Jan Hendrik Bruinier

Borcherds Products on O(2, l)

and Chern Classes of Heegner
Divisors
August 2, 2004

Springer
Berlin Heidelberg NewYork
Barcelona Hong Kong
London Milan Paris
Tokyo
To my parents
Table of Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1 Vector valued modular forms for the metaplectic group . . . 15

1.1 The Weil representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2 Poincaré series and Eisenstein series . . . . . . . . . . . . . . . . . . . . . . 19
1.2.1 Poincaré series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2.2 The Petersson scalar product . . . . . . . . . . . . . . . . . . . . . . 22
1.2.3 Eisenstein series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3 Non-holomorphic Poincaré series of negative weight . . . . . . . . . 27

2 The regularized theta lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.1 Siegel theta functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 The theta integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3 Unfolding against Fβ,m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.4 Unfolding against ΘL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3 The Fourier expansion of the theta lift . . . . . . . . . . . . . . . . . . . . 61

3.1 Lorentzian lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.1.1 The hyperbolic Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2 Lattices of signature (2, l) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3 Modular forms on orthogonal groups . . . . . . . . . . . . . . . . . . . . . . 82
3.4 Borcherds products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4 Some Riemann geometry on O(2, l) . . . . . . . . . . . . . . . . . . . . . . . 93

4.1 The invariant Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 Reduction theory and Lp -estimates . . . . . . . . . . . . . . . . . . . . . . . 101
4.3 Modular forms with zeros and poles on Heegner divisors . . . . . 110

5 Chern classes of Heegner divisors . . . . . . . . . . . . . . . . . . . . . . . . . 117

5.1 A lifting into the cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.1.1 Comparison with the classical theta lift . . . . . . . . . . . . . 133
5.2 Modular forms with zeros and poles on Heegner divisors II . . . 135

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
VIII Table of Contents

Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Introduction

Let (V, q) be a quadratic space of signature (2, l) and L ⊂ V an even lattice.

To simplify this exposition let us further assume that L is unimodular. This
implies that l is even. Around 1994 R. Borcherds discovered a lifting from
nearly holomorphic modular forms of weight 1 − l/2 for SL2 (Z) to mero-
morphic modular forms for the orthogonal group of L, which have infinite
product expansions analogous to the Dedekind eta function (see [Bo1]).
Here, by a nearly holomorphic modular form for SL2 (Z) we mean a func-
tion on the complex upper half plane H = {τ ∈ C; =(τ ) > 0} with the usual
transformation behavior, which is holomorphic on H, but may have a pole at
the cusp ∞. Such a modular form f has a Fourier expansion
X
f (τ ) = c(n)q n ,
n−∞

where q = e2πiτ as usual1 . The Fourier polynomial

X
c(n)q n
0>n−∞

is called the principal part of f .

To describe the image of f under the Borcherds lifting we first recall
some facts on the Hermitean symmetric space associated with the orthogonal
group O(V ) ∼
= O(2, l) of V . We extend the bilinear form (·, ·) attached to the
quadratic form q(·) to a C-bilinear form on the complexification V (C) of V .
The subset

K = [W ] ∈ P (V (C)); (W, W ) = 0, (W, W ) > 0

of the projective space P (V (C)) of V (C) has 2 components. We chose one of

them and denote it by K+ . It is easily verified that the connected component
O+ (V ) of the identity of O(V ) acts transitively on K+ . The stabilizer H of
a fixed point is a maximal compact subgroup and thereby K+ ∼ = O+ (V )/H.
Primitive isotropic vectors in L correspond to rational zero-dimensional
boundary components of K+ in the sense of Baily-Borel. Let z ∈ L be a
1
Confusion with the quadratic form q on V will not be possible.
2 Introduction

primitive isotropic vector, and z 0 ∈ L isotropic with (z, z 0 ) = 1. Then the

⊥
lattice K = L ∩ z ⊥ ∩ z 0 is unimodular and has signature (1, l − 1). The
assignment Z 7→ [Z + z 0 − q(Z)z] defines a biholomorphic map from the
set of Z ∈ K ⊗ C with positive imaginary part Y (i.e. q(Y ) > 0) to K.
One connected component Hl of this set is mapped isomorphically to K+ .
The domain Hl is a tube domain realization of O+ (V )/H in Cl and can be
viewed as a generalized upper half plane. The linear action of O+ (V ) on K+
induces an action on Hl by fractional linear transformations.
The orthogonal group O(L) of the lattice L is an arithmetic subgroup of
O(V ), and the subgroup Γ (L) = O(L) ∩ O+ (V ) acts on Hl . We are interested
in the geometry of the quotient XL = Hl /Γ (L). By the theory of Baily-Borel
it is a quasi-projective algebraic variety over C. There are certain special
divisors on XL , which arise from embedded quotients of type O(2, l − 1). Let
m be a negative integer. The orthogonal complement of a vector λ ∈ L with
q(λ) = m is a quadratic space of signature (2, l − 1). The sum
X
H(m) = λ⊥ ⊂ K+ ⊂ P (V (C))
λ∈L/{±1}
q(λ)=m

defines a Γ (L)-invariant divisor on Hl . It is the inverse image of an algebraic

divisor on XL (also denoted by H(m)). Following Borcherds we call it the
Heegner divisor of discriminant m.
Let f be a nearly holomorphic modular form of weight 1−l/2 with Fourier
coefficients c(n) as before, and assume that c(n) ∈ Z for n < 0. Then the
Borcherds lifting of f is given by the infinite product

Y
c(q(λ))
Ψ (Z) = e (%f (W ), Z) 1 − e((λ, Z))
λ∈K
(λ,W )>0

for Z ∈ Hl (as usual e(τ ) := e2πiτ ). Here %f (W ) ∈ K ⊗ Q is a so-called

Weyl vector, and (λ, W ) > 0 under the product means a certain positivity
condition.
Theorem 0.1 (Borcherds). (See Theorem 13.3 in [Bo2] or Theorem 3.22
in this text.) The product Ψ (Z) converges, if the imaginary part of Z is suf-
ficiently large. It can be continued to a meromorphic function on Hl with the
following properties:
1. The function Ψ is a meromorphic modular form for the group Γ (L) with
some multiplier system.
2. The weight of Ψ is given by the constant term of f . It equals c(0)/2.
3. The divisor of Ψ is determined by the principal part of f . It is given by
the linear combination of Heegner divisors
X
c(m)H(m).
m<0
 Introduction 3

Borcherds’ first proof of this result in [Bo1] was rather indirect. There
he writes down the infinite product, shows that it can be meromorphi-
cally continued with the right poles and zeros using the Hardy-Ramanujan-
Rademacher asymptotics, and proves that it is automorphic by looking at
the transformation behavior under suitable generators of Γ (L). Later the
physicists Harvey and Moore discovered that Borcherds’ lifting can be un-
derstood in the framework of the theta correspondence, if one regularizes the
wildly divergent integrals in the right way [HM, Kon]. This idea was used by
Borcherds in [Bo2] to develop a more conceptual account to his lifting and to
extend it in many ways. We briefly indicate the idea of the latter approach:
The groups SL2 (R) and O(2, l) form a dual reductive pair in the sense
of Howe [Ho]. Thus it is possible to lift automorphic forms on one group to
automorphic forms on the other by integrating against a certain kernel func-
tion, the Siegel theta function ΘL (τ, Z) of the lattice L. This theta function
transforms as a non-holomorphic modular form in the variable τ ∈ H and is
Γ (L)-invariant in the second variable Z ∈ Hl . If f is a nearly holomorphic
modular form of weight 1 − l/2 as above, then its theta lifting is formally
given by Z
dx dy
Φ(Z) = f (τ )ΘL (τ, Z)y ,
y2
F
where F = {τ = x + iy ∈ H; |x| ≤ 1/2, |τ | ≥ 1} denotes the standard
fundamental domain for the action of SL2 (Z) on H. Unfortunately, since f
grows exponentially as =(τ ) → ∞, the integral diverges. However, it can be
regularized by taking the constant term in the Laurent expansion at s = 0 of
the meromorphic continuation in s of
Z
dx dy
lim f (τ )ΘL (τ, Z)y 1+s ,
u→∞ y2
Fu

which converges for <(s)  0. Here Fu = {τ = x + iy ∈ F; y ≤ u} denotes

the truncated fundamental domain. The above limit roughly means that we
first integrate over x and afterwards over y.
The integral representation can be used to compute the singularities of
Φ(Z) and its Fourier expansion for q(Y )  0. It turns out that Φ(Z) =
log |Ψ (Z)|. In that way the above properties of Ψ can be deduced.
To illustrate this result, let us describe a famous example. Let K be
the even unimodular lattice II1,25 and L = K ⊕ II1,1 be the orthogonal
sum of K with a hyperbolic plane. Then l = 26, and by the above theorem
there is a lifting from nearly holomorphic modular forms of weight −12 for
SL
Q2∞(Z) to meromorphic modular forms for the group Γ (L) on H26 . If ∆(τ ) =
q n=1 (1 − q n )24 denotes the classical Delta-function, then we can consider
the Borcherds lifting of
X
1/∆(τ ) = p24 (n + 1)q n = q −1 + 24 + 324q + 3200q 2 + 25650q 3 + . . . .
n
4 Introduction

We find that

Y
p24 (1+q(λ))
Ψ (Z) = e (%, Z) 1 − e((λ, Z))
λ∈K
λ>0

is a holomorphic modular form of weight 12 for the group Γ (L). Its divisor is
equal to H(−1). Here we are in the special situation that 12 is the singular
weight for O(2, 26). Hence all Fourier coefficients a(λ) of Ψ must vanish unless
q(λ) = 0. The latter coefficients can be easily determined. One finds that
X

Ψ (Z) = det(w)∆ (w(%), Z) ,
w∈W

where W is the reflection group of the lattice II1,25 . If we compare the two
equalities, we get the Weyl denominator formula for the fake monster Lie
algebra (see [Bo1]).
In the present book we are mainly interested in geometric aspects of the
Borcherds lifting. The fact that it gives us explicit relations between Heegner
divisors in the divisor class group Cl(XL ) of XL makes it a very useful tool for
the study of the geometry of these divisors. In this context it is convenient to
work with a slightly modified divisor class group f Cl(XL ), which is defined as
the quotient of the group of divisors on XL modulo the subgroup of divisors
coming from meromorphic modular forms for Γ (L) of rational weight r with
some multiplier system.
By Theorem 0.1 we know that if m<0 c(m)q m is a Fourier polynomial,
P
which is the principal part of a nearly holomorphic Pmodular form of weight 1−
l/2, then the corresponding linear combination m<0 c(m)H(m) of Heegner
divisors is 0 in f
Cl(XL ). It is natural to ask, which Fourier polynomials occur
as principal parts of nearly holomorphic modular forms.
In the above example it is clear that every Fourier polynomial occurs as
the principal part of a nearly holomorphic modular form of weight −12. To
see this one can for instance apply the Hecke operator T (n) in weight −12 to
1/∆, to get a nearly holomorphic modular form with principal part q −n . More
generally, this argument shows that every Heegner divisor is trivial in f Cl(XL ),
if there exists a nearly holomorphic modular form of weight 1 − l/2 with
principal part q −1 . However, the situation gets more complicated, if there
are
P non-zero holomorphic cusp forms of complementary weight 1 + l/2. If g =
n n
P
n a(n)q is such a cusp form and f = n c(n)q any nearly holomorphic
modular form of weight 1 − l/2, then

f (τ )g(τ )dτ

is a meromorphic differential form on the Riemann surface H/ SL2 (Z) =

P 1 (C). It has just one pole, which lies at the cusp ∞. The residue is given
by
 Introduction 5
X
C[f g] = c(n)a(−n),
n<0

where C[h] means the constant term of a Laurent series h ∈ C((q)). By the
residue theorem this quantity has to vanish. We get a necessary condition for
the existence of nearly holomorphic modular forms. In particular we see that
there is no such form with principal part q −1 , if there are non-zero cusp forms
of weight 1 + l/2. As an application of Serre duality on Riemann surfaces,
Borcherds proved (more generally for vector valued modular forms) that the
above residue condition is also sufficient (see [Bo3] Theorem 3.1):

Proposition 0.2. There exists a nearly holomorphic P modular form f for

SL2 (Z) of weight k ≤ 2 with prescribed principal part n<0 c(n)q n , if and
only if X
c(n)a(−n) = 0
n<0

a(n)q n of weight 2 − k for SL2 (Z).

P
for all cusp forms g = n

In the easy special case of scalar valued modular forms for SL2 (Z) this
can be proved directly by induction: If k ≥ −12 and k 6= −10, there are
no non-zero cusp forms of weight 2 − k. The quotient of a modular form of
weight k + 12 with Fourier expansion 1 + O(q) and ∆ is a nearly holomorphic
modular form of weight k with principal part q −1 . For instance the above
Hecke operator argument shows that any principal part can be realized.
P let k = −10 or k < −12. Suppose that the Fourier polynomial
Now
p = n<0 c(n)q n satisfies the condition of the proposition. Then the principal
part of the Laurent series p∆ clearly satisfies the condition in weight k + 12.
By induction assumption there is a nearly holomorphic modular form f˜ of
weight k + 12 with the same principal part as p∆. We now show that the
constant terms of p∆ and f˜ also agree. If E is a modular form of weight
2 − k − 12 with E = 1 + O(q), then C[p∆ − f˜] = C[p∆E] − C[f˜E]. But the
first quantity on the right hand side vanishes by assumption and the second
by the residue theorem. Hence f˜ = p∆ + O(q), and f = f˜/∆ is a nearly
holomorphic modular form of weight k with principal part p.
The proposition in particular tells us that whenever there are cusp forms
of weight 1+l/2, then there are many linear combinations of Heegner divisors,
which are not the divisor of a Borcherds product. Here the natural question
arises, whether all relations between Heegner divisors in f
Cl(XL ) are given by
Borcherds products. In other words we ask:

Question 0.3. Let F be any meromorphic modular form for Γ (L) (with some
multiplier system), whose divisor is a linear combination of Heegner divisors
H(m). Is f then a Borcherds product?

The present work was motivated by this question. Before we describe

our approach to the problem, we rephrase it in a more algebraic way. Let
6 Introduction

κ = 1 + l/2 and denote by Sκ the space of cusp forms of weight κ for SL2 (Z).
We write Aκ (Z) for the Z-submodule of the dual space Sκ∗ of Sκ generated
by the functionals
X
ar : g 7→ b(r), for g = b(n)q n ∈ Sκ .
n≥1

The fact that Sκ has a basis of modular forms with coefficients in Z implies
that Aκ (Z) ⊗ C = Sκ∗ . Combining the Borcherds lifting with the proposition,
we get the following result.

Theorem 0.4. The assignment ar 7→ H(−r) defines a homomorphism

η : Aκ (Z) −→ f
Cl(XL ).

The above Question 0.3 is equivalent to the question whether η is injective.

Our approach to this problem consists in the following steps.

1. Extend the regularized theta lifting to Maass wave forms with singulari-
ties at the cusps to find for every Heegner divisor an interesting analytic
object, which can be used in the study of the next two steps.
2. Compose the map η with the Chern class map to the second cohomology
of the space XL .
3. Find an automorphic description of the resulting map from Sκ to the
cohomology in terms of square integrable harmonic differential forms.
Use its properties to obtain a criterion for the injectivity of η.
The first step is carried out in chapters 1 to 3, the second and third in
chapters 4 and 5. We now describe the content of this book in more detail.
We stick to our simplifying assumption that L be unimodular, a much more
general case is considered in the body of the text. Moreover, we assume from
now on that l > 2.
We want to extend the regularized theta lifting to Maass wave forms of
(negative) weight k = 1−l/2 for SL2 (Z) with singularities at the cusps. Recall
that such a Maass wave form is a function f on the upper half plane, which
transforms under SL2 (Z) like a modular form of weight k, but which is only an
eigenfunction of the hyperbolic Laplacian ∆k instead of being holomorphic.
Moreover, we require that f grows slower than O(e2πM y ) as y → ∞ for some
positive integer M .
It can be shown that the space of these Maass wave forms with fixed
eigenvalue is spanned by certain Poincaré series, which we now describe. Let
m be a negative integer, and Mν,µ (y) be the M -Whittaker function. If we
put Ms (y) = y −k/2 M−k/2, s−1/2 (y) for s ∈ C, then Ms (4π|m|y)e(mx) is an
eigenfunction of ∆k with eigenvalue s(1 − s) + (k 2 − 2k)/4. We consider the
Poincaré series of weight k and index m
 Introduction 7

1 X
Fm (τ, s) = [Ms (4π|m|y)e(mx)] |k M,
2Γ (2s)
M ∈Γ∞ \ SL2 (Z)

where Γ∞ = {( 10 n1 ) , n ∈ Z}, and |k denotes the usual Petersson slash

operator in weight k. It is easily seen that the series converges normally for
τ ∈ H and <(s) > 1. It is an eigenfunction of ∆k , since the action of ∆k and
the slash operator commute. The asymptotic behavior of the M -Whittaker
function implies that Fm (τ, s) increases exponentially as e2π|m|y for y → ∞.
We determine the Fourier expansion of Fm (τ, s) explicitly in Theorem 1.9.
For s = 1−k/2 the functions Fm (τ, s) are particularly interesting, because
Fm (τ, 1 − k/2) is annihilated by the Laplacian ∆k and thereby close to being
holomorphic on H. We prove that any nearly holomorphic modular form of
weight k is a linear combination of the Fm (τ, 1 − k/2).
The point is that (in any weight k) any Fourier polynomial can be realized
as the principal part of a linear combination of these Poincaré series. So they
can be viewed as a generalization of nearly holomorphic modular forms.
The Fourier coefficients of the non-holomorphic part of Fm (τ, 1−k/2) can
be identified with the coefficients of holomorphic Poincaré series of weight
κ = 1 + l/2 in Sκ (Proposition 1.16). As a first application of this crucial
observation we derive another proof of Proposition 0.2, which works more
generally for vector valued modular forms. We show that a finite linear com-
bination X
c(m)Fm (τ, 1 − k/2)
m<0
P
is nearly holomorphic, if and only if the functional m c(m)a−m is equal to
zero in Sκ∗ . Similar results in a slightly different setting are well known, see
for instance [He], [Ni].

In view of the above discussion it is natural to investigate the theta lifting

of Fm (τ, 1 − k/2) to the orthogonal group O+ (V ). Unfortunately, in the same
way as for the Borcherds lifting of nearly holomorphic modular forms, the
theta integral diverges because of the exponential growth of Fm (τ, 1 − k/2)
as y → ∞. In section 2.2 we show that it can be regularized as follows. For
s ∈ C with real part greater than 1 − k/2 we define a function Φm (Z, s) on
the generalized upper half plane Hl by
Z
dx dy
Φm (Z, s) = lim Fm (τ, s)ΘL (τ, v)y .
u→∞ y2
Fu

We show that Φm (Z, s) is holomorphic in s and can be continued to a holo-

morphic function on {s ∈ C; <(s) > 1, s 6= 1 − k/2} with a simple pole
at s = 1 − k/2. We define the regularized theta integral Φm (Z) to be the
constant term of the Laurent expansion in s of the function Φm (Z, s) at
s = 1 − k/2. Note that this regularization is different from the regularization
8 Introduction

due to Harvey, Moore and Borcherds described before. Nevertheless, up to an

additive constant it leads to the same lifting. (Hence the present work can be
viewed as an answer to Problem 16.13 in [Bo2].) We compute the singularities
of Φm (Z). We find that it is a real analytic function on Hl − H(m) with a
logarithmic singularity along H(m).
In section 2.3 the theta integral Φm (Z, s) is evaluated by unfolding it
against the Poincaré series Fm (τ, s). Since the integral is not absolutely con-
vergent, we have to make sure that the argument can still be justified. We
obtain a formula for Φm (Z, s) as a Poincaré series for the orthogonal group
Γ (L) involving the Gauss hypergeometric function (Theorem 2.14). Note that
in [Br1, Br2] we started with such a Poincaré series, calculated its Fourier
expansion, and thereby found a connection to certain elliptic modular forms
in a rather indirect way.
We then determine the Fourier expansion of the function Φm (Z, s) us-
ing the same method as Borcherds in [Bo2]. By applying a partial Fourier
transform and the Poisson summation formula, the theta series ΘL can be
rewritten as a Poincaré series involving similar theta series ΘK attached to
the Lorentzian sub-lattice K. Thus the theta integral can also be unfolded
against the kernel ΘL (Theorem 2.15).

In chapter 3 we investigate the Fourier expansion of Φm (Z) in more detail.

We find that it can be written as the sum of two real valued functions ψm (Z)
and ξm (Z). Here ξm (Z) is real analytic on the whole Hl , and − 41 ψm (Z) is the
logarithm of the absolute value of a holomorphic function Ψm (Z) on Hl whose
divisor equals H(m). As in [Br1] this splitting is fundamental for the results
of the present work. In the Fourier expansion of ξm (Z) a certain special
function Vκ (A, B) occurs. It can be regarded as a generalized Whittaker
function, which might be of independent interest.
Observe that the function Ψm (Z) is not necessarily automorphic; by con-
struction we only know that

|Ψm (Z)|e−ξm (Z)/4

is invariant under Γ (L). However, taking suitable finite products of the Ψm ,

one can attain that the main parts of the ξm cancel out. Thereby Theorem
0.1 can be recovered from a cohomological point of view (see Theorem 3.22).
If f is a nearly holomorphic modular form of weight k for SL2 (Z) as before,
then the function Y
Ψm (Z)c(m)
m<0

equals the Borcherds lifting Ψ (Z) of f up to a multiplicative constant of

absolute value 1.

Chapter 4 is the technical heart of this book. We consider the O(2, l)-
invariant Laplace operator Ω acting on functions on the generalized upper
 Introduction 9

half plane Hl . We prove that Φm (Z, s) is an eigenfunction of Ω (Theorem

4.6). This gives a partial answer to Problem 16.6 in [Bo2]. More precisely we
show

Theorem 0.5. The function Φm (Z, s) satisfies

1
ΩΦm (Z, s) = (s − 1/2 − l/4)(s − 1/2 + l/4)Φm (Z, s).
2
The proof relies on the following well known argument: Using the machin-
ery of the Weil representation it can be shown that the Siegel theta function
satisfies the differential equation

∆k ΘL (τ, Z)y l/2 = −2ΩΘL (τ, Z)y l/2

(cf. [Sn]). Thus the assertion formally follows from the self-adjointness of
the Laplacian ∆k and the fact that Fm (τ, s) is an eigenfunction of ∆k . But
since the theta integral Φm (Z, s) does not converge absolutely, we have to
prove step by step that this argument can still be justified for the regularized
integral. As a consequence we find for the regularized function Φm (Z) that

ΩΦm (Z) = c

with some explicit real constant c (Theorem 4.7).

Let F be a meromorphic modular form for Γ (L) of weight r. Suppose
that its divisor is a linear combination of Heegner divisors:
X
(F ) = c(m)H(m).
m<0

Then the function G(Z) = log(|F (Z)|q(Y )r/2 ) is Γ (L)-invariant and has
logarithmic singularities along Heegner divisors. It satisfies the differential
equation ΩG(Z) = −rl/8, similarly as the functions Φm (Z). We are now
ready to state the main result of section 4.3, which is a first step towards
answering Question 0.3 (see Theorem 4.23).

