Introduction
Arithmetic quotients and period mappings
It is a central theme in algebraic geometry to understand the relation between GIT quotients and compactifications arising from Hodge theory. In favorable situations, a moduli space admits a projective model arising from invariant theory. On the other hand, it is often realized as an arithmetic quotient of a Hermitian symmetric domain via a period mapping. A basic problem is then to compare these two descriptions not only on the open moduli space, but also at the level of compactifications, discriminant divisors, and natural finite coverings. In this paper, we study modular varieties of type IV attached to certain lattice-polarized K 3 K3 K 3 surfaces of Picard rank 16, which are naturally associated with configurations of six lines in P 2 \mathbb{P}^{2} P 2 and the hypergeometric system of type ( 3 , 6 ) (3,6) ( 3 , 6 ) .
The moduli space of cubic surfaces provides a classical example of this relation. Besides its construction by invariant theory [10 , 31 ] , Allcock, Carlson, and Toledo showed that the moduli space of stable cubic surfaces admits a complex hyperbolic uniformization by a ball quotient of dimension four [2 ] . This description was subsequently related to periods of K 3 K3 K 3 surfaces by Dolgachev, van Geemen, and Kondō [11 ] , while Allcock and Freitag used Borcherds products to construct automorphic forms reflecting the classical invariant theory of cubic surfaces [3 ] . Thus invariant theory, period mappings, and automorphic forms give complementary descriptions of the same moduli space.
A related period description arises from K 3 K3 K 3 surfaces obtained as the minimal resolutions of double coverings of the projective plane branched along six lines. Such a surface can be regarded as a natural generalization of an elliptic curve obtained as a double covering of P 1 \mathbb{P}^{1} P 1 branched at four points. Matsumoto, Sasaki, and Yoshida [28 ] studied this family through its periods and the associated Picard–Fuchs system. The periods are analogs of elliptic integrals, and their Picard–Fuchs system is the hypergeometric system of type ( 3 , 6 ) (3,6) ( 3 , 6 ) . The period domain for the family is the Hermitian symmetric domain D A 1 \mathcal{D}_{A_{1}} D A 1 of type IV and dimension four. By virtue of the Torelli theorem and the surjectivity of the period mapping for K 3 K3 K 3 surfaces, the moduli space of these K 3 K3 K 3 surfaces is identified with the complement of a Heegner divisor in the arithmetic quotient D A 1 / Γ B 1 \mathcal{D}_{A_{1}/\Gamma_{B_{1}}} D A 1 / Γ B 1 . Here, Γ B 1 \Gamma_{\mathrm{B}_1} Γ B 1 is the arithmetic group determined by the lattice polarization; its action on D A 1 \mathcal{D}_{\mathrm{A}_1} D A 1 is described in § 2.2 . The Baily–Borel compactification of this quotient was studied in detail in [28 ] . For applications of these periods, see [Mat93, Yos08, MT12, HLTY20].
We thus obtain two arithmetic quotients of dimension four: the quotient of the complex ball associated with cubic surfaces and the quotient of a domain of type IV attached to the lattice-polarized K 3 K3 K 3 surfaces. Projective duality between configurations of six ordered points in P 2 \mathbb{P}^2 P 2 and six ordered lines in the dual plane ( P 2 ) ∨ (\mathbb{P}^2)^\vee ( P 2 ) ∨ relates these two descriptions through the configuration space X ( 3 , 6 ) X(3, 6) X ( 3 , 6 ) . The aim of this paper is to make this relation precise by identifying the common open moduli space and the discriminant divisors that must be removed on the two sides, as suggested in [2 ] . We describe this common open subset inside the GIT quotient of six ordered points in P 2 \mathbb{P}^2 P 2 directly.
The family F 1 \mathcal{F}_1 F 1 of K 3 K3 K 3 surfaces
The family of K 3 K3 K 3 surfaces studied in [28 ] is characterized by the even lattice B 1 \mathrm{B}_1 B 1 of (11) . In this paper, we study the family F 1 \mathcal{F}_1 F 1 of K 3 K3 K 3 surfaces characterized by the lattice A 1 \mathrm{A}_1 A 1 of (10) . These surfaces are the van Geemen–Sarti partners, in the sense of [7 ] , of the K 3 K3 K 3 surfaces studied in [28 ] . The relation between the lattices B 1 \mathrm{B}_1 B 1 and A 1 \mathrm{A}_1 A 1 allows us to study the moduli space of the former family through the family F 1 \mathcal{F}_1 F 1 . In order to study smooth cubic surfaces using the periods of these K 3 K3 K 3 surfaces, as in § 1.1 , we must exclude configurations of six lines that are all tangent to a common conic in P 2 \mathbb{P}^2 P 2 . The explicit equation for the family F 1 \mathcal{F}_1 F 1 relates this condition to divisors defined by hyperplane arrangements in the period domain. We use Looijenga’s compactification to study modular forms with poles along the relevant arrangement; see § 4 .
Looijenga compactification
For arithmetic quotients equipped with hyperplane arrangements, Looijenga constructed natural compactifications interpolating between the Baily–Borel and toroidal compactifications [24 , 25 ] . This construction has been applied to the realization of GIT quotients as arrangement compactifications [4 , 5 , 15 , 20 ] . See also the subsequent developments in relation to the Hassett–Keel–Looijenga program [21 , 22 LO21] .
A fundamental feature of this construction is that the resulting compactification is closely related to the projective spectrum of a graded ring of automorphic forms allowed to have poles along the appropriate arrangement divisor [25 ] . For a conceptual overview, see [26 ] . For the lattice A 1 \mathrm{A}_1 A 1 , Nagano and Ueda [33 ] (resp. Wang and Williams [WW24]) determine the ring of holomorphic modular forms with characters (resp. meromorphic modular forms with trivial character and prescribed poles). We give a geometric realization of the generators of the latter ring as coefficients of a Weierstrass equation and study the corresponding forms with characters.
Overview of the results
We prove three main results, stated in § 2 . First, for the arrangement complement attached to the K 3 K3 K 3 surfaces of the family F 1 \mathcal{F}_1 F 1 , we give a geometric realization of the graded ring of meromorphic modular forms with trivial character and identify the resulting Looijenga compactification with P ( 2 , 4 , 6 , 8 , 10 ) \mathbb{P}(2, 4, 6, 8, 10) P ( 2 , 4 , 6 , 8 , 10 ) . The discriminant decomposes into two geometrically distinguished components. We determine the full graded ring by adjoining modular forms with characters that define the corresponding double coverings.
Second, we compare these orthogonal modular varieties with the complex hyperbolic and GIT quotients associated with cubic surfaces. We identify a common open subset of the GIT quotient of six ordered points in P 2 \mathbb{P}^2 P 2 and the ball quotient parametrizing marked nodal cubic surfaces. The collinearity strata in this open subset are precisely those whose associated roots in the root lattice of type E 6 E_6 E 6 are mutually orthogonal.
Third, starting from the description of the orbifold Picard group by Nagano and Ueda [33 ] , we interpret three natural double coverings of P ( 2 , 3 , 4 , 5 , 6 ) \mathbb{P}(2, 3, 4, 5, 6) P ( 2 , 3 , 4 , 5 , 6 ) in terms of modular forms with characters. Their branch divisors correspond to the locus of U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) U \oplus E_8(-1) \oplus E_7(-1) U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) -polarized K 3 K3 K 3 surfaces, a jump of the Néron–Severi lattice caused by the coalescence of two nodal fibers, and the symmetry exchanging the two A 1 ( − 1 ) A_1(-1) A 1 ( − 1 ) summands. We also describe these coverings in terms of quotients of the Coble fourfold and permutations of its coordinates.
Main results
Meromorphic modular forms for a family of lattice-polarized K3 surfaces
Let A 1 A_1 A 1 be an even lattice of signature ( 2 , 4 ) (2,4) ( 2 , 4 ) with the intersection matrix U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) U \oplus U \oplus A_1(-1) \oplus A_1(-1) U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) , where U U U is the hyperbolic lattice of rank two and A 1 ( − 1 ) A_1(-1) A 1 ( − 1 ) is the root lattice. Let D A 1 D_{A_1} D A 1 be a connected component of { [ ξ ] ∈ P ( A 1 ⊗ C ) ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ˉ ) > 0 } \{[\xi] \in\mathbb{P}(A_1 \otimes\mathbb{C}) \mid(\xi,\xi) = 0, (\xi,\bar{\xi}) > 0\} {[ ξ ] ∈ P ( A 1 ⊗ C ) ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ˉ ) > 0 } . Let O + ( A 1 ) O^+(A_1) O + ( A 1 ) be the subgroup of O ( A 1 ) O(A_1) O ( A 1 ) preserving the connected component D A 1 D_{A_1} D A 1 . Let Γ A 1 \Gamma_{A_1} Γ A 1 be the intersection of O + ( A 1 ) O^+(A_1) O + ( A 1 ) and the stable orthogonal subgroup of A 1 A_1 A 1 . Let H ∞ \mathcal{H}_\infty H ∞ and H 0 \mathcal{H}_0 H 0 be the Γ A 1 \Gamma_{A_1} Γ A 1 -orbits of linear hyperplanes orthogonal to the two summands isometric to A 1 ( − 1 ) A_1(-1) A 1 ( − 1 ) . Let H ∞ H_\infty H ∞ and H 0 H_0 H 0 be their respective supports in D A 1 D_{A_1} D A 1 (for details, see Definition 2.7 ).
In § 4 , we will determine the structure of the ring A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{A_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ of meromorphic modular forms, including all characters, with respect to Γ A 1 \Gamma_{A_1} Γ A 1 with poles along the divisor H ∞ H_\infty H ∞ . The following is our first main theorem.
Theorem 2.1. The ring A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{A_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ is isomorphic to
C [ r 2 , r 4 , r 6 , r 8 , r 10 , r ~ 4 , r ~ 30 ] / ( r ~ 4 2 − r 8 , r ~ 30 2 − d 60 ( r ) ) . \mathbb{C}[r_2,r_4,r_6,r_8,r_{10},\widetilde{r}_4,\widetilde{r}_{30}]/(\widetilde{r}_4^2-r_8,\widetilde{r}_{30}^2-d_{60}(r)). C [ r 2 , r 4 , r 6 , r 8 , r 10 , r 4 , r 30 ] / ( r 4 2 − r 8 , r 30 2 − d 60 ( r )) . Here, r k r_k r k for k ∈ { 2 , 4 , 6 , 8 , 10 } k \in\{2,4,6,8,10\} k ∈ { 2 , 4 , 6 , 8 , 10 } are modular forms of weight k k k with trivial character and d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] d_{60}(r) \in\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] . Also, r ~ 4 \widetilde{r}_4 r 4 (resp. r ~ 30 \widetilde{r}_{30} r 30 ) is a modular form of weight 4 4 4 (resp. 30 30 30 ) with a nontrivial character. The polynomial d 60 ( r ) d_{60}(r) d 60 ( r ) is defined by the discriminant factorization in Definition A.1 , and Proposition A.3 gives its resultant formula.
We briefly summarize here how we obtain the theorem. We remark that the lattice A 1 A_1 A 1 is a primitive sublattice of the K 3 K3 K 3 lattice L K 3 L_{K3} L K 3 , which is the even unimodular lattice of signature ( 3 , 19 ) (3,19) ( 3 , 19 ) . Let M 1 M_1 M 1 be the orthogonal complement of A 1 A_1 A 1 with respect to the unimodular lattice L K 3 L_{K3} L K 3 . Then, M 1 M_1 M 1 is a primitive sublattice of L K 3 L_{K3} L K 3 with signature ( 1 , 15 ) (1,15) ( 1 , 15 ) and we can study K 3 K3 K 3 surfaces polarized by the lattice M 1 M_1 M 1 . Unless otherwise stated, an M 1 M_1 M 1 -polarized K 3 K3 K 3 surface in this paper means a pseudo-ample M 1 M_1 M 1 -polarized K 3 K3 K 3 surface in the sense recalled in § 4 . We will obtain the family F 1 \mathcal{F}_1 F 1 of isomorphism classes of M 1 M_1 M 1 -polarized K 3 K3 K 3 surfaces explicitly. A member of F 1 \mathcal{F}_1 F 1 can be characterized as the quotient of a well-known K 3 K3 K 3 surface, which was studied in [28 ] and arises from the minimal resolution of the double covering of P 2 \mathbb{P}^2 P 2 branched along six lines, by a symplectic involution. Meanwhile, the moduli space of M 1 M_1 M 1 -polarized K 3 K3 K 3 surfaces is identified with the arithmetic quotient D A 1 / Γ A 1 D_{A_1}/\Gamma_{A_1} D A 1 / Γ A 1 by virtue of the Torelli theorem and the surjectivity of the period mapping for K 3 K3 K 3 surfaces. The generators of the ring A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{A_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ will appear as the deformation parameters for the family F 1 \mathcal{F}_1 F 1 . They are regarded as modular forms on D A 1 D_{A_1} D A 1 via the period mapping.
Our construction of the meromorphic modular forms with poles along the arrangement divisor H ∞ H_\infty H ∞ is based on the fact that members of F 1 \mathcal{F}_1 F 1 degenerate along H ∞ H_\infty H ∞ to the Shioda–Inose partners of principally polarized abelian surfaces. In particular, the Looijenga compactification of ( D A 1 ∖ H ∞ ) / Γ A 1 (D_{A_1} \setminus H_\infty)/\Gamma_{A_1} ( D A 1 ∖ H ∞ ) / Γ A 1 is identified with the weighted projective space P ( 2 , 4 , 6 , 8 , 10 ) \mathbb{P}(2,4,6,8,10) P ( 2 , 4 , 6 , 8 , 10 ) via the period mapping (see (32) ).
We remark that the structure of the subring C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] C [ r 2 , r 4 , r 6 , r 8 , r 10 ] of modular forms in Theorem 2.1 with trivial character is determined in [WW24]. Our theorem gives a geometric interpretation of this ring. Namely, there is a natural interpretation for the generators r k r_k r k (k ∈ { 2 , 4 , 6 , 8 , 10 } k \in\{2,4,6,8,10\} k ∈ { 2 , 4 , 6 , 8 , 10 } ). As we will see in § 4 , they appear as coefficients of a defining equation for the family of M 1 M_1 M 1 -polarized K 3 K3 K 3 surfaces. Our proof is based on a correspondence between the points of D A 1 D_{A_1} D A 1 and the coefficients via the period mapping. This is a natural counterpart of the classical fact that the ring of elliptic modular forms for S L 2 ( Z ) \mathrm{SL}_2(\mathbb{Z}) SL 2 ( Z ) is generated by g 2 g_2 g 2 and g 3 g_3 g 3 , which appear as coefficients of the Weierstrass equation y 2 = 4 x 3 − g 2 x − g 3 y^2 = 4x^3 - g_2x - g_3 y 2 = 4 x 3 − g 2 x − g 3 of the family of elliptic curves. Our approach also determines the generators with nontrivial characters by studying the period mapping for our K 3 K3 K 3 surfaces and the structure of the arithmetic quotient ( D A 1 \ H ∞ ) / Γ A 1 (D_{A_1} \backslash H_{\infty})/\Gamma_{A_1} ( D A 1 \ H ∞ ) / Γ A 1 .
Comparison with X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) and cubic surfaces
We next compare the modular varieties of type IV arising from ( 3 , 6 ) (3,6) ( 3 , 6 ) -K 3 K3 K 3 surfaces with the ball quotients and the GIT quotients associated with marked cubic surfaces. Put
X ( 3 , 6 ) : = { six labeled lines in P 2 in general position } / P G L 3 . X(3,6):=\{\text{six labeled lines in }\mathbb{P}^{2}\text{ in general position}\}/\mathrm{PGL}_{3}. X ( 3 , 6 ) := { six labeled lines in P 2 in general position } / PGL 3 .
Here “general position” means that no three of the six lines are concurrent. By projective duality, the same open variety can be described as
X ( 3 , 6 ) ≅ { ( p 1 , … , p 6 ) ∈ ( P 2 ) 6 ∣ p i ≠ p j , p i , p j , p k are not collinear } / P G L 3 . X(3,6)\cong\{(p_{1},\ldots,p_{6})\in(\mathbb{P}^{2})^{6}\mid p_{i}\ne p_{j},\ p_{i},p_{j},p_{k}\text{ are not collinear}\}/\mathrm{PGL}_{3}. X ( 3 , 6 ) ≅ {( p 1 , … , p 6 ) ∈ ( P 2 ) 6 ∣ p i = p j , p i , p j , p k are not collinear } / PGL 3 .
Let
X ( 3 , 6 ) ‾ G I T : = ( P 2 ) 6 / / S L 3 ( C ) \overline{X(3,6)}^{\mathrm{GIT}}:=(\mathbb{P}^{2})^{6}//\mathrm{SL}_{3}(\mathbb{C}) X ( 3 , 6 ) GIT := ( P 2 ) 6 // SL 3 ( C )
be the GIT quotient for the symmetric linearization O ( 1 , … , 1 ) \mathcal{O}(1,\ldots,1) O ( 1 , … , 1 ) . The GIT quotient is the Coble fourfold X ( 3 , 6 ) ‾ G I T ⊂ P ( 2 , 1 6 ) \overline{X(3,6)}^{\mathrm{GIT}}\subset\mathbb{P}(2,1^{6}) X ( 3 , 6 ) GIT ⊂ P ( 2 , 1 6 ) [11 Section 2.10] . In weighted homogeneous coordinates ( z ; x 1 , … , x 6 ) (z;x_{1},\ldots,x_{6}) ( z ; x 1 , … , x 6 ) , with deg z = 2 \deg z=2 deg z = 2 and deg x i = 1 \deg x_{i}=1 deg x i = 1 , its equations are
s 1 ( x ) = 0 , z 2 = 4 s 4 ( x ) − s 2 ( x ) 2 , (1) \begin{aligned}
s_{1}(x)&=0,\\
z^{2}&=4s_{4}(x)-s_{2}(x)^{2}, \tag*{(1)}
\end{aligned} s 1 ( x ) z 2 = 0 , = 4 s 4 ( x ) − s 2 ( x ) 2 , (1)
where s k s_{k} s k is the k k k -th elementary symmetric polynomial. We call ( z ; x 1 , … , x 6 ) (z;x_{1},\ldots,x_{6}) ( z ; x 1 , … , x 6 ) the Coble coordinates . The projection to ( x 1 : ⋯ : x 6 ) (x_{1}:\cdots:x_{6}) ( x 1 : ⋯ : x 6 ) is a double covering of P 4 = { s 1 = 0 } ⊂ P 5 \mathbb{P}^{4}=\{s_{1}=0\}\subset\mathbb{P}^{5} P 4 = { s 1 = 0 } ⊂ P 5 , branched along the Igusa quartic. Its deck transformation z ↦ − z z\mapsto-z z ↦ − z is the Gale involution.
We distinguish two actions of S 6 S_{6} S 6 on this fourfold. The action S 6 p t s S_{6}^{\mathrm{pts}} S 6 pts is induced by permutations of the six labeled points. For g ∈ S 6 g\in S_{6} g ∈ S 6 , it satisfies g ( x i ) = x φ ( g ) ( i ) g(x_{i})=x_{\varphi(g)(i)} g ( x i ) = x φ ( g ) ( i ) and g ( z ) = sgn ( g ) z g(z)=\operatorname{sgn}(g)z g ( z ) = sgn ( g ) z , where φ \varphi φ is the outer automorphism relating the point labels to the Coble coordinates; it preserves the sign character. The action S 6 f i x S_{6}^{\mathrm{fix}} S 6 fix has the same action on the x i x_{i} x i and fixes z z z . Equivalently, it is obtained by composing each odd permutation in S 6 p t s S_{6}^{\mathrm{pts}} S 6 pts with the Gale involution. See [37 Section 5.6, Theorems 41 and 43] . Unless a superscript is specified, the action of S 6 S_{6} S 6 on P 4 \mathbb{P}^{4} P 4 permutes the Coble coordinates.
Put
D c o n ∘ : = { p 1 , … , p 6 lie on a conic } ∩ X ( 3 , 6 ) . D_{\mathrm{con}}^{\circ}:=\{p_{1},\ldots,p_{6}\text{ lie on a conic}\}\cap X(3,6). D con ∘ := { p 1 , … , p 6 lie on a conic } ∩ X ( 3 , 6 ) .
Equivalently, in terms of configurations of six lines, D c o n ∘ D_{\mathrm{con}}^{\circ} D con ∘ is the locus where the six lines are tangent to a common conic. Then the locus of smooth marked cubic surfaces is
M c u b m ≅ X ( 3 , 6 ) ∖ D c o n ∘ . (2) \mathcal{M}_{\mathrm{cub}}^{\mathrm{m}}\cong X(3,6)\setminus D_{\mathrm{con}}^{\circ}. \tag*{(2)} M cub m ≅ X ( 3 , 6 ) ∖ D con ∘ . (2)
Let B 4 \mathbb{B}^{4} B 4 be the complex ball of dimension four associated with the Eisenstein lattice used in the ball uniformization of cubic surfaces, and let Γ m \Gamma_{\mathrm{m}} Γ m be the arithmetic group for marked cubic surfaces. By [11 Section 2.12, Corollary 9.3] , the moduli space of marked nodal cubic surfaces is identified with B 4 / Γ m \mathbb{B}^{4}/\Gamma_{\mathrm{m}} B 4 / Γ m and the discriminant divisor has 36 = 1 + 20 + 15 36=1+20+15 36 = 1 + 20 + 15 irreducible components:
D c o n , D c o l = ⋃ i < j < k D i j k , D p a i r = ⋃ i < j D i j . D_{\mathrm{con}},\qquad D_{\mathrm{col}}=\bigcup_{i<j<k}D_{ijk},\qquad D_{\mathrm{pair}}=\bigcup_{i<j}D_{ij}. D con , D col = i < j < k ⋃ D ij k , D pair = i < j ⋃ D ij .
These are divisors on B 4 / Γ m \mathbb{B}^{4}/\Gamma_{\mathrm{m}} B 4 / Γ m . Their root labels are specified in (45) . The geometric construction of D i j D_{ij} D ij is described in (48) .
For a lattice A A A of signature ( 2 , n ) (2,n) ( 2 , n ) , let D A \mathcal{D}_{A} D A be a chosen connected component of the associated domain of type IV. Let Γ ⊂ O + ( A ) \Gamma\subset O^{+}(A) Γ ⊂ O + ( A ) be an arithmetic subgroup, and let ℓ , δ ∈ Z > 0 \ell,\delta\in\mathbb{Z}_{>0} ℓ , δ ∈ Z > 0 . Write
D δ = { [ ξ ] ∈ D A ∣ ( ξ , δ ) = 0 } . D_{\delta}=\{[\xi]\in\mathcal{D}_{A}\mid(\xi,\delta)=0\}. D δ = {[ ξ ] ∈ D A ∣ ( ξ , δ ) = 0 } .
for δ ∈ A \delta\in A δ ∈ A . The divisor H ( ℓ ) ( Γ ) \mathcal{H}^{(\ell)}(\Gamma) H ( ℓ ) ( Γ ) is the reduced image in D A / Γ \mathcal{D}_A/\Gamma D A /Γ of the union of D δ \mathcal{D}_\delta D δ over primitive vectors δ ∈ A \delta\in A δ ∈ A with δ 2 = − ℓ \delta^2=-\ell δ 2 = − ℓ . The refined notation H A , div = d ( ℓ ) ( Γ ) \mathcal{H}^{(\ell)}_{A,\operatorname{div}=d}(\Gamma) H A , div = d ( ℓ ) ( Γ ) restricts this union to vectors satisfying
div A ( δ ) = gcd { ( δ , x ) ∣ x ∈ A } = d . (3) \operatorname{div}_A(\delta)=\gcd\{(\delta,x)\mid x\in A\}=d. \tag*{(3)} div A ( δ ) = g cd{( δ , x ) ∣ x ∈ A } = d . (3)
Put B 1 = U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) B_1=U(2)\oplus U(2)\oplus A_1(-1)\oplus A_1(-1) B 1 = U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) , and let Γ B 1 \Gamma_{B_1} Γ B 1 be the associated stable orthogonal subgroup as (12) . Let T : O + ( A 1 ) → ∼ O + ( B 1 ) T:O^+(A_1)\xrightarrow{\sim}O^+(B_1) T : O + ( A 1 ) ∼ O + ( B 1 ) be the isomorphism constructed in (41) . We identify the domains of type IV associated with A 1 A_1 A 1 and B 1 B_1 B 1 by the underlying rational similarity. Write A B 1 = B 1 ∨ / B 1 A_{B_1}=B_1^\vee/B_1 A B 1 = B 1 ∨ / B 1 and let q B 1 : A B 1 → Q / 2 Z q_{B_1}:A_{B_1}\to\mathbb{Q}/2\mathbb{Z} q B 1 : A B 1 → Q /2 Z be its discriminant quadratic form. Every primitive vector of B 1 B_1 B 1 has divisibility two. Let β 1 , β 2 \beta_1,\beta_2 β 1 , β 2 generate its two summands isometric to A 1 ( − 1 ) A_1(-1) A 1 ( − 1 ) , and let τ \tau τ interchange these generators and act trivially on U ( 2 ) ⊕ U ( 2 ) U(2)\oplus U(2) U ( 2 ) ⊕ U ( 2 ) . Put
γ τ = [ β 1 − β 2 2 ] = [ β 1 + β 2 2 ] ∈ A B 1 . \gamma_\tau=\left[\frac{\beta_1-\beta_2}{2}\right]=\left[\frac{\beta_1+\beta_2}{2}\right]\in A_{B_1}. γ τ = [ 2 β 1 − β 2 ] = [ 2 β 1 + β 2 ] ∈ A B 1 .
For γ ∈ A B 1 \gamma\in A_{B_1} γ ∈ A B 1 with q B 1 ( γ ) = − 1 q_{B_1}(\gamma)=-1 q B 1 ( γ ) = − 1 , define
H B 1 , γ ( 4 ) ( Γ B 1 ) = Im ( ⋃ r ∈ B 1 p r i m i t i v e r 2 = − 4 , [ r / 2 ] = γ D r ⟶ D A 1 / Γ B 1 ) r e d . \mathcal{H}^{(4)}_{B_1,\gamma}(\Gamma_{B_1})=\operatorname{Im}\left(\bigcup_{\substack{r\in B_1\ \mathrm{primitive}\\ r^2=-4,\ [r/2]=\gamma}}\mathcal{D}_r\longrightarrow\mathcal{D}_{A_1}/\Gamma_{B_1}\right)_{\mathrm{red}}. H B 1 , γ ( 4 ) ( Γ B 1 ) = Im r ∈ B 1 primitive r 2 = − 4 , [ r /2 ] = γ ⋃ D r ⟶ D A 1 / Γ B 1 red .
We distinguish the two parts
H B 1 , τ ( 4 ) ( Γ B 1 ) = H B 1 , γ τ ( 4 ) ( Γ B 1 ) , \mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1})=\mathcal{H}^{(4)}_{B_1,\gamma_\tau}(\Gamma_{B_1}), H B 1 , τ ( 4 ) ( Γ B 1 ) = H B 1 , γ τ ( 4 ) ( Γ B 1 ) ,
H B 1 , S 6 ( 4 ) ( Γ B 1 ) = ⋃ γ ∈ A B 1 q B 1 ( γ ) = − 1 , γ ≠ γ τ H B 1 , γ ( 4 ) ( Γ B 1 ) . \mathcal{H}^{(4)}_{B_1,S_6}(\Gamma_{B_1})=
\bigcup_{\substack{\gamma\in A_{B_1}\\q_{B_1}(\gamma)=-1,\ \gamma\ne\gamma_\tau}}
\mathcal{H}^{(4)}_{B_1,\gamma}(\Gamma_{B_1}). H B 1 , S 6 ( 4 ) ( Γ B 1 ) = γ ∈ A B 1 q B 1 ( γ ) = − 1 , γ = γ τ ⋃ H B 1 , γ ( 4 ) ( Γ B 1 ) .
These are the τ \tau τ -type and transposition parts of H B 1 ( 4 ) ( Γ B 1 ) \mathcal{H}^{(4)}_{B_1}(\Gamma_{B_1}) H B 1 ( 4 ) ( Γ B 1 ) . The subscript S 6 S_6 S 6 refers to the permutation action on the Coble coordinates. Proposition 5.3 gives their geometric interpretations and component counts.
On the orthogonal side, the period mapping for double coverings branched along six lines gives
X ( 3 , 6 ) ≅ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) . (4) X(3,6)\cong\mathcal{D}_{A_1}/\Gamma_{B_1}\setminus\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}). \tag*{(4)} X ( 3 , 6 ) ≅ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) . (4)
Under the extension of the period mapping in Theorem 5.1 , this arrangement corresponds to the twenty collinearity divisors in the GIT quotient of six ordered points in P 2 \mathbb{P}^2 P 2 . On the common locus of Theorem 5.4 , their images on the ball quotient are the divisors D i j k D_{ijk} D ij k . Under the period mapping Φ M S Y \Phi_{\mathrm{MSY}} Φ MSY of Theorem 5.1 , the conic locus satisfies
Φ M S Y ( D c o n ∘ ) = H B 1 , τ ( 4 ) ( Γ B 1 ) ∖ H B 1 ( 2 ) ( Γ B 1 ) . \Phi_{\mathrm{MSY}}(D^\circ_{\mathrm{con}})=\mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1})\setminus\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}). Φ MSY ( D con ∘ ) = H B 1 , τ ( 4 ) ( Γ B 1 ) ∖ H B 1 ( 2 ) ( Γ B 1 ) .
Under the comparison on the common open subset, it corresponds to D c o n ∖ ( D c o l ∪ D p a i r ) D_{\mathrm{con}}\setminus(D_{\mathrm{col}}\cup D_{\mathrm{pair}}) D con ∖ ( D col ∪ D pair ) on the ball quotient. Thus the locus that corresponds to smooth marked cubic surfaces may also be written as
M c u b m ≅ D A 1 / Γ B 1 ∖ ( H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ) . (5) \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}}\cong\mathcal{D}_{A_1}/\Gamma_{B_1}\setminus\left(\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1})\cup\mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1})\right). \tag*{(5)} M cub m ≅ D A 1 / Γ B 1 ∖ ( H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ) . (5)
Let X c o m X_{\mathrm{com}} X com be the quotient by P G L 3 \mathrm{PGL}_3 PGL 3 of ordered sextuples of distinct points in P 2 \mathbb{P}^2 P 2 such that no four are collinear and no partition into two triples has both triples collinear. The following theorem extends the comparison to this open subset of X ( 3 , 6 ) G I T X(3,6)^{\mathrm{GIT}} X ( 3 , 6 ) GIT .
Theorem 2.2 (Theorem 5.4 , Corollary 5.5). There is a natural isomorphism
X ( 3 , 6 ) ≅ B 4 / Γ m ∖ ( D c o l ∪ D p a i r ) ≅ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) . X(3,6)\cong\mathbb{B}^4/\Gamma_{\mathrm{m}}\setminus(D_{\mathrm{col}}\cup D_{\mathrm{pair}})\cong\mathcal{D}_{A_1}/\Gamma_{B_1}\setminus\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}). X ( 3 , 6 ) ≅ B 4 / Γ m ∖ ( D col ∪ D pair ) ≅ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) . Moreover, it extends to an isomorphism of algebraic varieties X c o m ≅ ( B 4 / Γ m ) ∖ D p a i r X_{\mathrm{com}}\cong(\mathbb{B}^4/\Gamma_{\mathrm{m}})\setminus D_{\mathrm{pair}} X com ≅ ( B 4 / Γ m ) ∖ D pair .
Theorem 2.2 gives an analog of the Janus-like phenomenon for the Segre cubic and Igusa quartic [17 , 2 ] . Further directions concerning an automorphic realization of the Coble map and the comparison of toroidal compactifications are discussed at the end of § 6.2 .
Remark 2.3. The quotient ( P 1 ) 6 / / P G L 2 (\mathbb{P}^{1})^{6}//\mathrm{PGL}_{2} ( P 1 ) 6 // PGL 2 is the Segre cubic, whose projective dual is the Igusa quartic [17 Sections 3.2 and 3.3] . These varieties are the Baily–Borel compactifications of a ball quotient of dimension three and a Siegel modular variety, respectively [18 , 17 ] . This is the classical example motivating the comparison in Theorem 2.2 .
