Positivity of vector bundles and Hodge theory
PPOSITIVITY OF VECTOR BUNDLES AND HODGE THEORY
MARK GREEN AND PHILLIP GRIFFITHS
Abstract.
It is well known that positivity properties of the curvature of a vectorbundle have implications on the algebro-geometric properties of the bundle, such asnumerical positivity, vanishing of higher cohomology leading to existence of globalsections etc. It is also well known that bundles arising in Hodge theory tend to havepositivity properties. From these considerations several issues arise:(i) For a bundle that is semi-positive but not strictly positive; what further nat-ural conditions lead to the existence of sections of its symmetric powers?(ii) In Hodge theory the Hodge metrics generally have singularities; what can besaid about these and their curvatures, Chern forms etc.?(iii) What are some algebro-geometric applications of positivity of Hodge bundles?The purpose of these partly expository notes is fourfold. One is to summarizesome of the general measures and types of positivity that have arisen in the litera-ture. A second is to introduce and give some applications of norm positivity . Thisis a concept that implies the different notions of metric semi-positivity that arepresent in many of the standard examples and one that has an algebro-geometricinterpretation in these examples. A third purpose is to discuss and compare someof the types of metric singularities that arise in algebraic geometry and in Hodgetheory. Finally we shall present some applications of the theory from both theclassical and recent literature.
Outline I. Introduction and notation and terminology
A. IntroductionB. General notations and terminologyC. Notations and terminology from Hodge theoryII.
Measures and types of positivity
A. Kodaira-Iitaka dimensionB. Metric positivityC. Interpretation of the curvature formD. Numerical positivityE. Numerical dimensionF. Tangent bundleG. Standard implicationsH. A further resultIII.
Norm positivity
A. Definition and first propertiesB. A result using norm positivityIV.
Singularities
A. Analytic singularities
These are notes prepared by the authors and largely based on joint work in progress with RaduLaza and Colleeen Robles (cf. [GGLR17]). a r X i v : . [ m a t h . AG ] O c t . Logarithmic and mild singularitiesV. Proof of Theorem IV.B.8
A. Reformulation of the resultB. Weight filtrations, representations of sl and limiting mixed Hodge struc-turesC. Calculation of the Chern forms Ω and Ω I VI.
Applications, further results and some open questions
A. The Satake-Baily-Borel completion of period mappingsB. Norm positivity and the cotangent bundle to the image of a period map-pingC. The Iitaka conjectureD. The Hodge vector bundle may detect extension dataE. The exterior differential system defined by a Chern formI.
Introduction and notation and terminology
I.A.
Introduction.
The general purpose of these notes is to give an account of some aspects andapplications of the concept of positivity of holomorphic vector bundles, especiallythose that appear in Hodge theory. The applications use the positivity of Hodge linebundle (Theorem I.A.14), the semi-positivity of the cotangent bundle to the imageof a period mapping (Theorem VI.B.1), and the semi-positivity of the Hodge vectorbundle (Theorem I.A.26). Also discussed is the numerical positivity of the Hodgebundles.The overall subject of positivity is one in which there is an extensive and richliterature and in which there is currently active research occurring with interestingresults appearing regularly. We shall only discuss a few particular topics and shalluse [Dem12a] as our reference for general background material concerning positivityin complex analytic geometry as well as an account of some recent work, and referto [P˘a16] as a source for a summary of some current research and as an overall guideto the more recent literature. The recent paper [Den18] also contains an extensivebibliography. For Hodge theory we shall use [PS08] and [CMSP17] as our main generalreferences.Following this introduction and the establishment of general notations and termi-nology in Sections I.B and I.C, in Section II we shall give a synopsis of some of thestandard measures and types of positivity of holomorphic vector bundles. In SectionII.H we present the further result (I.A.9) below. It is this result that enables us tolargely replace the cohomological notion of weak positivity in the sense of Viehweg[Vie83a], [Vie83b] with purely differential geometric considerations.One purpose of these notes is to give an informal account of some of the results in[GGLR17] and to discuss topics that are directly related to and/or grew out of thatwork. One application to algebraic geometry of Theorems 1.2.2 and 1.3.10 there is he following: (I.A.1) Let M be the KSBA moduli space for algebraic surfaces X of generaltype and with given p g ( X ) , q ( X ) and K X . Then the Hodge line bundle Λ e → M is defined over the canonical completion M of M . Moreover Proj(Λ e ) exists and defines the Satake-Baily-Borel completion M of theimage of M = Φ( M ) of the period mapping Φ : M → Γ \ D . The period mapping extends to Φ e : M → M and set-theoretically the image of the boundary ∂ M = M \ M consists of the associatedgraded to the limiting mixed structures along the boundary strata of ∂ M . As has beenfound in applying this result to the surfaces analyzed in [FPR15a], [FPR15b], [FPR17]and to similar surfaces studied in our work and in discussions with the authors ofthose papers, the extended period mapping may serve as an effective method fororganizing and understanding the boundary structure of M and in suggesting how todesingularize it. The construction of M is general and in brief outline proceeds as follows:(i) One begins with a variation of Hodge structure over a smooth quasi-projectivevariety B given by a period mapping(I.A.2) Φ : B → Γ \ D. Here B has a smooth completion B where Z = B \ B is a reduced normal crossingdivisor with irreducible component Z i around which the local monodromies T i areassumed to be unipotent. We may also assume that Φ has been extended across any Z i for which T i are the identity; then Φ is proper and the image Φ( B ) = M ⊂ Γ \ D is a closed analytic subvariety ([Som78]).(ii) Along the smooth points Z ∗ I of the strata Z I := (cid:84) i ∈ I Z i of Z there is a limit-ing mixed Hodge structure ([CKS86]), and passing to the associated graded gives avariation of Hodge structure Φ I : Z ∗ I → Γ I \ D I . Extending Φ I across the boundary components of Z ∗ I around which the monodromy isfinite, from the image of the extension of Φ I we obtain a complex analytic variety M I .Then as a set M = M ∪ (cid:16)(cid:91) I M I (cid:17) , Significant amplification and complete proofs of the results in [GGLR17] are currently beingprepared. Below we shall give a short algebro-geometric argument for the case when dim B = 2. We shall use [Kol13] as our general reference for the topic of moduli. The definition of a Satake-Baily-Borel completion will be explained below. Here we are mainlyconsidering that part of the period mapping that arises from the polarized Hodge structure on H ( X ). When p g ( X ) (cid:61) In contrast to the case of curves, M seems almost never to be smooth even when M is. here on the right-hand side there are identifications made corresponding to strata Z ∗ I , Z ∗ J where Z I ∩ Z J (cid:54) = ∅ and where Φ I , Φ J have both been extended across intersectionpoints.(iii) This defines M as a set, and it is not difficult to show that the resulting M hasthe structure of a compact Hausdorff topological space and that the extended periodmapping(I.A.3) Φ e : B → M is proper with compact analytic subvarieties F x := Φ − e ( x ), x ∈ M , as fibres. If wedefine O M,x = ring of functions f that are continuous ina neighborhood U of x ∈ M and whichare holomorphic in Φ − e ( U ) and constanton fibres of Φ e then the issue is to show that there are enough functions in O M,x to define the structureof a complex analytic variety on M . (iv) Using [CKS86] the local structure of Φ e along F x ⊂ Φ − e ( U ) can be analyzed;this will be briefly recounted in Section V below, the main point being to use im-plications of the relative weight filtration property (RWFP) of limiting mixed Hodgestructures.(v) This leaves the issue of the global structure of Φ e along F x . We note that thereis a map m x / m x → H (cid:16) N ∗ F x /B (cid:17) where m x ⊂ O M,x is the maximal ideal and N ∗ F x /B → F x is the co-normal bundle of F x in B . Thus one expects some “positivity” of N ∗ F x /B . From F x ⊂ Z ∗ I,e ⊂ B we obtain0 → N ∗ Z ∗ I,e /B (cid:12)(cid:12) F x → N ∗ F x /B → N ∗ F x /Z ∗ I,e → . Since Φ I : Z ∗ I,e → Γ I \ D I is defined and maps F x to a point, we obtain some positivityof N ∗ F x /Z ∗ I,e . Denoting by m I,x the maximal ideal in O Γ I \ D I,x at x = Φ I ( F x ), from thelocal knowledge of Φ e along F x we are able to infer that the positivity of N ∗ F x /Z ∗ I,e thatarises from m I,x / m I,x lifts to sections in H (cid:16) N ∗ F x /B (cid:17) ; thus the main issue is that ofthe positivity of N ∗ Z ∗ I,e /B (cid:12)(cid:12) F x .(vi) Here, as will be explained in the revised and expanded version of [GGLR17],some apparently new Hodge theoretic considerations arise. In the literature, and inthis work, there have been numerous consequences drawn from the positivity prop-erties of the Hodge line bundle, and also from some of the positivity properties ofthe cotangent bundle of the smooth points of the image M = Φ( B ) of a period map-ping. However we are not aware of similar applications of positivity properties of thebundles constructed from the extension data E associated to a limiting mixed Hodgestructure. As will be explained in loc. cit., there is an ample line bundle L → E and n isomorphism(I.A.4) ν ∗ L ∼ = N ∗ Z ∗ I,e/B (cid:12)(cid:12) Fx where(I.A.5) ν : F x → E associates to each point of F x the extension data associated to the LMHS at that point.Put another way, assuming Φ ∗ is generically injective, since the associated graded tothe limiting mixed Hodge structure is constant along F x one may expect non-trivialvariation in the extension data to the LMHS’s, i.e. in ν in (I.A.5) (or possibly somejet of ν ) should be non-constant. The map (I.A.4) then converts sections of the amplebundle L → E into sections of N ∗ F x /B . Returning to the discussion of (I.A.2), a central ingredient in its proof consists ofthe positivity properties of the Hodge line bundle Λ e . Referring to [GGLR17] and toSections IV.B, V, and VI.A below for details, the Chern form ω e of Λ e is a singulardifferential form on a desingularization B of M , and the exterior differential system(I.A.6) ω e = 0defines a complex analytic fibration whose quotient captures the polarized Hodgestructure on H ( X ) when X is smooth (or has canonical singularities), and when X = lim X t is a specialization of smooth X t ’s it captures the associated graded tothe limiting mixed Hodge structure. As discussed above it is the extension data inthe LMHS that is not detected by the extended period mapping. In this way theSatake-Baily-Borel completion is a minimal completion of the image of the periodmapping.The positivity of the Hodge line bundle raises naturally the issue of the degree ofpositivity of the Hodge vector bundle. It is this question that provides some of thebackground motivation for these notes. A vector bundle E → X over a complexmanifold will be said to be semi-positive if it has a Hermitian metric whose curvatureform(I.A.7) Θ E ( e, ξ ) (cid:61) E (cid:61)
0. The Hodge vector bundle is semi-positive, but the curvature form is generallynot positive, even at a general x ∈ X and general e ∈ E x . An interesting algebro-geometric question is: How positive is it? In the classical case of curves this extension data is given by the Jacobian variety of a smooth,but generally reducible, curve and L → E is the “theta” line bundle given by the polarization. Itis interesting and we think noteworthy that constructions akin to classical theta functions can begiven in non-classical situations. Of course one must make sense of (I.A.6) where ω e has singularities; this is addressed in SectionVI.E below. e shall say that a general bundle E → X is strongly semi-positive if there is ametric where (I.A.7) is satisfied and where(I.A.8) Tr Θ E = Θ det E > ∗ ,n of the differential of the period mapping is injective at a general point,a condition that is frequently satisfied in practice. One of the main observations inthese notes is(I.A.9) If X is compact and E → X is strongly semi-positive, then for some r (cid:53) rank E we have for m (cid:61) r Θ Sym m E > on an open set. As a corollary, Sym m E is big for m (cid:61) r . As the proof of (I.A.9) will show, although Sym m E “gets more sections” as m in-creases, in general Sym m E will never become entirely positive. We will see in SectionVI.E that the Sym m E have an intrinsic amount of “flatness” no matter what metricwe use for E . Among the algebro-geometric measures of positivity of a vector bundle E → X onemay single out(i) E is nef (see Section II.D);(ii) some Sym m E is big;(iii) some Sym m E is free (i.e., Sym m E is semi-ample).There are natural curvature conditions that imply both (i) and (ii), but other than theassumption of strict positivity which implies ampleness, we are not aware of naturalcurvature conditions that imply (iii). For a specific question concerning this point,suppose that L → X is a line bundle and Z ⊂ X is a reduced normal crossing divisor.Assume that h is a smooth metric in L → X whose Chern form has the properties(a) ω (cid:61) ξ ∈ T x X , ω ( ξ ) = 0 ⇐⇒ x ∈ Z and ξ ∈ T x Z ⊂ T x X .Briefly, ω (cid:61) ω = 0 defines Z ⊂ X . Then by (a)we see that L is nef, and as a consequence of (b) L is big (cf. II.G). Now ω defines aK¨ahler metric ω ∗ on X ∗ := X \ Z , and one may pose the(I.A.10) Question:
Are there conditions on the curvature R ω ∗ of ω ∗ that implythat L is free? For families of algebraic curves and surfaces this condition is the same as the differential beinginjective. In general, using the augmented Hodge bundle there is a similar interpretation using thefull differential Φ ∗ . This is not surprising as one can always add a trivial bundle to one that is strongly semi-positiveand it will remain strongly semi-positive. Somewhat more subtly, a strongly semi-positive bundlemay have “twisted” sub-bundles on which the curvature form vanishes. e note that by (b) the Chern form ω defines a norm in the normal bundle N Z/X .It seems reasonable to ask if the curvature form of this norm may be computed fromlim x → Z R ω ∗ ( x ), and if there are sign properties of this curvature form that implyfreeness? Remark that the construction of the completion M of the image of a period mapping(I.A.2) could have been accomplished directly if one could show that the canonicallyextended Hodge line bundle Λ e → B , which satisfies (i) and (ii), is free. Here onehas the situation of the above question in the case where h and ω have singularitiesalong Z . However, as discussed in Section V these singularities are mild; in fact thesingularities tend to “increase” the positivity of the Chern form of Λ e .In Section III we introduce the notation of norm positivity . This means that E → X has a metric whose curvature matrix is of the formΘ E = − t A ∧ A where A is a matrix of (1 ,
0) forms arising from a holomorphic bundle map(I.A.11) A : E ⊗ T X → G for G a holomorphic vector bundle having a Hermitian metric. The curvature formis then(I.A.12) Θ E ( e, ξ ) = (cid:107) A ( e ⊗ ξ ) (cid:107) G where (cid:107) (cid:107) G is the norm in G . Many of the bundles that arise naturally in algebraicgeometry, such as the Hodge bundle and any globally generated bundle, have thisproperty: The mapping A in (I.A.11) generally has algebro-geometric meaning, e.g.,as the differential of a map. Such bundles are semi-positive, and their degree ofpositivity will have an algebro-geometric interpretation. Remark:
One may speculate as to the reasons for what might be called “the unrea-sonable effectiveness of curvature in algebraic geometry.” After all, algebraic geometryis in some sense basically a 1 st order subject—the Zariski tangent space being the pri-mary infinitesimal invariant—whereas differential geometry is a 2 nd order subject—the principle invariants are the 2 nd fundamental form and the curvature. One expla-nation is that for bundles that have the norm positivity property, the curvature formmeasures the size of the 1 st order quantity A in (I.A.11). Another principal topic in these notes concerns the singularities of metrics and theircurvatures and Chern forms. This also is a very active area that continues to play aprominent role in complex algebraic geometry (cf. [Dem12a] and [P˘a16]). Reflectingthe fact that in families of algebraic varieties there are generally interesting singular In a special case this question has been treated in the interesting paper [Mok12]. We refer to(I.A.38) at the end of this introduction for similar questions concerning the Hodge vector bundles. Of course the semi-ampleness of Λ e → B is a posteriori a consequence of (I.A.2). The origin of the term is that a quantity (such as a Hermitian form) will be non-negative if itis expressed as a norm. Both in the sense of (I.A.7) and in the sense of Nakano positivity ([Dem12a]). In this regard in the Hodge-theoretic situation we note that curvature R ω ∗ in (I.A.10) is a 2 nd ,rather than a 3 rd , order invariant. embers, the role of the singularities of Hodge metrics is central in Hodge theory.Here there are two basic principles:(i) the singularities are mild, and(ii) singularities increase positivity.The first is explained in Section IV.B. Intuitively it means that although the Chernpolynomials are singular differential forms, they behave in their essential aspects asif they were smooth. In particular, although they have distribution coefficients theymay be multiplied and restricted to particular subvarieties as if they were smoothforms. This last property is central to the proof of (I.A.1) above.The first major steps in the general analysis of the singularities of the Chern formsof Hodge bundles for several parameter variations of Hodge structure were taken byCattani-Kaplan-Schmid ([CKS86]), with subsequent refinements and amplificationsby a number of people including Koll´ar ([Kol87]). In the paper [GGLR17] furthersteps are taken, ones that may be thought of as further refining the properties of thewave front sets of the Chern forms. This will be explained in Section IV.B and willbe applied in Section V where an alternate proof of one of the two main ingredientsin the proof of (I.A.14) below will be given.The result (I.A.1) is an application to a desingularization of a KSBA moduli spaceof a result that we now explain, referring to Sections I.B and I.C for explanationsof notation and terminology. Let B be a smooth projective variety with smoothcompletion B such that B \ B = Z = (cid:83) Z i is a reduced normal crossing divisor. Wedenote by Z I = (cid:84) i ∈ I Z i the strata of Z and by Z ∗ I = Z I, reg the smooth points of Z I .We consider a variation of Hodge structure over B given by a period mapping(I.A.13) Φ : B → Γ \ D. We assume that the local monodromies T i around the Z i are unipotent with logarithms N i , an assumption that may always be achieved by passing to a finite covering of B . We also suppose that the end pieceΦ ∗ ,n : T B → Hom( F n , F n − /F n )of the differential of Φ is generically injective. Finally assuming as we may that all N i (cid:54) = 0, the image Φ( B ) := M ⊂ Γ \ D of the period mapping (I.A.13) is a closed analytic subvariety of Γ \ D . Beginning with[Som78] there have been results stating that under certain conditions M is a quasi-projective variety and the Hodge line bundle Λ → M is at least big. The followingresult from [GGLR17] serves to extend and clarify the previous work in the literature: Theorem
I.A.14 : There exists a canonical completion M of M as a compact complexanalytic space that has the properties We will use [PS08], [CMSP17] and [CKS86] as general references for Hodge theory, includinglimiting mixed Hodge structures. By using the augmented Hodge line bundle (cid:98) n − / (cid:99) ⊗ p =0 det( F n − p ) rather than just the Hodge linebundle Λ = det F n , this assumption may be replaced by the injectivity of Φ ∗ (cf. [GGLR17]). i) the Hodge bundle extends to Λ e → M and there it is ample; and (ii) M is a Satake-Baily-Borel completion of M . The second statement means the following: The period mapping (I.A.13) extendsto(I.A.15) Φ e : B → M .
Along the non-singular strata Z ∗ I the exented period mapping Φ e induces variationsof graded polarized limiting mixed Hodge structures, and passing to the associatedgraded of these mixed Hodge structures gives period mappings(I.A.16) Φ I : Z ∗ I → Γ I \ D I . Then the restriction to Z ∗ I of Φ e in (I.A.15) may be identified with Φ I . Setting M I =Φ I ( Z ∗ I ) ⊂ Γ I \ D I , as a set M = M (cid:113) ( (cid:96) I M I ) The precise meaning of this is explainedin [GGLR17]; among other things it means that Proj(Λ e → B ) exists and on theboundary strata exactly detects the variation of the associated graded to the limitingmixed Hodge structures. The exterior differential system (I.A.6) defined by thesingular differential form ω e may be made sense of on B (cf. Section VI.E), and thereit defines a fibration by complex analytic subvarieties whose quotient is just M . Therestriction property of ω e referred to above may be summarized as saying that(I.A.17) ω e (cid:12)(cid:12) Z ∗ I is defined and is equal to ω I where ω I is the Chern form of the Hodge line bundle associated to (I.A.16).An implication of the above is(I.A.18) ω e is defined on M and there ω e > . One aspect of this is that given B and B as above the EDS(I.A.19) (cid:40) ω = 0 on Bω I = 0 on the Z ∗ I We remark that in the non-classical case when Γ \ D is not an algebraic variety, the constructionof M is necessarily accomplished by gluing together local extensions of M at the points of ∂M = M \ M . This requires both a local analysis, based on [CKS86], of local neighborhoods in β along thefibres F of the set-theoretically extended period map together with global analysis of those fibres, andespecially of the above mentioned positivity of the co-normal bundle N ∗ F/B . As mentioned above,global issues necessitate apparently new Hodge-theoretic constructions arising from the extensiondata in a limiting mixed Hodge structure. Even in the classical case this seems to give a newperspective on the Satake-Baily-Borel compactification. In general, given a fibration f : A → B between manifolds A, B and a smooth differential formΨ A on A , the necessary and sufficient conditions thatΨ A = f ∗ Ψ B for a smooth differential form Ψ B on B are that both Ψ A and its exterior derivative d Ψ A restrict tozero on ker { f ∗ : T A → T B } ⊂
T A.
