, ,

Research of the author is partially supported by Grant-in-Aid for Scientific Research A-14204001, Japan .

<ph f="ptmb7t">The moduli space of 8 points on </ph> <math xmlns="http://www.w3.org/1998/Math/MathML"> <msup> <mrow> <mo mathvariant="bold">P</mo> </mrow> <mrow> <mn>1</mn> </mrow> </msup> </math> <ph f="cmr"> </ph><ph f="ptmb7t">and automorphic forms</ph>

Shigeyuki Kond o ¯

Graduate School of Mathematics, Nagoya University, Nagoya, 464-8602, Japan E-mail address: kondo@math.nagoya-u.ac.jp

Abstract. First we give a complex ball uniformization of the moduli space of 8 ordered points on the projective line by using the theory of periods of $K 3$ surfaces. Next we give a projective model of this moduli space by using automorphic forms on a bounded symmetric domain of type $I V$ which coincides with the one given by cross ratios of 8 ordered points of the projective line.

Dedicated to Igor Dolgachev on his 60th birthday

1 Introduction

The main purpose of this paper is to give an application of the theory of automorphic forms on a bounded symmetric domain of type

I V

due to Gritsenko and Borcherds [Bor] for studying the moduli spaces. We consider the moduli space

P_{1}^{8}

of semi-stable 8 ordered points on the projective line. It is known that

P_{1}^{8}

is isomorphic to the Satake-Baily-Borel compactification of an arithmetic quotient of 5-dimensional complex ball by using the theory of periods of a family of curves which are the 4-fold cyclic covers of the projective line branched at eight points ([DM] ). Recently Matsumoto and Terasoma [MT] gave an embedding of

P_{1}^{8}

into

P^{104}

by using the theta constants related to the above curves. Their map coincides the one defined by the cross ratios of 8 points on the

P^{1}

. Here they used the fact that the complex ball is canonically embedded in a Siegel upper half plane.

In this paper, instead of the periods of curves, we use the periods of

K 3

surfaces. In our case, the complex ball is embedded in a bounded symmetric domain of type

I V

. In fact, to each stable point from

P_{1}^{8}

, we associate a

K 3

surface with a non-symplectic automorphism of order 4 (§ 2 ). The period domain of these

K 3

surfaces is a 5-dimensional complex ball

ℬ

(§ 3 ). This was essentially given in the paper [Kon1] . By using this, we shall see that

P_{1}^{8}

is isomorphic to the Satake-Baily-Borel compactification

\bar{ℬ} / Γ (1 - i)

ℬ / Γ (1 - i)

where

Γ (1 - i)

is an arithmetic subgroup of a unitary group of a hermitian form of signature

(1, 5)

defined over the Gaussiann integers (Theorems 3.7 , 4.11 ) . The symmetry group

S_{8}

of degree 8 naturally acts on

P_{1}^{8}

. On the other hand, there exists an arithmetic subgroup

Γ

acting on

ℬ

with

Γ / Γ (1 - i) \sim = S_{8}

. The above isomorphism

P_{1}^{8} \sim = \bar{ℬ} / Γ (1 - i)

S_{8}

-equivariant.

Next we apply the theory of automorphic forms [Bor] to this situation. The main idea comes from the paper of Allcock and Freitag [AF] in which they studied the same problem in the case of cubic surfaces. We shall show that there exists a 14-dimensional space of automorphic forms on

ℬ

which gives an

S_{8}

-equivariant map from the arithmetic quotient

\bar{ℬ} / Γ (1 - i)

into

P^{13}

(Theorem 7.6 ). Under the identification

P_{1}^{8} \sim = \bar{ℬ} / Γ (1 - i)

we show that this map coincides with the one defined by the cross ratios of 8 points on the projective line (Theorem 7.5 ). Thus our map coincides the one given by Matsumoto and Terasoma [MT] .

In this paper, a lattice means a

Z

-valued non-degenerate symmetric bilinear form on a free

Z

-module of finite rank. We denote by

U

the even lattice defined by the matrix

(\begin{matrix} 0 & 1 \\ 1 & 0 \end{matrix})

, and by

A_{m}

D_{n}

E_{l}

the even negative definite lattice defined by the Dynkin matrix of type

A_{m}

D_{n}

E_{l}

respectively. If

L

is a lattice and

m

is an integer, we denote by

L (m)

the lattice over the same

Z

-module with the symmetric bilinear form multiplied by

m

. We also denote by

L^{\oplus m}

the orthogonal direct sum of

m

copies of

L

and by

L^{*}

the dual of

L

2 $K 3$ surfaces associated with 8 points on the projective line

2.1

In this section, we shall construct a

K 3

surface associated to distinct 8 points from

P_{1}^{8}

. In section 4 , we shall generalize this to the cases of any stable points and semi-stable points from

P_{1}^{8}

. Let

{(λ_{i} : 1)}

be a set of distinct 8 points on the projective line. Let

(x_{0} : x_{1}, y_{0} : y_{1})

be the bi-homogenious coordinates on

P^{1} \times P^{1}

. Consider a smooth divisor

C

P^{1} \times P^{1}

of bidegree

(4, 2)

given by

\begin{matrix} y_{0}^{2} \cdot^{4} \prod_{i = 1} (x_{0} - λ_{i} x_{1}) + y_{1}^{2} \cdot^{8} \prod_{i = 5} (x_{0} - λ_{i} x_{1}) = 0 . \end{matrix}

(2.1)

Let

L_{0}

(resp.

L_{1}

) be the divisor defined by

y_{0} = 0

(resp.

y_{1} = 0

). Let

ι

be an involution of

P^{1} \times P^{1}

given by

\begin{matrix} (x_{0} : x_{1}, y_{0} : y_{1}) ⟶ (x_{0} : x_{1}, y_{0} : - y_{1}) \end{matrix}

(2.2)

which preserves

C

and

L_{0}, L_{1}

. Note that the double cover of

P^{1} \times P^{1}

branched along

C + L_{0} + L_{1}

has 8 rational double points of type

A_{1}

and its minimal resolution

X

is a

K 3

surface. This

K 3

surface

X

is obtained as follows: First blow up the 8 points which are the intersection of

C

and

L_{0} + L_{1}

Then

X

is the double cover branched along the proper transforms of

C

L_{0}

and

L_{1}

. We remark that the isomorphism class of

X

depends only on the 8 points in

P^{1}

(i.e. independent on the order of 8 points) because elementary transformations change the order of 8 points.

The involution

ι

lifts to an automorphism

σ

of order 4. We can easily see that

σ^{*} ω_{X} = \pm \sqrt{- 1} ω_{X}

where

ω_{X}

is a nowhere vanishing holomorphic 2-form on

X

. We denote by

S_{0}, S_{1}

the inverse image of

L_{0}, L_{1}

respectively. The projection

(x_{0} : x_{1}, y_{0} : y_{1}) ⟶ (x_{0} : x_{1})

from

P^{1} \times P^{1}

P^{1}

induces an elliptic fibration

π : X ⟶ P^{1}

which has 8 singular fibers of type

I I I

in the sense of Kodaira [Kod] and two sections

S_{0}, S_{1}

. Let

E_{i} + F_{i}

(

1 \leq i \leq 8

) be the 8 singular fibers of

π

. Then we may assume that

E_{i} \cdot S_{0} = F_{i} \cdot S_{1} = 1 .

Put

\begin{matrix} H^{2} (X, Z)^{+} = {x \in H^{2} (X, Z) | (σ^{2})^{*} (x) = x}; \end{matrix}

(2.3)

\begin{matrix} H^{2} (X, Z)^{-} = {x \in H^{2} (X, Z) | (σ^{2})^{*} (x) = - x} . \end{matrix}

(2.4)

We also denote by

S_{X}

T_{X}

the Picard lattice, the transcendental lattice of

X

respectively.

2.2 Lemma

(i)

H^{2} (X, Z)^{+} ≃ U (2) \oplus D_{4} \oplus D_{4}

H^{2} (X, Z)^{-} ≃ U \oplus U (2) \oplus D_{4} \oplus D_{4} .

(ii)

σ^{*}

acts on

H^{2} (X, Z)^{+}

trivially.

(iii) The following elements generate

(H^{2} (X, Z)^{+})^{*} / H^{2} (X, Z)^{+} ≃ (Z / 2 Z)^{6}

(F_{1} + F_{2}) / 2, (F_{1} + F_{3}) / 2, (F_{1} + F_{4}) / 2, (F_{1} + F_{5}) / 2, (F_{1} + F_{6}) / 2, (F_{1} + F_{7}) / 2 .