Modular forms with characters and branch geometry
The comparison in Theorem 2.2 leads us to study the finite coverings of the arithmetic quotients and their branch divisors. Let
X A 1 : = [ D A 1 / O + ( A 1 ) ] \mathcal{X}_{A_1}:=[\mathcal{D}_{A_1}/\mathrm{O}^{+}(A_1)] X A 1 := [ D A 1 / O + ( A 1 )]
be the quotient stack, with the central subgroup { ± i d } \{\pm\mathrm{id}\} { ± id } retained. Nagano and Ueda [33 Section 3] proved that
Pic ( X A 1 ) ≅ Z ⊕ ( Z / 2 Z ) ⊕ 3 \operatorname{Pic}(\mathcal{X}_{A_1})\cong\mathbb{Z}\oplus(\mathbb{Z}/2\mathbb{Z})^{\oplus3} Pic ( X A 1 ) ≅ Z ⊕ ( Z /2 Z ) ⊕ 3
and constructed three generators of its torsion subgroup using modular forms with characters. Starting from this description, we make the generators explicit in terms of reflections and identify the geometric meaning of the associated double coverings. This relates the Picard group to quotients of the Coble fourfold, symmetries of the lattice polarization, and degenerations of K 3 K3 K 3 surfaces.
By [33 Theorem 1.1] , the graded ring A ( O + ( A 1 ) ) \mathcal{A}(\mathrm{O}^{+}(A_1)) A ( O + ( A 1 )) of holomorphic modular forms, including all characters, is generated by
v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 4 , v ~ 10 , v ~ 30 , v_{4},\ v_{6},\ v_{8},\ v_{10},\ v_{12},\ \widetilde{v}_{4},\ \widetilde{v}_{10},\ \widetilde{v}_{30}, v 4 , v 6 , v 8 , v 10 , v 12 , v 4 , v 10 , v 30 ,
where v ~ 4 \widetilde{v}_{4} v 4 , v ~ 10 \widetilde{v}_{10} v 10 , and v ~ 30 \widetilde{v}_{30} v 30 are modular forms with nontrivial characters satisfying
v ~ 4 2 = v 8 , v ~ 10 2 = v 10 2 − 4 v 8 v 12 , v ~ 30 2 = d 60 ( v ) . (6) \widetilde{v}_{4}^{2}=v_{8},\qquad\widetilde{v}_{10}^{2}=v_{10}^{2}-4v_{8}v_{12},\qquad\widetilde{v}_{30}^{2}=d_{60}(v). \tag*{(6)} v 4 2 = v 8 , v 10 2 = v 10 2 − 4 v 8 v 12 , v 30 2 = d 60 ( v ) . (6)
Here v = ( v 4 , v 6 , v 8 , v 10 , v 12 ) v=(v_{4},v_{6},v_{8},v_{10},v_{12}) v = ( v 4 , v 6 , v 8 , v 10 , v 12 ) , and d 60 ( v ) d_{60}(v) d 60 ( v ) is the residual discriminant expressed in these coordinates. Its relation to d 60 ( r ) d_{60}(r) d 60 ( r ) is specified at the beginning of Appendix A. For an arithmetic quotient Z Z Z , let Z ∗ Z^{*} Z ∗ be its Baily–Borel compactification. Let
Y A 1 : = ( D A 1 / O + ( A 1 ) ) ∗ ≅ P ( 2 , 3 , 4 , 5 , 6 ) . Y_{A_1}:=(\mathcal{D}_{A_1}/\mathrm{O}^{+}(A_1))^{*}\cong\mathbb{P}(2,3,4,5,6). Y A 1 := ( D A 1 / O + ( A 1 ) ) ∗ ≅ P ( 2 , 3 , 4 , 5 , 6 ) .
For k ∈ { 4 , 10 , 30 } k\in\{4,10,30\} k ∈ { 4 , 10 , 30 } , let χ k ∈ Char ( O + ( A 1 ) ) \chi_{k}\in\operatorname{Char}(\mathrm{O}^{+}(A_1)) χ k ∈ Char ( O + ( A 1 )) be the character of v ~ k \widetilde{v}_{k} v k , and put Γ χ k = ker ( χ k ) \Gamma_{\chi_k}=\ker(\chi_k) Γ χ k = ker ( χ k ) . Since the weights are even, χ k ( − i d ) = 1 \chi_k(-\mathrm{id})=1 χ k ( − id ) = 1 . Thus each χ k \chi_k χ k defines a double covering
Y χ k : = ( D A 1 / Γ χ k ) ∗ ⟶ Y A 1 . (7) Y_{\chi_k}:=(\mathcal{D}_{A_1}/\Gamma_{\chi_k})^{*}\longrightarrow Y_{A_1}. \tag*{(7)} Y χ k := ( D A 1 / Γ χ k ) ∗ ⟶ Y A 1 . (7)
The following theorem describes these coverings and their branch divisors, using the actions S 6 p t s S_{6}^{\mathrm{pts}} S 6 pts and S 6 f i x S_{6}^{\mathrm{fix}} S 6 fix fixed in § 2.2 .
Theorem 2.4 (Theorem 5.7 , Proposition 5.6 ). The three distinguished double coverings Y χ k → Y A 1 Y_{\chi_k}\to Y_{A_1} Y χ k → Y A 1 , for k = 4 , 10 , 30 k=4,10,30 k = 4 , 10 , 30 , have the following reduced divisorial branch loci:
Table 2.4.
Moreover, Γ χ 10 = Γ A 1 \Gamma_{\chi_{10}}=\Gamma_{A_1} Γ χ 10 = Γ A 1 , and there are isomorphisms over Y A 1 Y_{A_1} Y A 1 :
Y χ 4 ≅ X ( 3 , 6 ) ‾ G I T / S 6 f i x ≅ P ( 2 , 2 , 3 , 5 , 6 ) . Y_{\chi_4}\cong\overline{\mathbb{X}(3,6)}^{\mathrm{GIT}}/S_{6}^{\mathrm{fix}}\cong\mathbb{P}(2,2,3,5,6). Y χ 4 ≅ X ( 3 , 6 ) GIT / S 6 fix ≅ P ( 2 , 2 , 3 , 5 , 6 ) . Under the identification Y A 1 ≅ P 4 / S 6 Y_{A_1}\cong\mathbb{P}^{4}/S_{6} Y A 1 ≅ P 4 / S 6 , the covering for χ 30 \chi_{30} χ 30 is isomorphic to P 4 / A 6 → P 4 / S 6 \mathbb{P}^{4}/A_{6}\to\mathbb{P}^{4}/S_{6} P 4 / A 6 → P 4 / S 6 , where A 6 ⊂ S 6 A_{6}\subset S_{6} A 6 ⊂ S 6 is the alternating group acting by even permutations of the six Coble coordinates. The
quotient X ( 3 , 6 ) G I T / S 6 p t s X(3,6)^{\mathrm{GIT}}/S_{6}^{\mathrm{pts}} X ( 3 , 6 ) GIT / S 6 pts instead corresponds to ker ( χ 4 χ 30 ) \ker(\chi_{4}\chi_{30}) ker ( χ 4 χ 30 ) and has reduced divisorial branch locus { v 8 d 60 ( v ) = 0 } \{v_{8}d_{60}(v)=0\} { v 8 d 60 ( v ) = 0 } .
Remark 2.5. The three branch divisors have the following geometric interpretations. For χ 4 \chi_{4} χ 4 , a general point corresponds to replacing a fiber of type I I I ∗ III^{*} I I I ∗ by one of type I I ∗ II^{*} I I ∗ , giving a K 3 K3 K 3 surface polarized by M 2 = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) M_{2}=U\oplus E_{8}(-1)\oplus E_{7}(-1) M 2 = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) . The character χ 10 \chi_{10} χ 10 records the symmetry exchanging the two summands isometric to E 7 ( − 1 ) E_{7}(-1) E 7 ( − 1 ) . For χ 30 \chi_{30} χ 30 , two fibers of type I 1 I_{1} I 1 coalesce into one of type I 2 I_{2} I 2 . The corresponding divisors for configurations of six lines are described in Theorem 5.7 and its proof.
Organization of the paper
The paper is organized as follows. In § 4 , we review the necessary background on lattice-polarized K 3 K3 K 3 surfaces, including the families F 1 \mathcal{F}_{1} F 1 and F 2 \mathcal{F}_{2} F 2 , the relevant hyperplane arrangements, and meromorphic modular forms with characters. In § 5 , we recall the basic geometry of cubic surfaces, their GIT quotients, and the ball quotient uniformization. In § 6 , we compare the arithmetic quotients via the configuration space X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) , study the associated compactifications, and analyze the three distinguished double coverings. Further directions concerning the Coble map and toroidal compactifications are discussed at the end of § 6.2 . Appendix A contains the calculation for discriminants.
Acknowledgments
The authors are grateful to Takuya Yamauchi for his valuable and insightful comments. The first author is grateful for the hospitality of Kanazawa University and is partially supported by the Alexander von Humboldt Foundation through a Humboldt research fellowship, by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the Collaborative Research Centre TRR 326 Geometry and Arithmetic of Uniformized Structures , project number 444845124, and by Waseda University Grant for Special Research Projects (Project number: 2026R022). The second author is supported by JST FOREST Program (JPMJFR2235) and JSPS Grant-in-Aid for Scientific Research (26K06725). The first author uses AI for editing the manuscript.
Notation and preliminaries
For domains of type IV, let H \mathcal{H} H be a collection of linear hyperplanes in the ambient complex vector space, and let H H H be its support in the period domain. The notation H \mathcal{H} H is used for reduced divisors on arithmetic quotients, both in the orthogonal and in the ball quotient settings. The lattice and the arithmetic subgroup are indicated when needed to distinguish the relevant divisors.
Lattice-polarized K 3 K3 K 3 surfaces. Let S S S be a K 3 K3 K 3 surface. Let NS ( S ) \operatorname{NS}(S) NS ( S ) be its Néron–Severi lattice: NS ( S ) : = H 2 ( S , Z ) ∩ H 1 , 1 ( S ) \operatorname{NS}(S):=H^{2}(S,\mathbb{Z})\cap H^{1,1}(S) NS ( S ) := H 2 ( S , Z ) ∩ H 1 , 1 ( S ) . Since H 2 ( S , Z ) H^{2}(S,\mathbb{Z}) H 2 ( S , Z ) can be identified with H 2 ( S , Z ) H_{2}(S,\mathbb{Z}) H 2 ( S , Z ) via Poincaré duality, we will often regard NS ( S ) \operatorname{NS}(S) NS ( S ) as a subgroup of H 2 ( S , Z ) H_{2}(S,\mathbb{Z}) H 2 ( S , Z ) .
Let H R 1 , 1 ( S ) : = H 1 , 1 ( S ) ∩ H 2 ( S , R ) H_{\mathbb{R}}^{1,1}(S):=H^{1,1}(S)\cap H^{2}(S,\mathbb{R}) H R 1 , 1 ( S ) := H 1 , 1 ( S ) ∩ H 2 ( S , R ) . Let V ( S ) + V(S)^{+} V ( S ) + be the positive cone of S S S . Namely, it is the connected component of
V ( S ) : = { x ∈ H R 1 , 1 ( S ) ∣ ( x , x ) > 0 } V(S):=\{x\in H_{\mathbb{R}}^{1,1}(S)\mid(x,x)>0\} V ( S ) := { x ∈ H R 1 , 1 ( S ) ∣ ( x , x ) > 0 }
which contains the class of a Kähler form on S S S . Let
Δ ( S ) : = { δ ∈ NS ( S ) ∣ ( δ , δ ) = − 2 } , \Delta(S):=\{\delta\in\operatorname{NS}(S)\mid(\delta,\delta)=-2\}, Δ ( S ) := { δ ∈ NS ( S ) ∣ ( δ , δ ) = − 2 } ,
and let Δ ( S ) + \Delta(S)^{+} Δ ( S ) + be the subset of effective classes in Δ ( S ) \Delta(S) Δ ( S ) . Put
C ( S ) : = { x ∈ V ( S ) + ∣ ( x , δ ) ≥ 0 for all δ ∈ Δ ( S ) + } . C(S):=\{x\in V(S)^{+}\mid(x,\delta)\geq0\text{ for all }\delta\in\Delta(S)^{+}\}. C ( S ) := { x ∈ V ( S ) + ∣ ( x , δ ) ≥ 0 for all δ ∈ Δ ( S ) + } .
Then NS ( S ) + : = C ( S ) ∩ H 2 ( S , Z ) \operatorname{NS}(S)^{+}:=C(S)\cap H_{2}(S,\mathbb{Z}) NS ( S ) + := C ( S ) ∩ H 2 ( S , Z ) is the set of nef divisor classes x x x with ( x , x ) > 0 (x,x)>0 ( x , x ) > 0 .
Let L K 3 L_{K3} L K 3 be the even unimodular lattice of signature ( 3 , 19 ) (3,19) ( 3 , 19 ) . For a sublattice M M M of L K 3 L_{K3} L K 3 of signature ( 1 , ρ − 1 ) (1,\rho-1) ( 1 , ρ − 1 ) , an M M M -polarized K 3 K3 K 3 surface is a pair ( S , j ) (S,j) ( S , j ) consisting of a K 3 K3 K 3 surface S S S and a primitive embedding j : M ↪ NS ( S ) j:M\hookrightarrow\operatorname{NS}(S) j : M ↪ NS ( S ) . Let A A A be the orthogonal complement of M M M in L K 3 L_{K3} L K 3 . Then A A A has signature ( 2 , 20 − ρ ) (2,20-\rho) ( 2 , 20 − ρ ) . The space
D A ′ : = { [ ξ ] ∈ P ( A ⊗ C ) ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ‾ ) > 0 } D'_{A}:=\{[\xi]\in\mathbb{P}(A\otimes\mathbb{C})\mid(\xi,\xi)=0,\;(\xi,\overline{\xi})>0\} D A ′ := {[ ξ ] ∈ P ( A ⊗ C ) ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ) > 0 }
has two connected components. Let D A \mathcal{D}_A D A be one of them. Then D A \mathcal{D}_A D A is the Hermitian symmetric domain of type IV of dimension 20 − ρ 20-\rho 20 − ρ .
The lattices used in this paper. For an integer n n n and a lattice Λ \Lambda Λ with intersection matrix ( c i , j ) (c_{i,j}) ( c i , j ) , let Λ ( n ) \Lambda(n) Λ ( n ) be the lattice determined by the intersection matrix ( n c i , j ) (nc_{i,j}) ( n c i , j ) . Let U U U be the unimodular hyperbolic lattice of rank two. Let A j A_j A j and E j E_j E j stand for the root lattices. The polarization lattices are
M 1 = U ⊕ E 7 ( − 1 ) ⊕ E 7 ( − 1 ) . (8) M_1=U\oplus E_7(-1)\oplus E_7(-1). \tag*{(8)} M 1 = U ⊕ E 7 ( − 1 ) ⊕ E 7 ( − 1 ) . (8)
M 2 = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) . (9) M_2=U\oplus E_8(-1)\oplus E_7(-1). \tag*{(9)} M 2 = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) . (9)
Their orthogonal complements and the lattice for six line configurations are
A 1 = U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) , (10) A_1=U\oplus U\oplus A_1(-1)\oplus A_1(-1), \tag*{(10)} A 1 = U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) , (10)
A 2 = U ⊕ U ⊕ A 1 ( − 1 ) , A_2=U\oplus U\oplus A_1(-1), A 2 = U ⊕ U ⊕ A 1 ( − 1 ) ,
B 1 = U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) . (11) B_1=U(2)\oplus U(2)\oplus A_1(-1)\oplus A_1(-1). \tag*{(11)} B 1 = U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) . (11)
For j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } , the lattice A j A_j A j is the orthogonal complement of M j M_j M j in L K 3 L_{K3} L K 3 . According to [36 Theorem 1.14.4] or [30 Theorem 2.8] , M j M_j M j has a unique primitive embedding into L K 3 L_{K3} L K 3 up to isometry. Throughout, quotients involving both A 1 A_1 A 1 and B 1 B_1 B 1 use the domain identification and transported group actions fixed in § 2.2 .
Modular forms. The orthogonal group O ( A ) \operatorname{O}(A) O ( A ) acts on D A ′ \mathcal{D}'_A D A ′ . Let O + ( A ) \operatorname{O}^{+}(A) O + ( A ) be the subgroup of O ( A ) \operatorname{O}(A) O ( A ) which preserves the connected component D A \mathcal{D}_A D A . Put
Γ A : = ker ( O + ( A ) → Aut ( A ∨ / A ) ) . (12) \Gamma_A:=\ker\left(\operatorname{O}^{+}(A)\to\operatorname{Aut}(A^\vee/A)\right). \tag*{(12)} Γ A := ker ( O + ( A ) → Aut ( A ∨ / A ) ) . (12)
where A ∨ : = { x ∈ A ⊗ Q ∣ ( x , A ) ⊂ Z } A^\vee:=\{x\in A\otimes\mathbb{Q}\mid(x,A)\subset\mathbb{Z}\} A ∨ := { x ∈ A ⊗ Q ∣ ( x , A ) ⊂ Z } . The finite abelian group A ∨ / A A^\vee/A A ∨ / A is called the discriminant group of the lattice A A A .
Also, let D ~ A \widetilde{\mathcal{D}}_A D A be the connected component of the affine cone
{ ξ ∈ A ⊗ C ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ‾ ) > 0 } (13) \{\xi\in A\otimes\mathbb{C}\mid(\xi,\xi)=0,\;(\xi,\overline{\xi})>0\} \tag*{(13)} { ξ ∈ A ⊗ C ∣ ( ξ , ξ ) = 0 , ( ξ , ξ ) > 0 } (13)
which projects to D A \mathcal{D}_A D A . We have the principal C × \mathbb{C}^{\times} C × -bundle D ~ A → D A \widetilde{\mathcal{D}}_A\to\mathcal{D}_A D A → D A .
Definition 2.6. Let Γ \Gamma Γ be an arithmetic subgroup of O + ( A ) \operatorname{O}^{+}(A) O + ( A ) and assume dim D A ≥ 3 \dim\mathcal{D}_A\geq3 dim D A ≥ 3 , as for all lattices used below. If a holomorphic function f : D ~ A → C f:\widetilde{\mathcal{D}}_A\to\mathbb{C} f : D A → C satisfies
(i) f ( λ ξ ) = λ − k f ( ξ ) f(\lambda\xi)=\lambda^{-k}f(\xi) f ( λ ξ ) = λ − k f ( ξ ) for any λ ∈ C × \lambda\in\mathbb{C}^{\times} λ ∈ C × ,
(ii) f ( γ ξ ) = χ ( γ ) f ( ξ ) f(\gamma\xi)=\chi(\gamma)f(\xi) f ( γ ξ ) = χ ( γ ) f ( ξ ) for any γ ∈ Γ \gamma\in\Gamma γ ∈ Γ ,
then f f f is called a modular form on D A \mathcal{D}_A D A with respect to Γ \Gamma Γ of weight k ∈ Z k\in\mathbb{Z} k ∈ Z and character χ ∈ Char ( Γ ) : = Hom ( Γ , C × ) \chi\in\operatorname{Char}(\Gamma):=\operatorname{Hom}(\Gamma,\mathbb{C}^{\times}) χ ∈ Char ( Γ ) := Hom ( Γ , C × ) . Note that the holomorphy at the cusps follows from the Koecher principle.
Let A k ( Γ , χ ) A_k(\Gamma,\chi) A k ( Γ , χ ) be the vector space of such modular forms. The graded ring of modular forms is given by
A ( Γ ) : = ⨁ k ∈ Z χ ∈ Char ( Γ ) A k ( Γ , χ ) . A(\Gamma):=\bigoplus_{\substack{k\in\mathbb{Z}\\ \chi\in\operatorname{Char}(\Gamma)}}A_k(\Gamma,\chi). A ( Γ ) := k ∈ Z χ ∈ Char ( Γ ) ⨁ A k ( Γ , χ ) .
The subring of modular forms with trivial character is
A ( Γ , id ) : = ⨁ k ∈ Z A k ( Γ , id ) . A(\Gamma,\operatorname{id}):=\bigoplus_{k\in\mathbb{Z}}A_k(\Gamma,\operatorname{id}). A ( Γ , id ) := k ∈ Z ⨁ A k ( Γ , id ) .
On the quotient stack [ D A / Γ ] [\mathcal{D}_A/\Gamma] [ D A /Γ ] , let L \mathcal{L} L be the Hodge line bundle induced by the tautological bundle O P ( A ⊗ C ) ( − 1 ) ∣ D A \mathcal{O}_{\mathbb{P}(A\otimes\mathbb{C})(-1)}|_{\mathcal{D}_A} O P ( A ⊗ C ) ( − 1 ) ∣ D A . For χ ∈ Char ( Γ ) \chi\in\operatorname{Char}(\Gamma) χ ∈ Char ( Γ ) , let L χ \mathcal{L}_\chi L χ be the line bundle induced by the trivial line on D A \mathcal{D}_A D A with Γ \Gamma Γ acting on its fiber through χ \chi χ . With the homogeneity convention above, modular forms of weight k k k and character χ \chi χ are sections of $\mathcal{L}^{\otimes k}\otimes\mathcal{L}_\chi.
Let Γ ⊂ O + ( A ) \Gamma\subset O^{+}(A) Γ ⊂ O + ( A ) be arithmetic, and let H \mathcal{H} H be a Γ \Gamma Γ -invariant collection of linear hyperplanes in A ⊗ C A \otimes\mathbb{C} A ⊗ C , locally finite on D A \mathcal{D}_{A} D A . For a hyperplane K K K , put D K = D A ∩ P ( K ) D_{K} = D_{A} \cap\mathbb{P}(K) D K = D A ∩ P ( K ) . In particular, D δ = D δ ⊥ D_{\delta} = D_{\delta^{\perp}} D δ = D δ ⊥ for a negative vector δ ∈ A \delta\in A δ ∈ A . The support of the arrangement is
H = ⋃ K ∈ H D K . H = \bigcup_{K \in\mathcal{H}} D_{K}. H = K ∈ H ⋃ D K .
Put
D A , H ∘ = D A ∖ H , X A , H ∘ = D A , H ∘ / Γ . (14) D^{\circ}_{A,\mathcal{H}} = D_{A} \setminus H,\qquad X^{\circ}_{A,\mathcal{H}} = D^{\circ}_{A,\mathcal{H}}/\Gamma. \tag*{(14)} D A , H ∘ = D A ∖ H , X A , H ∘ = D A , H ∘ /Γ. (14)
Let D ~ A , H ∘ \widetilde{D}^{\circ}_{A,\mathcal{H}} D A , H ∘ be the corresponding affine cone.
Definition 2.7. A meromorphic modular form of weight k ∈ Z k \in\mathbb{Z} k ∈ Z and character χ ∈ Char ( Γ ) \chi\in\operatorname{Char}(\Gamma) χ ∈ Char ( Γ ) with possible poles along H H H is a meromorphic function on D ~ A \widetilde{D}_{A} D A satisfying the functional equations of Definition 2.6, whose restriction to D ~ A , H ∘ \widetilde{D}^{\circ}_{A,\mathcal{H}} D A , H ∘ is holomorphic.
Let A k ∘ ( Γ , χ ) H A^{\circ}_{k}(\Gamma,\chi)_{\mathcal{H}} A k ∘ ( Γ , χ ) H be the space of these forms. Put
A ∘ ( Γ ) H = ⨁ k ∈ Z ⨁ χ ∈ Char ( Γ ) A k ∘ ( Γ , χ ) H . (15) A^{\circ}(\Gamma)_{\mathcal{H}} = \bigoplus_{k \in\mathbb{Z}}\bigoplus_{\chi\in\operatorname{Char}(\Gamma)} A^{\circ}_{k}(\Gamma,\chi)_{\mathcal{H}}. \tag*{(15)} A ∘ ( Γ ) H = k ∈ Z ⨁ χ ∈ Char ( Γ ) ⨁ A k ∘ ( Γ , χ ) H . (15)
Also put
A ∘ ( Γ , id ) H = ⨁ k ∈ Z A k ∘ ( Γ , id ) H . A^{\circ}(\Gamma,\operatorname{id})_{\mathcal{H}} = \bigoplus_{k \in\mathbb{Z}} A^{\circ}_{k}(\Gamma,\operatorname{id})_{\mathcal{H}}. A ∘ ( Γ , id ) H = k ∈ Z ⨁ A k ∘ ( Γ , id ) H .
For the arrangements considered below, the Looijenga condition and the corresponding Koecher principle imply that there is no negative weight modular form [WW24, Section 2]. For H ∞ H_{\infty} H ∞ , the required condition is verified in Lemma 3.12.
Hyperplane arrangements for A 1 A_{1} A 1 . Choose a basis e 1 , e 1 ′ , e 2 , e 2 ′ , α 1 , α 2 e_{1},e'_{1},e_{2},e'_{2},\alpha_{1},\alpha_{2} e 1 , e 1 ′ , e 2 , e 2 ′ , α 1 , α 2 compatible with
A 1 = U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) , A_{1} = U \oplus U \oplus A_{1}(-1) \oplus A_{1}(-1), A 1 = U ⊕ U ⊕ A 1 ( − 1 ) ⊕ A 1 ( − 1 ) ,
where α 1 2 = α 2 2 = − 2 \alpha_{1}^{2} = \alpha_{2}^{2} = -2 α 1 2 = α 2 2 = − 2 and ( α 1 , α 2 ) = 0 (\alpha_{1},\alpha_{2}) = 0 ( α 1 , α 2 ) = 0 . Write
ξ = ξ 1 e 1 + ξ 2 e 1 ′ + ξ 3 e 2 + ξ 4 e 2 ′ + ξ 5 α 1 + ξ 6 α 2 . \xi= \xi_{1}e_{1}+\xi_{2}e'_{1}+\xi_{3}e_{2}+\xi_{4}e'_{2}+\xi_{5}\alpha_{1}+\xi_{6}\alpha_{2}. ξ = ξ 1 e 1 + ξ 2 e 1 ′ + ξ 3 e 2 + ξ 4 e 2 ′ + ξ 5 α 1 + ξ 6 α 2 .
Define the collections of linear hyperplanes
H ∞ = { ( γ α 2 ) C ⊥ ∣ γ ∈ Γ A 1 } , H 0 = { ( γ α 1 ) C ⊥ ∣ γ ∈ Γ A 1 } . (16) \mathcal{H}_{\infty} = \{(\gamma\alpha_{2})_{\mathbb{C}}^{\perp} \mid\gamma\in\Gamma_{A_{1}}\},\qquad
\mathcal{H}_{0} = \{(\gamma\alpha_{1})_{\mathbb{C}}^{\perp} \mid\gamma\in\Gamma_{A_{1}}\}. \tag*{(16)} H ∞ = {( γ α 2 ) C ⊥ ∣ γ ∈ Γ A 1 } , H 0 = {( γ α 1 ) C ⊥ ∣ γ ∈ Γ A 1 } . (16)
Their representatives are { ξ 6 = 0 } \{\xi_{6}=0\} { ξ 6 = 0 } and { ξ 5 = 0 } \{\xi_{5}=0\} { ξ 5 = 0 } , respectively. Their supports in the period domain are
H ∞ = ⋃ γ ∈ Γ A 1 D γ α 2 , H 0 = ⋃ γ ∈ Γ A 1 D γ α 1 . H_{\infty} = \bigcup_{\gamma\in\Gamma_{A_{1}}}D_{\gamma\alpha_{2}},\qquad
H_{0} = \bigcup_{\gamma\in\Gamma_{A_{1}}}D_{\gamma\alpha_{1}}. H ∞ = γ ∈ Γ A 1 ⋃ D γ α 2 , H 0 = γ ∈ Γ A 1 ⋃ D γ α 1 .
The images of H ∞ H_{\infty} H ∞ and H 0 H_{0} H 0 in the quotient are described in Remark 3.10.
Ball quotients. Let ω \omega ω be a primitive third root of unity and E : = Z [ ω ] E := \mathbb{Z}[\omega] E := Z [ ω ] be the ring of Eisenstein integers. Let ( Λ , h ) (\Lambda,h) ( Λ , h ) be the Eisenstein Hermitian lattice of signature ( 1 , 4 ) (1,4) ( 1 , 4 ) used in the complex hyperbolic uniformization of cubic surfaces in [2 ] . Put Λ C : = Λ ⊗ E C \Lambda_{\mathbb{C}} := \Lambda\otimes_{E} \mathbb{C} Λ C := Λ ⊗ E C . The associated complex ball of dimension four is
B 4 : = { [ z ] ∈ P ( Λ C ) ∣ h ( z , z ) > 0 } . \mathbb{B}^{4} := \{[z] \in\mathbb{P}(\Lambda_{\mathbb{C}}) \mid h(z,z) > 0\}. B 4 := {[ z ] ∈ P ( Λ C ) ∣ h ( z , z ) > 0 } .
Let Γ u m : = U ( Λ , h ) \Gamma_{\mathrm{um}} := U(\Lambda,h) Γ um := U ( Λ , h ) be the arithmetic group whose quotient parametrizes unmarked stable cubic surfaces. The finite quotient acting on markings is the Weyl group W ( E 6 ) W(E_{6}) W ( E 6 ) . Let ρ : Γ u m → W ( E 6 ) \rho:\Gamma_{\mathrm{um}} \to W(E_{6}) ρ : Γ um → W ( E 6 ) be the corresponding quotient homomorphism. The marked subgroup is Γ m : = ker ( ρ ) \Gamma_{\mathrm{m}} := \ker(\rho) Γ m := ker ( ρ ) . Thus Γ u m / Γ m ≅ W ( E 6 ) \Gamma_{\mathrm{um}}/\Gamma_{\mathrm{m}} \cong W(E_{6}) Γ um / Γ m ≅ W ( E 6 ) . Let S 6 ⊂ W ( E 6 ) S_{6} \subset W(E_{6}) S 6 ⊂ W ( E 6 ) be the standard subgroup induced by permutations of the six exceptional curves in the blow-up. Put Γ S 6 : = ρ − 1 ( S 6 ) \Gamma_{S_{6}} := \rho^{-1}(S_{6}) Γ S 6 := ρ − 1 ( S 6 ) . Then Γ m ⊂ Γ S 6 ⊂ Γ u m \Gamma_{\mathrm{m}} \subset\Gamma_{S_{6}} \subset\Gamma_{\mathrm{um}} Γ m ⊂ Γ S 6 ⊂ Γ um and Γ S 6 / Γ m ≅ S 6 \Gamma_{S_{6}}/\Gamma_{\mathrm{m}} \cong S_{6} Γ S 6 / Γ m ≅ S 6 .
