Hodge theory and boundaries of moduli spaces
Phillip Griffiths∗
Based on correspondence and discussions with Mark Green, Colleen Robles and Radu Laza
∗Institute for Advanced Study
1/82 I The general theme of this talk is the use of Hodge theory to study questions in algebraic geometry.
I This is a vast and classical subject; here we shall focus on one particular aspect.
I In algebraic geometry a subject of central interest is the study of moduli spaces M and their compactifications M.
I Points in ∂M = M\M correspond to singular varieties X0 that arise from degenerations X → X0 of a smooth variety X .
2/82 I It is well known and classical that X0 may be “simpler” than X and understanding X0 may shed light on X . For an example, any smooth curve X may be degenerated to a stick curve
1 X0 = = configuration of P ’s
and deep questions about the geometry of a general X may frequently be reduced to essentially combinatorial ones on X0.
3/82 I The central thesis of this talk is Hodge theory may be used to guide the study of degenerations X → X0. Underlying this thesis are the points ∗ I the polarized Hodge structure on H (X , Z) (mod torsion) is the basic invariant of a smooth projective variety X ; I the study of degenerations (V , Q, F ) → (V , Q, W , Fo) of polarized Hodge structures is fairly highly developed;
I this understanding may be used to suggest what the possible degenerations X → X0 should be and where to look for them.
4/82 This thesis is well known and has been effectively used in the classical case when the Hodge structures are parametrized by a Hermitian symmetric domain and 1-parameter degenerations are used. It has been utilized in the non-classical case mainly for 1-parameter degenerations of Calabi-Yau threefolds, but to my knowledge not so much in other non-classical cases or for several parameter degenerations other than for the case of algebraic curves.
5/82 I the structure of this talk will be
I discussion of some of the basic aspects of Hodge theory;
I the example of algebraic curves, which indicates how the study might go in the classical case;
I discussion of the first non-classical case, namely that of an algebraic surface X of general type with h2,0(X ) = 2, 1,0 2 h (X ) = 0 and KX = 2; this is based on correspondence with Mark Green, and also with Colleen Robles and Radu Laza, and reflects developments in a very early stage.
6/82 I A final general point: a schematic of how Hodge theory interacts with algebraic geometry for the above surface is canonical 0 2 2,0 ←→ H (ΩX ) = H (X ) series |KX |
bi-canonical variation of Hodge structure ←→ 1 series |2KX | associated to H (C), C ∈ |KX | . . which arises from
KX = KC C. .
7/82 Definition A polarized Hodge structure (PHS) of weight n is given by (V , Q, F •) where
I V is a Q-vector space (in practice we will have V ⊃ VZ); I Q · V ⊗ V → Q is a non-degenerate form with Q(u, v) = (−1)nQ(v, u); • n n−1 0 I F is a decreasing filtration F ⊂ F ⊂ · · · ⊂ F = VC satisfying
p n−p+1 ∼ F ⊕ F −→ VC, 0 5 p 5 n;
p n−p+1 I Q(F , F ) = 0;
I Q(v, Cv¯) > 0 for v 6= 0, where C is the Weil operator.
8/82 q q,p Setting V p,q = F p ∩ F = V , the third condition above is
p,q p,q q,p VC = ⊕V where V = V and Cv = i p−q on V p,q. Example Let X ⊂ PN be a smooth projective variety of dimension d and L ∈ H2(X , Q) the restriction of the generator of a hyperplane class in H2(PN , Z). Then multiplication by L gives a nilpotent operator on H∗(X , Q)
m m+2 L : H (X , Q) → H (X , Q).
9/82 The Hard Lefschetz theorem k m−k ∼ m+k L : H (X , Q) −→ H (X , Q) implies that the action of L may be uniquely extended to the action of an sl2-triple {L, Y , Λ} where m Y = (m − d)Id on H (X , Q) ∗ is the grading. This sl2-action decomposes H (X , Q) into irreducibles whose basic building blocks are the lowest weight spaces
k+1 m−k n m−k L m+k+2 o H (X , Q)prim = ker H (X , Q) −−→ H (X , Q) = V of the irreducible summands.
