Mixed Hodge structures and log Hodge structures on log deformations Taro Fujisawa Tokyo Denki University, 2-2 Knada Nishiki-cho, Chiyoda-ku e-mail: fujisawa@mail.dendi.ac.jp Introduction (0.1) The notion of log Hodge structure, which is a generalization of variation of Hodge structures, is introduced by Kazuya Kato, and studied by Kato himself, T. Kajiwara, T. Matsubara, C. Nakayama and S. Usui in [KMN02] [KU09] etc. In the geometric setting it is natural to ask the follwoing question analogus to the classical case. (0.2) Question. Let f : X −→ S be a proper log smooth morphism. Does · the higher direct image sheaf Rq f∗ ωX/S underlies a log Hodge structure of weight q under some Kähler conditions? (0.3) [KMN02] gives an affirmative answer to this question, under the assumption that f is projective and that S is log smooth. In this case, the open subset Striv , the locus on which the log structure is trivial, plays an important role. Then, how about for the case where S is the standard log point? In this case Striv is empty, on the contrary. The talk which I presented in KUSATSU 2008 symposium is a trial to answer the question above for the case of log deformations. By definition, the base space S of a log deformation is the standard log point. 1 (0.4) As far as I understand, polarized log Hodge structure is a globalized notion of nilpotent orbit. In particular, the data of a log Hodge structure over the standard log point is almost the same as one dimensional nilpotent orbit (with no polarization). A nilpotent orbit yields a “limiting” mixed Hodge structure as proved by Schmid in [Sch73]. On the other hand, [FN03] showed that a proper log deformation with suitable conditions yields a mixed Hodge structure. In this sense, the problem which I will treat here concerns the reconstruction of a nilpotent orbit from a mixed Hodge structure. proper log deformation - mixed Hodge structure 6 ? nilpotent orbit + log Hodge structure on the standard log point (0.5) First, I misunderstood that the problem above is not so difficult by using Theorem (2.9) below. However, I found that I was wrong while I prepared the manuscript for the talk at Kusatsu. Even now, I have not succeeded to prove Conjecture (2.7) in Section 2. Here I will explain my present situation as I did in the talk. 1 Review on log Hodge structures (1.1) In this section, I will introduce the notion of log Hodge structure on a log deformation. First, I briefly recall log geometry, in particular, the standard log point. Then I present the notion of log Hodge structure on a log deformation. However, I will not give the precise definition here, but give the equivalent conditions for the case on the standard log point. (1.2) Definition (K. Kato [Kat88]). Let X be a complex analytic space. A pre-log structure on X is a pair (MX , αX ) of a sheaf of monoids MX on X and a morphism of monoid shaves αX : MX −→ OX , where OX is regarded as a monoid sheaf by its multiplication. A pre-log structure (MX , αX ) is said to −1 ∗ be a log structure if the morphism αX induces an isomorphism αX (OX ) −→ 2 ∗ OX . A log structure (MX , αX ) is simply denoted by MX if there is no danger of confusion. (1.3) Example. Let X be a complex manifold, D an effective divisor on X and j : X \ D −→ X an open immersion. We set ∗ ∩ OX , MX (D) = j∗ OX\D which is a monoid subsheaf of OX . Equipped with the inclusion MX (D) −→ OX , MX (D) turns out to be a log structure on X. This is a typical and most important example of log structures. (1.4) For a log complex analytic space X, the sheaf of log differential p-forms p p p+1 ωX and the differential d : ωX −→ ωX are defined in [Kat88]. It satisfies 2 the Leibniz rule and the equality d = 0 as usual. Thus we obtain the log de · Rham complex ωX . We can treat the ralative case by the similar way. See [Kat88] for the detail. (1.5) The ringed space X log . For