Normalize h h h so that the roots in the Eisenstein lattice Λ \Lambda Λ have norm − 1 -1 − 1 . For r ∈ Λ r \in\Lambda r ∈ Λ with h ( r , r ) = − 1 h(r,r)=-1 h ( r , r ) = − 1 , write B r : = { [ z ] ∈ B 4 ∣ h ( z , r ) = 0 } \mathrm{B}_r:=\{[z]\in\mathbb{B}^4\mid h(z,r)=0\} B r := {[ z ] ∈ B 4 ∣ h ( z , r ) = 0 } . For Γ ∈ { Γ u m , Γ m , Γ S 6 } \Gamma\in\{\Gamma_{\mathrm{um}},\Gamma_{\mathrm{m}},\Gamma_{\mathrm{S}_6}\} Γ ∈ { Γ um , Γ m , Γ S 6 } , the Heegner divisor on the quotient is
H ( Γ ) : = Im ( ⋃ r ∈ Λ h ( r , r ) = − 1 B r ⟶ B 4 / Γ ) r e d . (17) \mathcal{H}(\Gamma):=\operatorname{Im}\left(\bigcup_{\substack{r\in\Lambda\\h(r,r)=-1}}\mathrm{B}_r\longrightarrow\mathbb{B}^4/\Gamma\right)_{\mathrm{red}}. \tag*{(17)} H ( Γ ) := Im r ∈ Λ h ( r , r ) = − 1 ⋃ B r ⟶ B 4 /Γ red . (17)
Periods of lattice-polarized K3 surfaces
Family F 1 \mathcal{F}_1 F 1 of M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surfaces
3.1. Family F 1 \mathcal{F}_1 F 1 of M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surfaces. In this section, we will study a family of elliptic surfaces introduced in [33 Section 2] . A member of the family is given by the Weierstrass form
S ( u ) : z 2 = y 3 + ( u 5 , 3 x 5 w 3 + u 4 , 4 x 4 w 4 + u 3 , 5 x 3 w 5 ) y + ( u 7 , 5 x 7 w 5 + u 6 , 6 x 6 w 6 + u 5 , 7 x 5 w 7 ) , (18) S(u): z^2=y^3+(u_{5,3}x^5w^3+u_{4,4}x^4w^4+u_{3,5}x^3w^5)y+(u_{7,5}x^7w^5+u_{6,6}x^6w^6+u_{5,7}x^5w^7), \tag*{(18)} S ( u ) : z 2 = y 3 + ( u 5 , 3 x 5 w 3 + u 4 , 4 x 4 w 4 + u 3 , 5 x 3 w 5 ) y + ( u 7 , 5 x 7 w 5 + u 6 , 6 x 6 w 6 + u 5 , 7 x 5 w 7 ) , (18)
where
u = ( u 5 , 3 , u 4 , 4 , u 3 , 5 , u 7 , 5 , u 6 , 6 , u 5 , 7 ) u=(u_{5,3},u_{4,4},u_{3,5},u_{7,5},u_{6,6},u_{5,7}) u = ( u 5 , 3 , u 4 , 4 , u 3 , 5 , u 7 , 5 , u 6 , 6 , u 5 , 7 )
∈ U ∗ = Spec ( C [ u 5 , 3 , u 4 , 4 , u 3 , 5 , u 7 , 5 , u 6 , 6 , u 5 , 7 ] ) ∖ { u 3 , 5 = u 5 , 7 = 0 or u 5 , 3 = u 7 , 5 = 0 } . (19) \in U^*=\operatorname{Spec}\bigl(\mathbb{C}[u_{5,3},u_{4,4},u_{3,5},u_{7,5},u_{6,6},u_{5,7}]\bigr)\setminus\{u_{3,5}=u_{5,7}=0\ \text{or}\ u_{5,3}=u_{7,5}=0\}. \tag*{(19)} ∈ U ∗ = Spec ( C [ u 5 , 3 , u 4 , 4 , u 3 , 5 , u 7 , 5 , u 6 , 6 , u 5 , 7 ] ) ∖ { u 3 , 5 = u 5 , 7 = 0 or u 5 , 3 = u 7 , 5 = 0 } . (19)
When no confusion can arise, let S ( u ) S(u) S ( u ) also stand for the minimal resolution of the elliptic surface given by (18) . Let F 1 \mathcal{F}_1 F 1 be the family ϖ : F 1 ∋ S ( u ) ↦ u ∈ U ∗ \varpi:\mathcal{F}_1\ni S(u)\mapsto u\in U^* ϖ : F 1 ∋ S ( u ) ↦ u ∈ U ∗ of K 3 K3 K 3 surfaces. The above form (18) admits an action of ( κ , μ ) ∈ C κ × × C μ × (\kappa,\mu)\in\mathbb{C}_{\kappa}^{\times}\times\mathbb{C}_{\mu}^{\times} ( κ , μ ) ∈ C κ × × C μ × given by
κ : ( x , y , z , w ; u i , j ) ↦ ( x , κ 2 y , κ 3 z , w ; κ i + j 2 u i , j ) (20) \kappa:(x,y,z,w;u_{i,j})\mapsto(x,\kappa^2y,\kappa^3z,w;\kappa^{\frac{i+j}{2}}u_{i,j}) \tag*{(20)} κ : ( x , y , z , w ; u i , j ) ↦ ( x , κ 2 y , κ 3 z , w ; κ 2 i + j u i , j ) (20)
and
μ : ( x , y , z , w ; u i , j ) ↦ ( μ − 1 x , y , z , μ w ; μ i − j u i , j ) . (21) \mu:(x,y,z,w;u_{i,j})\mapsto(\mu^{-1}x,y,z,\mu w;\mu^{i-j}u_{i,j}). \tag*{(21)} μ : ( x , y , z , w ; u i , j ) ↦ ( μ − 1 x , y , z , μ w ; μ i − j u i , j ) . (21)
For u = ( u i , j ) i , j u=(u_{i,j})_{i,j} u = ( u i , j ) i , j , put κ ⋅ u = ( κ i + j 2 u i , j ) i , j \kappa\cdot u=(\kappa^{\frac{i+j}{2}}u_{i,j})_{i,j} κ ⋅ u = ( κ 2 i + j u i , j ) i , j and μ ⋅ u = ( μ i − j u i , j ) i , j \mu\cdot u=(\mu^{i-j}u_{i,j})_{i,j} μ ⋅ u = ( μ i − j u i , j ) i , j . There exists a unique holomorphic 2 2 2 -form η u \eta_u η u on S ( u ) S(u) S ( u ) up to a constant factor. The above action of κ \kappa κ and μ \mu μ induces the relation
η κ ⋅ u = κ − 1 η u , η μ ⋅ u = η u . (22) \eta_{\kappa\cdot u}=\kappa^{-1}\eta_u,\qquad\eta_{\mu\cdot u}=\eta_u. \tag*{(22)} η κ ⋅ u = κ − 1 η u , η μ ⋅ u = η u . (22)
In the affine coordinates ( x 0 , y 0 , z 0 ) = ( x w 4 , y w 6 , z w 9 ) (x_0,y_0,z_0)=\left(\frac{x}{w^4},\frac{y}{w^6},\frac{z}{w^9}\right) ( x 0 , y 0 , z 0 ) = ( w 4 x , w 6 y , w 9 z ) , the form η u \eta_u η u is given by d x 0 ∧ d y 0 z 0 \frac{dx_0\wedge dy_0}{z_0} z 0 d x 0 ∧ d y 0 .
Let π : S ( u ) → P 1 ( C ) = { ( w : x ) } \pi:S(u)\to\mathbb{P}^1(\mathbb{C})=\{(w:x)\} π : S ( u ) → P 1 ( C ) = {( w : x )} be the elliptic fibration given by the Weierstrass form (18) . The singular fibers of π \pi π for a very general member S ( u ) S(u) S ( u ) of F 1 \mathcal{F}_1 F 1 are of Kodaira type I I I ∗ + I I I ∗ + 6 I 1 III^*+III^*+6I_1 I I I ∗ + I I I ∗ + 6 I 1 , where π − 1 ( 0 : 1 ) \pi^{-1}(0:1) π − 1 ( 0 : 1 ) and π − 1 ( 1 : 0 ) \pi^{-1}(1:0) π − 1 ( 1 : 0 ) are of type I I I ∗ III^* I I I ∗ . Let F F F be a general fiber of π \pi π . Note that this gives a nef divisor. Let O O O be the identity element of the Mordell–Weil group of sections of π \pi π , and let ( O ) (O) ( O ) be the corresponding divisor. These sections and singular fibers for the elliptic fibration π \pi π are illustrated in Figure 1.
The divisors F , ( O ) , a 1 , … , a 7 , b 1 , … , b 7 F,(O),a_1,\ldots,a_7,b_1,\ldots,b_7 F , ( O ) , a 1 , … , a 7 , b 1 , … , b 7 in Figure 1 generate NS ( S ( u ) ) \operatorname{NS}(S(u)) NS ( S ( u )) of a very general member of F 1 \mathcal{F}_1 F 1 . Therefore, we have the following proposition.
Figure 1. Elliptic fibration π \pi π
Proposition 3.1. The Néron–Severi lattice NS ( S ( u ) ) \operatorname{NS}(S(u)) NS ( S ( u )) of a very general member S ( u ) S(u) S ( u ) of F 1 \mathcal{F}_1 F 1 is given by M 1 \mathrm{M}_1 M 1 of (8) . Its transcendental lattice is given by A 1 \mathrm{A}_1 A 1 of (10) .
Setting w = 1 w=1 w = 1 in (18) , we have the form z 2 = y 3 + g 2 ( x ) y + g 3 ( x ) z^2=y^3+g_2(x)y+g_3(x) z 2 = y 3 + g 2 ( x ) y + g 3 ( x ) . Let ℓ ( u ) \ell(u) ℓ ( u ) be a polynomial in u i , j u_{i,j} u i , j determined by the resultant of g 2 ( x ) x 3 \frac{g_2(x)}{x^3} x 3 g 2 ( x ) and g 3 ( x ) x 5 \frac{g_3(x)}{x^5} x 5 g 3 ( x ) . Then, letting R ( x ) R(x) R ( x ) be the discriminant of y 3 + g 2 ( x ) y + g 3 ( x ) y^3+g_2(x)y+g_3(x) y 3 + g 2 ( x ) y + g 3 ( x ) in y y y , the discriminant of 1 x 9 R ( x ) \frac{1}{x^9}R(x) x 9 1 R ( x ) in x x x is given by ℓ ( u ) 3 δ 60 ( u ) \ell(u)^3\delta_{60}(u) ℓ ( u ) 3 δ 60 ( u ) , up to a nonzero constant, where δ 60 ( u ) \delta_{60}(u) δ 60 ( u ) is a polynomial in u = ( u i , j ) u=(u_{i,j}) u = ( u i , j ) .
We have degenerate elliptic surfaces as follows.
Lemma 3.3. (1) Suppose u u u is a general point satisfying ℓ ( u ) = 0 \ell(u)=0 ℓ ( u ) = 0 . Then, a new singular fiber for π \pi π of Kodaira type II appears. Although this fiber is cuspidal, the Weierstrass surface is smooth at its cusp and acquires no additional rational double point there.
(2) Suppose u u u is a general point satisfying δ 60 ( u ) = 0 \delta_{60}(u)=0 δ 60 ( u ) = 0 . Then, two of the singular fibers of Kodaira type I1 _1 1 collapse into a singular fiber of type I2 _2 2 .
(3) If u u u satisfies u 5 , 3 = 0 u_{5,3}=0 u 5 , 3 = 0 (resp. u 3 , 5 = 0 u_{3,5}=0 u 3 , 5 = 0 ), π − 1 ( 0 : 1 ) \pi^{-1}(0:1) π − 1 ( 0 : 1 ) (resp. π − 1 ( 1 : 0 ) \pi^{-1}(1:0) π − 1 ( 1 : 0 ) ) of type III∗ ^* ∗ turns into a singular fiber of type II∗ ^* ∗ .
Proof. We can prove the lemma by an argument similar to [33 Section 3] or [35 Section 1.2] . □
Let us define pseudo-ample M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surfaces in the sense of [9 ] . Let S 0 : = S ( u ) S_0:=S(u) S 0 := S ( u ) be a very general member of the family F 1 \mathcal{F}_1 F 1 . Then, we have an isometry ψ 0 : H 2 ( S 0 , Z ) ≅ L K 3 \psi_0:H_2(S_0,\mathbb{Z})\cong L_{K3} ψ 0 : H 2 ( S 0 , Z ) ≅ L K 3 with ψ 0 − 1 ( M 1 ) = NS ( S 0 ) \psi_0^{-1}(\mathbf{M}_1)=\operatorname{NS}(S_0) ψ 0 − 1 ( M 1 ) = NS ( S 0 ) . Such a surface S 0 S_0 S 0 is called a reference surface. We use the positive cone V ( S 0 ) + V(S_0)^+ V ( S 0 ) + defined in Notation and preliminaries. Let Δ ( M 1 ) : = { δ ∈ M 1 ∣ ( δ , δ ) = − 2 } \Delta(\mathbf{M}_1):=\{\delta\in\mathbf{M}_1\mid(\delta,\delta)=-2\} Δ ( M 1 ) := { δ ∈ M 1 ∣ ( δ , δ ) = − 2 } . Set Δ ( M 1 ) + : = { δ ∈ Δ ( M 1 ) ∣ ψ 0 − 1 ( δ ) gives an effective class } \Delta(\mathbf{M}_1)^+:=\{\delta\in\Delta(\mathbf{M}_1)\mid\psi_0^{-1}(\delta)\text{ gives an effective class}\} Δ ( M 1 ) + := { δ ∈ Δ ( M 1 ) ∣ ψ 0 − 1 ( δ ) gives an effective class } . The set V ( M 1 ) = { y ∈ M 1 ⊗ Z R ∣ ( y , y ) > 0 } V(\mathbf{M}_1)=\{y\in\mathbf{M}_1\otimes_{\mathbb{Z}}\mathbb{R}\mid(y,y)>0\} V ( M 1 ) = { y ∈ M 1 ⊗ Z R ∣ ( y , y ) > 0 } has two connected components. Let V ( M 1 ) + V(\mathbf{M}_1)^+ V ( M 1 ) + be one of them containing ψ 0 ( x ) \psi_0(x) ψ 0 ( x ) for x ∈ V ( S 0 ) + ∩ NS ( S 0 ) R x\in V(S_0)^+\cap\operatorname{NS}(S_0)_{\mathbb{R}} x ∈ V ( S 0 ) + ∩ NS ( S 0 ) R . Moreover, put C ( M 1 ) + : = { y ∈ V ( M 1 ) + ∣ ( y , δ ) > 0 for all δ ∈ Δ ( M 1 ) + } C(\mathbf{M}_1)^+:=\{y\in V(\mathbf{M}_1)^+\mid(y,\delta)>0\text{ for all }\delta\in\Delta(\mathbf{M}_1)^+\} C ( M 1 ) + := { y ∈ V ( M 1 ) + ∣ ( y , δ ) > 0 for all δ ∈ Δ ( M 1 ) + } .
Definition 3.4. (1) A pseudo-ample M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surface ( S , j ) (S,j) ( S , j ) is a pair of a K 3 K3 K 3 surface S S S and a primitive embedding j : M 1 ↪ NS ( S ) j:\mathbf{M}_1\hookrightarrow\operatorname{NS}(S) j : M 1 ↪ NS ( S ) such that j ( C ( M 1 ) + ) ∩ NS ( S ) + ≠ ∅ j(C(\mathbf{M}_1)^+)\cap\operatorname{NS}(S)^+\ne\varnothing j ( C ( M 1 ) + ) ∩ NS ( S ) + = ∅ .
(2) Let ( S 1 , j 1 ) (S_1,j_1) ( S 1 , j 1 ) and ( S 2 , j 2 ) (S_2,j_2) ( S 2 , j 2 ) be two pseudo-ample M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surfaces. If there exists a biholomorphic mapping f : S 1 → S 2 f:S_1\to S_2 f : S 1 → S 2 satisfying j 2 = f ∗ ∘ j 1 j_2=f_*\circ j_1 j 2 = f ∗ ∘ j 1 , they are said to be isomorphic.
(3) Let S S S be a K 3 K3 K 3 surface. Suppose an isometry ψ : H 2 ( S , Z ) ≅ L K 3 \psi:H_2(S,\mathbb{Z})\cong L_{K3} ψ : H 2 ( S , Z ) ≅ L K 3 induces a pseudo-ample M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surface ( S , ψ − 1 ∣ M 1 ) (S,\psi^{-1}|_{\mathbf{M}_1}) ( S , ψ − 1 ∣ M 1 ) . Then, ( S , ψ ) (S,\psi) ( S , ψ ) is called a pseudo-ample marked M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surface.
(4) Let ( S 1 , ψ 1 ) (S_1,\psi_1) ( S 1 , ψ 1 ) and ( S 2 , ψ 2 ) (S_2,\psi_2) ( S 2 , ψ 2 ) be two pseudo-ample marked M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surfaces. If there exists a biholomorphic mapping f : S 1 → S 2 f:S_1\to S_2 f : S 1 → S 2 satisfying ψ 1 = ψ 2 ∘ f ∗ \psi_1=\psi_2\circ f_* ψ 1 = ψ 2 ∘ f ∗ , they are said to be isomorphic.
Let M M 1 p a \mathcal{M}_{\mathbf{M}_1}^{\mathrm{pa}} M M 1 pa be the set of isomorphism classes of pseudo-ample marked M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surfaces. Set Γ ( M 1 ) : = { γ ∈ O ( L K 3 ) ∣ γ ( m ) = m for all m ∈ M 1 } \Gamma(\mathbf{M}_1):=\{\gamma\in O(L_{K3})\mid\gamma(m)=m\text{ for all }m\in\mathbf{M}_1\} Γ ( M 1 ) := { γ ∈ O ( L K 3 ) ∣ γ ( m ) = m for all m ∈ M 1 } . By virtue of the Torelli theorem and the surjectivity of the period mapping for K 3 K3 K 3 surfaces, we have the bijection
Φ 1 : M M 1 p a / Γ ( M 1 ) ≅ D A 1 / Γ A 1 (23) \Phi_1:\mathcal{M}_{\mathbf{M}_1}^{\mathrm{pa}}/\Gamma(\mathbf{M}_1)\cong\mathcal{D}_{A_1}/\Gamma_{A_1} \tag*{(23)} Φ 1 : M M 1 pa /Γ ( M 1 ) ≅ D A 1 / Γ A 1 (23)
(see [9 Section 3] ). Therefore, the arithmetic quotient D A 1 / Γ A 1 \mathcal{D}_{A_1}/\Gamma_{A_1} D A 1 / Γ A 1 of the symmetric domain of dimension four can be identified with the set of isomorphism classes of pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surfaces via the period mapping.
Proposition 3.5. Every isomorphism class of pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surfaces is represented by an elliptic K 3 K3 K 3 surface defined by the Weierstrass form (18) .
Proof. The pair ( S , ψ ) (S,\psi) ( S , ψ ) is a pseudo-ample marked M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surface if and only if it satisfies the following conditions:
(i) ψ − 1 ( M 1 ) ⊂ N S ( S ) \psi^{-1}(\mathrm{M}_1) \subset\mathrm{NS}(S) ψ − 1 ( M 1 ) ⊂ NS ( S ) ,
(ii) ψ − 1 ∘ ψ 0 ( F ) \psi^{-1}\circ\psi_0(F) ψ − 1 ∘ ψ 0 ( F ) , ψ − 1 ∘ ψ 0 ( ( O ) ) \psi^{-1}\circ\psi_0((O)) ψ − 1 ∘ ψ 0 (( O )) , ψ − 1 ∘ ψ 0 ( a j ) \psi^{-1}\circ\psi_0(a_j) ψ − 1 ∘ ψ 0 ( a j ) , ψ − 1 ∘ ψ 0 ( b k ) \psi^{-1}\circ\psi_0(b_k) ψ − 1 ∘ ψ 0 ( b k ) are effective divisors on S S S for F , ( O ) , a j , b k F,(O),a_j,b_k F , ( O ) , a j , b k (j , k ∈ { 1 , … , 7 } j,k \in\{1,\ldots,7\} j , k ∈ { 1 , … , 7 } ) in Figure 1 ,
(iii) ψ − 1 ∘ ψ 0 ( F ) \psi^{-1}\circ\psi_0(F) ψ − 1 ∘ ψ 0 ( F ) is a nef divisor on S S S .
Such a pair is realized by an elliptic surface of the form (18) , as in the argument in [35 Section 2.2] . □
Proposition 3.5 enables us to study the period mapping Φ 1 \Phi_1 Φ 1 of (23) by our family F 1 \mathcal{F}_1 F 1 of (18) .
Proposition 3.6 ([33 Theorem 2.1] ). Set
v 4 = u 4 , 4 , v 6 = u 6 , 6 , v 8 = u 5 , 3 u 3 , 5 , v 10 = u 5 , 3 u 5 , 7 + u 3 , 5 u 7 , 5 , v 12 = u 7 , 5 u 5 , 7 , v ~ 10 = u 5 , 3 u 5 , 7 − u 3 , 5 u 7 , 5 . (24) \begin{aligned}
v_4 &= u_{4,4}, \quad v_6 = u_{6,6}, \quad v_8 = u_{5,3}u_{3,5},\\
v_{10} &= u_{5,3}u_{5,7}+u_{3,5}u_{7,5}, \quad v_{12}=u_{7,5}u_{5,7}, \quad\widetilde{v}_{10}=u_{5,3}u_{5,7}-u_{3,5}u_{7,5}. \tag*{(24)}
\end{aligned} v 4 v 10 = u 4 , 4 , v 6 = u 6 , 6 , v 8 = u 5 , 3 u 3 , 5 , = u 5 , 3 u 5 , 7 + u 3 , 5 u 7 , 5 , v 12 = u 7 , 5 u 5 , 7 , v 10 = u 5 , 3 u 5 , 7 − u 3 , 5 u 7 , 5 . (24) The ring A ( Γ A 1 , i d ) \mathcal{A}(\Gamma_{A_1},\mathrm{id}) A ( Γ A 1 , id ) of modular forms on D A 1 \mathcal{D}_{A_1} D A 1 with respect to Γ A 1 \Gamma_{A_1} Γ A 1 of character i d \mathrm{id} id is given by
C [ v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 10 ] / ( v ~ 10 2 − v 10 2 + 4 v 8 v 12 ) . \mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_{10}]/(\widetilde{v}_{10}^{2}-v_{10}^{2}+4v_8v_{12}). C [ v 4 , v 6 , v 8 , v 10 , v 12 , v 10 ] / ( v 10 2 − v 10 2 + 4 v 8 v 12 ) .
Remark 3.7. There exists an involution σ ∈ O + ( A 1 ) \sigma\in O^{+}(A_1) σ ∈ O + ( A 1 ) which interchanges two A 1 ( − 1 ) A_1(-1) A 1 ( − 1 ) summands and fixes U ⊕ U U\oplus U U ⊕ U in A 1 A_1 A 1 . The involution σ \sigma σ is not an element of Γ A 1 \Gamma_{A_1} Γ A 1 . Let σ ˉ \bar{\sigma} σ ˉ be its image in O + ( A 1 ) / Γ A 1 O^{+}(A_1)/\Gamma_{A_1} O + ( A 1 ) / Γ A 1 . Then
O + ( A 1 ) / Γ A 1 = ⟨ σ ˉ ⟩ ≅ Z / 2 Z . O^{+}(A_1)/\Gamma_{A_1}=\langle\bar{\sigma}\rangle\cong\mathbb{Z}/2\mathbb{Z}. O + ( A 1 ) / Γ A 1 = ⟨ σ ˉ ⟩ ≅ Z /2 Z .
We note that σ \sigma σ induces an element of O ( L K 3 ) O(L_{K3}) O ( L K 3 ) whose restriction to M 1 \mathrm{M}_1 M 1 interchanges two E 7 ( − 1 ) E_7(-1) E 7 ( − 1 ) summands in M 1 \mathrm{M}_1 M 1 . It comes from the involution indicated in Remark 3.2 .
Remark 3.8. Since the involution σ ∈ O + ( A 1 ) \sigma\in O^{+}(A_1) σ ∈ O + ( A 1 ) induces the transformation ( w : x ) ↦ ( x : w ) (w:x)\mapsto(x:w) ( w : x ) ↦ ( x : w ) on (18) , it induces the involution u i , j ↦ u j , i u_{i,j}\mapsto u_{j,i} u i , j ↦ u j , i for the parameters of the family F 1 \mathcal{F}_1 F 1 .
M2-polarized subfamily F2
We will make use of a subfamily F 2 \mathcal{F}_2 F 2 of F 1 \mathcal{F}_1 F 1 . In this subsection, we study the affine models of the members of F 1 \mathcal{F}_1 F 1 by setting w = 1 w=1 w = 1 in (18) . Let us consider the subfamily ϖ − 1 ( { u 5 , 3 = 0 } ) \varpi^{-1}(\{u_{5,3}=0\}) ϖ − 1 ({ u 5 , 3 = 0 }) of F 1 \mathcal{F}_1 F 1 . On this subfamily, the definition of U ∗ U^{*} U ∗ implies u 7 , 5 ≠ 0 u_{7,5}\ne0 u 7 , 5 = 0 . By putting u 5 , 3 = 0 u_{5,3}=0 u 5 , 3 = 0 and ( x , y , z ) ↦ ( x u 7 , 5 , y u 7 , 5 2 , z u 7 , 5 3 ) (x,y,z)\mapsto\left(\frac{x}{u_{7,5}},\frac{y}{u_{7,5}^{2}},\frac{z}{u_{7,5}^{3}}\right) ( x , y , z ) ↦ ( u 7 , 5 x , u 7 , 5 2 y , u 7 , 5 3 z ) , we have
S ‾ ( t ‾ ) : z ‾ 2 = y ‾ 3 + ( t ‾ 4 x 4 + t ‾ 10 x 3 ) y + ( x 7 + t ‾ 6 x 6 + t ‾ 12 x 5 ) . (25) \overline{S}(\overline{t}):\ \overline{z}^{2}=\overline{y}^{3}+(\overline{t}_{4}x^{4}+\overline{t}_{10}x^{3})y+(x^{7}+\overline{t}_{6}x^{6}+\overline{t}_{12}x^{5}). \tag*{(25)} S ( t ) : z 2 = y 3 + ( t 4 x 4 + t 10 x 3 ) y + ( x 7 + t 6 x 6 + t 12 x 5 ) . (25)
Here,
( t ‾ 4 , t ‾ 6 , t ‾ 10 , t ‾ 12 ) ∈ T ∗ = Spec ( C [ t ‾ 4 , t ‾ 6 , t ‾ 10 , t ‾ 12 ] ) ∖ { t ‾ 10 = t ‾ 12 = 0 } (\overline{t}_{4},\overline{t}_{6},\overline{t}_{10},\overline{t}_{12})\in T^{*}=\operatorname{Spec}(\mathbb{C}[\overline{t}_{4},\overline{t}_{6},\overline{t}_{10},\overline{t}_{12}])\setminus\{\overline{t}_{10}=\overline{t}_{12}=0\} ( t 4 , t 6 , t 10 , t 12 ) ∈ T ∗ = Spec ( C [ t 4 , t 6 , t 10 , t 12 ]) ∖ { t 10 = t 12 = 0 }
is given by
t ‾ 4 = u 4 , 4 , t ‾ 6 = u 6 , 6 , t ‾ 10 = u 3 , 5 u 7 , 5 , t ‾ 12 = u 5 , 7 u 7 , 5 . \overline{t}_{4}=u_{4,4},\qquad\overline{t}_{6}=u_{6,6},\qquad\overline{t}_{10}=u_{3,5}u_{7,5},\qquad\overline{t}_{12}=u_{5,7}u_{7,5}. t 4 = u 4 , 4 , t 6 = u 6 , 6 , t 10 = u 3 , 5 u 7 , 5 , t 12 = u 5 , 7 u 7 , 5 .
We have the family F 2 \mathcal{F}_2 F 2 given by S ‾ ( t ‾ ) ↦ t ‾ \overline{S}(\overline{t})\mapsto\overline{t} S ( t ) ↦ t .
Proposition 3.9 ([6 ] , see also [35 Proposition 2.1] ). *The Néron-Severi lattice of a very general member of F 2 \mathcal{F}_2 F 2 is given by M 2 \mathrm{M}_2 M 2 of (9) .
From the singular fibers for the elliptic fibration studied in Lemma 3.3 (3), a very general member of F 2 \mathcal{F}_2 F 2 gives a reference surface for pseudo-ample M 2 \mathrm{M}_2 M 2 -polarized K 3 K3 K 3 surfaces as in the case of M 1 \mathrm{M}_1 M 1 . Define M M 2 p a \mathcal{M}^{pa}_{\mathrm{M}_2} M M 2 p a as the set of isomorphism classes of pseudo-ample marked M 2 \mathrm{M}_2 M 2 -polarized K 3 K3 K 3 surfaces, and define Γ ( M 2 ) \Gamma(\mathrm{M}_2) Γ ( M 2 ) analogously. Also, we have the period mapping
Φ 2 : M M 2 p a / Γ ( M 2 ) ≅ D A 2 / Γ A 2 . \Phi_2:\mathcal{M}^{pa}_{\mathrm{M}_2}/\Gamma(\mathrm{M}_2)\cong\mathcal{D}_{A_2}/\Gamma_{A_2}. Φ 2 : M M 2 p a /Γ ( M 2 ) ≅ D A 2 / Γ A 2 .
This period mapping is closely related to Siegel modular forms. The period domain D A 2 \mathcal{D}_{A_2} D A 2 is biholomorphic to the Siegel upper half-space of genus two. We can regard the parameters t 4 , t 6 , t 10 , t 12 t_4,t_6,t_{10},t_{12} t 4 , t 6 , t 10 , t 12 of S ‾ ( t ) \overline{S}(t) S ( t ) given by (25) as well-known Siegel modular forms.
Remark 3.10. By considering the involution in Remark 3.2, the subfamily ϖ − 1 ( { u 3 , 5 = 0 } ) \varpi^{-1}(\{u_{3,5}=0\}) ϖ − 1 ({ u 3 , 5 = 0 }) also gives a family of M 2 \mathrm{M}_2 M 2 -polarized K 3 K3 K 3 surfaces. However, as in Remark 3.7, the subfamily ϖ − 1 ( { u 5 , 3 = 0 } ) \varpi^{-1}(\{u_{5,3}=0\}) ϖ − 1 ({ u 5 , 3 = 0 }) is not identified with ϖ − 1 ( { u 3 , 5 = 0 } ) \varpi^{-1}(\{u_{3,5}=0\}) ϖ − 1 ({ u 3 , 5 = 0 }) under the action of Γ A 1 \Gamma_{A_1} Γ A 1 . Namely, they are different subfamilies of degenerate pseudo-ample M 2 \mathrm{M}_2 M 2 -polarized K 3 K3 K 3 surfaces.
We use the arrangements H ∞ , H 0 \mathcal{H}_{\infty},\mathcal{H}_0 H ∞ , H 0 and their supports H ∞ , H 0 H_{\infty},H_0 H ∞ , H 0 . The distinction between the two supports corresponds to the two subfamilies discussed in Remark 3.10.
Proposition 3.11. The set of isomorphism classes of pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K 3 K3 K 3 surfaces which do not degenerate to pseudo-ample M 2 \mathrm{M}_2 M 2 -polarized K 3 K3 K 3 surfaces is given by
( D A 1 ∖ ( H ∞ ∪ H 0 ) ) / Γ A 1 (\mathcal{D}_{A_1}\setminus(H_{\infty}\cup H_0))/\Gamma_{A_1} ( D A 1 ∖ ( H ∞ ∪ H 0 )) / Γ A 1 via the period mapping Φ 1 \Phi_1 Φ 1 of (23) .
Proof. By Lemma 3.3(3) and Remark 3.10, the subfamilies ϖ − 1 ( { u 5 , 3 = 0 } ) \varpi^{-1}(\{u_{5,3}=0\}) ϖ − 1 ({ u 5 , 3 = 0 }) and ϖ − 1 ( { u 3 , 5 = 0 } ) \varpi^{-1}(\{u_{3,5}=0\}) ϖ − 1 ({ u 3 , 5 = 0 }) correspond under Φ 1 \Phi_1 Φ 1 to the images of H ∞ H_{\infty} H ∞ and H 0 H_0 H 0 , respectively, in D A 1 / Γ A 1 \mathcal{D}_{A_1}/\Gamma_{A_1} D A 1 / Γ A 1 . More precisely, the irreducible components of a singular fiber of Kodaira type I I ∗ II^* I I ∗ induce the E 8 ( − 1 ) E_8(-1) E 8 ( − 1 ) summand of the lattice M 2 \mathrm{M}_2 M 2 of (9) . The former (resp. latter) E 7 ( − 1 ) E_7(-1) E 7 ( − 1 ) summand of M 1 \mathrm{M}_1 M 1 of (8) turns into E 8 ( − 1 ) E_8(-1) E 8 ( − 1 ) if u 5 , 3 = 0 u_{5,3}=0 u 5 , 3 = 0 (resp. u 3 , 5 = 0 u_{3,5}=0 u 3 , 5 = 0 ).
We will study the meromorphic modular forms associated with the arrangement divisor H ∞ H_{\infty} H ∞ in the sense of Definition 2.7 via the period mapping of F 1 \mathcal{F}_1 F 1 .
Lemma 3.12. Every one-dimensional intersection of members of H ∞ \mathcal{H}_{\infty} H ∞ in A 1 ⊗ C \mathrm{A}_1\otimes\mathbb{C} A 1 ⊗ C is generated over C \mathbb{C} C by a vector v ∈ A 1 ⊗ Q v\in\mathrm{A}_1\otimes\mathbb{Q} v ∈ A 1 ⊗ Q with ( v , v ) < 0 (v,v)<0 ( v , v ) < 0 .
Proof. Apply [WW24, Lemma 5.3] to the positive-definite lattice L = A 1 ⊕ A 1 L=\mathrm{A}_1\oplus\mathrm{A}_1 L = A 1 ⊕ A 1 , so that A 1 = U ⊕ U ⊕ L ( − 1 ) \mathrm{A}_1=U\oplus U\oplus L(-1) A 1 = U ⊕ U ⊕ L ( − 1 ) . In their notation, H L , 0 H_{L,0} H L , 0 is the support H ∞ H_{\infty} H ∞ of our arrangement H ∞ \mathcal{H}_{\infty} H ∞ , and H L , 1 = ∅ H_{L,1}=\varnothing H L , 1 = ∅ . Their Looijenga condition, translated to our sign convention, gives the assertion.
Lemma 3.12 verifies the hypothesis of [25 Corollary 7.5] : no one-dimensional intersection of members of H ∞ \mathcal{H}_{\infty} H ∞ is positive semidefinite. Hence the arithmetic quotient X A 1 , H ∞ ∘ X^\circ_{A_1,H_{\infty}} X A 1 , H ∞ ∘ of (14) has a Looijenga compactification X ^ A 1 , H ∞ \widehat{X}_{A_1,H_{\infty}} X A 1 , H ∞ , which is the projective spectrum of the graded ring A ∘ ( Γ A 1 , i d ) H ∞ A^\circ(\Gamma_{A_1},\mathrm{id})_{H_{\infty}} A ∘ ( Γ A 1 , id ) H ∞ . The period mapping identifies this compactification with R ^ \widehat{\mathcal{R}} R in (32) .