10/82 Setting Q(u, v) = (Lk u · v)[X ] and using the decomposition n−k H (X , C)prim induced by the Hodge decomposition Hm−k (X , C) = ⊕ Hp,q(X ) gives a PHS of weight p+q=m−k m − k. Instead of the usual Hodge diamond, we will picture the groups in the first quadrant in the (p, q) plane. q
dim X = 1 L r r p ↑r 0 1 H (ΩX ) = holomorphic 1-forms on the compact Rieman surface X .
11/82 dim X = 2 and X is regular (H1,0(X ) = 0)
0 LH (X ) H rr HHj 1,1 H (X )prim r r
0 r 2 H (ΩX )
12/82 dim X = 1 The PHS on H1(X ) for X a compact Riemann surface is classically represented by the matrix of periods
γ 0 1 ω , , ω ∈ H ΩX ˆδ,γ ··· δ Even more classically, if X is given as the Riemann surface associated to an algebraic curve, e.g., the one of genus g given in C2 by
2g+2 2 Y f (x, y) = y − (x − ai ) = 0, ai distinct i=1 g(x) dx 2g(x) dx ω = = , 0 5 deg g(x) 5 g − 1 y fy (x, y)
13/82 the periods are δ ... a1 a2 a3 a4 γ
The closure in P2 of the above affine curve has a singular point at [0, 0, 1] and the condition deg g(x) 5 g − 1 is the one that gives a holomorphic 1-form on X .
14/82 dim X = 2 The surfaces we shall be considering are minimal, of general type and have
2,0 1,0 2 h (X ) = 2, h (X ) = 0, KX = 2.
They are extremal; i.e., they have the maximum h2,0(X ) subject to the other conditions above. They were analyzed classically by Castelnuovo and more recently by Horikawa. A general one is the minimal desingularization of the quartic 4 surface on P given in affine coordinates x1, x2, x3, x4 by ( f (x) = h(x) − k(x)2 = 0 2 g(x) = x1 − x2 = 0 where h(x), k(x) are a general cubic, quadric.
15/82 This surface has as singularity a double curve with eight pinch points. Think of it as an analogue of the smooth curve with affine equation 2g+2 2 Y y − (x − ai ) = 0 i=1 encountered above. It is the first regular, non-classical extremal surface. We will say more about it later.
16/82 The holomorphic 2-forms are
`(x)dx ∧ dx ω = 3 4 , deg `(x) 1 f (x)g (x) 5 x1 x2 and where `(x) vanishes on the double conic so that ω is holomorphic on X . With these conditions we find that h2,0(X ) = 2. Think of ω as the analogue of g(x) dx where fy (x) deg g(x) 5 g − 1 above. One has b2(X ) = 32, so that 1,1 h (X )prim = 31 and the PHS is given by the matrix of periods.
17/82 The surface is more conveniently given in the homogenous coordinate [t0, t1, x3, x4] in the weighted projective space P(1, 1, 2, 2) by
2 2 2 2 2 2 t0 G(t0 , t0t1, t1 , x3, x4) = F (t0 , t0t1, t1 , x3, x4) .
For fixed [t0, t1] we obtain genus 3 plane quartic curves which constitute the pencil |KX |.
18/82 Definition D = period domain given by {(V , Q, F •), dim V p,q = hp,q} Dˇ = compact dual given by {(V , Q, F •), Q(F p, F n−p+1) = 0}. If G = Aut(V , Q), then
D = GR/H, H compact ∩ ˇ D = GC/P, P parabolic and H = P ∩ GR.
19/82 Example
∼ t D = Hg = {Z = g × g matrix, Z = Z and Im Z > 0} = Sp(2g, R)/U(g) ˇ 2g 2g D = GrassL(g, C ) = Lagrangian g-planes in (C , Q) 0 I where Q = −I 0 .
The embedding D ,→ Dˇ is
Z Z → span of the columns of . Ig
1 For g = 1 we have H ,→ P and GR = SL2(R) acting by linear fractional transformations.
20/82 Example
ˇ D = GrassQ (2, VC) = {2-planes E : Q(E, E) = 0} ∪ D = 2-planes E with Q(E, E) < 0.