Theorem 0.6. Up to an additive constant the function G is equal to the

regularized theta lifting
1 X
Φ(Z) = − c(m)Φm (Z).
4 m<0

Hence, loosely speaking, any modular form, whose divisor is a linear combina-
tion of Heegner divisors, is given by the regularized theta lifting of a Maass
wave form with singularity at ∞. One might expect that this Maass form
actually has to be holomorphic on H. However, this seems far from being
obvious. We will come back to this problem in section 5.2 and Theorem 0.8.
10 Introduction

In the proof of Theorem 0.6 we first show that the difference G − Φ is a

square-integrable function on the quotient XL . (If l = 3, then it is only in
Lp (XL ) for p < 2.) This requires some reduction theory for the group O(2, l).
In a couple of technical lemmas we define Siegel domains in Hl and derive
some of their important properties. Moreover, we use the properties of the
splitting Φm = ψm + ξm and an application of the Koecher boundedness
principle. (Thus our proof in general breaks down for l ≤ 2. In fact, this is
not the only place where it breaks down.)
In the second step of the proof we exploit the fact that G − Φ is a smooth
solution of the differential equation Ω(G−Φ) = constant. By means of results
of Andreotti-Vesentini and Yau on (sub-) harmonic functions on complete
Riemann manifolds that satisfy certain integrability conditions, we may infer
that G − Φ is constant.
As a corollary (Corollary 4.29) we find that the weight of F is given by
a linear combination of the coefficients of the normalized Eisenstein series of
weight κ for SL2 (Z).

At the beginning of chapter 5 we determine the Chern class of the Heegner

divisor H(m) explicitly. The above properties of ξm and Ψm imply that Ψm (Z)
defines a trivialization of the sheaf L(H(m)) associated with H(m), and
eξm (Z)/4 a Hermitean metric on L(H(m)). Hence the (1, 1)-form
1 ¯ 1 ¯
hm (Z) = ∂ ∂ξm (Z) = ∂ ∂Φ m (Z)
4 4
represents the Chern class of H(m) in H 2 (XL , C) via the de Rham isomor-
phism (Theorem 5.3). Moreover, we show in Theorem 5.5 that hm (Z) is
actually a square integrable harmonic representative.
Together with Theorems 2.12 and 3.9 on the singularities and the Fourier
expansion of Φm (Z) this result implies that Φm (Z) is some kind of Green
current for the divisor H(m) in the sense of Arakelov geometry (see [SABK]).
More precisely this function should define a Green object with log-log-growth
in the extended arithmetic intersection theory due to Burgos, Kramer, and
Kühn [BKK].
Let H1,1 (XL ) denote the space of square integrable harmonic (1, 1)-forms
on XL . By a result of Borel it is a finite dimensional space of automorphic
forms [Bl3]. Write H e 1,1 (XL ) for the quotient of H1,1 (XL ) and the span of
¯
the Kähler form ∂ ∂ log q(Y ). It is a consequence of Theorem 0.6 that the
assignment
H(m) 7→ hm (Z)
defines a linear map from the subspace of f Cl(XL ) generated by the Heegner
divisors H(m) to H e 1,1 (XL ). If we compose it with the map η of Theorem 0.4
and tensor with C, we get a homomorphism ϑ̃ : Sκ∗ → H e 1,1 (XL ).
In section 5.1 we show that an automorphic kernel function for ϑ̃ is es-
sentially given by
 Introduction 11

1 ¯ X
Ω(Z, τ ) = ∂ ∂ log q(Y ) + h−n (Z)q n .
4 n>0

To this end we first observe that the function Ω(Z, τ ) converges normally
and is square integrable in Z. Hence it is clear that Ω(Z, τ ) lies in H1,1 (XL )
for fixed τ . Using the fact that the Fourier coefficients of the Poincaré series
Fm (τ, 1 − k/2) are equal to certain coefficients of holomorphic Poincaré se-
ries in Sκ , we may rewrite Ω(Z, τ ) such that it becomes a sum of Poincaré,
Eisenstein, and theta series in the space of modular forms Mκ of weight κ
for SL2 (Z). Hence for fixed Z the function Ω(Z, τ ) is contained in Mκ (The-
orem 5.8). The idea of rewriting an automorphic kernel function in terms of
Poincaré series appears already in Zagier’s work on the Doi-Naganuma lifting
[Za]. It is related to the “Zagier identity” in [RS]. We prove:

Theorem 0.7. By taking the Petersson scalar product in the τ -variable the
function Ω(Z, τ ) defines a lifting

ϑ : Sκ −→ H1,1 (XL ),


f 7→ f (τ ), Ω(Z, τ ) ,

with the following properties (see Theorem 5.9 for more details):
1. Let f = n c(n)q n be a cusp form in Sκ . Then the image of f has the
P
Fourier expansion
X X
ϑ(f )(Z) = ϑ0 (f )(Y ) − C(κ) |λ|−l nl−1 c(−q(λ)/n2 )
λ∈K n|λ
q(λ)<0
¯ κ (π|λ||Y |, π(λ, Y )) e((λ, X)),
× ∂ ∂V

where C(κ) = 2−κ π 1/2−κ is a constant, and ϑ0 (f )(Y ) the 0-th Fourier
coefficient, which can also be computed explicitly. Moreover, Vκ (A, B)
denotes the generalized Whittaker function defined in (3.25).
2. The function Ω(Z, τ ) is an automorphic kernel function for ϑ̃, that is the
diagram
ϑ /
SO κ H1,1 (XL )


Sκ∗
ϑ̃ /H
e 1,1 (XL )

commutes. Here the left vertical arrow denotes the identification via Pe-
tersson scalar product.

The lifting ϑ can be viewed as a generalization of the Doi-Naganuma map

(from cusp forms of weight 2 for Γ0 (D) to the cohomology of the Hilbert
modular surface in question) [DN, Na, Za, Br1]. Note that in the O(2, 3)-
case of Siegel modular forms of genus 2 such a generalization has been given
12 Introduction

by Piateskii-Shapiro using representation theoretic methods [PS1, PS2]. It

would be interesting to understand the map ϑ in the context of the theory
of Kudla and Millson on intersection numbers of cycles on locally symmetric
spaces (see for instance [KM1, KM2, KM3]).
The description of ϑ in terms of Fourier expansions given in Theorem 0.7
is vital for answering our initial Question 0.3. It can be used to prove that ϑ is
injective in the special case that L splits two hyperbolic planes over Z. (This
is automatically true, if L is unimodular.) Hence, by the second statement
of Theorem 0.7, ϑ̃ is also injective, which finally implies that η is injective as
well. We obtain the main result of section 5.2 (Theorem 5.12).
Theorem 0.8. Assume that L splits two hyperbolic planes over Z. Let F
be a meromorphic modular form for the group Γ (L) (with some multiplier
system), whose divisor is a linear combination of Heegner divisors
X
(F ) = c(m)H(m).
m<0

Then F is a Borcherds product in the sense of Theorem 0.1, i.e. there exists
a
Pnearly holomorphic modular form f of weight 1 − l/2 with principal part
m
m<0 c(m)q , and the Borcherds lifting of f equals F .
We consider the special O(2, 3)-case of the paramodular group of level
t as an example. By Theorem 0.7 we get a lifting from skew-holomorphic
Jacobi forms of index t and weight 3 to square integrable harmonic (1, 1)-
forms on the corresponding quotient, the moduli space of Abelian surfaces
with a (1, t)-polarization.

Except for Theorem 0.8, we prove all the above results under the mild
assumptions that l > 2, and that L splits two orthogonal hyperbolic planes
over Q. This is always true, if l > 4. (We do not require that L is unimodular.)
Then the discriminant group, the quotient L0 /L of the dual lattice L0 modulo
L, is usually non-trivial. This leads to several technical complications. For
instance, in general we have to replace the space Sκ of elliptic cusp forms by
a certain space Sκ,L of C[L0 /L]-valued cusp forms of (possibly half-integral)
weight κ for the metaplectic group Mp2 (Z). In the same way we have to
consider vector valued nearly holomorphic modular forms and Maass wave
forms. Therefore in section 1.2 and 1.3 we collect some basic facts on such
modular forms and carry out the standard constructions of Poincaré and
Eisenstein series. Moreover, many objects which are indexed by an integer in
the above exposition (as H(m) or Φm ) have to be replaced by objects which
are indexed by a tuple (β, m), where β ∈ L0 /L and m ∈ Z + q(β) (as H(β, m)
or Φβ,m ). For the group Γ (L) we need to take the discriminant kernel of the
orthogonal group of L.
Theorem 0.8 is proved under the assumption that L splits two orthogonal
hyperbolic planes over Z. It seems likely that (at least for large l) this result
 Introduction 13

can also be extended to a more general situation. However, this might require
additional considerations as some newform theory or Hecke theory for the
space Sκ,L and the lifting ϑ.

Acknowledgement. I am indebted to E. Freitag for his valuable suggestions and

encouragement. Without his support this work would not have been possible. I
thank R. Borcherds for suggesting to consider the functions Fβ,m . Moreover, I
would like to thank A. Deitmar, J. Funke, W. Kohnen, and P. Li for their help.
1 Vector valued modular forms for
the metaplectic group

In the following chapter we briefly recall some facts about the metaplectic
cover of SL2 (R), the Weil representation, and certain vector valued modular
forms. We essentially follow the terminology of Borcherds [Bo2]. Moreover,
we consider vector valued Poincaré and Eisenstein series in some detail.

1.1 The Weil representation

As usual, we denote by H = {τ ∈ C; =(τ ) > 0} the upper complex half plane.

Throughout (unless otherwise specified) we will use τ as a standard variable
on H and write x for its real part and y for its √ imaginary part, respectively.
For z ∈ C we put e(z) = e2πiz , and√ denote by z = z 1/2
the principal branch
of the square root, so that arg( z) ∈ (−π/2, π/2]. For any integer k we put
z k/2 = (z 1/2 )k . More generally if b ∈ C, then we define z b = eb Log(z) , where
Log(z) denotes the principal branch of the logarithm. If x is a non-zero real
number, we let sgn(x) = x/|x|.
We write Mp2 (R) for the metaplectic group, i.e. the double cover of
SL2 (R),
realized by the two choices of holomorphic square roots of τ 7→ cτ +d
for ac db ∈ SL2 (R). Thus the elements of Mp2 (R) are pairs

(M, φ(τ )),

a b


where M = c d ∈ SL2 (R), and φ denotes a holomorphic function on H
with
φ(τ )2 = cτ + d.
The product of (M1 , φ1 (τ )), (M2 , φ2 (τ )) ∈ Mp2 (R) is given by




M1 , φ1 (τ ) M2 , φ2 (τ ) = M1 M2 , φ1 (M2 τ )φ2 (τ ) ,
aτ +b
where M τ = cτ +d denotes the usual action of SL2 (R) on H. The map


√ 
7→ ^
a b

a b = a b
c d c d c d , cτ + d (1.1)

defines a locally isomorphic embedding of SL2 (R) into Mp2 (R).

16 1 Vector valued modular forms for the metaplectic group

Let Mp2 (Z) be the inverse image of SL2 (Z) under the covering map
Mp2 (R) → SL2 (R). It is well known that Mp2 (Z) is generated by
  
11
T = ,1 ,
01
0 −1 √
  
S= , τ .
1 0

One has the relations S 2 = (ST )3 = Z, where

  
−1 0
Z= ,i
0 −1

is the standard generator of the center of Mp2 (Z). We will often use the
abbreviations Γ1 := SL2 (Z), Γ∞ = {( 10 n1 ) ; n ∈ Z} ≤ Γ1 , and
   
1n
Γ̃∞ := hT i = ,1 ; n ∈ Z .
01

Throughout let L be an even lattice, i.e. a free Z-module of finite rank,

equipped with a symmetric Z-valued bilinear form (·, ·) such that the associ-
ated quadratic form
1
q(x) = (x, x)
2
takes its values in Z. We assume that L is non-degenerated and denote its
signature by (b+ , b− ). We write

L0 = {x ∈ L ⊗ Q; (x, y) ∈ Z for all y ∈ L}

for the dual lattice. Then the quotient L0 /L is a finite Abelian group, the so-
called discriminant group. The modulo 1 reduction of q(x) is a Q/Z-valued
quadratic form on L0 /L whose associated bilinear form is the modulo 1 re-
duction of the bilinear form (·, ·) on L0 .
Recall that there is a unitary representation %L of Mp2 (Z) on the group
algebra C[L0 /L]. If we denote the standard basis of C[L0 /L] by (eγ )γ∈L0 /L ,
then %L can be defined by the action of the generators S, T ∈ Mp2 (Z) as
follows (cp. [Bo2]):

%L (T )eγ = e(q(γ))eγ (1.2)

√ b− −b+
i X
%L (S)eγ = p e(−(γ, δ))eδ . (1.3)
|L0 /L| δ∈L0 /L

This representation is essentially the Weil representation attached to the

quadratic module (L0 /L, q) (cf. [No]). It factors through the finite group
 1.1 The Weil representation 17

SL2 (Z/N Z) if b+ + b− is even, and through a double cover of SL2 (Z/N Z) if

b+ + b− is odd, where N is the smallest integer such that N q(γ) ∈ Z for all
γ ∈ L0 . Note that
− +
%L (Z)eγ = ib −b e−γ . (1.4)
We write h·, ·i for the standard scalar product on C[L0 /L], i.e.
* +
X X X
λγ eγ , µγ eγ = λγ µγ .
γ∈L0 /L γ∈L0 /L γ∈L0 /L

For γ, δ ∈ L0 /L and (M, φ) ∈ Mp2 (Z) we define the coefficient %γδ (M, φ) of
the representation %L by

%γδ (M, φ) = h%L (M, φ)eδ , eγ i.

Shintani proved the following explicit formula for %L (cf. [Sn], Prop. 1.6).

Proposition 1.1. Let α, β ∈ L0 /L and M = ac db ∈ SL2 (Z). Then the

coefficient %αβ (M̃ ) is given by

√ (b− −b+ )(1−sgn(d))
i δα,aβ e(abq(α)), (1.5)

if c = 0, and by
√ (b− −b+ ) sgn(c)
i X  a(α + r, α + r) − 2(β, α + r) + d(β, β) 
p e ,
|c|(b− +b+ )/2 |L0 /L| r∈L/cL 2c
(1.6)
if c 6= 0. Here, δ∗,∗ denotes the Kronecker-delta.

Let κ ∈ 12 Z and f be a C[L0 /L]-valued function on H. For (M, φ) ∈

Mp2 (Z) we define the Petersson slash operator by

(f |κ (M, φ)) (τ ) = φ(τ )−2κ %L (M, φ)−1 f (M τ ). (1.7)

As usual, by f 7→ f |κ (M, φ) an operation of Mp2 (Z) on functions f : H →

C[L0 /L] is defined.
We denote by %∗L the dual representation of %L . If we think of %∗L (M, φ)
((M, φ) ∈ Mp2 (Z)) as a matrix with entries in C, then %∗L (M, φ) is simply
the complex conjugate of %L (M, φ). We have a “dual operation” of Mp2 (Z)
on functions f : H → C[L0 /L], given by

(f |∗κ (M, φ)) (τ ) = φ(τ )−2κ %∗L (M, φ)−1 f (M τ ). (1.8)

Now assume that f : H → C[L0 /L] is a holomorphic function which is

invariant under the |∗κ -operation of T ∈ Mp2 (Z). We denote the components
18 1 Vector valued modular forms for the metaplectic group
P
of f by fγ , so that f = γ∈L0 /L eγ fγ . The invariance of f under T implies
that the functions e(q(γ)τ )fγ (τ ) are periodic with period 1. Thus f has a
Fourier expansion
X X
f (τ ) = eγ c(γ, n)e(nτ ),
γ∈L0 /L n∈Z−q(γ)

R1
with Fourier coefficients c(γ, n) = 0
fγ (τ )e(−nτ ) dx. Using the abbreviation
eγ (τ ) := eγ e(τ ) we may write
X X
f (τ ) = c(γ, n)eγ (nτ ). (1.9)
γ∈L0 /L n∈Z−q(γ)

By means of the scalar product in C[L0 /L] the coefficients can be expressed
by
Z1
c(γ, n) = hf (τ ), eγ (nτ̄ )i dx. (1.10)
0

Definition 1.2. Let κ ∈ 12 Z. A function f : H → C[L0 /L] is called a modular

form of weight κ with respect to %∗L and Mp2 (Z) if
i) f |∗κ (M, φ) = f for all (M, φ) ∈ Mp2 (Z),
ii) f is holomorphic on H,
iii) f is holomorphic at the cusp ∞.
Here condition (iii) means that f has a Fourier expansion of the form
X X
f (τ ) = c(γ, n)eγ (nτ ).
γ∈L0 /L n∈Z−q(γ)
n≥0

Moreover, if all c(γ, n) with n = 0 vanish, then f is called a cusp form. The
C-vector space of modular forms of weight κ with respect to %∗L and Mp2 (Z)
is denoted by Mκ,L , the subspace of cusp forms by Sκ,L .

It is easily seen that Mκ,L is finite dimensional.

Example 1.3. Fix a positive integer t. Let L be the 1-dimensional lattice Z,

equipped with the positive definite quadratic form q(a) = ta2 . Then L0 /L ∼
=
Z/2tZ, and Mκ−1/2,L is isomorphic to the space Jκ,t of Jacobi forms of weight
κ and index t (cf. [EZ] Theorem 5.1). If L denotes the lattice Z with the
negative definite quadratic form q(a) = −ta2 , then Mκ−1/2,L is isomorphic
to the space Jκ,t of skew-holomorphic Jacobi forms of weight κ and index t
as defined in [Sk].
 1.2 Poincaré series and Eisenstein series 19

1.2 Poincaré series and Eisenstein series

In this section we carry out the standard construction of Poincaré and Eisen-
stein series for the space Mκ,L . For simplicity we assume that (b+ , b− ) = (2, l),
(1, l − 1), or (0, l − 2) with l ≥ 3. Moreover, we assume that κ = 1 + l/2 (as
we will only need this case later).

1.2.1 Poincaré series

Let β ∈ L0 /L and m ∈ Z − q(β) with m > 0. Then eβ (mτ ) is a holomorphic

function H → C[L0 /L] which is invariant under the |∗κ -operation of T ∈
Mp2 (Z). Moreover, by (1.4) we have

eβ (mτ ) |∗κ Z = i−2κ il−2 e−β (mτ ) = e−β (mτ ).

Thus eβ (mτ ) is also invariant under the action of Z 2 .

L
We define the Poincaré series Pβ,m of index (β, m) by
1 X
L
Pβ,m (τ ) = eβ (mτ ) |∗κ (M, φ) (1.11)
2
(M,φ)∈Γ̃∞ \ Mp2 (Z)

(recall that Γ̃∞ = hT i). Since κ ≥ 5/2, the usual argument shows that
this series converges normally on H and thereby defines a Mp2 (Z)-invariant
holomorphic function H → C[L0 /L]. We will often omit the superscript L, if
it is clear from the context.
Theorem 1.4. The Poincaré series Pβ,m has the Fourier expansion
X X
Pβ,m (τ ) = pβ,m (γ, n)eγ (nτ )
γ∈L0 /L n∈Z−q(γ)
n>0

with

pβ,m (γ, n) = δm,n (δβ,γ + δ−β,γ )

 n  κ−1
2
X 
4π √

+ 2π Hc∗ (β, m, γ, n)Jκ−1 mn . (1.12)
m |c|
c∈Z−{0}

Here Hc∗ (β, m, γ, n) denotes the generalized Kloosterman sum

e−πi sgn(c)κ/2
^  
X ab ma + nd
Hc∗ (β, m, γ, n) = %βγ e ,
|c| ∗
cd c
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d
(1.13)
and Jν the Bessel function of the first kind ([AbSt] Chap. 9). In particular
Pβ,m ∈ Sκ,L .
20 1 Vector valued modular forms for the metaplectic group

Recall the abbreviations Γ1 = SL2 (Z) and Γ∞ = {( 10 n1 ) ; n ∈ Z}.
The

sum in (1.13) runs over all primitive residues d modulo c and ac db is a
representative for the double coset in Γ∞ \Γ1 /Γ∞ with lower row (c d0 ) and
d0 ≡ d (mod c). Observe that the expression % ^ a b e ma+nd does not



βγ c d c
depend on the choice of the coset representative.
The coefficients %βγ ^a b


c d are universally bounded, because %L factors
through a finite group. Hence there exists a constant C > 0 such that
Hc∗ (β, m, γ, n) < C for all γ ∈ L0 /L, n ∈ Z − q(γ), and c ∈ Z − {0}.
This, together with the asymptotic behavior Jν (t) = O(tν ) (t → 0) of the
J-Bessel function, implies that the series (1.12) converges absolutely. If we
also use the asymptotic property Jν (t) = O(t−1/2 ) as t → ∞, we find that
pβ,m (γ, n) = O(nκ−1 ) for n → ∞. (This is actually a very poor bound. For
even l Deligne’s theorem, previously the Ramanujan-Petersson conjecture,
states that pβ,m (γ, n) = O(n(κ−1)/2+ε ).)

