Minimal Siegel modular threefolds
Extracted source text ยท arXiv alg-geom/9506017
%Paper: alg-geom/9506017
%From: hulek@math.uni-hannover.de
%Date: Fri, 23 Jun 1995 14:25:59 +0200 (MET DST)
%Date (revised): Wed, 10 Jan 1996 18:32:48 +0100 (MET)
\input amstex
\documentstyle{amsppt}
\magnification=1200
\catcode`\@=11
\redefine\logo@{}
\catcode`\@=13
%\pageno=3
\define\wed{\wedge}
\define \bn{\Bbb N}
\define \bz{\Bbb Z}
\define \bq{\Bbb Q}
\define \br{\Bbb R}
\define \bc{\Bbb C}
\define \bh{\Bbb H}
\define \bp{\Bbb P}
\define \ot{\bp\bh_t^{+}}
\define \gt{\Gamma_t}
\define \gts{\Gamma^*_t}
\define \gti{\widetilde\Gamma_t}
\define \km{\hbox{Km}}
\TagsOnRight
\NoBlackBoxes
\topmatter
\title
Minimal Siegel modular threefolds
\endtitle
\author
Valeri Gritsenko
\footnote{Supported by Institute Fourier, Grenoble,
the Schwerpunktprogramm
``Komplexe Mannigfaltigkeiten'' grant Hu 337/2-4 and DFG grant 436 Rus
17/108/95.
\hfill\hfill}
and
Klaus Hulek
\endauthor
\address
St. Petersburg Department Steklov Mathematical Institute
Fontanka 27,
\newline
${}\hskip 9pt $
191011 St. Petersburg, Russia
\endaddress
\email
gritsenk\@cfgauss.uni-math.gwdg.de;\ \
gritsenk\@pdmi.ras.ru
\endemail
\address
Institut f\"ur Mathematik, Universit\"at Hannover,
Postfach 6009,
\newline
${}\hskip 9pt $
D-30060 Hannover, Germany
\endaddress
\email
hulek\@math.uni-hannover.de
\endemail
\abstract
In this paper we study the maximal extension $\Gamma_t^*$ of the subgroup
$\Gamma_t$ of $\operatorname{Sp}_4
(\bq)$ which is conjugate to the paramodular group.
The index of this extension is $2^{\nu(t)}$ where $\nu(t)$ is the
number of prime divisors of $t$. The group $\Gamma_t^*$ defines the
minimal modular threefold ${\Cal A}_t^*$ which is a finite quotient of
the moduli space ${\Cal A}_t$ of $(1,t)$-polarized abelian surfaces.
A certain degree 2 quotient of
${\Cal A}_t$ is a moduli space of lattice polarized $K3$ surfaces.
The space ${\Cal A}_t^*$ can be interpreted as the space of Kummer
surfaces associated to $(1,t)$-polarized abelian surfaces.
Using the action of $\Gamma_t^*$
on the space of Jacobi forms we show that many spaces between ${\Cal A}_t$
and ${\Cal A}_t^*$ posess a non-trivial 3-form, i.e. the Kodaira
dimension of these spaces is non-negative. Finally we determine the
divisorial part of the ramification locus of the finite map ${\Cal
A}_t\rightarrow {\Cal A}_t^*$ which is a union of Humbert surfaces.
We interprete the corresponding
Humbert surfaces as Hilbert modular surfaces.
\endabstract
\rightheadtext
{Minimal Siegel modular threefolds}
\leftheadtext{V. Gritsenko and K. Hulek}
%\NoRunningHeads
\endtopmatter
\document
\document
\head
Introduction
\endhead
The moduli space ${\Cal A}_t$ of abelian surfaces with a
$(1,t)$-polarization is the quotient of the Siegel upper half plane
$\bh_2$ by a subgroup $\Gamma_t$ of
\,$\operatorname{Sp}_4(\bq)$ which is conjugate to the
paramodular group ${\widetilde \Gamma}_t$.
In \S\ 1 we define an
isomorphism between the symplectic group and the special
orthogonal group $\operatorname{SO}(3,2)$ over the integers.
This exhibits
$\Gamma_t/\{\pm E_4\}$ as a subgroup of the orthogonal group
$\operatorname{SO}(L_t)$ where
$L_t$ is the lattice of rank 5 equipped with the form
$<2t>\hskip-1pt\oplus\, 2\,U$
(here $U$ denotes the hyperbolic plane). Let ${\widehat L}_t$ be the
dual lattice of
$L_t$. The image of $\Gamma_t$ in $\operatorname{O}(L_t)$ acts trivially
on ${\widehat L}_t/L_t$. The orthogonal group
$\operatorname{O}({\widehat L}_t/L_t)$ is isomorphic to
$(\bz/2\bz)^{\nu(t)}$ where ${\nu}(t)$ is the number of prime divisors of
$t$. For every $d||t$ (i.e. $d|t$ and $(d,t/d)=1)$ we construct an
element $V_d$ in $\operatorname{Sp}_4(\br)$.
These elements $V_d$ define a normal
extension $\Gamma_t^*$ of $\Gamma_t$ of index $2^{\nu(t)}$ such that
$\Gamma_t^*/ \Gamma_t\cong \operatorname{O}({\widehat L}_t/L_t)$.
It turns out that
$\Gamma_t^*$ is the maximal normal extension of $\Gamma_t$ as
a discrete subgroup of $\operatorname{Sp}_4(\br)$.
Hence we can consider the moduli space
$\Cal A_t^*=\Gamma_t^*\backslash\bh_2$ as ``minimal'' Siegel modular
threefolds. In $\S\ 1$ we also give a geometric interpretation of the
action of $V_d$ on the moduli space
${\Cal A}_t$. In particular $V_t$ identifies a polarized abelian surface
with its dual. It also turns out that the space
$(\Gamma_t \cup \Gamma_tV_t)\backslash \bh_2$
is isomorphic to the moduli space of
lattice polarized $K3$ surfaces with a polarization of type
$<2t>\hskip-1pt\oplus\, 2E_8(-1)$.
Lattice polarized $K3$ surfaces have been studied by
Nikulin \cite{N2}.
They play a role in mirror symmetry for $K3$ surfaces
(see Dolgachev \cite{D}).
Moreover the variety ${\Cal A}_t^*$ is the space of Kummer surfaces
associated to $(1,t)$-polarized abelian surfaces.
(For a precise statement see Theorem 1.5.)
In $\S\ 2$ we study the action of the elements $V_d$ on the space
of Jacobi forms. This gives rise to a decomposition of the space of Jacobi
forms which was originally found by Eichler and Zagier \cite{EZ}.
Using lifting results due to the first author this enables us
to prove that many moduli spaces lying between ${\Cal A}_t$ and
the minimal Siegel modular threefold ${\Cal A}_t^*$ are not unirational,
resp. have non-negative Kodaira dimension.
This method, however, unfortunately does not give us
information about the Kodaira dimension of ${\Cal A}_t^*$ itself.
If one wants to determine the Kodaira dimension of ${\Cal A}_t^*$ one
needs precise information about the ramification locus of the finite map
${\Cal A}_t\rightarrow {\Cal A}_t^*$.
This turns out to be a difficult problem.
In $\S\ 3$ we determine the divisorial part of this ramification locus for
square free $t$ (the general case can be treated by the same method). The
divisorial part of this ramification locus is a finite union of Humbert
surfaces.
To determine these surfaces we reexamine the theory of Humbert
surfaces from the point of view of the orthogonal group.
This turns out to be a very useful way of studying Humbert surfaces.
An example, originally
due to Brasch, shows that the ramification locus can also contain curve
components. We finally interprete the Humbert surfaces in the ramification
locus as Hilbert modular surfaces.
\vskip 0.5truecm
\head
\S\ 1. The symplectic and orthogonal groups
\endhead
The local isomorphism between the symplectic group
$\operatorname{Sp }_4(\Bbb R)$
and the special orthogonal
group $\operatorname{SO }(3,2)_\Bbb R$ of signature $(3,2)$
is well known.
In this section we define this isomorphism over $\Bbb Z$.
Let us fix a lattice
$$
L=e_1\Bbb Z\oplus e_2\Bbb Z\oplus e_3\Bbb Z\oplus e_4 \Bbb Z.
$$
We identify $l\in L$ with a column-vector in the basis $\{e_i\}$.
$L^2=L\wedge L$ is the lattice of integral bivectors,
which is isomorphic to the lattice of integral skew-symmetric matrices.
The bivector $e_i\wedge e_j$ corresponds to the elementary
skew-symmetric matrix
$E_{ij}$, which has only two non-zero elements $e_{ij}=1$ and
$e_{ji}=-1$.
Any linear transformation $g:\,L\to L$ induces a linear map
$
\wedge^2 g: L\wedge L\to L\wedge L
$
on the $\Bbb Z$-lattice of bivectors.
If $g$ is represented with respect to the basis $\{e_i\}$ by
the matrix $G$, then
$$
(\wedge^2 g)\,(X)=G X \,{}^t G
\qquad\text{for any }\ X=\sum_{i<j}x_{ij}e_i\wedge e_j\in L\wedge L.
$$
One can define a symmetric bilinear form $(X,Y)$ on $L\wedge L$
$$
X\wedge Y=(X,Y)\,e_1\wedge e_2\wedge e_3\wedge e_4\in \wedge^4 L.
$$
It is known, that
$(X, X)=2\, \hbox{Pf\,} (X)$,
where $\hbox{Pf\,}(X)$ is the Pfaffian of the matrix $X$, and
$\hbox{Pf\,}(MX{}^tM)=\hbox{Pf\,}(X)\, \hbox{det\,} M$.
\definition{Definition}The group
$$
\gti=\{g:L\to L\,|\ \wedge^2 g\,(W_t)=W_t,\quad\text{where }
\ W_t=e_1\wedge e_3+t e_2\wedge e_4\} \tag{1.1}
$$
is called the integral {\it paramodular group} of level $t$.
\enddefinition
The lattice $L_t=W_t^{\perp}$ consisting of all elements of $L\wedge L$
orthogonal to $W_t$ has the following basis
$$
L_t=(e_1\wedge e_2,\, e_2\wedge e_3,\, e_1\wedge e_3 -te_2\wedge e_4,\,
e_4\wedge e_1,\, e_4\wedge e_3)\,\Bbb Z^5.
$$
We fix this basis for the rest of the paper.
The symmetric bilinear form $(\cdot, \cdot)$ defines a quadratic form
$S$ of signature $(3,2)$ on the lattice $L_t$, which has
the following form in the given basis
$$
S_t=\pmatrix
0&0&0&0&-1\\
0&0&0&-1&0\\
0&0&2t&0&0\\
0&-1&0&0&0\\
-1&0&0&0&0
\endpmatrix. \tag{1.2}
$$
The group of the real points of the paramodular group is conjugate
to $\operatorname{Sp }_4(\br)$. Thus the determinant of any element of
the paramodular group equals one and $\wed^2 g$ keeps
the bilinear form on $L\wedge L$.
This gives us a homomorphism from the symplectic group in the orthogonal
group of the isometries of the lattice $L_t$
$$
\wed^2:\, \gti \to \operatorname{O}(L_t).
$$
The paramodular group $\gti$ is conjugate to a
subgroup of the usual rational symplectic group:
$$
\gt:=I_t^{-1}\gti I_t=\left\{\pmatrix
* & * & * & t*\\
t* & * & t* & t*\\
{*} & * & * & t*\\
{*} & t^{-1}*& * & *
\endpmatrix
\in \operatorname{Sp }_4 (\bq) \right\},
$$
where all entries $*$ denote integers and
$I_t=\hbox{diag }(1,1,1,t)$.
The quotient space
$$
{\Cal A}_t=\gt\backslash \bh_2
$$
is the coarse moduli space of abelian surfaces with a polarization of
type $(1,t)$.
The composition of the conjugation
with the homomorphism $\wed^2$ defines a homomorphism
$$
\Psi: \gt\to \operatorname{O}(L_t)\qquad\text{where }\
\Psi(g)=\wed^2(I_t\, g \,I_t^{-1}).
\tag{1.3}
$$
One can extend $\Psi$ to the real symplectic group
$\Gamma_t(\br)\cong \operatorname{Sp }_4(\br)$.
Let
$
\widehat L_t=\{u\in L_t\otimes \Bbb Q\,|\
\forall\ l\in L_t\ (l,u)\in \Bbb Z\}
$
be the dual lattice of $L_t$. The discriminant group
$$
A_{t}:=\widehat L_t/L_t = (2t)^{-1} \bz/\bz \cong \bz/2t \bz
$$
is a finite abelian group equipped with a quadratic form
$$
q_t:\, A_t\times A_t\to (2t)^{-1} \bz/2\bz\qquad\qquad
q_t(l,l)\equiv (l,l)_{\widehat L_t}\text{\,mod\,}2\bz
$$
(see \cite{N1} for a general definition).
Any $g\in \operatorname{O}(L_t)$ acts
on the finite group $A_t$.
By
$$
\widehat{\hbox{O}}(L_t)=\{g\in \operatorname{O}(L_t)\,|\
\forall \ell\in \widehat L_t\quad g\ell-\ell\in L_t\}
$$
we denote the subgroup of the orthogonal group consisting of elements
which act identically on the discriminant group.