According to the proof of Proposition 3.11, via the period mapping Φ 1 \Phi_1 Φ 1 , every point of X A 1 , H ∞ ∘ X^\circ_{A_1,H_{\infty}} X A 1 , H ∞ ∘ corresponds to a K 3 K3 K 3 surface S ( u ) S(u) S ( u ) of (18) with u 5 , 3 ≠ 0 u_{5,3}\ne0 u 5 , 3 = 0 . From now on, let us study the affine equation of S ( u ) S(u) S ( u ) obtained by putting w = 1 w=1 w = 1 . If u 5 , 3 ≠ 0 u_{5,3}\ne0 u 5 , 3 = 0 , performing the transformation
( x , y , z ) ⟼ ( x u 5 , 3 , y u 5 , 3 2 , z u 5 , 3 3 ) (26) (x,y,z)\longmapsto\left(\frac{x}{u_{5,3}},\frac{y}{u_{5,3}^{2}},\frac{z}{u_{5,3}^{3}}\right) \tag*{(26)} ( x , y , z ) ⟼ ( u 5 , 3 x , u 5 , 3 2 y , u 5 , 3 3 z ) (26)
we have the equation
S p ( r ) : z 2 = y 3 + ( x 5 + r 4 x 4 + r 8 x 3 ) y + ( r 2 x 7 + r 6 x 6 + r 10 x 5 ) . (27) S_p(r):z^2=y^3+(x^5+r_4x^4+r_8x^3)y+(r_2x^7+r_6x^6+r_{10}x^5). \tag*{(27)} S p ( r ) : z 2 = y 3 + ( x 5 + r 4 x 4 + r 8 x 3 ) y + ( r 2 x 7 + r 6 x 6 + r 10 x 5 ) . (27)
where
r 2 = u 7 , 5 u 5 , 3 , r 4 = u 4 , 4 , r 6 = u 6 , 6 , r 8 = u 5 , 3 u 3 , 5 , r 10 = u 5 , 3 u 5 , 7 . (28) r_2=\frac{u_{7,5}}{u_{5,3}},\qquad r_4=u_{4,4},\qquad r_6=u_{6,6},\qquad r_8=u_{5,3}u_{3,5},\qquad r_{10}=u_{5,3}u_{5,7}. \tag*{(28)} r 2 = u 5 , 3 u 7 , 5 , r 4 = u 4 , 4 , r 6 = u 6 , 6 , r 8 = u 5 , 3 u 3 , 5 , r 10 = u 5 , 3 u 5 , 7 . (28)
For λ ∈ C × \lambda\in\mathbb{C}^{\times} λ ∈ C × , set
λ ⋅ r = ( λ 2 r 2 , λ 4 r 4 , λ 6 r 6 , λ 8 r 8 , λ 10 r 10 ) . (29) \lambda\cdot r=(\lambda^{2}r_2,\lambda^{4}r_4,\lambda^{6}r_6,\lambda^{8}r_8,\lambda^{10}r_{10}). \tag*{(29)} λ ⋅ r = ( λ 2 r 2 , λ 4 r 4 , λ 6 r 6 , λ 8 r 8 , λ 10 r 10 ) . (29)
The above action of λ \lambda λ is induced by the action of (20) and (21) on the surface (18) . Namely, it comes from the action of ( λ , λ − 1 ) ∈ C κ × × C μ × (\lambda,\lambda^{-1})\in\mathbb{C}^{\times}_{\kappa}\times\mathbb{C}^{\times}_{\mu} ( λ , λ − 1 ) ∈ C κ × × C μ × . From (22) , a holomorphic 2-form η r \eta_r η r on the K3 surface S p ( r ) S_p(r) S p ( r ) satisfies
η λ ⋅ r = λ − 1 η r (30) \eta_{\lambda\cdot r}=\lambda^{-1}\eta_r \tag*{(30)} η λ ⋅ r = λ − 1 η r (30)
for the λ \lambda λ -action of (29) .
Recalling (19) , we regard a tuple r = ( r 2 , r 4 , r 6 , r 8 , r 10 ) r=(r_2,r_4,r_6,r_8,r_{10}) r = ( r 2 , r 4 , r 6 , r 8 , r 10 ) of (28) as a point of
R ∗ : = Spec ( C [ r 2 , r 4 , r 6 , r 8 , r 10 ] ) ∖ { r 8 = r 10 = 0 } . \mathcal{R}^{*}:=\operatorname{Spec}\bigl(\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}]\bigr)\setminus\{r_8=r_{10}=0\}. R ∗ := Spec ( C [ r 2 , r 4 , r 6 , r 8 , r 10 ] ) ∖ { r 8 = r 10 = 0 } .
Put R ^ : = P ( 2 , 4 , 6 , 8 , 10 ) \widehat{\mathcal{R}}:=\mathbb{P}(2,4,6,8,10) R := P ( 2 , 4 , 6 , 8 , 10 ) and define
R = { [ r ] ∈ R ^ ∣ r ∈ R ∗ } . \mathcal{R}=\{[r]\in\widehat{\mathcal{R}}\mid r\in\mathcal{R}^{*}\}. R = {[ r ] ∈ R ∣ r ∈ R ∗ } .
Then R ^ ∖ R \widehat{\mathcal{R}}\setminus\mathcal{R} R ∖ R is the image of the locus { r 8 = r 10 = 0 } \{r_8=r_{10}=0\} { r 8 = r 10 = 0 } , and hence has codimension two in R ^ \widehat{\mathcal{R}} R . We naturally obtain a family { S p ( [ r ] ) ∣ [ r ] ∈ R } → R \{S_p([r])\mid[r]\in\mathcal{R}\}\to\mathcal{R} { S p ([ r ]) ∣ [ r ] ∈ R } → R . Due to the proof of Proposition 3.11, a member S p ( [ r ] ) S_p([r]) S p ([ r ]) is a pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K3 surface corresponding to a point of X A 1 , H ∞ ∘ X^{\circ}_{A_1,H_\infty} X A 1 , H ∞ ∘ of (14) . Recalling the period mapping Φ 1 \Phi_1 Φ 1 of (23) , set P 1 ∘ = Φ 1 − 1 ( X A 1 , H ∞ ∘ ) \mathcal{P}^{\circ}_1=\Phi^{-1}_1(X^{\circ}_{A_1,H_\infty}) P 1 ∘ = Φ 1 − 1 ( X A 1 , H ∞ ∘ ) , that is a subset of the moduli space of pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K3 surfaces.
Lemma 3.13. Suppose that S p ( [ r ] ) S_p([r]) S p ([ r ]) for [ r ] ∈ R [r]\in\mathcal{R} [ r ] ∈ R is isomorphic to S p ( [ r ′ ] ) S_p([r']) S p ([ r ′ ]) for [ r ′ ] ∈ R [r']\in\mathcal{R} [ r ′ ] ∈ R as pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K3 surfaces. Then [ r ] = [ r ′ ] [r]=[r'] [ r ] = [ r ′ ] .
Proof. Let π [ r ] : S p ( [ r ] ) → P 1 ( C ) \pi_{[r]}:S_p([r])\to\mathbb{P}^{1}(\mathbb{C}) π [ r ] : S p ([ r ]) → P 1 ( C ) be the elliptic fibration given by the Weierstrass form (27) , with base coordinate x x x . If S p ( [ r ] ) S_p([r]) S p ([ r ]) is isomorphic to S p ( [ r ′ ] ) S_p([r']) S p ([ r ′ ]) as pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K3 surfaces, then the types of the singular fibers for π [ r ] \pi_{[r]} π [ r ] must coincide with those of π [ r ′ ] \pi_{[r'] } π [ r ′ ] . Then, we can see that [ r ] = [ r ′ ] [r]=[r'] [ r ] = [ r ′ ] by observing the elliptic fibers as in the proof of [35 Lemma 1.1] . □
Theorem 3.14. The ring A ∘ ( Γ A 1 , i d ) H ∞ A^{\circ}(\Gamma_{A_1,\mathrm{id}})_{H_\infty} A ∘ ( Γ A 1 , id ) H ∞ of meromorphic modular forms of character id in the sense of Definition 2.7 is isomorphic to the polynomial ring C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] C [ r 2 , r 4 , r 6 , r 8 , r 10 ] via the period mapping Φ 1 \Phi_1 Φ 1 . Here, r k r_k r k gives a modular form ξ ↦ r k ( ξ ) \xi\mapsto r_k(\xi) ξ ↦ r k ( ξ ) of weight k k k .
Proof. By Lemma 3.13, together with Proposition 3.5, Remark 3.7 and Remark 3.10, we have an explicit realization
Φ 1 ∘ : R → ≅ X A 1 , H ∞ ∘ (31) \Phi^{\circ}_1:\mathcal{R}\xrightarrow{\ \cong\ }X^{\circ}_{A_1,H_\infty} \tag*{(31)} Φ 1 ∘ : R ≅ X A 1 , H ∞ ∘ (31)
given by the restriction of the period mapping Φ 1 \Phi_1 Φ 1 of (23) to P 1 ∘ \mathcal{P}^{\circ}_1 P 1 ∘ .
We remark that Φ 1 ∘ \Phi^{\circ}_1 Φ 1 ∘ is given by [ r ] ↦ [ ξ ] [r]\mapsto[\xi] [ r ] ↦ [ ξ ] , where ξ \xi ξ is a ratio of six integrals of the 2-form η r \eta_r η r on appropriate 2-cycles on S p ( r ) S_p(r) S p ( r ) . Recalling the action of λ ∈ C × \lambda\in\mathbb{C}^{\times} λ ∈ C × of (29) and (30) , Φ 1 ∘ \Phi^{\circ}_1 Φ 1 ∘ induces a correspondence R ∗ → D A 1 , H ∞ ∘ / Γ A 1 \mathcal{R}^{*}\to\mathcal{D}^{\circ}_{A_1,H_\infty}/\Gamma_{A_1} R ∗ → D A 1 , H ∞ ∘ / Γ A 1 given by r ↦ ξ r\mapsto\xi r ↦ ξ which sends λ − 1 ⋅ r \lambda^{-1}\cdot r λ − 1 ⋅ r to λ ξ \lambda\xi λ ξ as in the following diagram:
r → ξ ↓ ↓ λ − 1 ⋅ r → λ ξ \begin{CD}
r @>>> \xi\\
@VVV @VVV \\
\lambda^{-1}\cdot r @>>> \lambda\xi
\end{CD} r ↓ ⏐ λ − 1 ⋅ r ξ ↓ ⏐ λ ξ
Hence, the inverse of Φ 1 ∘ \Phi_1^\circ Φ 1 ∘ induces a tuple of the functions ξ ↦ r k ( ξ ) \xi\mapsto r_k(\xi) ξ ↦ r k ( ξ ) (k ∈ { 2 , 4 , 6 , 8 , 10 } k \in\{2,4,6,8,10\} k ∈ { 2 , 4 , 6 , 8 , 10 } ) satisfying r k ( λ ξ ) = λ − k r k ( ξ ) r_k(\lambda\xi)=\lambda^{-k}r_k(\xi) r k ( λ ξ ) = λ − k r k ( ξ ) for λ ∈ C × \lambda\in\mathbb{C}^{\times} λ ∈ C × and r k ( γ ξ ) = r k ( ξ ) r_k(\gamma\xi)=r_k(\xi) r k ( γ ξ ) = r k ( ξ ) for γ ∈ Γ A 1 \gamma\in\Gamma_{A_1} γ ∈ Γ A 1 .
Let X ∞ = [ D A 1 , H ∞ ∘ / Γ A 1 ] \mathcal{X}_\infty=[D_{A_1,H_\infty}^\circ/\Gamma_{A_1}] X ∞ = [ D A 1 , H ∞ ∘ / Γ A 1 ] and p ∞ : X ∞ → X A 1 , H ∞ ∘ p_\infty:\mathcal{X}_\infty\to X_{A_1,H_\infty}^\circ p ∞ : X ∞ → X A 1 , H ∞ ∘ be the quotient stack and its coarse moduli morphism. Using the Hodge line bundle L \mathcal{L} L fixed above, put A k = ( p ∞ ) ∗ L ⊗ k \mathcal{A}_k=(p_\infty)_*\mathcal{L}^{\otimes k} A k = ( p ∞ ) ∗ L ⊗ k . Since − id -\operatorname{id} − id acts on the fibers of L ⊗ k \mathcal{L}^{\otimes k} L ⊗ k by ( − 1 ) k (-1)^k ( − 1 ) k , the sheaf A k \mathcal{A}_k A k is zero for odd k k k . For even k k k , it is a reflexive sheaf of rank one on the quotient. A meromorphic modular form of weight k k k and trivial character in the sense of Definition 2.7 gives a section of A k \mathcal{A}_k A k on X A 1 , H ∞ ∘ X_{A_1,H_\infty}^\circ X A 1 , H ∞ ∘ . On R ^ = Proj C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \widehat{\mathcal{R}}=\operatorname{Proj}\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] R = Proj C [ r 2 , r 4 , r 6 , r 8 , r 10 ] , let O R ^ ( k ) \mathcal{O}_{\widehat{\mathcal{R}}}(k) O R ( k ) be the sheaf associated with the degree shift by k k k of this graded coordinate ring, where deg r i = i \deg r_i=i deg r i = i . This sheaf is zero for odd k k k and reflexive of rank one for even k k k . It need not be invertible. The equivariant period correspondence above identifies ( Φ 1 ∘ ) ∗ A k (\Phi_1^\circ)_*\mathcal{A}_k ( Φ 1 ∘ ) ∗ A k with O R ^ ( k ) ∣ R \mathcal{O}_{\widehat{\mathcal{R}}}(k)|_{\mathcal{R}} O R ( k ) ∣ R . Since R ^ \widehat{\mathcal{R}} R is normal and R ^ ∖ R \widehat{\mathcal{R}}\setminus\mathcal{R} R ∖ R has codimension two, restriction gives H 0 ( R ^ , O R ^ ( k ) ) ≅ H 0 ( R , O R ^ ( k ) ∣ R ) H^0(\widehat{\mathcal{R}},\mathcal{O}_{\widehat{\mathcal{R}}}(k))\cong H^0(\mathcal{R},\mathcal{O}_{\widehat{\mathcal{R}}}(k)|_{\mathcal{R}}) H 0 ( R , O R ( k )) ≅ H 0 ( R , O R ( k ) ∣ R ) . Therefore, the period mapping Φ 1 ∘ \Phi_1^\circ Φ 1 ∘ gives
⨁ k ∈ Z ≥ 0 H 0 ( X A 1 , H ∞ ∘ , A k ) ≅ ⨁ k ∈ Z ≥ 0 H 0 ( R ^ , O R ^ ( k ) ) = C [ r 2 , r 4 , r 6 , r 8 , r 10 ] . \bigoplus_{k\in\mathbb{Z}_{\geq0}} H^0(X_{A_1,H_\infty}^\circ,\mathcal{A}_k)\cong\bigoplus_{k\in\mathbb{Z}_{\geq0}} H^0(\widehat{\mathcal{R}},\mathcal{O}_{\widehat{\mathcal{R}}}(k))=\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}]. k ∈ Z ≥ 0 ⨁ H 0 ( X A 1 , H ∞ ∘ , A k ) ≅ k ∈ Z ≥ 0 ⨁ H 0 ( R , O R ( k )) = C [ r 2 , r 4 , r 6 , r 8 , r 10 ] .
Remark 3.15. The automorphic ring which appears in Theorem 3.14 is essentially the L 0 = A 1 L_0=A_1 L 0 = A 1 , L 1 = A 1 L_1=A_1 L 1 = A 1 case of [WW24, Theorems 1.2, 5.4 and Section 8.1]. More precisely, if one writes L = A 1 ⊕ A 1 L=A_1\oplus A_1 L = A 1 ⊕ A 1 with the convention that root lattices are positive definite, then, up to the sign convention for the quadratic form, A 1 ≅ 2 U ⊕ L ( − 1 ) A_1\cong2U\oplus L(-1) A 1 ≅ 2 U ⊕ L ( − 1 ) . The group Γ A 1 \Gamma_{A_1} Γ A 1 is the stable orthogonal subgroup O ~ + ( 2 U ⊕ L ( − 1 ) ) \widetilde{\mathrm{O}}^+(2U\oplus L(-1)) O + ( 2 U ⊕ L ( − 1 )) and the arrangement H ∞ H_\infty H ∞ is the arrangement H L , 0 H_{L,0} H L , 0 associated with the summand L 0 = A 1 L_0=A_1 L 0 = A 1 . In this case the second part H L , 1 H_{L,1} H L , 1 of their arrangement is empty. Their theorem therefore gives a free algebra of meromorphic modular forms with poles along H ∞ H_\infty H ∞ , with generator weights 2 , 4 , 6 , 8 , 10 2,4,6,8,10 2 , 4 , 6 , 8 , 10 , and hence gives the Looijenga compactification as the weighted projective space P ( 2 , 4 , 6 , 8 , 10 ) \mathbb{P}(2,4,6,8,10) P ( 2 , 4 , 6 , 8 , 10 ) . The point of Theorem 3.14 is the geometric identification of these abstract generators with the explicit period coordinates r 2 , r 4 , r 6 , r 8 , r 10 r_2,r_4,r_6,r_8,r_{10} r 2 , r 4 , r 6 , r 8 , r 10 coming from the Weierstrass family (27) .
By Lemma 3.12 , we can apply [25 Corollary 7.5] to the above result. Hence, we have the bijection
Φ ^ 1 ∘ : R ^ ≅ X ^ A 1 , H ∞ ∘ . (32) \widehat{\Phi}_1^\circ:\widehat{\mathcal{R}}\cong\widehat{X}_{A_1,H_\infty}^\circ. \tag*{(32)} Φ 1 ∘ : R ≅ X A 1 , H ∞ ∘ . (32)
This mapping is a natural extension of (31) .
Put R 0 = R − { r 8 = 0 } \mathcal{R}_0=\mathcal{R}-\{r_8=0\} R 0 = R − { r 8 = 0 } . From Proposition 3.11 and the procedure of (27) , the family { S p ( [ r ] ) ∣ [ r ] ∈ R 0 } \{\mathcal{S}_p([r])\mid[r]\in\mathcal{R}_0\} { S p ([ r ]) ∣ [ r ] ∈ R 0 } coincides with the family of the isomorphism classes of pseudo-ample M 1 \mathrm{M}_1 M 1 -polarized K3 surfaces which does not degenerate to pseudo-ample M 2 \mathrm{M}_2 M 2 -polarized K3 surfaces.
Characters and the orbifold Picard group
We retain the central subgroup { ± id } \{\pm\operatorname{id}\} { ± id } in Γ A 1 = [ O A 1 / O + ( A 1 ) ] \Gamma_{A_1}=[\mathrm{O}_{A_1}/\mathrm{O}^+(A_1)] Γ A 1 = [ O A 1 / O + ( A 1 )] and use its algebraic Picard group. The quotient by the full orthogonal group on both connected components used in [33 ] is equivalent to this quotient stack. Nagano and Ueda computed its Picard group and constructed the torsion generators from modular forms with characters. We recall their description below and give a direct argument using reflections for the character group, including the values of its generators on each reflection class. These values distinguish the three generators and relate them to the branch divisors in Theorem 3.7.
Proposition 3.16. The character group satisfies
Char ( O + ( A 1 ) ) ≅ O + ( A 1 ) a b ≅ ( Z / 2 Z ) ⊕ 3 \operatorname{Char}(\mathrm{O}^+(A_1))\cong\mathrm{O}^+(A_1)^{\mathrm{ab}}\cong(\mathbb{Z}/2\mathbb{Z})^{\oplus3} Char ( O + ( A 1 )) ≅ O + ( A 1 ) ab ≅ ( Z /2 Z ) ⊕ 3
and is generated by χ 4 \chi_4 χ 4 , χ 10 \chi_{10} χ 10 , and χ 30 \chi_{30} χ 30 . If s r s_r s r is the reflection in a primitive negative vector r ∈ A 1 r \in\mathbf{A}_1 r ∈ A 1 for which s r ∈ O + ( A 1 ) s_r \in\mathrm{O}^{+}(\mathbf{A}_1) s r ∈ O + ( A 1 ) , then its character values are
Table 3.16.
Proof. Using the generators α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 , we have A A 1 = A A 1 ∨ / A A 1 ≅ ( Z / 2 Z ) 2 A_{\mathbf{A}_1}=A_{\mathbf{A}_1}^{\vee}/A_{\mathbf{A}_1}\cong(\mathbb{Z}/2\mathbb{Z})^2 A A 1 = A A 1 ∨ / A A 1 ≅ ( Z /2 Z ) 2 is generated by the classes of α 1 / 2 \alpha_1/2 α 1 /2 and α 2 / 2 \alpha_2/2 α 2 /2 . Let q A 1 : A A 1 → Q / 2 Z q_{\mathbf{A}_1}: A_{\mathbf{A}_1}\to\mathbb{Q}/2\mathbb{Z} q A 1 : A A 1 → Q /2 Z be the discriminant quadratic form. It satisfies
q A 1 ( [ α 1 / 2 ] ) = q A 1 ( [ α 2 / 2 ] ) = − 1 2 , q A 1 ( [ ( α 1 + α 2 ) / 2 ] ) = − 1. q_{\mathbf{A}_1}([\alpha_1/2])=q_{\mathbf{A}_1}([\alpha_2/2])=-\frac{1}{2},\qquad q_{\mathbf{A}_1}([(\alpha_1+\alpha_2)/2])=-1. q A 1 ([ α 1 /2 ]) = q A 1 ([ α 2 /2 ]) = − 2 1 , q A 1 ([( α 1 + α 2 ) /2 ]) = − 1.
Hence O ( A A 1 ) ≅ Z / 2 Z \mathrm{O}(A_{\mathbf{A}_1})\cong\mathbb{Z}/2\mathbb{Z} O ( A A 1 ) ≅ Z /2 Z , generated by the transposition of α 1 / 2 \alpha_1/2 α 1 /2 and α 2 / 2 \alpha_2/2 α 2 /2 , which implies
O + ( A 1 ) = ⟨ Γ A 1 , σ ⟩ , O + ( A 1 ) / Γ A 1 ≅ Z / 2 Z . (33) \mathrm{O}^{+}(\mathbf{A}_1)=\langle\Gamma_{\mathbf{A}_1},\sigma\rangle,\qquad\mathrm{O}^{+}(\mathbf{A}_1)/\Gamma_{\mathbf{A}_1}\cong\mathbb{Z}/2\mathbb{Z}. \tag*{(33)} O + ( A 1 ) = ⟨ Γ A 1 , σ ⟩ , O + ( A 1 ) / Γ A 1 ≅ Z /2 Z . (33)
By [33 ] , the ring of modular forms with trivial character for O + ( A 1 ) \mathrm{O}^{+}(\mathbf{A}_1) O + ( A 1 ) is freely generated by v 4 , v 6 , v 8 , v 10 , v 12 v_4,v_6,v_8,v_{10},v_{12} v 4 , v 6 , v 8 , v 10 , v 12 . Applying [WW24, Theorem 2.6] with the empty arrangement shows that O + ( A 1 ) \mathrm{O}^{+}(\mathbf{A}_1) O + ( A 1 ) is generated by reflections.
Such reflections are defined by primitive negative vectors r ∈ A 1 r\in\mathbf{A}_1 r ∈ A 1 . The class of r / div A 1 ( r ) r/\operatorname{div}_{\mathbf{A}_1}(r) r / div A 1 ( r ) in A A 1 A_{\mathbf{A}_1} A A 1 has order div A 1 ( r ) \operatorname{div}_{\mathbf{A}_1}(r) div A 1 ( r ) . Since this discriminant group has exponent two, div A 1 ( r ) \operatorname{div}_{\mathbf{A}_1}(r) div A 1 ( r ) is either one or two. Integrality of the reflection implies r 2 ∣ 2 div A 1 ( r ) r^2\mid2\operatorname{div}_{\mathbf{A}_1}(r) r 2 ∣ 2 div A 1 ( r ) . Since the lattice is even, the possible pairs are
( r 2 , div A 1 ( r ) ) = ( − 2 , 1 ) , ( − 2 , 2 ) , ( − 4 , 2 ) . (r^2,\operatorname{div}_{\mathbf{A}_1}(r))=(-2,1),\qquad(-2,2),\qquad(-4,2). ( r 2 , div A 1 ( r )) = ( − 2 , 1 ) , ( − 2 , 2 ) , ( − 4 , 2 ) .
(See (3) for the definition of div A 1 \operatorname{div}_{\mathbf{A}_1} div A 1 .)
Eichler’s criterion [13 Proposition 3.3(i)] gives one Γ A 1 \Gamma_{\mathbf{A}_1} Γ A 1 -orbit for each of the first and third types. For the second type, the two discriminant classes are α 1 / 2 \alpha_1/2 α 1 /2 and α 2 / 2 \alpha_2/2 α 2 /2 . Their orbits are interchanged by σ \sigma σ . Together with (33) , this gives exactly three conjugacy classes of reflections in O + ( A 1 ) \mathrm{O}^{+}(\mathbf{A}_1) O + ( A 1 ) . The abelianization is consequently generated by three elements of order at most two. It is therefore a quotient of ( Z / 2 Z ) ⊕ 3 (\mathbb{Z}/2\mathbb{Z})^{\oplus3} ( Z /2 Z ) ⊕ 3 , and # Char ( O + ( A 1 ) ) ≤ 8 \#\operatorname{Char}(\mathrm{O}^{+}(\mathbf{A}_1))\leq8 # Char ( O + ( A 1 )) ≤ 8 .
By [33 Section 3] , the modular forms v ~ 4 , v ~ 10 \widetilde{v}_4,\widetilde{v}_{10} v 4 , v 10 , and v ~ 30 \widetilde{v}_{30} v 30 vanish simply on the reflection mirrors of types ( − 2 , 2 ) (-2,2) ( − 2 , 2 ) , ( − 4 , 2 ) (-4,2) ( − 4 , 2 ) , and ( − 2 , 1 ) (-2,1) ( − 2 , 1 ) , respectively, and do not vanish at a general point of a mirror of either of the other types. A reflection changes the sign of a local coordinate normal to its mirror. Since it fixes a lift of a point on the mirror, the transformation law of each form gives character value − 1 -1 − 1 in the case of simple vanishing and 1 1 1 in the case of nonvanishing. This proves (3.16). The three characters are therefore independent. Together with the upper bound, this gives
Char ( O + ( A 1 ) ) ≅ O + ( A 1 ) a b ≅ ( Z / 2 Z ) ⊕ 3 . (34) \operatorname{Char}(\mathrm{O}^{+}(\mathbf{A}_1))\cong\mathrm{O}^{+}(\mathbf{A}_1)^{\mathrm{ab}}\cong(\mathbb{Z}/2\mathbb{Z})^{\oplus3}. \tag*{(34)} Char ( O + ( A 1 )) ≅ O + ( A 1 ) ab ≅ ( Z /2 Z ) ⊕ 3 . (34)
To express the Picard group using the Hodge line bundle L \mathcal{L} L , let D k D_k D k be the zero divisor of v ~ k \widetilde{v}_k v k on X A 1 \mathcal{X}_{\mathbf{A}_1} X A 1 for k = 4 , 10 , 30 k=4,10,30 k = 4 , 10 , 30 . The description in [33 Section 3] gives L χ k ≅ O ( D k ) ⊗ L − k \mathcal{L}_{\chi_k}\cong\mathcal{O}(D_k)\otimes\mathcal{L}^{-k} L χ k ≅ O ( D k ) ⊗ L − k and
Pic ( X A 1 ) = Z [ L ] ⊕ ⨁ k ∈ { 4 , 10 , 30 } ( Z / 2 Z ) [ L χ k ] . \operatorname{Pic}(\mathcal{X}_{\mathbf{A}_1})=\mathbb{Z}[\mathcal{L}]\oplus\bigoplus_{k\in\{4,10,30\}}(\mathbb{Z}/2\mathbb{Z})[\mathcal{L}_{\chi_k}]. Pic ( X A 1 ) = Z [ L ] ⊕ k ∈ { 4 , 10 , 30 } ⨁ ( Z /2 Z ) [ L χ k ] .
In particular,
Pic ( X A 1 ) t o r s ≅ Char ( O + ( A 1 ) ) . (35) \operatorname{Pic}(\mathcal{X}_{\mathbf{A}_1})_{\mathrm{tors}}\cong\operatorname{Char}(\mathrm{O}^{+}(\mathbf{A}_1)). \tag*{(35)} Pic ( X A 1 ) tors ≅ Char ( O + ( A 1 )) . (35)
The generator [ L χ k ] [\mathcal{L}_{\chi_k}] [ L χ k ] determines the double covering of the quotient stack associated with ker ( χ k ) \ker(\chi_k) ker ( χ k ) . Passing to the underlying arithmetic quotients and taking Baily–Borel compactifications gives Y χ k → Y A 1 Y_{\chi_k}\to Y_{\mathbf{A}_1} Y χ k → Y A 1 as (7) . The geometric interpretation of these coverings is given in Theorem 2.4 and Section 5.3.
Let D D A 1 , H ∞ ∘ ∘ ⊂ D A 1 , H ∞ ∘ D^\circ_{D^\circ_{A_1,H_\infty}} \subset D^\circ_{A_1,H_\infty} D D A 1 , H ∞ ∘ ∘ ⊂ D A 1 , H ∞ ∘ be the countable union of all reflection hyperplanes derived from the action of Γ A 1 \Gamma_{A_1} Γ A 1 . Let D A 1 , H ∞ ∘ D^\circ_{A_1,H_\infty} D A 1 , H ∞ ∘ be the image of D D A 1 , H ∞ ∘ ∘ D^\circ_{D^\circ_{A_1,H_\infty}} D D A 1 , H ∞ ∘ ∘ by the projection D A 1 , H ∞ ∘ → D A 1 , H ∞ ∘ / Γ A 1 = X A 1 , H ∞ ∘ D^\circ_{A_1,H_\infty} \to D^\circ_{A_1,H_\infty}/\Gamma_{A_1}=X^\circ_{A_1,H_\infty} D A 1 , H ∞ ∘ → D A 1 , H ∞ ∘ / Γ A 1 = X A 1 , H ∞ ∘ . Let D R D_R D R be the reduced closure in R ^ \widehat{R} R of the inverse image of D X A 1 , H ∞ ∘ D^\circ_{X_{A_1,H_\infty}} D X A 1 , H ∞ ∘ under the period isomorphism Φ 1 ∘ \Phi^\circ_1 Φ 1 ∘ of (31) . Choose a square-free weighted homogeneous polynomial Δ R ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \Delta_R(r) \in\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] Δ R ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] defining D R D_R D R , so that
D R = { [ r ] ∈ R ^ ∣ Δ R ( r ) = 0 } . D_R=\{[r]\in\widehat{R}\mid\Delta_R(r)=0\}. D R = {[ r ] ∈ R ∣ Δ R ( r ) = 0 } .
Let S D A 1 , H ∞ ∘ ⊂ D A 1 , H ∞ ∘ S_{D^\circ_{A_1,H_\infty}}\subset D^\circ_{A_1,H_\infty} S D A 1 , H ∞ ∘ ⊂ D A 1 , H ∞ ∘ be the set of points whose stabilizer subgroup is neither { ± id A 1 } \{\pm\operatorname{id}_{A_1}\} { ± id A 1 } nor a group generated by − id A 1 -\operatorname{id}_{A_1} − id A 1 and s s s , where s ∈ Γ A 1 s\in\Gamma_{A_1} s ∈ Γ A 1 is a reflection. We remark that S D A 1 , H ∞ ∘ S_{D^\circ_{A_1,H_\infty}} S D A 1 , H ∞ ∘ is a countable union of analytic subsets of codimension at least two. The action of Γ A 1 / { ± id A 1 } \Gamma_{A_1}/\{\pm\operatorname{id}_{A_1}\} Γ A 1 / { ± id A 1 } on D A 1 , H ∞ ∘ ∖ ( D D A 1 , H ∞ ∘ ∘ ∪ S D A 1 , H ∞ ∘ ) D^\circ_{A_1,H_\infty}\setminus\left(D^\circ_{D^\circ_{A_1,H_\infty}}\cup S_{D^\circ_{A_1,H_\infty}}\right) D A 1 , H ∞ ∘ ∖ ( D D A 1 , H ∞ ∘ ∘ ∪ S D A 1 , H ∞ ∘ ) is free. Define S X A 1 , H ∞ ∘ ⊂ X A 1 , H ∞ ∘ S_{X^\circ_{A_1,H_\infty}}\subset X^\circ_{A_1,H_\infty} S X A 1 , H ∞ ∘ ⊂ X A 1 , H ∞ ∘ and S R ⊂ R ^ S_R\subset\widehat{R} S R ⊂ R analogously. Here, the weight d d d of Δ R ( r ) \Delta_R(r) Δ R ( r ) must be even, because [ r ] ∈ P ( 2 , 4 , 6 , 8 , 10 ) [r]\in\mathbb{P}(2,4,6,8,10) [ r ] ∈ P ( 2 , 4 , 6 , 8 , 10 ) . Since the sheaves considered below are reflexive, removing analytic subsets of codimension at least two does not affect the relevant spaces of sections. We will use this without further mention for subsets such as S R S_R S R .
