Modular forms and special cubic fourfolds
Abstract.
We study the degrees of special cubic divisors on moduli space of cubic fourfolds with at worst ADE singularities. In this paper, we show that the generating series of the degrees of such divisors is a level three modular form.
1. Introduction
The classical NoetherLefschetz locus for degree hypersurfaces in is the locus in the Hilbert scheme space where the Picard rank is greater than one. For , the NoetherLefschetz loci are known to be a countable union of proper subvarieties of by Griffiths and Harris [8]. A natural question is to find the degrees of these subvarieties. For quartic surfaces in , Maulik and Pandharipande [19] showed that the NoetherLefschetz loci are divisors and the degrees of these divisors are the Fourier coefficients of certain modular forms.
In higher dimensional cases, cubic fourfolds have received a lot of attention since their period map behaves quite nicely. Specifically, the period domain for cubic fourfolds is a bounded symmetric domain of type IV and the global Torelli theorem holds (cf. [25, 26]).
The analogues of NoetherLefschetz loci for surfaces are the loci of special cubic fourfolds studied by Hassett [9]. A smooth cubic fourfold in is special of discriminant if it contains an algebraic surface , and the discriminant of the saturated lattice spanned by and in is , where . The Zariski closure of the collection of such cubic fourfolds forms an irreducible divisor in the moduli space (cf. [15]) of cubic fourfolds with at worst isolated ADE singularities and it is nonempty if and only if . We interpret as the set of singular cubics in . The special cubic divisor in the Hilbert scheme of cubic hypersurfaces is the lift of (see 2 for more details). In the present paper, we study the degree of special cubic divisors . Our main result is:
Theorem 1.
Let be the generating series for the degrees of the special cubic divisors. Then is a modular form of weight and level with expansion:
where
(1.1) 
are level three modular forms of weight 1 and 3 that generate the space of modular forms with respect to the group (see 3.2). Here, denotes the Legendre symbol.
The approach to Theorem 1 is via the result of Borcherds [2] and KudlaMilson [14]. The degrees of are the Fourier coefficients of a vectorvalued modular form. As in [19], the NoetherLefschetz numbers are related to the reduced GromovWitten (GW) invariants of K3 surfaces. We hope there is a similar GWtheory interpretation of .
Outline of paper
In section 2, we review some classical result on cubic fourfolds and describe the special cubic divisors
from an arithmetic perspective. Section 3 is the central section of this paper. We recap Borcherds’ work on Heegner divisors to prove the modularity of a
vectorvalued generating series of . This vectorvalued modular form can be expressed explicitly in terms of some wellknown modular forms. The proof of our main theorem is presented in the last section.
After posting our paper on the arXiv server, we learned from Atanas Iliev that, in forthcoming work, Atanas Iliev, Emanuel Scheidegger,
and Ludmil Katzarkov in [11] have independently proved Theorem 3 using a different basis of the space of vectorvalued modular forms.
Acknowledgements. The first author was supported by NSF grant 0901645. The authors are grateful to their advisor Brendan Hassett for introducing this problem, and many useful discussions. We have benefited from discussions with Radu Laza. We would also like to thank the referee for many helpful comments.
2. Special Cubic fourfolds and Heegner divisors
In this section, we review some results on special cubic divisors and the relation with Heegner divisors associated to a signature lattice, which is defined from an arithmetic perspective. Throughout the paper, we denote by the dual of a lattice and its associated orthogonal group.
2.1. Period domain
Let be a smooth cubic fourfold in . Denote by the middle cohomology group containing . It is well known (e.g. [9],[26]) that the primitive middle cohomology is isometric to the lattice
(2.1) 
under the intersection form , with an associated period domain that is a connected component of
Here, and are root lattices corresponding to the root systems of the same names, and is the hyperbolic lattice of rank two. The monodromy group (i.e. the identity connected component of ) is defined by