Proof of Theorem 1.4. Let γ ∈ L0 /L and n ∈ Z − q(γ). According to (1.10)

the coefficient pβ,m (γ, n) can be computed as

Z1 * +
1 X
pβ,m (γ, n) = eβ (mτ ) |∗κ (M, φ), eγ (nτ̄ ) dx.
2
0 (M,φ)∈Γ̃∞ \ Mp2 (Z)

Since eβ (mτ ) is invariant under the action of Z 2 , this can be written in the
form
Z1 Z1
heβ (mτ ), eγ (nτ̄ )i dx + heβ (mτ ) |∗κ Z, eγ (nτ̄ )i dx
0 0
Z1 * X
+
|∗κ ^

a b , e (nτ̄ )
+ eβ (mτ ) c d γ dx
 
0 a b ∈Γ∞ \Γ1
c d
c6=0

= δm,n (δβ,γ + δ−β,γ )

Z∞

−1
X  
−κ
%∗L ^
a b a b


+ (cτ + d) c d eβ (m c d τ ), eγ (nτ̄ ) dx.
  c6=0 −∞
a b ∈Γ \Γ /Γ
∞ 1 ∞
c d

Since %L is unitary we have

−1
  ^    
ab ab
%∗L ^
a b a b


c d eβ (m c d τ ), eγ (nτ̄ ) = %βγ e m τ e(−nτ ),
cd cd

and thereby
 1.2 Poincaré series and Eisenstein series 21

pβ,m (γ, n) = δm,n (δβ,γ + δ−β,γ )

^ Z∞    
X ab −κ ab
+ %βγ (cτ + d) e m τ − nτ dx.
cd cd
  c6=0 −∞
a b ∈Γ \Γ /Γ
∞ 1 ∞
c d
(1.14)
√ √ p
We now use the elementary identities cτ + d = sgn(c) c τ + d/c and
 
ab a 1
τ= − 2 . (1.15)
cd c c (τ + d/c)
We find that the latter sum in (1.14) equals
^
X ab
|c|−κ sgn(c)κ %βγ
cd
  c6=0
a b ∈Γ∞ \Γ1 /Γ∞
c d
Z∞  
−κ ma m
× (τ + d/c) e − 2 − nτ dx
c c (τ + d/c)
−∞
^ 
X
−κ κ ab ma 
= |c| sgn(c) %βγ e
cd c
  c6=0
a b ∈Γ∞ \Γ1 /Γ∞
c d
Z∞  m 
× τ −κ e − 2 − n(τ − d/c) dx
c τ
−∞
^  
X X ab ma + nd
= |c|−κ sgn(c)κ %βγ e
∗
cd c
c6=0
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d
Z∞  m 
× τ −κ e − 2 − nτ dx.
c τ
−∞

We substitute τ = iw in the integral and obtain for the above expression

^  
−κ
X
−κ κ
X ab ma + nd
2πi |c| sgn(c) %βγ e
∗
cd c
c6=0
  d(c)
a b ∈Γ \Γ /Γ
∞ 1 ∞
c d
C+i∞
Z  
1 2πm
× w−κ exp − 2 + 2πnw dw,
2πi c w
C−i∞
22 1 Vector valued modular forms for the metaplectic group

where C is a positive real constant. We now use i−κ sgn(c)κ = e−πi sgn(c)κ/2
and the definition of the generalized Kloosterman sum (1.13) and find
X
pβ,m (γ, n) = δm,n (δβ,γ + δ−β,γ ) + 2π |c|1−κ Hc∗ (β, m, γ, n)
c6=0
C+i∞
Z  
1 −κ 2πm
× w exp − 2 + 2πnw dw. (1.16)
2πi c w
C−i∞

If n ≤ 0, we consider the limit C → ∞ and deduce that pβ,m (γ, n) = 0. If

n > 0, then the latter integral is an inverse Laplace transform that can be
found in any standard reference on integral transforms. According to [E2]
p. 245 (40) one has
C+i∞  2  κ−1
4π √
Z    
1 −κ 2πm c n 2
w exp − 2 + 2πnw dw = Jκ−1 mn .
2πi c w m |c|
C−i∞

If we insert this into (1.16) we obtain the assertion. t

u

1.2.2 The Petersson scalar product

Let f, g be modular forms in Mκ,L . Then it can be easily checked that

hf (τ ), g(τ )iy κ

is Mp2 (Z)-invariant. (As before, h·, ·i denotes the scalar product on C[L0 /L].)
If f or g is a cusp form, then hf (τ ), g(τ )iy κ is bounded on the standard
fundamental domain

F = {τ = x + iy ∈ H; |x| ≤ 1/2, |τ | ≥ 1}

for the action of Γ1 on H. (Hence the invariance property implies that

hf (τ ), g(τ )iy κ is bounded on the whole upper half plane H.) We define the
Petersson scalar product of f and g by
Z
dx dy
(f, g) = hf (τ ), g(τ )iy κ 2 . (1.17)
y
F

It is easily seen that the integral does not depend on the choice of the fun-
damental domain. The space Sκ,L endowed with the scalar product (·, ·) is a
finite dimensional Hilbert space.
The Petersson coefficient formula for Sκ,L has the following form.
 1.2 Poincaré series and Eisenstein series 23

Proposition 1.5. Let β ∈ L0 /L and m ∈ Z − q(β) with m > 0. Let f be a

cusp form in Sκ,L and denote its Fourier coefficients by c(γ, n) (as in (1.9)).
Then the scalar product of f with the Poincaré series Pβ,m is given by
Γ (κ − 1)
(f, Pβ,m ) = 2 c(β, m).
(4πm)κ−1

Proof. Let Γ 1 = Γ1 /{±1}. We use the usual unfolding argument and find
Z * +
1 X dx dy
(f, Pβ,m ) = f (τ ), eβ (mτ ) |∗κ (M, φ) y κ 2
2 y
Γ 1 \H (M,φ)∈Γ̃∞ \ Mp2 (Z)
Z X dx dy
= hf (M τ ), eβ (mM τ )i=(M τ )κ
y2
M ∈Γ∞ \Γ1
Γ 1 \H
Z
=2 hf (τ ), eβ (mτ )iy κ−2 dx dy.
Γ∞ \H

We insert the Fourier expansion of f and infer:

Z∞ Z1 X
(f, Pβ,m ) = 2 c(β, n)e(nτ )e(−mx + miy)y κ−2 dx dy
0 0 n∈Z−q(β)
n>0
Z∞
=2 c(β, m)e−4πmy y κ−2 dy
y=0
Γ (κ − 1)
=2 c(β, m).
(4πm)κ−1
t
u
As a corollary we obtain that the Poincaré series Pβ,m generate the space
Sκ,L .

1.2.3 Eisenstein series

Let β ∈ L0 /L with q(β) ∈ Z. Then the vector eβ = eβ (0) ∈ C[L0 /L], consid-
ered as a constant function H → C[L0 /L], is invariant under the |∗κ -action of
T, Z 2 ∈ Mp2 (Z). The Eisenstein series
1 X
EβL (τ ) = eβ |∗κ (M, φ) (1.18)
2
(M,φ)∈Γ̃∞ \ Mp2 (Z)

converges normally on H and therefore defines a Mp2 (Z)-invariant holomor-

phic function on H.
24 1 Vector valued modular forms for the metaplectic group

Theorem 1.6. The Eisenstein series Eβ has the Fourier expansion

X X
Eβ (τ ) = qβ (γ, n)eγ (nτ )
γ∈L0 /L n∈Z−q(γ)
n≥0

with

δ + δ−β,γ , if n = 0,
 β,γ


qβ (γ, n) = (2π)κ nκ−1 X (1.19)
 |c|1−κ Hc∗ (β, 0, γ, n), if n > 0.
 Γ (κ)

c∈Z−{0}

Here, Hc∗ (β, 0, γ, n) denotes the generalized Kloosterman sum defined in

(1.13). In particular Eβ is an element of Mκ,L .

Proof. We proceed in the same way as in the proof of Theorem 1.4. Let
γ ∈ L0 /L and n ∈ Z − q(γ). By (1.10) we have

Z1 * +
1 X
qβ (γ, n) = eβ |∗κ (M, φ), eγ (nτ̄ ) dx.
2
0 (M,φ)∈Γ̃∞ \ Mp2 (Z)

Because eβ is invariant under the action of Z 2 , this equals

Z∞

−1
X  
−κ ∗^a b
δ0,n (δβ,γ + δ−β,γ ) + (cτ + d) %L c d eβ , eγ (nτ̄ ) dx
  c6=0 −∞
a b ∈Γ∞ \Γ1 /Γ∞
c d
^ Z∞
X ab
= δ0,n (δβ,γ + δ−β,γ ) + %βγ (cτ + d)−κ e(−nτ ) dx.
cd
  c6=0 −∞
a b ∈Γ∞ \Γ1 /Γ∞
c d
√ √ p
We now evaluate the integral. Since cτ + d = sgn(c) c τ + d/c, we have

Z∞ Z∞
−κ −κ
(cτ + d) e(−nτ ) dx = |c| sgn(c) κ
(τ + d/c)−κ e(−nτ ) dx
−∞ −∞
  Z∞
−κ nd
= |c| sgn(c) eκ
τ −κ e(−nτ ) dx.
c
−∞

We substitute τ = iw and find

 1.2 Poincaré series and Eisenstein series 25

Z∞
(cτ + d)−κ e(−nτ ) dx
−∞
  C+i∞
Z
−κ −κ nd 1
= 2πi |c| κ
sgn(c) e w−κ e2πnw dw,
c 2πi
C−i∞

where C is a positive real constant. If n ≤ 0, one can consider the limit C →

∞ to infer that the integral is 0. It follows that qβ (γ, n) = δ0,n (δβ,γ + δ−β,γ )
in this case. Now let n > 0. Then the integral is an inverse Laplace transform.
According to [E2] p. 238 (1) we have
C+i∞
(2πn)κ−1
Z
1
w−κ e2πnw dw = .
2πi Γ (κ)
C−i∞

(In fact, one can deform the path of integration and essentially obtains the
Hankel integral for 1/Γ (κ).) Hence we get

(2π)κ nκ−1
^  
X
−κ −κ κ ab nd
qβ (γ, n) = |c| i sgn(c) %βγ e .
Γ (κ) cd c
  c6=0
a b ∈Γ∞ \Γ1 /Γ∞
c d

If we use i−κ sgn(c)κ = e−πi sgn(c)κ/2 and substitute in the definition of

Hc∗ (β, 0, γ, n), we obtain the assertion. u
t

For computational purposes the formula (1.19) for the coefficients qβ (γ, n)
is not very useful. An explicit formula, involving special values of Dirichlet
L-series and finitely many representation numbers modulo prime powers at-
tached to the lattice L, is derived in [BK].
To give an example we briefly consider the case that L is unimodular. Then
L0 /L is trivial and Mκ,L is just the space of elliptic modular forms of weight
1 + l/2. We have one Eisenstein series E with Fourier coefficients q(n) and
q(0) = 2. For n > 0 the above expression for the coefficients can be simplified
as follows. First note that l ≡ 2 (mod 8). This implies e−πi sgn(c)κ/2 = −1
and  
1 X nd 1 X
Hc∗ (β, 0, γ, n) = − e =− µ(|c|/d)d.
|c| ∗
c |c|
d(c) d|(c,n)
d>0

e( nd
P
Here we have used the evaluation of the Ramanujan sum d(c)∗ c ) by
means of the Moebius function µ. Thus we find
26 1 Vector valued modular forms for the metaplectic group
∞
(2π)κ nκ−1 X −κ X
q(n) = −2 c µ(c/d)d
Γ (κ) c=1 d|(c,n)
d>0
(2π)κ nκ−1 X 1−κ
= −2 d
Γ (κ)ζ(κ)
d|n

(2π)κ
= −2 σκ−1 (n),
Γ (κ)ζ(κ)

where σκ−1 (n) denotes the sum of the (κ−1)-th powers of the positive divisors
κ
of n and ζ(s) the Riemann zeta function. Using ζ(κ) = − (2πi)
2κ! Bκ with the
κ-th Bernoulli number Bκ , and the fact that κ ≡ 2 (mod 4), we get
4κ
q(n) = − σκ−1 (n).
Bκ
Thus in this case the Eisenstein series E(z) is the classical Eisenstein series
for Γ1 , normalized such that its constant term equals 2.
Let us now turn back to the general case and determine the number of
linearly independent Eisenstein series Eβ . The |∗κ -invariance of Eβ under Z
and the identity eβ |∗κ Z = e−β imply

1 X
Eβ = Eβ |∗κ Z = eβ |∗κ Z |∗κ (M, φ) = E−β .
2
(M,φ)∈Γ̃∞ \ Mp2 (Z)

Eis
Thus the space Mκ,L spanned by all Eisenstein series Eβ equals the space
spanned by the Eα where α runs through a set of representatives of {β ∈
L0 /L; q(β) ∈ Z} modulo the action of {±1}. Comparing constant terms one
finds that the set of Eisenstein series Eα with α as above is already linearly
independent. We obtain

Eis
dim(Mκ,L ) = #{β ∈ L0 /L; q(β) ∈ Z, 2β = 0 + L}
1
+ #{β ∈ L0 /L; q(β) ∈ Z, 2β 6= 0 + L}. (1.20)
2
Now let f ∈ Mκ,L and write the Fourier expansion of f in the form
X X
f (τ ) = c(γ, n)eγ (nτ ).
γ∈L0 /L n∈Z−q(γ)

The invariance of f under Z and (1.4) imply that c(γ, n) = c(−γ, n). Hence
1 X
f− c(β, 0)Eβ
2 0
β∈L /L
q(β)∈Z
 1.3 Non-holomorphic Poincaré series of negative weight 27

Eis
lies in Sκ,L . This shows that Mκ,L = Mκ,L ⊕ Sκ,L and (1.20) is a formula for
the codimension of Sκ,L in Mκ,L .

In later applications we will mainly be interested in the Eisenstein series

E0 (τ ) which we simply denote by E(τ ). In the same way we write q(γ, n) for
the Fourier coefficients q0 (γ, n) of E(τ ).
Let us finally cite the following result of [BK]:
Proposition 1.7. The coefficients q(γ, n) of E(τ ) are rational numbers.

1.3 Non-holomorphic Poincaré series of negative weight

As in the previous section we assume that L is an even lattice of signature
(b+ , b− ) = (2, l), (1, l − 1), or (0, l − 2) with l ≥ 3. Put k = 1 − l/2 and
κ = 1 + l/2. We now construct certain vector valued Maass-Poincaré series
for Mp2 (Z) of weight k. Series of a similar type are well known and appear
in many places in the literature (see for instance [He, Ni, Fa]).

Let Mν, µ (z) and Wν, µ (z) be the usual Whittaker functions as defined
in [AbSt] Chap. 13 p. 190 or [E1] Vol. I Chap. 6 p. 264. They are linearly
independent solutions of the Whittaker differential equation

d2 w µ2 − 1/4
 
1 ν
+ − + − w = 0. (1.21)
dz 2 4 z z2

The functions Mν, µ (z) and Mν, −µ (z) are related by the identity

Γ (−2µ) Γ (2µ)
Wν, µ (z) = Mν, µ (z) + Mν, −µ (z) (1.22)
Γ ( 12 − µ − ν) Γ ( 12 + µ − ν)

([AbSt] p. 190 (13.1.34)). This implies in particular Wν, µ (z) = Wν, −µ (z). As
z → 0 one has the asymptotic behavior

Mν, µ (z) ∼ z µ+1/2 / − 12 N),

(µ ∈ (1.23)
Γ (2µ)
Wν, µ (z) ∼ z −µ+1/2 (µ ≥ 1/2). (1.24)
Γ (µ − ν + 1/2)
If y ∈ R and y → ∞ one has
Γ (1 + 2µ)
Mν, µ (y) = ey/2 y −ν (1 + O(y −1 )), (1.25)
Γ (µ − ν + 1/2)
Wν, µ (y) = e−y/2 y ν (1 + O(y −1 )). (1.26)

For convenience we put for s ∈ C and y ∈ R>0 :

Ms (y) = y −k/2 M−k/2, s−1/2 (y). (1.27)

28 1 Vector valued modular forms for the metaplectic group

In the same way we define for s ∈ C and y ∈ R − {0}:

Ws (y) = |y|−k/2 Wk/2 sgn(y), s−1/2 (|y|). (1.28)

If y < 0 then equation (1.22) implies that

Γ (1 + k/2 − s) Γ (1 + k/2 − s)Γ (2s − 1)
Ms (|y|) = Ws (y) − M1−s (|y|).
Γ (1 − 2s) Γ (1 − 2s)Γ (s + k/2)
(1.29)
The functions Ms (y) and Ws (y) are holomorphic in s. Later we will be
interested in certain special s-values. For y > 0 we have

Mk/2 (y) = y −k/2 M−k/2, k/2−1/2 (y) = ey/2 , (1.30)

−k/2 −y/2
W1−k/2 (y) = y Wk/2, 1/2−k/2 (y) = e . (1.31)

Using the standard integral representation

Z∞
−z/2 µ+1/2
Γ (1/2 − ν + µ)Wν, µ (z) = e z e−tz t−1/2−ν+µ (1 + t)−1/2+ν+µ dt
0

(<(µ − ν) > −1/2, <(z) > 0) of the W -Whittaker function ([E1] Vol. I p. 274
(18)), we find for y < 0:

W1−k/2 (y) = |y|−k/2 W−k/2, 1/2−k/2 (|y|)

Z∞
−|y|/2
=e |y|1−k
e−t|y| (1 + t)−k dt
0
Z∞
= e|y|/2 |y|1−k e−t|y| t−k dt.
1

If we insert the definition of the incomplete Gamma function (cf. [AbSt] p. 81)
Z∞
Γ (a, x) = e−t ta−1 dt, (1.32)
x
we obtain for y < 0 the identity

W1−k/2 (y) = e−y/2 Γ (1 − k, |y|). (1.33)

The usual Laplace operator of weight k (cf. [Ma1])

 2
∂2
  
2 ∂ ∂ ∂
∆k = −y + 2 + iky +i (1.34)
∂x2 ∂y ∂x ∂y
acts on smooth functions f : H → C[L0 /L] component-wise. Let β ∈ L0 /L
and m ∈ Z+q(β) with m < 0. Then it can be easily checked that the function
 1.3 Non-holomorphic Poincaré series of negative weight 29

Ms (4π|m|y)eβ (mx)

is invariant under the |k -operation of T ∈ Mp2 (Z) (as defined in (1.7)) and
an eigenfunction of ∆k with eigenvalue

s(1 − s) + (k 2 − 2k)/4.
L
Definition 1.8. With the above notation we define the Poincaré series Fβ,m
of index (β, m) by

L 1 X
Fβ,m (τ, s) = [Ms (4π|m|y)eβ (mx)] |k (M, φ),
2Γ (2s)
(M,φ)∈Γ̃∞ \ Mp2 (Z)
(1.35)
where τ = x + iy ∈ H and s = σ + it ∈ C with σ > 1. If it is clear from the
context, to which lattice L the series (1.35) refers, we simply write Fβ,m (τ, s).

By (1.23) and (1.27) the above series (1.35) has the local majorant
X
=(M τ )σ .
(M,φ)∈Γ̃∞ \ Mp2 (Z)

Hence (1.35) converges normally for τ ∈ H, s ∈ C and σ > 1 and thereby

defines a Mp2 (Z)-invariant function on H. As in [Ma1] it can be seen that
∆k commutes with the action of Mp2 (Z), i.e. that

∆k (f |k (M, φ)) = (∆k f ) |k (M, φ)

for any smooth function f : H → C[L0 /L] and any (M, φ) ∈ Mp2 (Z). We may
infer that Fβ,m is an eigenfunction of ∆k :

∆k Fβ,m = s(1 − s) + (k 2 − 2k)/4 Fβ,m .

(Thus Fβ,m is automatically real analytic as a solution of an elliptic differ-

ential equation.) The invariance of Fβ,m under the action of Z ∈ Mp2 (Z)
implies that Fβ,m = F−β,m .
Let f : H → C[L0 /L] be a continuous function with the property f |k T =
f . By virtue of the same argument as in section 1.1 one finds that f has a
Fourier expansion of the form
X X
f (τ ) = c(γ, n, y)eγ (nx), (1.36)
γ∈L0 /L n∈Z+q(γ)

where the coefficients are given by

Z1
c(γ, n, y) = hf (τ ), eγ (nx)i dx. (1.37)
0
30 1 Vector valued modular forms for the metaplectic group

Theorem 1.9. The Poincaré series Fβ,m has the Fourier expansion

Γ (1 + k/2 − s)
Fβ,m (τ, s) = M1−s (4π|m|y)(eβ (mx) + e−β (mx))
Γ (2 − 2s)Γ (s + k/2)
X
+ b(γ, 0, s)y 1−s−k/2 eγ
γ∈L0 /L
q(γ)∈Z
X X
+ b(γ, n, s)Ws (4πny)eγ (nx) + Feβ,m (τ, s),
γ∈L0 /L n∈Z+q(γ)
n>0

where Feβ,m (τ, s) is the T -invariant function

X X
Feβ,m (τ, s) = b(γ, n, s)Ws (4πny)eγ (nx),
γ∈L0 /L n∈Z+q(γ)
n<0

and the Fourier coefficients b(γ, n, s) are given by

 k−1  
2π|n/m| 2 X 4π p
H (β, m, γ, n)I |mn| , if n > 0,

c 2s−1

Γ (s + k/2) |c|





 c∈Z−{0}
1−k/2 1+s−k/2
|m|s−k/2

 4 π X
|c|1−2s Hc (β, m, γ, 0),

if n = 0,


 (2s − 1)Γ (s + k/2)Γ (s − k/2)


c∈Z−{0}
 Γ (1 + k/2 − s)

 δm,n (δβ,γ + δ−β,γ )



 Γ (2s)Γ (1 − 2s)


 k−1  

 2π|n/m| 2 X 4π p
 + Γ (s − k/2) Hc (β, m, γ, n)J2s−1 |mn| , if n < 0.


|c|

c∈Z−{0}

Here Hc (β, m, γ, n) denotes the generalized Kloosterman sum

e−πi sgn(c)k/2
^  
X ab ma + nd
Hc (β, m, γ, n) = %−1
γβ e
|c| ∗
cd c
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d
(1.38)
and Jν (z), Iν (z) the usual Bessel functions as defined in [AbSt] Chap. 9.


The sum in (1.38) runs over all primitive residues d modulo c and ac db is a
representative for the double coset in Γ∞ \Γ1 /Γ∞ with lower row (c d0 ) and
d0 ≡ d (mod c). Observe that the expression %−1 ^ a b e ma+nd does not



γβ c d c
depend on the choice of the coset representative. The fact that Hc (β, m, γ, n)
is universally bounded implies that the series for the coefficients b(γ, n, s)
converge normally in s for σ > 1.
 1.3 Non-holomorphic Poincaré series of negative weight 31

Proof of Theorem 1.9. We split the sum in equation (1.35) into the sum over
1, Z, Z 2 , Z 3
∈ Γ̃∞ \ Mp2 (Z) and the sum over (M, φ) ∈ Γ̃∞ \ Mp2 (Z), where
M = ac db with c 6= 0. The latter sum will be denoted by H(τ, s). Since
eβ (mx) |k Z = e−β (mx), we find that the first term equals

1
Ms (4π|m|y)(eβ (mx) + e−β (mx)).
Γ (2s)

According to (1.29) this can be rewritten as

Γ (1 + k/2 − s)
M1−s (4π|m|y)(eβ (mx) + e−β (mx))
Γ (2 − 2s)Γ (s + k/2)
Γ (1 + k/2 − s)
+ Ws (4πmy)(eβ (mx) + e−β (mx)). (1.39)
Γ (2s)Γ (1 − 2s)

We now calculate the Fourier expansion of H(τ, s). Since eβ (mx) is in-
variant under the action of Z 2 , we may write H(τ, s) in the form
^
1 X ab
[Ms (4π|m|y)eβ (mx)] |k .
Γ (2s)   cd
a b ∈Γ∞ \Γ1
c d
c6=0

Let γ ∈ L0 /L and n ∈ Z + q(γ) and denote the (γ, n)-th Fourier coefficient of
H(τ, s) by c(γ, n, y) (cp. (1.36)). Then we have

Z1
c(γ, n, y) = hH(τ, s), eγ (nx)i dx
0
Z∞
X (cτ + d)−k 
y

= Ms 4π|m| |cτ +d| 2
Γ (2s)
  c6=0 −∞
a b ∈Γ \Γ /Γ
∞ 1 ∞
c d
 
× %−1 ^

a b e m< a b




L c d β c d τ , eγ (nx) dx.