(iv) Let

U \oplus A_{1}^{\oplus 8}

be the sublattice of

H^{2} (X, Z)^{+}

generated by the classes of a fiber,

S_{0}

and

F_{i}

(1 \leq i \leq 8)

. Then

H^{2} (X, Z)^{+}

is obtained from

U \oplus A_{1}^{\oplus 8}

by adding the vector

(F_{1} + \cdot \cdot \cdot + F_{8}) / 2

Proof. For the proof of the assertions (i)–(iii), see [Kon1] , Lemma 5.2. The sublattice $U \oplus A_{1}^{\oplus 8}$ has index 2 in $H^{2} (X, Z)^{+}$ . The later one is obtained from the former by adding the class of $S_{1}$ . Hence the last assertion follows from the fact that $S_{1} = 2 F + S_{0} + (F_{1} + \cdot \cdot \cdot + F_{8}) / 2$ where $F$ is a fiber. □

It follows that

H^{2} (X, Z)^{+} \subset S_{X}

and

T_{X} \subset H^{2} (X, Z)^{-}

2.3 A quadratic form

First of all, we define:

\begin{matrix} L = U^{3} \oplus E_{8}^{3}, M = U (2) \oplus D_{4} \oplus D_{4}, N = U \oplus U (2) \oplus D_{4} \oplus D_{4} . \end{matrix}

(2.5)

Recall that

H^{2} (X, Z) \sim = L

. We consider

M

as a sublattice of

L

and

N

is the orthogonal complement of

M

L

. It follows from Theorem 1.14.4 in Nikulin [N3] that the embedding of

M

into

L

is unique.

Let

A_{N} = N^{*} / N

which is isomorphic to a vector space

F_{2}^{6}

of dimension 6 over

F_{2}

(Lemma 2.2 ). The discriminant quadratic form

\begin{matrix} q_{N} : A_{N} ⟶ Q / 2 Z \end{matrix}

(2.6)

is defined by

q_{N} (x) = 〈 x, x 〉 m o d 2 Z

. In our situation, the image of

q_{N}

is contained in

Z / 2 Z

and hence

q_{N}

is a quadratic form on

A_{N}

defined over

F_{2}

whose associated bilinear form is given by

b_{N} (x, y) = 2 〈 x, y 〉 m o d 2 Z .

Let

u

be the hyperbolic plane defined over

F_{2}

, that is, the quadratic form of dimension 2 defined over

F_{2}

corresponding to the matrix

(\begin{matrix} 0 & 1 \\ 1 & 0 \end{matrix})

. The quadratic form

q_{N}

is isomorphic to the direct sum of 3 copies of

u

q_{N} \sim = u^{\oplus 3}

. It is known that

(A_{M} = M^{*} / M, q_{M}) \sim = (A_{N}, - q_{N})

([N3] , Corollary 1.6.2).

2.4 Lemma

(i)

O (q_{M}) ≃ O (q_{N}) ≃ O (u_{2}^{\oplus 3}) ≃ S_{8}

where

S_{8}

is the symmetric group of degree

8

(ii) The group

O (q_{M})

naturally isomorphic to the subgroup of

O (M)

generated by the permutations of the

8

components

A_{1}

U \oplus A_{1}^{\oplus 8}

Proof. The assertion (i) is well known (e.g. [Atlas] , page 22). For (ii), note that the permutations of 8 components $A_{1}$ can be extended to isometries of $M$ because they preserve $(F_{1} + \cdot \cdot \cdot + F_{8}) / 2$ (see Lemma 2.2 , (iv)). Now the assertion is obvious. □

2.5 A fundamental domain

Let

M = U (2) \oplus D_{4} \oplus D_{4}

. Let

P (M)

be a connected component of

{x \in M \otimes R : 〈 x, x 〉 > 0}

Let

W (M)

be the reflection group generated by

(- 2)

-reflections

s_{r} : x \to x + 〈 x, r 〉 r

for any

r \in M

with

r^{2} = - 2

. The group

W (M)

acts on

P (M)

discretely. Let

C (M)

be the finite polyhedral cone defined by the 18

(- 2)

-vectors which are corresponding to the 18 smooth rational curves

S_{0}, S_{1}, E_{i}, F_{i}

(1 \leq i \leq 8)

X

under an isomorphism

M \sim = H^{2} (X, Z)^{+}

2.6 Proposition

(i) The group

W (M)

is of finite index in the orthogonal group

O (M)

. Moreover the closure

\bar{C} (M)

C (M)

is a fundamental domain of

W (M)

. The symmetry group of

C (M)

is isomorphic to

S_{8} \times Z / 2 Z

(ii) If

S_{X} \sim = M

, then

X

contains exactly

18

smooth rational curves

S_{0}, S_{1}, E_{i}, F_{i}

(1 \leq i \leq 8)

Proof. (i) The first assertion follows from Nikulin's classification of such hyperbolic 2-elementary lattices ([N1] , Theorem 4.4.1). Moreover $C (M)$ satisfies the condition in Vinberg's theorem [V] , Theorem 2.6, that is, any maximal extended Dynkin diagram in these 18 $(- 2)$ -vectors is either ${\tilde{A}}_{1}^{\oplus 8}$ or ${\tilde{D}}_{4}^{\oplus 2}$ both of which have the maximal rank 8. Hence $C (M)$ is of finite volume. Let $C (M)^{'}$ be a fundamental domain of $W (M)$ with $C (M)^{'} \subset C (M)$ . Then [V] , Lemma 2.4 implies that $C (M) = C (M)^{'}$ . The last assertion is obvious.
(ii) It follows from a remark in Vinberg [V] , p. 335 that $C (M)$ has finite volume iff the polyherdal cone $\bar{C} (M)$ is contained in the closure $\bar{P} (M)$ of $P (M)$ . If there exists a smooth rational curve $E$ different from the above 18 curves. Then the intersection number of $E$ and any one of these 18 curves is non negative, that is, the class of $E$ is contained in $\bar{C} (M) \subset \bar{P} (M)$ . This implies that $E^{2} \geq 0$ , which is a contradiction. □

2.7

Let

X

be a

K 3

surface as above. Let

P (X)

be the component of

{x \in S_{X} \otimes R : 〈 x, x 〉 > 0}

which contains an ample class. Let

Δ (X)

be the set of all effective divisors

r

with

r^{2} = - 2

. Let

C (X)

be the polyhedral cone defined by:

C (X) = {x \in S_{X} \otimes R : 〈 x, r 〉 > 0, r \in Δ (X)} .

Note that the integral points in

C (X)

are nothing but the ample classes. If

S_{X} \sim = M

, then

C (X) = C (M)

2.8 Lemma

The orthogonal complement of

H^{2} (X, Z)^{+}

S_{X}

contains no

(- 2)

-vectors.

Proof. If $r \in (H^{2} (X, Z)^{+})^{⊥} \cap S_{X}$ with $r^{2} = - 2$ , then $(σ^{*})^{2} (r) = - r$ . On the other hand, Riemann-Roch theorem implies that $r$ is effective. This is a contradiction. □

2.9 Proposition

Assume that

S_{X} = H^{2} (X, Z)^{+}

. Then the automorphism group of

X

is finite. Moreover

X

has exactly

18

smooth rational curves which are components of singular fibers of

π

and two sections.

Proof. Recall that $A u t (X)$ is isomorphic to $O (S_{X}) / W (S_{X})$ up to finite groups ([PS] ). Here $W (S_{X})$ is the subgroup generated by $(- 2)$ -reflections. Hence the assertion follows from Proposition 2.6 . □

2.10 The automorphism of order 4

We shall study the action of

σ

H^{2} (X, Z)^{-}

. Recall that

D_{4} \sim = {(x_{1}, x_{2}, x_{3}, x_{4}) \in Z^{4} | x_{1} + x_{2} + x_{3} + x_{4} \equiv 0 (m o d 2)} .

Here we consider the standard inner product on

Z^{4}

with the negative sign. Let

ρ_{0}

be the isometry of

D_{4}

given by

ρ_{0} (x_{1}, x_{2}, x_{3}, x_{4}) = (x_{2}, - x_{1}, x_{4}, - x_{3}) .

Obviously

ρ_{0}

is of order 4 and fixes no non-zero vectors in

D_{4}

. Also an easy calculation shows that

ρ_{0}

acts trivially on

D_{4}^{*} / D_{4}

. Next let

e, f

(resp.

e^{'}, f^{'}

) be a basis of

U

(resp.

U (2)

). Define the isometry

ρ_{1}

U \oplus U (2)

ρ_{1} (e) = - e - e^{'}, ρ_{1} (f) = f - f^{'}, ρ_{1} (e^{'}) = e^{'} + 2 e, ρ_{1} (f^{'}) = 2 f - f^{'} .