The above results extend this to a situation where Ψ A is a possibly singular differential form. n the smooth strata defines on each a complex analytic fibration and a naturalquestion is(I.A.20) For I ⊂ J so that we have Z ∗ J ⊂ Z ∗ I , does the closure of a fibre of ω I = 0 in Z ∗ I intersect Z ∗ J in a fibre of ω J = 0 ? Are the limits of the fibres of Φ I contained in the fibres of Φ J ? In other words, do the period mappings given by Φ on B and Φ I on Z ∗ I fit together inan analytic way? That this is the case is proved in Section 3 of [GGLR17] (cf. Step 1in the proof of Theorem 3.15 there). Although not directly related to the positivityof Hodge bundles, because of the subtle way in which the relative weight filtrationproperty of a several parameter degeneration of polarized Hodge structures enters inthe argument we think the result is of interest in its own right and in Section VI.Awe have given a proof in the main special case of a positive answer to (I.A.20).In Section VI we shall give applications of some of the bundles that naturally arisein Hodge theory. As explained above, one such application is the use of the Hodgeline bundle in proving that its “Proj” defines the Satake-Baily-Borel completion ofthe image of a period mapping. The next application is to the cotangent bundle Ω M o over the smooth points M o of M . It is classical [CMSP17] that the induced metricon M o ⊂ Γ \ D is K¨ahler and has holomorphic sectional curvatures R ( ξ ) (cid:53) − c, ξ ∈ T M o bounded above by a constant − c where c >
0. On the other hand the more basiccurvature form for
T M o is given by the holomorphic bi-sectional curvatures R ( η, ξ ),and we shall prove (cf. Theorem VI.B.1) that they satisfy R ( η, ξ ) (cid:53) and R ( η, ξ ) < is an open U set in T M o × M o T M o . Moreover, U projects onto each factor in M o × M o . (I.A.21)An interesting point here is that the induced curvature on the horizontal sub-bundle ( T D ) h ⊂ T D is a difference of non-negative terms, each of which has thenorm positivity property (I.A.12). On integrable subspaces I of ( T D ) h the positiveterm drops out leading to (I.A.21). A consequence of (I.A.21) is(I.A.22) the curvature form Θ T ∗ M o (cid:61) and it is positive on an open set. To apply (I.A.22) two further steps are required. One is that curvatures decrease onholomorphic sub-bundles, which in the case at hand is used in the form(I.A.23) Θ I (cid:53) Θ ( T D ) h (cid:12)(cid:12) I . In general the curvature matrices of Hodge bundles are differences of non-negative curvatureoperators, and the paper [Zuo00] isolated the important point that on the kernels of Kodaira-Spencermaps one of the two terms drops out and on these subspaces the curvature forms have a sign. Herethe relevant bundles are constructed from the Hom( F p /F p +1 , F p − /F p ) bundles, and the relevantKodaira-Spencer maps vanish in the integrable subspaces (cf. Proposition 2.1 in loc. cit.). This issueis discussed below in some detail in Section VI.B. he second involves the singularities that arise both at the singular point M sing = M \ M o , and at the boundary ∂M = M \ M of M . Once these are dealt with wefind the following, which are variants and very modest improvements of results in theliterature due to Zuo in [Zuo00] and others (cf. [Bru16b] for a recent paper on thisgeneral topic and [Den18] for related results and further bibliography).(I.A.24) M is of log general type, and there exists an m such that(I.A.25) Sym m Ω M (log) is big for m (cid:61) m . The precise meaning of these statements will be explained in Section VI.B below. The theme of these notes is the positivity of vector bundles, especially those arisingfrom Hodge theory, and some applications of this positivity to algebraic geometry.Above we have discussed applications of the positivity of the Hodge line bundle andof the cotangent bundle at the smooth points to the image of a period mapping. Wenext turn to an application of positivity of the Hodge vector bundle to the
Iitakaconjecture , of which a special case is(I.A.26)
Let f : X → Y be a morphism between two smooth projective varietiesand assume that (i) the general fibre X y = f − ( y ) has Kodaira dimension κ ( X y ) =dim X g , (ii) Var f = dim Y . Then the Kodaira dimension is sub-additive (I.A.27) κ ( X ) (cid:61) κ ( Y ) + κ ( X y ) . This result was proved with one assumption, later seen to be unnecessary, by Viehweg([Vie83a], [Vie83b]). His work built on [Fuj78], [Uen74], [Uen78], [Kaw82], [Kaw83],and [Kaw85], and was extended by [Kol87]. The sub-additivity of the Kodaira dimen-sion plays a central role in the classification of algebraic varieties, and over the yearsrefinements of the result and alternative approaches to its proof have stimulated avery active and interesting literature (cf. [Kaw02] and the more recent papers [Sch15],[P˘a16]). It is not our purpose here to survey this work but rather it is to focus onsome of the Hodge theoretic aspects of the result.One of these aspects is that the original proofs showed the necessity for understand-ing the singularities of the Chern form ω of the Hodge line bundle. Here [CKS86] isfundamental; as noted there by those authors their work was in part motivated byestablishing what was required for the proof of the Iitaka conjecture.In this particular discussion we shall mainly concentrate on the Hodge-theoreticaspect of the problem where everything is smooth. The issues that arise from the In fact, as will be discussed there M is of stratified-log-general type, a notion that is a refinementof log-general type. This means that at a general point the Kodaira-Spencer map ρ y : T y Y → H ( T X y ) is injective. ingularities may be dealt with using Theorem IV.B.8. In Section VI.B followinggeneral remarks on what is needed to establish the result we first observe that ifgeneric local Torelli holds for f : X → Y , then (I.A.27) is a direct consequence ofTheorem III.B.7 applied to the Hodge vector bundle.However, in general the assumption κ ( X y ) = dim X y pertains to the H ( K mX y ) for m (cid:29)
0, not to H ( K X y ) itself. So the question becomes How can Hodge theory be used to study the pluricanonical series H ( K mX y ) ? To address this an idea in [Kaw82] was expanded and used in [Vie83b] with furtherrefinement and amplification in [Kol87]. It is this aspect that we shall briefly discusshere with further details given in Section VI.C.First a general comment. Given a smooth variety W of dimension n and an ampleline bundle L → W , there are two variations of Hodge structure associated to thelinear systems | mL | for m (cid:29) D ∈ | mL | carry polarized Hodge structures;(b) for each s ∈ H ( L m ) with smooth divisor ( s ) = D ∈ | mL | there is a cycliccovering (cid:102) W s π −→ W branched over D and with a distinguished section ˜ s ∈ H ( (cid:102) W s , π ∗ L ) where ˜ s m = π ∗ s .We may informally think of (b) as the correspondence(I.A.28) ( W, s /m ) ←→ ( (cid:102) W s , ˜ s )obtained by extracting an m th root of s .We note that for each λ ∈ C ∗ there is an isomorphism(I.A.29) (cid:102) W s ∼ −→ (cid:102) W λs , so that we should think of (b) as giving the smooth fibres in a family { (cid:102) W [ s ] : [ s ] ∈ P H ( L m ) } . A variant of the construction (b) is then given by(c) (cid:102) W (cid:15) (cid:15) = desingularization of (cid:91) [ s ] ∈ P H ( L m ) (cid:102) W [ s ] P H ( L m ) . It is classical that a suitable form of local Torrelli for the end piece of the differentialof the period mapping holds for the families of each of the types (a) and (c) when This means that the end piece Φ ∗ ,n of the differential of the period mapping is genericallyinjective. Here, as will be explained below, we are referring to both the traditional use of the positivityof the Hodge line bundle, and also to the semi-positivity of the Hodge vector bundle which may beused in place of the concept of weak positivity introduced by Vieweg [Vie83a] and [Vie83b]. (cid:29) local Torelli holds for both families of types (a) and (c) when W alsovaries. In the case L = K W there is the special feature(I.A.31) H ( K mW ) is related to the H n, -term of the polarized Hodge structure on H n ( (cid:102) W ) arising in the construction (c). This means that, setting P = P H ( K mW ) H ( K mW ) is a summand of F ⊗ O P ( − where F → P is the Hodge vectorbundle associated to the family. As a consequence(I.A.32) H ( K mW ) ⊗ H ( O P (1)) is a summand of H ( F ) . There is also a metric Hodge theoretic interpretation of the pluricanonical series.For ψ ∈ H ( K mW ) the Narashimhan-Simka [NS68] Finsler-type norm(I.A.33) (cid:107) ψ (cid:107) = (cid:90) W ( ψ ∧ ψ ) /m has the Hodge-theoretic interpretation(I.A.34) (cid:107) ψ (cid:107) = (cid:90) (cid:102) W ψ (cid:101) ψ ∧ (cid:101) ψ where the RHS is a constant times the square of the Hodge length of (cid:101) ψ ∈ H ( K (cid:102) W ψ ).The curvature properties of (cid:107) ψ (cid:107) arising from its Hodge-theoretic interpretation wereused by Kawamata [Kaw82] in in his proof of the Iitaka conjecture when Y is a curve.The metric properties of direct images of the pluricanonical systems have recentlybeen an active and highly interesting subject; cf. [P˘a16] for a summary of some ofthis work and references to the literature (cf. also [Den18]). It is possible that growingout of this work the Kawamata argument could be extended to give a proof of thefull Iitaka conjecture. We will comment further on this at the end of Section VI.Cwhere a precise conjecture (VI.C.30) will be formulated.Returning to the discussion of (I.A.27), the essential point is to show that(I.A.35) f ∗ ω mX/Y has lots of sections for m (cid:29) . The idea is that “positivity = ⇒ sections” and “assumptions (i) and (ii) in (I.A.26) = ⇒ f ∗ ω mX/Y has positivity.” As noted above, when m = 1 and local Torelli holds usingTheorem III.B.7 there is sufficient positivity to achieve (I.A.35). The issue then be-comes how to use the Hodge-theoretic interpretation of the pluricanonical series andsemi-positivity of Hodge vector bundles to also produce sections of f ∗ ω mX/Y . f we globalize the cyclic covering construction taking W to be a typical X y wearrive at a commutative diagram that is essentially (cid:101) X ˜ f (cid:31) (cid:31) g (cid:15) (cid:15) π (cid:47) (cid:47) X f (cid:15) (cid:15) P p (cid:47) (cid:47) Y where the fibres are given by • P y = p − ( y ) = P H ( ω mX y ), • (cid:101) X y = ˜ f − ( y ), where • π : (cid:101) X y → X y is the family of cyclic coverings (cid:101) X y, [ ψ ] → X y as [ ψ ] ∈ P H ( ω mX y )varies. If it were possible to say that from (I.A.31)(I.A.36) f ∗ ω mX is a direct factor of ˜ f ∗ ω (cid:101) X /Y then local Torelli-type statements would imply positivity of f ∗ ω mX/Y . However, due tothe twisting that occurs as λ varies in the scaling identification (I.A.29) essentiallywhat happens is that(I.A.37) f ∗ ω mX/Y is a direct factor of ˜ f ∗ ω (cid:101) X / P ⊗ p ∗ O P (1) . Now O P (1) is positive along the fibres of P → Y , but the hoped for positivity of f ∗ ω mX/Y means that O P (1) tends to be negative in the directions normal to P y in P .Thus the issue has an additional subtlety that, as will be explained in Section VI.C,necessitates bringing in properties of the mapsSym k ˜ f ∗ ω m (cid:101) X / P → ˜ f ∗ ω km (cid:101) X / P . A final question that arises is this:(I.A.38)
Let f : X → Y be a map between smooth, projective varieties, andassume that the locus of y ∈ Y where X y = f − ( y ) is singular is areduced normal crossing divisor. Then (i) if det f ∗ ω mX/Y is non-zero, is it free? (ii) if f ∗ ω mX/Y is non-zero, if it free? We note that (i) for m = 1 follows from (I.A.1). So far as we know, (ii) is not knownfor m = 1.Turning to Section VI.D, from the description following I.A.14 of the Satake-Baily-Borel completion M of the image M of a period mapping, we may say that theextended Hodge line bundle does not detect extension data in limiting mixed Hodgestructures. One may ask whether the same is or is not the case for the extendedHodge vector bundle. In Section VI.C we give examples to show that in fact thisvector bundle may detect both discrete and continuous extension data in limitingmixed Hodge structures. See (VI.C.19) for the complete definition and explanation of this diagram. s noted above the Hodge line bundle lives on the canonical completion M of theKSBA moduli space for surfaces of general type. Interestingly, as will be seen byexample in a sequel to [GGLR17] the Hodge vector bundle does not live on M . Thisis in contrast to the case of curves where the Hodge vector bundle is defined on themoduli space M g .In the final section VI.E we shall revisit the exterior differential system (I.A.6)defined by a Chern form, this time for the Chern form ω E of the line bundle O P E (1) → P E . For a bundle E → X with Θ E (cid:61) ω E (cid:61) P E , the failure of strictpositivity, or the degree of flatness, of the bundle E is reflected by the foliation givenby the integral varieties of the exterior differential system ω E = 0 . The result here is Proposition (VI.E.2), and it suggests a conjecture giving conditionsunder which equality might hold in the inequality κ ( E ) (cid:53) n ( E )between the Kodaira-Iitaka dimension κ ( E ) and numerical dimension n ( E ) of thebundle.I.B. General notations. • X, Y, W, . . . will be compact, connected complex manifolds.In practice they will be smooth, projective varieties. • E → X is a holomorphic vector bundle with fibres E x , x ∈ X and rank r = dim E x ; • A p,q ( X, E ) denotes the global smooth E -valued ( p, q ) forms; • we will not distinguish between a bundle and its sheaf of holomorphic sections;the context should make the meaning clear; • L → X will be a line bundle.Associated to L → X are the standard notions(i) ϕ L : X (cid:47) (cid:47) P H ( X, L ) ∗ is the rational mapping given for x ∈ X by(I.B.1) ϕ L ( x ) = [ s ( x ) , . . . , s N ( x )]where s , . . . , s N is a basis for H ( X, L ); in terms of a local holomorphic trivial-ization of L → X the s i ( x ) are given by holomorphic functions which are usedto give the homogeneous coordinates on the right-hand side of (I.B.1);(ii) the line bundle L → X is big if one of the equivalent conditions • h ( X, L m ) = Cm d + · · · where C >
0, dim X = d and · · · are lower orderterms; • dim ϕ L ( X ) = dim X is satisfied;(iii) L → X is free if one of the equivalent conditions The term semi-ample is also used. for some m >
0, the evaluation maps(I.B.2) H ( X, L m ) → L mx are surjective for all x ∈ X ; • ϕ mL ( x ) is a morphism; i.e., for all x ∈ X some s i ( x ) (cid:54) = 0; • the linear system | mL | := P H ( X, L m ) ∗ is base point free for m (cid:29) x ∈ X , we say that L m → X is generically globally generated .(iv) L → X is nef if deg (cid:0) L (cid:12)(cid:12) C (cid:1) (cid:61) C ⊂ X ; here L (cid:12)(cid:12) C = L ⊗ O X O C isthe restriction of L to C ;(v) we will say that L → X is strictly nef if deg (cid:0) L (cid:12)(cid:12) C (cid:1) > C ⊂ X ; • for a vector bundle E → X , we denote the k th symmetric product by S k E := Sym k E ; • P E π −→ X is the projective bundle of 1-dimensional quotients of the fibresof E → X ; thus for x ∈ X ( P E ) x = P E ∗ x ; • O P E (1) → P E is the tautological line bundle; then π ∗ O P E ( m ) = S m E gives H ( X, S m E ) ∼ = H ( P E, O P E ( m ))for all m . I.C.
Notations from Hodge theory.
We shall follow the generally standard no-tations and conventions as given in [CMSP17] and are used in [GGLR17]. Furtherdetails concerning the structure of limiting mixed Hodge structures (LMHS) will begiven in Section IV. • B will denote a smooth quasi-projective variety; • a variation of Hodge structure (VHS) parametrized by B will be given by theequivalent data(a) a period mapping Φ : B → Γ \ D where D is the period domain of weight n polarized Hodge structures ( V, Q, F • )with fixed Hodge numbers h p,q , and where the infinitesimal period relation(IPR) Φ ∗ : T B → I ⊂ T (Γ \ D )is satisfied; The higher direct images R qπ O P E ( m ) = 0 for q > m (cid:61) − r ; we shall note make use of this inthese notes. b) ( V , F • , ∇ ; B ) where V → B is a local system with Gauss-Manin connection ∇ : O B ( V ) → Ω B ( V )and F • = { F n ⊂ F n − ⊂ · · · ⊂ F } is a filtration of O ( V ) by holomorphicsub-bundles satisfying the IPR in the form ∇ F p ⊂ Ω B ( F p − ) , and where at each point b ∈ B the data ( V b , F • b ) defines a polarized Hodgestructure (PHS) of weight n .In the background in both (a) and (b) is a bilinear form Q that polarizes the Hodgestructures; we shall suppress the notation for it when it is not being explicitlyused. • The parameter space B will have a smooth projective completion B with theproperties Z := B \ B is a reduced normal crossing divisor Z = (cid:83) Z i having strata Z I := (cid:84) i ∈ I Z i with Z ∗ I ⊂ Z I denoting the non-singular points Z ∗ I, reg of Z I ,and where the local monodromies T i around the irreducible branches Z i of Z are unipotent with logarithms N i ; • the Hodge vector bundle F → B has fibres F b := F nb ; • the Hodge line bundle
Λ := det F = ∧ h n, F ; • the polarizing forms induce Hermitian metrics in F and Λ; • the differential of the period mapping isΦ ∗ : T B → (cid:77) p (cid:61) [ n ] Hom( F p , F p − /F p )where F p → B denotes the Hodge filtration bundles; • setting F = F n and G = F n − /F n the end piece of Φ ∗ isΦ ∗ ,n : T B → Hom(
F, G ); • the Hodge filtration bundles have canonical extensions F pe → B ;using the Hodge metrics on B , the holomorphic sections of F pe → B are thosewhose Hodge norms have logarithmic growth along Z ; • the geometric case is when the VHS arises from the cohomology along the fibresin a smooth projective family X f −→ B ; • such a family has a completion to X ¯ f −→ B where this map has the Abramovich-Karu ([AK00]) form of semi-stable reduction(cf. Section 4 in [GGLR17] for more details and for the notations to be used here);then the canonical extension of the Hodge vector bundle is given by F e = ¯ f ∗ ω X /B . e conclude this introduction with an observation and a question. For line bundles L → X over a smooth projective variety X of dimension d , there are three importantproperties:(i) L if nef;(ii) L is big;(iii) L if free (and therefore it is semi-ample).Clearly (iii) = ⇒ (i) and (ii), and (iii) = ⇒ | mL | gives a birational morphism for m (cid:29) L → X has a metric h thatmay be singular, with Chern ω that defines a (1 ,
1) current representing c ( L ). Thesingularities of h are generally of the following types: • h vanishes along a proper subvariety of X ; • h becomes infinite, either logarithmically or analytically (in the sense explainedbelow) along a proper subvariety of X .We note that ω (cid:61) ⇒ (i) , and that in a Zariski open X ⊂ X where h is a smooth metric ω d > ⇒ (ii).The Kodaira theorem states that if X = X , then h is ample. As discussed above inrelation to the question (I.A.10), for a number of purposes, including applications toHodge theory, it would be desirable to have conditions on ω that imply (iii).II. Measures of and types of positivity
In this section we will(i) define two measures of positivity—the Kodaira-Iitaka dimension and the numer-ical dimension;(ii) define two types of positivity—metric positivity and numerical positivity.Each of these will be done first for line bundles and then for vector bundles, and thedefinitions will be related via the canonical association E → X (cid:32) O P E (1) → P E of the tautological line bundle O P E (1) → P E to the vector bundle E → X . For each of the types of positivity there will be two notions, strict positivity denoted > semi-positivity denoted (cid:61)
0. For metric positivity there will be a third,denoted by E > L > (cid:61) > strong semi-positivity to mean that E met (cid:61) E >
0. Finally, We will use the convention whereby P E is the bundle of 1-dimensional quotients of the fibresof E ; thus the fibre P E x = P ( E ∗ x ) . ince metric positivity will be the main concept used in these notes, in some of thelater sections we will just write E > E > E met > E > Kodaira-Iitaka dimension.
In algebraic geometry positivity traditionally sug-gests “sections,” and one standard measure of the amount of sections of a line bundle L → X is given by its Kodaira-Iitaka dimension κ ( L ). This is defined by(II.A.1) κ ( L ) = max m dim ϕ mL ( X )where ϕ mL : X (cid:47) (cid:47) P H ( mL ) ∗ is the rational mapping given by the linear system | mL | . If h ( mL ) = 0 for all m weset κ ( L ) = −∞ . From [Dem12a] we have(II.A.2) h ( mL ) (cid:53) O ( m κ ( L ) ) for m (cid:61) , and κ ( L ) is the smallest exponent for which this estimate holds. We will sometimeswrite (II.A.2) as h ( mL ) ∼ Cm κ ( L ) , C > . We note that κ ( L ) = dim X ⇐⇒ L → X is big.II.B. Metric positivity.
Given a Hermitian metric h in the fibres of a holomorphicvector bundle E → X there is a canonically associated Chern connection D : A ( X, E ) → A ( X, E )characterized by the properties ([Dem12a])(II.B.1) (cid:40) D (cid:48)(cid:48) = ∂d ( s, s (cid:48) ) = ( Ds, s (cid:48) ) + ( s, Ds (cid:48) )where s, s (cid:48) ∈ A ( X, E ) and ( , ) denotes the Hermitian inner product in E . Thecurvature Θ E := D is linear over the functions; hence it is pointwise an algebraic operator. Using (II.B.1)it is given by a curvature operator Θ E ∈ A , ( X, End E )which satisfies (Θ E e, e (cid:48) ) + ( e, Θ E e (cid:48) ) = 0where e, e (cid:48) ∈ E x . Relative to a local holomorphic frame { s α } , h = (cid:107) h α ¯ β (cid:107) is a Hermitianmatrix and the corresponding connection and curvature matrices are given by θ = h − ∂h Θ E = ∂ ( h − ∂h ) = (cid:13)(cid:13)(cid:13) (cid:88) α,β,i,j Θ α ¯ βi ¯ j s α ⊗ s ∗ β ⊗ dz i ∧ d ¯ z j (cid:13)(cid:13)(cid:13) . Including κ ( L ) = −∞ where we set m −∞ = 0 for m > or line bundles the connection and curvature matrices are respectively θ = ∂ log h and Θ L = − ∂∂ log h . If h = e − ϕ , thenΘ L (cid:61) ⇐⇒ ( i/ ∂∂ϕ (cid:61) ⇐⇒ ϕ is plurisubharmonic. Definition:
The curvature form is given for x ∈ X , e ∈ E x and ξ ∈ T x X by(II.B.2) Θ E ( e, ξ ) = (cid:10) (Θ E ( e ) , e ) , ξ ∧ ¯ ξ (cid:11) . When written out in terms of the curvature matrix Θ E ( e, ξ ) is the bi-quadratic form (cid:88) α,β,i,j Θ α ¯ βi ¯ j e α ¯ e β ξ i ¯ ξ j . The bundle E → X is positive , written E met >
0, if there exists a metric such thatΘ E ( e, ξ ) > e, ξ . For simplicity we will write Θ E >
0. If we havejust Θ E ( e, ξ ) (cid:61)
0, then we shall say that E → X is semi-positive and write E met (cid:61) strongly semi-positive if E met (cid:61) E ) met > Nakano positive if there exists a metric such that for all non-zero ψ ∈ E x ⊗ T x X we have(II.B.3) (Θ E ( ψ ) , ψ ) > . The difference between positivity and Nakano positivity is that the former involvesonly the decomposable tensors in E ⊗ T X whereas the latter involves all tensors. In[Dem12a] there is the concept of m -positivity that involves the curvature acting ontensors of rank m and which interpolates between the two notions defined above.Positivity and semi-positivity have functoriality properties ([Dem12a]). For ourpurposes the two most important arethe tensor product of positive bundles is positive, and similarly forsemi-positive;(II.B.4) the quotient of a positive bundle is positive, and similarly for semi-positive.(II.B.5)The second follows from an important formula that we now recall (cf. [Dem12b]). Ifwe have an exact sequence of holomorphic vector bundles(II.B.6) 0 → S → E → Q → , then a metric in E induces metrics in S, Q and there is a canonical second fundamentalform β ∈ A , ( X, Hom(
S, Q ))that measures the deviation from being holomorphic of the C ∞ splitting of (II.B.6)given by the metric. For j : Q (cid:44) → E the inclusion given by the C ∞ splitting and q ∈ Q x , ξ ∈ T x X the formula is (loc. cit.)(II.B.7) Θ Q ( q ⊗ ξ ) = Θ E ( j ( q ) ⊗ ξ ) + (cid:107) β ∗ ( q ) ⊗ ξ (cid:107) We should say metrically positive , but since this is the main type of positivity used in thesenotes we shall drop the “metrically.” here by definition the last term is − (cid:10) ( β ∗ ( q ) , β ∗ ( q )) S , ξ ∧ ¯ ξ (cid:11) and ( , ) S is the inducedmetric in S . The minus sign is because the Hermitian adjoint β ∗ is of type (0,1). Examples. (i) The universal quotient bundle Q → G ( k, n ) with fibres Q Λ = C n / Λ over theGrassmannian G ( k, n ) of k -planes Λ ⊂ C n has a metric induced by that in C n , andwith this metric Θ Q (cid:61) Q > ⇐⇒ k = n − . Similarly, the dual S ∗ → G ( k, n ) of the universal sub-bundle has Θ S ∗ (cid:61) S ∗ > ⇐⇒ k = 1.Geometrically, for a k -plane Λ ∈ C n we have the usual identification T Λ G ( k, n ) ∼ = Hom(Λ , C n / Λ) . Then for ξ ∈ Hom(Λ , C n / Λ) and v ∈ ΛΘ S ∗ ( v, ξ ) = 0 ⇐⇒ ξ ( v ) = 0 . Here the RHS means that for the infinitesimal displacement Λ ξ of Λ given by ξ wehave v ∈ Λ ∩ Λ ξ . The picture for G (2 ,
4) viewed as the space of lines in P isΛ Λ ξ v There are similar semi-positivity properties for any globally generated vector bun-dle, since such bundles are induced from holomorphic mappings to a Grassmannian,and positivity and semi-positivity have the obvious functoriality properties.(ii) The Hodge bundle F → B with the metric given by the Hodge-Riemann bilinearrelation satisfies Θ F (cid:61) h n, = 1 very seldom do we have Θ F > X π −→ B with smooth fibres, the Narashimhan-Simka [NS68] Finsler type metrics (I.A.33) in π ∗ ω m X /B have the Hodge theoretic inter-pretation (I.A.34). As a consequence There is a metric h m in O P f ∗ ω mX/Y (1) whose Chern form ω m (cid:61) . Some care must both be taken here as although h m is continuous it is not smoothand so ω m = ( i/ ∂∂ log h m and the inequality ω m (cid:61) I.C.
Interpretation of the curvature form.
Given a holomorphic vector bundle E → X there is the associated projective bundle P E π −→ X of 1-dimensional quotientsof the fibres of E ; thus ( P E ) x = P E ∗ x . Over P E there is the tautological line bundle O P E (1). A metric in E → X induces one in O P E (1) → P E , and we denote by ω E the corresponding curvature form. Then Ω E := ( i/ π ) ω E represents the Chern class c ( O P E (1)) in H ( P E ).Since ω E (cid:12)(cid:12) ( P E ) x is a positive (1 ,
1) form, the vertical sub-bundle V := ker π ∗ : T P E → T X to the fibration P E → X has a C ∞ horizontal complement H = ω ⊥ E . Thus as C ∞ bundles (cid:40) T P E ∼ = V ⊕ H, and π ∗ : H ∼ −→ π ∗ T X.
In more detail, using the metric we have a complex conjugate linear identification E ∗ x ∼ = E x , and using this we shall write points in P E as ( x, [ e ]) where e ∈ E x is anon-zero vector. Then we have an isomorphism(II.C.1) π ∗ : H ( x, [ e ]) ∼ −→ T x X. Using this identification and normalizing to have (cid:107) e (cid:107) = 1, the interpretation of thecurvature form is given by the equation(II.C.2) Θ E ( e, ξ ) = (cid:10) ω E , ξ ∧ ¯ ξ (cid:11) =: ω E ( ξ )where ξ ∈ T x X ∼ = H ( x, [ e ]) and the RHS is evaluated at ( x, [ e ]). Thus(II.C.3) Θ E > ⇐⇒ ω E > , and similarly for (cid:61)
0. There are the evident extensions of (II.C.3) to open sets in P E lying over open sets in X . For semi-positive vector bundles we summarize by sayingthat the curvature form Θ E measures the degree of positivity of ω E in the horizontaldirections .For later use we conclude with the observation that using O X ( E ) ∼ = π ∗ O P E (1), given s ∈ O X,x ( E ) there is the identification of (1,1) forms(II.C.4) ( − ∂∂ log (cid:107) s (cid:107) )( x ) = ω E ( x, [ s ( x )])where the RHS is the (1,1) form ω E evaluated at the point ( x, [ s ( x )]) ∈ P E in thetotal tangent space (both vertical and horizontal directions).II.D. Numerical positivity.
In this section we shall discuss various measures of numerical positivity, one mainpoint being that these will apply to bundles arising from Hodge theory. The basicreference here is [Laz04]. A conclusion will be that the Hodge vector bundle F isnumerically semi-positive; i.e., F num (cid:61) I.D.1.
Definition of numerical positivity.
We first recall the definition of the cone C = ⊕ C d of positive polynomials P ( c , . . . , c r ) where c i has weighted degree i . Forthis we consider partitions λ = ( λ , . . . , λ n ) with 0 (cid:53) λ n (cid:53) λ n − (cid:53) · · · (cid:53) λ (cid:53) r ,Σ λ i = n , of n = dim X . For each such λ the Schur polynomial s λ is defined by thedeterminant(II.D.1) s λ = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) c λ c λ +1 · · c λ + n − · ·· · c λ n − n +1 · · · c λ n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . Then ([Laz04]) C is generated over Q > by the s λ . It contains the Chern monomials c i · · · c i r r , i + 2 i + · · · + ri r (cid:53) n as well as some combinations of these with negative coefficients, the first of which is c − c .For each P ∈ C d and d -dimensional subvariety Y ⊂ X we consider(II.D.2) (cid:90) Y P ( c ( E ) , . . . , c r ( E )) = P ( c ( E ) , . . . , c r ( E ))[ Y ]where the RHS is the value of the cohomology class P ( c ( E ) , . . . , c r ( E )) ∈ H d ( X )on the fundamental class [ Y ] ∈ H d ( X ). Definition: E → X is numerically positive , written E num >
0, if (II.D.2) is positivefor all P ∈ C d and subvarieties Y ⊂ X .We may similarly define E num (cid:61) For line bundles L → X , we have L num (cid:61) ⇐⇒ L is nef. The essential content of this statement is c ( L )[ C ] (cid:61) for all curves C = ⇒ c ( L ) d [ Y ] (cid:61) for all d -dimensionalsubvarieties Y ⊂ X. This is frequently formulated as saying that if L is nef then it is in the closure of theample cone.II.D.2. Relation between E num > and O P E (1) num > . For the fibration P E π −→ X there is a Gysin or integration over the fibre map π ∗ : H d − r +1) ( P E ) → H d ( X ) . It is defined by moving cohomology to homology via Poincar´e duality, taking theinduced map on homology and then again using Poincar´e duality. In de Rham coho-mology the mapping is given by what the name suggests. The d th Segre polynomial is defined by(II.D.4) S d ( E ) = π ∗ (cid:16) c ( O P E (1)) d − r +1 (cid:17) . hen ([Laz04]): (i) S d ( E ) is a polynomial in the Chern classes c ( E ) , . . . , c r ( E ), and(ii) S d ( c , . . . , c r ) ∈ C d . That it is a polynomial in the Chern classes is a consequenceof the Grothendieck relation (II.D.5) c ( O P E (1)) r − c ( O P E (1)) r − π ∗ e ( E ) + · · · + ( − r π ∗ c r ( E ) = 0 . The first few Segre polynomials are S = c S = c − c S = c c S = c − c c + c c − c . An important implication is(II.D.6) O P E (1) num > ⇒ E num > . Proof.
By Nakai-Moishezon (cf. (II.G.4) below), O P E (1) and hence E are ample. Then E num > (cid:3) As will be noted below, the converse implication is not valid.II.E.
Numerical dimension.
Let L → X be a line bundle with L num (cid:61)
0; i.e., L isnef. Definition (cf. [Dem12a]) : The numerical dimension is the largest integer n ( L ) suchthat c ( L ) n ( L ) (cid:54) = 0 . In practice in these notes there will be a semi-positive (1,1) form ω such that [( i/ π ) ω ] = c ( L ), and then n ( L ) is the largest integer such that ω n ( L )+1 ≡ ω n ( L ) (cid:54) = 0on an open set.Relating the Kodaira-Iitaka and numerical dimensions, from [Dem12a] we have(II.E.1) κ ( L ) (cid:53) n ( L )with equality if n ( L ) = dim X , but where equality may not hold if n ( L ) < dim X . Example (loc. cit.) : Let C be an elliptic curve and p, q ∈ C points such that p − q is not a torsion point. Then the line bundle [ p − q ] has a flat unitary metric, but h ( C, m [ p − q ]) = 0 for all m >
0. For any nef line bundle L (cid:48) → X (cid:48) we set X = C × X (cid:48) , L = [ p − q ] (cid:2) L (cid:48) . Then κ ( X ) = −∞ while n ( L ) may be any integer with n ( L ) (cid:53) dim X − κ ( L ) < n ( L ) seems to involve some sort of flatness as in the aboveexample. For a vector bundle E → X with E num (cid:61) Cf. [CD17a] and [CD17b] for a related discussion involving certain Hodge bundles. efinition: The numerical dimension n ( E ) of the vector bundle E → X is given by n ( O P E (1)).Since O P E (1) is positive on the fibers of P E → X we have r − (cid:53) n ( E ) (cid:53) dim P E = dim X + r − . Conjecture VI.E.6 below suggests conditions under which equality will hold.From (II.D.4) we have(II.E.2) n ( E ) is the largest integer with S n ( E ) − r +1 ( E ) (cid:54) = 0 , which serves to define the numerical dimension of a semi-positive vector bundle interms of its Segre classes. We remark that as noted above(II.E.3) O P E (1) num (cid:61) ⇒ S q ( E ) (cid:61) q (cid:61) Tangent bundle.