By (1.15) we find that

 
%−1 ^

a b e m< a b




L c d β c d τ , eγ (nx)
^   
ab ma m
= %−1
γβ e −< 2 e(−nx),
cd c c (τ + d/c)

and therefore
32 1 Vector valued modular forms for the metaplectic group
^ 
1 X ab ma 
c(γ, n, y) = %−1
γβ e
Γ (2s) cd c
  c6=0
a b ∈Γ \Γ /Γ
∞ 1 ∞
c d
Z∞      
y m
× (cτ + d)−k Ms 4π|m| e −< − nx dx.
|cτ + d|2 2
c (τ + d/c)
−∞

√ √ p
If we use cτ + d = sgn(c) c τ + d/c and substitute x by x − d/c in the
integral, we obtain
^  
1 X ab ma + nd
c(γ, n, y) = %−1
γβ e |c|−k sgn(c)k
Γ (2s) cd c
  c6=0
a b ∈Γ∞ \Γ1 /Γ∞
c d
Z∞    
−k y −mx
× τ Ms 4π|m| 2 2 e 2 2 − nx dx
c |τ | c |τ |
−∞
1 X 1−k k
= |c| i Hc (β, m, γ, n)
Γ (2s)
c6=0
Z∞    
−k y −mx
× τ Ms 4π|m| 2 2 e 2 2 − nx dx.
c |τ | c |τ |
−∞

By the definition of Ms this equals

(4π|m|y)−k/2 X
|c|Hc (β, m, γ, n)
Γ (2s)
c6=0
Z∞  −k/2    
τ y −mx
× M−k/2, s−1/2 4π|m| 2 2 e 2 2 − nx dx
−τ̄ c |τ | c |τ |
−∞
Z∞  −k/2
(4π|m|y)−k/2 X y − ix
= |c|Hc (β, m, γ, n)
Γ (2s) y + ix
c6=0 −∞
   
4π|m|y −mx
× M−k/2, s−1/2 e 2 2 − nx dx. (1.40)
c2 (x2 + y 2 ) c (x + y 2 )

We abbreviate the latter integral by I and substitute x = yu. Then I is equal

to
Z∞  −k/2    
1 − iu 4π|m| |m|u
y M−k/2, s−1/2 e − nyu du;
1 + iu c y(u2 + 1)
2 c y(u2 + 1)
2
−∞
 1.3 Non-holomorphic Poincaré series of negative weight 33
|m|
and if we set A = −ny and B = c2 y , we get

Z∞  −k/2    
1 − iu 4πB 2πiBu
I=y M−k/2, s−1/2 exp + 2πiAu du.
1 + iu u2 + 1 u2 + 1
−∞

The integral I/(yΓ (2s)) was evaluated by Hejhal in [He] p. 357. (Notice that
M−k/2, s−1/2 (y)/Γ (2s) is the same as Hejhal’s function G(y) with ` = −k/2.
Moreover, Wν, µ (y) equals e−y/2 y µ+1/2 Ψ (µ + 1/2 − ν, 1 + 2µ, y) in Hejhal’s
notation.) We find
 p
2π |B/A|  p 
W−k/2, s−1/2 (4π|A|)J2s−1 4π |AB| , if A > 0,






 Γ (s − k/2)

4π 1+s

I 
= Bs, if A = 0,
yΓ (2s)   (2s − 1)Γ (s + k/2)Γ (s − k/2)


 p
2π |B/A|

  p 
Wk/2, s−1/2 (4π|A|)I2s−1 4π |AB| , if A < 0.



Γ (s + k/2)

We resubstitute for A and B and obtain

 p  
2π |m/n| k/2 4π p
|mn| Ws (4πny), if n < 0,


 (4π|n|y) J 2s−1



 Γ (s − k/2) |c|

4π 1+s |m|s |c|1−2s y 1−s

|c|I 
= , if n = 0,
Γ (2s)   (2s − 1)Γ (s + k/2)Γ (s − k/2)


 p
  
 2π |m/n| k/2 4π p
(4π|n|y) I2s−1 |mn| Ws (4πny), if n > 0.



Γ (s + k/2) |c|

If we insert this into (1.40) we get c(γ, n, y) is given by

 k−1  
 2π|n/m| 2 X 4π p
H (β, m, γ, n)J |mn| Ws (4πny), n < 0,

c 2s−1

Γ (s − k/2) |c|




 c6=0

 41−k/2 π 1+s−k/2 |m|s−k/2
 X
|c|1−2s Hc (β, m, γ, 0)y 1−s−k/2 , n = 0,

(2s − 1)Γ (s + k/2)Γ (s − k/2)

 c6=0

 k−1  
2π|n/m| 2 X 4π p


Hc (β, m, γ, n)I2s−1 |mn| Ws (4πny), n > 0.



 Γ (s + k/2)

 |c|
c6=0

Combining this with (1.39) we obtain the assertion. t

u

Since ∆k Fβ,m (τ, 1 − k/2) = 0, the Poincaré series Fβ,m (τ, s) are in par-
ticular interesting in the special case s = 1 − k/2.
34 1 Vector valued modular forms for the metaplectic group

Proposition 1.10. For s = 1−k/2 the Fourier expansion of Fβ,m (τ, s) given
in Theorem 1.9 can be simplified as follows:
X X
Fβ,m (τ, 1 − k/2) = eβ (mτ ) + e−β (mτ ) + b(γ, n, 1 − k/2)eγ (nτ )
γ∈L0 /L n∈Z+q(γ)
n≥0

+ Feβ,m (τ, 1 − k/2).

The function Feβ,m (τ, 1 − k/2) has the expansion

X X
Feβ,m (τ, 1 − k/2) = b(γ, n, 1 − k/2)W1−k/2 (4πny)eγ (nx),
γ∈L0 /L n∈Z+q(γ)
n<0

and the Fourier coefficients b(γ, n, 1 − k/2) are given by

 k−1  
n 2 X 4π p
2π H (β, m, γ, n)I |mn| , if n > 0,


 c 1−k

 m |c|
c∈Z−{0}



(2π)2−k |m|1−k X



|c|k−1 Hc (β, m, γ, 0), if n = 0,


 Γ (2 − k)


c∈Z−{0}
−1
δm,n (δβ,γ + δ−β,γ )






 Γ (1 − k)

k−1
  
2π n 2 X 4π p


+ H (β, m, γ, n)J |mn| , if n < 0.


 Γ (1 − k) m c 1−k
|c|


c∈Z−{0}

Proof. By virtue of (1.30) and (1.31) this immediately follows from Theorem
1.9. t
u
Definition 1.11. A function f : H → C[L0 /L] is called a nearly holomorphic
modular form of weight k (with respect to %L and Mp2 (Z)), if
i) f |k (M, φ) = f for all (M, φ) ∈ Mp2 (Z),
ii) f is holomorphic on H,
iii) f has a pole in ∞, i.e. f has a Fourier expansion of the form
X X
f (τ ) = c(γ, n)eγ (nτ ).
γ∈L0 /L n∈Z+q(γ)
n−∞

!
The space of these nearly holomorphic modular forms is denoted by Mk,L .
The Fourier polynomial
X X
c(γ, n)eγ (nτ )
γ∈L0 /L n∈Z+q(γ)
n<0

is called the principal part of f .

 1.3 Non-holomorphic Poincaré series of negative weight 35

The following proposition is an analogue of the theorem in [He] on p. 660

(see also [Ni] Theorem 6).

Proposition 1.12. Let f (τ ) be a nearly holomorphic modular form of weight

k and denote its Fourier coefficients by c(γ, n) (γ ∈ L0 /L, n ∈ Z+q(γ)). Then
1 X X
f (τ ) = c(γ, n)Fγ,n (τ, 1 − k/2).
2 0
γ∈L /L n∈Z+q(γ)
n<0

Proof. Consider the function

1 X X
g(τ ) = f (τ ) − c(γ, n)Fγ,n (τ, 1 − k/2).
2 0
γ∈L /L n∈Z+q(γ)
n<0

The invariance of f under Z ∈ Mp2 (Z) implies that c(γ, n) = c(−γ, n).
According to Proposition 1.10 we have Fγ,n (τ, 1 − k/2) = eγ (nτ ) + e−γ (nτ ) +
O(1) for all γ, n. Hence g(τ ) is bounded as y → ∞. But g also satisfies

g |k (M, φ) = g ((M, φ) ∈ Mp2 (Z)),

∆k g = 0.

Thus g has to vanish identically (see [He] Chap. 9 Prop. 5.14c). t

u

Lemma 1.13. Let c ∈ Z − {0}, β, γ ∈ L0 /L, m ∈ Z + q(β) and n ∈ Z + q(β).

Then
Hc (β, m, γ, n) = H−c (β, m, γ, n).

Proof. From Proposition 1.1 it can be easily deduced that

%βγ ^ a −b
^ %−1 ^ −1 ^
a −b


a b =%

a b



c d βγ −c d and γβ c d = %γβ −c d .

Now the assertion immediately follows from the definition of Hc (β, m, γ, n).
t
u

Remark 1.14. This implies that the Fourier coefficients b(γ, n, 1 − k/2) of the
function Fβ,m (τ, 1 − k/2) are real numbers.

We now compare the Fourier coefficients of Fβ,m (τ, 1 − k/2) with the
coefficients of the holomorphic Poincaré series Pγ,n ∈ Sκ,L constructed in the
previous section.

Lemma 1.15. Let c ∈ Z − {0}, β, γ ∈ L0 /L, m ∈ Z + q(β) and n ∈ Z + q(β).

Then we have
∗
Hc (β, m, γ, n) = −H−c (γ, −n, β, −m).
36 1 Vector valued modular forms for the metaplectic group

Proof. We use the fact that (M̃ )−1 = M]−1 for any M ∈ Γ , and that κ =
1
2 − k. We find

e−πi sgn(c)k/2
^  
X ab ma + nd
Hc (β, m, γ, n) = %−1
γβ c d e
|c| ∗
c
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d

e−πi sgn(−c)(2−k)/2
^  
X db md + na
=− %−1
γβ e
|c| ∗
ca c
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d

e−πi sgn(−c)κ/2
^  
X a −b −na − md
=− %γβ e
|c| ∗
−c d −c
  d(c)
a b ∈Γ∞ \Γ1 /Γ∞
c d
∗
= −H−c (γ, −n, β, −m).

t
u
Proposition 1.16. Let β, γ ∈ L0 /L, n ∈ Z+q(γ), m ∈ Z+q(β), and m < 0.
Denote the (n, γ)-th Fourier coefficient b(γ, n, 1 − k/2) of Fβ,m (τ, 1 − k/2)
(see Proposition 1.10) by bβ,m (γ, n). If n < 0, then
1
bβ,m (γ, n) = − pγ,−n (β, −m),
Γ (1 − k)
where pγ,−n (β, −m) is the (β, −m)-th coefficient of the Poincaré series
Pγ,−n (τ ) ∈ Sκ,L (see 1.12). If n = 0, then

bβ,m (γ, 0) = −qγ (β, −m),

where qγ (β, −m) is the (β, −m)-th coefficient of the Eisenstein series Eγ (τ ) ∈
Mκ,L (see 1.19).
Proof. We compare the formulas for the Fourier coefficients given in Propo-
sition 1.10 and Theorem 1.4 (resp. Proposition 1.10 and Theorem 1.6) and
apply Lemma 1.15. Note that k − 1 = 1 − κ. t
u
For γ ∈ L0 /L and n ∈ Z − q(γ) with n > 0 let aγ,n : Sκ,L → C denote the
∗
functional in the dual space Sκ,L of Sκ,L which maps a cusp form f to its
(γ, n)-th Fourier coefficient aγ,n (f ). According to Proposition 1.5, aγ,n can
be described by means of the Petersson scalar product as
(4πn)κ−1
aγ,n = ( · , Pγ,n ). (1.41)
2Γ (κ − 1)
The following theorem can also be proved using Serre duality (see [Bo3]
Thm. 3.1). We deduce it from Proposition 1.16.
 1.3 Non-holomorphic Poincaré series of negative weight 37

!
Theorem 1.17. There exists a nearly holomorphic modular form f ∈ Mk,L
with prescribed principal part
X X
c(β, m)eβ (mτ )
β∈L0 /L m∈Z+q(β)
m<0

(c(β, m) ∈ C with c(β, m) = c(−β, m)), if and only if the functional

X X
c(β, m)aβ,−m ,
β∈L0 /L m∈Z+q(β)
m<0

∗
equals 0 in Sκ,L .
Proof. First, suppose that there exists such a nearly holomorphic modular
form f . Then by Proposition 1.12
1 X X
f (τ ) = c(β, m)Fβ,m (τ, 1 − k/2),
2 0
β∈L /L m∈Z+q(γ)
m<0

and in particular
∂ X X
c(β, m)Feβ,m (τ, 1 − k/2) = 0.
∂ τ̄
β∈L0 /L m∈Z+q(γ)
m<0

Since
∂
W1−k/2 (4πny)eγ (nx) 6= 0,
∂ τ̄
this implies that
X X
c(β, m)bβ,m (γ, n) = 0
β∈L0 /L m∈Z+q(γ)
m<0

for all γ ∈ L0 /L and n ∈ Z + q(γ) with n < 0. (Here we have used the same
notation as in Proposition 1.16.) Applying Proposition 1.16 we find that
X X X X
c(β, m)pγ,−n (β, −m) = c(β, m)aβ,−m (Pγ,−n )
β∈L0 /L m∈Z+q(γ) β∈L0 /L m∈Z+q(γ)
m<0 m<0
=0

for all γ ∈ L0 /L and n ∈ Z + q(γ) with n < 0. Since the Poincaré series Pγ,−n
generate the space Sκ,L we obtain
X X
c(β, m)aβ,−m = 0. (1.42)
β∈L0 /L m∈Z+q(β)
m<0
38 1 Vector valued modular forms for the metaplectic group

Now assume that (1.42) holds. Then we may reverse the above argument
to infer that
X X
c(β, m)Feβ,m (τ, 1 − k/2) = 0.
β∈L0 /L m∈Z+q(γ)
m<0

According to Proposition 1.10 this implies that

1 X X
c(β, m)Fβ,m (τ, 1 − k/2)
2 0
β∈L /L m∈Z+q(γ)
m<0

is a nearly holomorphic modular form with the right principal part. t

u
2 The regularized theta lift

In this chapter we consider a certain regularized theta lift of the Maass-

Poincaré series studied in the previous chapter.

2.1 Siegel theta functions

We use the same notation as in section 1.1. In particular L denotes an even
lattice of signature (b+ , b− ). The bilinear form on L induces a symmetric
bilinear form on V := L ⊗ R. We write Gr(L) for the Grassmannian of L, this
is the real analytic manifold of b+ -dimensional positive definite subspaces of
V . (For b+ = 0 or b− = 0 the Grassmannian is simply a point.) If v ∈ Gr(L),
we write v ⊥ for the orthogonal complement of v in V such that V = v ⊕ v ⊥ .
For any vector x ∈ V let xv resp. xv⊥ be the orthogonal projection of x to v
resp. v ⊥ . Then q(x) = 21 (x, x) = q(xv ) + q(xv⊥ ).
The Siegel theta function attached to L is defined by
X
θL (τ, v) = e (τ q(λv ) + τ̄ q(λv⊥ )) (τ ∈ H, v ∈ Gr(L)).
λ∈L

Following Borcherds we introduce a more general theta function. Let r, t ∈ V ,

γ ∈ L0 /L, τ ∈ H and v ∈ Gr(L). Define
X
θγ (τ, v; r, t) = e (τ q((λ + t)v ) + τ̄ q((λ + t)v⊥ ) − (λ + t/2, r)) (2.1)
λ∈γ+L

and X
ΘL (τ, v; r, t) = eγ θγ (τ, v; r, t). (2.2)
γ∈L0 /L

If r and t are both 0 we will omit them and simply write θγ (τ, v) resp. ΘL (τ, v).
Let Λ be the lattice Z2 equipped with the standard symplectic form and
W = Λ ⊗ R (such that Sp(Λ) = SL2 (Z)). Then W ⊗ V is a symplectic vec-
tor space. The usual theta function attached to the lattice Λ ⊗ L ⊂ W ⊗ V
restricted to the symmetric subspace H × Gr(L) ⊂ Sp(W ⊗ V )/K (where
K denotes a maximal compact subgroup) equals the Siegel theta function
θL (τ, v).
40 2 The regularized theta lift

It is well known that ΘL (τ, v; r, t) is a real analytic function in (τ, v) ∈

H × Gr(L).
Let

O(V ) = {g ∈ SL(V ); (gx, gy) = (x, y) for all x, y ∈ V }

be the (special) orthogonal group of V and

O(L) = {g ∈ O(V ); gL = L} .

be the orthogonal group of L. We denote by Od (L) the discriminant kernel of

O(L). This is the subgroup of finite index of O(L) consisting of all elements
which act trivially on the discriminant group L0 /L. Observe that Od (L) is
functorial in the following sense: If L̃ ⊂ L is a sublattice, then Od (L̃) ⊂ Od (L).
As a function in v ∈ Gr(L) the theta function ΘL (τ, v) is obviously in-
variant under the action of Od (L). As a function in τ ∈ H it transforms
like an elliptic modular form under the action of Mp2 (Z). Using the Poisson
summation formula the following transformation law can be established (see
[Bo2] Theorem 4.1).


Theorem 2.1. If (M, φ) ∈ Mp2 (Z) and M = ac db , then ΘL (τ, v; r, t) sat-
isfies
+ b−
ΘL (M τ, v; ar + bt, cr + dt) = φ(τ )b φ(τ ) %L (M, φ)ΘL (τ, v; r, t).

Borcherds showed that ΘL can be written as a Poincaré series involving

similar theta series ΘK which come from a smaller lattice K of signature
(b+ − 1, b− − 1) (cf. [Bo2] section 5). This is very useful for the evaluation of
“theta integrals” Z
+ dx dy
hF (τ ), ΘL (τ, v)i y b /2
y2
F

by the usual unfolding trick (Rankin-Selberg method).

In the rest of this section we review Borcherds construction in a slightly
more explicit way. First we need some basic facts about lattices.
Let z ∈ L be a primitive norm 0 vector (i.e. Qz ∩ L = Zz and q(z) = 0).
Then there exists a z 0 ∈ L0 with (z, z 0 ) = 1. Let N be the unique positive
integer with (z, L) = N Z. Then we have z/N ∈ L0 . Denote by K the lattice
⊥
K = L ∩ z⊥ ∩ z0 . (2.3)

Over Q we have
L ⊗ Q = (K ⊗ Q) ⊕ (Qz ⊕ Qz 0 ). (2.4)
Hence K has signature (b+ − 1, b− − 1). If n ∈ V = L ⊗ R then we write nK
for the orthogonal projection of n to K ⊗ R. It can be easily checked that nK
can be computed as
 2.1 Siegel theta functions 41

nK = n − (n, z)z 0 + (n, z)(z 0 , z 0 )z − (n, z 0 )z.

If n ∈ L0 then nK lies in the dual lattice K 0 of K. (Warning: Notice that K 0 is

not necessarily contained in L0 . Thus if n ∈ L, then nK does not necessarily
lie in K.)
Let ζ ∈ L be a lattice vector with (ζ, z) = N . It can be uniquely repre-
sented as
ζ = ζK + N z 0 + Bz (2.5)
with ζK ∈ K 0 and B ∈ Q.

Proposition 2.2. The lattice L can be written as L = K ⊕ Zζ ⊕ Zz.

Proof. Let n ∈ L. Then the vector

ñ = n − (n, z/N )ζ − (n, z 0 )z + (n, z/N )(ζ, z 0 )z

lies in L and an easy calculation shows that ñ ⊥ z and ñ ⊥ z 0 . Hence ñ ∈ K

and therefore n ∈ K + Zζ + Zz. The sum is obviously direct. t
u

If one uses the fact that |L0 /L| is given by the absolute value of the
Gram determinant of L, then the above proposition in particular shows that
|L0 /L| = N 2 |K 0 /K|.
Moreover, we may infer that γ 7→ γ − (γ, ζ)z/N defines an isometric
embedding K 0 → L0 . The kernel of the induced map K 0 → L0 /L equals
{γ ∈ K; (γ, ζ) ∈ N Z}.
Consider the sub-lattice

L00 = {λ ∈ L0 ; (λ, z) ≡ 0 (mod N )} (2.6)

of L0 . Obviously L is contained in L00 . Beside the orthogonal projection from

L00 to K 0 there is another natural projection L00 → K 0 : If we combine the
linear map L00 → L0 ∩ z ⊥ , λ 7→ λ − (λ,z)
N ζ with the orthogonal projection from
L0 ∩ z ⊥ to K 0 , we get a projection
(λ, z)
p : L00 −→ K 0 , p(λ) = λK − ζK . (2.7)
N
This map has the property that p(L) = K. In fact, if λ = k + aζ + bz ∈ L
with k ∈ K and a, b ∈ Z, then it is easily seen that p(λ) = k. Thus p induces
a surjective map L00 /L → K 0 /K which will also be denoted by p. Note that
L00 /L = {λ ∈ L0 /L; (λ, z) ≡ 0 (mod N )}.

We introduce some more notation. If x ∈ V then we sometimes sim-

ply write x2 instead of (x, x) or 2q(x). Furthermore, we abbreviate |x| :=
|(x, x)|1/2 . Let v ∈ Gr(L). We denote by w the orthogonal complement of zv
in v and by w⊥ the orthogonal complement of zv⊥ in v ⊥ . Hence one has the
orthogonal decomposition
42 2 The regularized theta lift

V = v ⊕ v ⊥ = w ⊕ Rzv ⊕ w⊥ ⊕ Rzv⊥ .

If x ∈ V , we write xw resp. xw⊥ for the orthogonal projection of x to w

resp. w⊥ . It can be easily verified that w and w⊥ are contained in (K⊗R)⊕Rz.
This implies that the orthogonal projection V → K ⊗ R induces an isometric
isomorphism w ⊕ w⊥ → wK ⊕ wK ⊥
= K ⊗ R. In particular we may identify w
with wK and consider w as an element of the Grassmannian Gr(K).
We denote by µ the vector
zv z ⊥
µ = −z 0 + + v2 (2.8)
2zv2 2zv⊥

in V ∩ z ⊥ = (K ⊗ R) ⊕ Rz.
Lemma 2.3. (cp. [Bo2] Lemma 5.1.) Let γ ∈ L0 /L. We have

1 X X
θγ (τ, v) = p e (τ q(λw ) + τ̄ q(λw⊥ ))
2yzv2 λ∈γ+K⊕Zζ d∈Z

|(λ, z)τ + d|2

 
d(λ, zv − zv⊥ )
×e − 2
− .
4iyzv 2zv2
Proof. For λ ∈ γ + K ⊕ Zζ and d ∈ R we define a function

g(λ, v; d) = e (τ q((λ + dz)v ) + τ̄ q((λ + dz)v⊥ )) .

Then X X
θγ (τ, v) = g(λ, v; d).
λ∈γ+K⊕Zζ d∈Z

We may apply the Poisson summation formula to rewrite the inner sum. We
find X X
θγ (τ, v) = ĝ(λ, v; d),
λ∈γ+K⊕Zζ d∈Z

where ĝ(λ, v; d) denotes the partial Fourier transform with respect to the
variable d. We now compute ĝ(λ, v; d). Using q(zv ) + q(zv⊥ ) = 0 we get

g(λ, v; d) = e d2 (τ − τ̄ )q(zv ) + d(τ (λ, zv ) + τ̄ (λ, zv⊥ )) + τ q(λv ) + τ̄ q(λv⊥ )

= e(Ad2 + Bd + C),

with A = (τ − τ̄ )q(zv ), B = (τ (λ, zv ) + τ̄ (λ, zv⊥ )), and C = τ q(λ

 v ) + τ̄ q(λ v ⊥ ).