The GR = SO(4, h) and D = GR/U(2) × SO(h) is not an Hermitian symmetric domain (HSD). Definition A variation of Hodge structure (VHS), or period mapping, is given by a holomorphically varying family of PHS’s parametrized by a complex manifold S and satisfying a differential constraint. In order to have the PHS’s occurring in a fixed vector space (V , Q) we need to use the universal covering Se → S.
21/82 Example S = ∆∗ = {t : 0 < |t| < 1}, Se = H = {z : Im z > 0}, and t = e2πiz . We may take products of this to have S = ∆∗`. Definition A VHS is given by Φ Se e / D
Φ S / Γ\D where we have the monodromy representation ρ : π1(S) → GZ and where Φe is an equivariant holomorphic mapping satisfying
(i) Φ(e γz) = ρ(γ)Φ(e z) for γ ∈ π1(S), z ∈ Se; p dFz p−1 (ii) dz ⊂ Fz (infinitesimal period relation = IPR).
22/82 Example 1 ∗ 2 Q S = ∆ and Xt = {y − x(x − t) (x − ai ) = 0}. The family X → ∆∗ is a differentiable fibre bundle, but we may only 1 2g identify the H (Xt , Z) with Z up to the action of monodromy when t turns around the origin
γ γ ··· −→ ··· δ γ→γ+δ
→·· −→·· ··· 0 t δ γ γ
23/82 At t = 0 we have the singular curve γ δ ··· ···
2 2 Q y = x (x − ai )
The IPR is trivial in this case, as it is whenever the period domain D is a Hermitian symmetric domain. γ γ δ ··· ··· δ
24/82 The pictures of the singularity in Example 1 are
←→ = neighborhood of the node
6 6 real picture topological picture
25/82 Example 2 This is similar to Example 1, but it is harder to draw the pictures. We will be interested in the degeneration of the surface above, e.g., given by
2 f (x, t) = (x1 − t)q(x) − k(x) = 0 where q(x) is a general quadric. It is much harder to draw the pictures.
26/82 The real picture of the singularity is
p
This is a double curve with a pinch point.
27/82 We remark that if we have a family of smooth projective varieties ∗ π ∗ −1 X −→ ∆ , Xt = π (t) the limit lim Xt t→0 is not well defined — it depends on how we complete ∗ ∗ X → ∆ to X → ∆ to have X0. Moreover, even if we have conditions on X that uniquely define X0, the geometry of X0 may be complicated. This situation is even more so when we have a higher dimensional parameter space, e.g., S = ∆∗`. This brings us to the main point of the talk.
28/82 ∗` I Degenerating Hodge structures given by Φ : ∆ → Γ\D are well understood when ` = 1 and a lot is known when ` > 1.
I One may hope that this knowledge can be used to help guide the geometric understanding of X → X0. We will try to (i) explain how this principle works for algebraic curves (classical), (ii) explain the issues when X is a surface of general type (like g = 2 for algebraic curves), and (iii) discuss part of the one surface example where there are very early indications of how the above philosophy might work out quite nicely. Implementing this philosophy in the end of course requires algebro-geometric arguments.
29/82 The analysis of degenerating PHS’s uses Lie theory and the geometry of homogeneous complex manifolds. The starting point is the Monodromy Theorem ∗ For Φ : ∆ → Γ\D the generator T ∈ Aut(VZ, Q) given by ∗ the image of the identity in π1(∆ ) is unipotent — i.e., the eigenvalues of T are roots of unity. Replacing t by tk we may assume that T = exp N where 1 2 N = log T = (T − I ) − 2 (T − I ) + · · · ∈ Aut(VQ, Q) is nilpotent. Moreover, Nn+1 = 0 where n = weight of the PHS’s.
30/82 Definition ∗ Φ k A nilpotent orbit is a VHS ∆r −→{T }\D given by
Φ(e z) = exp(zN) · F0 ˇ where F0 ∈ D and exp(zN) · F0 ∈ D for Im z > log(r). Rescaling t = λt we may assume r = 1 and F0 ∈ D. The IPR is equivalent to p p−1 NF0 ⊂ F0 .