an fs log analytic space X (‘fs’ stands for the properties ‘fine and satutrated’, which I do not mention in this article), a log topological space X log , a sheaf of rings OX on X log and a morphism of ringed log spaces τX : X −→ X are defined in [KN99]. We call this ringed space X log the Kato-Nakayama space associated to X. Here I do not give its definition, but I will describe the Kato-Nakayama space associated to the standard log point below. I remark that the construction of the Kato-Nakayama space is p,log log p ·,log functorial. We set ωX = OX ⊗τ −1 OX τX−1 ωX and obtain the complex ωX X on X log . (1.6) The standard log point ∗. A point is regraded as a complex manifold of dimension 0. The ring of “holomorphic functions” on a point is just C. We consider a monoid C∗ ⊕ N and a monoid homomorphism C∗ ⊕ N −→ C which sends (a, n) to 0 if n ̸= 0 and to a if n = 0. Then it is easy to see that this gives a log structure on a point. A point equipped with this log structure is called the standard log point and denoted by ∗ in this article. The C-vector space of the log differential 1-forms on ∗ is denoted by ω∗1 . We can see that ω∗1 is one dimensional C-vector space whose base is denoted by dlog t. Thus we have ω∗1 = C dlog t. The log de Rham complex ω∗· is equal to the complex 0 C ⊕ C dlog t[−1] = (0 −→ C −→ C dlog t −→ 0) 3 by definition. The notion of integrable log connection on ∗ is equivalent to the data of pairs (V, N ), where V is a finite dimensional C-vector space and N is a C-linear endomorphism of V . The endomorphism N is nilpotent if and only if the corresponding log connection is nilpotent. The associated Kato-Nakayama space ∗log is nothing but S1 = {z ∈ C : |z| = 1} with the trivial map τ . The universal covering space of ∗log = S1 is R equipped with the exponential map R ∋ x 7→ e2π √ −1x ∈ S1 , which is denoted by π : R −→ S1 . For a locally constant C-sheaf F on S1 , the C-vecotr space Γ(R, π −1 F) carries a natural automorphism corresponding to the action of the fundamental group Z of S1 . The correspondence above gives an equivalence between the category of locally constant C-sheaves on ∗log = S1 and the one of C-vector spaces equipped with automorphisms. For O∗log , which is locally constant C-sheaf on ∗log = S1 , Γ(R, π −1 O∗log ) is the polynomial ring of one variable C[u] equipped with the automorphism f (u) 7→ f (u − 1). For the complex ω∗·,log , we have π −1 ω∗·,log = (0 −→ C[u] −→ C[u] dlog t −→ 0) d with the differential ∂f dlog t 2π −1 ∂u for any f ∈ C[u]. For a nilpotent integrable log connenction (V, N ) the complex π −1 ω∗·,log ⊗C V is equal to df = 1 √ ∇log (0 −→ C[u] ⊗ V −→ C[u] dlog t ⊗ V −→ 0) where ∇log (f ⊗ v) = 1 √ ∂f dlog t ⊗ v + f dlog t ⊗ N (v). 2π −1 ∂u For the locally constant C-sheaf Ker(O∗log ⊗ V −→ ω∗1,log ⊗ V ), the corre⊗ V −→ ω∗1,log ⊗ V ))) is equal to sponding C-vector space Γ(R, π −1 (Ker(O∗log √ V equipped with the automorphism exp(2π −1N ). (1.7) By using the description above, the definition of log Hodge structures on the standard log point ∗ can be easily translated into the following form. For the precise definition of log (mixed) Hodge structure on an fs log complex analytic space, see [KU00], [KMN02], [KU09]. 4 (1.8) Definition. A Q-log Hodge structure of weight n on the standard log point ∗ is a triple (VQ , F, N ) where VQ is a finite dimensional Q-vector space, where F is a finite decreasing filtration on VC = VQ ⊗Q C and where N is a nilpotent endomorphism of VC satisfying the following conditions: √ (1.8.1) 2π −1N is defined over Q; (1.8.2) N (F p ) ⊂ F p−1 for every p; √ (1.8.3) The filtration e2π −1zN F defines a Hodge structure of weight n if Im z is sufficiently large. (1.9) According to the definition above, (V √Q , F, N ) is a Q-log Hodge structure of weight n on ∗ if and only if (VR , F, 2π −1N ) is a nilpotent orbit (with no polarization), where