Obviously

ρ_{1}

is of order 4, fixes no non-zero vectors in

U \oplus U (2)

and acts trivially on the discriminant group of

U \oplus U (2)

. Thus we have an isometry

ρ = ρ_{1} \oplus ρ_{0} \oplus ρ_{0}

N = U \oplus U (2) \oplus D_{4} \oplus D_{4}

which fixes no non-zero vectors in

N

and acts trivially on

N^{*} / N

. Then

ρ

can be extended to an isometry of the

K 3

lattice

L

acting trivially on

M

([N3] , Proposition 1.6.1).

2.11 Lemma

The isometry

ρ

is conjugate to

σ^{*}

under an isomorphism

H^{2} (X, Z) \sim = L

Proof. Let $ω$ be an eigenvector of $ρ$ which is sufficiently general, that is, satisfying the condition $ω^{⊥} \cap L = M$ . By the surjectivity of the period map for $K 3$ surfaces, there exists a $K 3$ surface $Y$ and an isometry $α_{Y} : H^{2} (Y, Z) \to L$ with $α_{Y} (ω_{Y}) = ω$ where $ω_{Y}$ is a nowhere vanishing holomorphic 2-form on $Y$ . By the condition $ω^{⊥} \cap L = M$ , we have $S_{Y} \sim = M$ . Consider the isometry $φ = α_{Y}^{- 1} \circ ρ \circ α_{Y}$ of $H^{2} (Y, Z)$ . Since $ρ$ acts trivially on $M$ , $φ$ preserves ample classes. Then it follows from the Torelli theorem [PS] that $φ$ is induced from an automorphism $g$ of $Y$ of order 4. On the other hand, Proposition 2.9 implies that $Y$ contains exactly 18 smooth rational curves whose dual graph is the same as that of the smooth rational curves on $X$ . In particular, $Y$ has an elliptic fibration with two sections and 8 singular fibers each of which is type $I I I$ or $I_{2}$ . Since $g$ acts trivially on the Picard lattice $M$ , $g$ preserves the elliptic fibration and the class of each component of singular fibers of type $I I I$ or $I_{2}$ . Since the elliptic fibration has 8 singular fibers, $g$ acts trivially on the base of the elliptic fibration, and hence induces an automorphism of each fiber. Hence all singular fibers are of type $I I I$ . It follows from Nikulin [N1] , Theorem 4.2.2 that the set of fixed points of the involution $g^{2}$ is the disjoint union of two smooth rational curves $R_{0}, R_{1}$ and a smooth curve $C$ of genus 3. Since $g$ acts trivially on the base, $R_{0}$ , $R_{1}$ are sections of the elliptic fibration. We can easily see that $C$ passes through singular points of singular fibers of type $I I I$ . Thus we have the same configuration of smooth rational curves on $Y$ as that of $X$ . By taking the quotient of $Y$ by $g^{2}$ , we can see that $Y$ is a deformation of $X$ . Hence we have the assertion. □

2.12 Markings

Recall that

H^{2} (X, Z)^{+} \sim = M = U (2) \oplus D_{4} \oplus D_{4}

(Lemma 2.2 ). We fix a fundamental domain

C (M)

(Proposition 2.6 ). It follows from Lemma 2.11 that there exists an isometry

α_{X} : H^{2} (X, Z) \to L

satisfying

ρ = α_{X} \circ σ^{*} \circ α_{X}^{- 1}

. We call

α_{X}

a marking and the pair

(X, α_{X})

a marked

K 3

surface.

Then

2.13 Proposition

There exists a marking

α_{X}

such that

α_{X} (C (X)) \cap M \otimes R \subset C (M)

Proof. It follows from Lemma 2.8 that $α_{X} (C (X)) \cap M \otimes R$ is an open polyhedral cone in $M \otimes R$ .
Hence Proposition 2.6 implies the assertion. □

3 A complex ball uniformization

In this section we construct an

S_{8}

-equivariant isomorphism between the moduli space of the projective equivalence classes of the set of distinct 8 ordered points in

P^{1}

and an open set of the arithmetic quotient of 5-dimensional complex ball.

3.1 The period domain

Let

(X, α_{X})

be a marked

K 3

surface and let

ω_{X}

be a nowhere vanishing holomorphic 2-form on

X

. Then

α_{X} (ω_{X})

is contained in the following domain:

\begin{matrix} D = {ω \in P (N \otimes C) : 〈 ω, ω 〉 = 0, 〈 ω, \bar{ω} 〉 > 0} . \end{matrix}

(3.1)

Note that

D

is a disjoint union of two copies of a bounded symmetric domain of type

I V

and of dimension 10. To get the period domain, we first define:

\begin{matrix} V_{\pm} = {z \in N \otimes C | ρ (z) = \pm \sqrt{- 1} z} . \end{matrix}

(3.2)

It follows from Nikulin [N2] , Theorem 3.1 that

N \otimes C = V_{+} \oplus V_{-}

. Now we may assume

σ^{*} (ω_{X}) = \sqrt{- 1} \cdot ω_{X}

. Then

α_{X} (ω_{X})

is, in fact, contained in

ℬ

defined by

\begin{matrix} ℬ = {z \in P (V_{+}) | 〈 z, \bar{z} 〉 > 0} . \end{matrix}

(3.3)

z \in ℬ

, then

〈 z, z 〉 = 〈 ρ (z), ρ (z) 〉 = 〈 \sqrt{- 1} z, \sqrt{- 1} z 〉 = - 〈 z, z 〉,

and hence

〈 z, z 〉 = 0 .

Thus we have

D \cap P (V_{+}) = ℬ .

We remark that

ℬ

is a 5-dimensional complex ball. We call

α_{X} (ω_{X})

the period of

(X, α_{X})

. We also define two arithmetic subgroups:

\begin{matrix} Γ = {γ \in O (N) | γ \circ ρ = ρ \circ γ}; \end{matrix}

(3.4)

\begin{matrix} Γ (1 - i) = K e r (Γ \to O (q_{N})) . \end{matrix}

(3.5)

We shall see that the quotient

ℬ / Γ

(resp.

ℬ / Γ (1 - i)

) is the coarse moduli space of distinct 8 unordered points on

P^{1}

(resp. distinct 8 ordered points on

P^{1}

) (see Theorem 3.7 ).

3.2 Hermitian form

We consider

N

as a free

Z [\sqrt{- 1}]

-module

Λ

(a + b \sqrt{- 1}) x = a x + b ρ (x) .

Let

h (x, y) = \sqrt{- 1} 〈 x, ρ (y) 〉 + 〈 x, y 〉 .

Then

h (x, y)

is a hermitian form on

Z [\sqrt{- 1}]

-module

Λ

. With respect to a

Z [\sqrt{- 1}]

-basis

(1, - 1, 0, 0)

(0, 1, - 1, 0)

D_{4}

, the hermitian matrix of

h | D_{4}

is given by

\begin{matrix} (\begin{matrix} - 2 & 1 - \sqrt{- 1} \\ 1 + \sqrt{- 1} & - 2 \end{matrix}) . \end{matrix}

(3.6)

And with respect to a

Z [\sqrt{- 1}]

-basis

e, e^{'}

U \oplus U (2)

, the hermitian matrix of

h | U \oplus U (2)

is given by

\begin{matrix} (\begin{matrix} 0 & 1 + \sqrt{- 1} \\ 1 - \sqrt{- 1} & 0 \end{matrix}) . \end{matrix}

(3.7)

Let

φ : Λ \to N^{*}

be a linear map defined by

φ (x) = (x + ρ (x)) / 2 .

Note that

φ ((1 - \sqrt{- 1}) x) = φ (x - ρ (x)) = x \in N

Hence

φ

induces an isomorphism

\begin{matrix} Λ / (1 - \sqrt{- 1}) Λ ≃ N^{*} / N . \end{matrix}

(3.8)

3.3 Remark

The hermitian form

h

coincides with the one of Matsumoto and Yoshida in [MY] , §6. This implies that our groups

Γ

Γ (1 - i)

coincide with the ones of Matsumoto and Yoshida in [MY] .