Theorem (Schmid) Any VHS over ∆∗ may be well approximated by a nilpotent orbit.
31/82 Classically it was known (theorem on regular singular points) that any period matrix may be expanded as a polynomial in log t with holomorphic coefficients — the nilpotent orbit is (with a suitable normalization) obtained by evaluating the coefficients at t = 0. The “suitable normalization” is subtle and is given a structure by the further approximating sl2-orbit. Example For y 2 = x(x − t)(x − a)(x − b), complex analysis may be used to show that t dx log t dx = + h(t), = k(t), k(0) 6= 0. ˆ0 y 2πi ˆγ y
32/82 The nilpotent orbit is obtained by using h(0), k(0). Rescaling t and dx/y, we may take h(0) = 0, k(0) = 1. The nilpotent orbit is then " # z 1 (∗) z → ∈ H ⊂ P . 1
0 Note that F0 = 1 ∈ ∂H; this will be a general situation. Definition
An sl2-orbit is given by a representation ρ : SL2(R) → GR with 0 1 ρ∗ ( 0 0 ) = N which induces an equivariant VHS
H ,→ D.
33/82 Theorem (Schmid)
A nilpotent orbit may be approximated by an sl2-orbit.
In this way we may understand limt→0 Φ(t). What is the limiting object? To explain this, we note that any N may be graded by a semi-simple Y with integral eigenvalues and weight spaces Vk where −n 5 k 5 n, and we define the weight filtration
W (N)m = ⊕ Vk+n k5m
W (N)0 ⊂ W (N)1 ⊂ · · · ⊂ W (N)2n = V .
With indices running opposite to the case of the hard Lefschetz theorem we have
N : W (N)k → W (N)k−2.
34/82 Definition A mixed Hodge structure (MHS) is given by (V , W•, F ) where W• is an increasing filtration on V and F is a decreasing
filtration on VC, and where F induces on each W Grk V = Wk /Wk−1 a Hodge structure of weight k. Definition A polarized limiting mixed Hodge structure (we shall just use LMHS) (V , Q, W (N), F ) is given by a MHS (V , W (N), F ) k such that for k = n the bilinear forms Qk (u, v) = Q(N u, v) W (N) induce PHS’s on Gk,primV .
35/82 W (N) Here the primitive spaces are defined on GrV as in the case of a smooth projective variety with ( N ↔ L Y ↔ Y , where as noted above N goes in the opposite direction to L. The above definition depends only on N and not on the Y . The sl2-orbit theorem gives us a choice of Y .
36/82 The punch line of the above is: A VHS
Φ : ∆∗ → Γ\D gives a LMHS
lim Φ(t) = (V , W (N), F0). t→0
The choice of F0 is not unique because rescaling t → λt gives F0 → exp(λN)F0. It may be normalized so that (among other things) F0 ∈ ∂D.
37/82 D
F∞ −1 + exp(iy N )F∞
F0 exp(iyN)F0
F∞ = lim exp(iyN)F0 y→∞ −1 + F0 = lim exp(iy N )F∞ y→0
+ 0 0 where N = ( 1 0 ). The point here is that a lot is known about degeneration of PHS’s.
38/82 Example When the weight n = 1 we have N2 = 0 and the weight filtration is
W0(N) ⊂ W1(N) ⊂ W2(N) = V .
W0(N) = Im N = isotropic subspace (Q = 0 on it) ⊥ W1(N) = Ker N = W0(N) .
39/82 Elementary linear algebra shows that we may choose a ∼ W (N) symplectic basis such that V = ⊕ Grk V and k 0 0 I ! 0 I g0 Q = 0 Q0 0 , Q0 = −Ig 0 −I 0 0 0 0 0 A t N = 0 0 0 , A = A, A > 0. 0 0 0
40/82 Recalling that the period domain D = Hg = {g × g matrices Z, Z = t Z and Im Z > 0}, the nilpotent orbit is