VR = VQ ⊗Q R. 2 Results (2.1) First, I will give the definition of log deformation. Then I will state the main topic of this article. (2.2) Definition. Let X be a complex manifold and ∆ be the unit disc in C. A morphism f : X −→ ∆ of complex manifolds is said to be a semi-stable degeneration, if the fiber Y = f −1 (0) is a reduced simple normal crossing divisor on X and if f is smooth over the punctrued disc ∆∗ . (2.3) Once a semi-stable degeneration f : X −→ ∆ is given, the simple normal crossing divisor f −1 (0) give rise to a log structure MX on X as in Example (1.3). Moreover the divisor {0} in ∆ gives a log structure M∆ on ∆. Note that the origin equipped with the ‘pull-back’ of the log structure M∆ is nothing but the standard log point. On the other hand, the complex analytic space Y = f −1 (0) is equipped with a log structure MY which is the pull-back log structure of MX . Then the semi-stable degeneration f induces a morphism of log complex analytic spaces Y −→ ∗, which we call the log central fiber of f . (2.4) Definition. Let Y be a reduced complex analytic space equipped with a log structure. A morphism of log complex analytic space f : Y −→ ∗ is said to be a log deformation (or simply Y is a log deformaiton) if locally on Y , 5 the morphism Y −→ ∗ is isomorphic to the log central fiber of a semi-stable degeneration. A log deformation f : Y −→ ∗ is said to be proper, if Y is compact. (2.5) Let f : Y −→ ∗ be a proper log deformation. Then we obtain a commutative diagram π τ Y Y Y∞ −−− → Y log −−− → log f∞ y yf Y f y R −−−→ S1 −−−→ ∗ τ τ as in [FN03]. Then a Q-vector space Γ(R, Rq(f∞ )∗ QY∞ ) = Hq (Y∞ , Q) is obtained. My aim is to show that this Q-vector space underlies a log Hodge structure of weihgt q under suitable assumption. More precisely, I will consider the following. (2.6) In the situation above, the relative log de Rham complex of Y over ∗ is denoted by ωY· /∗ . The cohomology of relative log de Rham complex Hq (Y, ωY· /∗ ) = Rq f∗ ωY· /∗ admits a nilpotent integrable log connection ∇ : Rq f∗ ωY· /∗ −→ Rq f∗ ωY· /∗ ⊗ ω∗1 which corresponds to the nilpotent endomorphism N : Hq (Y, ωY· /∗ ) −→ Hq (Y, ωY· /∗ ) as described in Section 1. The stupid filtration F on ωY· /∗ induces a finite filtration F on the relative log de Rham cohomology Hq (Y, ωY· /∗ ). On the other hand, there exists a natural isomorphism C[u] ⊗Q Hq (Y∞ , Q) ≅ C[u] ⊗C Hq (Y, ωY· /∗ ) by the log Riemann-Hilbert correcepondence proved by K. Kato and C. Nakayama in [KN99]. Substituting 0 for the variable u, we obtain an isomorphism C ⊗Q Hq (Y∞ , Q) ≅ Hq (Y, ωY· /∗ ). Then what I want to prove is summarized as follows. (2.7) Conjecture. Let f : Y −→ ∗ be a proper S log deformation. The irreducible decomposition of Y is denoted by Y = i∈I Yi . We assume that 6 all the irreducible components Yi are nonsingular. Moreover Y is assumed to be cohomologically Kähler, that is, there exits an element ℓ ∈ H2 (Y, R) such that the restriction ℓ|Yi is represented by a Kähler form for every i. Then the data (Hq (Y∞ , Q), (Hq (Y, ωY· /∗ ), N, F )) is a Q-log Hodge structure of weight q for every integer q. (2.8) At the time when I write this article, this conjecture is still open. In the remainder of this article, I explain my strategy toward a proof of it. The key is the following result. (2.9) Theorem ([CKS86],[KK87]). Let (V, W, F ) be an R-mixed Hodge structure, N a nilpotent endomorphism of V and S : V × V −→ R a bilinear form on V . We denote the C-linear (resp. C-bilinear) extension of N (resp. S) on VC = V ⊗R C by the same letter N (resp. S) for simplicity. Then (V, F, N ) is a nilpotent orbit of weight q if the following conditions are satisfied: (2.9.1) N q+1 = 0; (2.9.2) N (Wm ) ⊂ Wm−2 for every m; W (2.9.3) N m induces an isomorphism GrW m+q V −→ Gr−m+q V for every positive integre m (W is the monodromy weight filtration for N centerd at