When E = T X and the metric on E → X is given by aK¨ahler metric on X the curvature form has the interpretation(II.F.1) Θ T X ( ξ, η ) = holomorphic bi-sectional curvaturein the complex 2-plane ξ ∧ η spanned by ξ, η ∈ T x X . When ξ = η we have(II.F.2) Θ T X ( ξ, ξ ) = (cid:40) holomorphic sectional curvature inthe complex line spanned by ξ (cid:41) . Of particular interest and importance in Hodge theory and in other aspects of alge-braic geometry is the case when
T X has some form of negative curvature.
Proposition
II.F.3 ([BKT13]) : Assume there is c > such that (i) Θ T X ( ξ, ξ ) (cid:53) − c for all ξ , and (ii) Θ T X ( λ, η ) (cid:53) for all λ, η . Then there exists ξ, η such that Θ T X ( ξ, η ) (cid:53) − c/ for all η . In other words, if the holomorphic sectional curvatures are negative and the holo-morphic bi-sectional curvatures are non-positive, then they are negative on an openset in G (2 , T X ), the Grassmann bundle of 2-planes in T X , and this open set mapsonto X . Noting that Gr(2 , T X ) maps to an open subset of the horizontal sub-bundlein the fibration P T X → X , from (II.G.2) below we have the Corollary
II.F.4 : If the assumptions in II.F.3 are satisfied, then T ∗ X is big. II.G.
First implications.
In this section for easy reference we will summarize thefirst implications of the two types of positivity on the Kodaira-Iitaka dimension andnumerical dimension. These are either well known or easily inferred from what isknown. ase of a line bundle L → XL met > ⇒ L ample . (II.G.1)This is the Kodaira theorem which initiated the relation between metric positivityand sections.The next is the Grauert-Riemenschneider conjecture, established by Siu and De-mailly (cf. [Dem12a] and the references there):(II.G.2) L > ⇒ κ ( L ) = dim X. Here we recall that L > L → X whose curvatureform ω (cid:61) ω > L > ⇒ L is big.For these notes this variant of the Kodaira theorem will play a central role as bundlesconstructed from the extended Hodge vector bundle tend to be big and perhaps free, but just exactly what their “Proj” is seems to be an interesting issue. Because of thisfor later use, in the case when X is projective we now give a Proof of (II.G.2) . Let H → X be a very ample line bundle chosen so that H − K X isample. Setting F = L + H we have F met >
0. For D ∈ | F | smooth using the Kodairavanishing theorem we have (cid:40) h q ( X, mF ) = 0 , q > ,h q ( D, mF (cid:12)(cid:12) D ) = 0 , q > . We note that the vanishing theorems will remain true if we replace L by a positivemultiple.Let D, . . . , D m ∈ | H | be distinct smooth divisors. From the exact sequence 0 → m ( F − H ) → mF → m ⊕ j =1 mF (cid:12)(cid:12) D j we have0 → H ( X, m ( F − H )) → H ( X, mF ) → m ⊕ j =1 H ( D j , mF ) . This gives h ( X, mL ) = h ( X, m ( F − H )) (cid:61) h ( X, mF ) − mh ( D, mF ) . Using the above vanishing results h ( X, mL ) (cid:61) χ ( X, mF ) − mχ ( D, mF ) . The issue of additional conditions that will imply that an L which is nef and big is also free isa central one in birational geometry (cf. [Ko-Mo]). The results there seem to involve assumptionson mL − K X . In Hodge theory K X is frequently one of the things that one wishes to establishproperties of. or d = dim X and letting ∼ denote modulo lower order terms, from the Riemann-Roch theorem we have χ ( X, mF ) ∼ m d d ! F d mχ ( D, mF ) ∼ m d d ! ( dF d − · H ) . For m (cid:29) h ( X, mL ) (cid:61) m d d ! ( F d − dF d − · H ) + o ( m d ) . From F d − dF d − · H = ( L + H ) d − d ( L + H ) d − · H = ( L + H ) d − · ( L − ( d − H )replacing L by a multiple we may make this expression positive. (cid:3) The next result is(II.G.3) L met > ⇒ L num > , and similarly for (cid:61) . The inequality in (II.G.3) is sometimes phrased as L met (cid:61) ⇒ L is nef.The theorem of Nakai-Moishezon is L num > ⇐⇒ L is ample.(II.G.4)The next inequality was noted above: κ ( L ) (cid:53) n ( L ) , with equality if n ( L ) = dim X ;(II.G.5) Case of a vector bundle E → X (II.G.6) E met > ⇒ E ample, and E > ⇒ κ ( E ) = dim X + r − E > E is big.We next have(II.G.7) E met > ⇒ E num > . This result may be found in [Laz04]; the proof is not obvious from the definition. Weare not aware of any implication along the lines of E met (cid:61) ⇒ E num (cid:61) . Next we have(II.G.8) κ ( E ) (cid:53) n ( E ) with equality if n ( E ) = dim X + r − This argument is due to Catanese; cf. [Dem12a]. his leads to(II.G.9) E met (cid:61) E num > ⇒ κ ( E ) = dim X + r − . This follows from (II.E.2) and (II.G.8).It is not the case that E num > ⇒ E ample;there is an example due to Fulton of a numerically positive vector bundle over a curvethat is not ample (cf. [Laz04]).We will conclude this section with a sampling of well-known results whose proofsillustrate some of the traditional uses of positivity. Proposition
II.G.10 : If E → X is a Hermitian vector bundle with Θ E > , then H ( X, E ∗ ) = 0 .Proof. Using (II.C.4) applied to E ∗ → X , if s ∈ H ( X, E ∗ ) then when we evaluate ∂∂ log (cid:107) s (cid:107) at a strict maximum point where the Hessian is definite we obtain acontradiction. If the maximum is not strict then the usual perturbation argumentmay be used. (cid:3) Proposition
II.G.11 : If E → X is a Hermitian vector bundle of rank r (cid:53) dim X and with Θ E > , then every seciton s ∈ H ( X, E ) has a zero.Proof. The argument is similar to the preceding proposition, only this time we assumethat s has no zero and evaluate ∂∂ log (cid:107) s (cid:107) at a minimum. This (1,1) form is positivein the pullback to X of the vertical tangent space, and it is negative in the pullbackof the horizontal tangent space. The assumption r (cid:53) dim X then guarantees that ithas at least one negative eigenvalue. (cid:3) Proposition
II.G.12 : If E → X is a Hermitian vector bundle with Θ E (cid:61) and s ∈ H ( X, E ) satisfies Θ E ( s ) = 0 , then Ds = 0 .Proof. Let ω be a K¨ahler form on X . Then ∂∂ ( s, s ) = ( Ds, Ds ) + ( s, Θ E ( s )) = ( Ds, Ds ) (cid:61) , and if dim X = d using Stokes theorem we have0 = (cid:90) X ω d − ∧ ( i/ ∂∂ (cid:107) s (cid:107) = (cid:90) X ω d − ∧ ( i/ Ds, Ds )which gives the result. (cid:3)
II.H.
A further result.
As discussed above, for many purposes positivity is too strong (see the examplesin Section II.B) and semi-positivity is too weak (adding the trivial bundle to a semi-positive bundle gives one that is semi-positive). One desires a more subtle notionthan just Θ E (cid:61)
0. With this in mind, a specific guiding question for these notes hasbeen the uestion: Suppose that one has Θ E (cid:61) and for ∧ r E = det E we have Θ det E > on an open set; that is, E is strongly semi-positive. Does this enable one to producelots of sections of Sym m E for m (cid:29) ? The following is a response to this question:
Theorem
II.H.1 : Suppose that E → X is a Hermitian vector bundle of rank r thatis strongly semi-positive. Then Sym m E → X is big for any m (cid:61) r .Proof. Setting S r E = Sym r E , we have Θ S ∗ E (cid:61)
0. Let ω r be the curvature form for O P S r ( E ) (1). Then ω r (cid:61)
0, and we will show that(II.H.2) ω r > p = ( x, [ e · · · e r ]) ∈ P S r ( E ) x where (II.H.2)holds. Let x ∈ X be a point where (Tr Θ E )( x ) > e , . . . , e r be a unitarybasis for E x . Then some (cid:10) (Θ E ( e i ) , e i ) , ξ ∧ ξ (cid:11) > . We may assume that the Hermitian matrix (cid:10) (Θ E ( e i ) , e j ) , ξ ∧ ξ (cid:11) = δ ij λ i , λ i (cid:61) (cid:10) (Θ S r E ( e · · · e r ) , e · · · e r ) , ξ ∧ ξ (cid:11) = (cid:88) i (cid:68)(cid:16) Θ S r E ( e i ) e i ˆ · · · e r , e · · · e r (cid:17) , ξ ∧ ξ (cid:69) = (cid:88) i λ i > . The same argument works for any m (cid:61) r . (cid:3) Using this we are reduced to proving the
Lemma
II.H.3 : Let F → X be a Hermitian vector bundle with Θ F (cid:61) , and wherethis is a point x ∈ X and f ∈ F x such that the (1 , form Θ F ( f, · ) is positive. Thenfor the curvature form ω F of the line bundle O P F (1) → P F , we have ω F > at thepoint ( x, [ f ]) ∈ P F .Proof. Let f ∗ , . . . , f ∗ r be a local holomorphic frame for F ∗ → X and σ = u ([ a , . . . , a r ]; x ) (cid:88) i a i f ∗ i a local holomorphic section of O P E (1) → P F , where a , . . . , a r are variables definedmodulo the scaling action ( a , . . . , a r ) → λ ( a , . . . , a r ) and u ([ a , . . . , a r ] , x ) is holo-morphic. We have (cid:107) σ (cid:107) = | u | (cid:88) i,j h i ¯ j · a i ¯ a j where h i ¯ j = ( f ∗ i , f ∗ j ) is the metric in F → X ; up to a constant ω F = ∂∂ log (cid:107) σ (cid:107) . e may choose our frame and scaling parameter so that at the point ( x, [ f ])(II.H.4) h i ¯ j ( x ) = δ ij , dh ij ( x ) = 0 and (cid:107) σ ( x, [ f ]) (cid:107) = 1 . Computing ∂∂ log (cid:107) σ (cid:107) and evaluating at the point where (II.H.4) holds any cross-terms involving dh ij ( x ) drop out and we obtain ω F = (cid:88) ij ( ∂∂h i ¯ j )( x ) a i ¯ a j + (cid:32)(cid:88) i,j da i ∧ ¯ da i − (cid:16)(cid:88) a i · da i (cid:17) ∧ (cid:32)(cid:88) j a j da j (cid:33)(cid:33) . When we take the scaling action into account and use Cauchy-Schwarz it follows thatthat ω F > T ( x, [ f ]) P F . (cid:3) Remark that the point ( x, [ e · · · e r ]) corresponding to a decomposable tensor in S r E x is very special. Easy examples show that we do not expect to have ω r > ω r = 0is of interest and will be discussed in Section VI.E in the situation when the curvaturehas the norm positivity property to be introduced in Section III.A. Example II.H.5:
We will illustrate the mechanism of how passing to S r E increasesthe Kodaira-Iitaka dimension of the bundle. Let E → G (2 ,
4) denote the dual of theuniversal sub-bundle. As above, points of G (2 ,
4) will be denoted by Λ and thoughtof as lines in P . For v ∈ Λ we denote by [ v ] the corresponding line in C . Points of P E will be (Λ , v ) (cid:26)(cid:26)(cid:26)(cid:26)(cid:26) (cid:115) v Λand then the fibre of O P E (1) at (Λ , v ) is [ v ]. The fibre E Λ ∼ = Λ ∗ , and we have H ( G (2 , , E ) (cid:47) (cid:47) E Λ (cid:47) (cid:47) = = C ∗ (cid:47) (cid:47) Λ ∗ (cid:47) (cid:47) . The tangent space T Λ G (2 , ∼ = Hom(Λ , C / Λ)is isomorphic to the horizontal space H (Λ ,v ) ⊂ T (Λ ,v ) P E . As previously noted, for ξ ∈ T Λ G (2 ,
4) we have Θ E ( v, ξ ) = ω ( ξ ) = 0 ⇐⇒ ξ ( v ) = 0(II.H.6) (cid:26)(cid:26)(cid:26)(cid:26)(cid:26) (cid:115) v ΛΛ ξ Here Λ ξ is the infinitesimal displacement of Λ in the direction ξ . e observe that(II.H.7) ϕ O P E (1) : P E → P is the tautological map (Λ , v ) → [ v ], and consequently the fibre of (II.H.7) through v is the P of lines in P through v . The tangent space to this fibre are the ξ ’s aspictured above. We note that dim P E = 5 while κ ( O P E (1)) = n ( O P E (1)) = 3.Points of P S E are (Λ , v, v (cid:48) ) (cid:26)(cid:26)(cid:26)(cid:26)(cid:26) (cid:115) v (cid:115) v (cid:48) Λand unless v = v (cid:48) we have Θ E ( v · v (cid:48) , ξ ) (cid:54) = 0for any non-zero ξ ∈ T Λ G (2 , ω the curvature form of O P S E (1) we have ω > , v · v (cid:48) )unless v = v (cid:48) ; consequently S E is big.We shall give some further observations and remarks concerning the question posedat the beginning of this section. Proposition
II.H.8 : If E → X is generically globally generated and det E is big,then S k E is big for some k > .Proof. By standard arguments passing to a blowup of X and pulling E back, we mayreduce to the case where E is globally generated. Let N = h ( X, E ) and denote by Q → G ( N − r, N ) the universal quotient bundle over the Grassmannian. We thenhave a diagram(II.H.9) P E (cid:15) (cid:15) α (cid:47) (cid:47) P Q (cid:15) (cid:15) β (cid:47) (cid:47) P N − X f (cid:47) (cid:47) G ( N − r, N )where • f ∗ Q = E ; • β ◦ α = ϕ O P E (1)where ϕ P E (1) is the map induced by H ( O P E (1)). As a metric in E → X we use theone induced by the standard metric on Q → G ( N − r, N ). Then we claim that • Θ E (cid:61) • Tr Θ E > Q is an ample line bundle over G ( N − r, r )and det E = f ∗ det Q . Thus Θ det E = Tr Θ E (cid:61) d = dim X (Tr Θ E ) d ≡ ontradicts the assumption that det E is big. The proposition now follows fromTheorem II.H.1. (cid:3) This proposition will be used in connection with the following well known
Proposition
II.H.10 : If S k E is big for some k > , then there exist arbitrarily large (cid:96) such that S k(cid:96) E is generically globally generated.Proof. As a general comment, for any holomorphic vector bundle F → X (cid:40) F is genericallyglobally generated (cid:41) ⇐⇒ (cid:40) O P F (1) → P F is genericallyglobally generated (cid:41) . Since sufficiently high powers of a big line bundle are generically globally generated,and since by definition F → X is big if the line bundle O P F (1) is big, we have F big = ⇒ S n F → X is generically globally generated.Taking F = S k E , our assumption then implies that there are arbitrarily large (cid:96) suchthat S (cid:96) ( S k E ) is generically globally generated, consequently the direct summand S k(cid:96) E of S (cid:96) ( S k E ) is also generically globally generated. (cid:3) Corollary
II.H.11 : If E (cid:61) and det E > , then S m E is big for arbitrarily large m .Proof. By Theorem II.H.1, S m E is generically globally generated and det S m E > (cid:3)
Remark II.H.12:
For a line bundle L → X we consider the properties(i) L is nef ;(ii) L is big ;(iii) L is free (frequently expressed by saying that L is semi-ample ).For vector bundles E → X one has the corresponding properties using O (1). In(ii) and (iii) one generally uses Sym m P E E rather than just E itself.Although all these properties are important, in some sense (iii) is the most stable(note that (iii) = ⇒ (i)). Repeating a discussion from the introduction, suppose thatwe have in L → X a Hermitian metric with Chern form ω such that(II.H.13) ω (cid:61) ω > X \ Z where Z ⊂ X is a normal crossing divisor.Then ω defines a K¨ahler metric ω ∗ on X ∗ := X \ Z , and then as in (I.A.10) one mayask the(II.H.14) Question:
Are there properties of the metric ω ∗ , especially thoseinvolving its curvature R ω ∗ , that imply that L → X is free? III.
Norm positivity
III.A.
Definition and first properties.
As previously noted, for many purposes including those arising from Hodge theorystrict positivity of a holomorphic vector bundle in the sense of (II.B.2) and (II.B.3) istoo strong, whereas semi-positivity is too weak. The main observation of this sectionis that for many bundles that arise naturally in algebraic geometry the curvature has a pecial form, one that implies semi-positivity in both of the above senses and where inexamples the special form has Hodge-theoretic and algebro-geometric interpretations. Definition III.A.1:
Let E → X be a Hermitian vector bundle with curvature Θ E .Then Θ E has the norm positivity property if there is a Hermitian vector bundle G → X and a holomorphic bundle mapping(III.A.2) A : T X ⊗ E → G such that for x ∈ X and e ∈ E x , ξ ∈ T x X (III.A.3) Θ E ( e, ξ ) = (cid:107) A ( ξ ⊗ e ) (cid:107) G . Here we are identifying E x with E ∗ x using the metric, and (cid:107) (cid:107) G denotes the squarenorm in G . In matrix terms, relative to unitary frames in E and G there will be amatrix A of (1 ,
0) forms such that the curvature matrix is given by(III.A.4) Θ E = − t A ∧ A. We note that (III.A.3) will hold for any tensors in
T X ⊗ E , not just decomposableones. As a consequence E → X is semi-positive in both senses (II.B.2) and (II.B.3).The main implications of norm positivity will use the following observation:(III.A.5) If the curvatures of Hermitian bundles
E, E (cid:48) → X have the norm posi-tivity property, then the same is true for E ⊕ E (cid:48) → X and E ⊗ E (cid:48) → X ,as well as Hermitian direct summands of these bundles.Proof. If we have A : T X ⊗ E → G and A (cid:48) : T X ⊗ E (cid:48) → G (cid:48) , then Θ E ⊗ E (cid:48) =(Θ E ⊗ Id E (cid:48) ) ⊕ (Id E ⊗ Θ E (cid:48) ), and( A ⊗ Id E (cid:48) ) ⊕ (Id E ⊗ A (cid:48) ) : T X ⊗ E ⊗ E (cid:48) → ( G ⊗ E (cid:48) ) ⊕ ( E ⊗ G (cid:48) )leads to norm positivity for Θ E ⊗ E (cid:48) . The argument for ⊕ is evident. (cid:3) III.B.
A result using norm positivity.
The idea is this: For this discussion we abbreviate T = T x X, E = E x , G = G x and identify E ∼ = E ∗ using the metric. We have a linear mapping(III.B.1) A : T ⊗ E → G, and using (III.A.3) non-degeneracy properties of this mapping will imply positivityproperties of Θ E . Moreover, in examples the mapping A will have algebro-geometricmeaning so that algebro-geometric assumptions will lead to positivity properties ofthe curvature.The simplest non-degeneracy property of (III.B.1) is that A is injective; this seemsto not so frequently happen in practice. The next simplest is that A has injectivityproperties in each factor separately. Specifically we consider the two conditions(III.B.2) A : T → Hom(
E, G ) is injective; III.B.3) for general e ∈ E , the mapping A ( e ) : T → G given by A ( e )( ξ ) = A ( ξ ⊗ e ) , ξ ∈ T is injective.The geometric meanings of these are:(III.B.4) (III.B.2) is equivalent to having Θ det E = Tr Θ E > at x ; and(III.B.5) (III.B.3) is equivalent to having ω > at ( x, [ e ]) ∈ ( P E ) x . This gives the
Proposition
III.B.6 : If E → X has a metric whose curvature has the norm posi-tivity property, then (i) (II.B.2) = ⇒ det E is big; (ii) (III.B.3) = ⇒ E is big. A bit more subtle is the following result, which although it is a consequence ofTheorem II.H.1 and (III.B.3), for later use we shall give another proof.
Theorem
III.B.7 : If the rank r bundle E → X has a metric whose curvature hasthe norm positivity property, then (II.B.2) = ⇒ S r E is big. Corollary
III.B.8 : With the assumptions in (III.B.7) the map H ( X, S m ( S r E )) → S m ( S r E ) is generically surjective for m (cid:29) . Proof of Theorem III.B.7.
Keeping the above notations and working at a generalpoint in P E over x ∈ X , given ξ ∈ T x X and a basis e , . . . , e r of E x from (III.B.3) wehave(III.B.9) r (cid:88) i =1 (cid:107) A ( ξ ⊗ e i ) (cid:107) G (cid:54) = 0 . Then using (III.A.5) for the induced map A : T x X ⊗ S r E x → S r − E x ⊗ G x from (III.B.10) for ω r the canonical (1,1) form on P S r E at the point ( x, [ e · · · e r ]) (cid:10) ω r , ξ ∧ ¯ ξ (cid:11) > . (cid:3) Then this also implies that the map H ( X, S m ( S r E )) → S mr is generically surjective. emark: Viehweg ([Vie83a]) introduced the notion of weak positivity for a coherentsheaf. For vector bundles this means that for any ample line bundle L → X there isa k > H ( X, S (cid:96) ( S k E ⊗ L )) → S (cid:96) ( S k E ⊗ L )is generically surjective for (cid:96) (cid:29)
0. He then shows that for the particular bundles thatarise in the proof of the Iitaka conjecture if one has det
E > E is weakly positive. As willbe explained in Section V.A of these notes, (III.B.8) may be used to circumvent theneed for weak positivity in this case.We note that the ample line bundle L → X is not needed in (III.B.8). We alsopoint out that E met (cid:61) ⇒ E is weakly positive (cf. [P˘a16]).This is plausible since S k E (cid:61) L > ⇒ S k E ⊗ L >
0. In loc. cit. thisresult is extended to important situations where the metrics have certain types ofsingularities.We conclude this section with a discussion of the Chern forms of bundles havingthe norm positivity property, including the Hodge vector bundle.
Proposition
III.B.10 : The linear mapping A induces ∧ q A : ∧ q T → ∧ q G ⊗ S q E, and up to a universal constant c q (Θ) = (cid:107) ∧ q A (cid:107) . Proof.
The notation means (cid:107) ∧ q A (cid:107) = ( ∧ q A, ∧ q A )where in the inner product we use the Hermitian metrics in G and E , and we identify ∧ q T ∗ ⊗ ∧ q T ∗ ∼ = ( q, q )-part of ∧ q ( T ∗ ⊗ T ∗ ) . Then letting A ∗ denote the adjoint of A we have ∧ q Θ = ∧ q A ⊗ ∧ q A ∗ = ∧ q A ⊗ ( ∧ q A ) ∗ and c q (Θ) = Tr ∧ q (Θ) = ( ∧ q A, ∧ q A ) . (cid:3) In matrix terms, if A = dim G × dim T matrix with entries in E then ∧ q A = (cid:40) matrix whose entries are the q × q minors of A ,where the terms of E are multiplied as polynomials. (cid:41) t follows that up to a universal constant for the Hodge vector bundle Fc q (Θ F ) = (cid:88) a Ψ α ∧ Ψ α where the Ψ α are ( q,
0) forms. In particular, any monomial c I (Θ F ) (cid:61) ∧ q Φ ∗ ,n is not the same as rank Φ ∗ ,n < q . In fact,(III.B.11) rank Φ ∗ ,n < q ⇐⇒ c (Θ F ) q = 0 . In general we have the
Proposition
III.B.12 : If E → X has the norm positivity property, then P (Θ E ) (cid:61) for any P ∈ C . A proof of this appears in [Gri69].In the geometric case when we have a VHS arising from a family of smooth varietieswe have the period mapping Φ with the end piece of differential beingΦ ∗ ,n : T b B → Hom (cid:0) H (Ω nX b ) , H (Ω n − X b ) (cid:1) and the algebro-geometric interpretation of (III.B.11) is standard; e.g., Φ ∗ ,n injectiveis equivalent to local Torelli holding for the H n, -part of the Hodge structure.We conclude this subsection with a result that pertains to a question that wasraised above. Proposition
III.B.13 : If Φ : B → Γ \ D has no trivial factors, and if h n, (cid:53) dim B and H ( B, F e ) (cid:54) = 0 , then c h n, ( F ) (cid:54) = 0 . Proof.
We will first prove the result when B = B . We let s ∈ H ( F, B ) and assumethat c h n, ( F ) = 0. Then s is everywhere non-zero and we may go to a minimum of (cid:107) s (cid:107) . From Proposition II.G.12 we have Dσ = 0, which implies that the norm (cid:107) s (cid:107) isconstant and ∇ s = 0where ∇ is the Gauss-Manin connection. Using the arguments [Gri70] we may con-clude that the variation of Hodge structure has a trivial factor.If B (cid:54) = B , the arguments given in Section IV below may be adapted to show thatthe proof still goes through. The point is the equality of the distributional and formalderivatives that arise in integrating by parts. (cid:3) IV.
Singularities
In recent years the use of singular metrics and their curvatures in algebraic geometryhas become widespread and important. Here we shall discuss some aspects of thisdevelopment; the main objective is to define mild singularities and show that thesingularities that arise in Hodge theory have this property.One may roughly divide singularities into three classes: i) metrics with analytic singularities as defined in [Dem12a] and [P˘a16]; these arisein various extensions of the Kodaira vanishing theorem. We shall only brieflydiscuss these in part to draw a contrast with the next type of singularities thatwill play a central role in these notes;(ii) metrics with logarithmic singularities ; these arise in Hodge theory especially in[CKS86], and also in [Kol87], [Zuo00], [Bru16b] and [GGLR17], and for thosethat do arise in Hodge theory we shall show that they are mild as defined below;(iii) metrics with PDE singularities ; these arise in several places including the ap-plications to moduli where important classes of varieties have canonical specialmetrics. We shall not discuss these here but refer to the survey papers [Don16]and [Don15] and the Bourbaki talk [Dem16] for summaries of results and guidesto the literature.IV.A.
Analytic singularities.