Proposition 3.17. The weight of Δ R ( r ) \Delta_R(r) Δ R ( r ) is 68.
Proof. We compare canonical classes, as in [14 Proposition 5.2] , using the Hodge bundle L \mathcal{L} L on X ∞ X_\infty X ∞ . All comparisons may be made away from subsets of codimension at least two. For the coarse weighted projective space, the dualizing sheaf is
ω R ^ ≅ O R ^ ( − 30 ) . (36) \omega_{\widehat{R}}\cong\mathcal{O}_{\widehat{R}}(-30). \tag*{(36)} ω R ≅ O R ( − 30 ) . (36)
Here O R ^ ( − 30 ) \mathcal{O}_{\widehat{R}}(-30) O R ( − 30 ) has the interpretation in terms of graded sheaves given in the proof of Theorem 3.14 . On the other hand, D A 1 , H ∞ ∘ D^\circ_{A_1,H_\infty} D A 1 , H ∞ ∘ is open in a quadric hypersurface in P 5 \mathbb{P}^5 P 5 . The ordinary adjunction formula gives ω D A 1 , H ∞ ∘ ≅ O P 5 ( − 4 ) ∣ D A 1 , H ∞ ∘ \omega_{D^\circ_{A_1,H_\infty}}\cong\mathcal{O}_{\mathbb{P}^5}(-4)|_{D^\circ_{A_1,H_\infty}} ω D A 1 , H ∞ ∘ ≅ O P 5 ( − 4 ) ∣ D A 1 , H ∞ ∘ . Since the pullback of L \mathcal{L} L is the tautological O P 5 ( − 1 ) \mathcal{O}_{\mathbb{P}^5}(-1) O P 5 ( − 1 ) , the equivariant formula is
ω X ∞ ≅ L ⊗ 4 ⊗ L det , (37) \omega_{X_\infty}\cong\mathcal{L}^{\otimes4}\otimes\mathcal{L}_{\det}, \tag*{(37)} ω X ∞ ≅ L ⊗ 4 ⊗ L d e t , (37)
where det ( γ ) = det ( γ ∣ A 1 ) \det(\gamma)=\det(\gamma|_{A_1}) det ( γ ) = det ( γ ∣ A 1 ) and L det \mathcal{L}_{\det} L d e t is its twisted line bundle. In particular, L det ⊗ 2 ≅ O \mathcal{L}_{\det}^{\otimes2}\cong\mathcal{O} L d e t ⊗ 2 ≅ O . Let R r e f R_{\mathrm{ref}} R ref be the reduced reflection divisor on X ∞ X_\infty X ∞ . The generic effective stabilizer along each component has order two. Thus, using Φ 1 ∘ \Phi^\circ_1 Φ 1 ∘ to identify the quotient with R R R , one has p ∞ ∗ ( D R ∣ R ) = 2 R r e f p^*_\infty(D_R|_R)=2R_{\mathrm{ref}} p ∞ ∗ ( D R ∣ R ) = 2 R ref in codimension one. In codimension one, even powers of L \mathcal{L} L descend to the quotient and agree with the graded twists under the period correspondence. The equation of weight d d d defining D R D_R D R and the graded period identification give p ∞ ∗ [ D R ∣ R ] = d [ L ] p^*_\infty[D_R|_R]=d[\mathcal{L}] p ∞ ∗ [ D R ∣ R ] = d [ L ] . The ramification formula therefore yields
[ ω X ∞ ] = p ∞ ∗ [ ω R ] + [ R r e f ] = ( − 30 + d 2 ) [ L ] [\omega_{X_\infty}]=p^*_\infty[\omega_R]+[R_{\mathrm{ref}}]=\left(-30+\frac{d}{2}\right)[\mathcal{L}] [ ω X ∞ ] = p ∞ ∗ [ ω R ] + [ R ref ] = ( − 30 + 2 d ) [ L ]
in Pic ( X ∞ ) ⊗ Q \operatorname{Pic}(X_\infty)\otimes\mathbb{Q} Pic ( X ∞ ) ⊗ Q . By (37) , the left-hand side is 4 [ L ] 4[\mathcal{L}] 4 [ L ] . The class [ L ] [\mathcal{L}] [ L ] is nonzero: a positive multiple is the pullback of the restriction of an ample class on R ^ \widehat{R} R , whose complement has codimension two. Hence 4 = − 30 + d / 2 4=-30+d/2 4 = − 30 + d /2 , and d = 68 d=68 d = 68 .
Since d = 68 d=68 d = 68 , we have the double covering p : V → R − S R p:\mathcal{V}\to R-S_R p : V → R − S R branched along ( D R ∩ R ) ∖ S R (D_R\cap R)\setminus S_R ( D R ∩ R ) ∖ S R , where
V = { [ r : r ~ ] ∈ P ( 2 , 4 , 6 , 8 , 10 , d 2 ) ∣ [ r ] ∈ R ∖ S R , r ~ 2 = Δ R ( r ) } . (38) \mathcal{V}=\left\{[r:\widetilde r]\in\mathbb{P}\left(2,4,6,8,10,\frac{d}{2}\right)\mathrel{\big|}[r]\in R\setminus S_R,\ \widetilde r^2=\Delta_R(r)\right\}. \tag*{(38)} V = { [ r : r ] ∈ P ( 2 , 4 , 6 , 8 , 10 , 2 d ) [ r ] ∈ R ∖ S R , r 2 = Δ R ( r ) } . (38)
Proposition 3.18. There are two holomorphic functions r ~ 4 \widetilde r_4 r 4 and r ~ 30 \widetilde r_{30} r 30 on D A 1 , H ∞ ∘ D^\circ_{A_1,H_\infty} D A 1 , H ∞ ∘ satisfying
r ~ 4 2 = r 8 , r ~ 30 2 = d 60 ( r ) , (39) \widetilde r_4^2=r_8,\qquad\widetilde r_{30}^2=d_{60}(r), \tag*{(39)} r 4 2 = r 8 , r 30 2 = d 60 ( r ) , (39)
where r 2 , r 4 , r 6 , r 8 , r 10 r_2,r_4,r_6,r_8,r_{10} r 2 , r 4 , r 6 , r 8 , r 10 are regarded as modular forms by virtue of Theorem 3.14 and d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] d_{60}(r)\in\mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] is the weighted homogeneous polynomial of Definition A.1 .
Proof. According to Proposition 3.9 and Remark 3.10, the divisor { [ r ] ∈ R ^ ∣ r 8 = 0 } \{[r] \in\widehat{R} \mid r_{8}=0\} {[ r ] ∈ R ∣ r 8 = 0 } corresponds to the locus for the M 2 \mathbf{M}_{2} M 2 -polarized K 3 K3 K 3 surfaces. This comes from a reflection hyperplane in D A 1 , H ∞ ∘ \mathcal{D}^{\circ}_{A_{1},H_{\infty}} D A 1 , H ∞ ∘ for a ( − 2 ) (-2) ( − 2 ) -reflection of Γ A 1 \Gamma_{A_{1}} Γ A 1 .
Moreover, we obtain the polynomial d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] d_{60}(r) \in\mathbb{C}[r_{2},r_{4},r_{6},r_{8},r_{10}] d 60 ( r ) ∈ C [ r 2 , r 4 , r 6 , r 8 , r 10 ] from the polynomial δ 60 ( u ) \delta_{60}(u) δ 60 ( u ) , which appears in Lemma 3.3 , via the correspondence (26) and (28) . Equivalently, it is the residual discriminant polynomial of Definition A.1 . By Lemma 3.3 (2) and Lemma A.2 , the divisor { [ r ] ∈ R ^ ∣ d 60 ( r ) = 0 } \{[r] \in\widehat{R} \mid d_{60}(r)=0\} {[ r ] ∈ R ∣ d 60 ( r ) = 0 } is the locus where the Néron-Severi lattice M 1 \mathbf{M}_{1} M 1 of (8) for a very general member of { d 60 ( r ) = 0 } ∩ R \{d_{60}(r)=0\}\cap R { d 60 ( r ) = 0 } ∩ R degenerates to U ⊕ E 7 ( − 1 ) ⊕ E 7 ( − 1 ) ⊕ A 1 ( − 1 ) U \oplus E_{7}(-1) \oplus E_{7}(-1) \oplus A_{1}(-1) U ⊕ E 7 ( − 1 ) ⊕ E 7 ( − 1 ) ⊕ A 1 ( − 1 ) . This implies that the divisor originates from a reflection hyperplane in D A 1 , H ∞ ∘ \mathcal{D}^{\circ}_{A_{1},H_{\infty}} D A 1 , H ∞ ∘ for a ( − 2 ) (-2) ( − 2 ) -reflection of Γ A 1 \Gamma_{A_{1}} Γ A 1 .
Each of the polynomials r 8 r_{8} r 8 and d 60 ( r ) d_{60}(r) d 60 ( r ) divides the polynomial Δ R ( r ) \Delta_{R}(r) Δ R ( r ) . By virtue of Proposition 3.17 , we have
Δ R ( r ) = const ⋅ r 8 ⋅ d 60 ( r ) . \Delta_{R}(r)=\operatorname{const}\cdot r_{8}\cdot d_{60}(r). Δ R ( r ) = const ⋅ r 8 ⋅ d 60 ( r ) .
The existence of the two square roots follows from [33 Theorem 1.1] . In our notation, its forms v ~ 4 \widetilde{v}_{4} v 4 and v ~ 30 \widetilde{v}_{30} v 30 satisfy v ~ 4 2 = v 8 \widetilde{v}_{4}^{2}=v_{8} v 4 2 = v 8 and v ~ 30 2 = d 60 ( v ) \widetilde{v}_{30}^{2}=d_{60}(v) v 30 2 = d 60 ( v ) . By (24) and (28) , the coordinate substitution is
( v 4 , v 6 , v 8 , v 10 , v 12 ) = ( r 4 , r 6 , r 8 , r 10 + r 2 r 8 , r 2 r 10 ) . (v_{4},v_{6},v_{8},v_{10},v_{12})=(r_{4},r_{6},r_{8},r_{10}+r_{2}r_{8},r_{2}r_{10}). ( v 4 , v 6 , v 8 , v 10 , v 12 ) = ( r 4 , r 6 , r 8 , r 10 + r 2 r 8 , r 2 r 10 ) .
It gives v 8 = r 8 v_{8}=r_{8} v 8 = r 8 and d 60 ( v ) = d 60 ( r ) d_{60}(v)=d_{60}(r) d 60 ( v ) = d 60 ( r ) , with compatible normalizations, as specified in Appendix A. The restrictions of v ~ 4 \widetilde{v}_{4} v 4 and v ~ 30 \widetilde{v}_{30} v 30 to D A 1 , H ∞ ∘ \mathcal{D}^{\circ}_{A_{1},H_{\infty}} D A 1 , H ∞ ∘ therefore give the required functions r ~ 4 \widetilde{r}_{4} r 4 and r ~ 30 \widetilde{r}_{30} r 30 .
Recall the lattice B 1 \mathbf{B}_{1} B 1 of (11) and its stable orthogonal subgroup Γ B 1 \Gamma_{\mathbf{B}_{1}} Γ B 1 . Let τ ˉ \bar{\tau} τ ˉ be the image of the involution τ \tau τ in O + ( B 1 ) / Γ B 1 \mathrm{O}^{+}(\mathbf{B}_{1})/\Gamma_{\mathbf{B}_{1}} O + ( B 1 ) / Γ B 1 . Then
O + ( B 1 ) / Γ B 1 ≅ S 6 × ⟨ τ ˉ ⟩ (40) \mathrm{O}^{+}(\mathbf{B}_{1})/\Gamma_{\mathbf{B}_{1}}\cong S_{6}\times\langle\bar{\tau}\rangle \tag*{(40)} O + ( B 1 ) / Γ B 1 ≅ S 6 × ⟨ τ ˉ ⟩ (40)
Under the period identification, the factor S 6 S_{6} S 6 acts as S 6 f i x S_{6}^{\mathrm{fix}} S 6 fix on the Coble model of § 2.2 , and τ ˉ \bar{\tau} τ ˉ acts as the Gale involution. See [33 Proposition 4.1 and Section 4] . Let us consider the odd unimodular lattice T T T of signature ( 2 , 4 ) (2,4) ( 2 , 4 ) :
T ≅ 1 ⊕ 2 ⊕ ( − 1 ) ⊕ 4 ≅ U ⊕ 2 ⊕ ( − 1 ) ⊕ 2 . T\cong1^{\oplus2}\oplus(-1)^{\oplus4}\cong U^{\oplus2}\oplus(-1)^{\oplus2}. T ≅ 1 ⊕ 2 ⊕ ( − 1 ) ⊕ 4 ≅ U ⊕ 2 ⊕ ( − 1 ) ⊕ 2 .
Take a basis x 1 , x 2 , x 3 , x 4 , x 5 , x 6 \mathbf{x}_{1},\mathbf{x}_{2},\mathbf{x}_{3},\mathbf{x}_{4},\mathbf{x}_{5},\mathbf{x}_{6} x 1 , x 2 , x 3 , x 4 , x 5 , x 6 of T T T with the intersection matrix U ⊕ 2 ⊕ ( − 1 ) ⊕ 2 U^{\oplus2}\oplus(-1)^{\oplus2} U ⊕ 2 ⊕ ( − 1 ) ⊕ 2 . Then, the sublattice { x ∈ T ∣ ( x , x ) ∈ 2 Z } \{x\in T\mid(x,x)\in2\mathbb{Z}\} { x ∈ T ∣ ( x , x ) ∈ 2 Z } of index two is generated by x 1 , x 2 , x 3 , x 4 , x 5 + x 6 , x 5 − x 6 \mathbf{x}_{1},\mathbf{x}_{2},\mathbf{x}_{3},\mathbf{x}_{4},\mathbf{x}_{5}+\mathbf{x}_{6},\mathbf{x}_{5}-\mathbf{x}_{6} x 1 , x 2 , x 3 , x 4 , x 5 + x 6 , x 5 − x 6 . This sublattice is isomorphic to the lattice A 1 \mathbf{A}_{1} A 1 and O ( A 1 ) ≅ O ( T ) \mathrm{O}(\mathbf{A}_{1})\cong\mathrm{O}(T) O ( A 1 ) ≅ O ( T ) . Also, O ( T ) = O ( B 1 ) \mathrm{O}(T)=\mathrm{O}(\mathbf{B}_{1}) O ( T ) = O ( B 1 ) . Hence, we have an isomorphism of groups:
T : O + ( A 1 ) → ∼ O + ( B 1 ) . (41) T:\mathrm{O}^{+}(\mathbf{A}_{1})\xrightarrow{\sim}\mathrm{O}^{+}(\mathbf{B}_{1}). \tag*{(41)} T : O + ( A 1 ) ∼ O + ( B 1 ) . (41)
When comparing Heegner divisors through this similarity, we take a primitive normal vector separately in each lattice. Indeed, α 1 = x 5 + x 6 \alpha_{1}=x_{5}+x_{6} α 1 = x 5 + x 6 and α 2 = x 5 − x 6 \alpha_{2}=x_{5}-x_{6} α 2 = x 5 − x 6 have square − 2 -2 − 2 and divisibility two in A 1 \mathbf{A}_{1} A 1 , whereas their primitive representatives in B 1 \mathbf{B}_{1} B 1 have square − 4 -4 − 4 . The vector α 1 − α 2 = 2 x 6 \alpha_{1}-\alpha_{2}=2x_{6} α 1 − α 2 = 2 x 6 is primitive in A 1 \mathbf{A}_{1} A 1 , of square − 4 -4 − 4 and divisibility two. The primitive normal to the same mirror in B 1 \mathbf{B}_{1} B 1 is x 6 x_{6} x 6 , of square − 2 -2 − 2 . Finally, e 1 − e 1 ′ e_{1}-e_{1}^{\prime} e 1 − e 1 ′ has square − 2 -2 − 2 and divisibility one in A 1 \mathbf{A}_{1} A 1 , and remains primitive with square − 4 -4 − 4 in B 1 \mathbf{B}_{1} B 1 .
Put K B : = T − 1 ( Γ B 1 ) K_{B}:=T^{-1}(\Gamma_{\mathbf{B}_{1}}) K B := T − 1 ( Γ B 1 ) and Γ ′ : = Γ A 1 ∩ K B \Gamma^{\prime}:=\Gamma_{\mathbf{A}_{1}}\cap K_{B} Γ ′ := Γ A 1 ∩ K B . Since the involution σ \sigma σ exchanges α 1 \alpha_{1} α 1 and α 2 \alpha_{2} α 2 , we have T ( σ ) = diag ( 1 , 1 , 1 , 1 , 1 , − 1 ) T(\sigma)=\operatorname{diag}(1,1,1,1,1,-1) T ( σ ) = diag ( 1 , 1 , 1 , 1 , 1 , − 1 ) . Since B 1 = T ( 2 ) \mathbf{B}_{1}=T(2) B 1 = T ( 2 ) , its discriminant group is identified with T / 2 T T/2T T /2 T . The matrix T ( σ ) T(\sigma) T ( σ ) acts trivially on this group, so T ( σ ) ∈ Γ B 1 T(\sigma)\in\Gamma_{\mathbf{B}_{1}} T ( σ ) ∈ Γ B 1 . On the other hand, σ ∉ Γ A 1 \sigma\notin\Gamma_{\mathbf{A}_{1}} σ ∈ / Γ A 1 . Thus K B K_{B} K B is not contained in Γ A 1 \Gamma_{\mathbf{A}_{1}} Γ A 1 . Since Γ A 1 \Gamma_{\mathbf{A}_{1}} Γ A 1 is normal of index two in O + ( A 1 ) \mathrm{O}^{+}(\mathbf{A}_{1}) O + ( A 1 ) , we obtain Γ A 1 K B = O + ( A 1 ) \Gamma_{\mathbf{A}_{1}}K_{B}=\mathrm{O}^{+}(\mathbf{A}_{1}) Γ A 1 K B = O + ( A 1 ) . The second isomorphism theorem gives
K B / Γ ′ ≅ O + ( A 1 ) / Γ A 1 ≅ Z / 2 Z . K_{B}/\Gamma^{\prime}\cong\mathrm{O}^{+}(\mathbf{A}_{1})/\Gamma_{\mathbf{A}_{1}}\cong\mathbb{Z}/2\mathbb{Z}. K B / Γ ′ ≅ O + ( A 1 ) / Γ A 1 ≅ Z /2 Z .
Let G 1 \mathcal{G}_{1} G 1 be the family obtained by taking the minimal resolutions of double coverings of P 2 \mathbb{P}^{2} P 2 branched along six distinct ordered lines with no three concurrent. The transcendental lattice of @endsection a very general member is B 1 \mathbf{B}_1 B 1 . The corresponding period mapping and its extension to the GIT quotient are described in Theorem 5.1 .
The following is the main result of this section.
Theorem 3.19. The ring A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ of meromorphic modular forms with respect to Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 is generated by r 2 , r 4 , r 6 , r 8 , r 10 r_2,r_4,r_6,r_8,r_{10} r 2 , r 4 , r 6 , r 8 , r 10 , which are given via the period mapping Φ 1 \Phi_1 Φ 1 as in Theorem 3.14 , and r ~ 4 \widetilde{r}_4 r 4 , r ~ 30 \widetilde{r}_{30} r 30 satisfying (39) . Here, r ~ 4 \widetilde{r}_4 r 4 and r ~ 30 \widetilde{r}_{30} r 30 are modular forms of nontrivial characters of Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 . The character of r ~ 4 \widetilde{r}_4 r 4 is independent of that of r ~ 30 \widetilde{r}_{30} r 30 . In particular,
A ∘ ( Γ A 1 ) H ∞ ≅ C [ r 2 , r 4 , r 6 , r 8 , r 10 , r ~ 4 , r ~ 30 ] / ( r ~ 4 2 − r 8 , r ~ 30 2 − d 60 ( r ) ) . (42) \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} \cong\mathbb{C}[r_2,r_4,r_6,r_8,r_{10},\widetilde{r}_4,\widetilde{r}_{30}]/(\widetilde{r}_4^{\,2}-r_8,\widetilde{r}_{30}^{\,2}-d_{60}(r)). \tag*{(42)} A ∘ ( Γ A 1 ) H ∞ ≅ C [ r 2 , r 4 , r 6 , r 8 , r 10 , r 4 , r 30 ] / ( r 4 2 − r 8 , r 30 2 − d 60 ( r )) . (42)
Proof. By Proposition 3.18, we may choose r ~ k \widetilde{r}_k r k as the restriction of v ~ k \widetilde{v}_k v k for k ∈ { 4 , 30 } k \in\{4,30\} k ∈ { 4 , 30 } . For λ ∈ C × \lambda\in\mathbb{C}^{\times} λ ∈ C × and γ ∈ Γ A 1 \gamma\in\Gamma_{\mathrm{A}_1} γ ∈ Γ A 1 , these functions satisfy r ~ k ( λ ξ ) = λ − k r ~ k ( ξ ) \widetilde{r}_k(\lambda\xi)=\lambda^{-k}\widetilde{r}_k(\xi) r k ( λ ξ ) = λ − k r k ( ξ ) and r ~ k ( γ ξ ) = χ k ( γ ) r ~ k ( ξ ) \widetilde{r}_k(\gamma\xi)=\chi_k(\gamma)\widetilde{r}_k(\xi) r k ( γ ξ ) = χ k ( γ ) r k ( ξ ) . The character values in Proposition 3.16 show that χ 4 ∣ Γ A 1 \chi_4|_{\Gamma_{\mathrm{A}_1}} χ 4 ∣ Γ A 1 and χ 30 ∣ Γ A 1 \chi_{30}|_{\Gamma_{\mathrm{A}_1}} χ 30 ∣ Γ A 1 are nontrivial and independent: reflections of types ( − 2 , 2 ) (-2,2) ( − 2 , 2 ) and ( − 2 , 1 ) (-2,1) ( − 2 , 1 ) belong to Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 , and the pairs of values are ( − 1 , 1 ) (-1,1) ( − 1 , 1 ) and ( 1 , − 1 ) (1,-1) ( 1 , − 1 ) , respectively. Moreover, χ 4 χ 10 χ 30 = det \chi_4\chi_{10}\chi_{30}=\det χ 4 χ 10 χ 30 = det and χ 10 ∣ Γ A 1 = 1 \chi_{10}|_{\Gamma_{\mathrm{A}_1}}=1 χ 10 ∣ Γ A 1 = 1 . Thus r ~ 4 r ~ 30 \widetilde{r}_4\widetilde{r}_{30} r 4 r 30 has character det \det det .
For the interpretation in terms of (40) , write ( g γ , ϵ γ ) (g_\gamma,\epsilon_\gamma) ( g γ , ϵ γ ) for the image of T ( γ ) T(\gamma) T ( γ ) in S 6 f i x × ⟨ τ ‾ ⟩ S_6^{\mathrm{fix}}\times\langle\overline{\tau}\rangle S 6 fix × ⟨ τ ⟩ , where ϵ γ ∈ { ± 1 } \epsilon_\gamma\in\{\pm1\} ϵ γ ∈ { ± 1 } . The comparison with the Coble coordinates in Section 5.3 gives χ 4 ( γ ) = ϵ γ \chi_4(\gamma)=\epsilon_\gamma χ 4 ( γ ) = ϵ γ and χ 30 ( γ ) = sgn ( g γ ) \chi_{30}(\gamma)=\operatorname{sgn}(g_\gamma) χ 30 ( γ ) = sgn ( g γ ) . Indeed, under the graded period identification and compatible scalar normalizations, v ~ 4 \widetilde{v}_4 v 4 corresponds to z z z and v ~ 30 \widetilde{v}_{30} v 30 to ∏ i < j ( x i − x j ) \prod_{i<j}(x_i-x_j) ∏ i < j ( x i − x j ) . Their squares are v 8 v_8 v 8 and d 60 ( v ) d_{60}(v) d 60 ( v ) , respectively. The coordinate substitution in Proposition 3.18 identifies these with r 8 r_8 r 8 and d 60 ( r ) d_{60}(r) d 60 ( r ) . □ \square □
We also recall the following result of [33 ] on the structure of the ring A ( O + ( A 1 ) ) \mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) A ( O + ( A 1 )) of holomorphic modular forms on D A 1 \mathcal{D}_{\mathrm{A}_1} D A 1 with respect to the orthogonal group O + ( A 1 ) \mathrm{O}^{+}(\mathrm{A}_1) O + ( A 1 ) :
A ( O + ( A 1 ) ) ≅ C [ v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 4 , v ~ 10 , v ~ 30 ] / ( v ~ 4 2 − v 8 , v ~ 10 2 − v 10 2 + 4 v 8 v 12 , v ~ 30 2 − d 60 ( v ) ) . (43) \mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1))\cong\mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_4,\widetilde{v}_{10},\widetilde{v}_{30}]/(\widetilde{v}_4^{\,2}-v_8,\widetilde{v}_{10}^{\,2}-v_{10}^{\,2}+4v_8v_{12},\widetilde{v}_{30}^{\,2}-d_{60}(v)). \tag*{(43)} A ( O + ( A 1 )) ≅ C [ v 4 , v 6 , v 8 , v 10 , v 12 , v 4 , v 10 , v 30 ] / ( v 4 2 − v 8 , v 10 2 − v 10 2 + 4 v 8 v 12 , v 30 2 − d 60 ( v )) . (43)
Here, v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 10 v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_{10} v 4 , v 6 , v 8 , v 10 , v 12 , v 10 are the modular forms appearing in Proposition 3.6 and d 60 ( v ) d_{60}(v) d 60 ( v ) is the expression of the residual discriminant δ 60 ( u ) \delta_{60}(u) δ 60 ( u ) of Lemma 3.3 in the coordinates (24) ; see Appendix A. It is useful to compare the structure of A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ with that of A ( O + ( A 1 ) ) \mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) A ( O + ( A 1 )) :
The character of the modular form v ~ 10 ∈ A ( O + ( A 1 ) ) \widetilde{v}_{10}\in\mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) v 10 ∈ A ( O + ( A 1 )) corresponds to O + ( A 1 ) / Γ A 1 = ⟨ σ ‾ ⟩ ≅ Z / 2 Z \mathrm{O}^{+}(\mathrm{A}_1)/\Gamma_{\mathrm{A}_1}=\langle\overline{\sigma}\rangle\cong\mathbb{Z}/2\mathbb{Z} O + ( A 1 ) / Γ A 1 = ⟨ σ ⟩ ≅ Z /2 Z . Its restriction to Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 is therefore trivial. In the coordinates of (28) , we have v ~ 10 = r 10 − r 2 r 8 \widetilde{v}_{10}=r_{10}-r_2r_8 v 10 = r 10 − r 2 r 8 . Thus v ~ 10 \widetilde{v}_{10} v 10 belongs to C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] C [ r 2 , r 4 , r 6 , r 8 , r 10 ] and is not needed as an additional generator in (42) . All elements of our ring A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ of meromorphic modular forms are holomorphic on D A 1 , H ∞ ∘ = D A 1 ∖ H ∞ \mathcal{D}^{\circ}_{\mathrm{A}_1,H_\infty}=\mathcal{D}_{\mathrm{A}_1}\setminus H_\infty D A 1 , H ∞ ∘ = D A 1 ∖ H ∞ , where H ∞ H_\infty H ∞ is the support of H ∞ \mathcal{H}_\infty H ∞ . Recall that σ \sigma σ interchanges H ∞ H_\infty H ∞ and H 0 H_0 H 0 . Since H 0 H_0 H 0 is not removed from D A 1 , H ∞ ∘ \mathcal{D}^{\circ}_{\mathrm{A}_1,H_\infty} D A 1 , H ∞ ∘ , the group ⟨ σ ⟩ \langle\sigma\rangle ⟨ σ ⟩ does not act on D A 1 , H ∞ ∘ \mathcal{D}^{\circ}_{\mathrm{A}_1,H_\infty} D A 1 , H ∞ ∘ .
The subring A ( O + ( A 1 ) , i d ) \mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1),\mathrm{id}) A ( O + ( A 1 ) , id ) of A ( O + ( A 1 ) ) \mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) A ( O + ( A 1 )) is C [ v 4 , v 6 , v 8 , v 10 , v 12 ] \mathbb{C}[v_4,v_6,v_8,v_{10},v_{12}] C [ v 4 , v 6 , v 8 , v 10 , v 12 ] , whereas the subring A ∘ ( Γ A 1 , i d ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1},\mathrm{id})_{H_\infty} A ∘ ( Γ A 1 , id ) H ∞ of A ∘ ( Γ A 1 ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} A ∘ ( Γ A 1 ) H ∞ is C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \mathbb{C}[r_2,r_4,r_6,r_8,r_{10}] C [ r 2 , r 4 , r 6 , r 8 , r 10 ] . Forms in A ∘ ( Γ A 1 , i d ) H ∞ \mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1},\mathrm{id})_{H_\infty} A ∘ ( Γ A 1 , id ) H ∞ are allowed to have poles along H ∞ H_\infty H ∞ . After passing to Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 and allowing these poles, the weights of the generators change from ( 4 , 6 , 8 , 10 , 12 ) (4,6,8,10,12) ( 4 , 6 , 8 , 10 , 12 ) to ( 2 , 4 , 6 , 8 , 10 ) (2,4,6,8,10) ( 2 , 4 , 6 , 8 , 10 ) . An individual modular form retains its weight when regarded as a form for the subgroup Γ A 1 \Gamma_{\mathrm{A}_1} Γ A 1 . We emphasize that the removed arrangement divisor H ∞ H_\infty H ∞ does not give the zero set of a modular form on D A 1 \mathcal{D}_{\mathrm{A}_1} D A 1 (cf. [25 Corollary 7.5] , see also [24 Section 6] ).
The sum of the weights of the generators enters the calculation of canonical classes in the proof of Proposition 3.17. For the two rings with trivial character above, these sums are 40 40 40 and 30 30 30 , respectively.
The zero set of r ~ 4 ∈ A ∘ ( Γ A 1 ) H ∞ \widetilde{r}_4\in\mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} r 4 ∈ A ∘ ( Γ A 1 ) H ∞ gives the locus where the Néron–Severi lattice of members of F 1 \mathcal{F}_1 F 1 degenerate to M 2 \mathbf{M}_2 M 2 , as well as the zero set of v ~ 4 ∈ A ( O + ( A 1 ) ) \widetilde{v}_4\in\mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) v 4 ∈ A ( O + ( A 1 )) does. Moreover, both r ~ 30 ∈ A ∘ ( Γ A 1 ) H ∞ \widetilde{r}_{30}\in\mathcal{A}^{\circ}(\Gamma_{\mathrm{A}_1})_{H_\infty} r 30 ∈ A ∘ ( Γ A 1 ) H ∞ and v ~ 30 ∈ A ( O + ( A 1 ) ) \widetilde{v}_{30}\in\mathcal{A}(\mathrm{O}^{+}(\mathrm{A}_1)) v 30 ∈ A ( O + ( A 1 )) originate from the polynomial δ 60 ( u ) \delta_{60}(u) δ 60 ( u ) in Lemma 3.3. In this comparison, the generators with nontrivial characters retain their weights 4 4 4 and $30.
Moduli of cubic surfaces
In this section, we recall the basic geometry of cubic surfaces and descriptions of their moduli spaces.
Lattices and GIT quotients
A cubic surface is a hypersurface S = { F ( x , y , z , w ) = 0 } ⊂ P 3 S=\{F(x,y,z,w)=0\}\subset\mathbb{P}^{3} S = { F ( x , y , z , w ) = 0 } ⊂ P 3 with deg F = 3 \deg F=3 deg F = 3 . If S S S is smooth, then S S S is a del Pezzo surface of degree 3. The complete anticanonical linear system ∣ − K S ∣ \lvert-K_{S}\rvert ∣ − K S ∣ is very ample and gives the cubic embedding S ↪ P 3 S\hookrightarrow\mathbb{P}^{3} S ↪ P 3 .
Let p 1 , … , p 6 ∈ P 2 p_{1},\ldots,p_{6}\in\mathbb{P}^{2} p 1 , … , p 6 ∈ P 2 be distinct points such that no three are collinear and the six points do not lie on a conic. The blow-up of P 2 \mathbb{P}^{2} P 2 at these six points is a smooth cubic surface under the anticanonical embedding, and every smooth cubic surface arises in this way [11 Section 2.5] .