−1/2 (x+B)2
The Fourier transform of e(Ax2 + Bx + C) equals ( 2A i ) e − 4A + C
(see for instance [Bo2] Cor. 3.3). Hence we find that ĝ(λ, v; d) is equal to
 2
−d − 2d(τ (λ, zv ) + τ̄ (λ, zv⊥ )) − (τ (λ, zv ) + τ̄ (λ, zv⊥ ))2

1
e
2(τ − τ̄ )zv2
p
2yzv2
× e (τ q(λv ) + τ̄ q(λv⊥ )) .
 2.1 Siegel theta functions 43

Using q(λv ) = q(λw ) + (λ, zv )2 /2zv2 and q(λv⊥ ) = q(λw⊥ ) + (λ, zv⊥ )2 /2zv2⊥
we obtain
τ (λ, zv )2 − τ̄ (λ, zv⊥ )2
 
1
ĝ(λ, v; d) = p e + τ q(λ w ) + τ̄ q(λ w ⊥ )
2yzv2 2zv2
 2
−d − 2d(τ (λ, zv ) + τ̄ (λ, zv⊥ )) − (τ (λ, zv ) + τ̄ (λ, zv⊥ ))2

×e .
2(τ − τ̄ )zv2
(2.9)

Since

2d(τ (λ, zv ) + τ̄ (λ, zv⊥ )) = d [(λ, zv − zv⊥ )(τ − τ̄ ) + (λ, z)(τ + τ̄ )] ,

the right hand side of (2.9) can be written as

 
1 d(λ, zv − zv⊥ )
e τ q(λw ) + τ̄ q(λw⊥ ) −
2zv2
p
2yzv2
 2
d + d(λ, z)τ + d(λ, z)τ̄ + (λ, z)2 τ τ̄

×e − ,
2(τ − τ̄ )zv2

which is equal to

d(λ, zv − zv⊥ ) |d + (λ, z)τ |2

 
1
e τ q(λw ) + τ̄ q(λw⊥ ) − − .
2zv2 4iyzv2
p
2yzv2

This implies the assertion. t

u

Theorem 2.4. (cp. [Bo2] Theorem 5.2.) Let γ ∈ L0 . We have

|cτ + d|2
 
1 X
0 0
θL+γ (τ, v) = p e − − d(γ, z ) + cdq(z )
2yzv2 4iyzv2
c,d∈Z
c≡(γ,z) (N )

× θK+p(γ−cz0 ) (τ, w; dµ, −cµ),

where p denotes the projection L00 → K 0 defined in (2.7).

Remark 2.5. Regarding the definition (2.1) of ΘK we should write more cor-
rectly µK instead of µ (note that µK = µ − (µ, z 0 )z). Since µw = (µK )w = zw
0

and (µ, z) = (µK , z), we permit ourselves this abuse of notation.

Proof. We want to apply Lemma 2.3. Write γ = γK +aγ z 0 +bγ z with γK ∈ K 0 ,

2
aγ = (γ, z) ∈ Z and bγ = (γ, z 0 ) − (γ, z)z 0 ∈ Q and observe that every
λ ∈ γ + K ⊕ Zζ can be uniquely written as

λ = λK + cz 0 + f (c, γ)z
44 2 The regularized theta lift

with c ∈ Z, c ≡ aγ (mod N ), and λK ∈ K +γK +ζK (c−aγ )/N . The function

f (c, γ) is defined by f (c, γ) = bγ +(c−aγ )B/N , and ζK , B are given by (2.5).
It can be easily checked that γK + ζK (c − aγ )/N = p(γ − cz 0 ).
We find that θL+γ (τ, v) is given by

1 X X
p e (τ q((λK + cz 0 )w ) + τ̄ q((λK + cz 0 )w⊥ ))
2yzv2 c,d∈Z λK ∈K+p(γ−cz 0 )
c≡(γ,z) (N )

|(λK + cz 0 , z)τ + d|2 d(λK + cz 0 + f (c, γ)z, zv − zv⊥ )

 
×e − − ,
4iyzv2 2zv2

where we have used zw = zw⊥ = 0 and q(z) = 0. Regarding the definition

(2.1) of ΘK (τ, w; dµ, −cµ), we see that we have to prove
d(λK + cz 0 + f (c, γ)z, zv − zv⊥ )
− ≡ −(λK − cµ/2, dµ) − d(γ, z 0 ) + cdq(z 0 )
2zv2

modulo 1. Since (z, zv − zv⊥ )/2zv2 = 1, we find

(f (c, γ)z, zv − zv⊥ )
−d = −df (c, γ)
2zv2
(γ, z) − c
= −dbγ + dB
N
2 (γ, z) − c
= −d(γ, z 0 ) + d(γ, z)z 0 + dB.
N
2
From (ζ, z 0 ) ∈ Z it follows that B ≡ −N z 0 (mod 1), and this implies that
(f (c, γ)z, zv − zv⊥ ) 2
−d ≡ −d(γ, z 0 ) + cdz 0 (mod 1).
2zv2
Thus it suffices to show
(λK + cz 0 , zv − zv⊥ )
−d = −(λK − cµ/2, dµ) − cdq(z 0 ). (2.10)
2zv2
The left hand side of (2.10) equals

−d λK + cz 0 , zv /2zv2 + zv⊥ /2zv2⊥

= −d λK + c(z 0 − zv /2zv2 − zv⊥ /2zv2⊥ )/2, zv /2zv2 + zv⊥ /2zv2⊥

dc 0
z , zv /2zv2 + zv⊥ /2zv2⊥


−
2
dc
zv /2zv2 + zv⊥ /2zv2⊥ , zv /2zv2 + zv⊥ /2zv2⊥ .


− (2.11)
2
Since q(z) = 0, the third term on the right hand side of (2.11) is 0. The
second term can be written as
 2.1 Siegel theta functions 45

dc
−d λK + c(z 0 − zv /2zv2 − zv⊥ /2zv2⊥ )/2, −z 0 − (z 0 , z 0 ).
2
Thus the left hand side of (2.10) equals

−d λK + c(z 0 − zv /2zv2 + zv⊥ /2zv2⊥ )/2, −z 0 + zv /2zv2 + zv⊥ /2zv2⊥ − dcq(z 0 )

= −d(λK − cµ/2, µ) − cdq(z 0 ).

t
u

Let k ∈ 21 Z and FL (τ ) = δ∈L0 /L eδ fδ (τ ) be a function H → C[L0 /L]

P

with FL |k (M, φ) = FL for all (M, φ) ∈ Mp2 (Z). Let r, t be given integers.
We define a function
X
FK (τ ; r, t) = eγ fγ (τ ; r, t), (2.12)
γ∈K 0 /K

by putting
X
fγ (τ ; r, t) = e (−r(λ, z 0 ) − rtq(z 0 )) fL+λ+tz0 (τ )
λ∈L00 /L
p(λ)=γ

for γ ∈ K 0 /K. (This is just a different notation for the definition given in
[Bo2] on page 512.) Let γ ∈ K 0 . Then as a set of representatives for λ ∈ L00 /L
with p(λ) = γ +K we may take γ −(γ, ζ)z/N +bz/N where b runs modulo N .
A set of representatives for L00 /L is given by λ = γ − (γ, ζ)z/N + bz/N , where
γ runs through a set of representatives for K 0 /K and b runs modulo N . (Since
the index of L00 in L0 equals N , we find again that |L0 /L| = N 2 |K 0 /K|.)

Theorem
2.6. We use the same notation as above. Let (M, φ) ∈ Mp2 (Z),
M = ac db . Then FK (τ ; r, t) satisfies

FK (M τ ; ar + bt, cr + dt) = φ(τ )2k %K (M, φ)FK (τ ; r, t).

Proof. See [Bo2] Theorem 5.3. t

u
L
For the Poincaré series Fβ,m (τ, s) (see Definition 1.8) we denote the
0 K
C[K /K]-valued function constructed in (2.12) by Fβ,m (τ, s; r, t).

Proposition 2.7. Let β ∈ L0 /L and m ∈ Z + q(β) with m < 0.

K
i). If (β, z) 6≡ 0 (mod N ), then Fβ,m (τ, s; 0, 0) = 0.
K
ii). If (β, z) ≡ 0 (mod N ), then the function Fβ,m (τ, s; 0, 0) equals the
K
Poincaré series Fp(β),m (τ, s) of index (p(β), m) attached to the lattice K (see
Definition 1.35).
46 2 The regularized theta lift

K
Proof. i). Assume that σ ≥ 1 − k/2. From the definition of Fβ,m (τ, s; 0, 0),
L
the Fourier expansion of Fβ,m (τ, s) (see Theorem 1.9), and the asymptotic
property (1.26) of the Whittaker function Wν,µ (y) it follows that
K
Fβ,m (τ, s; 0, 0) = O(1)
K
for y → ∞. Since Fβ,m (τ, s; 0, 0) also satisfies
K K
Fβ,m (τ, s; 0, 0) |k (M, φ) = Fβ,m (τ, s; 0, 0), (M, φ) ∈ Mp2 (Z),
K 2

K
∆k Fβ,m (τ, s; 0, 0) = s(1 − s) + (k − 2k)/4 Fβ,m (τ, s; 0, 0),

we may infer that it vanishes identically (cf. [He] Chap. 9 Prop. 5.14c). Be-
K
cause Fβ,m (τ, s; 0, 0) is holomorphic in s, we find that it vanishes for all s.
ii) In the same way as in (i) it follows that
K
Fβ,m (τ, s; 0, 0)
Γ (1 + k/2 − s)
= M1−s (4π|m|y)(ep(β) (mx) + e−p(β) (mx)) + O(1)
Γ (2 − 2s)Γ (s + k/2)
for y → ∞, and thereby
K K
Fβ,m (τ, s; 0, 0) − Fp(β),m (τ, s) = O(1) (y → ∞).

The assertion can be deduced as in (i). t

u

2.2 The theta integral

As in section 1.3 let L be an even lattice of signature (b+ , b− ) = (2, l), (1, l−1),
or (0, l − 2) with l ≥ 3. Put k = 1 − l/2 and

σ0 = max{1, b+ /2 − k/2}.

Let β ∈ L0 /L and m ∈ Z + q(β) with m < 0. We consider the theta integral

Z
L L + dx dy
Φβ,m (v, s) = hFβ,m (τ, s), ΘL (τ, v)iy b /2 , (2.13)
y2
F

where Fβ,m (τ, s) denotes the Maass-Poincaré series of index (β, m) defined in
(1.35) and ΘL (τ, v) the Siegel theta function (2.2). If it is clear from the con-
text, to which lattice L the function (2.13) refers, we simply write Φβ,m (v, s).
According to Theorem 2.1, the integrand
+
hFβ,m (τ, s), ΘL (τ, v)iy b /2

is Mp2 (Z)-invariant. Because ΘL (τ, v) is invariant under the action of Od (L),

the function Φβ,m (v, s) is formally invariant under Od (L), too.
 2.2 The theta integral 47

Unfortunately, since |Fβ,m (τ, s)| increases exponentially as y → ∞, the

integral (2.13) diverges. However, we shall show that it can be regularized as
follows. Let Fu be the truncated fundamental domain

Fu = {τ = x + iy ∈ F; y ≤ u} (u ∈ R>0 ). (2.14)

Then for s = σ + it ∈ C with σ > σ0 we may define the regularized theta lift
of Fβ,m (τ, s) by
Z
+ dx dy
ΦL
β,m (v, s) = lim L
hFβ,m (τ, s), ΘL (τ, v)iy b /2 . (2.15)
u→∞ y2
Fu

Roughly speaking this means that we first carry out the integration over
x and afterwards the integration over y. We will show that Φβ,m (v, s) as
defined in (2.15) is holomorphic in s for σ > σ0 . It can be continued to a
holomorphic function on {s ∈ C; σ > 1, s 6= b+ /2 − k/2} with a simple
pole at s = b+ /2 − k/2. Hence for σ > 1 and s 6= b+ /2 − k/2 the function
Φβ,m (v, s) is given by holomorphic continuation. For s = b+ /2 − k/2 we may
define the regularized theta integral to be the constant term of the Laurent
expansion in s of Φβ,m (v, s) at the point s = b+ /2 − k/2.
This construction is very similar to the definition of the regularized theta
lift due to Harvey, Moore, and Borcherds [HM, Bo2]. Following Borcherds, one
could also define Φβ,m (v, s) to be the constant term of the Laurent expansion
in s0 of Z
+ 0 dx dy
lim hFβ,m (τ, s), ΘL (τ, v)iy b /2−s (2.16)
u→∞ y2
Fu
0 0
at s = 0. Here s is an additional complex variable. The above expression
makes sense if <(s0 )  0. Note that the integrand in (2.16) is only invariant
under Mp2 (Z) if s0 = 0. We will prove in Proposition 2.11 that (2.15) and
(2.16) essentially yield to the same Φβ,m (v, s). However, for our purposes the
definition of the regularized theta lift given in (2.15) is more natural.
As in [Bo2] we will show that Φβ,m (v, s) is a real analytic function on
Gr(L) with singularities along smaller sub-Grassmannians. To make this more
precise, we define a subset H(β, m) of the Grassmannian Gr(L) by
[
H(β, m) = λ⊥ . (2.17)
λ∈β+L
q(λ)=m

Here λ⊥ means the orthogonal complement of λ in Gr(L), i.e. the set of all
positive definite b+ -dimensional subspaces v ⊂ V with v ⊥ λ.1 It is known
that for any subset U with compact closure U ⊂ Gr(L) the set

S(β, m, U ) = {λ ∈ β + L; q(λ) = m, ∃v ∈ U with v ⊥ λ} (2.18)

1 +
If b = 0 we put H(β, m) = ∅.
48 2 The regularized theta lift

is finite. Hence H(β, m) is locally a finite union of codimension b+ sub-

Grassmannians. It will turn out that Φβ,m (v, s) is a real analytic function
on Gr(L) − H(β, m).

Proposition 2.8. Let v ∈ Gr(L)−H(β, m) and s ∈ C with σ > σ0 . Then the

regularized theta integral (2.15) converges and defines a holomorphic function
in s. It can be continued holomorphically to

{s ∈ C; σ > 1, s 6= b+ /2 − k/2},

and has a pole of first order with residue b(0, 0, b+ /2 − k/2) at s = b+ /2 − k/2
(supposed that b+ /2 − k/2 > 1).

Proof. Recall that Fβ,m (τ, s) is holomorphic in s for σ > 1. The integral
Z
+ dx dy
hFβ,m (τ, s), ΘL (τ, v)iy b /2
y2
F1

over the compact set F1 is clearly holomorphic in s for σ > 1. Thus it suffices
to consider the function
Z∞ Z1
+
ϕ(v, s) = hFβ,m (τ, s), ΘL (τ, v)iy b /2−2
dx dy. (2.19)
y=1 x=0

We insert the Fourier expansions

X X
Fβ,m (τ, s) = c(γ, n; y, s)eγ (nx)
γ∈L0 /L n∈Z+q(γ)
X
ΘL (τ, v) = e(iyq(λv ) − iyq(λv⊥ ))eλ (q(λ)x)
λ∈L0

into (2.19), carry out the integral over x, and obtain

Z∞ X
+
ϕ(v, s) = c(λ, q(λ); y, s) exp(−2πyq(λv ) + 2πyq(λv⊥ ))y b /2−2
dy
0
y=1 λ∈L
Z∞
+
= c(0, 0; y, s)y b /2−2
dy
y=1
Z∞ X +
+ c(λ, q(λ); y, s) exp(−4πyq(λv ) + 2πyq(λ))y b /2−2
dy.
y=1 λ∈L0 −0

According to Theorem 1.9 the Fourier expansion of Fβ,m (τ, s) can be written
more explicitly in the form
 2.2 The theta integral 49

1
Fβ,m (τ, s) = Ms (4π|m|y)(eβ (mx) + e−β (mx)) + H(τ, s),
Γ (2s)
where
X X
H(τ, s) = b̃(γ, 0, s)y 1−s−k/2 + b̃(γ, n, s)Ws (4πny)eγ (nx),
γ∈L0 /L γ∈L0 /L
q(γ)∈Z n∈Z+q(γ)
n6=0
Γ (1 + k/2 − s)
b̃(γ, n, s) = b(γ, n, s) − δm,n (δβ,γ + δ−β,γ ).
Γ (2s)Γ (1 − 2s)
Observe that all Fourier coefficients b̃(γ, n, s) are holomorphic in s for σ >
1. (This is not true for b(±β, m, s). That is why we have introduced the
b̃(γ, n, s).) We find that ϕ(v, s) is given by
Z∞
+
b(0, 0, s) yb /2−k/2−s−1
dy
1
Z∞ X
+
+ b(λ, 0, s) exp(−4πyq(λv ))y b /2−k/2−s−1
dy
0
1 λ∈L −0
q(λ)=0
Z∞ X
+
+ b̃(λ, q(λ), s)Ws (4πq(λ)y) exp(2πyq(λv⊥ ) − 2πyq(λv ))y b /2−2
dy
0
1 λ∈L
q(λ)6=0
Z∞ X
2 +
+ Ms (4π|m|y) exp(2πym − 4πyq(λv ))y b /2−2
dy. (2.20)
Γ (2s)
1 λ∈β+L
q(λ)=m

The first integral obviously converges for σ > b+ /2 − k/2 and equals
1
.
s − b+ /2 + k/2
Thus the first summand in (2.20) has a holomorphic continuation to {s ∈
C; σ > 1, s 6= b+ /2 − k/2}. If b+ /2 − k/2 > 1 it has a simple pole at
s = b+ /2 − k/2 with residue b(0, 0, b+ /2 − k/2).
Therefore it suffices to show that the remaining terms in (2.20) converge
locally uniformly absolutely for σ > 1.
For the convergence of the second term, notice that q(λ) = 0 implies
q(λv ) = 21 q(λv ) − 12 q(λv⊥ ). The latter quadratatic form is a multiple of the
(positive definite) majorant associated with v. Thus the sum over λ is domi-
nated by a subseries of a theta series attached to a definite lattice.
To prove the convergence of the third term we also use the majorant
associated with v, and in addition the asymptotic property (1.26) of the
Whittaker function Wν,µ (y) and the estimates
50 2 The regularized theta lift
 p 
b(γ, n, s) = O exp(4π |mn| ) , n → +∞, (2.21)
 
b(γ, n, s) = O |n|σ+k/2−1 , n → −∞ (2.22)

(locally uniformly in s for σ > 1). These bounds can be easily obtained from
the asymptotic behavior of the Iν - resp. Jν -Bessel function and the fact that
Hc (β, m, γ, n) is universally bounded.
If b+ = 0, then L is a definite lattice, and λv = 0 for all λ. The convergence
of the last term in (2.20) follows from the asymptotic property (1.25) of the
M -Whittaker function.
To prove the convergence of the last term for b+ = 1, 2 we again use the
asymptotic property (1.25) and the following fact: For any C ≥ 0 and any
compact subset B ⊂ Gr(L) the set

{λ ∈ L0 ; q(λ) = m, ∃v ∈ B with q(λv ) ≤ C}

is finite. By assumption v ∈ / H(β, m). Hence there exists an ε > 0 such that
q(λv ) > ε for all λ ∈ L0 with q(λ) = m. It follows that
X +
Ms (4π|m|y) exp(−4πyq(λv ) + 2πym)y b /2−2

λ∈β+L
q(λ)=m
X
 exp(−4πyq(λv ))
λ∈β+L
q(λ)=m
X
 e−2πεy e−π(m+q(λv )−q(λv⊥ ))
λ∈L0
q(λ)=m

uniformly for y ∈ [1, ∞) and locally uniformly in s. This implies the desired
convergence statement. t
u

Since Fβ,m = F−β,m , we have Φβ,m (v, s) = Φ−β,m (v, s).

Definition 2.9. Let D ⊂ C be an open subset, a ∈ D, and f a meromorphic

function on D. We denote the constant term of the Laurent expansion of f
at s = a by Cs=a [f (s)].

Definition 2.10. For v ∈ Gr(L) − H(β, m) we define

Φβ,m (v) = Cs=1−k/2 [Φβ,m (v, s)] .

If b+ = 0 or b+ = 1, then Φβ,m (v, s) is holomorphic at s = 1 − k/2 and

we may also write Φβ,m (v) = Φβ,m (v, 1 − k/2).

Proposition 2.11. Borcherds’ regularization of the theta integral as de-

scribed in (2.16) equals
 2.2 The theta integral 51

Φβ,m (v, s), if s 6= b+ /2 − k/2,

Cs=b+ /2−k/2 [Φβ,m (v, s)] − b0 (0, 0, b+ /2 − k/2), if s = b+ /2 − k/2.
Here b0 (0, 0, b+ /2 − k/2) means the derivative of the coefficient b(0, 0, s) at
s = b+ /2 − k/2.
Proof. The same argument as in Proposition 2.8 shows that Borcherds’ reg-
ularization of the theta integral can be written as
Z  
+ dx dy b(0, 0, s)
hFβ,m (τ, s), ΘL (τ, v)iy b /2 + C 0
s =0
y2 s0 + s − b+ /2 + k/2
F1
Z∞ X
+
+ c(λ, q(λ); y, s) exp(−4πyq(λv ) + 2πyq(λ))y b /2−2
dy.
0
1 λ∈L −0

If we compare this with the expression for ϕ(v, s) given in Proposition 2.8, we
immediately obtain the assertion in the case s 6= b+ /2 − k/2. For s = b+ /2 −
k/2 we find that the difference of our regularization Cs=b+ /2−k/2 [Φβ,m (v, s)]
and Borcherds’ regularization is given by
 
b(0, 0, s) +
h
0 −1
i
Cs=b+ /2−k/2 − b(0, 0, b /2 − k/2)Cs 0 =0 s
s − b+ /2 + k/2
= b0 (0, 0, b+ /2 − k/2).
t
u
+
For the rest of this section we assume that b = 1, 2. In Proposition 2.8
we considered Φβ,m (v, s) for a fixed v and varying s. Now we show that for
fixed s the function Φβ,m (v, s) is real analytic in v on Gr(L) − H(β, m).
Let U ⊂ Gr(L) be an open subset and f , g functions on a dense open
subset of U . Then we write
f ≈ g,
if f − g can be continued to a real analytic function on U . In this case we
will say that f has a singularity of type g.
Theorem 2.12. i) For fixed s ∈ {s ∈ C; σ > 1, s 6= b+ /2 − k/2} the
function Φβ,m (v, s) is real analytic in v on Gr(L) − H(β, m).
ii) Let U ⊂ Gr(L) be an open subset with compact closure U ⊂ Gr(L).
The function Φβ,m (v) is real analytic on Gr(L) − H(β, m). On U it has a
singularity of type X
−2 log q(λv ),
λ∈S(β,m,U )
+
if b = 2, and of type √ X
−4 2π |λv |,
λ∈S(β,m,U )

if b+ = 1.
52 2 The regularized theta lift

Proof. We use the same argument as in [Bo2] Theorem 6.2. We only give a
proof for (ii), because (i) can be treated similarly.
Since ΘL (τ, v) is real analytic and Fβ,m (τ, s) is holomorphic in s, the
integral Z
+ dx dy
hFβ,m (τ, s), ΘL (τ, v)iy b /2
y2
F1

is holomorphic in s and real analytic in v. As in the proof of Proposition 2.8

it suffices to consider the function ϕ(v, s) defined by (2.19). We insert the
Fourier expansions of ΘL (τ, v) and Fβ,m (τ, s) (as given in Theorem 1.9) and
infer as in the proof of Proposition 2.8 that ϕ(v, s) is equal to
Z∞ X
b(0, 0, s) +

+
+ b(λ, 0, s) exp(−4πyq(λv ))y b /2−k/2−s−1
dy
b /2 − k/2 − s 0
1 λ∈L −0
q(λ)=0
Z∞ X
+
+ b(λ, q(λ), s)Ws (4πq(λ)y) exp(−2πyq(λv ) + 2πyq(λv⊥ ))y b /2−2
dy
0
1 λ∈L
q(λ)6=0
Z∞ X
2Γ (1 + k/2 − s)
+ M1−s (4π|m|y)
Γ (2 − 2s)Γ (s + k/2)
1 λ∈β+L
q(λ)=m
+
× exp(−4πyq(λv ) + 2πym)y b /2−2
dy. (2.23)

The first quantity does not depend on v. The second and the third term are
holomorphic in s near s = 1 − k/2 and real analytic in v ∈ Gr(L). Hence they
do not contribute to the singularity. The last term is holomorphic in s near
s = 1 − k/2 and real analytic for v ∈ Gr(L) − H(β, m). We find that

Φβ,m (v) ≈ Cs=1−k/2 [ϕ(v, s)]

∞
X Z +
≈2 Mk/2 (4π|m|y) exp(−4πyq(λv ) + 2πym)y b /2−2 dy
λ∈β+L 1
q(λ)=m
Z∞
X +
≈2 exp(−4πyq(λv ))y b /2−2
dy.
λ∈β+L 1
q(λ)=m

In the last line we have used Mk/2 (y) = ey/2 . The integrals in the last line
only become singular if q(λv ) = 0. Thus for v ∈ U we get
 2.3 Unfolding against Fβ,m 53

Z∞
X +
Φβ,m (v) ≈ 2 exp(−4πyq(λv ))y b /2−2
dy
λ∈S(β,m,U ) 1
X +
≈2 (4πq(λv ))1−b /2
Γ (b+ /2 − 1, 4πq(λv )),
λ∈S(β,m,U )

where Γ (a, x) denotes the incomplete Gamma function defined in (1.32)

(cf. [AbSt] p. 81). Now the singularity can be read off from [Bo2] Lemma
6.1. t
u
Note that the functions Γ (0, x) and Γ (−1/2, x) that occur at the end
of the proof of Theorem 2.12 can be described more explicitly as follows:
According to [E1] Vol. II p. 143 (5) one has
∞
X (−x)n
Γ (0, x) = −γ − log(x) − , (2.24)
n=1
n! n

where γ denotes the Euler-Mascheroni constant. Using the recurrence relation

Γ (a + 1, x) = aΓ (a, x) + xa e−x

([E1] Vol. II p. 134 (3)) and the identity

√
Γ (1/2, x2 ) = π(1 − erf(x))

([AbSt] p. 82 (6.5.17)), one finds

2 √
Γ (−1/2, x2 ) = 2|x|−1 e−x − 2 π(1 − erf(x)). (2.25)
x 2
Here erf(x) = π2 0 e−t dt denotes the error function (cf. [AbSt] chapter 7).
R

It has the expansion

∞
2 X (−1)n x2n+1
erf(x) = .
π n=0 n!(2n + 1)

In particular, from (2.24) and (2.25) we can also read off the singularities of
Γ (0, x) and Γ (−1/2, x).