3.4 Reflections

For

r \in N

with

〈 r, r 〉 = - 2

, we define a reflection

s_{r} (x) = x + 〈 r, x 〉 r

which is contained in

\tilde{O} (N) = K e r (N \to O (q_{N}))

, but not in

Γ

. On the other hand, by considering

r

as in

Λ

, we define a reflection

\begin{matrix} R_{r, ε} (x) = x - (1 - ε) \frac{h (r, x)}{h (r, r)} r \end{matrix}

(3.9)

where

ε \neq 1

is a 4-th root of unity. We can easily see that

R_{r, - 1}

corresponds to the isometry in

Γ

x \to x + 〈 r, x 〉 r + 〈 ρ (r), x 〉 ρ (r)

which coincides with

s_{r} \circ s_{ρ (r)}

. Also

R_{r, \sqrt{- 1}}

corresponds to the isometry in

Γ

x \to x + 〈 r, x 〉 (r - ρ (r)) / 2 + 〈 ρ (r), x 〉 (r + ρ (r)) / 2

which induces a transvection of

A_{N}

defined by

t_{α} (x) = x + b_{N} (x, α) α

where

α \in A_{N}

is a non-isotropic vector

(r + ρ (r)) / 2 m o d N

3.5 Discriminant

Let

r \in N

with

r^{2} = - 2

. We denote by

H_{r}

the hyperplane of

ℬ

defined by

H_{r} = {z \in ℬ : 〈 z, r 〉 = 0} .

Let

ℋ

be the union of all hyperplanes

H_{r}

where

r

moves on the set of all

(- 2)

-vectors in

N

. We call

ℋ

the discriminant locus. By Lemma 2.8 , the periods of marked

K 3

surfaces as above are contained in

ℬ \ ℋ

Conversely let

ω \in ℬ \ ℋ

. Then by the surjectivity of the period map, there exists a marked

K 3

surface

(X, α_{X})

with

α_{X} (ω_{X}) = ω

. The condition

ω / \in ℋ

implies that Proposition 2.13 holds for this

K 3

surface. Hence, if necessary by replacing

α_{X}

, we may assume that the isometry

α_{X}^{- 1} \circ ρ \circ α_{X}

preserves the ample cone of

X

. It now follows from the Torelli type theorem ([PS] ) that there exists an automorphism

σ

of order 4 satisfying

α_{X}^{- 1} \circ ρ \circ α_{X} = σ^{*}

. Moreover the marking defines an elliptic fibration

π : X \to P^{1}

with a section

s

3.6 Lemma

(i)

π

has

8

singular fibers of type

I I I

and two sections; (ii) The set of fixed points of

σ^{2}

is the disjoint union of two sections and a smooth curve of genus

3

which passes through

8

singular points of

8

singular fibers.

Proof. It is known that the set of fixed points of the involution $σ^{2}$ is the disjoint union of two smooth rational curves $R_{0}$ , $R_{1}$ and a smooth curve $C$ of genus 3 (Nikulin [N1] , Theorem 4.2.2). Obviously the set $X^{σ}$ of fixed points of $σ$ is contained in $R_{0} + R_{1} + C$ . Since $σ^{*} = α_{X}^{- 1} \circ ρ \circ α_{X}$ , $X^{σ}$ has the Euler number 12. Since $σ$ acts on $M$ trivially, it preserves the section $s$ and the class of a fiber of $π$ . We show that $σ$ acts trivially on the base of $π$ . Assume otherwise, then $X^{σ}$ is contained in two invariant fibers, $F_{1}, F_{2}$ . Let $l$ be the number of irreducible one-dimensional components of $X^{σ}$ and $k$ the number of isolated fixed points of $σ$ . Then $2 l + k = 12$ . If we denote by $U$ the sublattice generated by the classes of a fiber and the section $s$ , then $U^{⊥} \cap H^{2} (X, Z)^{σ^{*}} = A_{1}^{8}$ . Hence the divisor $F_{1} + F_{2}$ contains at least 10 components. Assume $F_{1}$ contains at least 5 components. Note that $σ$ preserves the component of $F_{1}$ which meets with $s$ . Obviously there are no singular fibers with non-trivial symmetry of order 4. Hence the involution $σ^{2}$ preserves each component of $F_{1}$ . Then the sublattice generated by components of $F_{1}$ not meeting $s$ has at least rank 4 and is isomorphic to an indecomposable root lattice $R$ . Since $σ^{2}$ acts trivially on $R$ , $R$ is contained in $A_{1}^{8}$ which is impossible.
Thus $σ$ acts trivially on the base. This implies that each fiber has an automorphism of order 4. In particular, singular fibers of $π$ are either of type $I I I, I I I^{*}$ or $I_{0}^{*}$ . Since $U^{⊥} \cap H^{2} (X, Z)^{σ^{*}} = A_{1}^{8}$ and the singular fibers of type $I I I^{*}$ and $I_{0}^{*}$ have no non-trivial symmetry of order 4, every singular fiber is of type $I I I$ . Since $σ$ fixes two points on each component of a singular fiber one of which is the singular point, $R_{0}$ , $R_{1}$ or $C$ passes through these points. Now we can easily see the assertion (ii). □

3.7 Theorem

The period map induces an

S_{8}

-equivariant isomorphism

φ

between the moduli space

(P_{1}^{8})^{0}

of distinct ordered

8

points on the projective line and the quotient space

(ℬ \ ℋ) / Γ (1 - i)

Proof. As in 3.5 , for each $ω \in ℬ \ ℋ$ , we have a marked $K 3$ surface $(X, α_{X})$ with $α_{X} (ω_{X}) = ω$ .
Moreover $X$ has an automorphism $σ$ of order 4. By Lemma 3.6 , $X$ has an elliptic fibration with two section and 8 singular fibers of type $I I I$ . By taking the quotient of $X$ by $σ^{2}$ and contracting $(- 1)$ -curves, we have the embedding of $C$ as in 2.1 . This correspondence is the inverse of the period map. □

4 Discriminant locus

In this section we shall determine the discriminant locus

ℋ

ℬ

4.1

Let

r

be a

(- 2)

-vector in

N

. Let

ω \in ℬ

. Then

〈 r, ω 〉 = \sqrt{- 1} 〈 ρ (r), ω 〉

. Hence

r

and

ρ (r)

define the same hyperplane

H_{r}

ℬ

. Hence

H_{r}

corresponds to an embedding of the lattice

R_{r} \sim = A_{1} \oplus A_{1}

generated by

r

and

ρ (r)

into

N

. Obviously

R_{r} ≃ A_{1} \oplus A_{1}

. Also every embedding of

R_{r}

into

N

is primitive, that is,

N / R_{r}

is torsion free.

4.2 Lemma

Let

R_{r}^{⊥}

be the orthogonal complement of

R_{r}

N

. Then

R_{r}^{⊥} ≃ U \oplus U (2) \oplus D_{4} \oplus A_{1}^{2}

. In particular,

(r + ρ (r)) / 2 \in N^{*}

Proof. The proof for the first assertion is similar to those of [Kon1] , Lemmas 3.2, 3.3. Then $R_{r}^{⊥} \oplus R_{r}$ is a sublattice of $N$ of index 2 and $N$ is obtained from $R_{r}^{⊥} \oplus R_{r}$ by adding $(r + ρ (r)) / 2 + θ$ where $θ \in (R_{r}^{⊥})^{*}$ . We can see that $〈 (r + ρ (r)) / 2, x 〉 \in Z$ for $x \in R_{r} \oplus R_{r}^{⊥}$ and $x = (r + ρ (r)) / 2 + θ$ . Hence the second assertion holds. □

4.3

Note that

A_{N}

consists of the following 64 vectors:

Type

(00) : α = 0, # α = 1

(zero); Type

(0) : α \neq 0, q (α) = 0, # α = 35

(non-zero isotropic vector); Type

(1) : q (α) = 1, # α = 28

(non-isotropic vector).

4.4

By Lemma 4.2 , the vector

(r + ρ (r)) / 2

is contained in

N^{*}

. In particular it defines a non-isotropic vector

(r + ρ (r)) / 2 m o d N

A_{N}

. Conversely let

δ

be a

(- 4)

-vector in

N

with

δ / 2 \in N^{*}

. Since

ρ

acts trivially on

A_{N} = N^{*} / N

δ - ρ (δ) \in 2 N

. Put

r = (δ - ρ (δ)) / 2 \in N

. Since

〈 δ, ρ (δ) 〉 = 0

r^{2} = - 2

. Obviously

δ = r + ρ (r)

. Thus we have

4.5 Lemma

{δ \in N : δ^{2} = - 4, δ / 2 \in N^{*}} = {r + ρ (r) : r \in N, r^{2} = - 2} .

4.6 Lemma

Any non-isotropic vector in

A_{N}

is represented by

(r + ρ (r)) / 2

for a suitable

(- 2)

-vector

r

N

Proof. It follows from [N3] , Theorem 1.14.2 that the natural map from $O (N)$ to $O (A_{N})$ is surjective.
The group $O (A_{N}) \sim = S_{8}$ acts transitively on the set of non-isotropic vectors in $A_{N}$ . Combining these with Lemma 4.5 , we have the assertion. □

4.7 Proposition

(i)

Γ / Γ (1 - i) ≃ S_{8} .