For E → X a holomorphic vector bundle over acompact complex manifold, these singularities arise from metrics of the form(IV.A.1) h = e − ϕ h where h is an ordinary smooth metric in the bundle and ϕ is a weight function , whichin practice and will almost always locally be of the form(IV.A.2) ϕ = ϕ (cid:48) + ϕ (cid:48)(cid:48) where ϕ (cid:48) is plurisubharmonic (psh) and ϕ (cid:48)(cid:48) is smooth. Definition:
The metric (IV.A.1) has analytic singularities if it locally has the form(IV.A.2) with ϕ (cid:48) = α (cid:32) log m (cid:88) j =1 | f j ( z ) | α j (cid:33) , α, α j > f j ( z ) are holomorphic functions. Definition:
Given a weight function ϕ of the form (IV.A.2), we define the subsheaf I ( ϕ ) ⊂ O X by I ( ϕ ) = (cid:26) f ∈ O X : (cid:90) e − ϕ | f | < ∞ (cid:27) . Here the integral is taken over a relatively compact open set in the domain of definitionof f using any smooth volume form on X . Proposition
IV.A.3 (Nadel; cf. [Dem12a]) : I ( ϕ ) is a coherent sheaf of ideals. Example IV.A.4 ([Dem12a]) : For ϕ = log ( | z | α + · · · + | z n | α n ) , α j > I ( ϕ ) = (cid:40) z β · · · z β n n : (cid:88) j ( β j + 1) /α j > (cid:41) where the RHS is the ideal generated by the monomials appearing there. hus for β j = 1 and (cid:80) j α j = 2 + (cid:15)I ( ϕ ) = m is the maximal ideal at the origin in C n . The blow up using this ideal I ( ϕ ) is theusual blow up (cid:101) C n π −→ C n of C n at the origin.Variants of this construction give powers m k of the maximal ideal, and using z , . . . , z m for m (cid:53) n for suitable choices of the α j and β j leads to weighted blowupsof C n along the coordinate subspaces C n − m ⊂ C n . The use of weight functions withanalytic singularities provides a very flexible analytic alternative to the traditionaltechnique of blowing up along subvarieties. Example IV.A.5:
Quite different behavior occurs for the weight function ϕ = log (cid:0) ( − log | z | ) · · · ( − log | z k | ) (cid:1) . In this case I ( ϕ ) = O X . We will encounter weight functions of this type in Hodgetheory. Example IV.A.6: If L → X is a line bundle and s ∈ H ( X, L ) has divisor ( s ) = D ,we may define a metric (cid:107) (cid:107) in the line bundle by writing any local section s (cid:48) ∈ O X ( L )as s (cid:48) = f s where f is a meromorphic function and then setting (cid:107) s (cid:48) (cid:107) = | f | = e log | f | . By the Poincar´e-Lelong formula ([Dem12a]), formally ∂∂ log (cid:107) s (cid:48) (cid:107) = 0 but as currentsthe Chern form is the (1,1) current Ω L = [ D ]given by integration over the effective divisor D .We will be considering the case when E = L is a line bundle with singular metricsof the form (IV.A.1), (IV.A.2). In this case the curvature form is given by(IV.A.7) ω h = ( i/ ∂∂ log h = ( i/ ∂∂ϕ − ( i/ ∂∂ log h . For ϕ given by (IV.A.2) where ϕ (cid:48) is psh, the singular part of the curvature form isthe (1,1) current ( i/ ∂∂ϕ (cid:48) (cid:61) Theorem
IV.A.8 (Nadel vanishing theorem) : If ω h > in the sense of currents,then H q ( X, ( K X + L ) ⊗ I ( ϕ )) = 0 , q > . pplication: Assuming that ω L > ϕ in (IV.A.4), we may replace L by mL to make ( i/ ∂∂ϕ + ω K X + mω L > H ( X, mL ⊗ I x ) = 0 from which we inferthat H ( X, mL ) → mL x → . Thus mL is globally generated, and similar arguments show that for m (cid:29) ϕ mL : X → P N m is an embedding.This proof of the Kodaira embedding theorem illustrates the advantage of theflexibility provided by the choice of the weight function ϕ . Instead of blowing up asin the original Kodaira proof, the use of weight functions achieves the same effectwith greater flexibility.Another use of singular metrics is given by the Theorem
IV.A.9 (Kawamata-Viehweg vanishing theorem) : If L → X is big, then H q ( X, K X + L ) = 0 , q > . Proof (cf. [Dem12a]) . Let H → X be a very ample line bundle and D ∈ | H | a smoothdivisor. From the cohomology sequence of0 → mL − H → mL → mL (cid:12)(cid:12) D → h ( X, mL ) ∼ m d , h ( D, mL (cid:12)(cid:12) D ) ∼ m d − where d = dim X we have h ( mL − H ) (cid:54) = 0 for m (cid:29)
0. If E ∈ | mL − H | fromExample IV.A.6 there exists a singular metric in mL − H with Chern formΩ mL − H = [ E ] (cid:61) E . If ω H > H → X , then ω L = 1 m ([ E ] + ω H ) (cid:61) (cid:18) m (cid:19) ω H > , and Nadel vanishing gives the result. (cid:3) Remark:
A standard algebro-geometric proof of Kawamata-Viehweg vanishing usesthe branched covering method to reduce it to Kodaira vanishing. The above argumentagain illustrates the flexibility gained by the use of singular weight functions (we notethat D plays a role similar to that of the branch divisor in the branched coveringmethod). V.B.
Logarithmic and mild singularities.
As our main applications will be to Hodge theory, in this section we will use thenotations from Section I.C. We recall from that section that • B is a smooth quasi-projective variety; • B is a smooth projective completion of B ; • Z = B \ B is a divisor with normal crossings Z = ∪ Z i where Z I := (cid:84) i ∈ I Z i is a stratum of Z and Z ∗ I = Z I, reg are the smooth points of Z I ; • E → B is a holomorphic vector bundle.A neighborhood U in B of a point p ∈ Z will be U ∼ = ∆ ∗ k × ∆ (cid:96) with coordinates ( t, w ) = ( t , . . . , t k ; w , . . . , w (cid:96) ).We now introduce the co-frame in terms of which we shall express the curvatureforms in U . The Poincar´e metric in ∆ ∗ = { < | t | < } is given by the (1,1) form ω PM = ( i/ dt ∧ d ¯ t | t | ( − log | t | ) . We are writing − log | t | instead of just log | t | because we will want to have positivequantities in the computations below. As a check on signs and constants we note theformula(IV.B.1) ( i/ ∂∂ (cid:0) − log( − log | t | ) (cid:1) = (1 / ω PM . The inner minus sign is to have − log | t | > − log | t | ) is defined. Theouter one is to have the expression in parentheses equal to −∞ at t = 0 so that wehave a psh function. For ϕ = log( − log | t | )the curvature form in the trivial bundle over ∆ with the singular metric given by e − ϕ has curvature form(IV.B.2) ( i/ ∂∂ log( e − ϕ ) = (1 / ω PM . Remark:
The functions that appear as coefficients in formally computing (IV.B.1)using the rules of calculus are all in L and therefore define distributions. We maythen compute ∂ and ∂ either in the sense of currents or formally using the rules ofcalculus. An important observation is(IV.B.3) these two methods of computing ∂∂ϕ give the same result. This is in contrast with the situation when we take ϕ = log | t | in which case we have in the sense of currents the Poincar´e-Lelong formula(IV.B.4) ( i/π ) ∂∂ log | t | = δ here δ is the Dirac δ -function at the origin. Anticipating the discussion below,a charateristic feature of the metrics that arise in Hodge theory will be that theprinciple (IV.B.3) will hold. Definition:
The
Poincar´e coframe has as basis the (1,0) forms dt i t i ( − log | t i | ) , dw α and their conjugates. Definition:
A metric in the holomorphic vector bundle E → B is said to have logarithmic singularities along the divisor Z = B \ B if locally in an open set U asabove and in terms of a holomorphic frame for the bundles and the Poincar´e coframethe metric h , the connection matrix θ = h − ∂h , and the curvature matrix Θ E = ∂ ( h − ∂h ) have entries that are Laurent polynomials in the log | t i | with coefficientsthat are real analytic functions in U . Proposition
IV.B.5 : The Hodge metrics in the Hodge bundles F p → B have loga-rithmic singularities relative to the canonically extended Hodge bundles F pe → B . In the geometric case this result may be inferred from the theorem on regularsingular points of the Gauss-Manin connection ([Del70]) and (VI.C.1) above. Inthe general case it is a consequence of the several variable nilpotent orbit theorem([CKS86]). More subtle is the behavior of the coefficients of the various quantities,especially the Chern polynomials P (Θ F p ), when they are expressed in terms of thePoincar´e frame, a topic analyzed in [CKS86] and where the analysis is refined in[Kol87], and to which we now turn.We recall that a distribution Ψ on a manifold M has a singular support Ψ sing ⊂ M defined by the property that on any open set W ⊂ M \ Ψ sing in the complement therestriction Ψ (cid:12)(cid:12) W is given by a smooth volume form. A finer invariant of the singularitiesof Ψ is given by its wave front set W F (Ψ) ⊂ T ∗ M. Among other things the wave front set was introduced to help deal with two classicalproblems concerning distributions:(IV.B.6) (a) distributions cannot in general be multiplied;(b) in general distributions cannot be restricted to submanifolds N ⊂ M .For (a) the wave front sets should be transverse, and for (b) to define Ψ (cid:12)(cid:12) N it sufficesto have T N ⊂ W F (Ψ) ⊥ .In the case of currents represented as differential forms with distribution coeffi-cients, multiplication should be expressed in terms of the usual wedge product offorms. For restriction, if N is locally given by f = · · · = f m = 0, then for a currentΨ we first set df i = 0; i.e., we cross out any terms with a df i . Then the issue is to A good discussion of wave front sets and references to the literature is given in Wikipedia.We will not use them in a technical sense but rather as a suggestion of an important aspect to beanalyzed for the Chern polynomials of the Hodge bundles. estrict the distribution coefficients of the remaining terms to N . Thus the notion ofthe wave front set for a current Ψ involves both the differential form terms appearingin Ψ as well as the distribution coefficients of those terms. Definition:
The holomorphic bundle E → B has mild logarithmic singularities incase it has logarithmic singularities and the following conditions are satisfied:(i) the Chern polynomials P (Θ E ) are closed currents given by differential formswith L coefficients and which represent P ( c ( E ) , . . . , c r ( E )) in H ∗ DR ( B );(ii) the products P (Θ E ) · P (cid:48) (Θ E ) may be defined by formally multiplying them as L -valued differential forms, and when this is done we obtain a representativein cohomology of the products of the polynomials in the Chern classes;(iii) the restrictions P (Θ E ) (cid:12)(cid:12) Z ∗ I are defined and represent P (cid:16) c (cid:16) E (cid:12)(cid:12) Z ∗ I (cid:17) , . . . , c r (cid:0) E Z ∗ I (cid:1)(cid:17) .We note the opposite aspects of analytic singularities and mild logarithmic singu-larities: In the former one wants the singularities to create behavior different fromthat of smooth metrics, either with regard to the functions that are in L with respectto the singular metric, or to create non-zero Lelong numbers in the currents that arisefrom their curvatures. In the case of mild logarithmic singularities, basically one maywork with them as if there were no singularities at all. An important additional pointto be explained in more detail below is that the presence of singularities increases the positivity of the Chern forms of the Hodge bundles, so that in this sense one usessingularities to positive effect.The main result, stated below and which will be discussed in the next section, isthat the Hodge bundles have mild logarithmic singularities. This would follow if onecould show that(IV.B.7) When expressed in terms of the Poincar´e frame the polynomials P (Θ E ) have bounded coefficients. This is true when Z is a smooth divisor, but when Z is not a smooth divisor this isnot the case and the issue is more subtle. Main result.
Theorem
IV.B.8 ([CKS86], with amplifications in [Kol87], [GGLR17]) : The Hodgebundles have mild singularities.
The general issues (IV.B.6)(a), (IV.B.6)(b) concerning distributions were raisedabove. Since currents are differential forms with distribution coefficients, these issuesare also present for currents, where as noted above the restriction issue (IV.B.6)(b)involves both the differential form aspect and the distribution aspect of currents.This is part (iii) of the definition and is the property of the Chern polynomials thatappears in [GGLR17]. he proof of (i) and part of (ii) in Theorem IV.B.8 is based on the fundamentalresults in [CKS86], with refinements in [Kol87] concerning a particular multiplica-tive property (IV.B.6)(a) in the definition of mild logarithmic singularities, andin [GGLR17] the general multiplicative property and the restriction property of theChern polynomials is addressed. Both of these involve estimates in the ∆ ∗ -factorsin neighborhoods U ∼ = ∆ ∗ k × ∆ (cid:96) in B . In effect these estimates may be intrinsicallythought of as occurring in sectors in the co-normal bundle of the singular support ofthe Chern forms, and in this sense may be thought of as dealing with the wave frontsets of the these forms.A complete proof of Theorem IV.B.8 is given in Section 5 of [GGLR17]. In the nextsection we shall give the argument for the Chern form Ω = c (Θ det F ) of the Hodgeline bundle and in the special case when the localized VHS is a nilpotent orbit. Thecomputation will be explicit; the intent is to provide a perspective on some of thebackground subtleties in the general argument, one of which we now explain.We restrict to the case when U ∼ = ∆ ∗ k = { ( t , . . . , t k ) : 0 < | t j | < } , and setting (cid:96) ( t j ) = log t j / πi and x j = − log | t j | consider a nilpotent orbitΦ( t ) = exp (cid:32) k (cid:88) j =1 (cid:96) ( t j ) N j (cid:33) · F . Following explicit computations of the Chern form Ω and of the Chern form Ω I for therestriction of the Hodge line bundle to Z ∗ I , the desired result comes down to showingthat a limit(IV.B.9) lim x j →∞ Q ( x ) P ( x )exists where Q ( x ) , P ( x ) are particular homogeneous polynomials of the same degreewith P ( x ) > x j >
0. Limits such as (IV.B.9) certainly do not exist in general,and the issue to be understood is how in the case at hand the very special propertiesof several parameter limiting mixed Hodge structures imply the existence of the limit.As an application of Theorems III.B.7 and IV.B.8, using the notations from SectionI.C we consider a VHS given by a period mappingΦ : B → Γ \ D. Denoting by F e → B the canonically extended Hodge vector bundle we have Theorem
IV.B.10 : The specific result in [Kol87] is that the integral (cid:90) B c (Θ det E ) d < ∞ , dim B = d of the top power of the Chern form of the Hodge line bundle is finite. This result also follows fromthe analysis in Section V of [GGLR17]. i) The Kodaira-Iitaka dimension κ ( F e ) (cid:53) h n, − . (ii) Assuming the injectivity of the end piece Φ ∗ ,n of the differential of Φ , κ ( S h n, F e ) = dim P S h n, F e ; i.e., S h n, F e → B is big.Proof. It is well known [Gri70], [CMSP17] that the curvature of the Hodge vectorbundle has the norm positivity property. In fact, the curvature form is given by(IV.B.11) Θ F ( e, ξ ) = (cid:107) Φ ∗ ,n ( ξ )( e ) (cid:107) . Concerning the singularities that arise along B \ B , it follows from Theorem IV.B.8that we may treat the Chern form ω of O P F e (1) → B as if the singularities were notpresent.The linear algebra situation is(IV.B.12) T ⊗ F → G where dim T = dim B , dim F = h n, and dim G = h n − , . By (IV.B.11) condition(III.B.2) is equivalent to the injectivity of Φ ∗ ,n , and Theorem IV.B.10 is then a con-sequence of Theorem III.B.7. (cid:3) This result gives one answer to the question
The Hodge vector bundle is somewhat positive. Just how positive is it?
Since in the geometric case the linear algebra underlying the map (IV.B.12) is ex-pressed cohomologically, in particular cases the result (i) in (IV.B.10) can be consid-erably sharpened. For example, in the weight n = 1 case the method of proof of thetheorem gives the Proposition
IV.B.13 ([Bru16a]) : In weight n = 1 , (i) κ ( F e ) (cid:53) g − , and (ii) S F e → B is big.Proof. In this case D ⊂ H g where g = h , and H g is the Siegel generalized upper-half-plane. We then have • T ⊂ S V ∗ ; • G = V ∗ ; • T ⊗ V → G is induced by the natural contraction map S V ∗ ⊗ V → V ∗ .For any v ∈ V the last map has image of dimension (cid:53) g , and therefore the kernel hasdimension (cid:61) dim T − g . This gives (i) in the proposition.For (ii) we have T ⊗ S V (cid:47) (cid:47) V ∗ ⊗ V ∩ S V ∗ ⊗ S V. (cid:99) (cid:53) (cid:53) For a general q ∈ S V the contraction mapping (cid:99) is injective, and this implies (ii). (cid:3) At the other extreme we have the roposition IV.B.14 : Let M d,n denote the moduli space of smooth hypersurfaces Y ⊂ P n +1 of degree d = 2 n + 4 , n (cid:61) . Then the Hodge vector bundle F → M d,n isbig.Proof. Set V = C n +2 ∗ and let P ∈ V ( d ) be a homogeneous form of degree d thatdefines Y . Denote by J P ⊂ ⊕ k (cid:61) d − V ( k ) the Jacobian ideal. Then (cf. Section 5 in[CMSP17]) • T Y M d,n ∼ = V ( d ) /J ( d ) P ; • F Y = H n, ( Y ) ∼ = V ( d − n − ; • G Y = H n − , ( Y ) ∼ = V (2 d − n − /J (2 d − n − P .It will suffice to show(IV.B.15) For general P and general Q ∈ V ( d − n − the mapping V ( d ) /J ( d ) P Q −→ V (2 d − n − /J (2 d − n − P is injective. Noting that d − n − n + 2 and that it will suffice to prove the statement for one P and Q , we take Q = x · · · x n +1 ,P = x d + · · · + x dn +1 . Then J P = { x n +30 , . . . , x n +3 n +1 } and a combinatorial argument gives (IV.B.15). (cid:3) Remark IV.B.16:
The general principle that Proposition IV.B.14 illustrates is this:Let L → be an ample line bundle. Then both for general smooth sections Y ∈ | mL | and for cyclic coverings (cid:101) X Y → X branched over a smooth Y , as m increases theHodge vector bundle F → | mL | over the open set of smooth Y ’s becomes increasinglypositive in the sense that the k such that S k F is big decreases, and for m (cid:29) F itself is big. V. Proof of Theorem IV.B.8
V.A.
Reformulation of the result.
We consider a variation of Hodge structuregiven by a period mapping Φ : ∆ ∗ k → Γ loc \ D. Here we assume that the monodromy generators T i ∈ Aut Q ( V ) are unipotent withlogarithms N i ∈ End Q ( V ); Γ loc is the local monodromy group generated by the T i .For I ⊂ { , · · · , k } with complement I c = { , . . . , k }\ I we set∆ ∗ I = { ( t , . . . , t k ) : t i = 0 for i ∈ I and t j (cid:54) = 0 for j ∈ I c } . From the work of Cattani-Kaplan-Schmid [CKS86] the limit lim t → ∆ ∗ I Φ( t ) is definedas a polarized variation of limiting mixed Hodge structures on ∆ ∗ I . Passing to theprimitive parts of the associated graded polarized Hodge structures gives a periodmapping Φ I : ∆ ∗ I → Γ loc ,I \ D I here D I is a product of period domains and Γ loc ,I is generated by the T j for j ∈ I c .This may be suggestively expressed by writtinglim t → t I Φ( t ) = Φ I ( t I ) . However caution must be taken in interpreting the limit, as the “rate of convergence”is not uniform but depends on the sector in which the limit is taken in the mannerexplained in [CKS86].We denote by Λ → ∆ ∗ k and Λ I → ∆ ∗ I the Hodge line bundles. The Hodge-Riemann bilinear relations give metrics in these bundles and we denote by Ω and Ω I the respective Chern forms. The result to be proved is(V.A.1) lim t → ∆ I Ω = Ω I , where again care must be taken in interpreting this equation. In more detail, thismeans: In Ω set dt i = d ¯ t i = 0 for i ∈ I . Then the limit, in the usual sense, as t → ∆ I of the remaining terms exists and is equal to Ω I . We will write (V.A.1) as(V.A.2) Ω (cid:12)(cid:12) ∆ ∗ I = Ω I . The proof of (V.A.1) that we shall give can easily be adapted to the case when theperiod mapping depends on parameters.The limit can also be reduced to the case when Φ is a nilpotent orbit . This meansthat(V.A.3) Φ( t ) = exp (cid:32) k (cid:88) i =1 (cid:96) ( t i ) N i (cid:33) F where F ∈ ˇ D and the conditions(i) N : F p → F p − ,(ii) Φ( t ) ∈ D for 0 < | t | < (cid:15) are satisfied. This reduction is non-trivial and is given in Section 5 of [GGLR17].The main points in the proof of Theorem IV.B.8 in the nilpotent orbit case are asfollows:(a) without changing the associated graded’s to Φ and Φ I we may replace the F in(V.A.3) by an F such that the limiting mixed Hodge structure is R -split;(b) in this case N I can be completed to an sl which we denote by { N + I , Y I , N I } ; (c) the Y I -weight decomposition of N I c is N I c = N I c , + N I c , − + N I c , − + · · · where N I c , − m has Y I -weight − m , m (cid:61) N I c , − m = 0 for m >
0, then there is an sl c = { N + I c , Y I c , N I c } thatcommutes with the previous sl , and the result (V.A.1) is immediate; There are two ways of doing this—one is the method in [CKS86] and the other one, which ispurely linear algebra, is due to Deligne. e) in general, by direct computation we haveΩ I ≡ Ω + R mod dt i , d ¯ t i for i ∈ I where the remainder term R consists of expressions Q ( x ) /P ( x ) as in (IV.B.9),and then direct computation using the relative filtration property and the factthat for m > N I c , − m have negative Y I -weights gives the result.V.B. Weight filtrations, representations of sl and limiting mixed Hodgestructures. The proof of (V.A.1) will be computational, using only that Φ( t ) is a nilpotent orbit(V.A.3) and that the commuting N i ∈ End Q ( V ) have the relative weight filtrationproperty (RWFP), which will be reviewed below. The computation will be facilitatedby using the representation theory of sl adapted to the Hodge theoretic situation athand. The non-standard but hopefully suggestive notations for doing this will nowbe explained.(i) Given a nilpotent transformation N ∈ End Q ( V ) with N n +1 = 0 there is a uniqueincreasing weight filtration W ( N ) given by subspaces(V.B.1) V W ( N ) k := W k ( N ) V satisfying the conditions • N : V W ( N ) k → V W ( N ) k − , • N k : V W ( N ) n + k ∼ −→ V W ( N ) n − k (Hard Lefschetz property).Remark that the two standard choices for the ranges of indices in (V.B.1) are (cid:40) (cid:53) k (cid:53) n (Hodge theoretic) − n (cid:53) k (cid:53) n (representation theoretic) . We will use the first of these.The weight filtration is self-dual in the sense that using the bilinear form Q (V.B.2) V W ( N ) ⊥ k = V W ( N )2 n − k − which gives V W ( N ) ∗ k ∼ = V /V W ( N )2 n − k − . The associated graded to the weight filtation is the direct sum of theGr W ( N ) (cid:96) V := V W ( N ) (cid:96) /V W ( N ) (cid:96) − , and the primitive subspaces are defined for (cid:96) (cid:61) n byGr W ( N ) n + k, prim V = ker (cid:110) N k +1 : Gr W ( N ) n + k V → Gr W ( N ) n − k − V (cid:111) . The proof in [GGLR17] uses the detailed analysis of limiting mixed Hodge structures from[CKS86], of which the RWFP is one consequence. Part of the point for the argument given here isto isolate the central role played by that property. ii) A grading element for W ( N ) is given by a semi-simple Y ∈ End Q ( V ) withintegral eigenvalues 0 , , . . . , n , weight spaces V k ⊂ V W ( N ) k for the eigenvalue k , andwhere the induced maps V k ∼ −→ Gr W ( N ) k V are isomorphisms. Thus V W ( N ) k = k ⊕ (cid:96) =0 V (cid:96) . Grading elements always exists, and for any one such Y we have • [ Y, N ] = − N ; • there is a unique N + ∈ End Q ( V ) such that { N + , Y, N } is an sl -triple.The proof of the second of these uses the first together with the Hard Lefschetzproperty of W ( N ).We denote by U the standard representation of sl with weights 0 , ,
2. Thinkingof U as degree 2 homogeneous polynomials in x, y we have • weight x a y b = 2 a , a + b = 2; • N = ∂ x and N + = ∂ y .We denote by U i = Sym i U ∼ = (cid:40) homogeneous polynomialsin x, y of degree i + 1 (cid:41) the standard ( i + 1)-dimensional irreducible representation of sl . The N - string as-sociated to U i is { x i +1 } → { x i y } → · · · → { y i +1 } where N = ∂ x . The top of the N -string is the primitive space.Given ( V, Q, N ) as above and a choice of a grading element Y , for the sl -module V gr = n ⊕ k =0 V k we have a unique identification(V.B.3) V gr ∼ = n ⊕ i =0 H n − i ⊗ U i for vector spaces H n − i . The notation is chosen for Hodge-theoretic purposes. The N -string associated to H n − i ⊗ U i will be denoted by(V.B.4) H n − i ( − i ) N −→ H n − i ( − ( i − N −→ · · · N −→ H n − i and we define(V.B.5) the Hodge-theoretic weight of H n − i ( − j ) is n − i + 2 j . The representation-theoretic weight of H n − i ( − j ) is 2 j . It follows that H n − i ( − i ) isthe primitive part of the U i -component of V gr .Relative to Q the decomposition (V.B.3) is orthogonal and the pairing Q i : H n − i ( − i ) ⊗ H n − i ( − i ) → Q iven by(V.B.6) Q i ( u, v ) = Q ( N i u, v )is non-degenerate.(iii) We recall the Definition: A limiting mixed Hodge structure (LMHS) is a mixed Hodge structure( V, Q, W ( N ) , F ) with weight filtration W ( N ) defined by a nilpotent N ∈ End Q ( V )and Hodge filtration F which satisfies the conditions(a) N : F p → F p − ;(b) the form Q i in (V.B.6) polarizes Gr W ( N ) n + k, prim V ∼ = H n − k ( − k ).The MHS on V induces one on End Q ( V ), and (a) is equivalent to N ∈ F − End Q ( V ) . We denote by V C = ⊕ p,q I p,q the unique Deligne decomposition of V C that satisfies • W k ( N ) V = ⊕ p + q (cid:53) k I p,q ; • F p V = ⊕ p (cid:48) (cid:61) pq I p (cid:48) ,q ; • I q,p ≡ I p,q mod W p + q − ( N ) V .The LMHS is R - split if I p,q = I q,p . Canonically associated to a LMHS is an R -splitone ( V, Q, W ( N ) , F ) where F = e − δ F for a canonical δ ∈ I − , − End Q ( V R ). For this R -split LMHS there is an evidentgrading element Y ∈ I , (End Q ( V R )).Given a LMHS ( V, Q, W ( N ) , F ) there is an associated nilpotent orbit∆ ∗ (cid:47) (cid:47) Γ T \ D ∈ ∈ t (cid:47) (cid:47) exp( (cid:96) ( t ) N ) F where (cid:96) ( t ) = log t/ πi and Γ T = { T Z } . Conversely, given a 1-variable nilpotentorbit as described above there is a LMHS. We shall use consistently the bijectivecorrespondence LMHS’s ⇐⇒ . Since det F = det F without loss of generality for the purposes of this paper we will assume that ourLMHS’s are R -split and therefore have canonical grading elements. iv) Let N , N ∈ End Q ( V ) be commuting nilpotent transformations and set N = N + N . Then there are two generally different filtrations defined on the vector spaceGr W ( N ) V :(A) the weight filtration W ( N ) V induces a filtration on any sub-quotient spaceof V , and hence induces a filtration on Gr W ( N ) • V ;(B) N induces a nilpotent map N : Gr W ( N ) • V → Gr W ( N ) • V , and consequentlythere is an associated weight filtration W ( N ) Gr W ( N ) • V on Gr W ( N ) • V . Definition:
The relative weight filtration property (RWFP) is that these two filtra-tions coincide:(V.B.7) W ( N ) ∩ Gr W ( N ) • V = W ( N ) Gr W ( N ) • V. We note that N is the same as the map induced by N on Gr W ( N ) • V , so that (V.B.7)may be perhaps more suggestively written as(V.B.8) W ( N ) ∩ Gr W ( N ) • V = W ( N ) Gr W ( N ) • V. The RWFP is a highly non-generic condition on a pair of commuting nilpotent trans-formation, one that will be satisfied in our Hodge-theoretic context.(v) Suppose now that Y is a grading element for N so that the correspondingsl = { N +1 , Y , N } acts on V and hence on End Q ( V ). We observe that the Y -eigenspace deomposition of N is of the form(V.B.9) N = N , + N , − + · · · + N , − m , m > Y, N , − m ] = − mN , − m . The reason for this is that[ N , N ] = 0 = ⇒ (cid:26) N is at the bottom of the N -stringsfor N acting on End Q ( V ) (cid:27) . It can be shown that there is an sl (cid:48) = { N +2 , Y , N , } that commutes with the sl above. Thus(V.B.10) Given N , N as above, there are commuting sl ’s with N and N , asnil-negative elements. Moreover, N = N , +(terms of strictly negative weights) relative to { N +1 , Y , N } .It is the “strictly negative” that will be an essential ingredient needed to establishthat the limit exists in the main result.V.C. Calculation of the Chern forms Ω and Ω I .Step 1: For a nilpotent orbit (V.A.3) holomorphic sections of the canonically ex-tended VHS over ∆ are given byexp (cid:32) k (cid:88) j =1 (cid:96) ( t j ) N j (cid:33) v, v ∈ V C . There is a shift in indices that will not be needed here (cf. (VI.A.13) below). p to non-zero constants the Hodge metric is( u, v ) = Q (cid:32) exp (cid:32)(cid:88) j (cid:96) ( t j ) N j (cid:33) u, exp (cid:32)(cid:88) j (cid:96) ( t j ) N (cid:33) ¯ v (cid:33) = Q (cid:32) exp (cid:32)(cid:88) j log | t j | N j (cid:33) u, ¯ v (cid:33) . Using the notation (V.B.3) the associated graded to the LMHS as t → V gr = n ⊕ i =0 H n − i ⊗ U i and F n = n ⊕ i =0 H n − i, . For u ∈ H n − i, ( − i ) and v ∈ H n − i, Q (cid:32) exp (cid:32)(cid:88) j log | t j | N j (cid:33) u, ¯ v (cid:33) = (cid:18) i ! (cid:19) Q (cid:32)(cid:88) j log | t j | N j (cid:33) i u, ¯ v . Setting x j = − log | t j | the metric on the canonically extended line bundle is a non-zero constant times(V.C.2) P ( x ) = n (cid:89) i =0 det (cid:32)(cid:88) j x j N j (cid:12)(cid:12) H n − i, (cid:33) i . Here to define “det” we set N = (cid:80) N j and are identifying H n − i ( − i ) with H n − i using N i . Note that the homogenous polynomial P ( x ) is positive in the quadrant x j > ∂∂ log P ( x ) . Step 2:
Define N I = (cid:88) i ∈ I x i N i , N I c = (cid:88) j (cid:54)∈ I x j N j N = k (cid:88) i =1 x i N i = N I + N I c and set P = n (cid:89) i =0 det (cid:16) N (cid:12)(cid:12) H n − i, ( − i ) (cid:17) i . Denoting by V gr ,I = n ⊕ i =0 H n − i,iI ⊗ U i he associated graded to the LMHS as t → ∆ ∗ I , we define P I = n (cid:89) i =0 det (cid:16) N I (cid:12)(cid:12) H n − i, I ( − i ) (cid:17) i . Taking N I = N and N I c = N in (iv) in Section V.B, we have N I c , = weight zero component of N I c where weights are relative to the grading element Y I for N I . Decomposing the RHSof (V.C.1) using the sl × sl (cid:48) corresponding to N I and N I c , by (IV.B.10) we obtain(V.C.4) V gr ∼ = ⊕ i,j H n − i − ji,j ⊗ U i ⊗ U j where H n − i − ji,j is a polarized Hodge structure of weight n − i − j . Note that thisdecomposition depends on I . On H n − i − ji,j ⊗ U i ⊗ U j we have a commutative square H n − i − j ( − i − j ) N iI (cid:47) (cid:47) N jIc, (cid:15) (cid:15) H n − i − ji,j ( − j ) N jIc, (cid:15) (cid:15) H n − i − ji,j ( − i ) N iI (cid:47) (cid:47) H n − i − ji,j . Using (V.C.4) this gives P = (cid:89) i,j det (cid:16) N iI N jI c , (cid:12)(cid:12) H n − i − ji,j ( − i − j ) (cid:17) + R where the remainder term R involves the N I c , − m ’s for m >
0. We may factor theRHS to have P = (cid:89) i,j det (cid:16) N iI (cid:12)(cid:12) H n − i − ji,j ( − i − j ) (cid:17) (cid:89) i,j det (cid:16) N jI c , (cid:12)(cid:12) H n − i − ji,j ( − i − j ) (cid:17) + R which we write as(V.C.5) P = P I · P I c + R where P I and P I c are the two (cid:81) i,j factors. We note that(V.C.6) the remainder term R = 0 if we have commuting sl ’s. We next have the important observation
Lemma
V.C.7 : P I c is the Hodge metric in the line bundle Λ I → Z ∗ I .Proof. This is a consequence of the RWFP (V.B.7) applied to the situation at handwhen we take N = N I and N = N I c . (cid:3) To have commuting sl ’s means that N , − m = 0 for m > y (V.C.6), if we have commuting sl ’s, then R = 0Ω = − ∂∂ log P = − ∂∂ log P I − ∂∂R I c ≡ − ∂∂ log P I c modulo dt i , d ¯ t i for i ∈ I ≡ Ω I and we are done.In general, we have(V.C.8) Ω ≡ Ω I + S + S where(V.C.9) S = ∂P I c ∧ ∂R + ∂R ∧ ∂P I c − P I c ∂∂RP I P I c S = ∂R ∧ ∂RP I P I c . Step 3:
We will now use specific calculations to analyze the correction terms S , S .The key point will be to use that N = N I + N I c = N I + N I c , (cid:124) (cid:123)(cid:122) (cid:125) + (cid:88) m (cid:61) N I c , − m (cid:124) (cid:123)(cid:122) (cid:125) where the terms over the first brackets may be thought of as “the commuting sl -partof N I , N I c ” and the correction term over the second bracket has negative Y I -weights.We set h n − i, I = dim H n − i, I and for a monomial M = x (cid:96) · · · x (cid:96) k k we definedeg I M = (cid:88) i ∈ I (cid:96) i . Lemma
V.C.10 : (i) For any monomial M appearing in P deg I M (cid:53) nh I + ( n − h , I + · · · + h n − i, I = n (cid:88) i =1 ih n − i, I . (ii) If π is any permutation of , . . . , k and (cid:96) π,i = n (cid:88) j =1 j (cid:16) h n − j, { π (1) ,...,π ( i ) } − h n − j, { π (1) ,...,π ( i − } (cid:17) then M π := x (cid:96) π, π (1) · x (cid:96) π, π (2) · · · x (cid:96) π,k π ( k ) = x (cid:96) π,π − k ) · · · x (cid:96) π,π − k ) k appears with a non-zero coefficient in P . Corollary:
The monomials appearing in P are in the convex hull of the monomials M π . roof. For V gr ,I = Gr W ( N I ) V we have as { N + I , Y I , N I } -modules V gr ,I ∼ = n ⊕ i =0 H n − iI ⊗ U i . Decomposing the RHS as sl (cid:48) -modules we have V gr ,I ∼ = n ⊕ i =0 H n − ia,i − a ⊗ U a ⊗ U (cid:48) i − a where the H n − ia,i − a depend on I . The map N i : H n − i, ( − i ) → H n − i gives i ⊕ a =0 H n − i, a,i − a ( − i ) → i ⊕ a =0 H n − i, a,i − a . The Y I -weights of vectors in H n − i, a,i − a are equal to a , and thus ∧ h n − i, (cid:18) i ⊕ a =0 H n − i, a,i − a ( − i ) (cid:19) has weight (cid:80) ah n − i, a,i − a and ∧ h n − i, (cid:18) i ⊕ a =0 H n − i, a,i − a (cid:19) has weight − (cid:80) ah n − i, a,i − a . As a conse-quence any monomial in det (cid:16) N i (cid:12)(cid:12) H n − i, ( − i ) (cid:17) drops weights by (cid:80) h n − i, a,i − a . We have(V.C.11) det (cid:18)(cid:16) N (cid:12)(cid:12) H n − i, ( − i ) (cid:17) i (cid:19) = det (cid:18)(cid:16) ( N I + N I c , ) (cid:12)(cid:12) H n − i, ( − i ) (cid:17) i (cid:19) + T where T = terms involving N I c , neg . For any monomial M in a minor involving the N I c , neg of total weight − d ,2 deg I M + d = 2 i (cid:88) a =0 ih n − i, a,i − a , d > I M < i (cid:88) a =0 ah n − i, a,i − a . Putting everything together, we have (V.C.10) where(V.C.12) T = (cid:40) linear combinations of monomials M satisfying deg I M < (cid:80) ni =0 (cid:80) ia =0 ah n − i, a,i − a (cid:41) . Using the bookkeeping formula h n − i, I = (cid:80) ni = a h n − i, a,i − a we obtain n (cid:88) i =0 i (cid:88) a =0 ah n − i, a,i − a = i (cid:88) a =0 n (cid:88) i = a h n − i, a,i − a = n (cid:88) a =0 ah n − a, I hich gives P = P I P I c + (cid:32) correction term with deg I < n (cid:88) a =0 ah n − a, I (cid:33) where deg I P I = (cid:80) na =0 ah n − a, I and deg I P I c = 0, giving (i) in (V.C.10).A parallel argument shows that for I ∩ J = ∅ D I ∪ J := n (cid:89) i =0 det (cid:16)(cid:16) N I + N I, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17)(cid:17) i + a correction with deg I < m (cid:88) a =0 ah n − a, I . By the definition of H n − iI ∪ J ,det (cid:18)(cid:16) N I + N J, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17) i (cid:19) (cid:54) = 0and deg I (cid:18) det (cid:18)(cid:16) N I + N J, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17) i (cid:19)(cid:19) = n (cid:88) a =0 ah n − a, I while automatically deg I ∪ J (all terms of D I ∪ J ) = n (cid:88) a =0 ah n − a,oI ∪ J . Thus deg J det (cid:18)(cid:16) N I + N J, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17) i (cid:19) = deg I ∪ J det (cid:18)(cid:16) N I + N J, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17) i (cid:19) − deg I det (cid:18)(cid:16) N I + N J, (cid:12)(cid:12) H n − i, I ∪ J ( − i ) (cid:17) i (cid:19) = n (cid:88) a =0 a (cid:0) h n − a, I ∪ J − h n − a, I (cid:1) . Proceeding inductively on { π (1) } ⊂ { π (1) , π (2) } ⊂ · · · ⊂ { π (1) , . . . , π ( k ) } we obtain,if N { π (1) ,...,π ( (cid:96) ) } , = weight 0 piece of N { π (1) ,...,π ( (cid:96) ) } with respect to Gr W ( N ) { π (1) ,...,π ( k ) } then n (cid:89) i =0 det (cid:16)(cid:0) N { π (1) } + N { π (1) ,π (2) } , + · · · + N { π (1) ,...,π ( k ) } , (cid:12)(cid:12) H n − i, (cid:1) i (cid:17) is a non-zero multiple of x (cid:96) π (1) x (cid:96) π (2) · · x (cid:96) k π ( k ) . This is our M π . Tracking the correctionterms we have P = (cid:88) π C π M π + terms strictly in the convex hull of the M π where C π (cid:54) = 0 for all π . This proves (ii) in V.C.10. (cid:3) tep 4: Referring to (V.C.8) and (V.C.9), from Lemma V.C.10 we have:(a) R has deg I R < deg I P I , and all monomials satisfy (i) in V.C.10.(b) R is a sum of products of monomials M M where each deg I M < deg I P I and M i satisfies (i) in V.C.10.To complete the proof we have Lemma
V.C.13 : Given a monomial M in the I -variables satisfying deg I M < deg I P and (ii) in Lemma V.C.10, lim t → ∆ ∗ I M/P I = 0 . Proof.
Implicit in the lemma is that the limit exists. We note that t → ∆ ∗ I is thesame as x i → ∞ for i ∈ I . We also observe that the assumptions in the lemma implythat there is a positive degree monomial M (cid:48) with deg I ( M (cid:48) M ) = deg I P I and where M (cid:48) M lies in the convex hull of the M π ’s for P I . Using this convex hull property wewill show that(V.C.14) M (cid:48) M/P I is bounded as x i → ∞ for i ∈ I. Since lim x i →∞ M (cid:48) ( x ) = ∞ , this will establish the lemma.We now turn to the proof of (V.C.14). Because the numerator and denomina-tor are homogeneous of the same degree, the ratio is the same for ( x , . . . , x k ) and( λx , . . . , λx k ), λ > I = { , . . . , d } . Suppose that x ν = ( x ν , . . . , x νd ) isa sequence of points in ( x i > , i ∈ I ) such thatlim ν →∞ M (cid:48) M ( x ν ) D I ( x ν ) = ∞ . Consider a successive set of subsequences such that for all i, j , we have one of threepossibilities:(i) lim ν →∞ x νi /x νj = ∞ ;(ii) x νi /x νj is bounded above and below, which we write as x νi ≡ x νj ;(iii) lim ν →∞ x νi /x νj = 0. Now replace our sequence by this subsequence. Let I i , . . . , I r be the partition of I such that i ≡ j ⇐⇒ (ii) holds for i, j and order them so that (i) holds for i, j ⇐⇒ i ∈ I m , j ∈ I m and m < m . We maythus find a C > C (cid:53) x νi /x νj (cid:53) C if i, j in same I m , and for any B > x νi /x νj > B m − m if i ∈ I m , j ∈ I m , ν sufficiently large.By compactness, we may pick a subsequence so that lim ν →∞ ( x νi /x νj ) = C ij if i, j ∈ same I m .Now introduce variables y , . . . , y d and let x i = a i y m if i ∈ I m , a i /a j = C ij , a i > . In effect we are doing a sectoral analysis in the co-normal bundle to the stratum ∆ ∗ − I , whichexplains the wave front set analogy mentioned above. e may restrict our cone by taking (cid:101) N m = (cid:88) i ∈ I m a i N i . This reduces us to the case | I m | = 1 for all m , i.e.,lim ν →∞ x νi /x νj = ∞ if i < j . Thus for any B , x νi /x νj > B j − i for v (cid:29) . Now x m ν x m ν · · x m d ν d x (cid:96) ν x (cid:96) ν · · x (cid:96) d ν d → m + m + · · + m d = (cid:96) + (cid:96) + · · + (cid:96) d and m < (cid:96) , or m = (cid:96) and m < (cid:96) , · · .Thus P I = cM { , ,...,d } + terms of slower growth as ν → ∞ , c > M { , ,...,d } / other terms)( x ν ) > B. Since M (cid:48) M belongs to the convex hull of the M π , ( M (cid:48) M/M { , ,...,d } )( x ν ) is boundedas ν → ∞ . This proves the claim. (cid:3) Example V.C.15:
An example that illustrates most of the essential points in theargument is provided by a neighborhood ∆ of the dollar bill curve$ ←→ with the dual graphin M . The family may be pictured as follows: In effect we are making a generalized base change ∆ ∗ d → ∆ ∗ k such that for the pullback to ∆ ∗ d the coordinates y m go to infinity at different rates. ne nodal curveon the coordinateplanetwo nodal curve onthe coordinate axis dollar bill curveat the origin (cid:45) (cid:65)(cid:65)(cid:65)(cid:65)(cid:65)(cid:65)(cid:75) smooth curvein ∆ ∗ Figure 1.
With the picture δ δ δ each of the coordinate planes outside the axes is a family of nodal curves where oneof the vanishing cycles δ i has shrunk to a point. Along each of the coordinate axestwo of the three cycles have shrunk to a second node, and at the origin we have thedollar bill curve.We complete the δ i to a symplectic basis by adding cycles γ i . γ γ The corresponding monodromies around the coordinate axes are Picard-Lefschetztransformations with logarithms N i ( µ ) = ( µ, δ i ) δ i Here γ is not drawn in. nd with matrices N = (cid:32) (cid:33) , N = (cid:32) (cid:33) , N = (cid:32) (cid:33) . Setting as usual (cid:96) ( t ) = log t/ πi , the normalized period matrix isΩ( t ) = (cid:32) (cid:96) ( t ) + (cid:96) ( t ) (cid:96) ( t ) (cid:96) ( t ) (cid:96) ( t ) + (cid:96) ( t ) (cid:33) + (cid:32) holomorphicterm (cid:33) . The corresponding nilpotent orbit is obtained by taking the value at the t i = 0 of theholomorphic term, and by rescaling this term may be eliminated.Setting L ( t ) = ( − log | t | ) / π and P M ( t ) = ( i/ ∂∂ log L ( t ), the metric in thecanonically framed Hodge vector bundle is the 2 × H ( t ) = (cid:32) L ( t ) + L ( t ) L ( t ) L ( t ) L ( t ) + L ( t ) (cid:33) ;in the Hodge line bundle the metric is h ( t ) = L ( t ) L ( t ) + ( L ( t ) + L ( t )) L ( t )= L ( t ) L ( t ) + L ( t t ) L ( t ) . Setting ω = ∂∂ log h ( t ) = ∂∂ (cid:0) log( L ( t ) L ( t ) + L ( t t ) L ( t )) (cid:1) ω = ∂∂ log L ( t t )we will show that(V.C.16) ω (cid:12)(cid:12) t =0 is defined and is equal to ω . Proof.
Setting ψ = ∂h/h and η = ∂∂h/h we have ω = − ψ ∧ ψ + η. Now ψ = ∂ ( L ( t ) L ( t ) + L ( t t ) L ( t )) L ( t ) L ( t ) + L ( t t ) L ( t ) . Setting dt = 0 the dominant term of what is left is the left-hand term in ∂L ( t t ) L ( t ) L ( t ) L ( t ) + L ( t t ) −→ ∂L ( t t ) L ( t t ) , and the arrow means that the limit as t → η , letting ≡ denote modulo dt and dt and taking the limit as above η ≡ ∂∂L ( t t ) L ( t ) L ( t ) L ( t ) + L ( t t ) −→ ∂∂L ( t t ) L ( t t ) , which gives the result. (cid:3) e next observe that(V.C.17) ω (cid:12)(cid:12) t =0 is defined and is equal to zero.Proof. The computation is similar to but simpler than that in the proof of (V.C.16). (cid:3)
Interpretations : The curves pictured in Figure 1 map to an open set ∆ ⊂ M .The PHS’s of the smooth curves in ∆ ∗ vary with three parameters. Those on thecodimension 1 strata such as ∆ ∗ × ∆ vary in moduli with two parameters. Theirnormalizations are Eqp (cid:45) and their LMHS’s vary with two parameters with (cid:40) Gr (LMHS) ∼ = H ( E )Gr (LMHS) ∼ = Q and where the extension data in the LMHS is locally given by AJ E ( p − q ). ThusGr(LMHS) varies with one parameter and for the approximating nilpotent orbit isconstant along the curves t t = c . This local fibre of the map M → M is part ofthe closed fibre parametrized by E .Along the codimension 2 strata such as ∆ ∗ × ∆ the curves vary in moduli with1-parameter. Their normalizations are qp p (cid:48) q (cid:48) (cid:45) and the moduli parameter is locally the cross-ratio of { p, q ; p (cid:48) , q (cid:48) } . The LMHS’s arepurely Hodge-Tate and thus Gr(LMHS) has no continuous parameters. In summary • Φ e is locally 1-1 on ∆ ∗ ; • Φ e, ∗ has rank 1 on ∆ ∗ × ∆; • Φ e is locally constant on ∆ ∗ × ∆ .As c → e on the ∆ ∗ × ∆ tend to the coordinate axis ∆ ∗ × ∆ alongwhich Φ e is locally constant. roof of Theorem I.A.14 in the dim B = 2 case. We are given Φ : B → Γ \ D where B = B \ Z with Z = (cid:80) Z i a normal crossing divisor whose components areirreducible, smooth curves around which the monodromy T i is unipotent and non-trivial. We may also assume that Φ ∗ is generically 1-1. The mapping Φ is then properand the images M = Φ( B ) ⊂ Γ \ D is a closed complex surface. We want to showthat M has a canonical completion to a complex analytic surface M over which theHodge line bundle over M extends to an ample line bundle Λ e → M . We choose ourlabelling so that the LMHS is locally constant along the Z i for i = 1 , . . . , k and isnot locally constant along the remaining Z j . Then the curves Z i for i = 1 , . . . , m arethe ones to be contracted to give M . It is classical that the criterion to be able tocontract the Z i to normal singularities is that the intersection matrix(V.C.18) (cid:107) Z i · Z j (cid:107) ≤ i,j (cid:53) k < Z i is finite, the canonically extended Hodgeline bundle Λ e,i → Z i is of finite order, and moreover the canonically extended Chern ω e may be restrictedto Z i and ω i = ω e,i (cid:12)(cid:12) Z i = 0where ω i is the Chern form of the Hodge line bundle Λ i = Λ e (cid:12)(cid:12) Z i . Since ω e > e (cid:12)(cid:12) Z i is offinite order a power of it is trivial and therefore for some m the line bundle Λ me → B descends to a line bundle Λ me,M → M . Although M is singular and ω e only descendsto a singular differential form on M , the classical Kodaira theorem may be extendedto this case to give that Λ Me,M → M is ample (cf. Section VI in [GGLR17] for thedetails).VI. Applications, further results and some open questions
VI.A.
The Satake-Baily-Borel completion of period mappings.
We will discuss the proof of Theorem I.A.14 as stated in the introduction. Thereare three steps in the argument:(a) construction of the completion M of M ;(b) analysis of the extended Chern form ω e on B ; and(c) extension of the classical Kodaira theorem to the case of the singular variety M where the Chern form up on B has mild singularities.Step (b) was discussed in Section V above and using it step (c) can be done byextending essentially standard arguments. For this we refer to Section 6 in [GGLR17]and here shall take up step (a). What follows is not a formal proof; the intent is toillustrate some of the key ideas behind the argument given in [GGLR17]. There arefour parts to the argument: a1) localize the period mapping toΦ : ∆ ∗ k × ∆ (cid:96) → Γ loc \ D where Γ loc = { T , . . . , T k } is the local monodromy group and determine thestructure of local image Φ(∆ k × ∆ (cid:96) ) ⊂ M ;(a2) using the first step and the proper mapping theorem, show that the fibres F of the set-theoretic map Φ e : B → M are compact analytic subvarieties of B , and from this infer that M has thestructure of a compact Hausdorff topological space and Φ e is a proper map-ping;(a3) define the sheaf O M whose sections over an open set U ⊂ M are the continuousfunctions f such that f ◦ Φ e is holomorphic in Φ − e ( U ), and show that O M endows M with the structure of an analytic variety; and(a4) show that certain identifications of connected components of the fibres thatresult from the global action of monodromy give a finite equivalence relation(this uses [CDK95]).In this paper we shall mainly discuss the first step isolating the essential point of howthe relative weight filtration property enters in the analysis of how the local periodmapping in (a1) extends across the boundary strata of ∆ ∗ k × ∆ (cid:96) in ∆ k × ∆ (cid:96) .Step (a2) is based on the following (cf. [Som78]). LetΦ : B → Γ \ D be a period mapping where B = B \ Z with Z = (cid:80) i Z i a normal crossing divisorwith unipotent monodromy T i around Z i . The Φ may be extended across all the Z ∗ i := Z i \ ( ∩ j (cid:54) = i Z i ∩ Z j ) around which T i = Id and the resulting mappingΦ e : B e → Γ \ D is proper. Thus Φ e ( B e ) ⊂ Γ \ D is a closed analytic variety. This result will be usedfor B , and also for the period mappingsΦ I : Z ∗ I → Γ I \ D I given by taking the associated gradeds to the limiting mixed Hodge structures alongthe strata Z ∗ I := Z I \ ( ∩ J ⊃ I Z J ∩ Z I ).Step (a3), which is work still in progress, is based on analysis of the global structureof the fibres F , specifically positivity properties of the co-normal sheaf I F /I F (whichis the co-normal bundle N ∗ F /B in case F is smooth). This result uses that an integral element of the IPR the holomorphic sectional curvatures are (cid:53) − c for some c > urning to the discussion of step (a1) we shall only consider the key special casewhen (cid:96) = 0 (there are no parameters) and Φ is a nilpotent orbit (V.B.3). Then Φ isgiven by taking the orbit of the image of the homomorphism of complex Lie groups ρ : C ∗ k → G C whose differential is ρ ∗ ( t i ∂/∂t i ) = N i . Referring to the discussion following the statement of Theorem I.A.14 in the intro-duction and recalled below we will prove the following
Proposition
VI.A.1 : There exists a mapping µ : ∆ ∗ k → C N whose fibres are exactly the fibres of the set-theoretic mapping Φ e localized to ∆ ∗ k andwith monodromy group Γ loc (cf. VI.A.3 below).
The mapping µ in the proposition will be given by monomials, and for this reason itwill be called a monomial mapping .The mapping Φ e localized to ∆ ∗ k arises from a period mapping(VI.A.2) Φ : ∆ ∗ k → Γ loc \ D given by a nilpotent orbit. The corresponding variation of Hodge structure over∆ ∗ k induces variations of polarized limiting mixed Hodge structures along the openboundary strata ∆ ∗ I = { ( t , . . . , t k ) ∈ ∆ k : t i = 0 for i ∈ I and t j (cid:54) = 0 for j ∈ I c } .Passing to the primitive parts of the associated graded gives period mappingsΦ I : ∆ ∗ I → Γ I \ D I , where Γ I is generated by the T i for i ∈ I . The set-theoretic mapping Φ e is given by(VI.A.3) Φ e : ∆ k → Γ loc \ D (cid:113) (cid:32)(cid:97) I Γ I, loc \ D I (cid:33) where Φ e (cid:12)(cid:12) ∆ ∗ I and Φ I, loc is the monodromy group given by the VHS over ∆ ∗ I .The proof of Proposition VI.A.1 will be given in several steps, as follows.1. determine the connected components of the nilpotent orbit(VI.A.4) exp (cid:32)(cid:88) j (cid:96) ( t j ) N j (cid:33) · F : ∆ ∗ k → Γ loc \ D ;2. extend that argument to determining the connected components of the mappingsΦ I , which will also be given by nilpotent orbits(VI.A.5) Φ I ( t I ) = exp (cid:32)(cid:88) j ∈ I c (cid:96) ( t j ) N j (cid:33) · F I : ∆ ∗ I → Γ I \ D I ;3. show that for I ⊂ J the closure in ∆ ∗ I ∩ ∆ ∗ J of a fibre of Φ I is equal to a connectedcomponent of a fibre of Φ J ; and4. show that the limit of a sequence of fibres of Φ I is contained in a fibre of Φ J . s for removing the “connected component” qualifiers go, part of this deals withthe identifications induced by global monodromy that was mentioned above.The basic idea in the construction appears already in step 1; the most interestingpart of the argument is step 3 where the relative weight filtration property providesthe key. Step 1:
We consider the question:
What are the conditions that a monomial (VI.A.6) t B = t b · · · t b k k , b i ∈ Z be constant on the fibres of (VI.A.2) ? For this we let R = (cid:40) A = ( a , . . . , a k ) : (cid:88) i a i N i = 0 (cid:41) ⊂ R k be the set of relations on the N i ∈ g . We note that R is defined over Q . Proposition
VI.A.7 : The conditions that the monomial (VI.A.6) be constant on thefibres of (VI.A.2) are A · B = (cid:88) j a j b j = 0 , A ∈ R. Proof.