2.3 Unfolding against Fβ,m

As in section 1.3 let L be an even lattice of signature (b+ , b− ) = (2, l), (1, l−1),
or (0, l − 2) with l ≥ 3, and put k = 1 − l/2. In this section we find a nice
invariant expression for Φβ,m (v, s) by unfolding the theta integral Φβ,m (v, s)
against the Poincaré series Fβ,m (τ, s). In [Br1] we started with such an invari-
ant expression for Φβ,m (v, s), calculated its Fourier expansion, and thereby
obtained a connection to certain elliptic modular forms.
54 2 The regularized theta lift

Lemma 2.13. The Siegel theta function satisfies

+
/2−b− /2
θβ (τ, v) = O(y −b ) (y → 0)

uniformly in x.

Proof. This can be proved in the same way as the analogous statement for
holomorphic elliptic modular forms. u
t

Let F (a, b, c; z) denote the Gauss hypergeometric function

∞
X (a)n (b)n z n
F (a, b, c; z) = , (2.26)
n=0
(c)n n!

where (a)n = Γ (a + n)/Γ (a) (cf. [AbSt] Chap. 15 or [E1] Vol. I Chap. 2).
The circle of convergence of the series (2.26) is the unit circle |z| = 1. Using
the estimate ([E1] Vol. I p. 57 (5))

(a)n (b)n Γ (c)

na+b−c−1 1 + O(n−1 ) ,


= n → ∞,
(c)n n! Γ (a)Γ (b)

one finds that (2.26) converges absolutely for |z| = 1, if <(c − a − b) > 0.
Then the value for z = 1 is given by ([E1] Vol. I p. 104 (46))

Γ (c)Γ (c − a − b)
F (a, b, c; 1) = (c 6= 0, −1, −2, . . . ). (2.27)
Γ (c − a)Γ (c − b)

Theorem 2.14. Let β ∈ L0 /L and m ∈ Z + q(β) with m < 0. We have the

identity

Γ (b− /4 + b+ /4 + s − 1) X 1−b− /4−b+ /4−s

Φβ,m (v, s) = 2 k/2−s
(4π|q(λv⊥ )|)
Γ (2s)(4π|m|) λ∈β+L
q(λ)=m

× F (b /4 + b /4 + s − 1, s − b− /4 + b+ /4, 2s; m/q(λv⊥ )).

− +

The series converges normally for v ∈ Gr(L)−H(β, m) and σ > b− /4+b+ /4.
In particular, if (b+ , b− ) = (0, l − 2), then

8π|m|
Φβ,m (v, s) = · #{λ ∈ β + L; q(λ) = m}.
(s + l/4 − 3/2)Γ (s + 3/2 − l/4)

Proof. Let v ∈ Gr(L) − H(β, m) and s ∈ C with σ > b− /4 + b+ /4. By

definition the function Φβ,m (v, s) is equal to
Z
1 X D E + dx dy
[Ms (4π|m|y)eβ (mx)] |k M̃ , ΘL (τ, v) y b /2 ,
Γ (2s) y2
F M ∈Γ∞ \Γ1
 2.3 Unfolding against Fβ,m 55

where the integral is to be understood in the regularized sense (cp. (2.15)).

According to the theta transformation formula (Theorem 2.1) this can be
written as
Z
1 X +
Ms (4π|m|=(M τ ))(=M τ )b /2
Γ (2s)
F M ∈Γ∞ \Γ1
dx dy
× heβ (m<(M τ )), ΘL (M τ, v)i
y2
Z
2 + dx dy
= Ms (4π|m|y)y b /2 e(mx)θβ (τ, v) 2
Γ (2s) y
F
Z
1 X +
+ Ms (4π|m|=(M τ ))(=M τ )b /2
Γ (2s)  
F M = a b ∈Γ∞ \Γ1
c d
c6=0
dx dy
× e(m<(M τ ))θβ (M τ, v) . (2.28)
y2

The second integral on the right hand side of (2.28) can be evaluated by the
usual unfolding trick. It equals
Z
2 + dx dy
Ms (4π|m|y)y b /2 e(mx)θβ (τ, v) 2 , (2.29)
Γ (2s) y
G

where
G = {τ ∈ H; |x| ≤ 1/2, |τ | ≤ 1} (2.30)
S
is a fundamental domain for the action of Γ∞ on H − M ∈Γ∞ M F. By virtue
of (1.23) and Lemma 2.13, the integral in (2.29) converges absolutely if σ >
1 + b− /4 + b+ /4, and therefore the unfolding is justified. Combining (2.29)
with the first integral on the right hand side of (2.28), we obtain

Z∞ Z1
2 +
Φβ,m (v, s) = Ms (4π|m|y)y b /2−2
e(mx)θβ (τ, v) dx dy.
Γ (2s)
y=0 x=0

We now insert the Fourier expansion

X
θβ (τ, v) = exp(−2πyq(λv ) + 2πyq(λv⊥ ))e(q(λ)x)
λ∈β+L

of θβ (τ, v) and carry out the integration over x. We find

56 2 The regularized theta lift

Z∞ X
2 +
/2−2 −4πyq(λv )+2πym
Φβ,m (v, s) = Ms (4π|m|y)y b e dy
Γ (2s)
y=0 λ∈β+L
q(λ)=m
∞
2(4π|m|)−k/2 X Z −
/4+b+ /4−2
= M−k/2, s−1/2 (4π|m|y)y b
Γ (2s)
λ∈β+L 0
q(λ)=m

× e−4πyq(λv )+2πym dy. (2.31)

The integral in (2.31) is a Laplace transform. If λv 6= 0, it equals (cp. [E2]

p. 215 (11))

(4π|m|)s Γ (b− /4 + b+ /4 + s − 1)
(4π|q(λv⊥ )|)b− /4+b+ /4+s−1
× F b− /4 + b+ /4 + s − 1, s − b− /4 + b+ /4, 2s; m/q(λv⊥ ) . (2.32)

In the case (b+ , b− ) = (0, l − 2) the projection λv is always 0, and the

integral in (2.31) is
Z∞
M−k/2, s−1/2 (4π|m|y)y l/4−5/2 e2πym dy. (2.33)
0

By the asymptotic behavior of the M -Whittaker function, this integral con-

verges absolutely for σ > 3/2 − l/4. It can be easily checked that (2.32) still
holds:
Z∞
(2.33) = lim M−k/2, s−1/2 (4π|m|y)y l/4−5/2 e2πym−εy dy
ε→0+
0
3 l
= (4π|m|) 2 − 4 Γ (s + l/4 − 3/2)F (s + l/4 − 3/2, s + 1/2 − l/4, 2s; 1).

Using (2.27) we find

Γ (s + l/4 − 3/2)Γ (2s)

(2.33) = (4π|m|)3/2−l/4 . (2.34)
Γ (s − l/4 + 3/2)Γ (s + l/4 − 1/2)

If we substitute (2.32) resp. (2.34) into (2.31), we obtain the assertion. t

u

2.4 Unfolding against ΘL

Throughout the following section, let L be an indefinite even lattice of sig-
nature (b+ , b− ) = (2, l) or (1, l − 1) with l ≥ 3, and put k = 1 − l/2. Recall
the notation introduced in section 2.1.
 2.4 Unfolding against ΘL 57

Assume that L contains a primitive norm 0 vector z. Then we may use

the method of Borcherds to evaluate the theta integral Φβ,m (v, s) (cf. [Bo2]
Chap. 7). Thereby we obtain the Fourier expansion of Φβ,m (v, s) in terms of
the Fourier coefficients of Fβ,m (τ, s).
We denote the components of Fβ,m (τ, s) by fγ (τ, s), so that
X
Fβ,m (τ, s) = eγ fγ (τ, s).
γ∈L0 /L

Here we write the Fourier expansion of fγ (τ, s) in the form

X
fγ (τ, s) = c(γ, n; y, s)eγ (nx).
n∈Z+q(γ)

The coefficients c(γ, n; y, s) were calculated in section 1.3. For instance, for
γ ∈ L0 /L with q(γ) ∈ Z we found:

c(γ, 0; y, s) = b(γ, 0, s)y 1−s−k/2 ,

41−k/2 π 1+s−k/2 |m|s−k/2 X
b(γ, 0, s) = |c|1−2s Hc (β, m, γ, 0).
(2s − 1)Γ (s + k/2)Γ (s − k/2)
c∈Z−{0}

Theorem 2.15. (cp. [Bo2] Theorem 7.1.) Let β ∈ L0 /L and m ∈ Z + q(β),

1
m < 0. If v ∈ Gr(L) − H(β, m) and zv2 < 4|m| , then ΦL
β,m (v, s) is equal to

 2 s−b+ /4−b− /4
1 2 2zv
√ K
Φβ,m (w, s) + √ Γ (s + 1/2 − b+ /4 − b− /4)
2|zv | π π
X X + −
× b(`z/N, 0, s) e(`n/N )nb /2+b /2−1−2s
` (N ) n≥1
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))
|zv |
λ∈K 0 −0 n≥1 δ∈L00 /L
p(δ)=λ+K
Z∞
−πn2
 
b+ /2−5/2
× c(δ, q(λ); y, s)y exp − 2πyq(λw ) + 2πyq(λw⊥ ) dy.
2zv2 y
0
(2.35)

Here ΦK
β,m (w, s) is defined by
(
0, if (β, z) 6≡ 0 (mod N ),
ΦK
β,m (w, s) =
ΦK
p(β),m (w, s), if (β, z) ≡ 0 (mod N ).

The third summand in the above expression for ΦL

β,m (v, s) converges normally
in s for σ > 1.
58 2 The regularized theta lift

Proof. We use Theorem 2.4 to rewrite ΦL

β,m (v, s) as

|cτ + d|2
Z  
1 X X
0 0
fγ (τ, s) e − + d(γ, z ) − cdq(z )
4iyzv2
p
2zv2 0 c,d∈Z
F γ∈L /L
c≡(γ,z) (N )
+ dx dy
× θK+p(γ−cz0 ) (τ, w; dµ, −cµ)y (b −1)/2
y2
Z
1 X + dx dy
=p fγ (τ, s)θK+p(γ) (τ, w)y (b −1)/2
2
2zv y2
γ∈L0 /L
F
(γ,z)≡0 (N )
Z X
1 0 X
+p e (d(γ, z 0 ) − cdq(z 0 )) fγ (τ, s)
2zv2 γ∈L0 /L
F c,d∈Z
(γ,z)≡c (N )

|cτ + d|2
 
+
−1)/2 dx dy
×e − θK+p(γ−cz0 ) (τ, w; dµ, −cµ)y (b . (2.36)
4iyzv2 y2
X0
Here means that the sum runs over all pairs (c, d) ∈ Z2 with (c, d) 6=
c,d∈Z
K
(0, 0). If we substitute in the definition (2.12) of Fβ,m (τ, s; 0, 0), we obtain for
the first term on the right hand side of (2.36):
Z
1 + dx dy
K
Fβ,m (τ, s; 0, 0), ΘK (τ, w) y (b −1)/2 .
y2
p
2zv2
F

By Proposition 2.7 this equals √1 2 ΦK

β,m (w, s). 2zv
In the second term on the right hand side of (2.36) we change γ to γ + cz 0
K
and substitute in the definition (2.12) of Fβ,m (τ, s; −d, c). We find that it is
equal to

|cτ + d|2
Z X  
1 0
K
e − Fβ,m (τ, s; −d, c), ΘK (τ, w; dµ, −cµ)
4iyzv2
p
2zv2
F c,d∈Z
+ dx dy
−1)/2
× y (b
y2
1
Z X X  n2 |cτ + d|2 
=p e −
2zv2 4iyzv2
F (c,d)=1 n≥1

K +
−1)/2 dx dy
× Fβ,m (τ, s; −nd, nc), ΘK (τ, w; ndµ, −ncµ) y (b . (2.37)
y2
By Theorem 2.1 and Theorem 2.6 we get
 2.4 Unfolding against ΘL 59

−πn2
Z  
1 X X
p exp 2
2zv2 2zv =(M τ )
F M ∈Γ∞ \Γ1 n≥1

K +
−1)/2 dx dy
× Fβ,m (M τ, s; −n, 0), ΘK (M τ, w; nµ, 0) =(M τ )(b .
y2
As in the proof of Theorem 2.14 we split this up into
√ Z
−πn2
 
2 X + dx dy
exp K
Fβ,m (τ, s; −n, 0), ΘK (τ, w; nµ, 0) y (b −1)/2 2
|zv | 2zv2 y y
F n≥1
√ Z
−πn2
 
2 X X
+ exp
|zv | 2zv2 =(M τ )
F M = a b ∈Γ∞ \Γ 1 n≥1
 
c d
c6=0

K +
−1)/2 dx dy
× Fβ,m (M τ, s; −n, 0), ΘK (M τ, w; nµ, 0) =(M τ )(b , (2.38)
y2

where Γ 1 = Γ/{±1}. The first integral on the right hand side of (2.38) is to
be understood in the regularized sense. As in Lemma 2.13 it can be seen that
K
Fβ,m (τ, s; −n, 0) = O(e2π|m|/y ) (2.39)

for y → 0. This implies that the second integral converges absolutely if zv2 <
1
4|m| . By the usual unfolding argument it can be written in the form
√ Z
−πn2
 
2 X + dx dy
exp 2
K
Fβ,m (τ, s; −n, 0), ΘK (τ, w; nµ, 0) y (b −1)/2 ,
|zv | 2zv y y2
G n≥1
(2.40)
with G as in (2.30). By (2.39) the latter integral also converges absolutely for
1
zv2 < 4|m| . We find that (2.38) equals

√ Z∞ Z1
−πn2
 
2 X b+ 5
exp 2
K
Fβ,m (τ, s; −n, 0), ΘK (τ, w; nµ, 0) y 2 − 2 dx dy.
|zv | 2zv y
0 0 n≥1

We insert the Fourier expansions

X
ΘK (τ, w; nµ, 0) = e(iyq(λw ) − iyq(λw⊥ ) − (λ, nµ))eλ (q(λ)x)
λ∈K 0
X X X
K
Fβ,m (τ, s; −n, 0) = e(n(δ, z 0 ))c(δ, r; y, s)eγ (rx)
γ∈K 0 /K δ∈L00 /L r∈Z+q(δ)
p(δ)=γ

and carry out the integration over x to get

60 2 The regularized theta lift
√ Z∞ X
−πn2
  X
2 X
exp e(n(δ, z 0 ))c(δ, q(λ); y, s)
|zv | 2zv2 y
y=0 n≥1 λ∈K 0 δ∈L00 /L
p(δ)=λ+K
+
× e (iyq(λw ) − iyq(λw⊥ ) + (λ, nµ)) y b /2−5/2
dy. (2.41)

In the above sum we consider the λ = 0 term

√ Z∞ X
−πn2
 
2 X +
exp e(n(δ, z 0 ))c(δ, 0; y, s)y b /2−5/2
dy
|zv | 2zv2 y
y=0 n≥1 δ∈L00 /L
p(δ)=0+K

separately. A set of representatives for δ ∈ L00 /L with p(δ) = 0 + K is

given by `z/N where ` runs modulo N . We substitute in c(δ, 0; y, s) =
b(δ, 0, s)y 1−s−k/2 and obtain
√ Z∞
−πn2
 
2 X X + −
b(`z/N, 0, s) e(n`/N ) exp 2
y b /4+b /4−3/2−s dy.
|zv | 2zv y
` (N ) n≥1 y=0

Since the latter integral equals

s+1/2−b+ /4−b− /4
2zv2

Γ (s + 1/2 − b+ /4 − b− /4),
πn2

we find for the λ = 0 term:

s−b+ /4−b− /4
2zv2

2
√ Γ (s + 1/2 − b+ /4 − b− /4)
π π
X X + −
× b(`z/N, 0, s) e(`n/N )nb /2+b /2−1−2s .
` (N ) n≥1

Regarding the remaining sum over λ ∈ K 0 − 0 in (2.41) we observe the

following: Using the explicit formulas for the coefficients c(δ, r; y, s) (Theorem
1.9) and the asymptotic properties of the Whittaker functions, it can be seen
that the sum of the absolute values of the terms with λ 6= 0 is integrable
for σ > 1. Hence we may interchange integration and summation and obtain
the third term on the right hand side of (2.35). The resulting sum converges
normally in s. (If (β, z) ≡ 0 (mod N ) and if there is an λ ∈ p(β) + K with
q(λ) = m and λw = 0, then this λ gives a singular contribution in (2.41). How-
1
ever, this singularity cancels out with the singularity of √2|z ΦK
β,m (w, s).)
v|
t
u
3 The Fourier expansion of
the theta lift

We now determine the Fourier expansion of the regularized theta integral

more explicitly. We find that it consists of two different contributions. In
the case of signature (2, l) the first part gives rise to a certain “generalized
Borcherds product”, whereas the second part carries some cohomological in-
formation.

3.1 Lorentzian lattices

Throughout this section we assume that L is Lorentzian, i.e. has signature
(1, l − 1). Then Φβ,m (v, s) is holomorphic in s at 1 − k/2. Therefore the
function Φβ,m (v) (see Definition 2.10) is simply given by Φβ,m (v, 1 − k/2).
Moreover, Gr(L) is real hyperbolic space of dimension l − 1, and H(β, m)
is a union of hyperplanes of codimension 1. The set of points Gr(L)−H(β, m)
where Φβ,m (v) is real analytic is not connected. We call its components the
Weyl chambers of Gr(L) of index (β, m).
Observe that the smaller lattice K is negative definite. Hence the Grass-
mannian of K is a point, and the projection λw is 0 for any λ ∈ K 0 .
We now compute the Fourier expansion of ΦL β,m (v) more explicitly. We
L L
show that it can be written as the sum of two functions ψβ,m (v) and ξβ,m (v),
L L
where ξβ,m (v) is real analytic on the whole Gr(L) and ψβ,m (v) is the re-
L
striction of a continuous piecewise linear function on V . Both, ψβ,m (v) and
L
ξβ,m (v) are eigen functions of the hyperbolic Laplacian on Gr(L).
We need some basic properties of Bernoulli polynomials (cf. [E1] Vol. I
Chap. 1.13). For r ∈ N0 the r-th Bernoulli polynomial Br (x) is defined by
the generating series
∞
zexz X Br (x)
z
= zr (|z| < 2π).
e − 1 r=0 r!

The first Bernoulli polynomials are

B0 (x) = 1, B1 (x) = x − 1/2, B2 (x) = x2 − x + 1/6.

The identity
62 3 The Fourier expansion of the theta lift

Br0 (x) = rBr−1 (x)

implies that Br (x) is a polynomial of degree r. Let Br (x) be the 1-periodic
function with Br (x) = Br (x) for 0 ≤ x < 1. The Fourier expansion of Br (x)
can be easily determined (cf. [E1] Vol. I p. 37). For r ≥ 2 we have
X e(nx)
Br (x) = −r! . (3.1)
(2πin)r
n∈Z−{0}

If f (λ, β) is a function depending on λ ∈ K 0 and δ ∈ L0 /L, then we use

the abbreviation
 X X
X 
 f (λ, δ) + f (λ, −δ), if δ ∈ L00 /L,
f (λ, ±δ) = λ∈p(δ)+K λ∈−p(δ)+K
/ L00 /L.
if δ ∈

0,
λ∈±p(δ)+K

(3.2)
Recall that the projection p is only defined on L00 /L.
1
Proposition 3.1. Let v ∈ Gr(L) − H(β, m) with zv2 < 4|m| . Then

1 √ X
ΦL
β,m (v) = √ ΦK
β,m + 4 2π|zv | b(`z/N, 0)B2 (`/N )
2|zv | ` (N )
√ X
+ 4 2π|zv | B2 ((λ, µ) + (β, z 0 ))
λ∈p(β)+K
q(λ)=m
√ X X
+ 4 2(π/|zv |)−k b(δ, q(λ))|λ|1−k
λ∈K 0 −0 δ∈L00 /L
p(δ)=λ+K
X
× n−k−1 e(n(λ, µ) + n(δ, z 0 ))K1−k (2πn|λ|/|zv |), (3.3)
n≥1

where ΦK K
β,m denotes the constant Φβ,m (w, 1 − k/2), and b(γ, n) = b(γ, n, 1 −
L
k/2) denote the Fourier coefficients of the Poincaré series Fβ,m (τ, 1 − k/2).
The third term on the right hand side is to be interpreted as 0 if (β, z) 6≡
0 (mod N ). As usual Kν is the modified Bessel function of the third kind
(cf. [AbSt] Chap. 9).