acts on and the arithmetic quotient is a quasiprojective variety parametrizing the periods of cubics.
Hassett [9] has defined irreducible divisors as follows:
Definition.
Let be a ranktwo positive definite saturated sublattice of containing of discriminant . There is an associated hyperplane
in . Then is defined as the quotient by of the union of all such hyperplanes .
On the other hand, as constructed by Laza [15] via geometric invariant theory (GIT), the moduli space of cubic fourfolds with at worst isolated ADE singularities is a Zariski open subset of , where and denotes the GIT stable points of under the action of . Together with Voisin’s Global Torelli theorem (cf. [25, 26]), Laza [16] and also Looijenga [18] have shown that there is an extended period map
(2.2) 
which is an open immersion. The complement of the image of in is corresponding to degenerations of determinant cubic hypersurfaces (cf. [9] 4.4). Moreover, is exactly the pullback of via .
Let be the natural quotient map; then the special cubic divisor is the Zariski closure of the pullback of via .
2.2. Heegner divisors
In general, let be an even lattice of signature with an associated domain as a connected component of
there is an arithmetic group
acting on . For and , the Heegner divisor [4] on is defined by
(2.3) 
where is a hyperplane of . When , we take to be the Cartier divisor coming from on . Similarly to [19], it is clear from definition that is actually the Hodge bundle on , where is the universal family and is the relative sheaf of holomorphic 1forms.
Remark 1.
One can see that is the arithmetic quotient of a Hermitian symmetric subdomain of . They are called special cycle on in Kudla’s program [13].
Taking , we have and . Note that , one can choose representatives with for
Lemma 1.
The Heegner divisors on , where
Proof.
The redundancy is because of the symmetry . Let be a rank 2 negative sublattice containing and of discriminant . Assume that is generated by and . Then there is a bijection between the two sets of hyperplanes as follows:
since one can verify that
∎
As an application, we show the following result:
Proposition 1.
The Picard group has rank two and is spanned by Heegner divisors and .
Proof.
Let be the subgroup of generated by the Heegner divisors. By applying the general formula from Bruinier [4] 5.2 to the lattice , we get
On the other hand, the moduli space of cubic fourfolds with at worst isolated ADE singularities is an open subset of via the extended period map, and the complement of is the irreducible Heegner divisor . If the Picard number of is at most one, then is at most two and has to be the same as by dimension considerations.
We prove that the dimension of is at most one. Observe that is constructed via the GIT quotient , where is the open subset of the Hilbert scheme parameterizing all cubic hypersurfaces with at worst ADE singularities. Then has rank one since the boundary of in has codimension at least two.
Let be the set of linearized line bundles on . There is an injection
by [12, Proposition 4.2.]. Our assertion follows from the fact the forgetful map is an injection.
∎
Remark 2.
The moduli space of quasipolarized K3 surface is a 19dimensional locally Hermitian symmetric variety associated to . It is conjectured by Maulik and Pandharipande that the Picard group of is rationally spanned by NoetherLefschetz divisors. This conjecture has been verified for low degree surfaces (cf. [23], [24], [17]). We also refer the readers to [10] and [1] for some recent results in this subject.
2.3. The degree of special cubic divisors
Analogous to NoetherLefschetz numbers of K3 surfaces [19], the degree of can be computed via intersection with a test curve. Let be a Lefschetz pencil of cubic hypersurfaces in . It yields a natural morphism
which factors through the rational map . It is not difficult to see that is the same as the intersection number .
Let be the composition of and . If we set
(2.4) 
then we have for , and since there are no determinantal cubic fourfolds in a Lefschetz pencil . The generating series can be rewritten as
(2.5) 
Some examples
One can see that there is a natural enumerate geometry interpretation of , and some of them can be computed using geometric methods when is small.