(ii)

Γ

acts transitively on the set of cusps of

ℬ

and on the set of

ρ

-invariant

R = A_{1} \oplus A_{1}

(iii)

Γ (1 - i)

-orbits of

ρ

-invariant

R = A_{1} \oplus A_{1}

bijectively correspond to non-isotropic vectors in

A_{N}

. Also

Γ (1 - i)

-orbits of cusps of

ℬ

bijectively correspond to non-zero isotropic vectors in

A_{N}

Proof. Recall that the pair $(N, ρ)$ naturally corresponds to the hermitian form $h$ (see Remark 3.3 ).
Hence the assertions follow from [MY] . □

4.8 Stable points

Next we shall construct a

K 3

surface associated to each stable point from

P_{1}^{8}

. Recall that 8 points is stable (resp. semi-stable) iff no four points (resp. five points) coincide (e.g. [DO] , Chap. I, §4, Example 2 (page 31)). We denote by the symbol

(11111111)

for distinct 8 points in

P^{1}

. If two points (resp. three points) coincide, then we denote it by

(2111111)

(resp.

(311111)

4.9 Example: $(2111111)$

We use the same notation as in Section 2 . We assume that

(λ_{1} : 1)

has multiplicity 2. Then the curve

C

in 2.1 degenerates to one of the following two types:

C_{1} : (x_{0} - λ_{1} x_{1}) ({y_{0}^{2}}^{4} \prod_{i = 2} (x_{0} - λ_{i} x_{1}) + y_{1}^{2} \cdot^{7} \prod_{i = 5} (x_{0} - λ_{i} x_{1})) = 0;

C_{2} : y_{0}^{2} (x_{0} - λ_{1} x_{1} {)^{2}}^{3} \prod_{i = 2} (x_{0} - λ_{i} x_{1}) + y_{1}^{2} \cdot^{7} \prod_{i = 4} (x_{0} - λ_{i} x_{1}) = 0 .

The minimal resolution

Y_{i}

of the double covering of

P^{1} \times P^{1}

branched along

C_{i} + L_{0} + L_{1}

is a

K 3

surface and has an elliptic fibration

π

which has 6 singular fibers of type

I I I

, one singular fiber of type

I_{0}^{*}

and two sections

S_{0}, S_{1}

(

i = 1, 2

). We remark that

Y_{1}

and

Y_{2}

are isomorphic because

C_{1}

and

C_{2}

are mutually transformed under elementary transformations. Thus we denote by

Y

instead of

Y_{1}, Y_{2}

. Denote by

2 R_{0} + R_{1} + R_{2} + R_{3} + R_{4}

the singular fiber of type

I_{0}^{*}

. Assume that

S_{0}

meets

R_{1}

and

S_{1}

meets

R_{2}

. Then the normalization

\tilde{C}

C_{1}

meets

R_{3}, R_{4}

. The involution

ι

given in 2.2 induces an automorphism

σ^{'}

Y

of order 4. Note that the restriction

σ^{'}

\tilde{C}

is the hyperelliptic involution of the smooth curve of genus 2. This implies that

σ^{'}

switches

R_{3}

and

R_{4}

. Let

U

be the sublattice generated by the classes of a fiber and

S_{0}

. Then 6 components in the fibers of type

I I I

not meeting to

S_{0}

and

R_{2}

2 R_{0} + R_{2} + R_{3} + R_{4}

generate the sublattice isomorphic to

A_{1}^{8}

. This gives an isometry from

M

into

S_{X}

such that

M^{⊥} \cap S_{X}

contains

(- 2)

vectors

R_{3}

R_{4}

. Thus the period of

Y

is contained in

ℋ

For other stable points, the process is similar. We have several types of the branch curve

C

depending on the order of 8 points, however, they are transformed each other under elementary transformations. Hence the corresponding

K 3

surface is determined by the isomorphism class of 8 points (independent of the order of 8 points).

If three points coincide, then the corresponding elliptic fibration has a singular fiber of type

I I I^{*}

All cases except

(2222)

, the elliptic fibration has two sections. In case of

(2222)

, it has four sections.

The next Table 1 lists the type of 8 stable points on the projective line, type of singular fibers of the elliptic fibration, the Picard lattice and the transcendental lattice of a generic member.

\begin{matrix} 8 p o i n t s & S i n g u l a r f i b e r s & P i c a r d l a t t i c e & T r a n s c e n d e n t a l l a t t i c e \end{matrix}

\begin{matrix} 1) & (11111111) & 8 I I I & U (2) \oplus D_{4} \oplus D_{4} & U \oplus U (2) \oplus D_{4} \oplus D_{4} \end{matrix}

\begin{matrix} 2) & (2111111) & I_{0}^{*}, 6 I I I & U \oplus D_{4} \oplus D_{4} \oplus A_{1}^{\oplus 2} & U \oplus U (2) \oplus D_{4} \oplus A_{1}^{\oplus 2} \end{matrix}

\begin{matrix} 3) & (221111) & 2 I_{0}^{*}, 4 I I I & U \oplus D_{6} \oplus D_{4} \oplus A_{1}^{\oplus 2} & U \oplus U (2) \oplus A_{1}^{\oplus 4} \end{matrix}

\begin{matrix} 4) & (22211) & 3 I_{0}^{*}, 2 I I I & U \oplus D_{6} \oplus D_{6} \oplus A_{1}^{\oplus 2} & A_{1} (- 1)^{\oplus 2} \oplus A_{1}^{\oplus 4} \end{matrix}

\begin{matrix} 5) & (2222) & 4 I_{0}^{*} & U \oplus D_{8} \oplus D_{8} & U (2) \oplus U (2) \end{matrix}

\begin{matrix} 6) & (311111) & I I I^{*}, 5 I I I & U \oplus D_{8} \oplus D_{4} & U \oplus U (2) \oplus D_{4} \end{matrix}

\begin{matrix} 7) & (32111) & I I I^{*}, I_{0}^{*}, I I I & U \oplus E_{8} \oplus D_{4} \oplus A_{1}^{\oplus 2} & U \oplus U (2) \oplus A_{1}^{\oplus 2} \end{matrix}

\begin{matrix} 8) & (3221) & I I I^{*}, 2 I_{0}^{*}, I I I & U \oplus E_{8} \oplus D_{6} \oplus A_{1}^{\oplus 2} & A_{1} (- 1)^{\oplus 2} \oplus A_{1}^{\oplus 2} \end{matrix}

\begin{matrix} 9) & (3311) & 2 I I I^{*}, 2 I I I & U \oplus E_{8} \oplus D_{8} & U \oplus U (2) \end{matrix}

\begin{matrix} 10) & (332) & 2 I I I^{*}, I_{0}^{*} & U \oplus E_{8} \oplus D_{10} & A_{1} (- 1)^{\oplus 2} \end{matrix}

\begin{matrix}  \end{matrix}

Table 1

4.10 Strictly semi-stable points: $(44)$

In this case, we have the following 3 cases of curves in the quadric corresponding to the strictly semi-stable points with unique minimal closed orbit:

C_{3} : (x_{0} - λ_{1} x_{1})^{2} (x_{0} - λ_{2} x_{1})^{2} (y_{0}^{2} + y_{1}^{2}) = 0;

C_{4} : (x_{0} - λ_{1} x_{1}) (x_{0} - λ_{2} x_{1}) (y_{0}^{2} (x_{0} - λ_{1} x_{1})^{2} + y_{1}^{2} (x_{0} - λ_{2} x_{1})^{2}) = 0;

C_{5} : y_{0}^{2} (x_{0} - λ_{1} x_{1})^{4} + y_{1}^{2} (x_{0} - λ_{2} x_{1})^{4} = 0 .

These curves appear in the list of Shah's classification of semistable

K 3

surfaces of degree 4. See Shah [S] , Theorem 4.8, B, Type II, (i)–(iii).