One can easily prove the next lemma (see \cite{G1}).
\proclaim{Lemma 1.1}The following relations are valid
{\rm 1}. $\Psi(\gt)\subset \widehat{\hbox{\rm SO}}(L_t)
=\widehat{\hbox{\rm O}}(L_t)\cap \operatorname{SO }(L_t)$;
\smallskip
{\rm 2}. $\hbox{\rm Ker}\,\Psi=\{\pm E_4\}$.
\endproclaim
The finite orthogonal group $\operatorname{O}(A_t)$
can be described as follows.
For every $d||t$ (i.e. $d|t$ and $(d,\dsize\frac{t}d)=1$)
there exists a unique ($\operatorname{mod\,} 2t$) integer $\xi_d$
satisfying
$$
\xi_d=-1\, \operatorname{mod }\, 2 d,\quad \xi_d=
1\, \operatorname{mod }\, 2t/d.
$$
All such $\xi_d$ form the group
$$
\Xi (t)=\{\,\xi \operatorname{mod} 2 t\ |\
\xi^2= 1 \operatorname{mod }4 t\,\}
\cong (\bz/2\bz)^{\nu (t)}, \tag{1.4}
$$
where $\nu(t)$ is the number of prime divisors of $t$. It is evident
that
$\operatorname{O}(A_t)\cong \Xi (t)$.
One can take an element in $\operatorname{SO }(L_t)$ realising the
multiplication by
$\xi_d$ on $A_t$. It gives us an element in $\operatorname{Sp }_4(\br)$
with integral $\Psi$-image. For example,
for every $d||t$ we can define $x,y \in \bz$ (which are
not uniquely determined) such that
$$
xd-yt_d=1\quad \operatorname{where }\ t_d=\frac t d.
$$
The matrix
$$
\widetilde{V}_d=\pmatrix
dx & -1 & 0 & 0\\
-yt & d & 0 & 0\\
0 & 0 & d & yt\\
0 & 0 & 1 & dx
\endpmatrix
$$
is an integral symplectic similitude of degree $d$.
We put
$$
V_d=\frac 1{\sqrt d} {\widetilde V}_d \in \operatorname{Sp }_4 (\br).
$$
$V_d$ has the following $\Psi$-image
$$
\Psi({V}_d)=
\pmatrix
1 & 0& 0 & 0 &0\\
0 &d & -2yt &y^2t_d&0\\
0 &-1 &dx+t_dy& -xy &0\\
0 &t_d& -2tx &x^2d &0\\
0 & 0 & 0 & 0 &1
\endpmatrix. \tag{1.5}
$$
We note here that
$$
xd+t_dy= -1\, \operatorname{mod }\, 2d
\qquad\text{and}\qquad xd+t_dy= 1
\, \operatorname{mod }\, 2t_d,
$$
thus $V_d$ induces the multiplication by $\xi_d$ on $A_t$.
It is easy to see, that for all $V_d$ ($d||t$)
$$
V_d^2\in \gt, \qquad V_d\gt V_d=\gt.
$$
I.e. $V_d$ are involutions modulo $\gt$.
Therefore
one can define the following normal extension of the paramodular group
$\gt$.
\definition{Definition}
$\gts$ is the group generated by the elements of $\gt$ and $V_d$
for all $d||t$.
\enddefinition
In accordance with Lemma 1.1 any element
in $\Psi(V_d\gt)$ defines
the same automorphism of $A_t$,
thus
$$
\gts/\gt\cong \operatorname{O}(\widehat L_t/L_t)\cong \Xi (t)
\cong (\bz/ 2\bz)^{\nu(t)}. \tag{1.6}
$$
\medskip
The real orthogonal group
$\operatorname{O}_{\br}(L_t)=\operatorname{O}(L_t\otimes \br)$
acts on a domain
lying on a projective quadric, more exactly on
$$
\bp\bh_t^{\,3}=\bp\bh_{\,L_t}^{\,3}=
\{Z\in \bp(L_t\otimes \bc)\,|\ (Z,Z)=0,\ (Z,\overline Z)<0\}=
\bp{\bh}_t^{+} \cup \overline{\bp\bh}_t^{+},
$$
where
$$
{\bp\bh}_t^{+}=
\{Z={}^t\bigl((t z_2^2-z_1z_3),\,z_3,\,z_2,\,z_1,\,1\bigr)
\cdot z_0\in \bp\bh_t^{3}\
|\ \hbox{Im}\,(z_1)>0\}. \tag{1.7}
$$
This is a classical homogeneuos domain of type IV.
The condition $(Z,\overline Z)<0$ is equivalent to
$$
y_1y_3-ty_2^2>0
\qquad\text{where }\ y_i=\hbox{Im}\,(z_i).
$$
Taking $z_0=1$ one gets the corresponding cylindric domain in the affine
coordinates $(z_i)_{1\le i\le 3}$
$$
{\bh}_t^{+}=
\{Z={}^t\bigl(z_3,\,z_2,\,z_1)\in \bc^3\,|\
\ y_1y_3-ty_2^2>0,\ \ y_1=\hbox{Im}\,(z_1)>0\}.
$$
The domain ${\bh}_t^{+}$ for $t=1$ coincides with
the Siegel upper half-plane $\bh_2$.
For a general $t$ one can define the following isomorphism
of the complex domains
$$
\psi_t: \bh_2\to \bh_t^+\qquad
\psi_t(\pmatrix \tau_1&\tau_2\\\tau_2&\tau_3\endpmatrix)=
{}^t\bigl(\frac{\tau_3}t,\, \frac{\tau_2}t,\, \tau_1\bigr). \tag{1.8}
$$
The linear action of the real orthogonal group
$\operatorname{O}_\br(L_t)$
on $\bp\bh_t^{+}$ defines ``fractional-linear'' transformations
on $\bh_t^+$.
By $\operatorname{O}_\br^+(L_t)$ we denote the subgroup of
index $2$ of
the orthogonal group
consisting of elements which leave $\bp\bh_t^+ $ invariant.
(This is the subgroup of the elements with
real spin norm equal one.)
\proclaim{Proposition 1.2}Let $t$ be square free.
Then $\Psi$ defines the following isomorphisms
$$
\Psi:\ \gts/\{\pm E_4\} \to \operatorname{SO }^+(L_t),
$$
where $\operatorname{SO }^+(L_t)
=\operatorname{SO}(L_t)\cap \operatorname{O }^+_\br(L_t)$, and
$$
\Psi:\ \gt/\{\pm E_4\} \to \widehat{\hbox{\rm SO}}^+(L_t),
$$
where $\widehat{\hbox{\rm SO}}^+(L_t)=\operatorname{SO }^+
(L_t)\cap \widehat{\hbox{\rm SO}}(L_t)$.
Moreover the following diagram is commutative
$$
\CD
\bh_2@>g>>\bh_2\\
@V\psi_t VV @V\psi_t VV\\
\bh_t^+@>\Psi(g)>>\bh_t^+.
\endCD
$$
\endproclaim
\demo{Proof}The diagram is commutative for any
$g\in \operatorname{Sp }_4(\br)$.
To prove this one has to calculate
the images of standard generators of $\operatorname{Sp }_4(\br)$
under $\Psi$.
It is known that for square free $t$,
the group $\hbox{P}\gts\cong \gts/\{\pm E_4\}$ is
a maximal discrete subgroup of the group of analytic
automorphisms of $\bh_2$ and
%\newline
$[\gts : \gt]=2^{\nu(t)}$
(see for example \cite{Al}, \cite{Gu}).
{}From the description of the finite orthogonal group
$\operatorname{O}(A_t)$ given
in \thetag{1.4} we obtain that
$[\operatorname{SO }^+(L_t):\widehat{\hbox{SO}}^+\hskip -2pt(L_t)]
=2^{\nu(t)}$.
The statement of the proposition about the isomorphism of the groups
follows
from the maximality of $\gts$ and Lemma 1.1.
\newline\qed\enddemo
The coset $V_t\,\gt$ (in the case $d=t$ we may take $x=0, y=-1$)
can also be written in the form
$$
V_t\,\gt=
\pmatrix
0 & \sqrt{t}^{-1} & 0 & 0\\
\sqrt{t} & 0 & 0 & 0\\
0 & 0 & 0 & \sqrt{t}\\
0 & 0 & \sqrt{t}^{-1} & 0
\endpmatrix
\gt.
$$
According to \thetag{1.3} and \thetag{1.5} $\Psi(V_t)$ defines
the multiplication by $-1$ on $\widehat{L}_t/L_t$, i.e.
$V_t$ corresponds to the element
$\xi_t=-1$ of $\Xi(t)$ (see \thetag{1.4}).
Therefore
$$
-\Psi({V}_t)\in \widehat{\hbox{O}}(L_t).
$$
Elements $M$ and $-M\in \operatorname{O}(L_t)$ define the same
transformation of the domain $\bh_t^+$. Thus we have
\proclaim{Corollary 1.3}Let $t$ be square free. The groups
$$
\Psi(\gt\cup\gt V_t)\qquad\text{and}\qquad
\operatorname{O }^*(L_t)=\widehat{\hbox{\rm O}}(L_t)\cap
\operatorname{O }^+_\br(L_t)
$$
coincide, if we consider them as groups of analytic
transformations of $\ot $.
\endproclaim
\proclaim{Proposition 1.4}The quotient
$$
(\gt\cup\gt V_t)\setminus \bh_2
$$
is isomorphic to the moduli space of polarized $K3$ surfaces with
a polarization of type $<2t>\oplus\, 2E_8(-1)$.
\endproclaim
\demo{Proof}
It is known that a moduli space of polarized $K3$ surfaces is a quotient
of a $19$-dimensional homogeneous domain of type IV
by an arithmetic group. In the proposition we consider
polarized $K3$ surfaces with a condition on the Picard group
or equivalently on the lattice of its trancendental cycles.
To formulate these conditions we need some definitions (see
\cite{N2}, \cite{D}).
Let $X$ be a $K3$ surface.
Let us take a sublattice $D_t=<2t>\oplus \,E_8(-1)\oplus E_8(-1)$
of the lattice
$$
L_{K3}=
U\oplus U\oplus U\oplus E_8(-1)\oplus E_8(-1)
\cong H^2(X,\Bbb Z),
$$
where $<2t>$ ($t\in \bz$) denotes the one-dimensional lattice
generated by a vector $l$
such that $l^2=2t$,
$U$ is the hyperbolic plane with quadratic form
$\left(\smallmatrix 0&1\\1&0\endsmallmatrix\right)$
and $E_8(-1)$ is the even unimodular lattice of
dimension 8 with the negative definite quadratic form.
We note that
$\hbox{sign\,}(D_t)=(1,16)$ and
$D_{t}^\perp\cong U\oplus U\oplus <-2t>=L'_t=L_t(-1)$.
(Notation $L_t(-1)$ means that we multiply the quadratic form $S_t$
on the lattice $L_t$ by $-1$.)
We recall (see \cite{N1}) that the orthogonal group
$$
\operatorname{O}(D_t,\,L_{K3})=
\{\,g:L_{K3}\to L_{K3}\ |\ g\,|_{D_t}\equiv \hbox{id}\}
$$
is isomorphic to the group
$$
\widehat{\hbox{O}}(L'_t)=\{\,g:\,L'_t\to L'_t\ |\ \forall\ l\in
\widehat{L}'_t\ \ gl- l \in L'_t\,\}.
$$
A marked $D_t$-polarized $K3$-surface is defined by the following
datum (see \cite{N2} and \cite{D} for more details):
a surface $X$, a fundamental domain $C(M)^{+}$ of a group
generated by some $2$-reflections of the lattice $D_t$ acting on
a connected component $V(D_t)^+$ of the cone
$V(D_t)=\{v\in D_t\otimes \br\ |\ (v,v)>0\}$
and an isomorphism of the lattices
$\phi:\,H^2(X,\bz)\to L_{K3}$,
such that
$\phi^{-1}(D_t)\subset\hbox{Pic\,}(X)$,
$\phi^{-1}(V(D_t)^+)\subset V(X)^+$ and
$\phi^{-1}(C(M)^+)$
contains at least one numerically effective divisor class.
By $V(X)^+$ one denotes the connected component of the cone
$$
V(X)=\{v\in H^{1,1}_\br (X)\ |\ (v,\,v)>0\}
$$
containing the cohomology class of a K\"ahler form on $X$.
Let us denote by $\omega_X$ a holomorphic $2$-form
which generates $H^{2,0}(X)$.
Its image under the isometry $\phi$ belongs
to the following domain in the projective space $\Bbb P^{4}$
$$
\phi(\omega_X)\in \Cal D_m=\{v\in \Bbb P(L'_t\otimes\Bbb C):\ (v, v)=0,
\quad (v,\bar v)>0\}.
$$
This domain is an example of the domains of type IV.
Its connected components are isomorphic to the domain
$\Cal H_t^+$ introduced in \thetag{1.7}.
The quotient
$$
\Cal M(<2t>\oplus \,2E_8(-1))=
\operatorname{O}^*(L'_t)\setminus \Cal H_t^{+}
$$
is the moduli space of isomorphism classes of
$<2t>\oplus\, 2E_8(-1)$-polarized $K3$ surfaces.