The degree of counts the number of singular fibers in . The first jet bundle [21] of parametrizes all nodal cubic hypersurfaces in . Then equals to the top Chern class of , which is .

The degree of counts the number of planes contained in the fibers of . Let be the Grassmannian parametrizing all planes in . The planes contained in a cubic hypersurface of are parametrized by certain vector bundle on . Via standard Schubert calculus, one can show equals , which is the top chern class of .

The degree of counts the number of Pfaffian cubic fourfolds (cf. [9]) in , which equals to according to our main theorem.
3. Modular forms associated to signature lattices
In this section, we introduce the vectorvalued modular form associated to an even lattice and prove the modularity of the generating series of special cubic divisors from Borcherds’ work [2].
3.1. Vector valued modular forms
The metaplectic double cover of consists of pairs , where
It is wellknown that is generated by
Let be the complex upper halfplane. Suppose is a representation of on a finite dimensional complex vector space , such that factors through a finite quotient. For any , a vectorvalued modular form of weight and type on is a holomorphic function on , such that
When , this recovers the definition of scalarvalued modular forms with a character.
Given a lattice M of signature with a bilinear form , there is a Weil representation of on the group ring defined by the action of the generators as follows:
where is the standard basis of for . We denote by the space of modular forms of weight and type .
Now we take and denote by the standard basis of corresponding to element as in 2.2. Our first result is:
Theorem 2.
Proof.
In general, as shown in [2, Theorem 4.5] and [20, Theorem 5.6], the generating series for Heegner divisors associated to a lattice of signature
(3.2) 
is an element in .
In our situation , we have by Lemma 1 and thus the generating series
(3.3) 
is a vectorvalued modular form of weight and type with coefficients in .
Next, let be a linear function defined by
Then as shown in 2.2, we have and
which is the degree of the Hodge bundle . It follows that is an element in . ∎
Remark 3.
In Borcherds’ setting, the lattice has signature and the generating series of Heegner divisors are vectorvalued modular forms of type (dual of ). For with signature , one can get (3.2) by transferring the lattice to , which has signature and .
3.2. Construction of modular forms
Here we introduce some modular forms which will be used later.
3.2.1. Scalarvalued Eisenstein series
The classical Eisenstein series
(3.4) 
is a modular form of weight for for , where is the Bernoulli number.
Let us denote by (resp. ) the arithmetic subgroups in defined by
and let be the nontrivial Dirichlet character modulo 3 on , i.e.
(3.5) 
where is the nontrivial Dirichlet character modulo 3.
Proposition 2.
Assume that is an odd integer. Then the Eisenstein series
(3.6) 
is a modular form of weight with character for congruence group . Moreover, and .
Respectively, is a modular form of weight with character for congruence group .
3.2.2. Vectorvalued Eisenstein series
Let and an even lattice. The vectorvalued Eisenstein series on constructed via Petersson slash operator [4] are vectorvalued modular forms of weight and type . The equivalence of Weil representations and implies
(3.7) 
Let be the vectorvalued Eisenstein series associated to of weight . For , it is given by [6] that
Here, denotes the Dirichlet Lseries with character and is the local Euler product defined as following:
When ,
where is the Bernoulli polynomial. Thus one obtains that
3.2.3. RankinCohen bracket
Given any two level scalarvalued modular forms on the upper half plane of weight and . The nth RankinCohen bracket is defined as follows:
where denotes the rth differential of with respect to .
For a vectorvalued modular form
one can extend the RankinCohen bracket to and as follows,
(3.8) 
According to [19, Lemma 5], we have the following result:
Lemma 2.
The vectorvalued functions
(3.9) 
are vectorvalued modular forms of weight and type .
3.3. Expression of the generating series
Now we are ready to give an explicit expression of . From the dimension formula of Bruinier in [5], we know that
(3.10) 
It follows that
Theorem 3.
Let be the vectorvalued modular forms constructed in Lemma 2. Then is a basis of and
(3.11)  
Proof.
By checking the coefficients of the term , we know that and are linearly independent. Thus they form a basis of by dimension considerations. To obtain the expression (3.11), it suffices to use the following two constraints:

The degree of the Hodge bundle is , which gives the coefficient of . By GrothendieckRiemannRoch, we have the following Chern character computation:
where is the relative tangent bundle. Since is trivial by the Lefschetz hyperplane theorem, we get .

The coefficient of is as shown in 2.3.
∎
4. Proof of Theorem 1
To prove our main theorem, we first start with a Lemma on the modularity on the components of a vectorvalued modular form:
Lemma 3.
Let be an element in . Then the following are true:

is a scalarvalued modular form for of weight with character .