We denote by

\bar{ℬ} / Γ (1 - i)

the Satake-Baily-Borel compactification of

ℬ / Γ (1 - i)

whose boundary consists of 35 cusps. Then we conclude:

4.11 Theorem

The

S_{8}

-equivariant isomorphism

φ

in Theorem 3.7 can be extended to an

S_{8}

-equivariant isomorphism

\tilde{φ}

between

P_{1}^{8}

and

\bar{ℬ} / Γ (1 - i)

. Moreover

\tilde{φ}

sends strictly semistable points to cusps and stable but not distinct

8

points into

ℋ / Γ (1 - i)

Proof. We can apply the argument of Horikawa's proof of the main theorem in [H] . Let $ℳ$ be the space of all 8 semi-stable points on $P^{1}$ and $ℳ_{0}$ the space of all distinct 8 points on $P^{1}$ . We can easily see that $ℳ \ ℳ_{0}$ is locally contained in a divisor with normal crossing. By construction, $φ$ is locally liftable to $ℬ$ . It now follows from a theorem of Borel [Bo] that $φ$ can be extended to a holomorphic map from $ℳ$ to $\bar{ℬ} / Γ (1 - i)$ which induces a holomorphic map $\tilde{φ} : P_{1}^{8} \to \bar{ℬ} / Γ (1 - i) .$ By using the same argument as in the proof of [H] , Theorem 2.2, we can see that $\tilde{φ}$ sends stable, but non-distinct 8 points to $ℋ$ . More precisely we can choose a marking for $K 3$ surfaces corresponding to stable, but non-distinct 8 points, and define the period for them. For each stratification as in Table 1, we can prove the similar statement as in Lemma 3.6 . Then, as in the generic case (Theorem 3.7 ), by case by case argument according to strata, we can see that the map $\tilde{φ}$ is injective over $ℋ$ . Moreover Shah's classification [S] implies the image of strictly semi-stable points go to the boundaries. Hence the Zariski Main theorem implies that $\tilde{φ}$ is an isomorphism. The $S_{8}$ -equivariantness is obvious. □

5 The Weil representation

In this section we shall study the quadratic form

(A_{N}, q_{N})

over

F_{2}

given in 2.6 and the Weil representation of

S L (2, Z)

on the group ring

C [A_{N}]

5.1

Let

\begin{matrix} T = (\begin{matrix} 1 & 1 \\ 0 & 1 \end{matrix}), S = (\begin{matrix} 0 & - 1 \\ 1 & 0 \end{matrix}), \end{matrix}

(5.1)

We denote by

{e_{α}}_{α \in A_{N}}

the satndard basis of

C [A_{N}]

. Let

ρ

be the Weil representation of

S L (2, Z)

C [A_{N}]

which factors through

S L (2, Z / 2 Z)

([Bor] ):

\begin{matrix} ρ (T) (e_{α}) = (- 1)^{q_{N} (α)} e_{α}; ρ (S) (e_{α}) = \frac{1}{8} \sum_{β} (- 1)^{b_{N} (β, α)} e_{β} . \end{matrix}

(5.2)

Representatives of the conjugacy classes of

S L (2, Z / 2 Z) ≃ S_{3}

consist of

E, T, S T

. A direct calculation shows that the traces of the action of

E, T, S T

C [A_{N}]

are

t r (E) = 2^{6}, t r (T) = 8, t r (S T) = 1 .

Let

χ_{i}

(1 \leq i \leq 3)

be the characters of irreducible representations of

S L (2, Z / 2 Z)

χ_{1}, χ_{2}

χ_{3}

is the trivial, alternating character or the character of 2-dimensional irreducible representation respectively.

Let

χ

be the character of the Weil representation of

S L (2, Z / 2 Z)

C [A_{N}]

and let

χ = \sum_{i} m_{i} χ_{i}

be its decomposition into irreducible characters. Then an elementary calculation shows that

\begin{matrix} χ = 15 χ_{1} + 7 χ_{2} + 21 χ_{3} . \end{matrix}

(5.3)

We call a subspace

I

A_{N}

totally isotropic if

q_{N}

vanishes on

I

, and

I

maximal if it has dimension 3.

5.2 Lemma

For each maximal totally isotropic subspace

I

A_{N}

\sum_{α \in I} e_{α} \in C [A_{N}]^{S L (2, Z)} .

Proof. The proof is the same as that of [Kon2] , Lemma 3.2. □

Let

α \in A_{N}

with

q_{N} (α) = 1

. Then

t_{α} : x ⟶ x + b_{N} (x, α) α

is called a transvection and contained in

O (q_{N})

. Note that

t_{α}

is induced from a reflection

s_{r}

associated with a

(- 4)

-vector

r

N

with

r / 2

mod

N = α

and these

t_{α}

(

α \in A_{N}

with

q_{N} (α) = 1

) generate

O (q_{N})

. These 28 transvections in

O (A_{N})

correspond to the 28 transpositions of

S_{8}

5.3 Definition

Let

q_{s}

be a quadratic form on

F_{2}^{3}

given by

q_{s} (x) = \sum_{i = 1}^{3} x_{i}, x = (x_{1}, x_{2}, x_{3}) \in F_{2}^{3} .

Note that the associated bilinear form of

q_{s}

is identically zero. Let

V

be a 3-dimensional subspace of

A_{N}

. We call

V

maximal totally singular if

(V, q_{N} | V)

is isomorphic to

(F_{2}^{3}, q_{s})

. Obviously

V

has a basis consisting of 3 mutually orthogonal non-isotropic vectors

{α_{1}, α_{2}, α_{3}}

. We remark that

V

consists of 4 non-isotropic vectors

α_{i} (1 \leq i \leq 3), α_{1} + α_{2} + α_{3}

and 4 isotropic vectors

0, α_{i} + α_{j}, (1 \leq i < j \leq 3)

. For each maximal totally singular subspace

V

, we define a vector

f_{V} \in C [A_{N}]^{S L (2, Z)}

on which the transvection

t_{α}

(α \in V)

acts as

- 1

. Let

I

be the kernel of

q_{N} | V

Then

I

is a totally isotropic subspace of dimension 2 in

A_{N}

and there exist exactly two maximal totally isotropic subspaces

I^{+}, I^{-}

A_{N}

which contain

I

. We define

\begin{matrix} f_{V} = \sum_{α \in I^{+}} e_{α} - \sum_{α \in I^{-}} e_{α} \in C [A_{N}] . \end{matrix}

(5.4)

5.4 Theorem

Let

V

be a maximal totally singular subspace. Then

f_{V}

is contained in

C [A_{N}]^{S L (2, Z)}

satisfying the following condition:

f_{V}

is the unique vector (up to constant) in

C [A_{N}]

on which transvections

t_{α} (α \in V, q_{N} (α) = 1)

act as

- 1

Proof. The proof is the same as that of [Kon2] , Theorem 3.4. □

5.5 Remark-Definition

The group

O (q_{N}) (≃ S_{8})

naturally acts on

C [A_{N}]^{S L (2, Z)}

with character

χ_{1} + χ_{14}

, where

χ_{1}

is the trivial character and

χ_{14}

is the character of an irreducible representation of

S_{8}

of degree 14.

This follows from Lemma 5.2 and [Atlas] , page 22. Moreover it follows from Theorem 5.4 that the multiplicity of the irreducible representation of degree 14 on

C [A_{N}]

is one. We denote by

W

the subspace of dimention 14 in

C [A_{N}]^{S L (2, Z)}

with character

χ_{14}

5.6 Lemma

The number of maximal totally singular subspaces of

A_{N}

is equal to

105

Proof. We can easily count the number of mutually orthogonal three non-isotropic vectors in $A_{N}$ which is $2^{3} \cdot 3^{2} \cdot 5 \cdot 7$ . On the other hand, the automorphism group of a maximal totally singular subspace has order $2^{3} \cdot 3$ . Hence the assertion follows. □

In the Lemma 7.3 , we shall give a geometric interpretation of maximal totally singular subspaces.

5.7 Heegner divisors

Let

δ \in N

be a

(- 4)

-vector with

δ / 2 \in N^{*}

. Let

D_{δ}

be the hyperplane of

D

defined by

D_{δ} = δ^{⊥} \cap D .