In accordance with Corollary 1.3
$$
\Cal M(<2t>\oplus \,2E_8(-1))\cong (\gt\cup \gt V_t) \setminus \bh_2.
$$
\qed\enddemo
We now want to explain the relationship between the variety
${\Cal A}_t^*=\Gamma_t^*\setminus \bh_2$ and the space of
Kummer surfaces associated to $(1,t)$-polarized abelian surfaces.
For an abelian surface $A$ we denote by $X=\km (A)$
its associated Kummer surface.
\proclaim{Theorem 1.5} (i) Let $A$, $A'$ be two $(1,t)$-polarized
abelian surfaces which define the same point in ${\Cal A}_t^*$.
Then their Kummer surfaces $X$, $X'$ are isomorphic.
(ii) Assume that the Neron-Severi group of $A$, resp. $A'$ is
generated by the polarization. Then the converse is true:
If $A$ and $A'$ have isomorphic Kummer surfaces,
then $A$ and $A'$ define the same point in ${\Cal A}_t^*$.
\endproclaim
\demo{Proof}We consider the commutative diagram
$$
\CD
\tilde A@>p>>X=\km (A)\\
@V\sigma VV @VVV\\
A@>>>Y=A/\iota
\endCD
$$
where $\tilde A$ is the blow-up of $A$ in the $16$ points of order $2$.
By \cite{BPV, p. 246} the map
$$
\alpha=p_!\circ \sigma^*:H^2(A,\bz)\to H^2(X,\bz)
$$
multiplies the intersection form by $2$ and hence is in particular a
monomorphism. Let $T\subset H^2(X,\bz)$ be the orthogonal complement of
the sublattice of $H^2(X,\bz)$ generated by the polarization
and the nodal classes. Using \cite{BPV, Corollary VIII.5.6}
it follows that $T\cong 2U(2)\oplus <-4t>$.
If $A$ is generic, i.e. the Neron-Severi
group is generated by the polarization,
then $T$ is the transcendental lattice of $X$.
(i) Assume that $A$ and $A'$ define the same point in ${\Cal A}_t^*$.
Then there exists an element $V$ in $\Gamma_t^*$ which induces an
isomorphism of the lattices $T$ and $T'$ of $X$ and $X'$ respectively.
By the Torelli theorem for Kummer surfaces it now suffices
to prove
\enddemo
\proclaim{Claim}The isomorphism $\ V:T\to T'$ can be extended to a
Hodge isometry
\newline
$\widetilde V: H^2(X,\bz)\to H^2(X',\bz)$.
\endproclaim
\demo{Proof of Claim}By definition $T$ is a primitive
non-degenerate sublattice of
the unimodular lattice $H^2(X,\bz)$.
Its orthogonal complement $T^\perp$ in $H^2(X,\bz)$
is an even indefinite lattice of signature $(1,16)$.
The discriminant groups $A_T$ and $A_{T^\perp}$ are isomorphic
and have five generators. By \cite{N1, Theorem 1.14.2}
the homomorphism
$\operatorname{O}(T^\perp)\to
\operatorname{O}(A_{T^\perp},q_{T^\perp})$ is surjective.
Using \cite{N1, Corollary 1.5.2} one can, therefore, construct
an extension of the isomorphism $V$.
\enddemo
(ii) An isomorphism $f:X\to X'$ induces an isomorphism of
transcendental lattices. Under the assumption stated
this defines an isometry $V:T\to T'$ where
$T\cong T'\cong 2U(2)\oplus <-4t>=L_t(-2)$, which is an element
of the group $\operatorname{O}^+(L_t(-2))$.
Since $\operatorname{O}^+(L_t(-2))=\operatorname{O}^+(L_t)$
it follows from Proposition 1.2 that there is an element
$\overline V\in \Gamma_t^*$ which defines one of the elements
$\pm V$. The element
$\overline V$ identifies the points defined by $A$ and $A'$ resp..
\newline\qed
\remark{Remark}The above Theorem justifies it to consider
$\Cal A_t^*$ as the moduli space of Kummer surfaces
associated to abelian surfaces with a $(1,t)$-polarization.
\endremark
\remark{Remark}In \cite{D, Example 6.5} Dolgachev considered the
space of $M_t$-polarized abelian surfaces where $M_t$ is
the orthogonal complement of $2U(2)\oplus <-4t>$ in $L_{K3}$.
This leads to a subgroup of finite index of $\operatorname{O}^+(L_t)$
and hence a covering space of $\Cal A_t^*$.
\endremark
Our next aim is to interprete the involutions $V_d$ geometrically.
Because of Proposition 1.2 the element $V_d$ induces a map from
${\Cal A}_t^*$ to itself.
Let $(A,H)$ be a $(1,t)$-polarized abelian surface. The polarization $H$
defines an isogeny
$$
\align
\lambda_H:\quad A & \rightarrow {\widehat A} = \hbox{Pic}^0 A\\
x & \mapsto T_x^*{\Cal L}\otimes{\Cal L}^{-1}
\endalign
$$
where ${\Cal L}$ is a line bundle representing $H$ and $T_x$
denotes translation by $x$. The map $\lambda_H$ only depends on $H$, not on
the line bundle
${\Cal L}$. There is a (non-canonical) isomorphism
$\hbox{ker\,} \lambda_H\cong
\bz_t \times \bz_t$. For every divisor $d$ of $t$ there is a unique
subgroup $G(d)\subset \hbox{ker\,} \lambda_H$
which is isomorphic to $\bz_d \times
\bz_d$. This subgroup defines a quotient
$$
\lambda_d :\quad A \rightarrow A/G(d)=A'.
$$
If $A$ is given by the period matrix
$$
\Omega = \pmatrix
1 & 0 & \tau_1 & \tau_2\\
0 & t & \tau_2 & \tau_3
\endpmatrix, \qquad \tau =
\pmatrix
\tau_1 & \tau_2\\
\tau_2 & \tau_3
\endpmatrix
\in \bh_2
$$
then $A'$ is given by
$$
\Omega ' = \pmatrix
d & 0 & d\tau_1 & \tau_2\\
0 & t_d & \tau_2 & \tau_3/d
\endpmatrix.
$$
The abelian surface $A'$ carries a uniquely determined
polarization $H'$ with
$$
dH=\lambda_d^* (H').
$$
The polarization $H'$ is of type $e\cdot (1,t/e^2)$ where $e=(d,t_d)$.
Altogether this shows that we have a morphism of moduli spaces
$$
\align
\Phi=\Phi(d):\quad {\Cal A}_t & \rightarrow {\Cal A}_{t/e^2}\\
(A,H) & \mapsto (A',H').
\endalign
$$
If $d=t$ we obtain as a special case the map
$$
\align
\Phi(t):\quad {\Cal A}_t & \rightarrow {\Cal A}_t\\
(A,H) & \mapsto ({\widehat A},{\widehat H})
\endalign
$$
which maps an abelian surface to its dual polarized abelian surface.
\proclaim{Proposition 1.6}
Let $d$ be a divisor of $t$ with $(d,t_d)=1$. Then the map
$$
\Phi(d):\quad {\Cal A}_t \rightarrow {\Cal A}_t
$$
is the map induced by $V_d$.
\endproclaim
\demo{Proof}
For $\tau=
\pmatrix
\tau_1 & \tau_2\\
\tau_2 & \tau_3
\endpmatrix
\in \bh_2$ we have the following formula for the action
$$
V_d<\tau>=
\pmatrix
x & -1\\
-y t_d & d
\endpmatrix
\pmatrix
d\tau_1 & \tau_2\\
\tau_2 & \tau_3/d
\endpmatrix
\pmatrix
x & -y t_d\\
-1 & d
\endpmatrix.
$$
Now consider the matrix
$$
\pmatrix
1 & t_d & 0 & 0\\
y & xd & 0 & 0\\
0 & 0 & x & -y t_d\\
0 & 0 & -1 & d
\endpmatrix
\in \operatorname{SL}_4(\bz).
$$
This matrix transforms the symplectic form
$d e_1\wedge e_3 + t_d e_2\wedge e_4$
into $W_t$ (see \thetag{1.1}).
The claim now follows from the equality
$$
\gather
\pmatrix
x & -1\\
-y t_d & d
\endpmatrix \hskip-1.5pt
\pmatrix
d & 0 & d\tau_1 & \tau_2\\
0 & t_d & \tau_2 & \tau_3/d
\endpmatrix
\hskip-1.5pt\pmatrix
1 & t_d & 0 & 0\\
y & xd & 0 & 0\\
0 & 0 & x & -y t_d\\
0 & 0 & -1 & d
\endpmatrix\\
= \left(\matrix
1 & 0\\
0 & t
\endmatrix\ \ V_d<\tau>\right).
\endgather
$$
\qed\enddemo
\remark{Remark}In view of Theorem 1.5 this shows in particular
that a $(1,t)$-polarized abelian surface and its dual
polarized abelian surface have isormorphic Kummer surfaces.
\endremark
\head
\S\ 2. Nonunirationality of some quotient spaces
\endhead
In accordance
with Proposition 1.4 a moduli space of special
$K3$ surfaces is a quotient of a moduli space of polarized abelian
surfaces. It gives us a double covering
$$
{\Cal A}_{t}\to ({\Gamma}_{t}\cup {\Gamma}_{t} {V}_{t}) \setminus \bh_2.
$$
The degree of the covering
$
{\Cal A}_{t} \to {\Cal A}_{t}^{*}={\Gamma}_{t}^{*}\setminus \bh_{2}
$
has order $2^{\nu(t)}$, where $\nu(t)$ is the number of prime divisors
of $t$. Since for square free $t$ the extension
$\hbox{P}\Gamma_{t}^*$ is
the maximal discrete subgroup of $\operatorname{PSp }_4(\br)$
containing $\hbox{P}\Gamma_t$ the quotient
${\Cal A}_{t}^*$ is the minimal Siegel threefold
associated to the polarization $(1,t)$.
There are $2^{\nu(t)}-2$
other threefolds between ${\Cal A}_{t}$ and ${\Cal A}_{t}^*$.
Let us take for example two primes $p\neq q$ and let
$t=pq$. The involutions $V_p$ and $V_q$ give rise to the following moduli
spaces
$$
{\Cal A}^{(p)}_{pq} = {\Cal A}_{pq} /<V_p>,\
{\Cal A}^{(q)}_{pq}= {\Cal A}_{pq} /<V_q>,\
{\Cal A}^*_{pq} = {\Cal A}_{pq} /<V_p,V_q>\,
=\Gamma^*_{pq}\backslash \bh_2
$$
resp. a commutative diagram
$$
\matrix
& & {\Cal A}_{pq}& &\\
& \swarrow & & \searrow &\\
{\Cal A}^{(p)}_{pq} & & & & {\Cal A}^{(q)}_{pq}\\
& \searrow & & \swarrow & \\
& & {\Cal A}^*_{pq} & &
\endmatrix
$$
where all maps are 2:1.
Using the modular forms constructed in \cite{G1}
we can obtain information about the geometrical type
of some of these moduli spaces.
By $J_{k,t}^{cusp}$ we denote the space of Jacobi cusp forms of weight $k$
and index $t$. In \cite{G1} a lifting was constructed which associates to
a Jacobi form $\Phi \in J_{k,t}^{cusp}$ a cusp form
$F_{\Phi}\in {\Cal M}_k(\widehat{\Gamma}_t)$
of weight $k$ with respect to the group
$$
\widehat{\Gamma}_t=
\left\{\pmatrix
* & t* & * & *\\
{*} & * & * & *t^{-1}\\
{*} & t* & * & *\\
t* & t* & t* & *
\endpmatrix
\in \operatorname{Sp }_4
(\bq) \right\},
$$
where all entries $*$ denote integers.
The groups $\Gamma_t$ and $\widehat{\Gamma}_t$ are conjugate.
Indeed if
$C_t=\hbox{diag\,}(1,\,t^{-1},\,1,\,t)$
then
$
\widehat\Gamma_ t=C_t \Gamma_t C_t^{-1}
$.
For
$$
\widehat{V}_t={}^tV_t=
\pmatrix
0 & \sqrt{t} & 0 & 0\\
\sqrt{t}^{-1} & 0 & 0 & 0\\
0 & 0 & 0 & \sqrt{t}^{-1}\\
0 & 0 & \sqrt{t} & 0
\endpmatrix
$$
it was proved in \cite{G1, formula (2.8)} that
$$
F_{\Phi}(Z)=(-1)^k F_{\Phi}|_k\, \widehat{V}_t(Z).
$$
or equivalently
$$
F_{\Phi} (\pmatrix
\tau_1 & \tau_2\\
\tau_2 & \tau_3
\endpmatrix) =
F_{\Phi} (\pmatrix
t\tau_3 & \tau_2\\
\tau_2 & t^{-1} \tau_1
\endpmatrix).
$$
In this section we describe the behavior of $F_{\Phi}$ with respect
to the group
$$
\widehat\Gamma_t^*=\,<\widehat\Gamma_t, \, \widehat{V}_d\ |\ d||t>,
$$
where $\ \widehat{V}_d=C_t V_d C_t^{-1}$.
