πΎπ))β¨ defined by
π(πΎ) = β«
πΎ
. It is injective, and the quotient
π»0(π,O(πΎπ))β¨ π(π»1(π, β€))
is a complex torus of dimension π (seeβTori, complex - and Abelian varietiesβ). The Jacobian of π is the complex torus above endowed with the following polarization:
let πΈ be the alternating form on π»0(π,O(πΎπ))β¨obtained extending on β the form on π»1(π, β€) given by the intersection of 1-cycles; let
π» : π»0(π,O(πΎπ))β¨Γ π»0(π,O(πΎπ))β¨σ³¨β β, defined by
π»(π£, π€) = πΈ(ππ£, π€) + ππΈ(π£, π€),
for any π£, π€ β π»0(π,O(πΎπ))β¨. We endow the complex torus above with the polarization given by π». One can prove that it is a principal polarization. Thus the Jacobian of π is a principally polarized Abelian variety.
The Jacobian of π coincides with the Albanese variety of π (seeβAlbanese varietiesβ).
100 | Jacobians of compact Riemann surfaces
Definition. Let π·ππ£π(π) be the set of divisors of degree π on π. The map π·ππ£0(π) σ³¨β π½(π)
defined by
β
π=1,...,π
(ππβ ππ) σ³¨σ³¨β β
π=1,...,π ππ
β«
ππ
,
for any π β β, ππ, ππ β π, is called the AbelβJacobi map of π and denoted by π.
Obviously, if we fix a point π on π, we can also define a map π·ππ£π(π) σ³¨β π½(π)
(again called the AbelβJacobi map and denoted by π) by composing the map π·ππ£π(π) σ³¨β π·ππ£0(π),
π· σ³¨σ³¨β π· β ππ with the AbelβJacobi map π·ππ£0(π) σ³¨β π½(π).
AbelβJacobi theorem. The AbelβJacobi map defines an isomorphism (again called the AbelβJacobi map)
πππ0(π) σ³¨β π½(π) .
(See βEquivalence, algebraic, rational, linear -, Chow, NeronβSeveri and Picard groupsβfor the definition of the Picard group πππ0(π); in case π is a compact Riemann surface, πππ0(π) is the set of the divisors of degree 0 up to linear equivalence.) Notation. Define ππ := π(π(π)), where π(π)is the symmetric π-product of π, i.e., the set of effective divisors of degree π on π.
Proposition. The dimension of ππis π if π β€ π, while it is π if π β₯ π. If π· β π(π), then πβ1(π(π·)) = |π·| = β(π»0(π,O(π·)),
where |π·| denotes the linear system of π· (seeβLinear systemsβ) Let π β₯ 1; then the map π : π σ³¨β π½(π) is injective and the map π : π(π)σ³¨β π½(π) is generically injective for π β€ π.
Theorem. Let π© be the divisor in π½(π) associated to a section of a holomorphic line bundle on π½(π) defining the polarization. We have that the intersection number of π1 and π© is π:
π1β π© = π.
In addition, π© is a translate of the image through the AbelβJacobi map of the set of the
Jacobians, Weil and Griffiths intermediate - | 101
effective divisors of degree π β 1 on π, i.e., π© = ππβ1+ πΎ, for some πΎ β π½(π).
Torelliβs theorem. The map from the set of compact Riemann surfaces up to isomor-phisms to the set of the principally polarized Abelian varieties up to isomorisomor-phisms, associating to every Riemann surface π its Jacobian π½(π) is injective.
PoincarΓ©βs formula. Let [ππ] be the class of ππ in the singular homology group π»2πβ2π(π½(π), β€). We have that
[ππ] = [π©](π β π)!πβπ,
where π© is the divisor associated to a section of a holomorphic line bundle defining the polarization on π½(π).
Jacobians, Weil and Griffiths intermediate -.