It follows from Lemma 4.5 that

H_{r} = D_{δ} \cap ℬ

where

r = (δ - ρ (δ)) / 2

is a

(- 2)

-vector in

N

and

H_{r}

is as in 3.5 . For

α \in A_{N}

with

q_{N} (α) = 1

, we define Heegner divisors

D_{α}

and

ℋ_{α}

D_{α} = \sum_{δ} D_{δ}, ℋ_{α} = \sum_{δ} H_{δ}

where

δ

varies over the set of

(- 4)

-vectors in

N

with

δ / 2 m o d N = α

. Since

H_{r} = D_{δ} \cap ℬ = D_{ρ (δ)} \cap ℬ

, we have

\begin{matrix} 2 ℋ_{α} = D_{α} | ℬ . \end{matrix}

(5.5)

6 Automorphic forms

6.1

Let

{f_{α}}_{α \in A_{N}}

be a vector valued elliptic modular form of weight

- 4

and of type

ρ

, i.e.,

f_{α}

is a holomorphic function on the upper half plane satisfying

\begin{matrix} f_{α} (τ + 1) = e^{2 π i q (α)} f_{α} (τ), f_{α} (- 1 / τ) = \frac{τ^{- 4}}{8} \sum_{β \in A_{N}} e^{- 2 π i 〈 α, β 〉} f_{β} . \end{matrix}

(6.1)

Recall that there are three types of vectors in

A_{N}

denoted by type 00, 0 or 1 according to zero, non-zero isotropic or non-isotropic respectively (see 4.3 ) . For each

α \in A_{N}

, we denote by

m_{0}

m_{1}

the number of vectors

β \in A_{N}

with

b (α, β) = 0

1

respectively which is as in the following Table 2:

\begin{matrix} α & 00 & 00 & 00 & 0 & 0 & 0 & 1 & 1 & 1 \end{matrix}

\begin{matrix} β & 00 & 0 & 1 & 00 & 0 & 1 & 00 & 0 & 1 \end{matrix}

\begin{matrix} m_{0} & 1 & 35 & 28 & 1 & 19 & 12 & 1 & 15 & 16 \end{matrix}

\begin{matrix} m_{1} & 0 & 0 & 0 & 0 & 16 & 16 & 0 & 20 & 12 \end{matrix}

\begin{matrix}  \end{matrix}

Table 2 We shall find a modular form

h

such that the components

h_{α}

are given by functions

h_{00}, h_{0}, h_{1}

depending only on the type of

α

. Then it follows from 6.1 and Table 2 that

h = {h_{α}}

satisfies:

h_{00} (τ + 1) = h_{00} (τ), h_{00} (- 1 / τ) = \frac{τ^{- 4}}{8} (h_{00} (τ) + 35 h_{0} (τ) + 28 h_{1} (τ)),

h_{0} (τ + 1) = h_{0} (τ), h_{0} (- 1 / τ) = \frac{τ^{- 4}}{8} (h_{00} (τ) + 3 h_{0} (τ) - 4 h_{1} (τ)),

h_{1} (τ + 1) = - h_{1} (τ), h_{1} (- 1 / τ) = \frac{τ^{- 4}}{8} (h_{00} (τ) - 5 h_{0} (τ) + 4 h_{1} (τ)) .

6.2 Lemma

One solution of these equations is given as follows:

h_{00} (τ) = 56 η (2 τ)^{8} / η (τ)^{16} = 56 + 896 q + 8064 q^{2} + \cdot \cdot \cdot,

h_{0} (τ) = - 8 η (2 τ)^{8} / η (τ)^{16} = - 8 - 128 q - 1152 q^{2} - \cdot \cdot \cdot,

h_{1} (τ) = 8 η (2 τ)^{8} / η (τ)^{16} + η (τ / 2)^{8} / η (τ)^{16} = q^{- 1 / 2} + 36 q^{1 / 2} + 402 q^{3 / 2} + \cdot \cdot \cdot

where

η (τ)

is the Dedekind eta function and

q = e^{2 π \sqrt{- 1} τ}

Proof. The proof is the same as that of [Kon2] , Lemma 4.3. □

6.3

By applying Borcherds [Bor] , Theorem 13.3, for the vector valued modular form

h

given in Lemma 6.2 , we have

6.4 Theorem

There exists an automorphic form of weight

28 (= 56 / 2)

D

which vanishes exactly on Heegner divisors corresponding 28 non-isotoropic vectors in

A_{N}

6.5

On the other hand, by Borcherds [Bor] , Theorem 14.3, we have an

S_{8}

-equivariant map

φ : W \to A_{4} (\tilde{O} (N))

where

W

is the 14-dimensional subspace of

C [A_{N}]^{S L (2, Z)}

given in Remark-Definition 5.5 ,

A_{4} (\tilde{O} (N))

is the space of automorphic forms on

D

of weight 4 with respect to

\tilde{O} (N)

and

\tilde{O} (N) = K e r (O (N) \to O (q_{N})) .

It follows from [N3] , Theorem 1.14.2 that the map from

O (N)

O (q_{N})

is surjective and hence

S_{8}

(\sim = O (q_{N}) \sim = O (N) / \tilde{O} (N))

naturally acts on

A_{4} (\tilde{O} (N))

. On the other hand,

S_{8}

acts on

W

. With respect to these actions,

φ

S_{8}

-equivariant.

6.6 Lemma

The map

φ

is injective.

Proof. The proof is the same as that of [Kon2] , Lemma 4.1. □

6.7 Theorem

Let

V

be a maximal totally singular subspace of

A_{N}

. Let

F_{V}

be an automorphic form of weight

4

D

associated with

f_{V} \in C [A_{N}]^{S L (2, Z)} : F_{V} = φ (f_{V})

. Then

(F_{V}) = \sum_{α \in V, q (α) = 1} D_{α}

where

D_{α}

is the Heegner divisor associated with

α

Proof. Let $Φ$ be the product of all $F_{V}$ where $V$ varies over all maximal totally singular subspaces.
Then $Φ$ is an automorphic form of weight $105 \times 4$ (Lemma 5.6 ). By Theorem 5.4 and the $S_{8}$ -equivariantness of $φ$ , we have $\sum_{α \in V, q (α) = 1} D_{α} \subset (F_{V}) .$ Hence $Φ$ vanishes along $D_{α}$ with vanishing order $\geq 4 \times 105 / 28 = 15$ . On the other hand, the $15$ -th power of the automorphic form given in Theorem 6.4 has the same weight and vanishes on $D_{α}$ with multiplicity 15. The assertion now follows from the Koecher principle. □

7 Cross ratios

7.1

Let

\begin{matrix} τ = (\begin{matrix} τ_{11} & τ_{12} \\ τ_{21} & τ_{22} \\ τ_{31} & τ_{32} \\ τ_{41} & τ_{42} \end{matrix}), τ_{i j} \in {1, 2, . . ., 8} . \end{matrix}

(7.1)

We call

τ

a tableau. A tableau

τ

is called standard if

τ_{i j} < τ_{i j + 1}, τ_{i j} \leq τ_{i + 1 j}

for any

i, j

. The number of tableaus is 105 and the number of standard tableaus is 14 (e.g. [DO] , Chap. I). For each

τ

we define

μ_{τ} = \prod_{1 \leq i \leq 4} d e t (v^{τ_{i 1}} v^{τ_{i 2}})

where

v^{i} \in C^{2}

is a column vector. If

\begin{matrix} v^{i} = (\begin{matrix} 1 \\ x^{i} \end{matrix}), \end{matrix}

(7.2)

then,

μ_{τ} = \prod_{1 \leq i \leq 4} (x^{τ_{i 2}} - x^{τ_{i 1}}) .

These

μ_{τ}

define an

S_{8}

-equivariant map

Θ

from

P_{1}^{8}

P^{13}

(e.g. [DO] ). We identify

P_{1}^{8}

with

\bar{ℬ} / Γ (1 - i)

under the isomorphism given in Theorem 4.11 . In the following, we shall discuss a relation between

Θ

and the map defined by 14-dimensional space

W

of automorphic forms given in § 6 .

7.2

We give a relation between the set of tableaus and the set of totally singular subspaces. Let

K = U \oplus A_{1}^{\oplus 8}

be the sublattice given in Lemma 2.2 . Then

A_{K} = K^{*} / K \sim = (F_{2})^{8}

is generated by

F_{i} / 2

(1 \leq i \leq 8)

corresponding to 8 points on the projective line. The discriminant quadratic form

q_{K}

K

is a map

q_{K} : (F_{2})^{8} \to Q / 2 Z

defined by

q_{K} (x) = 〈 x, x 〉 m o d 2 Z

. Let

θ = (F_{1} + \cdot \cdot \cdot + F_{8}) / 2

which is perpendicular to all vectors in

A_{K}

. Then it is known (Nikulin [N3] , Proposition 1.4.1) that the discriminant quadratic form of

M

is obtained by

q_{M} = q_{K} | θ^{⊥} / θ .