Let H H H be the pull-back of the class of a line in P 2 \mathbb{P}^{2} P 2 along the blow-up Bl p 1 , … , p 6 P 2 → P 2 \operatorname{Bl}_{p_{1},\ldots,p_{6}}\mathbb{P}^{2}\to\mathbb{P}^{2} Bl p 1 , … , p 6 P 2 → P 2 , and let E i E_{i} E i be the exceptional classes. Then
Pic ( S ) ≅ Z H ⊕ ⨁ i = 1 6 Z E i , − K S = 3 H − ∑ i = 1 6 E i . \operatorname{Pic}(S)\cong\mathbb{Z}H\oplus\bigoplus_{i=1}^{6}\mathbb{Z}E_{i},\qquad-K_{S}=3H-\sum_{i=1}^{6}E_{i}. Pic ( S ) ≅ Z H ⊕ i = 1 ⨁ 6 Z E i , − K S = 3 H − i = 1 ∑ 6 E i .
with intersection pairing H 2 = 1 H^{2}=1 H 2 = 1 , E i 2 = − 1 E_{i}^{2}=-1 E i 2 = − 1 , H ⋅ E i = 0 H\cdot E_{i}=0 H ⋅ E i = 0 , E i ⋅ E j = 0 E_{i}\cdot E_{j}=0 E i ⋅ E j = 0 for i ≠ j i\ne j i = j . Let I 1 , 6 I_{1,6} I 1 , 6 be the lattice Z 7 \mathbb{Z}^{7} Z 7 with Gram matrix diag ( 1 , − 1 , − 1 , − 1 , − 1 , − 1 , − 1 ) \operatorname{diag}(1,-1,-1,-1,-1,-1,-1) diag ( 1 , − 1 , − 1 , − 1 , − 1 , − 1 , − 1 ) . A marking of a smooth cubic surface S S S is an isometry ϕ : Pic ( S ) → ∼ I 1 , 6 \phi:\operatorname{Pic}(S)\xrightarrow{\sim}I_{1,6} ϕ : Pic ( S ) ∼ I 1 , 6 that maps − K S -K_{S} − K S to ( 3 , − 1 , − 1 , − 1 , − 1 , − 1 , − 1 ) (3,-1,-1,-1,-1,-1,-1) ( 3 , − 1 , − 1 , − 1 , − 1 , − 1 , − 1 ) . A more geometric but equivalent way to think about markings, compatible with the lattice marking above, uses the blow-up description. A choice of six pairwise disjoint lines on S S S determines a blow-down S → P 2 S\to\mathbb{P}^{2} S → P 2 , and an ordering of these lines determines a basis { H , E 1 , … , E 6 } \{H,E_{1},\ldots,E_{6}\} { H , E 1 , … , E 6 } of Pic ( S ) \operatorname{Pic}(S) Pic ( S ) . Different choices of blow-down correspond to different chambers of the Weyl group W ( E 6 ) W(E_{6}) W ( E 6 ) . A further classical feature is the presence of twenty-seven ( − 1 ) (-1) ( − 1 ) -curves on S S S , which are represented by the classes E i E_{i} E i ( 1 ≤ i ≤ 6 ) (1\le i\le6) ( 1 ≤ i ≤ 6 ) , H − E i − E j H-E_{i}-E_{j} H − E i − E j ( 1 ≤ i < j ≤ 6 ) (1\le i<j\le6) ( 1 ≤ i < j ≤ 6 ) , and 2 H − ∑ k ≠ i E k 2H-\sum_{k\ne i}E_{k} 2 H − ∑ k = i E k ( 1 ≤ i ≤ 6 ) (1\le i\le6) ( 1 ≤ i ≤ 6 ) .
Let V : = H 0 ( P 3 , O P 3 ( 3 ) ) V:=H^{0}(\mathbb{P}^{3},\mathcal{O}_{\mathbb{P}^{3}}(3)) V := H 0 ( P 3 , O P 3 ( 3 )) so that P ( V ) ≅ P 19 \mathbb{P}(V)\cong\mathbb{P}^{19} P ( V ) ≅ P 19 parametrizes cubic hypersurfaces in P 3 \mathbb{P}^{3} P 3 . The group GL 4 ( C ) \operatorname{GL}_{4}(\mathbb{C}) GL 4 ( C ) acts naturally on P ( V ) \mathbb{P}(V) P ( V ) , and geometric invariant theory produces a projective quotient
M c u b G I T : = P ( V ) / / SL 4 ( C ) , \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}}:=\mathbb{P}(V)//\operatorname{SL}_{4}(\mathbb{C}), M cub GIT := P ( V ) // SL 4 ( C ) ,
which contains the moduli space M c u b s m \mathcal{M}_{\mathrm{cub}}^{\mathrm{sm}} M cub sm of smooth cubic surfaces as a Zariski open dense subset. The boundary of M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub GIT parametrizes certain singular cubic surfaces, which correspond to GIT polystable orbits, and M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub GIT contains a unique point corresponding to the strictly semistable locus. For general background on GIT quotients and their use in moduli problems we refer to [32 ] .
An ordering of six distinct points with no three collinear and not all on a conic determines a geometric marking of their blow-up. Conversely, a marked smooth cubic surface determines such a blow-down. This gives the identification
M c u b m ≅ X ( 3 , 6 ) ∖ D c o n ∘ \mathcal{M}_{\mathrm{cub}}^{\mathrm{m}}\cong X(3,6)\setminus D_{\mathrm{con}}^{\circ} M cub m ≅ X ( 3 , 6 ) ∖ D con ∘
of (2) . See, for instance, [1 Section 2.3] or [12 Definition 4.1] . Forgetting the marking gives a finite morphism
M c u b m ⟶ M c u b s m ⊂ M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{m}}\longrightarrow\mathcal{M}_{\mathrm{cub}}^{\mathrm{sm}}\subset\mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub m ⟶ M cub sm ⊂ M cub GIT
whose Galois group is the Weyl group W ( E 6 ) W(E_{6}) W ( E 6 ) . In this sense M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub GIT is the moduli space of unmarked cubic surfaces, while M c u b m \mathcal{M}_{\mathrm{cub}}^{\mathrm{m}} M cub m is the moduli space of marked ones. We refer to [2 , 38 ] for concrete realizations of M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub GIT and for the relation between different notions of marking.
Proposition 4.1 ([31 Theorems 7.14, 7.23, 7.24] ). On M c u b G I T \mathcal{M}_{\mathrm{cub}}^{\mathrm{GIT}} M cub GIT , the following properties hold.
(1) A cubic surface S ⊂ P 3 S\subset\mathbb{P}^{3} S ⊂ P 3 is stable if and only if S S S has at worst ordinary double points, that is, Du Val singularities of type A 1 A_{1} A 1 .
(2) The strictly semistable locus collapses to one point S 3 A 2 : = { y 3 = w x z } ⊂ P 3 S_{3A_{2}}:=\{y^{3}=wxz\}\subset\mathbb{P}^{3} S 3 A 2 := { y 3 = w x z } ⊂ P 3 , which has exactly 3 A 2 3A_{2} 3 A 2 singularities.
Ball quotient uniformization
The fundamental result of Allcock, Carlson, and Toledo identifies the moduli space of stable cubic surfaces with B 4 / Γ u m \mathbb{B}^{4}/\Gamma_{\mathrm{um}} B 4 / Γ um [2 ] . The locus of smooth cubic surfaces corresponds to the complement of the Heegner divisor
M c u b s m ≅ B 4 / Γ u m ∖ H ( Γ u m ) , \mathcal{M}_{\mathrm{cub}}^{\mathrm{sm}} \cong\mathbb{B}^{4}/\Gamma_{\mathrm{um}} \setminus\mathcal{H}(\Gamma_{\mathrm{um}}), M cub sm ≅ B 4 / Γ um ∖ H ( Γ um ) ,
where H ( Γ u m ) \mathcal{H}(\Gamma_{\mathrm{um}}) H ( Γ um ) is defined in (17) . For the marked subgroup Γ m = ker ( ρ ) \Gamma_{\mathrm{m}}=\ker(\rho) Γ m = ker ( ρ ) , the corresponding identification is
M c u b m ≅ B 4 / Γ m ∖ H ( Γ m ) . (44) \mathcal{M}_{\mathrm{cub}}^{\mathrm{m}} \cong\mathbb{B}^{4}/\Gamma_{\mathrm{m}} \setminus\mathcal{H}(\Gamma_{\mathrm{m}}). \tag*{(44)} M cub m ≅ B 4 / Γ m ∖ H ( Γ m ) . (44)
By [11 Section 2.12] , the marked discriminant H ( Γ m ) \mathcal{H}(\Gamma_{\mathrm{m}}) H ( Γ m ) has thirty-six irreducible components, indexed by the root pairs of E 6 E_{6} E 6 . The components D c o n D_{\mathrm{con}} D con , D i j k D_{ijk} D ij k , and D i j D_{ij} D ij introduced in § 2.2 correspond to
{ ± ( 2 H − ∑ i = 1 6 E i ) } , { ± ( H − E i − E j − E k ) } , { ± ( E i − E j ) } \left\{\pm\left(2H-\sum_{i=1}^{6}E_i\right)\right\}, \qquad\left\{\pm\left(H-E_i-E_j-E_k\right)\right\}, \qquad\left\{\pm\left(E_i-E_j\right)\right\} { ± ( 2 H − i = 1 ∑ 6 E i ) } , { ± ( H − E i − E j − E k ) } , { ± ( E i − E j ) }
respectively. Here 1 ≤ i < j < k ≤ 6 1\leq i<j<k\leq6 1 ≤ i < j < k ≤ 6 for D i j k D_{ijk} D ij k and 1 ≤ i < j ≤ 6 1\leq i<j\leq6 1 ≤ i < j ≤ 6 for D i j D_{ij} D ij . All these divisors lie on B 4 / Γ m \mathbb{B}^{4}/\Gamma_{\mathrm{m}} B 4 / Γ m .
With the unions D c o l D_{\mathrm{col}} D col and D p a i r D_{\mathrm{pair}} D pair fixed there, we have
H ( Γ m ) = D c o n ∪ D c o l ∪ D p a i r . (45) \mathcal{H}(\Gamma_{\mathrm{m}})=D_{\mathrm{con}}\cup D_{\mathrm{col}}\cup D_{\mathrm{pair}}. \tag*{(45)} H ( Γ m ) = D con ∪ D col ∪ D pair . (45)
We also need the Baily–Borel compactification of the marked ball quotient. We use the Coble model of X ‾ ( 3 , 6 ) G I T \overline{X}(3,6)^{\mathrm{GIT}} X ( 3 , 6 ) GIT and the actions S 6 p t s S_{6}^{\mathrm{pts}} S 6 pts and S 6 f i x S_{6}^{\mathrm{fix}} S 6 fix fixed in § 2.2 .
Coble constructed forty elements of the linear system ∣ I Sing ( X ‾ ( 3 , 6 ) G I T ) ( 3 ) ∣ \left|I_{\operatorname{Sing}(\overline{X}(3,6)^{\mathrm{GIT}})}(3)\right| I Sing ( X ( 3 , 6 ) GIT ) ( 3 ) which span a projective space P 9 \mathbb{P}^{9} P 9 . This induces a rational map
Φ C o b : X ‾ ( 3 , 6 ) G I T ⇢ P 9 . (46) \Phi_{\mathrm{Cob}}:\overline{X}(3,6)^{\mathrm{GIT}}\dashrightarrow\mathbb{P}^{9}. \tag*{(46)} Φ Cob : X ( 3 , 6 ) GIT ⇢ P 9 . (46)
This map is resolved by first blowing up the special points indexed by pair partitions of { 1 , … , 6 } \{1,\ldots,6\} { 1 , … , 6 } , then blowing up the fifteen singular curves ℓ i j \ell_{ij} ℓ ij , parametrizing the locus where p i = p j p_i=p_j p i = p j , and finally blowing up the remaining base surfaces. Let X 3 X_{3} X 3 be the resulting resolution. The proper transform of the Coble linear system is base point free and defines
Y m c u b : = Im ( X 3 → P 9 ) . (47) Y_{\mathrm{mcub}}:=\operatorname{Im}(X_{3}\to\mathbb{P}^{9}). \tag*{(47)} Y mcub := Im ( X 3 → P 9 ) . (47)
The conic and collinearity loci are mapped to the closures of D c o n D_{\mathrm{con}} D con and D i j k D_{ijk} D ij k , respectively. The collision locus has codimension 2 2 2 in ( P 2 ) 6 (\mathbb{P}^{2})^{6} ( P 2 ) 6 , and in the GIT quotient it is not a divisor. It appears as one of the fifteen singular lines ℓ i j \ell_{ij} ℓ ij . The divisor extracted over ℓ i j \ell_{ij} ℓ ij , followed by its proper transform in X 3 X_{3} X 3 , is mapped onto the closure of D i j D_{ij} D ij under the resolved Coble map. Thus
D ‾ i j ⊂ Y m c u b , D i j = D ‾ i j ∩ ( B 4 / Γ m ) , (48) \overline{D}_{ij}\subset Y_{\mathrm{mcub}}, \qquad D_{ij}=\overline{D}_{ij}\cap(\mathbb{B}^{4}/\Gamma_{\mathrm{m}}), \tag*{(48)} D ij ⊂ Y mcub , D ij = D ij ∩ ( B 4 / Γ m ) , (48)
where closures are taken in Y m c u b Y_{\mathrm{mcub}} Y mcub . Dolgachev, van Geemen, and Kondō showed that
Y m c u b ∖ M c u b m = D ‾ c o n ∪ D ‾ c o l ∪ D ‾ p a i r , (49) Y_{\mathrm{mcub}}\setminus\mathcal{M}_{\mathrm{cub}}^{\mathrm{m}}=\overline{D}_{\mathrm{con}}\cup\overline{D}_{\mathrm{col}}\cup\overline{D}_{\mathrm{pair}}, \tag*{(49)} Y mcub ∖ M cub m = D con ∪ D col ∪ D pair , (49)
and that Y m c u b Y_{\mathrm{mcub}} Y mcub has forty cusp singularities, all contained in these closures. Furthermore, by comparing this Coble image with the modular embedding of Allcock and Freitag by Borcherds products, one obtains
Y m c u b ≅ ( B 4 / Γ m ) ∗ . (50) Y_{\mathrm{mcub}}\cong(\mathbb{B}^{4}/\Gamma_{\mathrm{m}})^{*}. \tag*{(50)} Y mcub ≅ ( B 4 / Γ m ) ∗ . (50)
See [11 Section 2.11] for the extension of the ball quotient isomorphism to marked nodal cubic surfaces.
Comparison of the arithmetic quotients
The configuration space X(3,6) and the GIT quotient
We use the configuration space X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) and the GIT quotient X ( 3 , 6 ) ‾ G I T \overline{X(3,6)}^{\mathrm{GIT}} X ( 3 , 6 ) GIT fixed in § 2.2 . The open locus parametrizing smooth marked cubic surfaces is
M c u b m ≅ X ( 3 , 6 ) ∖ D c o n ∘ . (51) \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} \cong X(3,6) \setminus D^{\circ}_{\mathrm{con}}. \tag*{(51)} M cub m ≅ X ( 3 , 6 ) ∖ D con ∘ . (51)
The divisorial part of its complement in X ( 3 , 6 ) ‾ G I T \overline{X(3,6)}^{\mathrm{GIT}} X ( 3 , 6 ) GIT consists of the closure of the conic locus and the twenty collinearity divisors. The period mapping identifies this GIT quotient with the Baily–Borel compactification as follows.
Theorem 5.1 [28 Proposition 2.10.1] . The period mapping induces an isomorphism Φ M S Y : X ( 3 , 6 ) → ∼ ( D A 1 / Γ B 1 ) ∖ H B 1 ( 2 ) ( Γ B 1 ) \Phi_{\mathrm{MSY}}: X(3,6) \xrightarrow{\sim} (\mathcal{D}_{A_1}/\Gamma_{B_1}) \setminus\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}) Φ MSY : X ( 3 , 6 ) ∼ ( D A 1 / Γ B 1 ) ∖ H B 1 ( 2 ) ( Γ B 1 ) . It extends to an isomorphism X ( 3 , 6 ) ‾ G I T → ∼ ( D A 1 / Γ B 1 ) ∗ \overline{X(3,6)}^{\mathrm{GIT}} \xrightarrow{\sim} (\mathcal{D}_{A_1}/\Gamma_{B_1})^* X ( 3 , 6 ) GIT ∼ ( D A 1 / Γ B 1 ) ∗ . The extension is equivariant for permutations of the six labels.
Remark 5.2. The sublattice U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 2 ) U(2) \oplus U(2) \oplus A_1(-2) U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 2 ) (resp. U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) U(2) \oplus U(2) \oplus A_1(-1) U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 1 ) ) of B 1 B_1 B 1 gives the transcendental lattice of a very general member of the subfamily of G 1 \mathcal{G}_1 G 1 corresponding to the fixed locus of the involution τ \tau τ (resp. T ( σ ) T(\sigma) T ( σ ) ) in D B 1 \mathcal{D}_{B_1} D B 1 via the period mapping. The periods of these subfamilies are studied in [28 0.15, 0.19] . In particular, the subfamily with transcendental lattice U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 2 ) U(2) \oplus U(2) \oplus A_1(-2) U ( 2 ) ⊕ U ( 2 ) ⊕ A 1 ( − 2 ) parametrizes the Kummer surfaces of principally polarized abelian surfaces. This subfamily corresponds to the condition that the six lines in P 2 \mathbb{P}^2 P 2 are tangent to a conic.
By [33 Section 4] , the divisorial fixed locus of τ \tau τ on D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 is H B 1 , τ ( 4 ) ( Γ B 1 ) \mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1}) H B 1 , τ ( 4 ) ( Γ B 1 ) . Under Φ M S Y \Phi_{\mathrm{MSY}} Φ MSY , the conic locus D c o n ∘ D^{\circ}_{\mathrm{con}} D con ∘ corresponds to its intersection with the image of X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) . Therefore
M c u b m ≅ D A 1 / Γ B 1 ∖ ( H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ) . (52) \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} \cong\mathcal{D}_{A_1}/\Gamma_{B_1} \setminus\left(\mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}) \cup\mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1})\right). \tag*{(52)} M cub m ≅ D A 1 / Γ B 1 ∖ ( H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ) . (52)
The following proposition determines the irreducible components of the relevant Heegner divisors.
Proposition 5.3. On D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 , the reduced divisor
H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ∪ H B 1 , S 6 ( 4 ) ( Γ B 1 ) \mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}) \cup\mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1}) \cup\mathcal{H}^{(4)}_{B_1,S_6}(\Gamma_{B_1}) H B 1 ( 2 ) ( Γ B 1 ) ∪ H B 1 , τ ( 4 ) ( Γ B 1 ) ∪ H B 1 , S 6 ( 4 ) ( Γ B 1 ) has
20 + 1 + 15 = 36 20 + 1 + 15 = 36 20 + 1 + 15 = 36 labeled irreducible components.
Proof. Choose a 3 × 6 3 \times6 3 × 6 matrix whose i i i -th column is a homogeneous coordinate vector for p i p_i p i . The 3 × 3 3 \times3 3 × 3 minor indexed by { i , j , k } \{i,j,k\} { i , j , k } vanishes precisely when p i , p j , p k p_i,p_j,p_k p i , p j , p k are collinear. Under the extended period mapping of Theorem 5.1 , the Heegner divisor H B 1 ( 2 ) ( Γ B 1 ) \mathcal{H}^{(2)}_{B_1}(\Gamma_{B_1}) H B 1 ( 2 ) ( Γ B 1 ) is the union of the closures, in D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 , of the twenty loci where exactly one such minor vanishes.
For a fixed triple, consider the incidence variety parametrizing a line, three distinct labeled points on that line, and three further labeled points in P 2 \mathbb{P}^2 P 2 . This incidence variety is irreducible. The condition that the chosen triple be the only collinear triple defines a nonempty open subset. Its geometric quotient is therefore irreducible, as is its closure. Moreover, a general configuration in this locus has no other collinear triple, so different triples give distinct divisors. Thus there are exactly twenty irreducible components.
The closure of H B 1 , τ ( 4 ) ( Γ B 1 ) \mathcal{H}^{(4)}_{B_1,\tau}(\Gamma_{B_1}) H B 1 , τ ( 4 ) ( Γ B 1 ) in the Coble fourfold is the ramification divisor of its double covering of P 4 \mathbb{P}^4 P 4 . This divisor maps isomorphically onto the irreducible Igusa quartic. The other fifteen classes correspond to the inverse images of the transposition hyperplanes x i = x j x_i=x_j x i = x j in P 4 \mathbb{P}^4 P 4 [33 Section 4] . Each inverse image is irreducible because the restriction of the branch quartic 4 s 4 − s 2 2 4s_4-s_2^2 4 s 4 − s 2 2 is not a square. For example, on x 1 = x 2 x_1=x_2 x 1 = x 2 the specialization ( x 1 , … , x 6 ) = ( t , t , 1 , 2 , 3 , − 2 t − 6 ) (x_1,\ldots,x_6)=(t,t,1,2,3,-2t-6) ( x 1 , … , x 6 ) = ( t , t , 1 , 2 , 3 , − 2 t − 6 ) gives − ( 9 t 4 + 120 t 3 + 570 t 2 + 1128 t + 769 ) -(9t^4+120t^3+570t^2+1128t+769) − ( 9 t 4 + 120 t 3 + 570 t 2 + 1128 t + 769 ) , whose discriminant is nonzero. The other hyperplanes are obtained by permutations. Their inverse images are distinct, so they give fifteen irreducible components. □
The fifteen transposition components just described are distinct from the collision divisors D i j D_{ij} D ij on B 4 / Γ m \mathbb{B}^{4}/\Gamma_{m} B 4 / Γ m . The same transposition hyperplanes determine the branch divisor of the covering associated with the sign character, as follows.
By [33 Section 4] , the compactified quotient ( D A 1 / ⟨ Γ B 1 , τ ⟩ ) ∗ (\mathcal{D}_{A_{1}}/\langle\Gamma_{B_{1}},\tau\rangle)^{*} ( D A 1 / ⟨ Γ B 1 , τ ⟩ ) ∗ is P 4 \mathbb{P}^{4} P 4 , with the coordinates x 1 , … , x 6 x_{1},\ldots,x_{6} x 1 , … , x 6 satisfying s 1 ( x ) = 0 s_{1}(x)=0 s 1 ( x ) = 0 fixed in (1) . The quotient of P 4 \mathbb{P}^{4} P 4 by this S 6 S_{6} S 6 -action is P ( 2 , 3 , 4 , 5 , 6 ) \mathbb{P}(2,3,4,5,6) P ( 2 , 3 , 4 , 5 , 6 ) , with homogeneous coordinate ring C [ v 4 , v 6 , v 8 , v 10 , v 12 ] \mathbb{C}[v_{4},v_{6},v_{8},v_{10},v_{12}] C [ v 4 , v 6 , v 8 , v 10 , v 12 ] . Here v 2 i v_{2i} v 2 i are suitable symmetric generators of homogeneous degree i i i in the Coble coordinates and modular weight 2 i 2i 2 i . They need not be the individual elementary symmetric polynomials. Let A 6 = ker ( sgn : S 6 → { ± 1 } ) A_{6}=\ker(\operatorname{sgn}:S_{6}\to\{\pm1\}) A 6 = ker ( sgn : S 6 → { ± 1 }) . The corresponding double covering is P 4 / A 6 → P 4 / S 6 \mathbb{P}^{4}/A_{6}\to\mathbb{P}^{4}/S_{6} P 4 / A 6 → P 4 / S 6 . Its branch divisor is the image of the hyperplanes x i = x j x_{i}=x_{j} x i = x j , since the generic stabilizer on such a hyperplane is generated by the transposition ( i j ) (ij) ( ij ) . It is defined by the invariant polynomial Δ 60 = ∏ i < j ( x i − x j ) 2 \Delta_{60}=\prod_{i<j}(x_{i}-x_{j})^{2} Δ 60 = ∏ i < j ( x i − x j ) 2 , of homogeneous degree 30 in the Coble coordinates and modular weight 60.
Common open subsets and collinearity strata
We compare the Coble map of (46) with the period mapping of Theorem 5.1 . We use the subset X c o m X_{\mathrm{com}} X com defined in § 2.2 . For the symmetric linearization, the Hilbert–Mumford criterion says that a sextuple is stable if and only if its points are distinct and no four are collinear. The additional condition excludes a closed subset of the stable quotient. Thus X c o m X_{\mathrm{com}} X com is an open subset of X ‾ ( 3 , 6 ) G I T \overline{X}(3,6)^{\mathrm{GIT}} X ( 3 , 6 ) GIT containing X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) .
On ball quotients, we consider ( B 4 / Γ m ) ∖ D p a i r (\mathbb{B}^{4}/\Gamma_{m})\setminus D_{\mathrm{pair}} ( B 4 / Γ m ) ∖ D pair . We use the divisors D c o n D_{\mathrm{con}} D con , D i j k D_{ijk} D ij k , and D i j D_{ij} D ij fixed in § 2.2 ; their root labels are described in § 5 .
Theorem 5.4. The restriction of the Coble map of (46) is an isomorphism of algebraic varieties Φ C o b : X c o m → ∼ ( B 4 / Γ m ) ∖ D p a i r \Phi_{\mathrm{Cob}}:X_{\mathrm{com}}\xrightarrow{\sim}(\mathbb{B}^{4}/\Gamma_{m})\setminus D_{\mathrm{pair}} Φ Cob : X com ∼ ( B 4 / Γ m ) ∖ D pair .
Proof. Let p = ( p 1 , … , p 6 ) p=(p_{1},\ldots,p_{6}) p = ( p 1 , … , p 6 ) represent a point of X c o m X_{\mathrm{com}} X com , and set Y p = Bl p 1 , … , p 6 P 2 Y_{p}=\operatorname{Bl}_{p_{1},\ldots,p_{6}}\mathbb{P}^{2} Y p = Bl p 1 , … , p 6 P 2 . Using the classes introduced in § 5 , we have − K Y p = 3 H − ∑ i = 1 6 E i -K_{Y_{p}}=3H-\sum_{i=1}^{6}E_{i} − K Y p = 3 H − ∑ i = 1 6 E i and K Y p 2 = 3 K_{Y_{p}}^{2}=3 K Y p 2 = 3 . The divisor − K Y p -K_{Y_{p}} − K Y p is nef and the complete linear system ∣ − K Y p ∣ \lvert-K_{Y_{p}}\rvert ∣ − K Y p ∣ is base point free of dimension three and defines a birational morphism onto a normal cubic surface, contracting precisely the ( − 2 ) (-2) ( − 2 ) -curves [8 ] . The root classes listed in § 5 determine the possible ( − 2 ) (-2) ( − 2 ) -curves. Since the six points are distinct, neither E i − E j E_{i}-E_{j} E i − E j nor E j − E i E_{j}-E_{i} E j − E i is effective.
If the six points lie on a conic, that conic is smooth. A reducible conic would contain either four of the points on one line or three points on each of its two components, both of which are excluded. Its strict transform has the class 2 H − ∑ i = 1 6 E i 2H-\sum_{i=1}^{6}E_{i} 2 H − ∑ i = 1 6 E i . By Bézout’s theorem, no three of the six points are collinear. Together with the preceding classification of root classes, this shows that the strict transform of the conic is the unique ( − 2 ) (-2) ( − 2 ) -curve on Y p Y_{p} Y p .
Otherwise, the ( − 2 ) (-2) ( − 2 ) -curves are exactly the strict transforms of the lines through collinear triples, with classes H − E i − E j − E k H-E_{i}-E_{j}-E_{k} H − E i − E j − E k . Two distinct such triples have exactly one common index. Two common indices would force four collinear points, whereas disjoint triples are excluded by the definition of X c o m X_{\mathrm{com}} X com . For such sets I I I and J J J of indices, the intersection number is
( H − ∑ i ∈ I E i ) ⋅ ( H − ∑ j ∈ J E j ) = 1 − ∣ I ∩ J ∣ = 0. \left(H-\sum_{i\in I}E_{i}\right)\cdot\left(H-\sum_{j\in J}E_{j}\right)=1-\lvert I\cap J\rvert=0. ( H − i ∈ I ∑ E i ) ⋅ H − j ∈ J ∑ E j = 1 − ∣ I ∩ J ∣ = 0.
Thus these ( − 2 ) (-2) ( − 2 ) -curves are disjoint. The image of the morphism defined by ∣ − K Y p ∣ \lvert-K_{Y_{p}}\rvert ∣ − K Y p ∣ therefore has at worst ordinary double points. The ordered blow-up supplies a geometric marking, and the absence of effective classes ± ( E i − E j ) \pm(E_{i}-E_{j}) ± ( E i − E j ) shows that the resulting point belongs to ( B 4 / Γ m ) ∖ D p a i r (\mathbb{B}^{4}/\Gamma_{m})\setminus D_{\mathrm{pair}} ( B 4 / Γ m ) ∖ D pair .
Conversely, let S S S be a marked nodal cubic surface corresponding to a point of ( B 4 / Γ m ) ∖ D p a i r (\mathbb{B}^{4}/\Gamma_{m})\setminus D_{\mathrm{pair}} ( B 4 / Γ m ) ∖ D pair . Its geometric marking presents the minimal resolution as a succession of six labeled blow-ups of P 2 \mathbb{P}^{2} P 2 . If two blow-up centers had the same image in P 2 \mathbb{P}^{2} P 2 , one of the classes ± ( E i − E j ) \pm(E_{i}-E_{j}) ± ( E i − E j ) would be effective. The moduli point of S S S would then belong to D p a i r D_{\mathrm{pair}} D pair , contrary to our assumption. Hence the images of the six centers in P 2 \mathbb{P}^{2} P 2 are pairwise distinct. Nefness of the anticanonical class excludes four collinear points, since the strict transform of a line through four of the points would have negative anticanonical degree. If a triple and its complementary triple were both collinear, the strict transforms of the two lines would be ( − 2 ) (-2) ( − 2 ) -curves with intersection number one. This is impossible because the exceptional curves over distinct nodes are disjoint. The resulting configuration therefore belongs to X c o m X_{\mathrm{com}} X com . By the correspondence between geometric markings and equivalence classes of lattice markings in [11 Proposition 2.4] , the two constructions above induce mutually inverse bijections between the closed points of X c o m X_{\mathrm{com}} X com and those of ( B 4 / Γ m ) ∖ D p a i r (\mathbb{B}^{4}/\Gamma_{\mathrm{m}})\setminus D_{\mathrm{pair}} ( B 4 / Γ m ) ∖ D pair .
We now verify that this correspondence is realized by the Coble map. Over the space of configurations representing X c o m X_{\mathrm{com}} X com , the blow-ups form an algebraic family. For every member Y p Y_{p} Y p , one has h 0 ( Y p , O Y p ( − K Y p ) ) = 4 h^{0}(Y_{p},\mathcal{O}_{Y_{p}}(-K_{Y_{p}}))=4 h 0 ( Y p , O Y p ( − K Y p )) = 4 and H 1 ( Y p , O Y p ( − K Y p ) ) = 0 H^{1}(Y_{p},\mathcal{O}_{Y_{p}}(-K_{Y_{p}}))=0 H 1 ( Y p , O Y p ( − K Y p )) = 0 . Moreover, ∣ − K Y p ∣ |-K_{Y_{p}}| ∣ − K Y p ∣ is base point free. Cohomology and base change therefore give a locally free sheaf of rank four with a surjective relative evaluation map. The resulting relative anticanonical morphism gives an algebraic family of nodal cubic surfaces. This construction is invariant under P G L 3 \mathrm{PGL}_{3} PGL 3 and hence induces a morphism from X c o m X_{\mathrm{com}} X com to the stable locus of M c u b G I T \mathcal{M}^{\mathrm{GIT}}_{\mathrm{cub}} M cub GIT .
By the construction in [11 Section 2] , Y m c u b Y_{\mathrm{mcub}} Y mcub is the normalization of M c u b G I T \mathcal{M}^{\mathrm{GIT}}_{\mathrm{cub}} M cub GIT in the function field of M c u b m \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} M cub m . Since X c o m X_{\mathrm{com}} X com is an open subset of the GIT quotient X ( 3 , 6 ) G I T X(3,6)^{\mathrm{GIT}} X ( 3 , 6 ) GIT , it is normal. Since the normalization morphism Y m c u b → M c u b G I T Y_{\mathrm{mcub}}\to\mathcal{M}^{\mathrm{GIT}}_{\mathrm{cub}} Y mcub → M cub GIT is finite, the lift on the dense open subset M c u b m \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} M cub m extends uniquely to X c o m X_{\mathrm{com}} X com by the universal property of normalization. The lifted morphism assigns to each configuration its cubic surface equipped with the geometric marking induced by the ordered blow-up. On M c u b m \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} M cub m , this morphism agrees with the Coble map. It therefore gives a regular extension of the Coble map to X c o m X_{\mathrm{com}} X com .