q); (2.9.4) N (F p ) ⊂ F p−1 for every p; (2.9.5) S(x, y) = (−1)q S(y, x) for every x, y ∈ V ; (2.9.6) S(F p , F q−p+1 ) = 0 for every p; (2.9.7) The Hodge structure of weight q + l on the primitive part Pq+l = W l Ker(N l+1 : GrW q+l V −→ Grq−l−2 V ) is polarized by the form S(·, N ·) for every positive integer l. (2.10) I try to apply this theorem to our situation. Form now on, I consider the C-structure only for simplicity. In [FN03], the natural mixed Hodge structure on Hq (Y, ωY· /∗ ) is constructed. Then the conditions (2.9.2) and (2.9.4) can be easily seen by recalling the construction in [FN03]. The condition (2.9.3) is the consequence of Morihiko Saito’s result in [Sai88]. So the main point remained is to construct an appropriate bilinear form on the cohomology group Hq (Y, ωY· /∗ ). 7 (2.11) For a compact Kähler complex manifold X of dimension n, the polarization on H∗ (X, C) is constructed by using the cup product, Lefschetz operator and the trace map H2n (X, C) −→ C, which is an isomorphism. Comparing to this procedure, I consider multiplicative structure or “cup product” on H∗ (Y, ωY· /∗ ) and the trace map. As for the trace map, I have got the following: (2.12) Lemma. Let f : Y −→ ∗ be a proper log deformation such that all the irreducible components of Y is nonsingular and Kähler. We denote the dimension of Y by n. (By definition, Y is pure dimensional.) Then the equality dim H2n (Y, ωY· /∗ ) = 1 holds. Proof. Compute the weight spectral sequence associated to the complex AC in [Ste76]. (2.13) By the lemma above, I can expect to find a natural trace isomorphism H2n (Y, ωY· /∗ ) −→ C. (2.14) A candidate on the multiplicative structure is the morphism ωY· /∗ ⊗ ωY· /∗ −→ ωY· /∗ given by the wedge product. We can easily see that the wedge product above give rise to a bilinear form satisfying the coditions (2.9.5), (2.9.6) for any trace map chosen. Thus, the remaining problem is the following: (2.15) Problem. In the situation in Conjecture (2.7), find an appropriate trace map H2n (Y, ωY· /∗ ) −→ C and prove the condition (2.9.7). (2.16) After the talk, Prof. Usui suggested me a different approach to find a bilinear form, which uses the canonical splitting of a mixed Hodge structure. In this approach, the difficult point is to prove the condition (2.9.6) because the canonical splitting does note preserve the Hodge filtration F . 8 References [CKS86] E. Cattani, A. Kaplan, and W. Schmid, Degeneration of Hodge structures, Ann. of Math. 123 (1986), 457–535. [FN03] T. Fujisawa and C. Nakayama, Mixed Hodge Structures on Log Deformations, Rend. Sem. Mat. Univ. Padova 110 (2003), 221– 268. [KK87] M. Kashiwara and T. Kawai, The Poincaré lemma for variations of polarized Hodge structures, Publ. RIMS 23 (1987), 345–407. [Kat88] K. Kato, Logarithmic structures of Fontaine-Illusie, Algebraic Analysis, Geometry and Number Theory (J.-I. Igusa, ed.), Johns Hopkins Univ., 1988, pp. 191–224. [KMN02] K. Kato, T. Matsubara, and C. Nakayama, Log C ∞ -Functions and Degenerations of Hodge structures, Algebraic Geometry 2000, Azumino, Advanced Studies in Pure Mathematics, vol. 36, Mathematical Society of Japan, 2002, pp. 269–320. [KN99] K. Kato and C. Nakayama, Log Betti cohomology, log étale cohomology, and log de Rham cohomology of log schemes over C, Kodai Math. J. 22 (1999), 191–224. [KU00] K. Kato and S. Usui, Logarithmic Hodge structures and classifying spaces, CRM Proceedings and Lecture Notes 24 (2000), 115–130. [KU09] , Classifying Spaces of Degenerating Polarized Hodge Structures, Annales of Math. Stud., vol. 169, Princeton Univ. Press, 2009. [Sai88] M. Saito, Modules de Hodge Polarisable, Publ. RIMS. 24 (1988), 849–921. [Sch73] W. Schmid, Variation of Hodge structure : the singularities of the period mapping, Invent. Math. 22 (1973), 211–319. [Ste76] J. Steenbrink, Limits of Hodge Structures, Invent. Math. 31 (1976), 229–257. 9