Finally

q_{N}

is canonically isomorphic to

- q_{M}

([N3] , Corollary 1.6.2). Thus non-isotropic vectors with respect to

q_{N}

bijectively corresponds to the vectors

(F_{i} + F_{j}) / 2

(i \neq j)

A_{K}

. Each column

(τ_{i 1}, τ_{i 2})

τ

in 7.1 defines a vector in

(F_{2})^{8}

whose nonzero entries are indexed by

τ_{i 1}, τ_{i 2}

. Thus four columns of

τ

corresponds to mutually orthogonal 4 non-isotropic vectors which generate a maximal totally singular subspace in

A_{N}

. This implies the following:

7.3 Lemma

The set of 105 tableaus

τ

bijectively corresponds to the set of maximal totally singular subspaces of

A_{N}

. Under this correspondence, the zero of

μ_{τ}

coincides with the zero of

F_{V}

where

V

is a maximal totally singular subspace corresponding to

τ

7.4

Consider the linear system of automorphic forms

F_{V}

of dimension 14 defined by

W

(see 6.5 ). Note that the divisor

(F_{V} | ℬ)

is given by

2 \sum_{α \in V, q (α) = 1} ℋ_{α}

(see 5.5 , Theorem 6.7 ). Since

ℬ

is simply connected, we can take a square root of

F_{V}

. Thus we have an automorphic form

G_{V}

of weight 2 on

ℬ

with

\begin{matrix} (G_{V}) = \sum_{α \in V, q (α) = 1} ℋ_{α} . \end{matrix}

(7.3)

Then

{G_{V}}_{V}

defines a map

Ψ

from

\bar{ℬ} / Γ (1 - i)

P^{13}

7.5 Theorem

The map

Θ

coincides with

Ψ

Proof. We shall show that

μ_{τ_{1}} / μ_{τ_{2}}

coincides with

G_{V_{1}} / G_{V_{2}}

for suitable tableaux

τ_{1}, τ_{2}

and the corresponding maximal totally singular subspaces

V_{1}, V_{2}

. We consider the following tableaux:

\begin{matrix} τ_{1} = (\begin{matrix} 1 & 2 \\ 3 & 4 \\ 5 & 6 \\ 7 & 8 \end{matrix}), τ_{2} = (\begin{matrix} 1 & 2 \\ 3 & 4 \\ 5 & 7 \\ 6 & 8 \end{matrix}), τ_{3} = (\begin{matrix} 1 & 2 \\ 3 & 4 \\ 5 & 8 \\ 6 & 7 \end{matrix}) . \end{matrix}

(7.4)

We take a decomposition of

A_{N} = u_{1} \oplus u_{2} \oplus u_{3}

into three hyperbolic planes

u_{1}, u_{2}, u_{3}

defined over

F_{2}

. Let

{e_{i}, f_{i}}

be a basis of

u_{i}

with

〈 e_{i}, e_{i} 〉 = 0, 〈 f_{i}, f_{i} 〉 = 0, 〈 e_{i}, f_{i} 〉 = 1 .

Let

α_{i} = e_{i} + f_{i}

be the non-isotropic vector in

u_{i}

. We may assume that

V_{1} = 〈 α_{1}, α_{2}, α_{3} 〉, V_{2} = 〈 α_{1}, α_{2}, α_{1} + e_{3} 〉, V_{3} = 〈 α_{1}, α_{2}, α_{1} + f_{3} 〉

correspond to

τ_{1}, τ_{2}, τ_{3}

respectively. Then by 7.3 , we can see that

(G_{V_{1}} / G_{V_{2}}) = (μ_{τ_{1}} / μ_{τ_{2}})

as divisors.

Note that the function

μ_{τ_{1}} / μ_{τ_{2}}

takes the value 1 on the divisors defined by

x_{5} - x_{8}

and

x_{6} - x_{7}

. On the other hand, easy calculation shows that

f_{V_{1}} - f_{V_{2}} = f_{V_{3}}

and hence

F_{V_{1}} - F_{V_{2}}

vanishes on the divisor of

F_{V_{3}}

. This implies that

G_{V_{1}} / G_{V_{2}} = 1

on the divisors

ℋ_{a_{1} + f_{3}}

and

ℋ_{a_{2} + f_{3}}

. Hence

G_{V_{1}} / G_{V_{2}} = μ_{τ_{1}} / μ_{τ_{2}}

For any pair of

τ_{1}^{'}, τ_{2}^{'}

, the ratio

μ_{τ_{1}^{'}} / μ_{τ_{2}^{'}}

can be written as the product of some

μ_{τ_{1}} / μ_{τ_{2}}

as above type.

Hence the assertion follows. □

7.6 Theorem

Ψ

is an embedding from

\bar{ℬ} / Γ (1 - i)

into

P^{13}

. The image satisfies

2^{2} \cdot 3 \cdot 5 \cdot 7

quartic relations.

Proof. It is known that $Θ$ is embedding (Koike [Koi] ). The proof of the second assertion is the same as that of [Kon2] , Theorem 7.2, that is, for each non-isotropic vector in $A_{N}$ we have 15 quartic relations.
Since the number of non-isotropic vectors is 28, the second assertion follows. □

7.7 Remark

In the paper [Koi] , by using a computer, he showed that the image is the intersection of 14 quadrics.

7.8 Remark

In the paper [MT] , Matsumoto and Terasoma constructed an

S_{8}

-equivariant map from the moduli space of ordered 8 points on

P^{1}

P^{104}

by using the theta constants related to the curve which is the 4-fold covering of

P^{1}

branched at 8 points. They showed that this map coincides with the map defined by the above 105

μ_{τ}

7.9 Remark

Since 8 points on

P^{1}

naturally correspond to hyperelliptic curves of genus three, we can consider that our case is a degenerate one of smooth curves of genus three. The moduli space of non-hyperelliptic curves of genus three can be also described as an arithmetic quotient of a complex ball ([Kon1] ). On the other hand, Coble constructed a map from the moduli space of curves of genus 3 with level 2-structure to

P^{14}

by using Göpel functions (Coble [C] , Dolgachev, Ortland [DO] , Chap. IX). It would be interesting to extend the result in this paper to the case of curves of genus three.

References

D. Allcock, E. Freitag, Cubic surfaces and Borcherds products, Comm. Math. Helv., 77 (2002), 270–296.
J. H. Conway et al., Atlas of finite groups, Oxford 1985.
R. Borcherds, Automorphic forms with singularities on Grassmannians, Invent. Math. 132 (1998), 491–562.
A. Borel, Some metric properties of arithmetic quotients of symmetric spaces and an extension theorem, J. Diff. Geometry 6 (1972), 543–560.
A. Coble, Algebraic geometry and theta functions, Amer. Math. Soc. Coll. Publ. 10 Providence, R.I., 1929 (3rd ed., 1969).
P. Deligne, G. W. Mostow, Monodromy of hypergeometric functions and non-lattice integral monodromy, Publ. Math. IHES, 63 (1986), 5–89
I. Dolgachev, D. Ortland, Point sets in projective spaces and theta functions, Astérisque 165(1988).
E. Horikawa, On the periods of Enriques surfaces. II, Math. Ann. 235 (1978), 217–246.
K. Kodaira, On compact complex analytic surfaces II, Ann. Math., 77(1963), 563–626. III, Ann. Math., 78(1963), 1–40.
K. Koike, The projective embedding of the configuration space $X (2, 8)$ , preprint.
S. Kond $\bar{o}$ , A complex hyperbolic structure for the moduli space of curves of genus three, J. reine angew. Math., 525(2000), 219–232.
S. Kond $\bar{o}$ , The moduli space of Enriques surfaces and Borcherds products, J. Algebraic Geometry, 11 (2002), 601–627.
K. Matsumoto, T. Terasoma, Theta constants associated to coverings of $P^{1}$ branching at $8$ points, Compositio Math., 140 (2004), 1277–1301.
K. Matsumoto, M. Yoshida, Configuration space of $8$ points on the projective line and a $5$ -dimensional Picard modular group, Compositio Math., 86 (1993), 265–280.
V. V. Nikulin, Factor groups of groups of automorphisms of hyperbolic forms with respect to subgroups generated by $2$ -reflections, J. Soviet Math., 22 (1983), 1401–1475.
V. V. Nikulin, Finite automorphism groups of Kähler $K 3$ surfaces, Trans. Moscow Math. Soc., 38 (1980), 71–135.
V. V. Nikulin, Integral symmetric bilinear forms and its applications, Math. USSR Izv., 14 (1980), 103–167.
I. Piatetski-Shapiro, I. R. Shafarevich, A Torelli theorem for algebraic surfaces of type $K 3$ , Math. USSR Izv., 5 (1971), 547–587.
J. Shah, Degenerations of $K 3$ surfaces of degree $4$ , Trans. A. M. S., 263 (1981), 271–308.
E. B. Vinberg, Some arithmetic discrete groups in Lobachevskii spaces, in ”Discrete subgroups of Lie groups and applications to moduli”, Tata-Oxford (1975), 323–348.

Graduate School of Mathematics, Nagoya University, Nagoya, 464-8602, Japan E-mail address: kondo@math.nagoya-u.ac.jp