The vector field induced by N = (cid:80) a j N j ∈ g R is nowhere vanishing on D . Thus on the one hand ρ ∗ (cid:32)(cid:88) j a j t j ∂/∂t j (cid:33) is tangent to a fiber of Φ ⇐⇒ (cid:88) j a j N j = 0 . On the other hand, the condition that the monomial (VI.A.6) be constant on theorbits of the vector field (cid:80) j a j t j ∂/∂t j on ∆ ∗ k is (cid:32)(cid:88) j a j t j ∂/∂t j (cid:33) t B = ( A · B ) t B = 0 . (cid:3) This simple computation contains one of the key ideas in the construction of themonomial mapping µ : ∆ k → C N . We next consider the question:
Are there enough monomials (VI.A.6) satisfying A · B = 0 for all A ∈ R andwhere b j ∈ Z (cid:61) to separate the connected components of the fibres of (VI.A.4) ? This is the existence result that is needed to give local charts that will define thefibres of (VI.A.4) up to connected components. It is a consequence of the following
Proposition
VI.A.8 : The subspace R ⊥ is spanned by vectors B where all b i ∈ Q (cid:61) . This is a consequence of a result in linear programming, known as
Farkas’ alterna-tive theorem . We refer to Section 3 in [GGLR17] for details and a reference. Here we use the [GGLR17] notations g = End Q ( V ) and g R = End Q ( V R ). This is because D ∼ = G R /H where H is a compact subgroup of G R . The Lie algebra h thencontains no non-zero nilpotent elements in End Q ( V R ). tep 2: We consider the period mappings (VI.A.5)Φ I : Z ∗ I → Γ loc ,I \ D I given by the variation of polarized limiting mixed Hodge structures on the opensmooth strata Z ∗ I . Again restricting to the case of a nilpotent orbit (VI.A.5) we mayask for the analogue of the question in Step 1 for this nilpotent orbit.The key observations here are that here both the weight and the Hodge filtrationsenter , and since Φ I maps to the associated graded relative to the weight filtration W ( N I ) any operation that decreases W ( N I ) has no effect. Recalling from SectionV.B our notations • N I = (cid:80) i ∈ I N i ; • Y I = grading element for N I and { N I , Y I , N + I } is the resulting sl ; • for j ∈ I c we have N j = N j, + N j, − + · · · where N j, − m is the − m weightspace for Y I ;it follows that the nilpotent orbit (VI.A.5) is the same as the nilpotent orbit using N j, in place of N j , and in place of Proposition VI.A.7 we have Proposition
VI.A.9 : The connected components of the fibres of (VI.A.5) are thelevel sets of monomials t B where b j ∈ Z (cid:61) and (cid:88) j ∈ I c b j N j ∈ W − ( N I ) g . Moreover, recalling that the Chern form of the Hodge line bundle is given by ω I = ω e (cid:12)(cid:12) Z ∗ I these connected components are exactly the connected integral varieties of theexterior differential system ω I = 0 . We recall our notation ∆ ∗ I = { t ∈ ∆ ∗ k : t i = 0 for i ∈ I and t j (cid:54) = 0 for j ∈ I c } anddenote by(VI.A.10) µ I : ∆ ∗ I → C N I the monomial map constructed in the same way as the monomial map constructedfrom Propositions VI.A.7 and VI.A.8 using the t B for a generating set of vectors B ∈ R ⊥ with the b i ∈ Z (cid:61) . The to be constructed compact analytic variety is M isset theoretically the disjoint union(VI.A.11) M = M (cid:113) (cid:32)(cid:97) I M I (cid:33) where M I is a finite quotient of the union of the imagesΦ I ( Z ∗ I ) ⊂ Γ loc ,I \ D I . To complete the proof of the construction of M as compact analytic variety two issuesneed to be addressed: i) set-theoretically, the inverse image of Φ I (∆ ∗ I ) ⊂ M I is a finite cover of µ I (∆ ∗ I )and we need to describe analytic functions that will separate the sheets of thiscovering, and(ii) we need to show that the analytic varieties M I fit together to give the structureof an analytic variety on M . Step 3:
For the second of the two issues above we observe that the restriction t B (cid:12)(cid:12) ∆ ∗ I = 0 if b i > i ∈ I. To establish (ii) we have the
Proposition
VI.A.12 : For I (cid:36) J so that ∆ ∗ J ⊂ ∆ ∗ I , the closure of a level set of µ I is contained in a level set of µ J . Moreover the limit in ∆ ∗ J of level sets in ∆ ∗ I iscontained in a level set of µ J .Proof. This will be a consequence of the relative weight filtration property (RWFP)(V.B.7) that we now recall in a form adapted to the proof of (VI.A.12).Given
A, B ⊂ { , . . . , k } with A ∩ B = ∅ , we denote by N B = N B, + N B, − + N B, − + · · · the Y A -eigenspace decomposition of N B relative to the sl { N + A , Y A , N A } . Then thenilpotent operator N B, (cid:12)(cid:12) Gr W ( NA ) m V : Gr W ( N A ) m V → Gr W ( N A ) m V induces a weight filtration on Gr W ( N A ) • V . Another weight filtration on this vectorspace is defined by W • ( N A + N B ) ∩ W m ( N A ) W • ( N A + N B ) ∩ W m − ( N A ) . The RWFP is that these two filtrations coincide; i.e.,(VI.A.13) W m + m (cid:48) ( N A + N B ) ∩ W m ( N A ) W m + m (cid:48) ( N A + N B ) ∩ W m − ( N A ) = W m (cid:48) (cid:16) N B, (cid:12)(cid:12) Gr W ( NA ) m V (cid:17) . Returning to the proof of (VI.A.12), what must be proved is that for I (cid:36) J (VI.A.14) (cid:88) j ∈ I c a j N j ∈ W − ( N I ) = ⇒ (cid:88) j ∈ J c a j N j ∈ W − ( N J ) . On the RHS we have used that N j ∈ W − ( N J ) for j ∈ J , so the sum is really over j ∈ J c . What (VI.A.14) translates into is that if t A is a monomial that is constanton the fibres of Φ I , then the restriction t A (cid:12)(cid:12) ∆ ∗ J is constant on the level sets of Φ J .We let X = (cid:80) j ∈ I c a j N j ∈ W − ( N I ) g , where the “ ∈ ” is because X ∈ Z ( N I ) = ⇒ X has only negative weights in the N I -string decomposition of g . Here there is both a local issue dealing with the fibres of Φ I : ∆ ∗ I → Γ I \ D I , and a global issuearising from the possibility that two connected components of the fibre of Φ I : ∆ ∗ I → Γ I \ D I maybe subsets of a single fibre of Φ I on Z ∗ I due to the global action of monodromy. rite X = X + X − + X − + · · · in terms of the eigenspace decomposition of Y J . Since N J − I is in the − Y I , we have [ N J − I , X ] = 0 . Now decompose X into Y I eigenspace components X = X , − m + X , − ( m +1) + · · · , m (cid:61) . Then [ N J − I, , X , − m ] = 0 = ⇒ X , − m ∈ Gr W ( N I ) − m g lies in Z (cid:16) N J − I, (cid:12)(cid:12) Gr W ( NI ) − m g (cid:17) and consequently X , − m ∈ W (cid:16) N J − I, (cid:12)(cid:12) Gr W ( NI ) − m g (cid:17) . Applying (VI.A.13) with A = I , B = J − I gives (VI.A.14), which proves VI.A.12. (cid:3) Step 4:
At this point we have constructed a monomial mapping µ : C k → C N whose fibres are unions of the fibres of the nilpotent orbit (V.A.3). The final localstep is to refine the construction to have C k η (cid:47) (cid:47) µ (cid:33) (cid:33) C k ˜ µ (cid:125) (cid:125) C N (VI.A.15)where ˜ µ is a monomial mapping with connected fibres. The basic idea already occurswhen k = N = 1 and µ ( t ) = t m ; then “ t /m ” separates the fibres of µ .In general we suppose that µ ( t ) = ( t I , . . . , t I N )where t I j = t i j · · · t i jk k . Define a map Z k → Z N by e j → ( i j , . . . , i N j ) . Then identifying Z N with Hom( Z N , Z ), up to a finite group the image Λ ⊂ Z N isdefined by Λ ⊥ ⊂ Z N . Setting (cid:101) Λ = (Λ ⊥ ) ⊥ , for the finite abelian group (cid:101) Λ / Λ we have(VI.A.16) (cid:101) Λ / Λ ∼ = ⊕ Z /d i Z . he mapping η in (VI.A.15) will be a | (cid:101) Λ / Λ | -to-1 monomial map. To construct it, atthe level of exponents of monomials we will have Z k B (cid:47) (cid:47) A (cid:33) (cid:33) Z k (cid:101) A (cid:125) (cid:125) Z N (VI.A.17)where Im( A ) = Λ, Im( (cid:101) A ) = (cid:101) Λ and Z k / Im( B ) ∼ = (cid:101) Λ / Λ. Such a map always existsand may be constructed using (VI.A.16). We then use (VI.A.17) to define (VI.A.15)where ˜ µ is a monomial map with connected fibres. (cid:3) For the issue arising from the global action of monodromy we refer to Section 3 in[GGLR17].
Example VI.A.18:
This is a continuation of Example V.C.15 above. Since N , N , N are linearly independent, the mapping ∆ → Γ loc \ D given by the corresponding nilpo-tent orbit is 1-1. The interesting situation is on the face { } × ∆ ∗ given by t = 0. There we observefrom the computations in Example V.C.15 that the induced maps N , N : Gr W ( N )1 V → Gr W ( N )1 V are equal. This gives the relation N − N = 0 which leads to the monomial t t thatis constant on the fibres of the period mappingΦ : { } × ∆ ∗ → Γ , loc \ D . On the codimension 2 strata the associated graded to the LMHS’s are Hodge-Tate,so the corresponding period mappings are constant. We conclude this section by discussing the following question:
What are the Zariski tangent spaces to M ? More precisely,(VI.A.19)
What is the kernel of the mapping T b B → T Φ e ( b ) M ? Recalling the notation ω e for the Chern form of the canonically extended Hodgeline bundle Λ e → B , we want to define in each tangent space to B the meaning ofthe equations(VI.A.20) ω e ( ξ ) = 0 , ξ ∈ T b B. The issue is that ω e is not smooth, continuous, or even bounded. If b ∈ B , then(VI.A.20) has the usual meaning Φ ∗ ( ξ ) = 0. If b ∈ Z ∗ I and ξ ∈ T b Z ∗ I is tangent to Z ∗ I ,then since ω e (cid:12)(cid:12) Z ∗ I = ω I (VI.A.20) means that Φ I, ∗ ( ξ ) = 0. In this case D = H . In this case D = H . We note that the limit of the parabolas t t = c as c → hus the interesting case is when b ∈ Z ∗ I and ξ is a normal vector to Z ∗ I in B . Thisamounts to the situation of a 1-parameter VHSΦ : ∆ ∗ → Γ T \ D, Γ T = T Z and (VI.A.20) becomes the condition(VI.A.21) Φ e, ∗ ( ∂/∂t ) (cid:12)(cid:12) t =0 = 0 . In this case we have the
Proposition
VI.A.22 : If T (cid:54) = Id , then Φ e, ∗ ( ∂/∂t ) (cid:12)(cid:12) t =0 (cid:54) = 0 .Proof. If T = exp N is unipotent, then the methods used in Section V.A above andSection 3 in [GGLR17] give for the Chern form on ∆ ∗ ω e (cid:61) C dt ∧ d ¯ t | t | ( − log | t | ) , C > . In general T is quasi-unipotent and after a base change t (cid:48) = t m we will have∆ (cid:48) ∗ π (cid:15) (cid:15) Φ (cid:48) (cid:47) (cid:47) Γ T (cid:48) \ D (cid:15) (cid:15) ∆ ∗ Φ (cid:47) (cid:47) Γ T \ D where T (cid:48) is unipotent. Then from π ∗ (cid:18) dt ∧ d ¯ t | t | ( − log | t | ) (cid:19) = C (cid:48) dt (cid:48) ∧ d ¯ t (cid:48) | t (cid:48) | ( − log | t (cid:48) | ) , C (cid:48) > (cid:3) From VI.A.22 we may draw the
Conclusions
VI.A.23 : (i) If the N j , j ∈ I c , are linearly independent modulo W − ( N ) ,then ω e > in the normal spaces to Z ∗ I . (ii) For nilpotent orbits, ω e ( ξ ) = 0 ⇐⇒ µ ∗ ( ξ ) = 0 where µ : ∆ ∗ k → C N is themonomial map. We comment that (ii) holds in the general situation in Section 3 of [GGLR17] wherelocal quasi-charts are constructed for arbitrary VHS over ∆ ∗ k × ∆ (cid:96) .Finally we use the results of Section V above and Section 3 in [GGLR17] to sum-marize the properties of the EDS (VI.A.20):(VI.A.24) (a) (VI.A.20) defines a coherent, integrable sub-sheaf I ⊂ O B ( T B ) ; (b) the maximal leaves of I are closed , complex analytic subvarietiesof B ; (c) as a set, M is the quotient of the fibration of B given by the leavesof I . It is interesting to note that the pullback of smooth forms under a branched covering mapvanish along the branch locus. For the Poincar´e metric this is not the case, illustrating again thegeneral principle that in Hodge theory singularities increase positivity. ingular integrable foliations and their quotients have been introduced and studiedin [Dem12b].VI.B. Norm positivity and the cotangent bundle to the image of a periodmapping. (i)
Statement of results.
Let Φ : B → Γ \ D be a period mapping with image aquasi-projective variety M ⊂ Γ \ D . The G R -invariant metric on D constructed fromthe Cartan-Killing form on g R induces a K¨ahler metric on the Zariski open set M o of smooth points of M . We denote by R ( η, ξ ) and R ( ξ ) the holomorphic bi-sectionaland holomorphic sectional curvatures respectively. Theorem
VI.B.1 : There exists a constant c > such that (i) R ( ξ ) (cid:53) − c for all ξ ∈ T M o ; (ii) R ( η, ξ ) (cid:53) for all η, ξ ∈ T M o × M o T M o ; (iii) For any b ∈ M o there exists a ξ ∈ T b M o such that R ( η, ξ ) (cid:53) − c/ for all η ∈ T b M o . Observe that using (II.G.3) from [BKT13] (iii) follows from (i) and (ii). As acorollary to (ii) we have(iv) R ( η, ξ ) (cid:53) − c/ T M o × M o T M o .We note that (iii) implies that this open set projects onto each factor in M o × M o .As applications of the proof of Theorem VI.B.1 and consideration of the singularityissues that arise we have the following extensions of some of the results of Zuo [Zuo00]and others (cf. Chapter 13 in [CMSP17]): M is of log-general type, (VI.B.2) Sym m Ω M (log) is big for m (cid:61) m . (VI.B.3)The result in (VI.B.2) means that for any desingularization (cid:102) M of M with (cid:102) M lyingover M and (cid:101) Z = (cid:102) M \ (cid:102) M , the Kodaira dimension κ (cid:16) K (cid:102) M ( (cid:101) Z ) (cid:17) = dim M. The result in (VI.B.3) means thatSym m Ω (cid:102) M (log (cid:101) Z ) is big for m (cid:61) m . The proof will show that we may choose m to depend only on the Hodge numbersfor the original VHS.The proof will also show that(VI . B . S M is of stratified-log-general type, (VI . B . S Sym m Ω M (log) is stratified-big for m (cid:61) m . Here stratified-log-general type means that there is a canonical stratification { M ∗ I } of M such that each stratum M ∗ I is of log-general type. There is the analogous definitionfor stratified big. ithout loss of generality, using the notations above we may take (cid:102) M = B , (cid:102) M = B and (cid:101) Z = Z ; we shall assume this to be the case. Remark:
The results of Zuo, Brunebarbe and others are essentially that K M (log)and Ω M (log) are weakly positive in the sense of Viehweg. This is implied by (VI.B.3).The proof of Theorem VI.B.1 will be done first in the case(VI.B.4) B = B and Φ ∗ is everywhere injective. It is here that the main ideas and calculations occur.The singularities that arise are of the types(VI.B.5) (a) where Φ ∗ fails to be injective (e.g., on M sing ),(b) on Z = B \ B where the VHS has singularities,(c) the combination of (a) and (b).As will be seen below, there will be a coherent sheaf I withΦ ∗ ( T B ) ⊂ I ⊂ Φ ∗ ( T (Γ \ D )) . Denoting by I o the open set where I is locally free, there is an induced metric andcorresponding curvature form for I o , and with the properties (i), (ii) in the theorem for I o Theorem VI.B.1 will follow from the curvature decreasing property of holomorphicsub-bundles, which gives R ( η, ξ ) = Θ T M o ( η, ξ ) (cid:53) Θ I o ( η, ξ ) . As for the singularities, if we show that κ (det I o (log)) = dim B (VI.B.6) Sym m I o (log) is big (VI.B.7)then (VI.B.2) and (VI.B.3) will follow from the general result: If over a projectivevariety Y we have line bundles L, L (cid:48) and a morphism L → L (cid:48) that is an inclusion overan open set, then(VI.B.8) L → Y big = ⇒ L (cid:48) → Y is big. We will explain how (VI.B.6) and (VI.B.7) will follow from (VI.B.8) for suitablechoices of
Y, L and L (cid:48) .(ii) Basic calculation.
It is conventient to use Simpson’s system of Higgs bundlesframework (cf. [Sim92] and Chapter 13 in [CMSP17]) whereby a VHS is given by asystem of holomorphic vector bundles E p , and maps E p +1 θ p +1 −−→ E p ⊗ Ω B θ p −→ E p − ⊗ ∧ Ω X Here we are identifying a coherent sub-sheaf of a vector bundle with the corresponding familyof linear subspaces in the fibres of the vector bundle. The coherent sheaf I will be a subsheaf of thepull-back Φ ∗ T ( D \ Γ) h of the horizontal tangent spaces to Γ \ D . The critical step in the calculationwill be that it is integrable as a subsheaf Φ ∗ T (Γ \ D ) h . hat satisfy(VI.B.9) θ p ∧ θ p +1 = 0 . Thus there is induced E p +1 θ p +1 −−→ E p ⊗ Ω B θ p −→ E p − ⊗ Sym Ω B , and the data ( ⊕ p E p ⊗ Sym k − p Ω B , ⊕ p θ p ) for any k with k (cid:61) p is related to the notionof an infinitesimal variation of Hodge structure (IVHS) (cf. 5.5 ff. in [CMSP17]).In our situation the vector bundles E p will have Hermitian metrics with Chernconnections D p . The metrics define adjoints θ p ∗ : E p → E p +1 ⊗ Ω B , and in the cases we shall consider if we take the direct sum over p we obtain( E, ∇ = θ ∗ + D + θ ) , with (VI.B.9) equivalent to ∇ = 0 . The properties uniquely characterizing the Chern connection together with ∇ = 0give for the curvature matrix of E p the expression(VI.B.10) Θ E p = θ p +1 ∧ θ p +1 ∗ + θ p ∗ ∧ θ p , which is a difference of non-negative terms each of which has the norm positivityproperty (III.A.3) (cf. [Zuo00] and Chapter 13 in [CMSP17]).For a PVHS ( V, Q, ∇ , F ) we now set E p = Gr p Hom Q ( V, V ) , − n (cid:53) p (cid:53) n where Gr p is relative to the filtration induced by F on Hom Q ( V, V ). At each point b of B there is a weight zero PHS induced on Hom Q ( V, V ) = g and E pb = g p, − p with the bracket [ , ] : E p ⊗ E q → E p + q . Thinking of θ as an element in g ⊗ Ω B , the integrability condition VI.B.9 translatesinto(VI.B.11) [ θ, θ ] = 0 . We shall use the notation Gr p = Gr p Hom Q ( V, V )rather than E p for this example.The differential of Φ gives a mapΦ ∗ : T B → Gr − . Definition: I ⊂ Gr − is the coherent subsheaf generated by the sections of Gr − that are locally in the image of Φ ∗ over the Zariski open set where Φ ∗ is injective.For ξ a section of I we denote by ad ξ the corresponding section of Gr − . Theintegrability condition (VI.B.11) then translates into the first part of the roposition VI.B.12 : I is a sheaf of abelian Lie sub-algebras of ⊕ p Gr p . For η, ξ sections of I Θ Gr − ( η, ξ ) = −(cid:107) ad ∗ ξ ( η ) (cid:107) . Proof.
For η, ξ ∈ Gr − the curvature formula (VI.B.10) isΘ Gr − ( η, ξ ) = (cid:107) ad ξ ( η ) (cid:107) − (cid:107) ad ∗ ξ ( η ) (cid:107) . The result then follows from ad ξ ( η ) = [ ξ, η ] = 0 for η, ξ ∈ I . (cid:3) On the open set where I o is a vector bundle with metric induced from that on Gr − we have Θ I o ( η, ξ ) (cid:53) Θ Gr − ( η, ξ ) (cid:53) . The first term is the holomorphic bi-sectional curvature for the indued metric onΦ( B ).To complete the proof of Theorem VI.B.10 we need to show the existence of c > ξ of unit length(VI.B.13) (cid:107) ad ∗ ξ ( ξ ) (cid:107) (cid:61) c. The linear algebra situation is this: At a point of B we have V = ⊕ p + q = n V p,q and ξ is given by maps A p : V p,q → V p − ,q +1 , (cid:106) n + 12 (cid:107) (cid:53) p (cid:53) n. In general a linear map A : E → F between unitary vector spaces has principal values λ i defined by Ae i = λ i f i , λ i real and non-zerowhere e i is a unitary basis for (ker A ) ⊥ and f i is a unitary basis for Im A . The squarenorm is (cid:107) A (cid:107) = Tr A ∗ A = (cid:88) i λ i . We denote by λ p,i the principal values of A p . The λ p,i depend on ξ , and the squarenorm of ξ as a vector in T p B ⊂ T Φ( p ) (Γ \ D ) is (cid:107) ξ (cid:107) = (cid:88) p (cid:88) i λ p,i . In the above we now replace V by Hom Q ( V, V ) and use linear algebra to determinethe principal values of ad ∗ ξ . These will be quadratic in the λ p,i ’s, and then (cid:107) ad ξ ∗ ( ξ ) (cid:107) ill be quartic in the λ p,i . A calculation gives(VI.B.14) (cid:107) ad ξ ∗ ( ξ ) (cid:107) = (cid:88) p (cid:32) (cid:80) i a p λ p,i (cid:0)(cid:80) i λ p,i (cid:1) (cid:33) where the a p are non-negative integers that are positive if A p (cid:54) = 0, and from this byan elementary algebra argument we may infer the existence of the c > (cid:3) At this point we have proved the theorem. The basic idea is very simple:
For a VHS the curvature (VI.B.10) of the Hodge bundles is a difference of non-negative terms, each of which is of norm positivity type where the “ A ” in Def-inition III.A.3 is a Kodaira-Spencer map or its adjoint. For the Hom Q ( V, V ) variation of Hodge structure, A ( ξ )( η ) = [ ξ, η ] = 0 by integrability. Conse-quently the curvature form has a sign, and a linear algebra calculation givesthe strict negativity Θ I ( ξ, ξ ) (cid:53) − c (cid:107) ξ (cid:107) for some c > . (iii) Singularities.
The singularity issues were identified in (VI.B.5), and we shallstate a result that addresses them. The proof of this result follows from the resultsin [CKS86] as extended in [GGLR17], [Kol87] and the arguments in [Zuo00]. Using the notations introduced in (ii) above, a key observation is that the differ-ential Φ ∗ : T B → Gr − extends to Φ ∗ : T B (cid:104)− Z (cid:105) → Gr − e where T B (cid:104)− Z (cid:105) = Ω B (log Z ) ∗ and Gr − e is the canonical extension to B of Gr − → B . This is just a reformulation of the general result (cf. [CMSP17]) that for all p , θ p : E p → E p ⊗ Ω B extends to(VI.B.15) θ pe : E pe → E p − e ⊗ Ω B (log Z ) . As noted above, the image Φ ∗ T B ⊂ Gr − generates a coherent subsheaf I ⊂ Gr − and from (VI.B.15) we may infer that I extends to a coherent subsheaf I e ⊂ Gr − e .As in [Zuo00] we now blow up B to obtain a vector sub-bundle of the pullback ofGr − and note that I e ⊂ Gr − e will be an integrable sub-bundle. The metric on Gr − induces a metric in I and we use the notations • ϕ = Chern form of det I o ∗ ; • ω = Chern form of O P I o ∗ (1). Theorem
VI.B.16 : Both ϕ and ω extend to closed, (1 , currents ϕ e and ω e on B and P I ∗ e that respectively represent c (det I ∗ e ) and c ( O P I ∗ e )(1) . They have mildsingularities and satisfy This first proof of the result that appeared in the literature was Lie-theoretic where the metricon g was given by the Cartan-Killing form. As will be illustrated below the above direct algebraargument is perhaps more amenable to the computation in examples. These arguments have been amplified at a number of places in the literature; cf. [VZ03] and[P˘a16]. ϕ e (cid:61) and ϕ e > on an open set; • ω e (cid:61) and ω e > on an open set. With one extra step this result follows from singularity considerations similar tothose in Section IV above. The extra step is that I e is not a Hodge bundle, but rather it is the kernel of the map θ − : Gr − e → Gr − e ⊗ Ω e ⊗ Ω B (log Z ) . As was noted in [Zuo00], either directly or using (5.20) in [Kol87] we may infer thestated properties of ϕ e and ω e . (cid:3) Remark:
It is almost certainly not the case that any sub-bundle G ⊂ Gr − e willhave Chern forms with mild singularities. The bundle I e is special in that it is thekernel of the map Gr − e → Gr − e ⊗ Ω B (log Z ). Although we have not computed the2 nd fundamental form of I e ⊂ Gr − e , for reasons to be discussed below it is reasonableto expect it to also have good properties.The issue of the curvature form of the induced metric on the image M = Φ( B ) ⊂ Γ \ D seems likely to be interesting. Since the metric on the smooth points M o ⊂ M is the K¨ahler metric given by the Chern form of the augmented Hodge line bundle,the curvature matrix of T M o is computed from a positive (1,1) form that is itself thecurvature of a singular metric. In the 1-parameter case the dominant term in ω isthe Poincar´e metric P M = dt ⊗ d ¯ t/ | t | ( − log | t | ) , and the curvature of the P M is apositive constant times − P M . One may again suspect that the contributions of thelower order terms in ω are less singular than P M . This issue may well be relevant toQuestion I.A.10.(iv)
Examples.