Eichler and Zagier \cite{EZ, \S\ 5}, have constructed a decomposition of
the space of Jacobi forms
$$
J_{k,t}^{cusp}=\bigoplus\limits_{\epsilon}\, J_{k,t}^{\epsilon}
$$
where $\epsilon$ runs over all characters of the group
$$
\Xi (t)=\{\,\xi \operatorname{mod\,} 2 t\ |\ \xi^2\equiv 1
\operatorname{mod\,}4t\,\}\cong (\bz/2\bz)^{\nu (t)}
$$
satisfying $\ \epsilon(-1)=(-1)^k$.
For any $d$ with $d || t$ one can define an operator $W_d$
acting on $J_{k,t}$ in the following way \cite{EZ, \S 5}.
For
$$
\Phi(\tau_1,\tau_2)=\sum\Sb n,\,l\in \bz\\ 4nt >l^2 \endSb
f(n,l) \exp (2\pi i (n\tau_1 + l\tau_2))\in J_{k,t}^{cusp}
$$
we put
$$
(\Phi\,|\,W_d)(\tau_1,\tau_2)=\sum\Sb n,\,l\in \bz\\ 4nt > l^2 \endSb
f(n', l') \exp (2\pi i (n \tau_1 + l\tau_2))\in J_{k,t}^{cusp}
$$
where $l', n'$ are determined by
$$
l'\equiv -l \operatorname{\,mod\,} 2d,
\quad l'\equiv l \operatorname{\,mod\,} 2t/d,
\quad 4n' t-{l'}^{2}= 4nt-l^2.
$$
All $W_d$ are involutions. They form a group isomorphic to $\Xi(t)$.
The subspaces $J_{k,t}^{\epsilon}$ are eigenspaces of the operation
$W_d$, namely
$$
J_{k,t}^{\epsilon}=\{\Phi \in J^{cusp}_{k,t}\ |\ \Phi\,|\, W_d
= \epsilon(W_d)\,\Phi\}.
$$
Note that if
$$
\Phi(\tau_1, \tau_2)=\sum_{\mu \operatorname{mod} 2 t} \varphi_{\mu} (\tau_1)
\theta_{t,\mu}(\tau_1,\tau_2)\in J_{k,t}^{\epsilon}
$$
is the standard decomposition of the Jacobi form $\Phi$ with respect to
the theta-functions $\theta_{t,\mu}(\tau_1, \tau_2)$ then for
$\xi_d\in \Xi(t)$
$$
\varphi_{\xi_d\mu}(\tau_1)=\epsilon(\xi_d) \varphi_{\mu} (\tau_1).
$$
\proclaim{Theorem 2.1}
Let $\Phi\in J_{k,t}^{\epsilon}$ be a Jacobi form and $F_{\Phi}\in {\Cal
M}_k (\widehat{\Gamma}_t )$ be its lifting. For any divisor $d$ of $t$ with
$(d, t_d)=1$ the following equality holds
$$
F_{\Phi}|_k \,\widehat{V}_d = \epsilon (\xi_d) F_{\Phi}.
$$
\endproclaim
\demo{Proof}
Let us recall the definition of the lifting $F_{\Phi}$ in terms of the
Fourier expansion \cite{G2}. If
$$
\Phi(\tau_1,\tau_2)=\sum\Sb n,\,l\in \bz\\ 4nt > l^2 \endSb
f(n, l) \exp (2\pi i (n \tau_1 + l\tau_2))\in J_{k,t}^{cusp}
$$
then
$$
F_{\Phi}(Z)=\sum_{N\in \frak{A}_t} b(N)\exp (2\pi i\ tr (NZ))
$$
where summation is taken over all positive definite symmetric matrices
of the following form
$$
N\in {\frak{A}_t}=\left\{ \pmatrix
n & l/2\\
l/2 & mt
\endpmatrix > 0\ |\ n, l, m \in \bz \right\}
$$
and
$$
b( \pmatrix
n & l/2\\
l/2 & mt
\endpmatrix)= \sum_{a\,|\,(n, l, m)} a^{k-1}\
f \left(\frac {n m}{ a^2}, \frac l a \right).
$$
The action of $\widehat{V}_d$ on $F_{\Phi}$ is given by
$$
(F_{\Phi} |_k\, \widehat{V}_d)(Z)
=F_{\Phi}(d^{-1} A_d\ Z\ {} ^tA_d)\quad \hbox{ where }
A_d= \pmatrix
dx & -t\\
-y & d
\endpmatrix
$$
or
$$
\align
(F_{\Phi} |_k \,\widehat{V}_d)(Z)& = \sum_{N\in\frak{A}_t} b(N)
\exp(2\pi i\ tr (d^{-1}\ {} ^tA_d N A_dZ))\\
{}& = \sum_{N\in\frak{A}_t} b(d^{-1}\ {} ^t{\tilde A}_d N
{\tilde A}_d) \exp (2 \pi i\ tr (NZ))
\endalign
$$
where
${\tilde A}_d = d A_d^{-1} =
\pmatrix
d & t\\
y & dx
\endpmatrix$.
Let
$
{\widetilde N}= d^{-1}\ {} ^t{\tilde A}_d N {\tilde A}_d =
\pmatrix
{\tilde n} & {\tilde l}/2\\
{\tilde l}/2 & {\widetilde m} t
\endpmatrix
$.
Clearly $\hbox{det\,} N =\hbox{det\,} {\widetilde N}$.
It is easy to see that the elements
$n$, $l$, $m$ and ${\tilde n}$, ${\tilde l}$, ${\widetilde m}$
have the same set of
common divisors. Moreover
$$
{\tilde l} = l (y t_d + dx) + 2 (nt + xymt)\quad \hbox{and}
\quad{\tilde l}\equiv
\cases -l &\operatorname{\,mod\,} 2d\\
\hphantom{-}l &\operatorname{\,mod\,} 2t_d
\endcases
$$
Hence, by the definition of $J_{k,t}^{\epsilon}$ we have
$$
b(d^{-1}\ {} ^t{\tilde A}_d N {\tilde A}_d)=\sum_{a\,|\,(n, l, m)}
a^{k-1} f\left( \frac {{\tilde n}{\widetilde m}}{a^2},\,
\frac {{\tilde l}}{a}\right) =
\epsilon (\xi_d) b(N)
$$
which proves the theorem.
\newline\qed\enddemo
This result can be used to gain some information on
the Kodaira dimension of moduli spaces.
Whenever we speak of the {\it Kodaira dimension} of some moduli space
${\Cal A}$ we mean the Kodaira dimension of a desingularization of a
projective compactification of ${\Cal A}$.
\proclaim{Corollary 2.2}
Let $p\neq q$ be primes $\ge 5$. Then the Kodaira dimension of at least
one of the spaces ${\Cal A}^{(p)}_{pq}$ or ${\Cal A}^{(q)}_{pq}$ is
$\ge 0$.
\endproclaim
\demo{Proof}
The Eichler-Zagier decomposition gives a decomposition
$$
J_{3, pq} = J_{3, pq}^{(+,-)} \oplus J _{3, pq}^{(-,+)}.
$$
If $p,q \ge 5$ then there exists a cusp form in $J _{3, pq}$ and hence
in $J_{3, pq}^{(+,-)}$ or $J_{3, pq}^{(-,+)}$. By \cite{G1},
\cite{G2} this
can be lifted to a weight 3 cusp form with respect to
$<\Gamma_t,\, V_p>$, resp. $<\Gamma_t,\, V_q>$.
By Freitag's extension theorem this
defines a differential form on any desingularization of a projective
compactification of ${\Cal A}^{(p)}_{pq}$, resp. ${\Cal A}^{(q)}_{pq}$.
This gives the result.
\newline\qed\enddemo
For any integer $t$
let us take a character of the group $\Xi(t)$ isomorphic to
the orthogonal group of the discriminant group of $L_t$
(see \thetag{1.4})
$$
\epsilon:\ \Xi(t)\to \{\pm 1\}.
$$
We define a set $U(\epsilon)=\{\,V_d\in \gts |\ \epsilon(\xi_d)=1\,\}$
and a subgroup
$$
\gt^{\epsilon}=\,<\gt,\ \xi_d\ |\ \xi_d\in U(\epsilon)>\,\subset \gts
$$
of $\gts$.
Theorem 2.1 and the method of the proof of Corollary 2.2 gives us
the next result
\proclaim{Corollary 2.3}If $\hbox{\rm dim\,}(J_{3,t}^{\epsilon})>0$,
then the Kodaira dimension of the quotient space
$$
\Cal A_t^{\epsilon}= \gt^{\epsilon}\,\backslash \bh_2
$$
is nonnegative.
\endproclaim
\remark{Remark}Corollary 2.3 gives information about
$\Cal A_t^{\epsilon}$ only if $V_t\notin U(\epsilon)$.
If $\epsilon(V_t)=\epsilon(-1)=1$, then
$\hbox{dim\,}(J_{3,t}^{\epsilon})=0$.
\endremark
\proclaim{Corollary 2.4}Let $t\ge 21$ ($t\ne 30,\, 36$)
and let its number of prime divisors
$\nu(t)\ge 2$.
Then there exists a finite quotient of
$\Cal A_t$ of degree $2^{\nu(t)-1}$ which is not unirational.
\endproclaim
\demo{Proof}For any integer $t$ from the corollary the dimension
of $J_{3,t}^{cusp}$ is positive. Thus there is a character
$\epsilon$ of $\Xi(t)$ such that $\epsilon(\xi_t)=-1$
and $\hbox{dim}\,J_{3,t}^{\epsilon}>0$.
$\Xi(t)\cong (\bz/2\bz)^{\nu(t)}$ therefore
$[\Gamma_t^{\epsilon}:\gt]=2^{\nu(t)-1}$.
\newline\qed\enddemo
Using dimension formulae for the spaces $J^{\epsilon}_{k,t}$
one can obtain more precise results.
It is easy to get an exact dimension formula using
the trace formula of the operator $W_d$ on the space
$J_{k,t}^{cusp}$ given in \cite{SZ}. By definition of
$W_d$ we have
$$
\hbox{tr\,}(W_d,\ J_{3,t}^{cusp})=
\sum_{\epsilon}\,\epsilon(\xi_d)\,
\hbox{dim\,}(J_{3,t}^{\epsilon}),
$$
where the sum is taken over all characters of $\Xi(t)$.
Therefore
$$
\hbox{dim\,}(J_{3,t}^{\epsilon})=
2^{-\nu(t)}\sum_{d||t}\epsilon(\xi_d) \,
\hbox{tr\,}(W_d,\ J_{3,t}^{cusp}).
$$
For weight $3$ the trace formula of $W_d$ on $J_{3,t}^{cusp}$
(we recall that $d|t$ and $(d,t_d)=1$, where $t_d=\dsize\frac t{d}$\,)
proved in \cite{SZ, Theorem 1} can be reduced to the following
expression
$$
\multline
\hbox{tr\,}(W_d,\ J_{3,t}^{cusp})=
\frac 1{4}\sum_{e|d}H_{t_d}(-4e)-\frac 1{4}\sum_{e'|t_d}H_{d}(-4e')+
\frac 3{2}(H_d(0)-H_{t_d}(0))
\\
+\frac 1{2}\bigr(\delta_2(t_d)H_d(-4)-\delta_2(d)H_{t_d}(-4)\bigl)+
\bigr(\delta_3(t_d)H_d(-3)-\delta_3(d)H_{t_d}(-3)\bigl)\\
+\frac 1{4}\biggr((Q(t_d),2)Q(d) -(Q(d),2)Q(t_d)\biggl).
\endmultline
$$
We denote by $Q(n)$ the greatest integer whose square divides $n$;
$\delta_a(b)=1$ or $0$ if $a|b$ or $a\not| \ b$
and $H_n(\Delta)$ is a generalization of the Hurwitz-Kronecker
class number, i.e $H_1(0)=-\frac 1{12}$
and $H_1(\Delta)$ for $\Delta<0$ is the number of equivalence classes
with respect to $\operatorname{SL}_2(\bz)$ of integral, positive definite,
binary quadratic forms of discriminant $\Delta$,
counting forms equivalent to a multiple of $x^2+y^2$
(resp. $x^2+xy+y^2$) with multiplicity $\frac 1{2}$ (resp. $\frac 1{3}$).
For $n\ge 2$ with $(n,\Delta)=a^2b$ and square free $b$
$$
H_n(\Delta)=\cases
a^2b\biggl(\dsize\frac{\Delta/a^2b^2}{n/a^2b}\biggr)
H_1(\Delta/a^2b^2)
\quad&\text{if } a^2b^2|\Delta\\
0&\text{otherwise,}
\endcases
$$
where $\bigl(\frac{\cdot}{\cdot}\bigr)$ is the generalized
Kronecker symbol.
We note that the trace formula has the simplest form for square free
$t$ coprime to $6$:
$$
\hbox{tr\,}(W_d,\ J_{3,t}^{cusp})=
\frac 1{4}\biggl(\sum_{e|d}\biggl(\frac{-4e}{t_d}\biggr)H_{1}(-4e)-
\sum_{e'|t_d}\biggl(\frac{-4e'}{d}\biggr)H_{1}(-4e')\biggr)+
\frac{t_d-d}{8}.
$$
\example{Example 2.5}The calculation gives us only
thirteen different threefolds
of type $\Cal A_t^{\epsilon}$ with $t$ having only two prime divisors
($t=p^a q^b$),
whose geometric genus could be equal to zero.