is a scalarvalued modular form for of weight with character , where

The sum is a scalarvalued modular form for of weight with character .
Proof.
Statement (i) and (ii) follows from [22] 2. For (iii), since the generators of the congruence subgroup are
(4.1) 
then statement (iii) follow from a direct computation by checking the modularity on these generators. ∎
Proof of Theorem 1. Let us write . Then it is not difficult to see that
(4.2) 
Next, let (resp. ) denote the space of scalarvalued modular forms with character for (resp. ). It is known by [3] 12 that is a polynomial ring generated by two modular forms of weight and weight .
By Proposition 2, we thus get that is generated by of weight and of weight . Since has weight by Lemma 3, can be expressed as a linear combination of
The coefficients computation shows that
(4.3) 
Similarly, is a polynomial ring generated by and . Then the modular form has weight and can be expressed as
(4.4)  
References
 [1] Nicolas Bergeron, John Millson, and Colette Moeglin. Hodge type theorems for arithmetic manifolds associated to orthogonal groups. arXiv:1110.3049, 2011.
 [2] Richard E. Borcherds. The GrossKohnenZagier theorem in higher dimensions. Duke Math. J., 97(2):219–233, 1999.
 [3] Richard E. Borcherds. Reflection groups of Lorentzian lattices. Duke Math. J., 104(2):319–366, 2000.
 [4] Jan Hendrik Bruinier. Borcherds products on O(2, ) and Chern classes of Heegner divisors, volume 1780 of Lecture Notes in Mathematics. SpringerVerlag, Berlin, 2002.
 [5] Jan Hendrik Bruinier. On the rank of Picard groups of modular varieties attached to orthogonal groups. Compositio Math., 133(1):49–63, 2002.
 [6] Jan Hendrik Bruinier and Michael Kuss. Eisenstein series attached to lattices and modular forms on orthogonal groups. Manuscripta Math., 106(4):443–459, 2001.
 [7] Fred Diamond and Jerry Shurman. A first course in modular forms, volume 228 of Graduate Texts in Mathematics. SpringerVerlag, New York, 2005.
 [8] Phillip Griffiths and Joe Harris. On the NoetherLefschetz theorem and some remarks on codimensiontwo cycles. Math. Ann., 271(1):31–51, 1985.
 [9] Brendan Hassett. Special cubic fourfolds. Compositio Math., 120(1):1–23, 2000.
 [10] William Hoffman and He Hongyu. Picard groups of Siegel modular threefolds, and theta lifting. arXiv:0712.3718, 2007.
 [11] Atanas Iliev, Emanuel Scheidegger, and Ludmil Katzarkov. Pencils of cubic fourfolds. in preparation.
 [12] Friedrich Knop, Hanspeter Kraft, and Thierry Vust. The Picard group of a variety. In Algebraische Transformationsgruppen und Invariantentheorie, volume 13 of DMV Sem., pages 77–87. Birkhäuser, Basel, 1989.
 [13] Stephen S. Kudla. Algebraic cycles on Shimura varieties of orthogonal type. Duke Math. J., 86(1):39–78, 1997.
 [14] Stephen S. Kudla and John J. Millson. Intersection numbers of cycles on locally symmetric spaces and Fourier coefficients of holomorphic modular forms in several complex variables. Inst. Hautes Études Sci. Publ. Math., (71):121–172, 1990.
 [15] Radu Laza. The moduli space of cubic fourfolds. J. Algebraic Geom., 18(3):511–545, 2009.
 [16] Radu Laza. The moduli space of cubic fourfolds via the period map. Ann. of Math. (2), 172(1):673–711, 2010.
 [17] Zhiyuan Li and Zhiyu Tian. Picard groups on moduli space of complete intersections. arXiv:1304.3219, 2013.
 [18] Eduard Looijenga. The period map for cubic fourfolds. Invent. Math., 177(1):213–233, 2009.
 [19] Davesh Maulik and Rahul Pandharipande. GromovWitten theory and NoetherLefschetz theory. Proceedings of Conference: A Celebration of Algebraic Geometry, 2013.
 [20] William J. McGraw. The rationality of vector valued modular forms associated with the Weil representation. Math. Ann., 326(1):105–122, 2003.
 [21] David John Saunders. The geometry of jet bundles, volume 142 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 1989.
 [22] Nils R. Scheithauer. Some construction of modular forms for the Weil representation of . in preprint, 2012.
 [23] Jayant Shah. A complete moduli space for surfaces of degree . Ann. of Math. (2), 112(3):485–510, 1980.
 [24] Jayant Shah. Degenerations of surfaces of degree . Trans. Amer. Math. Soc., 263(2):271–308, 1981.
 [25] Claire Voisin. Théorème de Torelli pour les cubiques de . Invent. Math., 86(3):577–601, 1986.
 [26] Claire Voisin. Erratum: “A Torelli theorem for cubics in ” (French) [Invent. Math. 86 (1986), no. 3, 577–601; no 0860684]. Invent. Math., 172(2):455–458, 2008.