On the smooth points of M o of the image of a period mapping theholomorphic bi-sectional curvature satisfies(VI.B.17) R ( η, ξ ) (cid:53) , and for η, ξ in an open set in T M o × M o T M o it is strictly negative. This raises theinteresting question of the degree of flatness of T ∗ M o . In the classical case when D is a Hermitian symmetric domain and B = Γ \ D is compact this question hasbeen studied by Mok [Mok87] and others. In case B is a Shimura variety the relatedquestion of the degree of flatness of the extended Hodge bundle F e over a toriodalcompactification of Γ \ D is one of current interest (cf. [Bru16a], [Bru16b] and thereferences cited there). This issue will be further discussed in Section VI.E.Here we shall discuss the equationΘ I ( η, ξ ) = 0over the smooth locus M o of M . In view of (VI.B.10) this equation is equivalent toad ξ ∗ ( η ) = 0 , η ∈ I. To compute the dimension of the solution space to this equation, we use the dualityker(ad ∗ ξ ) = (Im(ad ξ )) ⊥ o have(VI.B.18) dim ker(ad ∗ ξ ) = dim (cid:0) coker (cid:0) Im { ad ξ : Gr → Gr − } (cid:1)(cid:1) . Since I depends on the particular VHS, at least as a first step it is easier to studythe equation(VI.B.19) Ad ξ ∗ ( η ) = 0 , η ∈ Gr − . Because the curvature form decreases on the sub-bundle I ⊂ Gr − , over M o we have(VI.B.18) = ⇒ (VI.B.19)but in general not conversely. Example 1:
For weight n = 1 with h , = g , with a suitable choice of coordinatesthe tangent vector ξ is given by g × g symmetric matrix A , and on Gr − we have(VI.B.20) dim ker(ad ∗ ξ ) = (cid:32) g − rank A + 12 (cid:33) Proof.
At a point we may choose a basis for that Q = (cid:0) Ig − Ig (cid:1) and F is given by (cid:32) Ω I g (cid:33) , Im Ω > ξ ∈ Gr − is given by (cid:32) A (cid:33) , A = t Aη ∈ Gr is given by (cid:32) C − t C (cid:33) . Then [ ξ, η ] = (cid:32) AC + t CA (cid:33) . Diagonalizing A and using (VI.B.18) we obtain (VI.B.20). Example 2:
For weight n = 2, ξ is given by A = h , × h , matrix . We will show that on Gr − (VI.B.21) dim ker(ad ∗ ξ ) = ( h , -rank A )( h , -rank A ) . roof. We may choose bases so that Q = diag( I h , , − I h , , I h , ) and F is given by Ω0 i Ω , Ω non-singular ,ξ is given by A
00 0 t A ,η is given by C D
00 0 − t C . Then [ ξ, η ] = AC − DA
00 0 t AD + t ( AC )0 0 0 . Choosing bases so that A = ( I
00 0 ), C = (cid:0) C C C C (cid:1) and D = (cid:0) D D − t D D (cid:1) , we have AC − DA = (cid:32) C − D D − t D (cid:33) . Setting rk( E ) = rank E for a matrix E , this givesrk A h , -rk A rk Ah , -rk A (cid:32) ∗ ∗∗ (cid:33) where the ∗ ’s are arbitrary. (cid:3) As in the n = 1 case we note that(VI.B.22) A of maximal rank ⇐⇒ ker(ad ∗ ξ ) = 0 . Example 3:
Associated to a several parameter nilpotent orbitexp (cid:32)(cid:88) i (cid:96) ( t i ) N i (cid:33) · F is a nilpotent cone σ = { N λ = (cid:80) λ i N i , λ i > } and the weight filtration W ( N )is independent of N ∈ σ . As discussed in Section 2 of [GGLR17], without loss ofgenerality in what follows here we may assume that the LMHS associate to N ∈ σ is R -split. Thus there is a single Y ∈ Gr Hom Q ( V, V ) such that for any N ∈ σ [ Y, N ] = − N, nd using the Hard Leftschetz Property N k : Gr W ( N ) n + k ( V ) ∼ −→ Gr W ( N ) n − k ( V ) we mayuniquely complete Y, N to an sl { N, Y, N + } . Let g σ ⊂ End(Gr W ( N ) • V ) be the Liealgebra generated by the N i and Y . The properties of this important Lie algebrawill be discussed elsewhere; here we only note that g σ is semi-simple and that thenilpotent orbit gives a period mapping∆ ∗ k Φ σ −→ Γ loc \ D σ where D σ = G σ, R /H σ is a Mumford-Tate sub-domain of D . Of interest are theholomorphic bi-sectional curvatures of Φ σ (∆ ∗ k ). We shall not completely answer this,but shall give a proof of the Proposition
VI.B.23 : Θ I ( η, N ) = 0 for all N ∈ σ , if and only if, η ∈ Z ( g σ ) .Proof. We denote by g C = ⊕ p g p, − p the Hodge decomposition on the associated gradedto the limiting mixed Hodge structure defined by σ . The Hodge metric is given on g C by the Cartan-Killing form, and its restriction to g − , − is non-degenerate. Thedecomposition of g C into N -strings for the sl given by { N, Y, N + } is orthogonal withrespect to the Hodge metric, from which we may infer that the adjoint ad N ∗ actsseparately on each N -string. The picture is something like η ◦ η (cid:47) (cid:47) ◦ N ∗ (cid:106) (cid:106) η (cid:47) (cid:47) ◦ η (cid:47) (cid:47) N ∗ (cid:104) (cid:104) ◦ . N ∗ (cid:104) (cid:104) Because N is an isomorphism the same is true of N ∗ ; consequentlyΘ I ( η, N ) = 0 ⇐⇒ η belongs to an N -string of length 1,and this implies that [ η, Y ] = [ η, N + ] = 0. By varying N over σ we may conclude theproposition. (cid:3) Example 4:
One of the earliest examples of the positivity of the Hodge line bundlearose in the work of Arakelev ([Ara71]). For 1-parameter families it gives an upper bound on the degree of the Hodge line bundle in terms of the degree of the logarithmiccanonical bundle of the parameter spaces. This result has been extended in a numberof directions; we refer to [CMSP17], Section 13.4 for further general discussion andreferences to the literature.One such extension is due to [Zuo00], [VZ03] and [VZ06]. This proof of that resultcenters around the above observation that the curvature of Hodge bundles has asign on the kernels of Kodaira-Spencer mappings. There is a new ingredient in theargument that will be useful in other contexts and we shall now explain this. Asabove there are singularity issues that arise where the differential of Φ fails to beinjective. These may be treated in a similar manner to what was done above, andfor simplicity of exposition and to get at the essential new point we shall assume The decomposition of g C into the primitive sub-spaces and their images under powers of N depends on the particular N . The Hodge metric on g − , − is only definite on the subspaces arisingfrom the primitive decomposition for such an N . Using the above notations, the logarithmic canonical bundle of the parameter space is K B ( Z ). hat Φ ∗ is everywhere injective and that the relevant Kodaira-Spencer mappings haveconstant rank.The basic Arakelev-type inequality then exists at the curvature level. For a varia-tion of Hodge structure ( V, Q, ∇ , F ) over B with a completion to B with Z = B \ B a reduced normal crossing divisor, the inequality is(VI.B.24) (cid:32) curvature ofdet Gr p V (cid:33) (cid:53) C p (cid:32) curvature ofdet Ω B (log Z ) (cid:33) where C p is a positive constant that depends on the ranks of the Kodaira-Spencermappings. Here we will continue using the notations(VI.B.25) (cid:40) Gr p V = F p V /F p +1 V, Gr p V θ −→ Gr p − V ⊗ Ω B (log Z ) . The second of these was denoted by θ p above; we shall drop the “ p ” here but notethat below we shall use θ (cid:96) to denote the (cid:96) th iterate of θ . Proof of (VI.B.24) . Using the integrability condition (VI.B.9) the iterates of (VI.B.25)give Gr p V θ (cid:96) −→ Gr p − (cid:96) ⊗ Sym (cid:96) Ω B (log Z )We use the natural inclusion Sym (cid:96) Ω B (log Z ) ⊂ (cid:96) ⊗ Ω B (log Z ) and consider this mapas giving(VI.B.26) Gr p V θ (cid:96) −→ Gr p − (cid:96) V ⊗ (cid:18) (cid:96) ⊗ Ω B (log Z ) (cid:19) . There is a filtration ker θ ⊂ ker( θ ) ⊆ · · · ⊆ ker θ p +1 = Gr p V and Gr p V has graded quotientsker θ, ker θ ker θ , . . . , Gr p V ker θ p . The crucial observation (and what motivates the above use of (cid:96) ⊗ rather than Sym (cid:96) )is ker θ (cid:96) ker θ (cid:96) +1 (cid:44) → Gr p − (cid:96) +1 V ⊗ (cid:18) (cid:96) ⊗ Ω B (log Z ) (cid:19) lies in K p − (cid:96) +1 ⊗ (cid:18) (cid:96) − ⊗ Ω B (log Z ) (cid:19) where K p − (cid:96) +1 = ker (cid:110) Gr p − (cid:96) +1 V θ −→ Gr p − (cid:96) ⊗ Ω B (log Z ) (cid:111) . (VI.B.27)From this we infer that(i) K p , K p − , . . . , K o all have negative semi-definite curvature forms; ii) ker θ (cid:96) ker θ (cid:96) − (cid:44) → K p − (cid:96) +1 ⊗ (cid:18) (cid:96) − ⊗ Ω B (log Z ) (cid:19) which gives(iii) det (cid:16) ker θ (cid:96) ker θ (cid:96) − (cid:17) (cid:44) → ∧ d p,(cid:96) (cid:18) K p − (cid:96) +1 ⊗ (cid:18) (cid:96) − ⊗ Ω B (log Z ) (cid:19)(cid:19) .Using(iv) det Gr p V ∼ = p +1 ⊗ (cid:96) =1 det (cid:16) ker θ (cid:96) ker θ (cid:96) − (cid:17) and combining (iv), (iii) and (ii) at the level of curvatures gives (VI.B.24). (cid:3) Note : In [GGK08] there are results that in the 1-parameter case express the “errorterm” in the Arakelov inequality by quantities involving the ranks of the Kodaira-Spencer maps and structure of the monodromy at the singular points.VI.C.
The Iitaka conjecture.
One of the main steps in the general classification theory of algebraic varietieswas provided by a proof of the Iitaka conjecture. An important special case of thisconjecture is the
Theorem
VI.C.1 : Let f : X → Y be a morphism between smooth projective varietiesand assume that (i) Var f = dim Y (i.e., the Kodaira-Spencer maps are generically 1-1); (ii) the general fibre X y = f − ( y ) is of general type.Then the Kodaira-Iitaka dimensions satisfy (VI.C.2) κ ( X ) (cid:61) κ ( X y ) + κ ( Y ) . As noted in the introduction, this result was proved with one assumption (laterseen to not be necessary) by Viehweg ([Vie83a], [Vie83b]), and in general by Koll´ar[Kol87]. The role of positivity of the Hodge vector bundle had earlier been identifiedin [Fuj78], [Kaw81], [Kaw83], [Kaw85] and Ueno [Uen74], [Uen78]. Over the yearsthere has been a number of interesting results concerning the positivity of the Hodgevector bundle and, in the geometric case, the positivity of the direct images of thehigher pluricanonical series; cf. [PT14] and also [P˘a16] and [Sch15] for recent resultsand a survey of some of what is known together with further references.To establish (VI.C.2) one needs to find global sections of ω mX . From h ( ω mX ) = h ( f ∗ ω mX )and f ∗ ω mX = f ∗ ω mX/Y ⊗ ω mY , the issue is to find sections of f ∗ ω mX/Y . If for example f ∗ ω mX/Y is generically globallygenerated, then we have at least approximately h ( ω mX y ) · h ( ω mY ) sections of f ∗ ω mX which leads to the result. o find sections of f ∗ ω mX/Y , if for example Sym m f ∗ ω X/Y has sections, then sincethe multiplication mapping(VI.C.3) Sym m f ∗ ω X/Y → f ∗ ω mX/Y is injective on decomposable tensors, we get sections of the image. In general theissue is to find sections of Sym m (cid:48)(cid:48) f ∗ ω m (cid:48) X/Y and use an analogue of (VI.C.3).The arguments in [Vie83b], [Kol87] have two main aspects:(i) the use of Hodge theory;(ii) algebro-geometric arguments using Viehweg’s notion of weak positivity.In the works cited above, Hodge theory is used to show that under the assumptionsin (VI.C.1) for m (cid:29) κ (det f ∗ ω mX/Y ) = dim Y. From this one wants to infer that Sym m (cid:48)(cid:48) f ∗ ω m (cid:48) X/Y has sections for m (cid:48) , m (cid:48)(cid:48) (cid:29)
0. Thisis where (ii) comes in.The objectives of this section are twofold. One is to show that in case local Torelliholds for f : X → Y step (ii) may be directly circumvented by using the specialform Θ F = − t A ∧ A of the curvature in the Hodge vector bundle where A hasalgebro-geometric meaning that leads to positivity properties. The other is to dis-cuss Viehweg’s branched covering construction, which provides a mechanism to applythe positivity properties of the Hodge vector bundle to the pluri-canonical series.One issue has been that the assumptions (i), (ii) plus Viehweg’s branched coveringmethod give(VI.C.4) κ (cid:0) det (cid:0) f ∗ ω mX/Y (cid:1)(cid:1) = dim Y. From this one wants to show that(VI.C.5) (cid:16) det f ∗ ω mX/Y (cid:17) m (cid:48) haslots of sections = ⇒ (cid:40) Sym m (cid:48)(cid:48) (cid:16) f ∗ ω mX/Y (cid:17) haslots of sections (cid:41) . The Viehweg-Koll´ar method proves (VI.C.4) using the positivity properties of theHodge line bundle, and from this goes on to infer (VI.C.5) by an algebro-geometricargument involving Vieweg’s concept of weak positivity for a coherent sheaf. Thepositivity of the Hodge vector bundle does not enter directly.We will give a four-step sketch of the proof of Theorem VI.C.1, one that avoidsthe use of weak positivity. The first three steps follow from the discussions above.The fourth step uses a variant of Viehweg’s argument to derive Hodge theoreticinformation from the pluri-canonical series together with (III.B.7) and (III.B.8). As noted above, A is the end piece of the differential of the period mapping. The definition for a vector bundle was recalled above. tep one (already noted above): Suppose that the fibres of X f −→ Y are smooth andthat Φ ∗ ,n is injective at a general point. Then by Theorem III.B.1 κ (cid:16) Sym h n, ( f ∗ ω X/Y ) (cid:17) = dim P f ∗ ω X/Y (1) > dim Y. Step two : This is the same as step one but where we allow singular fibres. What isneeded to handle these follows from the discussion in Section IV.B.
Reformulation of step two : We set S k = Sym k and • F e = f ∗ ω X/Y ; • ω k = curvature form of O P S k F e (1)where ω k is the (1,1) form computed using the induced Hodge metric in S k F . Then ω k > P S k F e → Y . Using theinjectivity of the Kodaira-Spencer map T y Y → Hom( F ny , F n − y /F ny )at a general point y ∈ Y , from Theorem III.B.1 it follows that for all k (cid:61) h n, (VI.C.6) ω k > in the horizontal space at ageneral point in ( P S k F e ) y . This gives ω > P S k F e , which implies that O P S k F e (1) is big. This in turn implies the same for S k F e . Step three : Since S k f ∗ ω X/Y is big, S (cid:96) ( S k f ∗ ω X/Y ) is generically globally generated for (cid:96) (cid:29)
0. It follows that the direct summand S k(cid:96) f ∗ ω X/Y is generically globally generatedfor (cid:96) (cid:29)
0. Since S k(cid:96) f ∗ ω X/Y → f ∗ ω k(cid:96)X/Y is injective on decomposable tensors, we obtain at least approximately h (cid:16) ω k(cid:96)X y (cid:17) h (cid:0) ω k(cid:96)Y (cid:1) sections of ω k(cid:96)X . Remark VI.C.7:
In the geometric case an alternative geometric argument that S k f ∗ ω X/Y is big may be given as follows:First, for a family W g −→ Y with smooth general fibre W y = g − ( y ), the condition(III.B.3) for bigness of g ∗ ω W/Y is:(i) for general y ∈ Y the Kodaira-Spencer map ρ y : T y Y → H ( T W y ) should beinjective;(ii) for general ψ ∈ ( g ∗ ω W/Y ) y = H (Ω nW y ) the map H ( T W y ) ψ −→ H (Ω n − W y )should be injective on the image ρ y ( T y Y ) ⊂ H ( T W y ). ext, for a famliy X → Y with generically injective Kodaira-Spencer mappings,we set W = X k (cid:122) (cid:125)(cid:124) (cid:123) × Y × · · · × Y X. Then H (Ω nW ) contains k ⊗ H (Ω nX ) as a direct summand and the same argument asin the proof of Theorem (III.B.7) gives that k ⊗ f ∗ ω X/Y is big.Finally, we may apply a similar argument to a desingularization of the quotient of W → Y by the action of the symmetric group. Then the general fibre is X ( k ) y = desingularization of Sym k X y with S k H (Ω nX y ) = direct summand of H (Ω nX ( k ) y )and again the argument in the proof of Theorem III.B.7 will apply. Step four : The idea is to apply the reformulation of step two with f ∗ ω mX/Y replacing f ∗ ω X/Y . This will be discussed below in which the pluricanonical series of a smoothvariety will be seen to have Hodge theoretic interpretations. An alternate more direct approach would be to have a metric in O P S k f ∗ ω mX/Y (1)whose Chern form ω is positive in the space at a general point of P S k f ∗ ω mX/Y . Forthis approach to work one would need to have a metric in f ∗ ω mX/Y that has a property analogous to that obtained using the Kodaira-Spencer map inthe case m = 1 considered above. Here one possibility might be to use the relativeBergman kernel metrics that have appeared from the recent work of a number ofpeople; cf. [PT14] and the references cited therein. This possibility will be discussedfurther at the end of Section VI.C.Before giving the detailed discussion we will give some general comments.(a) A guiding heuristic principle is(VI.C.8) For families f : X → Y of varieties of general type the f ∗ ω kX/Y become more positive as k increases. The reasons are An alternative approach to the Iitaka conjecture which replaces the use of Hodge theory byvanishing theorems and also uses the cyclic covering trick has been used by Koll´ar (
Ann. of Math. (1986), 11–42). The relation between vanishing theorems and Hodge theory is classical datingto Kodaira-Spencer and Akizuki-Nakano in the 1950’s. As noted by a number of people, Hodgetheory in some form and the curvature properties of the Hodge bundles seem to generally be lurkingbehind the positivity of direct images of pluricanonical sheaves. f ∗ ω X/Y (cid:61) S k f ∗ ω X/Y becomes more positive with k by the argument in the proof of Theorem III.B.7; • the map S (cid:96) f ∗ ω kX/Y → f ∗ ω k · (cid:96)X/Y is non-trivial since it is injective on decomposable tensors (similar to the proofof Clifford’s theorem); • passing to the quotient increases positivity (curvatures increase on quotientbundles).(b) In what follows we will without comment make simplifying modifications byreplacing f : X → Y by f (cid:48) : X (cid:48) → Y (cid:48) where all the varieties are smooth and where • Y (cid:48) → Y is an isomorphism outside a codimension 2 subvariety of Y ; • X (cid:48) is a desingularization of X × Y (cid:48) Y and f (cid:48) is flat; • the fibres X (cid:48) y (cid:48) = f (cid:48) − ( y (cid:48) ) are smooth outside a normal crossing divisor in Y (cid:48) around which the local monodromies are unipotent ([Vie83b], page 577).(c) We now recall the two constructions from the introduction that associate to f ∗ ω mX/Y a variation of Hodge structure; these constructions will be done first for afixed smooth W and then for a family f : X → Y where W is a typical general fibre X y . The objectives are to illustrate how pluricanonical series can give rise to Hodgestructures. (i) The case of fixed W , Hodge structures associated to H ( K mW ) . We summarize and establish notation for the standard construction of a cycliccovering (cid:102) W ψ → W associated to ψ ∈ H ( K mW ) whose divisor ( ψ ) ∈ | mK X | is smooth. The constructionis(VI.C.9) (cid:102) W ψπ (cid:15) (cid:15) ⊂ K W (cid:15) (cid:15) = total space of the line bundle W = W where (cid:102) W ψ = { ( w, η ) : η ∈ K W,w , η m = ψ ( w ) } . Then the direct image(VI.C.10) ψ ∗ K (cid:102) W ψ ∼ = m − ⊕ i =0 K m − iW which gives(VI.C.11) H ( K (cid:102) W ψ ) ∼ = m − ⊕ i =0 H ( K m − iW ) . In this way pluri-differentials on W become ordinary differentials on (cid:102) W ψ ; this is theinitial step in the relation between the pluricanonical series and Hodge theory. e observe that the cyclic group G m acts on (cid:102) W ψ → W with the action by ζ = e πi/m on H ( K (cid:102) W ψ ) and by ζ i +1 on H ( K m − iX ) in (VI.C.11).We also observe the diagram (cid:102) W ψ (cid:15) (cid:15) ∼ (cid:47) (cid:47) (cid:102) W λψ (cid:15) (cid:15) W W (VI.C.12)induced by scaling the action of λ ∈ C ∗ on K W → W in (VI.C.9). The isomorphism ∼ −→ in (VI.C.12) depends on the choice of λ /m . We denote by H ( K mW ) ⊂ H ( K mW ) the ψ ’s with smooth divisor ( ψ ), and we set P = P H ( K mW ) ⊂ P = H ( K mW ) . Then H ( O P (1)) ∼ = H ( K mW ) ∗ , and the identity gives a canonical section(VI.C.13) Ψ ∈ H ( W × P , K mW (cid:2) O P (1)) . Definition: (Ψ) is the universal divisor of the ( ψ ) ’s for ψ ∈ H ( K mW ).To construct a universal family of cyclic coverings (cid:102) W ψ → W it is necessary tochoose an auxiliary cyclic covering (cid:101) P q −→ P , with q ∗ O P (1) = L m for an ample line bundle L → (cid:101) P . From W × (cid:101) P id × g −−−→ W × P we obtain H ( W × (cid:101) P , K mW (cid:2) L m ) ∼ = H ( W × (cid:101) P , ( K W (cid:2) L ) m )and using the pullback to W × (cid:101) P of Ψ in (VI.C.13) there is a cyclic covering (cid:102) W Ψ := (cid:94) W × P h −→ W × P branched over the universal divisor (Ψ) and depending on the choice of (cid:101) P → P . Inthis way the choice of isomorphisms in (VI.C.12) necessitated by choosing an m th root of λ may be made uniform. We observe that(VI.C.14) (cid:40) (cid:101) Ψ ∈ H ( K (cid:102) W Ψ / (cid:101) P ) (cid:101) Ψ m = p ∗ Ψ . Setting (cid:102) W = h − ( W × P ) the total space and the fibres of (cid:102) W → (cid:101) P are irreducible and smooth. This gives a period mapping, or equivalently a VHS,(VI.C.15) Φ : (cid:101) P → Γ \ D. roposition VI.C.16 : The Hodge vector bundle F Ψ ∼ = m − ⊕ i =0 H ( K m − iW ) ⊗ L i +1 . Arguments as in sections (1.4)–(1.7) of [Vie83b] show that for m (cid:29) Proposition
VI.C.17 : For m (cid:29) the part Φ ∗ ,n : T (cid:101) P → Hom (cid:0) H ( K mW ) ⊗ L, F n − /F n Ψ (cid:1) of the end piece of Φ ∗ ,n is injective. This is proved by showing that Φ ∗ is injective on certain of the eigenspaces for the G -equivariant action in the picture, and that the ζ -eigenspace is among those thatare included. Discussion of singularities : For ψ ∈ H ( K mW ) the Finsler-type norm (cid:107) ψ (cid:107) = (cid:90) W ( ψ ∧ ψ ) /m = (cid:90) (cid:102) W ψ (cid:101) ψ ∧ (cid:101) ψ is equal to the square of the Hodge length of (cid:101) ψ = ψ /m ∈ H ( K (cid:102) W ψ ). Even when thedivisor ( ψ ) acquires singularities so that (cid:102) W ψ becomes singular, the Hodge length ofthe canonical section (cid:101) ψ will remain finite. However, although (cid:107) ψ (cid:107) is continuous in ψ , it is not smooth as its derivatives detect singularities of the degenerating Hodgestructures. In terms of limiting mixed Hodge structures, (cid:101)
Ψ lies in the lowest possibleweight part.The second method of associating Hodge theoretic data to the pluricanonical series H ( K mW ) is the following: Let (cid:102) W Ψ be a desingularization of a completion of (cid:102) W and (cid:102) W Ψ (cid:101) π −→ P the resulting fibration. If dim (cid:102) W Ψ = ˜ n , then H ˜ n ( (cid:102) W Ψ ) has a polarized Hodge structure.The general Hodge theory of maps such as (cid:101) π is contained in the decomposition theorem ([dCM09]). In the case at hand a special feature arises in that from PropositionVI.C.16 we may infer that(VI.C.18) H ˜ n, ( (cid:102) W Ψ ) contains as a direct summand m − ⊕ i =0 H ( K m − iW ) ⊗ H ( O P ( n i )) . Here the n i > ν i > q ∗ L i +1 ∼ = ⊕ O P ( ν i ) . Norms of this type appear in [NS68] and have been used extensively in the recent literature (cf.[P˘a16] for a summary and survey). ii) The case of a family f : X → Y ; Hodge structures associated to f ∗ ω mX/Y . The following is a sketch of the proof of Theorem VI.C.1. It is intended to pointout some of the geometric aspects of the arguments in [Vie83b], in particular the wayin which the Hodge-theoretic interpretations enter into those arguments, referring tothat paper for the details.Since the publication of [Vie83b] several important general results concerning thepluricanonical series have been established (cf. [Dem12a]), and we shall assume thefollowing: • for general y ∈ Y , X y is smooth and there is an (cid:96) such that for all m = k(cid:96) , k (cid:61)
1, the linear system | mK X y | is ample; whenever an m appears below itwill be of this form; and • the assumptions to have local Torelli in the form used in [Vie83b] are satisfiedfor f ∗ ω mX/Y .The basic diagram is (cid:101) X ˜ f (cid:15) (cid:15) (cid:47) (cid:47) X h (cid:15) (cid:15) (cid:47) (cid:47) X (cid:47) (cid:47) h (cid:15) (cid:15) X f (cid:15) (cid:15) (cid:101) P g (cid:47) (cid:47) (cid:101) P q (cid:47) (cid:47) P p (cid:47) (cid:47) Y (VI.C.19)where • P = P f ∗ ω mX/Y , so that p ∗ O P (1) = f ∗ ω m ∗ X/Y is the dual of f ∗ ω mX/Y ; • X = X × Y P ; • (cid:101) P q −→ P is a cyclic branched covering where there is an ample line bundle L → (cid:101) P with q ∗ O P (1) = L m ; • X = X × P (cid:101) P ; and • (cid:101) X → X is the cyclic covering obtained by globalizing the construction of (cid:102) W Ψ → W × (cid:101) P given by the completion of (cid:102) W in (VI.C.14) above.The players in the basic diagram are • f − ( y ) = X y ; • p − ( y ) = P H ( ω mX y ), whose points are [ ψ ] where ψ ∈ H ( ω mX y ); • q − ([ ψ ]) = { [ ψ i ] } where [ ψ i ] q −→ [ ψ ] under the cyclic covering; • Ψ is the tautological section of ω m X / P ⊗ h ∗ O P (1); • (cid:101) Ψ is the tautological section of ω m X / (cid:101) P ⊗ ˜ h ∗ L m = (cid:16) ω X / (cid:101) P ⊗ ˜ h ∗ L (cid:17) m ; • (cid:101) X → X is the m -sheeted cyclic covering obtained by extracting an m th rootof (cid:101) Ψ. e will denote the fibre over y of (cid:101) X → Y by(VI.C.20) (cid:101) X y = (cid:91) [ ψ ] ∈ P H (cid:16) ω mXy (cid:17) (cid:101) X y,ψ where X y corresponds to W and (cid:101) X y,ψ to (cid:102) W ψ above.There are two families of varieties constructed from the basic diagram. Denotingthe composition q ◦ g ◦ ˜ f by G for the first we have(VI.C.21) G : (cid:101) X → P whose fibre over ( y, [ ψ ]) ∈ P is (cid:101) X y,ψ . The second is(VI.C.22) F : (cid:101) X → Y whose fibre over y ∈ Y is the variety (VI.C.20).There are two important observations concerning these families: Generic local Torelli holds for both families; and(VI.C.23)
The basic diagram (VI.C.19) is commutative. (VI.C.24)The fibres of G ∗ ω (cid:101) X / P are given by (cid:0) G ∗ ω (cid:101) X / P (cid:1) ( y, [ ψ ]) = H (cid:0) ω (cid:101) X y,ψ (cid:1) . From (VI.C.23) we have(VI.C.25) G ∗ ω (cid:101) X / P (cid:61) and det G ∗ ω (cid:101) X / P > on an open set. From (VI.C.25) we have(VI.C.26)
For k (cid:29) , both S k G ∗ ω (cid:101) X / P and S k F ∗ ω (cid:101) X /Y are big. To complete the proof of Theorem VI.C.1 the argument one first might try to makeis this: H ( ω mX y ) is a direct factor of H ( ω (cid:101) X y,ψ ) , and from the commutativityof the basic diagram (VI.C.19) it follows that f ∗ ω mX/Y is a direct factorof the Hodge vector bundle associated to the family (cid:101) X → Y . Bythe local Torelli property (VI.C.23) with the implication (VI.C.26) , itfollows that S k f ∗ ω mX/Y is big. However, this argument is not correct; the issue is more subtle. The problem is thatunder the mapping (cid:101) X H −→ X in (VI.C.19) we do not have H ∗ ω (cid:101) X / P ∼ = m − ⊕ i =0 ω m − i X / P , but rather H ∗ ω (cid:101) X / P ∼ = m − ⊕ i =0 ω m − i X / P ⊗ h ∗ O P ( i + 1) here the O P ( i + 1)’s reflect the global twisting of the identification (VI.C.10). Thiswould not be an issue if O P (1) were positive. This positivity trivially holds alongthe fibres of P → Y , but since p ∗ O P (1) = f ∗ ω m ∗ X/Y any positivity of f ∗ ω mX/Y becomesnegativity of O P (1) in directions normal to fibres of P → Y . An additional step isrequired.The key observation is that (cf. [Vie83b], page 587)For a > ω ma + a +1 X / P ⊗ h ∗ O P ( a ) is a direct summand of G ∗ ω a +1 (cid:101) X / P .It is the additional factor of a + 1 in ω ma + a +1 X / P that offsets the negativity of O P ( a ) inthe normal direction of the fibres of P → Y .Thus what needs to be shown is(VI.C.27) S k G ∗ ω a +1 (cid:101) X / P is big for k (cid:29) . This follows from (VI.C.27) if we have S a +1 G ∗ ω (cid:101) X / P → G ∗ ω a +1 (cid:101) X / P → , and this may be accomplished generically in Y by choosing m (cid:29) Remark VI.C.28:
We conclude with a comment and a question. The comment isthat perhaps the most direct way to prove the Iitaka Conjecture VI.C.2 would beto use the curvature properties of the Finsler-type metric in f ∗ ω mX/Y . Specifically,referring to [Ber09], [BP˘a12] and [PT14] for details we set P ∗ = P ( f ∗ ω mX/Y ) ∗ . Then O P ∗ (1) ˆ π −→ Y has fibres(VI.C.29) O P ∗ (1) ( y, [ η ]) = H ( ω mX y ) /η ⊥ , η ∈ H ( ω mX y ) ∗ and ˆ π ∗ O P ∗ (1) = f ∗ ω mX/Y . Following [Kaw82] and [PT14], we define a metric in O P ∗ (1) by taking the infinimumof the (cid:107) ψ (cid:107) ’s where ψ ∈ H ( ω mX y ) projects to a fixed vector in the quotient (VI.C.29).Leaving aside the question of singularities, for this metric the form ω m satisfies ω m (cid:61) . We do not expect to have ω m > P ∗ : this is already not thecase when m = 1. Instead we carry out a similar construction replacing f ∗ ω mX/Y by S m (cid:48) f ∗ ω mX/Y and denote by ω m (cid:48) ,m the Chern form of the corresponding O (1)-bundle. Question:
Assuming
Var f = dim Y , for m (cid:48) (cid:29) do we have ω m (cid:48) ,m > in an openset? This is true when m = 1, and if it holds for general m and the issue of singularitiescan be handled one would have a direct “curvature” proof of VI.C.2. In fact, heuristicreasoning suggests the following onjecture VI.C.30 : Under the assumptions in Theorem VI.C.1, let m (cid:61) besuch that for general y ∈ Y the bundle K mX y is globally generated. Then for P m ( X y ) = h ( K mX y ) Sym m (cid:48) f ∗ ω mX/Y is big for m (cid:48) (cid:61) P m ( X y ) . VI.D.