They correspond to the following {\it trivial\,} subspaces of
$J_{3,t}^{cusp}$ of type $J_{3,p^a q^b}^{\ \,\,-\,+}$
(this notation means that
$J_{3,p^a q^b}^{\ \,\,-\,+}=J_{3,p^a q^b}^{\epsilon}$ with
$\epsilon(\xi_{p^a})=-1$ and $\epsilon(\xi_{q^b})=1$):
$$
\gather
J_{3,2\cdot 11}^{\ \,-+},\ \
J_{3,2\cdot 13}^{\ \,-+},\ \ J_{3,2\cdot 17}^{\ \,-+},\
\ J_{3,2\cdot 19}^{\ \,-+},\ \ J_{3,2\cdot 25}^{\ \,-+},\ \
J_{3,2\cdot 27}^{\ \,-+},\\
J_{3, 3\cdot 7}^{\ \,-+},\ \ J_{3, 3\cdot 13}^{\ \,-+},\ \
J_{3, 3\cdot 16}^{\ \,-+},\ \
J_{3, 5\cdot 7}^{\ \,-+},\ \ J_{3, 5\cdot 8}^{\ \,-+},\ \
J_{3, 7\cdot 4}^{\ \,-+},\ \ J_{3, 7\cdot 8}^{\ \,-+}.
\endgather
$$
We may add to this list twenty threefolds $\Cal A_t$
with
$t= 1,\,\dots\,,12$, $14$, $15$, $16$, $18$, $20$, $24$,
$30$, $36$ (see \cite{G2})
whose geometric genus could be zero. According to classical
results and new results of M. Gross and S. Popescu (see \cite{GP})
it really is for $t=1,\dots\,,12$, $14$, $16$, $18$, $20$.
\endexample
\example{Example 2.6}We have the following quotients of order
$4$ and $8$ which are not unirational. One has
$$
\hbox{dim}\,J_{3,42}^{cusp}=
\hbox{dim}\,J_{3,2\cdot 3\cdot 7}^{\ \,++-}=1.
$$
Thus
for
$\Cal A_{42}^{(2,3)}=<V_2, V_3>\setminus \,\Cal A_{42}$
we obtain $h^{3,0}(\Cal A_{42}^{(2,3)})\ge 1$.
The geometric genus of all four threefolds of type
$<V_a, V_b, V_c>\setminus\, \Cal A_{210}$, where
$a,\ b,\ c\in \{ 2,\,3,\,5,\,7 \}$ is positive.
\endexample
\vskip0.5truecm
\head
\S\ 3. The Humbert surfaces and the ramification locus
\endhead
If one wants to determine the Kodaira dimension of the variety
${\Cal A}^*_t=\gts\backslash \bh_2$ it is important to
know the ramification locus of the covering map ${\Cal A}_t\rightarrow
{\Cal A}^*_t$, i.e. the locus where the stabilizer of the finite group
$\gts/ \gt$ is not trivial.
Unfortunately this turns out to be a difficult question.
Here we shall give a partial answer, i.e. we shall determine
the divisorial part of the ramification locus which is a union of
a finite numbers of Humbert surfaces. We shall restrict ourselves
to $t$ square free.
First we collect some known facts about divisors
on the homogeneous domain $\ot$.
For any $v\in L_t\otimes\br$ we set
$$
\Cal H_v=\{Z\in \ot \ |\ v\cdot Z=0\},
$$
where $v\cdot u=(v, u)_t$ is the bilinear product
corresponding to the quadratic form $S_t$ (see \thetag{1.2}).
\proclaim{Lemma 3.1}
{\rm 1}. $\Cal H_{g v}=g^{-1}\Cal H_v$ for any
$g\in \hbox{\rm O}_\br^+(L_t)$.
{\rm 2.} Let $v\not= 0$, then $\Cal H_v\not= \varnothing$
if and only if $v^2>0$.
{\rm 3}. $\Cal H_{v}\cap \Cal H_{u} \not=\varnothing$
if and only if the matrix
$\pmatrix v^2&v\cdot u\\v\cdot u&u^2\endpmatrix$
is positive define.
\endproclaim
\demo{Proof} 1. The first property is trivial.
2. The orthogonal group $\operatorname{O }^+_\br(L_t)$
acts transitively on $\ot$.
Thus any $Z\in \ot$ can be reduced to
$Z_{\bold i}={}^t(1,i,0,i,1)$ in the coordinates
$(z_i)$ from \thetag{1.7}.
If $v=(a,b,c,d,e)$ and $v\cdot Z_{\bold i}=0$,
then $a=-e$ and $b=-d$.
Thus $v^2=2a^2+2b^2+2tc^2>0$.
Let $L_t\otimes \br=\br v\oplus V$ with $v^2>0$. One has
$\hbox{sign\,}(V)=(2,2)$.
The group
$\operatorname{SO }^+(V)\cong \operatorname{SO }_\br^+(2,2)$
is locally isomorphic to
$\operatorname{SL}_2(\br)\times \operatorname{SL}_2(\br)$.
Thus
$$
\Cal H_v\cong \bp\bh_{V}^+\cong \bh_1\oplus \bh_1,
$$
where $\bh_1$ is the usual upper half-plane.
This proves the second statement.
3. Let us suppose that
$\Cal H_{v}\cap \Cal H_{u} \not=\varnothing$.
It follows from $(xu-v)^2>0$ that the matrix in 3 is
positive definite.
If the symmetric bilinear form on the plane
$P=\br v\oplus \br u\ $ is
positive definite,
then $\hbox{sign\,}(P^{\perp})=(1,2)$.
The group
$\operatorname{SO }^+(P^{\perp})\cong
\operatorname{SO }^+(1,2)_{\br}$
is locally isomorphic to $\operatorname{SL}_2(\br)$
and
$$
\Cal H_{v}\cap \Cal H_{u}\cong \bp\bh^+_{P^{\perp}}\cong \bh_1.
$$
\qed\enddemo
\remark{Remark}For $l\in L_t$ ($l^2>0$)
the group $\widehat{\hbox{SO}}^+\hskip -2pt(l^{\perp})$
is isomorphic to a subgroup
of $\operatorname{SL}_2(\bz)\times \operatorname{SL}_2(\bz)$ or
to a subgroup of a Hilbert modular group.
\endremark
\definition{Definition}Let $\ell\in \widehat L_t$ be a vector
in the dual lattice.
The {\it Humbert surface} $H_\ell$ is defined by
$$
H_\ell =\pi\,( \bigcup\Sb
g\in \widehat{SO}^+\hskip-1pt (L_t)\endSb \Cal H_{g \ell}\,),
$$
where $\pi:\, \ot\to \widehat{\hbox{SO}}^+\hskip -2pt(L_t)\setminus\ot$
is the natural projection.
\enddefinition
$\Cal H_\ell$ depends only on the one dimensional lattice
$\Bbb Z \ell$, thus we can restrict ourselves to primitive
vectors $\ell \in \widehat{L}_t$. The primitivity means that
$\ell/d\not\in \widehat{L}_t$ for any interger $d>1$.
The first statement of Lemma 3.1 says that there is a one to one
correspondence between the
$\widehat{\hbox{SO}}^+\hskip -2pt(L_t)$-orbits
of primitive vectors $\ell\in \widehat{L}_t$
with positive norm and the Humbert surfaces.
It is well known that for any even integral lattice $L$
with two hyperbolic planes
(in particular for $L_t$) the
$\widehat{\hbox{SO}}(L)$-orbit of any $l\in L$ depends only on
the norm of $l$ and its canonical image
$l^*:=l/\hbox{div\,}(l)$ in
the discriminant group $\widehat L/L$,
where the {\it divisor}
$\hbox{div\,}(l)\in \bn$ of $l$ is the positive generator
of the ideal
$\{(x,l)_L\ |\ x\in L\}$. As a corollary we have
\proclaim{Lemma 3.2}Let $\ell_1$, $\ell_2\in \widehat L_t$ be two
primitive vectors
with the same image in the discriminant group
(i.e. $\ell_1-\ell_2\in L_t$). If $\ell_1^2=\ell_2^2$, then
$H_{\ell_1}=H_{\ell_2}$.
\endproclaim
\demo{Proof}If $\ell_1-\ell_2\in L_t$, then
$\hbox{div\,}(2t\ell_1)=\hbox{div\,}(2t\ell_2)$
and
$\widehat{\hbox{SO}}^+\hskip -2pt(L_t)\,\ell_1
=\widehat{\hbox{SO}}^+\hskip -2pt(L_t)\,\ell_2$.
\newline\qed\enddemo
\definition{Definition}Let $\ell$ be a primitive vector of
$\widehat{L}_t$.
The integer $\Delta(\ell)=2t\ell^2$ is called
the {\it discriminant} of $H_\ell$.
\enddefinition
{}From the isomorphism $\widehat{L}_t/L_t\cong \bz/ 2t\bz$
one gets
\proclaim{Corollary 3.3}The number of surfaces $H_\ell$
with fixed discriminant $\Delta=2t\ell^2$,
which are not $\gt$-equivalent,
is equal to the number of solutions
$$
\#\,\{b\operatorname{mod } 2t\,
|\ b^2\equiv \Delta\operatorname{mod } 4t\}.
$$
\endproclaim
The standard definition of the Humbert surfaces
(see \cite{vdG}, \cite{F})
is given in terms of the moduli space of abelian surfaces
with polarization $(1,t)$.
Let us compare both definitions.
According to \thetag{1.8} and Proposition 1.2
we may rewrite the definition of $\Cal H_\ell$
with $\ell=(e, a, -\frac{b}{2t}, c, f)\in \widehat{L}_t$
($(e,a,b,c,f)=1$) in coordinates $(\tau_i)$ of $\bh_2$:
$$
\Cal H'_x=\psi_t^{-1}(\Cal H_\ell)=\{\,
\pmatrix \tau_1&\tau_2\\ \tau_2&\tau_3\endpmatrix\in \bh_2\ |\
(\tau_2^2-\tau_1\tau_3)f+c\tau_3+b\tau_2+ta\tau_1+te=0\,\},
$$
where $x=(te, ta, b, c, f)$.
The number
$$
2t\ell^2=b^2-4f(te)-4c(ta)=\Delta(\Cal H'_x).
$$
is by definition the discriminant of $\Cal H'_x$.
Let us introduce a lattice
$$
N_t=\{(e,a,b,c,f)\in \bz^5\,|\ e,\,a
\equiv 0\operatorname{mod } t\}.
$$
In accordance with
Proposition 1.2 and Lemma 3.1 we have the following decomposition of
the usual (in sense of \cite{vdG}, \cite{F}) Humbert surface
$H_\Delta\subset \Cal A_t=\gt\setminus \bh_2$:
$$
H_\Delta=\pi_t\,\bigl(\bigcup
\Sb x\in N_t,\ primitive\\ \vspace{1\jot}\Delta(x)=\Delta\endSb
\Cal H'_x\,\bigr)\cong
\pi\,\bigl( \bigcup
\Sb 2t\ell^2=\Delta \endSb
\ \bigcup \Sb
g\in \widehat{SO}^+\hskip -1pt(L_t)\endSb
\Cal H_{g \ell}\,\bigr),
$$
where one takes the summation over representatives $\ell$
from the distinct orbits
and $\pi_t$ is the natural projection $\pi_t:\bh_2\to \Cal A_t$.
Thus the surface $\Cal H_\ell$ defined above corresponds to
an irreducible component of the surface $H_\Delta$.
Corollary 3.3 tells us that
the number of the irreducible components of the $H_\Delta$ is equal to
$$
\#\,\{b\operatorname{mod } 2t\,|\ b^2\equiv
\Delta\operatorname{mod } 4t\}.
$$
This gives a new proof of Theorem 2.4 in \cite{vdG} (see p. 212).
In \S\ 1 we fixed a basis of the lattice $L_t$ such that
$$
L_t=U(-1)\oplus U(-1)\oplus <2t>,
$$
where $U(-1)$ is the integral hyperbolic plane with the quadratic form
$\left(\smallmatrix 0&-1\\-1&0\endsmallmatrix\right)$ and $<2t>$ is
the one dimensional
$\bz$-lattice
with even quadratic form $2t$.
By
$L^{(3)}_t$
we denote the orthogonal component in $L_t$
of the first hyperbolic plane
$$
L^{(3)}_t=\bigl(e_2\wedge e_3,\, e_1\wedge e_3 -
te_2\wedge e_4,\,e_4\wedge e_1\bigr)\bz^3\subset L_t.
$$
It is easy to see that in any orbit $\widehat{\hbox{SO}}(L_t)\ell$
there is a vector from $L^{(3)}_t$ and
\newline
$\widehat{\hbox{SO}}^+\hskip -2pt(L_t)\ell=\widehat{\hbox{SO}}(L_t)\ell$.