The resulting morphism to ( B 4 / Γ m ) ∖ D p a i r (\mathbb{B}^{4}/\Gamma_{\mathrm{m}})\setminus D_{\mathrm{pair}} ( B 4 / Γ m ) ∖ D pair is bijective on closed points, hence quasi-finite. It is birational because it is an isomorphism on the locus parametrizing smooth cubic surfaces. Its target is normal, and the morphism is separated. Zariski’s Main Theorem therefore makes it an open immersion. Surjectivity then makes it an isomorphism. □
Corollary 5.5. The preceding isomorphism restricts to X ( 3 , 6 ) ≅ ( B 4 / Γ m ) ∖ ( D c o l ∪ D p a i r ) X(3,6)\cong(\mathbb{B}^{4}/\Gamma_{\mathrm{m}})\setminus(D_{\mathrm{col}}\cup D_{\mathrm{pair}}) X ( 3 , 6 ) ≅ ( B 4 / Γ m ) ∖ ( D col ∪ D pair ) . Together with Theorem 5.1 , this identifies the same open subset of B 4 / Γ m \mathbb{B}^{4}/\Gamma_{\mathrm{m}} B 4 / Γ m and D A 1 / D B 1 D_{A_{1}}/D_{B_{1}} D A 1 / D B 1 . Removing D c o n ∘ D_{\mathrm{con}}^{\circ} D con ∘ gives M c u b m \mathcal{M}^{\mathrm{m}}_{\mathrm{cub}} M cub m of (51) and (52) .
Proof. Inside X c o m X_{\mathrm{com}} X com , the condition that no three points are collinear defines X ( 3 , 6 ) X(3,6) X ( 3 , 6 ) . Under the preceding isomorphism this is the complement of D c o l D_{\mathrm{col}} D col . □
The isomorphism of Theorem 5.4 has the following properties.
(1) Its restriction to the conic locus gives an isomorphism
D c o n ∘ ⟶ ~ D c o n ∖ ( D c o l ∪ D p a i r ) . D_{\mathrm{con}}^{\circ}\ \widetilde{\longrightarrow}\ D_{\mathrm{con}}\setminus(D_{\mathrm{col}}\cup D_{\mathrm{pair}}). D con ∘ ⟶ D con ∖ ( D col ∪ D pair ) .
(2) Fix a nonempty collection of triples realized by a configuration in X c o m X_{\mathrm{com}} X com . The locus on which precisely these triples are collinear is mapped isomorphically onto the locus contained in all the corresponding divisors D i j k D_{ijk} D ij k and in no other component of H ( Γ m ) \mathcal{H}(\Gamma_{\mathrm{m}}) H ( Γ m ) .
At a general point of a collinearity divisor, the effective local stabilizer for the marked ball quotient is cyclic of order three, generated by a triflction, that is, a complex reflection of order three. See [2 Sections 3.18–3.20 and Theorem 7.25] . The effective local stabilizer along the corresponding divisor in D A 1 / D B 1 D_{A_{1}}/D_{B_{1}} D A 1 / D B 1 has order two, since it is generated by a reflection. Consequently, these isomorphisms of algebraic varieties do not give isomorphisms of the arithmetic quotient orbifolds along these divisors. They therefore do not, by themselves, identify the automorphic line bundles.
For the collinearity loci just described, the possible collections of distinct triples are exactly those in which any two triples have one common index. Such a collection has at most four members. Indeed, three triples cannot have a common index, since this would require seven distinct labels. Up to relabeling, three triples are therefore { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } , { 1 , 4 , 5 } \{1,4,5\} { 1 , 4 , 5 } , and { 2 , 4 , 6 } \{2,4,6\} { 2 , 4 , 6 } . The only possible additional triple is { 3 , 5 , 6 } \{3,5,6\} { 3 , 5 , 6 } .
Every such collection is realized by a configuration in X c o m X_{\mathrm{com}} X com . Up to relabeling, the constructions and their dimensions before quotienting by PGL 3 \operatorname{PGL}_3 PGL 3 are as follows.
For one collinear triple, choose a line, three distinct points on it, and three further general points. The dimension is 2 + 3 + 6 = 11 2+3+6=11 2 + 3 + 6 = 11 .
For two collinear triples, choose two distinct lines and use their intersection as the common labeled point. Choose two further points on each line and one further general point in P 2 \mathbb{P}^2 P 2 . The dimension is 4 + 4 + 2 = 10 4+4+2=10 4 + 4 + 2 = 10 .
For three collinear triples, choose three lines forming a triangle. Use their vertices and one further point on each side. Choose the three additional points to be noncollinear. The dimension is 6 + 3 = 9 6+3=9 6 + 3 = 9 .
For four collinear triples, take the six intersection points of four lines in general position. The dimension is 8 8 8 .
In each case, the requirement that no additional triple be collinear is an open condition. Since these configurations are stable and dim PGL 3 = 8 \dim\operatorname{PGL}_3=8 dim PGL 3 = 8 , the locus with precisely a prescribed collection of r r r collinear triples has dimension 4 − r 4-r 4 − r .
Consider six distinct points such that p 1 , p 2 , p 3 p_1,p_2,p_3 p 1 , p 2 , p 3 lie on one line and p 4 , p 5 , p 6 p_4,p_5,p_6 p 4 , p 5 , p 6 lie on another, with none of the points at the intersection of the two lines. These configurations are stable for the GIT quotient of six ordered points in P 2 \mathbb{P}^2 P 2 . Their images form a locus of dimension 4 + 6 − 8 = 2 4+6-8=2 4 + 6 − 8 = 2 in the GIT quotient.
On the blow-up, the strict transforms of the two lines are ( − 2 ) (-2) ( − 2 ) -curves meeting once. Their contraction produces an A 2 A_2 A 2 singularity. Thus these configurations do not give nodal cubic surfaces, which explains their exclusion from X c o m X_{\mathrm{com}} X com .
There are ten such loci, indexed by partitions of the six labels into two triples. Under the resolved Coble map, the proper transform of each closure is contracted to a cusp of Y m c u b Y_{\mathrm{mcub}} Y mcub [11 Section 2] .
The comparison is summarized by the following diagram.
M c u b m → X ( 3 , 6 ) → Φ M S Y ∼ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) ↓ ↓ ↓ Y m c u b = ( B 4 / Γ m ) ∗ → Φ C o b − 1 X ( 3 , 6 ) G I T → ∼ ( D A 1 / Γ B 1 ) ∗ \begin{CD}
\mathcal{M}_{\mathrm{cub}}^{m} @>>> X(3,6) @>{\Phi_{\mathrm{MSY}}}_{\sim}>> \mathcal{D}_{A_1}/\Gamma_{B_1}\setminus\mathcal{H}_{B_1}^{(2)}(\Gamma_{B_1}) \\
@VVV @VVV @VVV \\
Y_{\mathrm{mcub}}=(\mathbb{B}^{4}/\Gamma_m)_* @>{\Phi_{\mathrm{Cob}}^{-1}}>> X(3,6)^{\mathrm{GIT}} @>{\sim}>> (\mathcal{D}_{A_1}/\Gamma_{B_1})_*
\end{CD} M cub m ↓ ⏐ Y mcub = ( B 4 / Γ m ) ∗ Φ Cob − 1 X ( 3 , 6 ) ↓ ⏐ X ( 3 , 6 ) GIT Φ MSY ∼ ∼ D A 1 / Γ B 1 ∖ H B 1 ( 2 ) ( Γ B 1 ) ↓ ⏐ ( D A 1 / Γ B 1 ) ∗
The restriction of the Coble map Φ C o b \Phi_{\mathrm{Cob}} Φ Cob in (46) to the closure of the conic locus in X ( 3 , 6 ) G I T X(3,6)^{\mathrm{GIT}} X ( 3 , 6 ) GIT gives the classical dual map from the Igusa quartic to the Segre cubic [11 Section 2] . Kondō constructed an automorphic realization of the dual map in the opposite direction, from the Segre cubic to the Igusa quartic, using Borcherds’ theory [19 ] . An explicit automorphic description of the Coble map would require formulas for a linear system of modular forms for Γ B 1 \Gamma_{B_1} Γ B 1 on D A 1 \mathcal{D}_{A_1} D A 1 , together with its weight, character, and base locus.
A related direction concerns the comparison of toroidal compactifications. In the Segre–Igusa setting, the relevant Siegel and ball quotients admit isomorphic toroidal compactifications; see [18 ] or [17 Theorem 2.4.11] . In the present setting, Naruki’s cross-ratio variety is a smooth projective compactification of the moduli space of marked cubic surfaces whose boundary is a divisor with normal crossings [34 ] . It is isomorphic to the toroidal compactification of the ball quotient B 4 / Γ m \mathbb{B}^{4}/\Gamma_m B 4 / Γ m for marked cubic surfaces [12 ] . It is therefore natural to investigate how the Coble map relates Naruki’s compactification to toroidal compactifications of D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 associated with suitable admissible fans.
Three double modular coverings of P ( 2 , 3 , 4 , 5 , 6 ) \mathbb{P}(2,3,4,5,6) P ( 2 , 3 , 4 , 5 , 6 )
We use the two actions S 6 p t s S_{6}^{\mathrm{pts}} S 6 pts and S 6 f i x S_{6}^{\mathrm{fix}} S 6 fix fixed in § 2.2 .
Recall also that K B = T − 1 ( Γ B 1 ) K_{B}=T^{-1}(\Gamma_{B_{1}}) K B = T − 1 ( Γ B 1 ) , and let
ρ B : O + ( A 1 ) ⟶ O + ( B 1 ) / Γ B 1 , γ ⟼ T ( γ ) Γ B 1 . \rho_{B}:\mathrm{O}^{+}(A_{1})\longrightarrow\mathrm{O}^{+}(B_{1})/\Gamma_{B_{1}},\qquad\gamma\longmapsto T(\gamma)\Gamma_{B_{1}}. ρ B : O + ( A 1 ) ⟶ O + ( B 1 ) / Γ B 1 , γ ⟼ T ( γ ) Γ B 1 .
Thus ker ( ρ B ) = K B \ker(\rho_{B})=K_{B} ker ( ρ B ) = K B . By Theorem 5.1 , the extended period mapping gives an isomorphism
X ‾ ( 3 , 6 ) G I T ⟶ ∼ ( D A 1 / K B ) ∗ . \overline{X}(3,6)^{\mathrm{GIT}}\overset{\sim}{\longrightarrow}(\mathcal{D}_{A_{1}}/K_{B})^{*}. X ( 3 , 6 ) GIT ⟶ ∼ ( D A 1 / K B ) ∗ .
Recall that τ ˉ \bar{\tau} τ ˉ is the class of the involution τ ∈ O + ( B 1 ) \tau\in\mathrm{O}^{+}(B_{1}) τ ∈ O + ( B 1 ) in O + ( B 1 ) / Γ B 1 \mathrm{O}^{+}(B_{1})/\Gamma_{B_{1}} O + ( B 1 ) / Γ B 1 . Under the period isomorphism, τ ˉ \bar{\tau} τ ˉ acts as the Gale involution ( z ; x 1 , … , x 6 ) ↦ ( − z ; x 1 , … , x 6 ) (z;x_{1},\ldots,x_{6})\mapsto(-z;x_{1},\ldots,x_{6}) ( z ; x 1 , … , x 6 ) ↦ ( − z ; x 1 , … , x 6 ) . Its quotient is P 4 \mathbb{P}^{4} P 4 , and the double covering is branched along the Igusa quartic.
The finite quotient in (40) acts on the Coble fourfold. In terms of the Coble coordinates, it has the decomposition O + ( B 1 ) / Γ B 1 ≅ S 6 f i x × ⟨ τ ˉ ⟩ \mathrm{O}^{+}(B_{1})/\Gamma_{B_{1}}\cong S_{6}^{\mathrm{fix}}\times\langle\bar{\tau}\rangle O + ( B 1 ) / Γ B 1 ≅ S 6 fix × ⟨ τ ˉ ⟩ . The corresponding arithmetic subgroups of O + ( A 1 ) \mathrm{O}^{+}(A_{1}) O + ( A 1 ) will be obtained by taking inverse images under ρ B \rho_{B} ρ B . The other quotient O + ( A 1 ) / Γ A 1 \mathrm{O}^{+}(A_{1})/\Gamma_{A_{1}} O + ( A 1 ) / Γ A 1 is generated by the class of the involution σ \sigma σ of Remark 3.7.
Proposition 5.6. The quotient X ‾ ( 3 , 6 ) G I T / S 6 f i x \overline{X}(3,6)^{\mathrm{GIT}}/S_{6}^{\mathrm{fix}} X ( 3 , 6 ) GIT / S 6 fix is isomorphic to P ( 2 , 2 , 3 , 5 , 6 ) \mathbb{P}(2,2,3,5,6) P ( 2 , 2 , 3 , 5 , 6 ) . With Δ 60 \Delta_{60} Δ 60 as in § 6.1 , the natural quotient X ‾ ( 3 , 6 ) G I T / S 6 p t s \overline{X}(3,6)^{\mathrm{GIT}}/S_{6}^{\mathrm{pts}} X ( 3 , 6 ) GIT / S 6 pts is the projective spectrum of
C [ s 2 , s 3 , s 4 , s 5 , s 6 , w 34 ] / ( w 34 2 − ( 4 s 4 − s 2 2 ) Δ 60 ) . \mathbb{C}[s_{2},s_{3},s_{4},s_{5},s_{6},w_{34}]/\left(w_{34}^{2}-(4s_{4}-s_{2}^{2})\Delta_{60}\right). C [ s 2 , s 3 , s 4 , s 5 , s 6 , w 34 ] / ( w 34 2 − ( 4 s 4 − s 2 2 ) Δ 60 ) .
Proof. Set S = C [ x 1 , … , x 6 ] / ( s 1 ) S=\mathbb{C}[x_{1},\ldots,x_{6}]/(s_{1}) S = C [ x 1 , … , x 6 ] / ( s 1 ) . The homogeneous coordinate ring of the Coble fourfold is R = S ⊕ z S R=S\oplus zS R = S ⊕ z S , with z 2 = 4 s 4 − s 2 2 z^{2}=4s_{4}-s_{2}^{2} z 2 = 4 s 4 − s 2 2 . For the action fixing z z z , the invariants are generated by z , s 2 , s 3 , s 4 , s 5 , s 6 z,s_{2},s_{3},s_{4},s_{5},s_{6} z , s 2 , s 3 , s 4 , s 5 , s 6 with that relation. Eliminating s 4 s_{4} s 4 gives C [ z , s 2 , s 3 , s 5 , s 6 ] \mathbb{C}[z,s_{2},s_{3},s_{5},s_{6}] C [ z , s 2 , s 3 , s 5 , s 6 ] and proves the first assertion.
Put Δ ~ 30 = ∏ i < j ( x i − x j ) \widetilde{\Delta}_{30}=\prod_{i<j}(x_{i}-x_{j}) Δ 30 = ∏ i < j ( x i − x j ) , so that Δ ~ 30 2 = Δ 60 \widetilde{\Delta}_{30}^{2}=\Delta_{60} Δ 30 2 = Δ 60 . For the action induced by permutations of the six points, the invariants in z S zS z S are z z z times alternating polynomials in the x i x_{i} x i , since the outer automorphism relating the labels preserves parity. Every alternating polynomial is Δ ~ 30 \widetilde{\Delta}_{30} Δ 30 times a symmetric polynomial. Consequently the invariant algebra is C [ s 2 , … , s 6 ] ⊕ z Δ ~ 30 C [ s 2 , … , s 6 ] \mathbb{C}[s_{2},\ldots,s_{6}]\oplus z\widetilde{\Delta}_{30}\mathbb{C}[s_{2},\ldots,s_{6}] C [ s 2 , … , s 6 ] ⊕ z Δ 30 C [ s 2 , … , s 6 ] . Setting w 34 = z Δ ~ 30 w_{34}=z\widetilde{\Delta}_{30} w 34 = z Δ 30 proves the second assertion. □ \square □
In the above proposition, deg s i = i \deg s_{i}=i deg s i = i and deg w 34 = 17 \deg w_{34}=17 deg w 34 = 17 . This agrees with [37 Theorem 43] . We now compare the defining equations on the common quotient Y A 1 ≅ P 4 / S 6 Y_{A_{1}}\cong\mathbb{P}^{4}/S_{6} Y A 1 ≅ P 4 / S 6 . We use the graded period identification, under which homogeneous degree i i i in the Coble coordinates corresponds to modular weight 2 i 2i 2 i . The polynomials 4 s 4 − s 2 2 4s_{4}-s_{2}^{2} 4 s 4 − s 2 2 and v 8 v_{8} v 8 have modular weight 8 8 8 and define the same reduced divisor. Likewise, Δ ~ 30 2 \widetilde{\Delta}_{30}^{2} Δ 30 2 and d 60 ( v ) d_{60}(v) d 60 ( v ) have modular weight 60 60 60 and define the same reduced transposition branch divisor. For the latter assertion, let E i j E_{ij} E ij be the reduced inverse image of the hyperplane x i = x j x_{i}=x_{j} x i = x j in D A 1 / Γ B 1 \mathcal{D}_{A_{1}}/\Gamma_{B_{1}} D A 1 / Γ B 1 . The Coble double covering is unramified at a general point of E i j E_{ij} E ij , and Δ ~ 30 \widetilde{\Delta}_{30} Δ 30 vanishes there to order one. Thus, on this arithmetic quotient, div ( Δ ~ 30 ) = ∑ i < j E i j \operatorname{div}(\widetilde{\Delta}_{30})=\sum_{i<j}E_{ij} div ( Δ 30 ) = ∑ i < j E ij and div ( Δ 60 ) = 2 ∑ i < j E i j \operatorname{div}(\Delta_{60})=2\sum_{i<j}E_{ij} div ( Δ 60 ) = 2 ∑ i < j E ij . On P 4 / S 6 \mathbb{P}^{4}/S_{6} P 4 / S 6 , a local parameter transverse to the transposition branch divisor is t 2 t^{2} t 2 , where t = x i − x j t=x_{i}-x_{j} t = x i − x j is a transverse parameter before taking the quotient. Consequently, Δ 60 \Delta_{60} Δ 60 defines the reduced transposition branch divisor on P 4 / S 6 \mathbb{P}^{4}/S_{6} P 4 / S 6 .
Consequently, the ratios ( 4 s 4 − s 2 2 ) / v 8 (4s_{4}-s_{2}^{2})/v_{8} ( 4 s 4 − s 2 2 ) / v 8 and Δ ~ 30 2 / d 60 ( v ) \widetilde{\Delta}_{30}^{2}/d_{60}(v) Δ 30 2 / d 60 ( v ) are rational functions on Y A 1 Y_{A_{1}} Y A 1 with no zeros or poles in codimension one on its open modular quotient. The Baily–Borel boundary has codimension at least two. Normality therefore implies that both ratios extend to invertible regular functions on Y A 1 Y_{A_{1}} Y A 1 . They are constant because Y A 1 Y_{A_{1}} Y A 1 is projective.
After compatible scalar normalizations and choices of square roots, z z z and Δ ~ 30 \widetilde{\Delta}_{30} Δ 30 correspond to v ~ 4 \widetilde{v}_{4} v 4 and v ~ 30 \widetilde{v}_{30} v 30 , respectively. The characters χ 4 \chi_{4} χ 4 and χ 30 \chi_{30} χ 30 are trivial on K B K_{B} K B . Indeed, Γ B 1 \Gamma_{B_{1}} Γ B 1 is generated by the ( − 2 ) (-2) ( − 2 ) -reflections described in [33 Proposition 4.1 and its proof] . Under T T T , these correspond to reflections in primitive vectors of A 1 A_{1} A 1 of square − 4 -4 − 4 and divisibility two. Neither v ~ 4 \widetilde{v}_{4} v 4 nor v ~ 30 \widetilde{v}_{30} v 30 vanishes at a general point of these mirrors. Their characters are therefore trivial on these reflections, and hence on K B K_B K B . Thus both characters factor through ρ B \rho_B ρ B . Using the preceding comparison with z z z and Δ ~ 30 \widetilde{\Delta}_{30} Δ 30 , we obtain χ 4 ( γ ) = ε \chi_4(\gamma)=\varepsilon χ 4 ( γ ) = ε and χ 30 ( γ ) = sgn ( g ) \chi_{30}(\gamma)=\operatorname{sgn}(g) χ 30 ( γ ) = sgn ( g ) if ρ B ( γ ) = ( g , ε ) ∈ S 6 f i x × { ± 1 } \rho_B(\gamma)=(g,\varepsilon)\in S_6^{\mathrm{fix}}\times\{\pm1\} ρ B ( γ ) = ( g , ε ) ∈ S 6 fix × { ± 1 } .
Let A 6 f i x ⊂ S 6 f i x A_6^{\mathrm{fix}}\subset S_6^{\mathrm{fix}} A 6 fix ⊂ S 6 fix be the alternating subgroup. The formula for χ 30 \chi_{30} χ 30 gives ker ( χ 30 ) = ρ B − 1 ( A 6 f i x × { ± 1 } ) \ker(\chi_{30})=\rho_B^{-1}(A_6^{\mathrm{fix}}\times\{\pm1\}) ker ( χ 30 ) = ρ B − 1 ( A 6 fix × { ± 1 }) . Since the quotient of the Coble fourfold by the Gale involution is P 4 \mathbb{P}^4 P 4 , it follows that
Y χ 30 ≅ X ( 3 , 6 ) G I T / ( A 6 f i x × ⟨ τ ˉ ⟩ ) ≅ P 4 / A 6 Y_{\chi_{30}}\cong X(3,6)^{\mathrm{GIT}}/(A_6^{\mathrm{fix}}\times\langle\bar{\tau}\rangle)\cong\mathbb{P}^4/A_6 Y χ 30 ≅ X ( 3 , 6 ) GIT / ( A 6 fix × ⟨ τ ˉ ⟩) ≅ P 4 / A 6
over Y A 1 ≅ P 4 / S 6 Y_{A_1}\cong\mathbb{P}^4/S_6 Y A 1 ≅ P 4 / S 6 . Thus this covering distinguishes even and odd permutations of the Coble coordinates.
The action induced by permutations of the six points corresponds to the subgroup S 6 p t s = { ( g , sgn ( g ) ) ∣ g ∈ S 6 f i x } S_6^{\mathrm{pts}}=\{(g,\operatorname{sgn}(g))\mid g\in S_6^{\mathrm{fix}}\} S 6 pts = {( g , sgn ( g )) ∣ g ∈ S 6 fix } of this finite quotient. Consequently, Γ χ 4 = ρ B − 1 ( S 6 f i x × { 1 } ) \Gamma_{\chi_4}=\rho_B^{-1}(S_6^{\mathrm{fix}}\times\{1\}) Γ χ 4 = ρ B − 1 ( S 6 fix × { 1 }) and Γ p t s : = ρ B − 1 ( S 6 p t s ) = ker ( χ 4 χ 30 ) \Gamma_{\mathrm{pts}}:=\rho_B^{-1}(S_6^{\mathrm{pts}})=\ker(\chi_4\chi_{30}) Γ pts := ρ B − 1 ( S 6 pts ) = ker ( χ 4 χ 30 ) . In particular, K B ⊂ Γ p t s K_B\subset\Gamma_{\mathrm{pts}} K B ⊂ Γ pts and Γ p t s / K B ≅ S 6 p t s \Gamma_{\mathrm{pts}}/K_B\cong S_6^{\mathrm{pts}} Γ pts / K B ≅ S 6 pts . Thus X ( 3 , 6 ) G I T / S 6 p t s ≅ ( D A 1 / Γ p t s ) ∗ X(3,6)^{\mathrm{GIT}}/S_6^{\mathrm{pts}}\cong(\mathcal{D}_{A_1}/\Gamma_{\mathrm{pts}})^* X ( 3 , 6 ) GIT / S 6 pts ≅ ( D A 1 / Γ pts ) ∗ over Y A 1 Y_{A_1} Y A 1 . The covering is the projective spectrum of C [ v 4 , v 6 , v 8 , v 10 , v 12 , w 34 ] \mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},w_{34}] C [ v 4 , v 6 , v 8 , v 10 , v 12 , w 34 ] with relation w 34 2 = v 8 d 60 ( v ) w_{34}^2=v_8d_{60}(v) w 34 2 = v 8 d 60 ( v ) , where w 34 = v ~ 4 v ~ 30 w_{34}=\widetilde{v}_4\widetilde{v}_{30} w 34 = v 4 v 30 has modular weight 34. Its divisorial branch locus is { v 8 d 60 ( v ) = 0 } r e d \{v_8d_{60}(v)=0\}_{\mathrm{red}} { v 8 d 60 ( v ) = 0 } red . The same comparison identifies the quotient by the action fixing z z z with the covering for χ 4 \chi_4 χ 4 over this base.
The covering associated with Γ A 1 \Gamma_{A_1} Γ A 1 is instead the covering for χ 10 \chi_{10} χ 10 , with branch equation v 10 2 − 4 v 8 v 12 = 0 v_{10}^2-4v_8v_{12}=0 v 10 2 − 4 v 8 v 12 = 0 . Theorem 5.7 (2) gives its presentation. Our coordinates v 4 , v 6 , v 8 , v 10 , v ~ 10 , v 12 v_4,v_6,v_8,v_{10},\widetilde{v}_{10},v_{12} v 4 , v 6 , v 8 , v 10 , v 10 , v 12 are the coordinates t 4 , t 6 , t 8 , t 10 , s 10 , t 12 t_4,t_6,t_8,t_{10},s_{10},t_{12} t 4 , t 6 , t 8 , t 10 , s 10 , t 12 , respectively, of [33 Theorem 2.1] . These group comparisons do not assert T − 1 ( Γ B 1 ) ⊂ Γ A 1 T^{-1}(\Gamma_{B_1})\subset\Gamma_{A_1} T − 1 ( Γ B 1 ) ⊂ Γ A 1 . Indeed, T ( σ ) ∈ Γ B 1 T(\sigma)\in\Gamma_{B_1} T ( σ ) ∈ Γ B 1 , whereas χ 10 ( σ ) = − 1 \chi_{10}(\sigma)=-1 χ 10 ( σ ) = − 1 , as established in Section 3.5.
The subgroup Γ S 6 \Gamma_{S_6} Γ S 6 corresponds to the permutations of the six exceptional classes. Equivalently, S 6 ⊂ W ( E 6 ) S_6\subset W(E_6) S 6 ⊂ W ( E 6 ) is the Weyl group of the root subsystem { E i − E j } \{E_i-E_j\} { E i − E j } of type A 5 A_5 A 5 [11 Section 2.13] . It gives finite morphisms
B 4 / Γ m ⟶ B 4 / Γ S 6 ⟶ B 4 / Γ u m . \mathbb{B}^4/\Gamma_m\longrightarrow\mathbb{B}^4/\Gamma_{S_6}\longrightarrow\mathbb{B}^4/\Gamma_{\mathrm{um}}. B 4 / Γ m ⟶ B 4 / Γ S 6 ⟶ B 4 / Γ um .
The image of the marked discriminant on B 4 / Γ S 6 \mathbb{B}^4/\Gamma_{S_6} B 4 / Γ S 6 has three irreducible components, namely the images of D c o n D_{\mathrm{con}} D con , D c o l D_{\mathrm{col}} D col , and D p a i r D_{\mathrm{pair}} D pair . Indeed, S 6 S_6 S 6 fixes the conic root and acts transitively on triples and pairs of indices. In the unmarked quotient, the image is irreducible because W ( E 6 ) W(E_6) W ( E 6 ) acts transitively on the roots.
The union of these three images is H ( Γ S 6 ) \mathcal{H}(\Gamma_{S_6}) H ( Γ S 6 ) . For smooth cubic surfaces, put
M c u b S 6 : = M c u b m / S 6 ≅ ( B 4 / Γ S 6 ) ∖ H ( Γ S 6 ) . \mathcal{M}_{\mathrm{cub}}^{S_6}:=\mathcal{M}_{\mathrm{cub}}^{\mathrm{m}}/S_6\cong(\mathbb{B}^4/\Gamma_{S_6})\setminus\mathcal{H}(\Gamma_{S_6}). M cub S 6 := M cub m / S 6 ≅ ( B 4 / Γ S 6 ) ∖ H ( Γ S 6 ) .
Let π p t s : D A 1 / Γ p t s → D A 1 / O + ( A 1 ) \pi_{\mathrm{pts}}:\mathcal{D}_{A_1}/\Gamma_{\mathrm{pts}}\to\mathcal{D}_{A_1}/O^+(A_1) π pts : D A 1 / Γ pts → D A 1 / O + ( A 1 ) be the natural projection. Set X p t s ∘ : = ( D A 1 / Γ p t s ) ∖ π p t s − 1 ( { v 10 2 − 4 v 8 v 12 = 0 } ) X_{\mathrm{pts}}^\circ:=(\mathcal{D}_{A_1}/\Gamma_{\mathrm{pts}})\setminus\pi_{\mathrm{pts}}^{-1}(\{v_{10}^2-4v_8v_{12}=0\}) X pts ∘ := ( D A 1 / Γ pts ) ∖ π pts − 1 ({ v 10 2 − 4 v 8 v 12 = 0 }) . Then X ( 3 , 6 ) / S 6 p t s ≅ X p t s ∘ X(3,6)/S_6^{\mathrm{pts}}\cong X_{\mathrm{pts}}^\circ X ( 3 , 6 ) / S 6 pts ≅ X pts ∘ and M c u b S 6 ≅ X p t s ∘ ∖ π p t s − 1 ( { v 8 = 0 } ) \mathcal{M}_{\mathrm{cub}}^{S_6}\cong X_{\mathrm{pts}}^\circ\setminus\pi_{\mathrm{pts}}^{-1}(\{v_8=0\}) M cub S 6 ≅ X pts ∘ ∖ π pts − 1 ({ v 8 = 0 }) . Let Φ ‾ C o b \overline{\Phi}_{\mathrm{Cob}} Φ Cob be the rational map induced by Φ C o b \Phi_{\mathrm{Cob}} Φ Cob on the quotients by S 6 p t s S_6^{\mathrm{pts}} S 6 pts . We have the following diagram.
M c u b S 6 → X ( 3 , 6 ) / S 6 p t s → ∼ X p t s ∘ ↓ ↓ ↓ ( B 4 / Γ S 6 ) ∗ → Φ ‾ C o b − 1 X ( 3 , 6 ) G I T / S 6 p t s → ∼ ( D A 1 / Γ p t s ) ∗ ↓ 2 : 1 Y A 1 \begin{CD}
\mathcal{M}_{\mathrm{cub}}^{S_6} @>>> X(3,6)/S_6^{\mathrm{pts}} @>\sim>> X_{\mathrm{pts}}^\circ\\
@VVV @VVV @VVV \\
(\mathbb{B}^4/\Gamma_{S_6})^* @>{\overline{\Phi}_{\mathrm{Cob}}^{-1}}>> X(3,6)^{\mathrm{GIT}}/S_6^{\mathrm{pts}} @>\sim>> (\mathcal{D}_{A_1}/\Gamma_{\mathrm{pts}})^* \\
@. @VV{2:1}V @. \\
@. Y_{A_1} @.
\end{CD} M cub S 6 ↓ ⏐ ( B 4 / Γ S 6 ) ∗ Φ Cob − 1 X ( 3 , 6 ) / S 6 pts ↓ ⏐ X ( 3 , 6 ) GIT / S 6 pts ↓ ⏐ 2 : 1 Y A 1 ∼ ∼ X pts ∘ ↓ ⏐ ( D A 1 / Γ pts ) ∗
Put δ σ = α 1 − α 2 \delta_\sigma=\alpha_1-\alpha_2 δ σ = α 1 − α 2 . Then δ σ 2 = − 4 \delta_\sigma^2=-4 δ σ 2 = − 4 and div A 1 ( δ σ ) = 2 \operatorname{div}_{A_1}(\delta_\sigma)=2 div A 1 ( δ σ ) = 2 . The image of its mirror is contained in H A 1 , div = 2 ( O + ( A 1 ) ) \mathcal{H}_{A_1,\operatorname{div}=2}(O^+(A_1)) H A 1 , div = 2 ( O + ( A 1 )) . The following theorem identifies the reduced divisorial branch loci of the three coverings with closures of Heegner divisors in Y A 1 Y_{A_1} Y A 1 .
Theorem 5.7. The characters χ k \chi_k χ k with k ∈ { 4 , 10 , 30 } k \in\{4,10,30\} k ∈ { 4 , 10 , 30 } define double coverings
( D A 1 / Γ χ k ) ∗ → ( D A 1 / O + ( A 1 ) ) ∗ . (\mathcal{D}_{A_1}/\Gamma_{\chi_k})^* \to(\mathcal{D}_{A_1}/O^+(A_1))^*. ( D A 1 / Γ χ k ) ∗ → ( D A 1 / O + ( A 1 ) ) ∗ . Moreover the following statements hold. All closures below are taken in Y A 1 Y_{A_1} Y A 1 . The branch statements concern the reduced divisorial branch loci.