The Hodge vector bundle may detect extension data.
Proposition
VI.D.1 : On a subvariety Y ⊂ Z ∗ I along which the period mapping Φ I is locally constant, the extended Hodge bundle F e,I → Y is flat. It may however havenon-trivial monodromy.Proof. F e,I is filtered by W • ( N I ) ∩ F e,I . For V I the local system corresponding to Φ I : Z ∗ I → Γ I \ D I , the Gauss-Manin connection ∇ acting on O Z ∗ I ( V I ) preserves W • ( N I ),and on the associated graded to W • ( N I ) V I it preserves the associated graded to W • ( N I ) ∩ F e,I . It follows that ∇ preserves F e,I ⊂ O Y ( V ). (cid:3) Example VI.D.2:
On an algebraic surface S suppose we have a smooth curve (cid:101) C of genus g ( (cid:101) C ) ≥
2, and that through each pair of distinct points p, q ∈ (cid:101) C there is aunique rational curve meeting (cid:101) C at two points. (cid:101) C P p q Suppose moreover that along the diagonal D ⊂ (cid:101) C × (cid:101) C the P becomes simply tangent.Then (cid:101) C + P is a nodal curve, and for m (cid:29) | mω (cid:101) C + P | contracts the P and we obtain an irreducible stable curve C p,q with arithmetic genus p a ( C p,q ) = 3. p = q As P becomes tangent we obtain a cusp.The extension data for the MHS is given (cf. [Car80]) byAJ (cid:101) C ( p − q ) ∈ J ( (cid:101) C ) . When we turn around p = q we interchange p, q ; after base change t → t theextension class is well defined locally. Globally, it is another story.The vector space H ( ω C p,q ), which is the fibre of the canonically extended Hodgebundle, has the 2-dimensional fixed subspace H (Ω (cid:101) C ) and variable 1-dimensionalquotient represented by a differential of the third kind ϕ p,q ∈ H (Ω (cid:101) C ( p + q )) havingnon-zero residues at p, q . As p → q the differential ϕ p,q tends to a differential of he second kind ϕ p ∈ H (Ω (cid:101) C (2 p )) with non-zero polar part at p . For the globalmonodromy over Y = (cid:101) C × (cid:101) C \ D , the extension class is given for ψ ∈ H (Ω (cid:101) C ) by ψ → (cid:90) qp ψ (modulo periods).The action of π ( Y ) on H ( (cid:101) C, { p, q } ) may then be shown to give an infinite subgroupof H ( (cid:101) C, Z ), which implies the assertion. Claim
VI.D.3 : The Hodge vector bundle may detect continuous extension data.
Here continuous extension data means the extension classes that arise in the inducedfiltration of F e → Y where there is a variation of limiting mixed Hodge structure over Y with F e the Hodge vector bundle. Example VI.D.4:
We let (cid:101) C be a smooth curve with g ( (cid:101) C ) ≥ p ∈ (cid:101) C a fixedpoint. We construct a family C q , p ∈ (cid:101) C , of stable curves as follows. • For q (cid:54) = p we identify p, qpq (cid:101) C (cid:45) C q • For p = q , we obtain a curve Ep r (cid:101) C (VI.D.5)In this way we obtain a VMHS parametrized by (cid:101) C and with trivial monodromy.For the filtration on the canonically extended Hodge bundle there is a fixed part W ( N ) ∩ F e ∼ = H (Ω (cid:101) C ) , and a variable part whose quotient is W ( N ) ∩ F e /W ( N ) ∩ F e = (cid:40) C ϕ p,q for q (cid:54) = p C ϕ E,r for q = p. The notation means that ϕ E,r is the differential of the third kind on the normalizationof E and with residues at ± ϕ p,q and ϕ E,r . ombining the above we have over (cid:101) C an exact sequence0 → H (Ω (cid:101) C ) ⊗ O (cid:101) C → F e → O (cid:101) C → e = “Identity” ∈ H ( O (cid:101) C ) ⊗ H (Ω (cid:101) C ) . VI.E.
The exterior differential system defined by a Chern form.
In this sec-tion we will discuss the exterior differential system(VI.E.1) ω = 0defined by the Chern form of the line bundle O P E (1) where E → X is a Hermitianvector bundle whose curvature has the norm positivity property (III.A.1). Withoutassuming the norm positivity property, this type of EDS has been previously studiedin [BK77] and [Som59] and also appeared in [Kol87].Here our motivation is the following question:(VI.E.2) Under what conditions can one say that the Kodaira-Iitaka dimensionof E → X is equal to its numerical dimension? Proposition
VI.E.3 : The exterior differential system (VI.E.1) defines a foliation of P E by complex analytic subvarieties W ⊂ P E with the properties(i) W meets the fibres of P E π −→ X transversely; thus W → π ( W ) is an ´etal´emap;(ii) the restriction E (cid:12)(cid:12) π ( W ) is flat.Proof. Since ω > P E → X , the vectors ξ ∈ T ( x, [ e ]) P E that satisfy ω ( ξ ) = 0 project isomorphically to T X . The image of these vectors is the subspace(here identifying ξ with π ∗ ( ξ ))(VI.E.4) { ξ ∈ T x X : A ( e ⊗ ξ ) = 0 } . This is the same as the subspace of T x X defined byΘ E ( e ⊗ ξ ) = 0 , which implies that E (cid:12)(cid:12) π ( w ) is flat. (cid:3) Remark:
Given any holomorphic bundle map(VI.E.5) A : T X ⊗ E → G, if we have a metric in E → X we may use it to identify E ∼ = E ∗ and then definethe horizontal sub-bundle H ⊂ T O P E (1). It follows that (VI.E.4) defines a C ∞ distribution (with jumping fibre dimensions) in T O P E (1), and when the map (VI.E.5)arises from the curvature of the metric connection as in (III.A.3) this distribution isintegrable and the maximal leaves of the corresponding foliation of P E by complexanalytic subvarieties are described by Proposition VI.E.3.The restrictions E (cid:12)(cid:12) π ( W ) being flat, the monodromy is discrete. Heuristic argumentssuggest that the maximal leaves W ⊂ P E are closed analytic subvarieties. onjecture VI.E.6 : Finite monodromy provides the necessary and sufficient con-dition to have the equality κ ( E ) = n ( E ) of Kodaira-Iitaka and numerical dimensions of a holomorphic vector bundle having aHermitian metric whose curvature satisfies the norm positivity condition. The idea is that the quotient P E/ ∼ , where ∼ is the equivalence relation given bythe connected components of the foliation defined by (VI.E.1), exists as a complexanalytic variety of dimension equal to n ( E ), and there is a meromorphic mapping P E (cid:57)(cid:57)(cid:75) P E/ ∼ together with an ample line bundle on P E/ ∼ that pulls back to O P E (1). The rathersimple guiding model here is the dual of the universal sub-bundle over the Grassman-nian that was discussed above. In fact, the conjecture holds if E → X is globallygenerated with metrics induced from the corresponding mapping to a Grassmannian.We note that the foliation defined by the null space of the holomorphic bi-sectionalcurvature on quotients of bounded symmetric domains has been studied in [Mok87].In this case the leaves are generally not closed.Finally we point out the very interesting papers [CD17a] and [CD17b]. In thesepapers the authors construct examples of smooth fibrations f : X → B of a surface over a curve such that for E = f ∗ ω X/B one has E = A ⊕ Q where A is an ample vector bundle and Q is a flat U ( m, C )-bundle with infinitemonodromy group. In this case the leaves of the EDS (VI.E.1) may be described asfollows: For each b ∈ B we have P Q ∗ b ⊂ P E ∗ b and using the flat connection on Q ∗ the parallel translate of any point in P Q ∗ b definesan integral curve of the EDS. References [AK00]
D. Abramovich and
K. Karu , Weak semistable reduction in characteristic 0,
In-vent. Math. no. 2 (2000), 241–273. MR 1738451. Zbl 0958.14006. Available athttps://doi.org/10.1007/s002229900024.[Ara71] A. Arakelev, Formulas of algebraic curves with fixed degeneracies,
Izv. Akad. Nauk. SSSRSoc. Math. (1971), 1277–1302.[BK77] E. Bedford and
M. Kalka , Foliations and complex Monge-Amp`ere equations,
Comm.Pure Appl. Math. no. 5 (1977), 543–571. MR 0481107. Zbl 0351.35063. Available athttps://doi.org/10.1002/cpa.3160300503. We note that Q ⊂ f ∗ ω X/B ⊂ R f C X is not flat relative to the Gauss-Manin convention on R f C X . Ber09]
B. Berndtsson , Curvature of vector bundles associated to holomorphic fibrations,
Ann.of Math. (2) no. 2 (2009), 531–560. MR 2480611. Zbl 1195.32012. Available athttps://doi.org/10.4007/annals.2009.169.531.[BP˘a12]
B. Berndtsson and
M. P˘aun , Quantitative extensions of pluricanonical formsand closed positive currents,
Nagoya Math. J. (2012), 25–65. MR 2891164.Zbl 1248.32012. Available at http://projecteuclid.org/euclid.nmj/1330611001.[Bru16a]
Y. Brunebarbe , A strong hyperbolicity property of locally symmetric varieties, 2016.arXiv 1606.03972.[Bru16b] , Symmetric differentials and variations of Hodge structures, 2016,
J. reine angew.Math. , published on-line. https://doi.org/10.1515/crelle-2015-0109.[BKT13]
Y. Brunebarbe , B. Klingler , and
B. Totaro , Symmetric differentials and thefundamental group,
Duke Math. J. no. 14 (2013), 2797–2813. MR 3127814.Zbl 1296.32003. Available at https://doi.org/10.1215/00127094-2381442.[CMSP17]
J. Carlson , S. M¨uller-Stach , and
C. Peters , Period mappings and period domains , Cambridge Studies in Advanced Mathematics , 2nd edition, Cambridge UniversityPress, Cambridge, 2017. MR 2012297. Zbl 1030.14004.[Car80] J. A. Carlson , Extensions of mixed Hodge structures, in
Journ´ees de G´eometrieAlg´ebrique d’Angers, Juillet 1979/Algebraic Geometry, Angers, 1979 , Sijthoff & No-ordhoff, Alphen aan den Rijn—Germantown, Md., 1980, pp. 107–127. MR 0605338.Zbl 0471.14003.[dCM09]
M. A. A. de Cataldo and
L. Migliorini , The decomposition theorem, per-verse sheaves and the topology of algebraic maps,
Bull. Amer. Math. Soc.(N.S.) no. 4 (2009), 535–633. MR 2525735. Zbl 1181.14001. Available athttps://doi.org/10.1090/S0273-0979-09-01260-9.[CD17a] F. Catanese and M. Dettweiler, Answer to a question by Fujita on variation of Hodgestructures, in Higher Dimensional Algebraic Geometry—In Honour of Professor YujiroKawamata’s Sixtieth Birthday , 73–102,
Adv. Stud. Pure Math. , Math. Soc. Japan,Tokyo, 2017.[CD17b] ,[CDK95] E. Cattani , P. Deligne , and
A. Kaplan , On the locus of Hodge classes,
J.Amer. Math. Soc. no. 2 (1995), 483–506. MR 1273413. Zbl 0851.14004. Available athttps://doi.org/10.2307/2152824.[CKS86] E. Cattani , A. Kaplan , and
W. Schmid , Degeneration of Hodge structures,
Ann.of Math. (2) no. 3 (1986), 457–535. MR 0840721. Zbl 0617.14005. Available athttps://doi.org/10.2307/1971333.[Del70]
P. Deligne , ´Equations diff´erentielles `a points singuliers r´eguliers , Lecture Notesin Mathematics, Vol. 163 , Springer-Verlag, Berlin-New York, 1970. MR 0417174.Zbl 0244.14004.[Dem12a]
J.-P. Demailly , Analytic methods in algebraic geometry , Surveys of Modern Mathe-matics , International Press, Somerville, MA; Higher Education Press, Beijing, 2012.MR 2978333. Zbl 1271.14001.[Dem12b] , Hyperbolic algebraic varieties and holomorphic differential equations, ActaMath. Vietnam. no. 4 (2012), 441–512. MR 3058660. Zbl 1264.32022.[Dem16] , Variational approach for complex Mondge-Amp`ere equations and geometric ap-plications [after Berman, Berndtsson, Boucksom, Eyssidieux, Guedj, Jonsson, Zeri-ahi,. . . ] , Sem. Bourbaki no. 1112, 2016.[Den18] Ya Deng,
Pseudo Kobayashi hyperbolicity of base spaces of families of minimal projectivemanifolds with maximal variation , arXiv:1809.005891v1 [math.AG] 16Sep2018.[Don15]
S. Donaldson , Stability of algebraic varieties and K¨ahler geometry, to appear in
Proc.Amer. Math. Soc. Sumer School in Algebraic Geometry (Salt Lake City 2015). Don16] , K¨ahler-Einstein metrics and algebraic geometry, in
Current developmentsin mathematics 2015 , Int. Press, Somerville, MA, 2016, pp. 1–25. MR 3642542.Zbl 06751629.[FPR15a]
M. Franciosi , R. Pardini , and
S. Rollenske , Computing invariants of semi-log-canonical surfaces,
Math. Z. no. 3-4 (2015), 1107–1123. MR 3369370. Zbl 1329.14076.Available at https://doi.org/10.1007/s00209-015-1469-9.[FPR15b] , Log-canonical pairs and Gorenstein stable surfaces with K X = 1, Com-pos. Math. no. 8 (2015), 1529–1542. MR 3383166. Zbl 1331.14037. Available athttps://doi.org/10.1112/S0010437X14008045.[FPR17] , Gorenstein stable surfaces with K X = 1 and p g > Math. Nachr. no. 5-6 (2017), 794–814. MR 3636379. Zbl 06717887. Available at https://doi.org/10.1002/mana.201600090.[Fuj78]
T. Fujita , On K¨ahler fiber spaces over curves,
J. Math. Soc. Japan no. 4 (1978), 779–794. MR 0513085. Zbl 0393.14006. Available at https://doi.org/10.2969/jmsj/03040779.[GGK08] M. Green, P. Griffiths and M. Kerr, Some enumerative global properties of variations ofHodge structure, Moscow Math. (2009), 469–530.[GGLR17] M. Green , P. Griffiths , R. Laza , and
C. Robles , Completion of period mappingsand ampleness of the Hodge bundle, 2017. arXiv 1708.09523v1.[Gri69]
P. A. Griffiths , Hermitian differential geometry, Chern classes, and positive vectorbundles, in
Global Analysis (Papers in Honor of K. Kodaira) , Univ. Tokyo Press, Tokyo,1969, pp. 185–251. MR 0258070.[Gri70] , Periods of integrals on algebraic manifolds. III. Some global differential-geometric properties of the period mapping,
Inst. Hautes ´Etudes Sci. Publ. Math. no. 38(1970), 125–180. MR 0282990. Zbl 0212.53503.[Kaw81]
Y. Kawamata , Characterization of abelian varieties,
Compositio Math. no. 2 (1981),253–276. MR 0622451. Zbl 0471.14022.[Kaw82] , Kodaira dimension of algebraic fiber spaces over curves, Invent.Math. no. 1 (1982), 57–71. MR 0652646. Zbl 0461.14004. Available athttps://doi.org/10.1007/BF01404756.[Kaw83] , Kodaira dimension of certain algebraic fiber spaces, J. Fac. Sci. Univ. TokyoSect. IA Math. no. 1 (1983), 1–24. MR 0700593. Zbl 0516.14026.[Kaw85] , Minimal models and the Kodaira dimension of algebraic fiber spaces, J.Reine Angew. Math. (1985), 1–46. MR 0814013. Zbl 0589.14014. Available athttps://doi.org/10.1515/crll.1985.363.1.[Kaw02] , On algebraic fiber spaces, in
Contemporary trends in algebraic geometryand algebraic topology (Tianjin, 2000) , Nankai Tracts Math. , World Sci. Publ.,River Edge, NJ, 2002, pp. 135–154. MR 1945358. Zbl 1080.14507. Available athttps://doi.org/10.1142/97898127774160006.[Kol87] J. Koll´ar , Subadditivity of the Kodaira dimension: fibers of general type, in
Algebraicgeometry, Sendai, 1985 , Adv. Stud. Pure Math. , North-Holland, Amsterdam, 1987,pp. 361–398. MR 0946244. Zbl 0659.14024.[Kol13] , Moduli of varieties of general type, in Handbook of moduli. Vol. II , Adv.Lect. Math. (ALM) , Int. Press, Somerville, MA, 2013, pp. 131–157. MR 3184176.Zbl 1322.14006.[Ko-Mo] J. Koll´ar and S. Mori, Birational geometry of algebraic varieties, Cambridge Tracts inMath. , Cambridge Univ. Press, 1998.[Laz04]
R. Lazarsfeld , Positivity in algebraic geometry. II , Ergebnisse der Mathematik undihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results inMathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics] , Springer-Verlag, Berlin, 2004, Positivity for vector bundles, and multiplier ideals.MR 2095472. Zbl 1093.14500. Available at https://doi.org/10.1007/978-3-642-18808-4. Mok87]
N. Mok , Uniqueness theorems of Hermitian metrics of seminegative curvature on quo-tients of bounded symmetric domains,
Ann. of Math. (2) no. 1 (1987), 105–152.MR 0873379. Zbl 0616.53040. Available at https://doi.org/10.2307/1971290.[Mok12] , Projective-algebraisity of minimal compactifications of complex hyperbolicspace forms of finite volume,
Perspectives in Analysis, Geometry and Topology ,Birkh¨auser/Springer, NY (2012), 331–354.[MT07]
C. Mourougane and
S. Takayama , Hodge metrics and positivity of direct images,
J. Reine Angew. Math. (2007), 167–178. MR 2337646. Zbl 1128.14030. Available athttps://doi.org/10.1515/CRELLE.2007.039.[MT08] , Hodge metrics and the curvature of higher direct images,
Ann. Sci. ´Ec.Norm. Sup´er. (4) no. 6 (2008), 905–924. MR 2504108. Zbl 1167.14027. Availableat https://doi.org/10.24033/asens.2084.[NS68] M. S. Narasimhan and
R. R. Simha , Manifolds with ample canonical class,
Invent. Math. (1968), 120–128. MR 0236960. Zbl 0159.37902. Available athttps://doi.org/10.1007/BF01425543.[PS08] C. A. M. Peters and
J. H. M. Steenbrink , Mixed Hodge structures , Ergebnisse derMathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics[Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys inMathematics] , Springer-Verlag, Berlin, 2008. MR 2393625. Zbl 1138.14002.[P˘a16] M. P˘aun , Singular Hermitian metrics and positivity of direct images of pluricanonicalbundles, 2016. arXiv 1606.00174.[PT14]
M. P˘aun and
S. Takayama , Positivity of twisted relative pluricanonical bundles andtheir direct images, 2014. arXiv 1409.5505v1.[Sch15]
C. Schnell , Weak positivity via mixed Hodge modules, in
Hodge theory andclassical algebraic geometry , Contemp. Math. , Amer. Math. Soc., Prov-idence, RI, 2015, pp. 129–137. MR 3445002. Zbl 1360.14027. Available athttps://doi.org/10.1090/conm/647/12957.[Sim92] ??? Simpson, Higgs bundles and local systems,
Publ. Math. IHES (1992), 5–95.[Som59] F. Sommer , Komplex-analytische Bl¨atterung reeler Hyperfl¨achen im C n , Math. Ann. (1959), 392–411. MR 0108821. Zbl 0092.29903. Available athttps://doi.org/10.1007/BF01360840.[Som78]
A. J. Sommese , On the rationality of the period mapping,
Ann. Scuola Norm. Sup.Pisa Cl. Sci. (4) no. 4 (1978), 683–717. MR 0519890. Zbl 0396.32004.[Uen74] K. Ueno , Introduction to classification theory of algebraic varieties and compact com-plex spaces, in
Classification of algebraic varieties and compact complex manifolds , Lec-ture Notes in Math. , Springer, Berlin, 1974, pp. 288–332. MR 0361174.[Uen78] , Classification of algebraic varieties. II. Algebraic threefolds of parabolic type,in
Proceedings of the International Symposium on Algebraic Geometry (Kyoto Univ.,Kyoto, 1977) , Kinokuniya Book Store, Tokyo, 1978, pp. 693–708. MR 0578882.[Vie83a]
E. Viehweg , Weak positivity and the additivity of the Kodaira dimension for certainfibre spaces, in
Algebraic varieties and analytic varieties (Tokyo, 1981) , Adv. Stud. PureMath. , North-Holland, Amsterdam, 1983, pp. 329–353. MR 0715656. Zbl 0513.14019.[Vie83b] , Weak positivity and the additivity of the Kodaira dimension. II. The local Torellimap, in Classification of algebraic and analytic manifolds (Katata, 1982) , Progr. Math. , Birkh¨auser Boston, Boston, MA, 1983, pp. 567–589. MR 0728619. Zbl 0543.14006.[VZ03] E. Viehweg and K. Zuo, On the Brady hyperbolicity of moduli spaces for canonicallypolarized manifolds, Duke Math. J. (2003), 103–150.[VZ06] , Numerical bounds for semi-stable families of curves or of certain higher dimen-sional manifolds,
J. Diff. Geom. (2006), 291–352. Zuo00]
K. Zuo , On the negativity of kernels of Kodaira-Spencer maps on Hodge bundles andapplications,
Asian J. Math. no. 1 (2000), 279–301, Kodaira’s issue. MR 1803724.Zbl 0983.32020. Available at https://doi.org/10.4310/AJM.2000.v4.n1.a17.no. 1 (2000), 279–301, Kodaira’s issue. MR 1803724.Zbl 0983.32020. Available at https://doi.org/10.4310/AJM.2000.v4.n1.a17.