Thus
any Humbert surfaces can be given in the form
$$
\align
\Cal H_\ell&=\{a z_1+b z_2+c z_3=0\}\subset \ot,
\qquad \Delta(\Cal H_\ell)=2t \ell^2\\
\intertext{or}
\Cal H'_{x}&=\{ta\tau_1+b\tau_2+c\tau_3=0\}\subset \bh_2,\qquad
\Delta(\Cal H'_x)=b^2-4tac=2t \ell^2
\endalign
$$
where $\ell={}^t(0,a,-\frac{b}{2t},c,0)\in \widehat{L}_t$ and
$x=(ta,b,c)\in N_t$.
\medskip
For any $d||t$ we define the following subgroup of $\gts$
and the corresponding quotient space
of the moduli space $\Cal A_t$
$$
\gt^{(d)}={\gt\cup \gt V_{d}},\qquad \qquad
\Cal A_t^{(d)}=\gt^{(d)}\setminus \bh_2.
$$
The ramification locus of the map
${\Cal A}_t\rightarrow {\Cal A}^{(d)}_t$
can consist of components of different dimension. In the next theorem
we describe its {\it divisorial part} $D_t^{(d)}$.
\proclaim{Theorem 3.4}Let $t$ be square free, $d>1$ and
$t_d=\dsize\frac{t}{d}$. Then
$$
D_t^{(d)}=
\cases H_{4d}\cup H_{d}
&\hbox{\rm if \ } \ \biggl(\dsize\frac {d}{4t_d}\biggr)=1\\
H_{4d} & \hbox{\rm if \ } \ \biggl(\dsize\frac {d}{4t_d}\biggr)\ne 1
\ \ \text{\rm and }\ \biggl(\dsize\frac {d}{t_d}\biggr)=1,
\endcases
$$
where $ \biggl(\dsize\frac {a}{b}\biggr)$
is the generalized Kronecker symbol of
the quadratic residue.
\endproclaim
\remark{Remarks}1. For $d=t$
$$
D_t^{(t)}
=\cases H_{4t}\cup H_{t}&
\hbox{if \ } t\equiv 1\operatorname{mod }4\\
H_{4t} &
\hbox{otherwise}.
\endcases
$$
In particular $D_t^{(t)}$ is irreducible if $t\equiv 2$ or
$3\operatorname{mod } 4$ (see Corollary 3.3).
2. For $d=1$ Theorem 3.4 is still true if we denote
by $D_t^{(1)}$ the divisorial part of the branch locus
of the covering $\bh_2\to \Cal A_t$. We note that
$D_t^{(1)}$ was found in \cite{Br} by another method.
\endremark
\proclaim{Corollary 3.5}Let $t$ be square free. The divisorial part
$D^*_t$ of the ramification locus of the map
${\Cal A}_t\rightarrow \Cal A_t^*$,
where
$\Cal A_t^* =\gts\setminus \bh_2$
is the ``minimal'' Siegel modular threefold
corresponding to polarization of type $(1,t)$,
is the union of the following Humbert surfaces
$$
D^*_t=\bigcup_{d|t}\,\bigl( \varepsilon_1(d) H_{4d} \cup
\varepsilon_2(d) H_{d}\bigr),
$$
where $\varepsilon_1(d)=1$ if $\biggl(\dsize\frac d{t_d}\biggr)=1$,
$\varepsilon_2(d)=1$ if $d$ is odd and
$\biggl(\dsize\frac d{4t_d}\biggr)=1$
and they equal $0$ in all other cases. Moreover none of
the above Humbert surfaces are $\gt$-equivalent.
\endproclaim
\demo{Proof of Corollary}We have to prove only the last statement.
If $\ell_1$ and $\ell_2\in\widehat{L}_t$ are two
primitive vectors with norms $\ell_1^2=2/t_{d_1}$,
$\ell_2^2=1/(2t_{d_2})$, then $\ell_1^2\not=\ell_2^2$, since
$t$ is square free.
\newline\qed\enddemo
We break up the proof of Theorem 3.4 into several lemmas.
Let us consider a reflection with respect to a vector
$v\in L_t\otimes \br$:
$$
\sigma_v (x)=x-\frac{2(x,v)}{(v,v)}\,v.
$$
It is known that $\sigma_v\in \operatorname{O}_\br^+(L_t)$ if and only if
$v^2>0$. (This follows from the definition of the real spin norm.)
If $\sigma_v\in \operatorname{O}_\br^+(L_t)$,
then the set $\hbox{Fix\,}_{\ot}(\sigma_v)$
of fix points of $\sigma_v$ on $\bp\bh_t^+$ is a complex surface
$\Cal H_v$. The opposite statement is also true.
\proclaim{Lemma 3.6}Let us suppose that the set of fix points of
$\sigma\in \operatorname{SO }^+_\br(L_t)$ on $\ot$ is a complex surface.
Then $-\sigma$ is a reflection with respect to a vector
$v\in L_t\otimes \br$.
\endproclaim
\demo{Proof}Over $\br$ one can reduce the quadratic form $S_t$
to $S=\hbox{diag\,}(E_3,-E_2)$.
The maximal compact subgroup $K_\br$ of the orthogonal group
$\operatorname{SO }_\br^+(S)$
is isomorphic to $\operatorname{SO }(3)\times \operatorname{SO }(2)$
consisting of all elements which
fix the point $Z_{\bold i}={}^t(0,0,0,i,1)\in \bp\bh_t^+$.
Since the group $\operatorname{SO }^+_\br(S)$ acts
transitively on the homogeneous domain
we can suppose that $\sigma=\hbox{diag\,}(A,B)\in K$ where
$A\in \operatorname{SO }(3)$ and $B\in \operatorname{SO }(2)$.
If $B\not=\pm E_2$, then $B$ has only
one fix point ${\bold i}={}^t(i,1)$ on the projective line.
If $\sigma=\hbox{diag\,}(A,B)$
has at least three fixed points, then $B$ has an
eigenvalue $\lambda$ of order two. $A$ and $B$ are orthogonal,
thus all eigenvalues of $\sigma$ are equal to $\pm 1$.
There are two possibilities for the set of eigenvalues of $\sigma$
$$
\{\,\lambda(\sigma)\,\}=\{\,1, -1, -1, -1, -1\}\quad\text{or}
\quad \{\,1, 1, 1, -1, -1\}.
$$
In the first case $-\sigma$ is a reflection. In the second
case $\sigma$ can be written as a product of two reflections
$\sigma_v \sigma_u$
with orthogonal $u$ and $v$. Thus
$\hbox{Fix\,}_{\bp\bh_t^+}(\sigma)=\Cal H_u\cap \Cal H_v$
and we have proved the lemma
for non-trivial $B$.
If $B=\pm E_2$, then the same arguments show that $\sigma$ is conjugate
to
$$
D=\pmatrix B_1&0&0\\0&\pm 1&0\\0&0&\pm E_2\endpmatrix \
\qquad B_1\in \operatorname{SO }(2) \tag{3.1}
$$
if $\sigma$ has at least two fixed points. If $B_1\not=\pm E_2$,
then $\hbox{Fix\,}_{\bp\bh_t^+}(D)$ is a subset of
$\Cal H_x\cap \Cal H_y$,
where $x$ and $y$ form an orthogonal basis of the plane
of rotation of $B_1\in \operatorname{SO }(2)$.
\newline\qed\enddemo
\proclaim{Lemma 3.7}There is a one to one correspondence between
the irreducible components $H$ of the divisorial part $D_t^{(d)}$
and the surfaces
$H_\ell$ defined by reflections $\sigma_\ell\in \gt V_d$.
\endproclaim
\demo{Proof}By $\Cal H$ we denote an irreducible surface in $\bh_2$
whose image is $H$.
Let us suppose that $\Cal H=\hbox{Fix\,}_{\bh_2}(G)$ with $G\in\gt V_d$.
In accordance with Proposition 1.2 and Lemma 3.6
$\Psi(G)=\sigma_\ell$ is a reflection.
Moreover $\psi_t(\Cal H)=\Cal H_\ell\subset \ot$ and $\sigma_\ell$
induces multiplication by
$\xi_d$ on the discrimnant group $A_{t}$.
The reflection
$$
\sigma_\ell(x)=x-\frac{2(x,\ell)}{(\ell,\ell)}\,\ell
$$
depends only on the line $<\ell>$ defined by $\ell\in L_t$.
It follows from the definition that $\sigma_\ell$
keeps the lattice $L_t$ invariant if and only if
$\ell^2\,|\,2D$ where $D=\hbox{div\,}(\ell)$
(see the definition before Lemma 3.2).
The surface $H$ depends only on the class
$\{\,\gamma G\gamma^{-1}\,|\,\gamma\in \gt\}$ and
$$
\Psi(\gamma G \gamma ^{-1})= \beta \sigma_\ell \beta^{-1}=
\sigma_{\beta \ell}
\qquad \bigr(\gamma \in \gt,\ \beta=\Psi(\gamma )
\in \widehat{\hbox{SO}}^+\hskip -2pt(L_t)\bigl).
$$
\qed\enddemo
Therefore in order to find all Humbert surfaces in the divisorial part
we have to classify the $\widehat{\hbox{SO}}(L_t)$-orbits of vectors
$\ell\in L_t$ with the additional condition
$\ell^2|\,2\hbox{div\,} (\ell)$.
\proclaim{Lemma 3.8}Let $t$ be an arbitrary positive integer and $d||t$.
There is a one to one correspondence between the
$\widehat{\hbox{\rm SO}}(L_t)$-conjugacy classes of
reflections $\sigma_\ell$ in the coset $(-\gt V_d)$
and the orbits of the primitive vectors in $L_t$, which satisfy
the following conditions:
$$
\ell^2=2d \quad\text{ and}\quad
\cases
\hbox{\rm div\,}(\ell)=2t_d&\hbox{ if \ }
\biggl(\dsize\frac {d}{4t_d}\biggr)=1
\\
\hbox{\rm div\,}(\ell)=t_d & \hbox{ if \ }
\biggl(\dsize\frac {d}{t_d}\biggr)=1.
\endcases
$$
\endproclaim
\demo{Proof}
One can suppose that $\ell={}^t(0,a,b,c,0)\in L_t$
and $(a,b,c)=1$. For such $\ell$ the matrix of $\sigma_\ell$
has the following form
$$
\sigma_\ell=\pmatrix
1&0&0&0&0\\
0&1+\frac{2ca}{\ell^2}&-\frac{4tba}{\ell^2}&\frac{2a^2}{\ell^2}&0\\
0&\frac{2cb}{\ell^2}&1-\frac{4tb^2}{\ell^2}&\frac{2ab}{\ell^2}&0\\
0&\frac{2c^2}{\ell^2}&-\frac{4tbc}{\ell^2}&1+\frac{2ac}{\ell^2}&0\\
0&0&0&0&1
\endpmatrix,
$$
where $\ell^2=2tb^2-2ac$.
$-\sigma_\ell\in \operatorname{SO }_\br^+$ if and only if $\ell^2>0$.
On the discriminant group $-\sigma_\ell$ defines
multiplication by
$$
\xi(\ell)=\frac{4t}{\ell^2}b^2-1.
$$
By definition $D=\hbox{div\,}(\ell)=(a,\,2tb,\,c)$, therefore
$D\,|\,2t$ and $D\,|\,\ell^2\,|\,2D$. We put
$$
\ell = {}^t(0,Da_1,\, b,\, Dc_1,0)
\qquad\text{with \ } (a_1, \,\frac{2tb}{D},\, c_1)=1.
$$
We have to consider four cases:
$$
\ell^2=D\quad\text{or}\quad \ell^2=2D\qquad\text{and}
\qquad D\, |\,t\quad\text{or}\quad D\hskip-2pt\not| \ t.
$$
1). Let us suppose that $D\,|\,t$ ($t_D=\dsize\frac{t}D$).
Then we have
$$\gather
\ell^2=2D\ \Longleftrightarrow \ 1=t_Db^2-Da_1c_1\
\Longleftrightarrow \ (D,t_D)=1\,\&\,\biggl( \frac{t_D}{D}\biggr)=1,\\
\xi(-\sigma_\ell)=2t_D b^2 -1 =
\cases -1&\operatorname{mod }\ 2t_D\\
\hphantom{-} 1&\operatorname{mod }\ 2D
\endcases
\Longrightarrow -\sigma_\ell\in V_{t_D}\gt
\endgather
$$
(see \thetag{1.5}).
The case $\ell^2=D$ leads trivially to a contradiction.
2). Let us suppose that $D\,|\,2t$, but $\dsize\frac{t}D\not\in\bz$.
In this case
$D=2D_1$, $D_1|\,t$ and $t_{D_1}$ is odd. For such $D$ we have
$$
\gather
\ell^2=D\Longleftrightarrow 1=t_{D_1}b^2 - 4D_1 a_1c_1
\Longleftrightarrow \ (D_1,t_{D_1})=1\,\&\
(t_{D_1} \hbox{ odd})\,
\&\, \biggl( \dsize\frac{t_{D_1}}{4D_1}\biggr)=1,\\
\xi(-\sigma_\ell)=2t_{D_1} b^2 -1=
\cases -1&\operatorname{mod }\ 2t_{D_1}\\
\hphantom{-} 1&\operatorname{mod }\ 2D_1
\endcases
\Longrightarrow -\sigma_\ell\in V_{t_{D_1}}\gt.