Proof. We apply Theorem 2.15. Since L is Lorentzian, the first and the second
term on the right hand side of (2.35) are holomorphic in s near 1 − k/2. We
find
 3.1 Lorentzian lattices 63
√
1 2 2|zv | X X
ΦL
β,m (v) = √ ΦK
β,m + b(`z/N, 0) e(`n/N )n−2
2|zv | π
` (N ) n≥1
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))
|zv | 0 0
λ∈K −0 n≥1 δ∈L0 /L
p(δ)=λ+K
Z∞
−πn2
 
−2
× c(δ, q(λ); y, 1 − k/2)y exp + 2πyq(λ) dy. (3.4)
2zv2 y
0

According to Proposition 1.10 we have

c(γ, q(λ); y, 1 − k/2)

= (δβ,γ + δ−β,γ )δm,q(λ) e−2πmy + b(γ, q(λ))W1−k/2 (4πq(λ)y)

for λ ∈ K 0 − 0. Moreover, we know that W1−k/2 (y) = e−y/2 Γ (1 − k, |y|) for

y < 0 by (1.33). Hence the last term in (3.4) can be rewritten as
√ Z∞
−πn2
 
2 X X
0
e(n(λ, µ) + n(±β, z )) exp y −2 dy
|zv | 2zv2 y
λ∈±p(β)+K n≥1 0
q(λ)=m
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))b(δ, q(λ))
|zv |
λ∈K 0 −0 n≥1 δ∈L00 /L
p(δ)=λ+K
Z∞
−πn2
 
× Γ (1 − k, 4π|q(λ)|y) exp y −2 dy. (3.5)
2zv2 y
0

2z 2
The first integral in (3.5) equals πnv2 . If we replace the sum over n ≥ 1 by a
sum over n 6= 0 and substitute in identity (3.1) for B2 (x), we find
1 √ X
ΦL
β,m (v) = √ ΦK
β,m + 4 2π|zv | b(`z/N, 0)B2 (`/N )
2|zv | ` (N )
√ X
+ 4 2π|zv | B2 ((λ, µ) + (β, z 0 ))
λ∈p(β)+K
q(λ)=m
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))
|zv |
λ∈K 0 −0 n≥1 δ∈L00 /L
p(δ)=λ+K
Z∞
−πn2
 
× b(δ, q(λ)) Γ (1 − k, 4π|q(λ)|y) exp y −2 dy. (3.6)
2zv2 y
0
64 3 The Fourier expansion of the theta lift

We now compute the latter integral, which we abbreviate by

Z∞
I(A, B) = Γ (1 − k, Ay)e−B/y y −2 dy (A, B > 0).
0

The integral representation (1.32) of the incomplete Gamma function obvi-

ously implies
Γ (a, x)0 = −xa−1 e−x .
If we integrate by parts, we find
Z∞
−1
I(A, B) = A 1−k
B e−B/y−Ay y −k dy.
0

Using the standard integral representation

Z∞ √
e−Ay−B/y y ν−1 dy = 2(B/A)ν/2 Kν (2 AB ) (3.7)
0

for the K-Bessel function ([E2] Vol. I p. 313 (6.3.17)), we get

√
I(A, B) = 2A(1−k)/2 B −(1+k)2 K1−k (2 AB ).

If we put this into (3.6), we obtain the assertion. t

u

The set of norm 1 vectors in V = L ⊗ R has two components, one being

given by
V1 = {v1 ∈ V ; v12 = 1, (z, v1 ) > 0}.
We may identify Gr(L) with V1 via

V1 −→ Gr(L), v1 7→ Rv1 .

This is the “hyperboloid model” of hyperbolic space. In this identification we

have

zv = (z, v1 )v1 ,
|zv | = (z, v1 ).

The vector µ defined in (2.8) equals

(z, v1 )v1 z − (z, v1 )v1

µ = −z 0 + 2
−
2(z, v1 ) 2(z, v1 )2
v1 z
= − z0 − .
(z, v1 ) 2(z, v1 )2
 3.1 Lorentzian lattices 65

After a short calculation one finds that v1 can be expressed in terms of (z, v1 )
and the projection µK = µ − (µ, z 0 )z ∈ K ⊗ R by

v1 = (z, v1 )µK + (z, v1 )z 0 + bz. (3.8)

Here b ∈ R is uniquely determined by the condition v12 = 1. Conversely every

x0 > 0, µ ∈ K ⊗ R defines a unique v1 ∈ V1 as in (3.8). To be precise we may
identify

R>0 × (K ⊗ R) −→ V1 , (x0 , µ) 7→ µ/x0 + z 0 /x0 + bz = v1 , (3.9)

where b is again determined by v12 = 1. Thereby we have realized Gr(L) as

the subset
H = {(x0 , . . . , xl−2 ) ∈ Rl−1 ; x0 > 0}
of Rl−1 = R × (K ⊗ R). This is known as the “upper half space model” of
hyperbolic space.
Let % ∈ V with % = %K + %1 z 0 + %2 z, %K ∈ K ⊗ R, and %1 , %2 ∈ R. The
scalar product of (x0 , µ) ∈ H with % is given by
1  2

((x0 , µ), %) = 2(µ, %K ) + %1 z 0 + 2%2 − %1 µ2 + %1 x0 /2. (3.10)
2x0
Let us now consider the Weyl chambers of index (β, m), i.e. the connected
components of Gr(L)−H(β, m), in the hyperboloid model. For a Weyl cham-
ber W ⊂ V1 and λ ∈ L0 we write (λ, W ) > 0, if λ has positive inner product
with all elements in the interior of W .

Lemma 3.2. Let W ⊂ V1 be a Weyl chamber of index (β, m) and assume

that λ ∈ L0 with q(λ) ≥ 0 or q(λ) = m, λ + L = ±β. Suppose that (λ, v1 ) > 0
for a v1 ∈ W . Then (λ, W ) > 0.

Proof. Obviously λ 6= 0. Assume that there is a v2 ∈ W with (λ, v2 ) ≤ 0.

Then, since W is connected, there exists a v0 ∈ W satisfying (λ, v0 ) = 0,
i.e. λ ∈ v0⊥ . Because v0⊥ ⊂ V is a negative definite subspace, it follows that
q(λ) < 0. Without any restriction we may conclude that q(λ) = m and
λ + L = β. But then (λ, v0 ) = 0 implies v0 ∈ H(β, m), a contradiction. t
u

In the coordinates of V1 the Fourier expansion of ΦL

β,m is given by
66 3 The Fourier expansion of the theta lift

1 √ X
ΦL
β,m (v1 ) = √ ΦK
β,m + 4 2π(z, v1 ) b(`z/N, 0)B2 (`/N )
2(z, v1 ) ` (N )
√
 
X (λ, v1 )
+ 4 2π(z, v1 ) B2 + (β, z 0 )
(z, v1 )
λ∈p(β)+K
q(λ)=m
√ X X
+ 4 2(π/(z, v1 ))−k b(δ, q(λ))|λ|1−k
λ∈K 0 −0 δ∈L00 /L
p(δ)=λ+K
 
X (λ, v1 )
× n−k−1 e n + n(δ, z 0 ) K1−k (2πn|λ|/(z, v1 )).
(z, v1 )
n≥1
(3.11)
L L
Definition 3.3. Using the above notation we define two functions ξβ,m , ψβ,m
on Gr(L) by
√
ΦK
 
β,m 1 4 2π X
L
ξβ,m (v1 ) = √ − 2(z 0 , v1 ) + (λ, v1 )2
2 (z, v1 ) (z, v1 )
λ∈p(β)+K
q(λ)=m
√ X X
+ 4 2(π/(z, v1 ))−k b(δ, q(λ))|λ|1−k
λ∈K 0 −0 δ∈L00 /L
p(δ)=λ+K
 
X (λ, v1 )
× n−k−1 e n + n(δ, z 0 ) K1−k (2πn|λ|/(z, v1 )) (3.12)
(z, v1 )
n≥1

and
L
ψβ,m (v1 ) = ΦL L
β,m (v1 ) − ξβ,m (v1 ).

Note that this definition depends on the choice of the vectors z and z 0 . It
L L
is easily checked that ξβ,m = ξ−β,m .
By Theorem 2.14, Φβ,m is a real valued function and ΦK
L
β,m a real constant.
According to Lemma 1.13 the coefficients b(δ, q(λ)) are real numbers. Hence
L L
ξβ,m and ψβ,m are real-valued functions.
Using the boundedness of the coefficients b(δ, n) with n < 0 (cf. (2.22))
and the asymptotic property
r
π −y
Kν (y) ∼ e (y → ∞) (3.13)
2y
L
of the K-Bessel function, one can show that ξβ,m (v1 ) is real analytic on the
whole Gr(L).
L
Theorem 3.4. The function ψβ,m is the restriction of a continuous piece-
wise linear function on V . Its only singularities lie on H(β, m). For v1 ∈ V1
1
with (z, v1 )2 < 4|m| it equals
 3.1 Lorentzian lattices 67
√ √ X
L
ψβ,m (v1 ) = 2(z 0 , v1 )ΦK
β,m + 4 2π(z, v1 ) b(`z/N, 0)B2 (`/N )
` (N )
√ (λ, v1 )2
   
X (λ, v1 ) 0
+ 4 2π(z, v1 ) B2 + (β, z ) − .
(z, v1 ) (z, v1 )2
λ∈p(β)+K
q(λ)=m
(3.14)

Proof. Equality (3.14) immediately follows from (3.11) and the definition of
L
ψβ,m (v1 ). Using the fact that

B2 (x) = B2 (x − [x]) = x2 − (2[x] + 1)x + [x]2 + [x] + 1/6,

one finds that the quadratic terms in the sum on the right hand side of (3.14)
L 1
cancel out. Hence ψβ,m (v1 ) is piecewise linear on {v1 ∈ V1 ; (z, v1 )2 < 4|m| }.
L L
Since ξβ,m (v1 ) is real analytic on Gr(L), the only singularities of ψβ,m (v1 )
L
lie on H(β, m) by Theorem 2.12. Moreover, ψβ,m (v1 ) is continuous on Gr(L).
Let W1 and W2 be two different Weyl chambers of index (β, m) such that
the interior of W1 ∪W2 is connected. Then there is a λ ∈ β +L with q(λ) = m
and W1 ∩ W2 ⊂ λ⊥ . If Φ1 resp. Φ2 is a real analytic function on W1 ∪ W2 with
Φ1 |W1 = ΦL L
β,m |W1 resp. Φ2 |W2 = Φβ,m |W2 then Theorem 2.12 implies that

Φ1 (v1 ) − Φ2 (v1 ) = const · (λ, v1 ).

L L
The functions ψ1 = Φ1 − ξβ,m and ψ2 = Φ2 − ξβ,m are real analytic on
L
W1 ∪W2 , differ by a constant multiple of (λ, v1 ), and satisfy ψ1 |W1 = ψβ,m |W1
L L L
resp. ψ2 |W2 = ψβ,m |W2 . We may conclude: If ψβ,m |W1 is linear then ψβ,m |W2
is linear, too.
L
This fact, together with the first observation, implies that ψβ,m (v1 ) is
piecewise linear on the whole Grassmannian V1 . t
u

Definition 3.5. Let W be a Weyl chamber of index (β, m). Then we write
%β,m (W ) for the unique vector in V with the property
L
√
ψβ,m (v1 ) = 8 2π(v1 , %β,m (W ))

on W . We call %β,m (W ) the Weyl vector of W .

Later we will need the following theorem.

Theorem 3.6. (cp. [Bo2] Theorem 10.3.) Let

X
ξ L (v1 ) = L
cβ,m ξβ,m (v1 ) (cβ,m ∈ C)
β,m

L
be a finite linear combination of the ξβ,m . If ξ L (v1 ) is a rational function on
V1 , then it vanishes identically.
68 3 The Fourier expansion of the theta lift

Remark 3.7. The condition that ξ L is rational is obviously equivalent to say-

ing that X
cβ,m b(δ, q(λ)) = 0
β,m

for all λ ∈ K − 0 and δ ∈ L00 /L with p(δ) = λ + K.

0

Proof of Theorem 3.6. We use the following notation: Let f and g be functions
on V1 . Then we write
.
f = g,
if f − g is the restriction of a piecewise linear function on V .
.
We first prove that ξ L (v1 ) = 0. Let
X
ΦL (v1 ) = cβ,m ΦL
β,m (v1 ).
β,m

.
By Theorem 3.4 we have ΦL (v1 ) = ξ L (v1 ). Hence it suffices to show that
.
ΦL (v1 ) = 0.
The assumption that ξ L is rational implies that
" #
K
. 1 X Φβ,m 8π X
ΦL (v1 ) = √ cβ,m − 2(z 0 , v1 )ΦK
β,m + (λ, v1 )2 .
2 (z, v1 ) (z, v1 )
β,m λ∈p(β)+K
q(λ)=m
(3.15)
This formula for ΦL (v1 ) depends on the choice of a primitive norm 0 vector
z ∈ L. We compute ΦL (v1 ) using a different norm 0 vector and compare the
result with (3.15).
02
The vector z 0 − z2 z ∈ L ⊗ Q has norm 0. Let a be the unique positive
integer such that  
02
z̃ = a z 0 − z2 z

is a primitive element of L. Define z̃ 0 ∈ L0 , K̃ = L ∩ z̃ ⊥ ∩ z̃ 0⊥ , and p̃ analogous

to z 0 , K, and p (see section 2.1). Since ΦL (v1 ) is the sum of a rational function
and a piecewise linear function, we find by Proposition 3.1 and Theorem 3.4
that
" #
L . 1 X
K̃
X
2
Φ (v1 ) = √ cβ,m Φβ,m + 8π (λ, v1 ) . (3.16)
2(z̃, v1 ) β,m
λ∈p̃(β)+K̃
q(λ)=m

It is easily seen that (L⊗Q)∩ z̃ ⊥ = (K ⊗Q)⊕Qz̃. Thus K̃ ⊗Q ⊂ (K ⊗Q)⊕Qz̃.

Hence every λ ∈ K̃ 0 in the sum in (3.16) can be written as λ = λK + bz̃ with
λK ∈ K ⊗ Q and b ∈ Q. We have
(λ, v1 )2 . (λK , v1 )2
= ,
(z̃, v1 ) (z̃, v1 )
 3.1 Lorentzian lattices 69

and (3.16) can be rewritten in the form

" #
. 1 X X
ΦL (v1 ) = √ cβ,m ΦK̃
β,m + 8π (λK , v1 )2 . (3.17)
2(z̃, v1 ) β,m 0
λ∈K̃

We now compare (3.15) and (3.17) in the upper half space model H. If v1 is
represented by (x0 , µ) as in (3.9), we have

a(x20 − µ2 )
(z̃, v1 ) = ,
2x0
µ2 1
= − 2(z 0 , v1 ). (3.18)
x0 (z, v1 )

We find
 
1 X X
cβ,m µ2 ΦK
β,m + 8π (λ, µ)2 
x0 0
β,m λ∈K
 
. 2x0 X 8π X
= cβ,m ΦK̃
β,m + 2 (λK , µ)2  .
a(x20 − µ2 ) x0 0
β,m λ∈K̃

We multiply this equation by (x20 −µ2 ) and consider x0 as a complex variable.

Putting x0 = i|µ| we obtain
X X X
8π cβ,m (λK , µ)2 = µ2 cβ,m ΦK̃
β,m .
β,m λ∈K̃ 0 β,m

Inserting this into (3.17) we find

√ √
L . 2X K̃ 2(z, v1 ) X .
Φ (v1 ) = cβ,m Φβ,m = cβ,m ΦK̃
β,m = 0
ax0 a
β,m β,m

.
and thereby ξ L (v1 ) = 0.
In the upper half space model we have
 
1 X X
ξ L ((x0 , µ)) = √ cβ,m µ2 ΦK
β,m + 8π (λ, µ)2  . (3.19)
2x0 β,m λ∈K 0

.
Comparing coefficients in (3.19) and (3.10), we may infer that ξ L ((x0 , µ)) = 0
L
already implies ξ ((x0 , µ)) = 0. t
u
70 3 The Fourier expansion of the theta lift

3.1.1 The hyperbolic Laplacian

L L
We now consider the functions ψβ,m and ξβ,m in the upper half space model
H and determine the action of the of the hyperbolic Laplacian. For (x0 , µ) ∈
H = R>0 × (K ⊗ R) we have
√
L µ2 K 4 2π X
ξβ,m ((x0 , µ)) = √ Φβ,m + (λ, µ)2
2x0 x0
λ∈p(β)+K
q(λ)=m
√ X X
+ 4 2(πx0 )l/2−1 b(δ, q(λ))|λ|l/2
λ∈K 0 −0 δ∈L00 /L
p(δ)=λ+K
X
× n l/2−2
e (n(λ, µ) + n(δ, z 0 )) Kl/2 (2πn|λ|x0 ). (3.20)
n≥1

In the coordinates of H the hyperbolic Laplace operator has the form

(cp. [Ma2])
l−2 l−2
X ∂ 3−l ∂ X ∂2 ∂
∆L = −xl−1
0 x0 = −x20 2 + (l − 3)x0 .
j=0
∂xj ∂xj j=0
∂xj ∂x0

It is invariant under the action of SO(1, l − 1) on H.

L L
Theorem 3.8. The functions ψβ,m ((x0 , µ)) and ξβ,m ((x0 , µ)) are eigenfunc-
tions of the hyperbolic Laplacian with eigenvalue 1 − l.
L
Proof. First we consider ψβ,m . Since it is piecewise linear, it suffices to show
that
∆L ((x0 , µ), %) = (1 − l)((x0 , µ), %)
for any % ∈ V . Using (3.10) this can be verified by a straightforward compu-
tation.
L
To prove the statement for ξβ,m we compute the action of ∆L on the
different terms of the Fourier expansion (3.20). It is easily seen that

µ2 µ2
∆L = (1 − l) + 2(l − 2)x0 . (3.21)
x0 x0
Regarding the second term we find
√ √
4 2π X
2 4 2π X
∆L (λ, µ) = (1 − l) (λ, µ)2
x0 x0
λ∈p(β)+K λ∈p(β)+K
q(λ)=m q(λ)=m
√
− 16 2π|m|x0 · #{λ ∈ p(β) + K; q(λ) = m}. (3.22)

Moreover, a lengthy but trivial calculation shows that

 3.2 Lattices of signature (2, l) 71
l/2−1 l/2−1
∆L x0 Kl/2 (2πn|λ|x0 )e(n(λ, µ)) = (1 − l)x0 Kl/2 (2πn|λ|x0 )e(n(λ, µ))
(3.23)
for λ ∈ K 0 − 0. Here one has to use that Kν is a solution of the differential
equation
d2 f df
z2 2 + z − (z 2 + ν 2 )f = 0.
dz dz
If we put (3.21), (3.22), and (3.23) together, we see that
L L
√
∆L ξβ,m ((x0 , µ)) = (1 − l)ξβ,m ((x0 , µ)) + (l − 2) 2x0 ΦK β,m
√
− 16 2π|m|x0 · #{λ ∈ p(β) + K; q(λ) = m}. (3.24)

But Theorem 2.14 tells us that

8π|m|
ΦK
β,m = · #{λ ∈ p(β) + K; q(λ) = m}.
l/2 − 1
Thus the last two terms in (3.24) cancel and
L L
∆L ξβ,m ((x0 , µ)) = (1 − l)ξβ,m ((x0 , µ)).

This proves the assertion. t

u

3.2 Lattices of signature (2, l)

For the rest of this book let L denote a lattice of signature (2, l) with l ≥ 3.
Then the Grassmannian Gr(L) is a Hermitean symmetric space. Let z ∈ L
be a primitive norm 0 vector and choose a z 0 ∈ L0 with (z, z 0 ) = 1. Then
⊥
K = L ∩ z ⊥ ∩ z 0 is a Lorentzian lattice. We assume that K also contains a
primitive norm 0 vector. We set k = 1 − l/2 and κ = 1 + l/2.
If v ∈ Gr(L), then we write w for the corresponding element in the Grass-
mannian Gr(K) as in section 2.1 and w1 for its realization in the hyperboloid
model.
We shall determine the Fourier expansion of the function Φβ,m (v) (see
Definition 2.10) more explicitly. In the same way as in the previous section, we
show that it can be written as the sum of two real valued functions functions
L L L
ψβ,m (v) and ξβ,m (v). Here ξβ,m (v) is real analytic on the whole Gr(L), and
L
ψβ,m (v) is essentially the logarithm of the absolute value of a holomorphic
L
function on Gr(L) whose only zeros lie on H(β, m). These functions ψβ,m ,
L
ξβ,m generalize the functions ψm and ξm considered in [Br1, Br2].
L
In the Fourier expansion of ξβ,m (v) the following special function will
occur. For A, B ∈ R and κ ∈ R with κ > 1 we define
Z∞
2
Vκ (A, B) = Γ (κ − 1, A2 y)e−B y−1/y −3/2
y dy. (3.25)
0
72 3 The Fourier expansion of the theta lift

Let ε > 0. Then it can be easily checked that

√
A2 +B 2
Vκ (A, B) ε e−2(1−ε) . (3.26)
For κ ∈ N the function Vκ (A, B) can be expressed as a finite sum of K-
Bessel functions: If n ∈ N0 then by repeated integration by parts one finds
that Γ (1+n, x) = n!e−x en (x), where en (x) denotes the truncated exponential
series
n
X xr
en (x) =
r=0
r!
(also see [E1] Vol. II p. 136 (18)). Hence we have for integral κ:
Z∞
2
+B 2 )y−1/y −3/2
Vκ (A, B) = (κ − 2)! eκ−2 (A2 y)e−(A y dy
0
κ−2 Z∞
X A2r 2
+B 2 )y−1/y r−3/2
= (κ − 2)! e−(A y dy.
r=0
r!
0

By virtue of (3.7) we obtain

κ−2
X A2r 2 p
Vκ (A, B) = 2(κ − 2)! (A + B 2 )1/4−r/2 Kr−1/2 (2 A2 + B 2 ).
r=0
r!
1
Theorem 3.9. Let v ∈ Gr(L) − H(β, m) with zv2 < 2|m| . Then ΦL
β,m (v) is
equal to
1
√ ΦK
β,m (w, 1 − k/2) + Cβ,m + b(0, 0) log(zv )
2
2|zv |
X
−2 Log (1 − e((±β, z 0 ) + (λ, µ) + i|λw |/|zv |))
λ∈±p(β)+K
q(λ)=m
X X
−2 b(δ, q(λ)) Log (1 − e((δ, z 0 ) + (λ, µ) + i|λw |/|zv |))
0
λ∈K −0 δ∈L00 /L
q(λ)≥0 p(δ)=λ+K

2 X X X1
+√ b(δ, q(λ)) e(n(δ, z 0 ) + n(λ, µ))
π 0
n
λ∈K δ∈L00 /L n≥1
q(λ)<0 p(δ)=λ+K

× V2−k (πn|λ|/|zv |, πn|λw |/|zv |) ,

with

Cβ,m = −b(0, 0) (log(2π) + Γ 0 (1)) + b0 (0, 0)

X
−2 b(`z/N, 0) log |1 − e(`/N )|.
` (N )
`6≡0 (N )
 3.2 Lattices of signature (2, l) 73

Here b(γ, n) = b(γ, n, 1 − k/2) denote the Fourier coefficients of the Poincaré
L
series Fβ,m (τ, 1 − k/2) as in Proposition 1.10, and b0 (0, 0) means the deriva-
tive of b(0, 0, s) at s = 1 − k/2. The sum over λ in the second line is to be
interpreted as in (3.2).
1
Proof. We first assume zv2 < 4|m| and apply Theorem 2.15. The only term
on the right hand side of (2.35) that is not holomorphic in s near 1 − k/2 is
the second one
2
s−1/2−l/4 X X
h(s) = √ 2zv2 /π Γ (s − l/4) b(`z/N, 0, s) e(`n/N )nl/2−2s .
π
` (N ) n≥1

Here the summand with ` ≡ 0 (mod N ) gives the singular contribution

2
s−1/2−l/4
√ 2zv2 /π Γ (s − l/4)b(0, 0, s)ζ(2s − l/2), (3.27)
π

where ζ(s) denotes the Riemann zeta function. The remaining terms with
` 6≡ 0 (mod N ) converge for σ > l/4 and are therefore holomorphic. Their
value at s = 1 − k/2 = 1/2 + l/4 is
X
−2 b(`z/N, 0) log |1 − e(`/N )|. (3.28)
` (N )
`6≡0 (N )

To compute the constant term in the Laurent expansion of (3.27) at s =

1 − k/2 we note that
1
ζ(2s − l/2) = (s − 1/2 − l/4)−1 − Γ 0 (1) + . . . ,
2
2

s−1/2−l/4
= 1 + (s − 1/2 − l/4) log 2zv2 /π + . . . ,


2zv /π
√
Γ (s − l/4) = π + Γ 0 (1/2)(s − 1/2 − l/4) + . . . ,
b(0, 0, s) = b(0, 0) + b0 (0, 0)(s − 1/2 − l/4) + . . . .