([92], [93], [166], [250]). We follow mainly [166].Let π be a compact KΓ€hler manifold of dimension π (seeβHermitian and KΓ€hlerian metricsβ). Let 1 β€ π β€ π. The Griffiths π-th intermediate Jacobian of π is the complex torus (seeβTori, complex - and Abelian varietiesβ)
π½π(π) = βπ=π,...,2πβ1π»2πβ1βπ,π(π, β) π(π»2πβ1(π, β€)) ,
where π : π»2πβ1(π, β€) β βπ=π,...,2πβ1π»2πβ1βπ,π(π, β) is given by the composition of the canonical map π’ : π»2πβ1(π, β€) β π»2πβ1(π, β) with the projection π»2πβ1(π, β) β
βπ=π,...,2πβ1π»2πβ1βπ,π(π, β) induced by the Hodge decomposition (seeβHodge theoryβand
βSingular homology and cohomologyβ, in particular the Universal Coefficient Theo-rem). If π = 1 we have the Picard variety of π (seeβEquivalence, algebraic, rational, linear -, Chow, NeronβSeveri and Picard groupsβ):
π½1(π) = π»1(π,O)
π(π»1(π, β€)) = πππ0(π).
If π = π we have the Albanese variety of π (seeβAlbanese varietiesβ):
π½π(π) = π»πβ1,π(π, β)
π(π»2πβ1(π, β€))β π»1,0(π, β)β¨
π(π»1(π, β€)) β π»0(π, πΊ1)β¨
π(π»1(π, β€)) = π΄ππ(π),
where π is the map πΎ σ³¨β β«πΎ(seeβDolbeaultβs theoremβ,βHodge theoryβ,βSerre dualityβ
to understand the isomorphisms above).
We have
πππ0(π΄ππ(π)) = π»0,1(π΄ππ(π), β)
π(π»1(π΄ππ(π), β€)) β π»ππβ(π»1,0(π, β), β) π(π»1(π, β€))
β π»1(π,O)
π(π»1(π, β€)) = πππ0(π),
102 | Jacobians, Weil and Griffiths intermediate
-and analogously we also have an isomorphism π΄ππ(πππ0(π)) β π΄ππ(π). So πππ0(π) and π΄ππ(π) are dual complex tori (seeβTori, complex - and Abelian varietiesβ). More gen-erally, we can prove that π½π(π) and π½πβπ+1(π) are dual complex tori.
Now let π be a smooth complex projective algebraic variety, and let π be the Fubini-Study form restricted to π (thus π is a closed positive integer (1, 1)-form).
If 2π β 1 β€ π, we can consider on π½π(π) the polarization with index defined by the following Hermitian form (if 2π β 1 > π, we define the polarization with index as the dual polarization with index of the one on π½πβπ+1(π)):
π»(π, π) = 2π(β1)πβ«
π
ππβ2π+1β§ π β§ π;
the form π» is Hermitian; in fact it is β-linear in the first variable and π»(π, π) = 2π(β1)πβ«
In general it is not positive definite, so in general the Griffiths intermediate Jacobian is not an Abelian variety; the Albanese variety and the Picard variety (of a smooth complex projective algebraic variety) are Abelian varieties.
Let 1 β€ π β€ π. Let πΆ : π»π(π, β) β π»π(π, β) be the linear operator defined to be the multiplication by ππβπ on π»π,π(π); it takes π»π(π, β) to π»π(π, β) and, if π is odd, we have πΆ2 = β1. The operator πΆ is called Weilβs operator. The π-th Weil intermediate Jacobian of π is the following torus:
ππ(π) = π»2π β1(π, β) π’(π»2π β1(π, β€)),
with the complex structure given by βπΆ; here π’ : π»2π β1(π, β€) β π»2π β1(π, β) is the canonical map; its image is isomorphic to the free part of π»2π β1(π, β€). We can prove that ππ(π) and ππβπ+1(π) are dual complex tori.
Jumping lines and splitting type of a vector bundle on βπ | 103 For 2π β 1 > π, we define the polarization on ππ(π) to be the dual polarization of the one on ππβπ+1(π).
One can prove that the Weil intermediate Jacobian is an Abelian variety (with the po-larization we have just defined).
As we have already said, Griffiths intermediate Jacobian is not an Abelian variety in general. The advantage of Griffiths Jacobian with respect to Weil Jacobian is that Grif-fiths Jacobian varies holomorphically in a family of smooth projective algebraic vari-eties (while Weil Jacobian does not).
See alsoβJacobians of compact Riemann surfacesβ.