\endgather
$$
The case $\ell^2=2D$ leads to a contradiction to the primitivity of
$\ell$.
\newline\qed\enddemo
For square free $t$ the system of the surfaces $\{\,H_{\ell^*}\}$,
where
$\ell^*=\dsize\frac{\ell}{\hbox{div\,}(\ell)}$
and $\ell$ satisfies
the condition of Lemma 3.8, contains all irreducible components
of the Humbert surfaces from Theorem 3.4.
This finishes the proof of Theorem 3.4.
\newline\qed
The next corollary follows immediately from the proof of Lemma 3.8.
\proclaim{Corollary 3.9}Let $t$ be square free and $d||t$.
If $\biggl(\dsize\frac {d}{t_d}\biggr)=1$ and
$\biggl(\dsize\frac {d}{4t_d}\biggr)\ne 1$ then there is,
up to conjugation with respect to $\gt$,
exactly one, and if $\biggl(\dsize\frac {d}{4t_d}\biggr)=1$
then there are exactly two
involutions in $\Gamma_t V_d$.
They are
$\Psi^{-1}(-\sigma_{\ell_1})$ (in the both cases),
and
$\Psi^{-1}(-\sigma_{\ell_2})$ (in the second case),
where
$$\align
\ell_1&={}^t(0,a_1,\dsize\frac {b_1}{t_d}, c_1,0)\in\widehat{L}_t,
\qquad
\,(a_1,b_1,c_1)=1,\quad db_1^2-t_da_1c_1=1\\
\ell_2&={}^t(0,a_2,\dsize\frac {b_2}{2t_d}, c_2,0)\in\widehat{L}_t,
\qquad
(a_2,b_2,c_2)=1,\quad db_2^2-4t_da_2c_2=1.
\endalign
$$
\endproclaim
\remark{Remarks}
1. It is possible to apply Lemma 3.8, which has been proved for any
integer $t$,
to classify the divisorial part of the ramification locus
of the covering $\Cal A_t\to \Cal A_t^*$ for any integer $t$.
2. Using the same method one can construct divisors
on a homogeneous domain of type IV of any dimension $n$.
\endremark
\medskip
The ramification locus can also have components of smaller
dimension.
The proof of Lemma 3.6 shows us that the orthogonal group
can contain a rotation in the positive definite subplane
of the lattice $L_t$.
\example{Example}(Brasch)
The following example is due to Brasch. It shows that in general the
ramification locus of the map ${\Cal A}_t \rightarrow {\Cal A}^*_t$
contains other components apart from the divisorial part described above.
Let $t\equiv 1 \operatorname{\,mod\,}4$. For an integer $f>0$ put
$$
c=-f^2 t-1,\quad g=f^2.
$$
Then the matrix
$$
N=\pmatrix
-f\sqrt{t} & 1/\sqrt{t} & 0 & f\sqrt{t}\\
c\sqrt{t} & 0 & f\sqrt{t} & f^2 t\sqrt{t}\\
c\sqrt{t} & 0 & f\sqrt{t} & -c\sqrt{t}\\
0 & 1/\sqrt{t} & -1/\sqrt{t} & 0
\endpmatrix
$$
is an element of $\Gamma_t^*$. One immediately checks that
$N^2=-E_4$. The fixed point set $\hbox{Fix } N$ is a curve
(cf. \cite{B, Hilfssatz 2.5.3}). Moreover the curve $\hbox{Fix } N$ is not
contained in the fixed point set $\hbox{Fix } I$ of an involution $I$ in
$\Gamma_t^*$.
This would namely imply that $IN=-NI$. Now using
the explicit form for $N$ which follows from \cite{B, Hilfssatz 2.8}, a
lengthy but straightforward calculation shows $c=-1$, a contradiction.
\endexample
As a next step we want to interprete the surfaces $H_t$, resp. $H_{4t}$ as
moduli spaces of abelian surfaces with real multiplication. It is well
known that there is a close connection between Hilbert modular surfaces
and Humbert surfaces \cite{F}, \cite{vdG, chapter IX}. Here we want to
determine precisely which Hilbert modular surfaces correspond to $H_t$,
resp. $H_{4t}$. Consider the ring $\frak{o}$ of integers in the
number field $\bq(\sqrt{t})$ and recall that
$$
\frak{o} = \bz + \bz \omega,\quad \omega=\frac 1 2 (1+\sqrt{t})\quad
\hbox{if } t\equiv 1 \operatorname{mod} 4
$$
resp.
$$
\frak{o} = \bz + \bz \omega,\quad \omega=\sqrt{t} \quad
\hbox{if } t\not\equiv 1
\operatorname{mod} 4.
$$
The {\it Hilbert modular group} $\operatorname{SL}_2(\frak{o})$ acts on
$\bh_1\times \bh_1$ by
$$
\pmatrix
\alpha & \beta\\
\gamma & \delta
\endpmatrix
(z_1, z_2) = \left(\frac{\alpha z_1 + \beta}{\gamma z_1 + \delta},
\ \frac{\alpha ' z_2 + \beta '}{\gamma ' z_2 + \delta '}\right)
$$
where $\ '$ denotes the Galois automorphism
$\ \sqrt{t} \mapsto - \sqrt{t}$.
\ The quotient space
\newline
$Y=\operatorname{SL}_2(\frak{o})\backslash \bh_1\times \bh_1$ is the
standard {\it Hilbert modular surface} associated to $\bq(\sqrt{t})$.
Let $\sigma$ be the involution which interchanges the two factors of
$\bh_1 \times \bh_1$, i.e.
$$
\sigma (z_1, z_2)=(z_2, z_1).
$$
Then the {\it symmetric} Hilbert modular group is
$$
\operatorname{SL}_2^{\sigma}(\frak{o})
=\operatorname{SL}_2 (\frak{o}) \cup
\sigma \operatorname{SL}_2 (\frak{o})
$$
and $Y^{\sigma} =\operatorname{SL}_2^{\sigma}
(\frak{o})\backslash \bh_1 \times
\bh_1$ is the corresponding symmetric Hilbert modular surface. We shall
first consider the Humbert surface $H_t$. In particular we assume that
$t\equiv 1 \operatorname{mod} 4$. To every point $(z_1, z_2) \in \bh_1
\times \bh_1$ one can associate the lattice
$$
\Lambda_{(z_1, z_2)}=\frak{o}\pmatrix
z_1\\ z_2
\endpmatrix+
\frak{o}\pmatrix
1\\1
\endpmatrix
= \bz
\pmatrix
z_1\\z_2
\endpmatrix+
\bz\pmatrix
\eta z_1\\ \eta ' z_2
\endpmatrix
+ \bz \pmatrix
-\eta ' / \sqrt{t}\\
\eta / \sqrt{t}
\endpmatrix
+ \bz
\pmatrix
\sqrt{t}\\ -\sqrt{t}
\endpmatrix
$$
where $\eta = \sqrt{t} \omega$. The form
$$
E((x_1, x_2), (y_1, y_2))=\operatorname{Im} \left( \frac {x_1
\bar{y}_1}{\operatorname{Im} z_1} + \frac {x_2\bar{y}_2}{\operatorname{Im}
z_2}\right)
$$
defines a Riemann form for $\Lambda_t$. With respect to the basis given
above this is just the alternating form $W_t$. The torus
$$
A_{(z_1, z_2)}=\bc^2/ \Lambda_{(z_1, z_2)}
$$
is hence a $(1,t)$-polarized abelian surface with real multiplication in
$\frak{o}$. The Hilbert modular surface $Y$ is the moduli space of these
objects. The 2:1 cover $Y\rightarrow Y^{\sigma}$ identifies abelian
surfaces whose real multiplication differs by the Galois conjugation. We
have a ``forgetful" map
$$
\Phi : Y^{\sigma} \rightarrow {\Cal A}_t.
$$
\proclaim{Theorem 3.10}
Assume $t\equiv 1 \operatorname{mod} 4$. The Humbert
surface $H_t$ is the image of the symmetric Hilbert modular surface
$Y^{\sigma}$ under the natural morphism $\Phi : Y^{\sigma}
\rightarrow {\Cal A}_t$ which is of degree 1 onto its image.
\endproclaim
Before giving the proof we turn to the Humbert surfaces $H_{4t}$.
Consider the ring
$$
\frak{o}_2 = \bz + \bz \sqrt{t}.
$$
Note that this is an order in $\frak{o}$ if $t\equiv 1
\operatorname{mod} 4$, whereas $
\frak{o}=\frak{o}_2$ if $t\not\equiv 1 \operatorname{mod} 4$. Let
$$
{\tilde \frak{o}}_2 = \frac 12 \bz + \frac 12 \sqrt{t} \bz,\quad {\tilde
\frak{o}}_2^{-1} = 2 \bz + 2\sqrt{t} \bz.
$$
The group
$$
\operatorname{SL}_2 (\frak{o}_2,{\tilde{\frak{o}}}_2)=
\{\pmatrix
a & b\\
c & d
\endpmatrix
|\ a, d \in \frak{o}_2,\ b \in {\tilde{\frak{o}}}_2,\ c \in
{\tilde\frak{o}}_2^{-1}, \ ad-bc=1 \}
$$
acts on $\bh_1 \times \bh_1$ as well as its symmetric counterpart
$$
\operatorname{SL}_2^{\sigma} (\frak{o}_2,
{\tilde{\frak{o}}}_2)=\operatorname{SL}_2 (\frak{o}_2,
{\tilde{\frak{o}}}_2) \cup \sigma \operatorname{SL}_2 (\frak{o}_2,
{\tilde{\frak{o}}}_2).
$$
Let
$$
{\widetilde Y}
=\operatorname{SL}_2(\frak{o}_2,
{\tilde{\frak{o}}}_2)\backslash \bh_1\times \bh_1,
\quad {\widetilde Y}^{\sigma} =
\operatorname{SL}^{\sigma}_2(\frak{o}_2,
{\tilde{\frak{o}}}_2)\backslash \bh_1\times \bh_1
$$
be the corresponding Hilbert modular surfaces.
Again the Riemann form $E$
induces a $(1,t)$-polarization on the tori
$$
A_{(z_1, z_2)}=\bc^2 / \Lambda_{(z_1, z_2)}
$$
where for $t\equiv 1\operatorname{mod} 4$
$$
\Lambda_{(z_1, z_2)}=\bz\pmatrix
z_1\\
z_2
\endpmatrix + \bz\pmatrix
2 \eta z_1\\
2 \eta ' z_2
\endpmatrix+ \bz \pmatrix
- \eta '/ \sqrt{t}\\
\eta/ \sqrt{t}
\endpmatrix + \bz \pmatrix
\sqrt{t} /2\\
-\sqrt{t}/2
\endpmatrix
$$
resp. $t\not\equiv 1 \operatorname{mod} 4$
$$
\Lambda_{(z_1, z_2)}=\bz\pmatrix
z_1\\
z_2
\endpmatrix+ \bz \pmatrix
\omega z_1\\
\omega ' z_2
\endpmatrix+ \bz \pmatrix
1/2\\
1/2
\endpmatrix+ \bz \pmatrix
\omega /2\\
\omega '/2
\endpmatrix\quad
$$
Hence ${\widetilde Y}$, resp. ${\widetilde Y}^{\sigma}$ are moduli spaces
of $(1,t)$-polarized abelian surfaces with real multiplication in
$\frak{o}_2$ and as before we have a canonical map
$$
{\widetilde{\Phi}} : {\widetilde Y}^{\sigma} \rightarrow {\Cal A}_t.
$$
\proclaim{Theorem 3.11}
The Humbert surface $H_{4t}$ is the image of the symmetric Hilbert modular
surface ${\widetilde Y}^{\sigma}$ under the natural map
$\widetilde{\Phi} : {\widetilde Y}^{\sigma}\rightarrow {\Cal A}_t$
which is of degree 1 onto its image.
\endproclaim
\demo{Proof of Theorems 3.10 and 3.11}
We shall treat the case of $H_{4t}$ and $t\equiv 1 \operatorname{mod }4$
in detail and then comment on the other cases. The proof is similar to
the proofs in \cite{HL \S 0}, cf. also \cite{F, Abschnitt 3}. Let
$$
R=\pmatrix
1 & 2\eta\\
1 & 2 \eta '
\endpmatrix
$$
and consider the map
$$
\align
{\widehat{\Phi}} : \bh_1 \times \bh_1 &\rightarrow \bh_2\\
(z_1, z_2) &\mapsto\ ^tR
\pmatrix
z_1 & 0\\
0 & z_2
\endpmatrix R.