(1) The quotient X ( 3 , 6 ) ‾ G I T / S 6 f i x \overline{X(3,6)}^{\mathrm{GIT}}/S_6^{\mathrm{fix}} X ( 3 , 6 ) GIT / S 6 fix is isomorphic to ( D A 1 / Γ χ 4 ) ∗ (\mathcal{D}_{A_1}/\Gamma_{\chi_4})^* ( D A 1 / Γ χ 4 ) ∗ :
( D A 1 / Γ χ 4 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 4 ] / ( v ~ 4 2 − v 8 ) ≅ P ( 2 , 2 , 3 , 5 , 6 ) . (\mathcal{D}_{A_1}/\Gamma_{\chi_4})^* \cong\operatorname{Proj}\mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_4]/(\widetilde{v}_4^2-v_8) \cong\mathbb{P}(2,2,3,5,6). ( D A 1 / Γ χ 4 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v 4 ] / ( v 4 2 − v 8 ) ≅ P ( 2 , 2 , 3 , 5 , 6 ) . The associated double covering is branched along H A 1 , d i v = 2 ( 2 ) ( O + ( A 1 ) ) = { v 8 = 0 } \mathcal{H}_{A_1,\mathrm{div}=2}^{(2)}(O^+(A_1))=\{v_8=0\} H A 1 , div = 2 ( 2 ) ( O + ( A 1 )) = { v 8 = 0 } .
(2) The arithmetic subgroup Γ χ 10 \Gamma_{\chi_{10}} Γ χ 10 coincides with Γ A 1 \Gamma_{A_1} Γ A 1 and the double covering is given by
( D A 1 / Γ χ 10 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 10 ] / ( v ~ 10 2 − v 10 2 + 4 v 8 v 12 ) . (\mathcal{D}_{A_1}/\Gamma_{\chi_{10}})^* \cong\operatorname{Proj}\mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_{10}]/(\widetilde{v}_{10}^2-v_{10}^2+4v_8v_{12}). ( D A 1 / Γ χ 10 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v 10 ] / ( v 10 2 − v 10 2 + 4 v 8 v 12 ) . It is branched along H A 1 , d i v = 2 ( 4 ) ( O + ( A 1 ) ) = { v 10 2 − 4 v 8 v 12 = 0 } \mathcal{H}_{A_1,\mathrm{div}=2}^{(4)}(O^+(A_1))=\{v_{10}^2-4v_8v_{12}=0\} H A 1 , div = 2 ( 4 ) ( O + ( A 1 )) = { v 10 2 − 4 v 8 v 12 = 0 } .
(3) The double covering associated with χ 30 \chi_{30} χ 30 is given by
( D A 1 / Γ χ 30 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v ~ 30 ] / ( v ~ 30 2 − d 60 ( v ) ) . (\mathcal{D}_{A_1}/\Gamma_{\chi_{30}})^* \cong\operatorname{Proj}\mathbb{C}[v_4,v_6,v_8,v_{10},v_{12},\widetilde{v}_{30}]/(\widetilde{v}_{30}^2-d_{60}(v)). ( D A 1 / Γ χ 30 ) ∗ ≅ Proj C [ v 4 , v 6 , v 8 , v 10 , v 12 , v 30 ] / ( v 30 2 − d 60 ( v )) . Its branch divisor is H A 1 , d i v = 1 ( 2 ) ( O + ( A 1 ) ) = { d 60 ( v ) = 0 } \mathcal{H}_{A_1,\mathrm{div}=1}^{(2)}(O^+(A_1))=\{d_{60}(v)=0\} H A 1 , div = 1 ( 2 ) ( O + ( A 1 )) = { d 60 ( v ) = 0 } .
Proof. For χ 4 \chi_4 χ 4 , the relation v ~ 4 2 = v 8 \widetilde{v}_4^2=v_8 v 4 2 = v 8 gives the branch divisor { v 8 = 0 } \{v_8=0\} { v 8 = 0 } . By the period and character comparison following Proposition 5.6 , this covering is identified with X ( 3 , 6 ) ‾ G I T / S 6 f i x \overline{X(3,6)}^{\mathrm{GIT}}/S_6^{\mathrm{fix}} X ( 3 , 6 ) GIT / S 6 fix over Y A 1 Y_{A_1} Y A 1 . Eliminating v 8 v_8 v 8 gives P ( 2 , 2 , 3 , 5 , 6 ) \mathbb{P}(2,2,3,5,6) P ( 2 , 2 , 3 , 5 , 6 ) . On D A 1 / Γ A 1 \mathcal{D}_{A_1}/\Gamma_{A_1} D A 1 / Γ A 1 , the reduced inverse image has the two components coming from H ∞ H_\infty H ∞ and H 0 H_0 H 0 . In the original coefficients of (18) , their equations are u 5 , 3 = 0 u_{5,3}=0 u 5 , 3 = 0 and u 3 , 5 = 0 u_{3,5}=0 u 3 , 5 = 0 , respectively. At a general point of either component, one of the two I I I ∗ III^* I I I ∗ fibers of (18) is enhanced to I I ∗ II^* I I ∗ , by Lemma 3.3 (3). After removing H ∞ H_\infty H ∞ , the remaining component has equation r 8 = 0 r_8=0 r 8 = 0 on R \mathcal{R} R . This gives the general degeneration from M 1 \mathbf{M}_1 M 1 -polarized K 3 K3 K 3 surfaces to M 2 \mathbf{M}_2 M 2 -polarized ones.
For χ 10 \chi_{10} χ 10 , the relation v ~ 10 2 = v 10 2 − 4 v 8 v 12 \widetilde{v}_{10}^2=v_{10}^2-4v_8v_{12} v 10 2 = v 10 2 − 4 v 8 v 12 shows that the branch divisor is { v 10 2 − 4 v 8 v 12 = 0 } \{v_{10}^2-4v_8v_{12}=0\} { v 10 2 − 4 v 8 v 12 = 0 } . Since v ~ 10 \widetilde{v}_{10} v 10 is invariant under Γ A 1 \Gamma_{A_1} Γ A 1 and changes sign under the nontrivial coset in O + ( A 1 ) / Γ A 1 ≅ Z / 2 Z O^+(A_1)/\Gamma_{A_1}\cong\mathbb{Z}/2\mathbb{Z} O + ( A 1 ) / Γ A 1 ≅ Z /2 Z , one has Γ χ 10 = Γ A 1 \Gamma_{\chi_{10}}=\Gamma_{A_1} Γ χ 10 = Γ A 1 . The intersection of this branch divisor with D A 1 / O + ( A 1 ) \mathcal{D}_{A_1}/O^+(A_1) D A 1 / O + ( A 1 ) is the image of the mirror D δ σ \mathcal{D}_{\delta_\sigma} D δ σ . Its reduced inverse image in D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 is H B 1 ( 2 ) ( Γ B 1 ) \mathcal{H}_{B_1}^{(2)}(\Gamma_{B_1}) H B 1 ( 2 ) ( Γ B 1 ) , with twenty labeled collinearity components.
For χ 30 \chi_{30} χ 30 , the relation v ~ 30 2 = d 60 ( v ) \widetilde{v}_{30}^2=d_{60}(v) v 30 2 = d 60 ( v ) shows that the branch divisor is { d 60 ( v ) = 0 } \{d_{60}(v)=0\} { d 60 ( v ) = 0 } . On R \mathcal{R} R its reduced inverse image is { d 60 ( r ) = 0 } \{d_{60}(r)=0\} { d 60 ( r ) = 0 } , where at a general point two fibers of type I 1 I_1 I 1 coalesce into a fiber of type I 2 I_2 I 2 . Under the B 1 \mathbf{B}_1 B 1 -description this is the union of the fifteen transposition components, namely H B 1 , S 6 ( 4 ) ( Γ B 1 ) \mathcal{H}_{B_1,S_6}^{(4)}(\Gamma_{B_1}) H B 1 , S 6 ( 4 ) ( Γ B 1 ) .
The polynomial v 10 2 − 4 v 8 v 12 v_{10}^2-4v_8v_{12} v 10 2 − 4 v 8 v 12 is irreducible. The irreducibility and reducedness of d 60 ( v ) d_{60}(v) d 60 ( v ) follow from the independent discriminant calculation in Lemma A.2 . □
For reference, the correspondence of the three types is summarized below. The equations are on D A 1 / O + ( A 1 ) \mathcal{D}_{A_1}/O^+(A_1) D A 1 / O + ( A 1 ) , and the numbers of components refer to their reduced inverse images on D A 1 / Γ B 1 \mathcal{D}_{A_1}/\Gamma_{B_1} D A 1 / Γ B 1 . The normal vectors in the two columns are primitive in their respective lattices.
Table 1.
The corresponding characters are χ 4 \chi_4 χ 4 , χ 10 \chi_{10} χ 10 , and χ 30 \chi_{30} χ 30 , respectively.
Appendix A. On the factor d 60 d_{60} d 60
We compute the residual discriminant δ 60 ( u ) \delta_{60}(u) δ 60 ( u ) of Lemma 3.3 in two systems of coordinates. Its expressions in the normalized coefficients of (28) and in the coordinates on the quotient by O + ( A 1 ) \mathrm{O}^{+}(A_{1}) O + ( A 1 ) introduced in Proposition 3.6 are d 60 ( r ) d_{60}(r) d 60 ( r ) and d 60 ( v ) d_{60}(v) d 60 ( v ) , respectively. The argument r r r or v v v specifies the coordinate system. With compatible normalizations, δ 60 ( u ) = d 60 ( r ( u ) ) = d 60 ( v ( u ) ) \delta_{60}(u)=d_{60}(r(u))=d_{60}(v(u)) δ 60 ( u ) = d 60 ( r ( u )) = d 60 ( v ( u )) on u 5 , 3 ≠ 0 u_{5,3}\ne0 u 5 , 3 = 0 . More explicitly, the substitution is d 60 ( r 2 , r 4 , r 6 , r 8 , r 10 ) = d 60 ( r 4 , r 6 , r 8 , r 10 + r 2 r 8 , r 2 r 10 ) d_{60}(r_{2},r_{4},r_{6},r_{8},r_{10})=d_{60}(r_{4},r_{6},r_{8},r_{10}+r_{2}r_{8},r_{2}r_{10}) d 60 ( r 2 , r 4 , r 6 , r 8 , r 10 ) = d 60 ( r 4 , r 6 , r 8 , r 10 + r 2 r 8 , r 2 r 10 ) , where the right-hand side is evaluated in the ordered variables ( v 4 , v 6 , v 8 , v 10 , v 12 ) (v_{4},v_{6},v_{8},v_{10},v_{12}) ( v 4 , v 6 , v 8 , v 10 , v 12 ) .
We prove the discriminant factorization in the normalized coordinates of (27) and derive a resultant formula for d 60 ( r ) d_{60}(r) d 60 ( r ) . Resultants and discriminants are computed in the polynomial coefficient rings, and the resulting identities specialize to individual parameter values. Put A r ( T ) = T 2 + r 4 T + r 8 A_{r}(T)=T^{2}+r_{4}T+r_{8} A r ( T ) = T 2 + r 4 T + r 8 , B r ( T ) = r 2 T 2 + r 6 T + r 10 B_{r}(T)=r_{2}T^{2}+r_{6}T+r_{10} B r ( T ) = r 2 T 2 + r 6 T + r 10 , and
P r ( T ) : = 4 A r ( T ) 3 + 27 T B r ( T ) 2 . (53) P_{r}(T):=4A_{r}(T)^{3}+27TB_{r}(T)^{2}. \tag*{(53)} P r ( T ) := 4 A r ( T ) 3 + 27 T B r ( T ) 2 . (53)
The singular fibers away from 0 0 0 and ∞ \infty ∞ occur at the zeros of P r ( T ) P_{r}(T) P r ( T ) . Indeed, after setting w = 1 w=1 w = 1 , the discriminant of the Weierstrass cubic in (27) is, up to a nonzero constant, 4 g 2 ( T ) 3 + 27 g 3 ( T ) 2 = T 9 P r ( T ) 4g_{2}(T)^{3}+27g_{3}(T)^{2}=T^{9}P_{r}(T) 4 g 2 ( T ) 3 + 27 g 3 ( T ) 2 = T 9 P r ( T ) . Also put ℓ 20 ( r ) : = Res T ( A r ( T ) , B r ( T ) ) \ell_{20}(r):=\operatorname{Res}_{T}(A_{r}(T),B_{r}(T)) ℓ 20 ( r ) := Res T ( A r ( T ) , B r ( T )) . Explicitly,
ℓ 20 ( r ) = r 10 2 + r 10 r 2 r 4 2 − 2 r 10 r 2 r 8 − r 10 r 4 r 6 + r 2 2 r 8 2 − r 2 r 4 r 6 r 8 + r 6 2 r 8 . (54) \ell_{20}(r)=r_{10}^{2}+r_{10}r_{2}r_{4}^{2}-2r_{10}r_{2}r_{8}-r_{10}r_{4}r_{6}+r_{2}^{2}r_{8}^{2}-r_{2}r_{4}r_{6}r_{8}+r_{6}^{2}r_{8}. \tag*{(54)} ℓ 20 ( r ) = r 10 2 + r 10 r 2 r 4 2 − 2 r 10 r 2 r 8 − r 10 r 4 r 6 + r 2 2 r 8 2 − r 2 r 4 r 6 r 8 + r 6 2 r 8 . (54)
The subscript indicates the weight with respect to (29) .
Definition A.1. Define the polynomial d 60 ( r ) d_{60}(r) d 60 ( r ) of weight 60, uniquely up to multiplication by a nonzero constant, by the factorization
Disc T ( P r ( T ) ) = const ⋅ ℓ 20 ( r ) 3 d 60 ( r ) . (55) \operatorname{Disc}_{T}(P_{r}(T))=\operatorname{const}\cdot\ell_{20}(r)^{3}d_{60}(r). \tag*{(55)} Disc T ( P r ( T )) = const ⋅ ℓ 20 ( r ) 3 d 60 ( r ) . (55)
Lemma A.2. For a weighted homogeneous polynomial f ( r ) f(r) f ( r ) , let Z ( f ) Z(f) Z ( f ) denote its zero locus in R \mathcal{R} R .
(1) One has
Z ( ℓ 20 ) = { [ r ] ∈ R ∣ ∃ a ∈ C × , A r ( a ) = B r ( a ) = 0 } . Z(\ell_{20})=\{[r]\in\mathcal{R}\mid\exists a\in\mathbb{C}^{\times},\ A_{r}(a)=B_{r}(a)=0\}. Z ( ℓ 20 ) = {[ r ] ∈ R ∣ ∃ a ∈ C × , A r ( a ) = B r ( a ) = 0 } . For a general point [ r ] ∈ Z ( ℓ 20 ) [r]\in Z(\ell_{20}) [ r ] ∈ Z ( ℓ 20 ) , the polynomials A r A_{r} A r and B r B_{r} B r have a unique common root a a a , and the fiber of the elliptic fibration over T = a T=a T = a is of Kodaira type II.
(2) One has
Z ( d 60 ) = { [ r ] ∈ R ∖ Z ( ℓ 20 ) ∣ P r ( a ) = P r ′ ( a ) = 0 for some a ∈ C × } R , Z(d_{60})=\{[r]\in\mathcal{R}\setminus Z(\ell_{20})\mid P_{r}(a)=P_{r}'(a)=0\text{ for some }a\in\mathbb{C}^{\times}\}^{\mathcal{R}}, Z ( d 60 ) = {[ r ] ∈ R ∖ Z ( ℓ 20 ) ∣ P r ( a ) = P r ′ ( a ) = 0 for some a ∈ C × } R , where the closure is taken in the Zariski topology on R \mathcal{R} R . At a general point of Z ( d 60 ) Z(d_{60}) Z ( d 60 ) , the polynomial P r P_{r} P r has exactly one root a ∈ C × a\in\mathbb{C}^{\times} a ∈ C × of multiplicity two, and all its other roots are simple. The fiber over T = a T=a T = a is of Kodaira type I 2 I_{2} I 2 .
(3) The polynomials d 60 ( r ) d_{60}(r) d 60 ( r ) and d 60 ( v ) d_{60}(v) d 60 ( v ) are irreducible in the respective rings C [ r 2 , r 4 , r 6 , r 8 , r 10 ] \mathbb{C}[r_{2},r_{4},r_{6},r_{8},r_{10}] C [ r 2 , r 4 , r 6 , r 8 , r 10 ] and C [ v 4 , v 6 , v 8 , v 10 , v 12 ] \mathbb{C}[v_{4},v_{6},v_{8},v_{10},v_{12}] C [ v 4 , v 6 , v 8 , v 10 , v 12 ] . Their zero loci in the respective affine coefficient spaces are irreducible reduced hypersurfaces.
Proof. Assigning weight 4 to T T T , the polynomials A r , B r , P r A_{r},B_{r},P_{r} A r , B r , P r have weights 8, 10, 24, respectively. The sextic P r P_{r} P r has constant leading coefficient 4. Its six roots have weight 4, so its discriminant has weight 6 ⋅ 5 ⋅ 4 = 120 6\cdot5\cdot4=120 6 ⋅ 5 ⋅ 4 = 120 . Since ℓ 20 \ell_{20} ℓ 20 has weight 20, the quotient in (A.3) has weight 60.
The resultant ℓ 20 \ell_{20} ℓ 20 vanishes precisely when A r A_{r} A r and B r B_{r} B r have a common root. The incidence variety of such roots is parametrized by a , r 2 , r 4 , r 6 a,r_{2},r_{4},r_{6} a , r 2 , r 4 , r 6 , with r 8 = − a 2 − r 4 a r_{8}=-a^{2}-r_{4}a r 8 = − a 2 − r 4 a and r 10 = − r 2 a 2 − r 6 a r_{10}=-r_{2}a^{2}-r_{6}a r 10 = − r 2 a 2 − r 6 a . Its projection to the coefficient space is finite, since a a a satisfies the equation A r ( a ) = 0 A_{r}(a)=0 A r ( a ) = 0 . Thus the resultant divisor has irreducible support. At a general point there is one common root, simple for both polynomials, and the resultant vanishes simply. Over R \mathcal{R} R this root is nonzero, since a common root at zero would require r 8 = r 10 = 0 r_{8}=r_{10}=0 r 8 = r 10 = 0 . There is no common root at infinity because A r A_{r} A r is monic. Since g 2 ( T ) = T 3 A r ( T ) g_{2}(T)=T^{3}A_{r}(T) g 2 ( T ) = T 3 A r ( T ) and g 3 ( T ) = T 5 B r ( T ) g_{3}(T)=T^{5}B_{r}(T) g 3 ( T ) = T 5 B r ( T ) , this also identifies the locus where g 2 g_{2} g 2 and g 3 g_{3} g 3 have a common zero away from 0 and ∞ \infty ∞ . We compute the multiplicity of this component of the discriminant. At a general point of it, let a 0 a_0 a 0 be the common root. Choose an analytic curve transverse to the resultant divisor. The simple root of B r B_r B r near a 0 a_0 a 0 varies analytically; write it as a ( r ) a(r) a ( r ) and use s = A r ( a ( r ) ) s=A_r(a(r)) s = A r ( a ( r )) as the parameter on the curve. The function y = B r ( T ) y=B_r(T) y = B r ( T ) is a local coordinate near a 0 a_0 a 0 . Writing r = r ( s ) r=r(s) r = r ( s ) and T = T ( y , s ) T=T(y,s) T = T ( y , s ) , we have
A r ( s ) ( T ( y , s ) ) = s + c y + O ( y 2 , s y ) , T ( y , s ) = a 0 + O ( y , s ) , A_{r(s)}(T(y,s))=s+cy+O(y^2,sy),\qquad T(y,s)=a_0+O(y,s), A r ( s ) ( T ( y , s )) = s + cy + O ( y 2 , sy ) , T ( y , s ) = a 0 + O ( y , s ) ,
where a 0 c ≠ 0 a_0c\ne0 a 0 c = 0 . Put F ( y , s ) = P r ( s ) ( T ( y , s ) ) F(y,s)=P_{r(s)}(T(y,s)) F ( y , s ) = P r ( s ) ( T ( y , s )) . Then F y y ( 0 , 0 ) = 54 a 0 ≠ 0 F_{yy}(0,0)=54a_0\ne0 F y y ( 0 , 0 ) = 54 a 0 = 0 and F y ( 0 , s ) = O ( s 2 ) F_y(0,s)=O(s^2) F y ( 0 , s ) = O ( s 2 ) . The equation F y = 0 F_y=0 F y = 0 therefore has a unique solution y = y ( s ) = O ( s 2 ) y=y(s)=O(s^2) y = y ( s ) = O ( s 2 ) near the origin, and F ( y ( s ) , s ) = 4 s 3 + O ( s 4 ) F(y(s),s)=4s^3+O(s^4) F ( y ( s ) , s ) = 4 s 3 + O ( s 4 ) . By the holomorphic Morse lemma with parameters, the local discriminant is a unit times this critical value. The other roots of P r P_r P r are simple at a general point; for example, this holds when A r = ( T − 1 ) ( T − 2 ) A_r=(T-1)(T-2) A r = ( T − 1 ) ( T − 2 ) and B r = T − 1 B_r=T-1 B r = T − 1 . Consequently, ℓ 20 \ell_{20} ℓ 20 occurs with multiplicity three in (A.3). At the common root, the vanishing orders of g 2 , g 3 g_2,g_3 g 2 , g 3 and the elliptic discriminant are ( 1 , 1 , 2 ) (1,1,2) ( 1 , 1 , 2 ) , which gives type II.
Consider the incidence variety
I = { ( r , a ) ∈ A 5 × C × ∣ P r ( a ) = P r ′ ( a ) = 0 , A r ( a ) B r ( a ) ≠ 0 } . I=\{(r,a)\in\mathbb{A}^5\times\mathbb{C}^{\times}\mid P_r(a)=P'_r(a)=0,\ A_r(a)B_r(a)\ne0\}. I = {( r , a ) ∈ A 5 × C × ∣ P r ( a ) = P r ′ ( a ) = 0 , A r ( a ) B r ( a ) = 0 } .
Put u = A r ( a ) u=A_r(a) u = A r ( a ) and v = B r ( a ) v=B_r(a) v = B r ( a ) . The equation P r ( a ) = 0 P_r(a)=0 P r ( a ) = 0 gives a = − 4 u 3 / ( 27 v 2 ) a=-4u^3/(27v^2) a = − 4 u 3 / ( 27 v 2 ) , and P r ′ ( a ) = 0 P'_r(a)=0 P r ′ ( a ) = 0 determines b = B r ′ ( a ) b=B'_r(a) b = B r ′ ( a ) by 12 u 2 ( 2 a + r 4 ) + 27 v 2 + 54 a v b = 0 12u^2(2a+r_4)+27v^2+54avb=0 12 u 2 ( 2 a + r 4 ) + 27 v 2 + 54 a v b = 0 . The coefficients are recovered from
r 8 = u − a 2 − r 4 a , r 6 = b − 2 r 2 a , r 10 = v − r 2 a 2 − r 6 a . r_8=u-a^2-r_4a,\qquad r_6=b-2r_2a,\qquad r_{10}=v-r_2a^2-r_6a. r 8 = u − a 2 − r 4 a , r 6 = b − 2 r 2 a , r 10 = v − r 2 a 2 − r 6 a .
These formulas give an isomorphism I ≅ ( C × ) 2 × A 2 I\cong(\mathbb{C}^{\times})^2\times\mathbb{A}^2 I ≅ ( C × ) 2 × A 2 , with parameters u , v , r 4 , r 2 u,v,r_4,r_2 u , v , r 4 , r 2 . The projection of I I I to the coefficient space is quasi-finite, since a a a is a root of the nonzero sextic P r P_r P r . Its image therefore has irreducible closure of dimension four.
Outside the resultant divisor, every multiple root of P r P_r P r belongs to this incidence variety. Indeed, a multiple root at zero would require r 8 = r 10 = 0 r_8=r_{10}=0 r 8 = r 10 = 0 and hence ℓ 20 = 0 \ell_{20}=0 ℓ 20 = 0 ; no root at infinity occurs because the leading coefficient is constant. Thus the closure of the image of I I I is the only residual discriminant component. For r = ( 0 , 0 , − 3 , − 4 , 5 ) r=(0,0,-3,-4,5) r = ( 0 , 0 , − 3 , − 4 , 5 ) , one has ℓ 20 ( r ) = − 11 \ell_{20}(r)=-11 ℓ 20 ( r ) = − 11 and
P r ( T ) = ( T − 1 ) 2 Q ( T ) , Q ( T ) = 4 T 4 + 8 T 3 − 36 T 2 + 163 T − 256. P_r(T)=(T-1)^2Q(T),\qquad Q(T)=4T^4+8T^3-36T^2+163T-256. P r ( T ) = ( T − 1 ) 2 Q ( T ) , Q ( T ) = 4 T 4 + 8 T 3 − 36 T 2 + 163 T − 256.
Here Q ( 1 ) = − 117 Q(1)=-117 Q ( 1 ) = − 117 and gcd ( Q , Q ′ ) = 1 \gcd(Q,Q')=1 g cd( Q , Q ′ ) = 1 . This gives a point of the residual component at which P r P_r P r has exactly one double root and all other roots are simple. At such a double root a a a , the normal direction to the discriminant hypersurface of sextics is measured by the variation of P r ( a ) P_r(a) P r ( a ) . Since ∂ P r ( a ) / ∂ r 10 = 54 a B r ( a ) ≠ 0 \partial P_r(a)/\partial r_{10}=54aB_r(a)\ne0 ∂ P r ( a ) / ∂ r 10 = 54 a B r ( a ) = 0 , our family is transverse to that hypersurface. The residual component therefore has multiplicity one, so d 60 ( r ) d_{60}(r) d 60 ( r ) is irreducible and its zero divisor is reduced. At a general point of this divisor, the elliptic discriminant has order two at the double root and the Weierstrass coefficients do not vanish there, giving type I 2 I_2 I 2 . This proves (1), (2), and the assertion about d 60 ( r ) d_{60}(r) d 60 ( r ) in (3).
Finally, the substitution from r r r to v v v at the beginning of this appendix is dominant and preserves the positive weights. Indeed, for v 8 ≠ 0 v_8\ne0 v 8 = 0 one can solve v 8 r 2 2 − v 10 r 2 + v 12 = 0 v_8r_2^2-v_{10}r_2+v_{12}=0 v 8 r 2 2 − v 10 r 2 + v 12 = 0 , set r 10 = v 10 − r 2 v 8 r_{10}=v_{10}-r_2v_8 r 10 = v 10 − r 2 v 8 , and take ( r 4 , r 6 , r 8 ) = ( v 4 , v 6 , v 8 ) (r_4,r_6,r_8)=(v_4,v_6,v_8) ( r 4 , r 6 , r 8 ) = ( v 4 , v 6 , v 8 ) . A nontrivial factorization of d 60 ( v ) d_{60}(v) d 60 ( v ) could be chosen with weighted homogeneous factors of positive weight. Dominance and preservation of weights would make their pullbacks nonconstant factors of d 60 ( r ) d_{60}(r) d 60 ( r ) , a contradiction. Thus d 60 ( v ) d_{60}(v) d 60 ( v ) is irreducible and its zero divisor is reduced.
Proposition A.3. Define the polynomial
C r ( T ) : = 3 T A r ′ ( T ) B r ( T ) − A r ( T ) B r ( T ) − 2 T A r ( T ) B r ′ ( T ) C_r(T):=3T A'_r(T)B_r(T)-A_r(T)B_r(T)-2T A_r(T)B'_r(T) C r ( T ) := 3 T A r ′ ( T ) B r ( T ) − A r ( T ) B r ( T ) − 2 T A r ( T ) B r ′ ( T ) = r 2 T 4 + ( − 2 r 2 r 4 + 3 r 6 ) T 3 + ( 5 r 10 − 5 r 2 r 8 ) T 2 + ( 2 r 10 r 4 − 3 r 6 r 8 ) T − r 10 r 8 . (56) =r_2T^4+(-2r_2r_4+3r_6)T^3+(5r_{10}-5r_2r_8)T^2+(2r_{10}r_4-3r_6r_8)T-r_{10}r_8. \tag*{(56)} = r 2 T 4 + ( − 2 r 2 r 4 + 3 r 6 ) T 3 + ( 5 r 10 − 5 r 2 r 8 ) T 2 + ( 2 r 10 r 4 − 3 r 6 r 8 ) T − r 10 r 8 . (56) Then one has the polynomial identity
r 8 Disc T ( P r ( T ) ) = − 2 4 3 12 ℓ 20 ( r ) Res T ( P r ( T ) , C r ( T ) ) . (57) r_8\operatorname{Disc}_T(P_r(T))=-2^4 3^{12}\ell_{20}(r)\operatorname{Res}_T(P_r(T),C_r(T)). \tag*{(57)} r 8 Disc T ( P r ( T )) = − 2 4 3 12 ℓ 20 ( r ) Res T ( P r ( T ) , C r ( T )) . (57)
Consequently,
d 60 ( r ) = const ⋅ Res T ( P r ( T ) , C r ( T ) ) r 8 ℓ 20 ( r ) 2 , (58) d_{60}(r) = \operatorname{const}\cdot\frac{\operatorname{Res}_{T}(P_r(T),C_r(T))}{r^{8}\ell_{20}(r)^{2}}, \tag*{(58)} d 60 ( r ) = const ⋅ r 8 ℓ 20 ( r ) 2 Res T ( P r ( T ) , C r ( T )) , (58)
where the right-hand side is a polynomial of weight 60. Equivalently, d 60 ( r ) d_{60}(r) d 60 ( r ) is, up to a nonzero constant, the determinant of the Bezout matrix of P r ( T ) P_r(T) P r ( T ) and C r ( T ) C_r(T) C r ( T ) divided by r 8 ℓ 20 ( r ) 2 r^{8}\ell_{20}(r)^{2} r 8 ℓ 20 ( r ) 2 .
Proof. Modulo P r ( T ) P_r(T) P r ( T ) one has
T B r ( T ) P r ′ ( T ) ≡ 4 A r ( T ) 2 C r ( T ) . (59) T B_r(T)P_r'(T) \equiv4A_r(T)^2C_r(T). \tag*{(59)} T B r ( T ) P r ′ ( T ) ≡ 4 A r ( T ) 2 C r ( T ) . (59)
Indeed, on using P r ( T ) = 4 A r ( T ) 3 + 27 T B r ( T ) 2 P_r(T)=4A_r(T)^3+27TB_r(T)^2 P r ( T ) = 4 A r ( T ) 3 + 27 T B r ( T ) 2 , the congruence follows from
T B r P r ′ = 12 T A r 2 A r ′ B r + 27 T B r 3 + 54 T 2 B r 2 B r ′ = 4 A r 2 ( 3 T A r ′ B r − A r B r − 2 T A r B r ′ ) . T B_rP_r'=12T A_r^2A_r'B_r+27T B_r^3+54T^2B_r^2B_r'=4A_r^2(3T A_r'B_r-A_rB_r-2T A_rB_r'). T B r P r ′ = 12 T A r 2 A r ′ B r + 27 T B r 3 + 54 T 2 B r 2 B r ′ = 4 A r 2 ( 3 T A r ′ B r − A r B r − 2 T A r B r ′ ) .
Taking resultants with P r P_r P r gives Res ( P r , T B r P r ′ ) = Res ( P r , 4 A r 2 C r ) \operatorname{Res}(P_r,TB_rP_r')=\operatorname{Res}(P_r,4A_r^2C_r) Res ( P r , T B r P r ′ ) = Res ( P r , 4 A r 2 C r ) . Since deg P r = 6 \deg P_r=6 deg P r = 6 and the leading coefficient of P r P_r P r is 4, we have
Res ( P r , T ) = 4 r 8 3 , Res ( P r , B r ) = 16 ℓ 20 ( r ) 3 , Res ( P r , P r ′ ) = − 4 Disc T ( P r ) . \operatorname{Res}(P_r,T)=4r_8^3,\qquad\operatorname{Res}(P_r,B_r)=16\ell_{20}(r)^3,\qquad\operatorname{Res}(P_r,P_r')=-4\operatorname{Disc}_{T}(P_r). Res ( P r , T ) = 4 r 8 3 , Res ( P r , B r ) = 16 ℓ 20 ( r ) 3 , Res ( P r , P r ′ ) = − 4 Disc T ( P r ) .
Similarly, Res ( P r , A r ) = 27 2 r 8 ℓ 20 ( r ) 2 \operatorname{Res}(P_r,A_r)=27^2r_8\ell_{20}(r)^2 Res ( P r , A r ) = 2 7 2 r 8 ℓ 20 ( r ) 2 . Substituting these identities into the previous resultant equality yields (A.5). Combining this with (A.3) gives (A.6). The weight is wt Res ( P r , C r ) − 8 − 2 ⋅ 20 = 60 \operatorname{wt}\operatorname{Res}(P_r,C_r)-8-2\cdot20=60 wt Res ( P r , C r ) − 8 − 2 ⋅ 20 = 60 , or equivalently follows from Lemma A.2. □
References
Email address: y.maeda.math@gmail.com, atsuhira.nagano@gmail.com