\endalign
$$
Then
$$
\operatorname{Im }\widehat{\Phi} =
\{-(t^2-t)\tau_1+2t\tau_2-\tau_3=0\}={\Cal H}'_{4t}
$$
and modulo
$$
X=\pmatrix
1 & 0 & 0 & 0\\
t & 1 & 0 & 0\\
0 & 0 & 1 & -t\\
0 & 0 & 0 & 1
\endpmatrix
\in \Gamma_t
$$
this is equivalent to ${\Cal H}_{4t}=\{t \tau_1-\tau_3=0\}$. Let
$A_{\widehat{\Phi}(z_1, z_2)}$ be the abelian surface associated to the
period matrix
$$
\left( ^tR \pmatrix
z_1 & 0\\
0 & z _2
\endpmatrix \
R \pmatrix
1 & 0\\
0 & t
\endpmatrix \right).
$$
Then $A_{(z_1, z_2)}$ and $A_{\widehat{\Phi}(z_1, z_2)}$ are
isomorphic as polarized abelian surfaces since
$$
^tR
\pmatrix
z_1 & 2\eta z_1 & -\eta'/\sqrt{t} & \sqrt{t}/2\\
z_2 & 2\eta' z_2 & \eta/\sqrt{t} & -\sqrt{t}/2
\endpmatrix=\left(
^tR \pmatrix
z_1 & 0\\
0 & z _2
\endpmatrix\ R
\pmatrix
1 & 0\\
0 & t
\endpmatrix
\right).
$$
Hence $\widehat{\Phi}$ is a lift of the map ${\widetilde{\Phi}}$.
Next we consider the homomorphism
$$
\Psi : \operatorname{SL}_2(\br) \times
\operatorname{SL}_2(\br)\longrightarrow \operatorname{Sp }_4(\br)
$$
$$
\Psi(
\pmatrix
a_1 & b_1\\
c_1 & d_1
\endpmatrix,\
\pmatrix
a_2 & b_2\\
c_2 & d_2
\endpmatrix)=
\pmatrix
^tR & 0\\
0 & R^{-1}
\endpmatrix
\pmatrix
d(a_1, a_2) & d(b_1, b_2)\\
d(c_1, c_2) & d(d_1, d_2)
\endpmatrix
\pmatrix
^tR & 0\\
0 & R
\endpmatrix
$$
where $d(a_1, a_2)=
\pmatrix
a_1 & 0\\
0 & a_2
\endpmatrix
$, etc.
Via the embedding
$$
\align
\operatorname{SL}_2(\bq(\sqrt{t})) & \rightarrow
\operatorname{SL}_2(\br) \times \operatorname{SL}_2(\br)\\
\pmatrix
\alpha & \beta\\
\gamma & \delta
\endpmatrix & \mapsto
(\pmatrix
\alpha & \beta\\
\gamma & \delta
\endpmatrix,\
\pmatrix
\alpha' & \beta'\\
\gamma' & \delta'
\endpmatrix )
\endalign
$$
this also defines a homomorphism
$$
\widehat{\Psi} : \operatorname{SL}_2(\bq(\sqrt{t})) \rightarrow
\operatorname{Sp}_4(\br).
$$
For $J=
\left(\smallmatrix
0 & 1\\
1 & 0
\endsmallmatrix\right)$
we find that
$$
S=\pmatrix
^tRJ\ ^tR^{-1} & 0\\
0 & R^{-1} JR
\endpmatrix \in \Gamma_t.
$$
Clearly $S$ is an involution. Setting $\widehat{\Psi}(\sigma)=S$ we can
extend $\widehat{\Psi}$ to a homomorphism
$$
\widehat{\Psi}_{\sigma} : \operatorname{SL}_2^{\sigma}
(\bq(\sqrt{t}))\rightarrow \operatorname{Sp}_4(\br)
$$
and one checks easily that $\widehat{\Phi}$ is
$\widehat{\Psi}_{\sigma}$-equivariant. Let $G$, resp. $G_{\br}$ be the
stabilizer of ${\Cal H}'_{4t}$ in $\Gamma_t$, resp.
$\operatorname{Sp}_4(\br)$. As in \cite{HL, Lemma 0.8}, cf. also
\cite{F, Korollar 3.2.8} one shows that $G_{\br}$ is the group generated
by the image of $\Psi$ and $S$. The result follows if we can prove that
$G=\widehat{\Psi}_{\sigma} (\operatorname{SL}_2^{\sigma}(\frak{o}_2,
{\tilde\frak{o}}_2))$. For this it is now enough to prove the following.
\enddemo
\proclaim{Lemma 3.12}Let
$\pmatrix
a_i & b_i\\
c_i & d_i
\endpmatrix\in \operatorname{SL}_2(\br); \ i=1,2$ and assume that
$$
M=\pmatrix
^tR & 0\\
0 & R^{-1}
\endpmatrix
\pmatrix
d(a_1, a_2) & d(b_1, b_2)\\
d(c_1, c_2) & d(d_1, d_2)
\endpmatrix
\pmatrix
^tR^{-1} & 0\\
0 & R
\endpmatrix
\in \Gamma_t.
$$
Then $a_1, d_1 \in \frak{o}_2,\ b_1 \in {\tilde\frak{o}}_2,\ c_1 \in
{\tilde\frak{o}}_2^{-1}$ and $ a_2=a_1',\ b_2=b_1',\ c_2=c_1',\ d_2=d_2'$.
\endproclaim
\demo{Proof of the lemma}
We write
$M=
\pmatrix
A & B\\
C & D
\endpmatrix$.
Straightforward calculation gives
$$
A=\frac 1{2(\eta' -\eta)}
\pmatrix
2 \eta' a_1- 2\eta a_2 & -a_1 + a_2\\
4 \eta \eta' (a_1 - a_2) & -2\eta a_1 + 2 \eta' a_2
\endpmatrix.
$$
{}From $A_{12} \in \bz$ we find $(a_1-a_2)/ (2\sqrt{t}) \in \bz$, i.e.
$$
a_2=a_1+2 n \sqrt{t},\quad n\in \bz.
$$
$A_{11}\in \bz$ gives $a_1 \omega' + a_2 \omega \in \bz$.
Hence
$$
a_1+2n\sqrt{t} \omega' = a_1(\omega + \omega') + 2n\sqrt{t} \omega'=
a_1\omega +a_2\omega' \in \bz. \tag{3.2}
$$
I.e. $ a_1\in \bz + \bz\sqrt{t}=\frak{o}_2$. We can write $a_1=\alpha +
\beta \sqrt{t}$ with $ \alpha, \beta \in \bz$. By \thetag{3.2}
we have $a_1+n(\sqrt{t} + t) \in \bz$ and hence $\beta=-n$.
But then $a_2=a_1-2\beta
\sqrt{t} = \alpha-\beta\sqrt{t}=a_1'$. Note also that for $a_1\in
\frak{o}_2$ and
$a_2=a_1'$ one has $A\in \pmatrix
\bz & \bz\\
t\bz & \bz
\endpmatrix$.
Similarly we obtain
$$
B=\pmatrix
b_1 + b_2 & 2 \eta b_1 + 2\eta' b_2\\
2\eta b_1+ 2\eta' b_2 & 4 \eta^2 b_1 + 4 {\eta'}^{2} b_2
\endpmatrix.
$$
Since $B_{11} \in \bz$ we find that $b_2=n-b_1$ for some $n\in \bz$. Using
this and $B_{12}\in t \bz$
one concludes that
$$
(b_1\omega - b_2\omega') \in \frac{\sqrt{t}}{2} \bz.
$$
Hence
$$
b_1-n\omega' = b_1(\omega+\omega')-n\omega'=b_1 \omega - b_2\omega' \in
\frac{\sqrt{t}}{2} \bz. \tag{3.3}
$$
This shows $b_1, b_2 \in {\tilde \frak{o}}_2$. Writing $b_1=(\alpha +
\beta \sqrt{t})/2$ with $\alpha, \beta \in \bz$ one has from \thetag{3.3}
$$
\frac{\alpha}2 + \frac{\beta}2 \sqrt{t} - n\omega' \in \frac{\sqrt{t}}2
\bz
$$
i.e. $\alpha = n$. Hence $b_2=n-b_1=(\alpha-\beta \sqrt{t})/2=b_1'$.
Conversely if $ b_1\in {\tilde\frak{o}}_2$ and $b_2=b_1'$, then
$B\in
\pmatrix
\bz & t\bz\\
t\bz & t\bz
\endpmatrix$.
For $C$ one computes
$$
C=\frac{1}{4t}
\pmatrix
4{\eta'}^{2} c_1+4\eta^2 c_2 & -2\eta' c_1 - 2\eta c_2\\
-2\eta' c_1-2\eta c_2 & c_1 + c_2
\endpmatrix.
$$
Comparing this with the situation for B one finds that $c_1 \in {\tilde
\frak{o}}_2^{-1}, c_2=c_1'$.
Finally
$$
D=\frac{1}{2(\eta'-\eta)}
\pmatrix
2\eta' d_1 - 2\eta d_2 & 4\eta \eta' (d_1-d_2)\\
-d_1 + d_2 & -2\eta d_1 + 2\eta' d_2
\endpmatrix.
$$
This case is analogous to A and one obtains $d_1 \in \frak{o}_2,\
d_2=d_1'$. This proves Lemma 3.12.
\newline\qed\enddemo
The case of the Humbert surface $H_t$, $t\equiv 1 \operatorname{mod } 4$
can be treated in the same manner if we choose
$R=\pmatrix
1 & \eta\\
1 & \eta'
\endpmatrix$.
This is analogous to \cite{HL, \S 0} where the case
$t=5$ was done.
Finally we have to deal with $H_{4t}$ in case $t\not\equiv 1
\operatorname{mod } 4$. Here we can choose
$R=\pmatrix
1 & \omega\\
1 & \omega'
\endpmatrix, \omega = \sqrt{t}$.
In this case
$$
\operatorname{Im } \widehat{\Phi}=\{t\tau_1 - \tau_3=0\}={\Cal H}_{4t}
$$
and the above arguments go through essentially unchanged.
\newline\qed
\Refs
\widestnumber\key {AMRT}
\ref
\key Al
\by N.D. Allan
\paper The problem of the maximality of arithmetic groups
\inbook Algebraic groups and discontinuous subgroups
\bookinfo Proc. of Sympos. Pure Math., vol. IX
\yr 1966
\publ Amer. Math. Soc.
\publaddr Providence
\pages 104--109
\endref
\ref
\key BVP
\by W. Barth, C. Peters, A. van de Ven
\book Compact complex surfaces
\bookinfo Erg. Math. Grenzgeb., 3.Folge, ${\bold 4}$
\yr 1984
\publ Springer Verlag
\endref
\ref
\key B
\by H.-J. Brasch
\book Singularit\"aten von Modulr\"aumen abelscher
Fl\"achen mit Polarisierung von Typ (1,p)
\bookinfo Thesis
\yr 1994
\publ Erlangen
\endref
\ref
\key D
\by I. Dolgachev
\book Mirror symmetry for lattice polarized K3-surfaces
\bookinfo Duke e-prints: alg-geom/9502005
\endref
\ref
\key EZ
\by M. Eichler, D. Zagier
\book The theory of Jacobi forms
\bookinfo Progress in Math. 55
\yr 1985
\publ Birkh\"auser
\endref
\ref
\key F
\by H.-G. Franke
\paper Kurven in Hilbertschen Modulfl\"achen und Humbertsche
Fl\"achen im Siegelraum
\jour Bonner Math. Schriften
\vol 104
\yr 1978
\endref
\ref\key G1
\by V. Gritsenko
\paper Modular forms and moduli spaces of abelian and K3 surfaces
\jour St.Petersb. Math. Jour.
\vol 6:6
\yr 1994
%\lang Russian
\pages 65--102
\endref
\ref\key G2
\by V. Gritsenko
\paper Irrationality of the moduli spaces of polarized abelian surfaces
\jour The International Mathematics Research Notices
\vol 6
\yr 1994
\pages 235--243,
In full form in the Proceedings of the Egloffstein conference (1993)
de Gruyter, Berlin, 1995, pp. 63--81
\endref
\ref
\key GP
\by M. Gross, S. Popescu
\book Calabi Yau treefolds and moduli of abelian surfaces
\bookinfo Preprint
\yr 1995
\publaddr Harvard
\endref
\ref
\key Gu
\by L. Gutnik
\paper On extension of integral subgroups of some groups.
\jour Vestnik Leningradskogo Universiteta, Ser. Matemat.
\vol 12
\yr 1957
\pages 47--78
\endref
\ref
\key HL
\by K. Hulek, H. Lange
\paper The Hilbert modular surface for the ideal $(\sqrt{5})$
and the Horrocks--Mumford bundle
\jour Math. Z.
\vol 198
\yr 1988
\pages 95--116
\endref
\ref
\key vdG
\by G. van der Geer
\book Hilbert modular surfaces
\bookinfo Erg. Math. Grenzgeb., 3.Folge, ${\bold 16}$
\yr 1988
\publ Springer Verlag
%\publaddr Berlin
\endref
\ref
\key N1
\by V. Nikulin
\paper Integral quadratic forms and some of its geometric
applications
\jour Izv. Akad. Nauk SSSR, Ser. Math.
\vol 43
\yr 1979
\pages 103--167
\endref
\ref
\key N2
\by V. Nikulin
\paper Finite groups of automorphisms of K\"ahler K3 surfaces
\jour Proc. Moscow Math. Society
\vol 38
\yr 1980
\pages 71--135
\endref
\ref
\key SZ
\by N.-P. Skoruppa, D. Zagier
\paper Jacobi forms and a certain space of modular forms
\jour Invent. Math.
\vol 94
\yr 1988
\pages 113--146
\endref
\endRefs
\enddocument
\end