(For the first expansion for instance see [E1] p. 34 (17); note that −Γ 0 (1) =
γ = Euler-Mascheroni constant.) We may infer that it is equal to

Γ 0 (1/2)
  2 
2zv
b(0, 0) log + √ − 2Γ 0 (1) + b0 (0, 0).
π π
√
From the duplication formula Γ (ν)Γ (ν√+ 1/2) = 21−2ν πΓ (2ν) for the
gamma function it follows that Γ 0 (1/2)/ π = Γ 0 (1) − 2 log 2. Thus the above
expression for the constant term in the Laurent expansion of (3.27) can be
simplified to

b(0, 0) log(zv2 ) − log(2π) − Γ 0 (1) + b0 (0, 0).

(3.29)
74 3 The Fourier expansion of the theta lift

Putting (3.28) and (3.29) together, we get

Cs=1−k/2 [h(s)] = b(0, 0) log(zv2 ) + Cβ,m .
Hence the function ΦL L
β,m (v) = Cs=1−k/2 [Φβ,m (v, s)] is given by

1
√ ΦK 2
β,m (w, 1 − k/2) + b(0, 0) log(zv ) + Cβ,m
2|zv |
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))
|zv | 0 0
λ∈K −0 n≥1 δ∈L0 /L
p(δ)=λ+K
Z∞
−πn2
 
× c(δ, q(λ); y, 1 − k/2) exp − 4πyq(λ w ) + 2πyq(λ) y −3/2 dy.
2zv2 y
0
(3.30)
We now insert the explicit formulas for the Fourier coefficients c(γ, q(λ); y, 1−
k/2) given in Proposition 1.10. We saw that c(γ, q(λ); y, 1 − k/2) equals
b(γ, q(λ))e−2πq(λ)y ,


 if q(λ) ≥ 0,

b(γ, q(λ))W1−k/2 (4πq(λ)y), if q(λ) < 0, q(λ) 6= m,


(δβ,γ + δ−β,γ )e−2πmy + b(γ, q(λ))W1− k (4πq(λ)y), if q(λ) = m.

2

Furthermore, according to (1.33) we have

W1−k/2 (y) = e−y/2 Γ (1 − k, |y|)
for y < 0. Hence the last summand in (3.30) can be rewritten in the form
√
2 X X X
e(n(λ, µ)) e(n(δ, z 0 ))b(δ, q(λ))
|zv | 0 0
λ∈K −0 n≥1 δ∈L0 /L
q(λ)≥0 p(δ)=λ+K
Z∞
−πn2
 
dy
× exp − 4πyq(λ w )
2zv2 y y 3/2
0
√ Z∞
−πn2
 
2 X X 0 dy
+ e(n(λ, µ) + n(±β, z )) exp 2y
− 4πyq(λ w ) 3/2
|zv | 0
2z v y
λ∈K n≥1 0
λ+K=±β
q(λ)=m
√
2 X X X
+ e(n(λ, µ)) e(n(δ, z 0 ))b(δ, q(λ))
|zv | 0
λ∈K n≥1 δ∈L00 /L
q(λ)<0 p(δ)=λ+K
Z∞
−πn2
 
dy
× Γ (1 − k, 4π|q(λ)|y) exp − 4πyq(λ w ) . (3.31)
2zv2 y y 3/2
0
 3.2 Lattices of signature (2, l) 75

Here the sum over λ in the second term is to be understood as in (3.2).

We now compute the integrals in the above expression. Using (3.7) and
the identity
r
2z
K−1/2 (z) = e−z (3.32)
π
(cf. [AbSt] p. 160 (10.2.17)), we obtain for A, B > 0:
Z ∞ √
√
e−Ay−B/y y −3/2 dy = πB −1/2 e−2 AB .
0

For instance by considering the limit A → 0 it can be shown that this formula
still holds for A = 0. We may conclude that
Z∞ √
−πn2
 
−3/2 2|zv | −2πn|λw |/|zv |
exp − 4πyq(λw ) y dy = e .
2zv2 y n
0

In the second integral in (3.31)

Z∞
−πn2
 
Γ (1 − k, 4π|q(λ)|y) exp − 4πyq(λw ) y −3/2 dy
2zv2 y
0

2zv2
we substitute u = πn2 y and find that it is equal to
√
2|z |
√ v V2−k (πn|λ|/|zv |, πn|λw |/|zv |) .
πn
We may rewrite (3.31) in the form
X X X1
2 b(δ, q(λ)) e(n(λ, µ) + n(δ, z 0 ))e−2πn|λw |/|zv |
0 0
n
λ∈K −0 δ∈L0 /L n≥1
q(λ)≥0 p(δ)=λ+K
X X1
+2 e(n(λ, µ) + n(±β, z 0 ))e−2πn|λw |/|zv |
n
λ∈±p(β)+K n≥1
q(λ)=m
2 X X X1
+√ b(δ, q(λ)) e(n(λ, µ) + n(δ, z 0 ))
π 0
n
λ∈K δ∈L00 /L n≥1
q(λ)<0 p(δ)=λ+K

× V2−k (πn|λ|/|zv |, πn|λw |/|zv |) . (3.33)

If we insert the power series expansion of the principal branch of the logarithm
1
and put everything together, we obtain the stated identity for zv2 < 4|m| .
Using the estimates (2.21) and (2.22) it can be seen that (3.33) even con-
1
verges and defines a real analytic function on zv2 < 2|m| . Hence the assertion
follows by real analytic continuation. t
u
76 3 The Fourier expansion of the theta lift

We now introduce a complex structure on the Grassmannian Gr(L). Let

V (C) = V ⊗ C be the complexification of V = L ⊗ R, and extend the bilinear
form (·, ·) on V to a C-bilinear form on V (C). Let P (V (C)) be the associated
projective space and denote the canonical projection by

V (C) → P (V (C)), ZL 7→ [ZL ].

In the zero-quadric

N = {[ZL ] ∈ P (V (C)); (ZL , ZL ) = 0}

we consider the open subset


K = [ZL ] ∈ N ; (ZL , Z̄L ) > 0 .

It is easily seen that K is a complex manifold that consists of 2 connected

components. The action of the orthogonal group O(V ) on V induces an action
on K. The connected component O+ (V ) of the identity (i.e. the subgroup of
elements with positive spinor norm) preserves the components of K, whereas
O(V )\ O+ (V ) interchanges them. We choose one fixed component of K and
denote it by K+ .
If we write ZL = XL + iYL with XL , YL ∈ V , then [ZL ] ∈ K is equivalent
to
XL ⊥ YL , and XL2 = YL2 > 0. (3.34)
But this means that XL and YL span a two dimensional positive definite
subspace of V and thereby define an element of Gr(L). Conversely for a
given v ∈ Gr(L) we may choose an orthogonal base XL , YL as in (3.34) and
obtain a unique [ZL ] = [XL + iYL ] in K+ . (Then [Z̄L ] ∈ K corresponds to
the same v ∈ Gr(L).) We get a bijection between Gr(L) and K+ and hereby
a complex structure on Gr(L).

We may realize K+ as a tube domain in the following way. Suppose that

z ∈ L is a primitive norm 0 vector and z 0 ∈ L0 with (z, z 0 ) = 1. As in section
⊥
2.1, define a sub-lattice K by K = L ∩ z ⊥ ∩ z 0 . Then K is Lorentzian and

V = (K ⊗ R) ⊕ Rz 0 ⊕ Rz.

If ZL ∈ L ⊗ C and ZL = Z + az 0 + bz with Z ∈ K ⊗ C and a, b ∈ C, then we

briefly write ZL = (Z, a, b).
We have a map from the set

{Z = X + iY ∈ K ⊗ C; Y 2 > 0} (3.35)

of vectors in K ⊗ C with positive imaginary part to K given by

Z 7→ [ZL ] = [(Z, 1, −q(Z) − q(z 0 ))]. (3.36)

 3.2 Lattices of signature (2, l) 77

Conversely assume that [ZL ] ∈ K, ZL = XL + iYL . From the fact that XL ,

YL span a two dimensional positive definite subspace of V it follows that
(ZL , z) 6= 0. Thus [ZL ] has a unique representative of the form (Z, 1, b).
The condition q(ZL ) = 0 implies b = −q(Z) − q(z 0 ), and thereby [ZL ] =
[(Z, 1, −q(Z) − q(z 0 ))]. Moreover, from (ZL , Z̄L ) > 0 one easily deduces Y 2 >
0. We may infer that the map (3.36) is biholomorphic.
The set (3.35) has two components. Let Hl be the component which is
mapped to K+ under (3.36). Then Hl is a realization of K+ as a tube domain
and can be viewed as a generalized upper half plane.
The cone iC = Hl ∩ i(K ⊗ R) is given by one of the two components of
the set {Y ∈ K ⊗ R; Y 2 > 0} of positive norm vectors of K ⊗ R and Hl =
K ⊗ R + iC. Without any restriction we may assume that {Y /|Y |; Y ∈ C}
coincides with the hyperboloid model of Gr(K) that we used in section 3.1.
Let Z = X + iY be a point of Hl and v = RXL + RYL the corresponding
element of the Grassmannian. We have

ZL = XL + iYL = (Z, 1, −q(Z) − q(z 0 )),

XL = (X, 1, q(Y ) − q(X) − q(z 0 )),
YL = (Y, 0, −(X, Y )).

Using XL2 = YL2 = Y 2 we find

zv = (z, XL )XL /XL2 + (z, YL )YL /YL2 = XL /Y 2 ,

zv2 = 1/Y 2 .

The vector
zv z ⊥ zv z
µ = −z 0 + 2
+ v2 = −z 0 + 2 − 2
2zv 2zv⊥ zv 2zv
is given by
µ = (X, 0, −q(X) − q(z 0 ))
and its orthogonal projection µK by µK = µ − (µ, z 0 )z = X. The subspace
w ⊂ V , i.e. the orthogonal complement of zv in v, is equal to RYL . Recall
from section 2.1 that we have identified w with its orthogonal projection
wK ⊂ K ⊗ R. Obviously wK = RY , w1 = Y /|Y |, and thereby

λw = (λ, Y )Y /Y 2 ,
|λw | = |(λ, Y )||zv |.

The intersection
Γ (L) = O+ (V ) ∩ Od (L) (3.37)
of the discriminant kernel Od (L) with O (V ) is a discrete subgroup of O+ (V ).
+

We will be interested in its action on K+ and Hl .

Let λ ∈ L0 be a vector of negative norm. If we write λ = λK + az 0 + bz
with λK ∈ K 0 , a ∈ Z, and b ∈ Q, then
78 3 The Fourier expansion of the theta lift

λ⊥ = {Z ∈ Hl ; aq(Z) − (Z, λK ) − aq(z 0 ) − b = 0}

in the coordinates of Hl . This set defines a prime divisor on Hl . The sum
X
λ⊥
λ∈β+L
q(λ)=m

is a Γ (L)-invariant divisor on Hl with support H(β, m). Following Borcherds

we call it the Heegner divisor of discriminant (β, m). The proof of the next
lemma will be left to the reader.
Lemma 3.10. Let λ ∈ L0 with q(λ) = m < 0 and (λ, z) = a 6= 0. Assume
that Z ∈ λ⊥ ⊂ Hl . Then q(Y ) < |m|/a2 .
Assume that (β, z) ≡ 0 (mod N ). Let W be a Weyl chamber of Gr(K) of
index (p(β), m). Then we also call the subset
{Z = X + iY ∈ Hl ; Y /|Y | ∈ W } ⊂ Hl
a Weyl chamber of index (β, m) and denote it by W . For the corresponding
Weyl vector in K ⊗ R we write %β,m (W ).
If (β, z) 6≡ 0 (mod N ), then by a Weyl chamber of index (β, m) we simply
mean Hl and put %β,m (W ) = 0. (In this case there is no contribution to ΦL
β,m
from the smaller lattice K.)

We may consider ΦL β,m as a function on Hl . By construction it is invariant

under the action of Γ (L).
Let W ⊂ Hl be a Weyl chamber of index (β, m). Suppose that Z ∈ W
with q(Y ) > |m|. Then Lemma 3.10 implies that Z ∈ / H(β, m). According to
Theorem 3.9 the Fourier expansion of ΦL β,m (Z) is given by
|Y | K
√ Φβ,m (Y /|Y |, 1 − k/2) + Cβ,m − b(0, 0) log(Y 2 )
2
X
−4 log |1 − e((±β, z 0 ) + (λ, Z))|
λ∈±p(β)+K
(λ,W )>0
q(λ)=m
X X
−4 b(δ, q(λ)) log |1 − e((δ, z 0 ) + (λ, Z))|
0
λ∈K δ∈L00 /L
(λ,W )>0 p(δ)=λ+K
q(λ)≥0
2 X X X1
+√ b(δ, q(λ)) e(n(δ, z 0 ) + n(λ, X))
π 0
n
λ∈K δ∈L00 /L n≥1
q(λ)<0 p(δ)=λ+K

× V2−k (πn|λ||Y |, πn(λ, Y )) .

Here we have used b(γ, n) = b(−γ, n) and the fact that |λw | = (λ, Y )|zv | for
any λ ∈ K 0 with (λ, W ) > 0.
 3.2 Lattices of signature (2, l) 79

L
Definition 3.11. We define a function ξβ,m : Hl → R by

L |Y | K 2 X X
ξβ,m (Z) = √ ξβ,m (Y /|Y |)−b(0, 0) log(Y 2 )+ √ b(δ, q(λ))
2 π
λ∈K 0 δ∈L00 /L
q(λ)<0 p(δ)=λ+K
X1
× e(n(δ, z 0 ) + n(λ, X))V2−k (πn|λ||Y |, πn(λ, Y ))
n
n≥1

L
and a function ψβ,m : Hl − H(β, m) → R by
L
ψβ,m (Z) = ΦL L
β,m (Z) − ξβ,m (Z).

K
Here ξβ,m is given by
(
K
0, if (β, z) 6≡ 0 (mod N ),
ξβ,m =
K
ξp(β),m , if (β, z) ≡ 0 (mod N ) (see Def. 3.3).

Note that this definition depends on the choice of z and z 0 . It is easily

L L L L
checked that ξβ,m = ξ−β,m and ψβ,m = ψ−β,m
Using the boundedness of the coefficients b(δ, n) with n < 0 (cf. (2.22))
and the asymptotic property (3.26) of the function V2−k (A, B), one can show
L
that ξβ,m is real analytic on the whole Hl .
According to Lemma 1.13 the coefficients b(δ, n) are real numbers. We
L L
may infer that ξβ,m and ψβ,m are real-valued functions.

Definition 3.12. Let D be a domain in Cl and f : D → R a twice continu-

ously differentiable function. Then f is called pluriharmonic, if

∂2
f =0
∂zi ∂ z̄j

for all 1 ≤ i, j ≤ l.

Lemma 3.13. Let D ⊂ Cl be a simply connected domain and f : D → R

a twice continuously differentiable function. Then f is pluriharmonic if and
only if there is a holomorphic function h : D → C with f = <(h).

Proof. See [GR] chapter IX section C. t

u

Now let W ⊂ Hl be a fixed Weyl chamber of index (β, m) and %β,m (W )

the corresponding Weyl vector. For Z ∈ W with q(Y ) > |m| we may write
L
the Fourier expansion of ψβ,m (Z) in the form
80 3 The Fourier expansion of the theta lift
X
Cβ,m + 8π(%β,m (W ), Y ) − 4 log |1 − e((±β, z 0 ) + (λ, Z))|
λ∈±p(β)+K
(λ,W )>0
q(λ)=m
X X
−4 b(δ, q(λ)) log |1 − e((δ, z 0 ) + (λ, Z))| . (3.38)
0
λ∈K δ∈L00 /L
(λ,W )>0 p(δ)=λ+K
q(λ)≥0

Definition 3.14. For Z ∈ Hl with q(Y ) > |m| we define

Y
1 − e((±β, z 0 ) + (λ, Z))



Ψβ,m (Z) = e (%β,m (W ), Z)
λ∈±p(β)+K
(λ,W )>0
q(λ)=m
Y Y
b(δ,q(λ))
× 1 − e((δ, z 0 ) + (λ, Z)) . (3.39)
λ∈K 0 δ∈L00 /L
(λ,W )>0 p(δ)=λ+K
q(λ)≥0

Lemma 3.15. The infinite product (3.39) converges normally for Z ∈ Hl

with q(Y ) > |m|.
Proof. It suffices to show that the two series
X X X
e−2π(λ,Y ) and b(δ, q(λ))e−2π(λ,Y )
0
λ∈±p(β)+K λ∈K δ∈L00 /L
(λ,W )>0 (λ,Y )>0 p(δ)=λ+K
q(λ)=m q(λ)≥0

converge normally. The convergence of the second series for q(Y ) > |m| fol-
lows from the asymptotic behavior (2.21) of the coefficients b(γ, n).
It is easily seen that the first series converges normally for Z ∈ W . Now
suppose that U is an arbitrary open subset of Hl with compact closure U ⊂
Hl . Then there are only finitely many λ ∈ K 0 with q(λ) = m and λ + K =
±p(β) such that (λ, W ) > 0 and (λ, Y ) ≤ 0 for some Z = X + iY ∈ U .
Using this fact, one immediately obtains the convergence of the first series
on U . t
u
By Lemma 3.15 the function Ψβ,m (Z) is holomorphic on {Z ∈ Hl ; q(Y ) >
|m|} and satisfies
1 L
log |Ψβ,m (Z)| = − (ψβ,m (Z) − Cβ,m ) (3.40)
4
on the complement of H(β, m).
Theorem 3.16. The function Ψβ,m (Z) has a holomorphic continuation to
Hl , and (3.40) holds on Hl − H(β, m). Let U ⊂ Hl be an open subset with
compact closure U ⊂ Hl and denote by S(β, m, U ) the finite set
 3.2 Lattices of signature (2, l) 81

S(β, m, U ) = {λ ∈ β + L; q(λ) = m, ∃Z ∈ U with (ZL , λ) = 0} .

Then Y
Ψβ,m (Z) (λ, ZL )−1 (3.41)
λ∈S(β,m,U )

is a holomorphic function without any zeros on U .

Proof. It suffices to show that Ψβ,m (Z) can be continued holomorphically to

any simply connected open set U with compact closure U ⊂ Hl and nonempty
intersection U ∩ {Z ∈ Hl ; q(Y ) > |m|}, and that (3.40), (3.41) hold on U .
According to Theorem 2.12 we may consider
X
L
ψβ,m (Z) + 4 log |(λ, ZL )| (3.42)
λ∈S(β,m,U )

as a real analytic function on U . Moreover, the Fourier expansion (3.38) of

L
ψβ,m on {Z ∈ W ; q(Y ) > |m|} implies that (3.42) is even pluriharmonic (by
real analytic continuation on U ). By Lemma 3.13 there exists a holomorphic
function f : U → C with
X
L
ψβ,m (Z) + 4 log |(λ, ZL )| = <(f (Z)).
λ∈S(β,m,U )

On the non-empty intersection of the open sets U and {Z ∈ Hl ; q(Y ) > |m|}
we therefore have
   
Y 1
< Log Ψβ,m (Z) (λ, ZL )−1  − Cβ,m /4 = − <(f (Z)).
4
λ∈S(β,m,U )

But now − 41 f can only differ by an additive constant from the expression in
brackets on the left-hand side. We may assume that this constant equals zero
and find Y
Ψβ,m (Z) (λ, ZL )−1 = eCβ,m /4 e−f (Z)/4 .
λ∈S(β,m,U )

Since e−f (Z)/4 is a holomorphic function without any zeros on U , we obtain

the assertion. t
u

If 2β = 0 in L0 /L, then all zeros of Ψβ,m (Z) have order 2, because λ ∈

S(β, m, U ) implies −λ ∈ S(β, m, U ). If 2β 6= 0 in L0 /L, then the zeros of
Ψβ,m have order 1, and Ψβ,m = Ψ−β,m .
According to (3.40) the function Ψβ,m is independent of the choice of W
up to multiplication by a constant of absolute value 1.
82 3 The Fourier expansion of the theta lift

3.3 Modular forms on orthogonal groups

In this section we review a few basic facts on modular forms for orthogonal
groups. The theory can be developed similarly as in [Fr1], [Fr2].
Recall the realizations K+ and Hl of the Hermitean symmetric space
Gr(L) introduced in the previous section. Let

e + = {W ∈ V (C) − {0};
K [W ] ∈ K+ } ⊂ V (C) (3.43)

be the cone over K+ ⊂ P (V (C)). The map

Z 7→ ZL = (Z, 1, −q(Z) − q(z 0 )) (3.44)

defines an embedding of Hl into K e + , and the assignment Z 7→ [ZL ] ∈

P (V (C)) induces an isomorphism Hl → K+ . The linear action of the or-
thogonal group O+ (V ) on K+ induces an action on Hl such that the diagram

[W ]7→[σW ]
KO + / K+ (3.45)
O
Z7→[ZL ]

Hl
Z7→σZ / Hl

commutes for all σ ∈ O+ (V ). It is easily verified that σZ is the unique

element of Hl with the property

(σ(ZL ), z)(σZ)L = σ(ZL ). (3.46)

The function
j(σ, Z) = (σ(ZL ), z)
on O+ (V ) × Hl is an automorphy factor for O+ (V ), i.e. it satisfies the cocycle
relation
j(σ1 σ2 , Z) = j(σ1 , σ2 Z)j(σ2 , Z)
and does not vanish on Hl .
Let r be a rational number. If σ ∈ O+ (V ) and Z ∈ Hl , we define

j(σ, Z)r := er Log j(σ,Z) ,

where Log j(σ, Z) denotes a fixed holomorphic logarithm of j(σ, Z). There
exists a map wr from O+ (V ) × O+ (V ) to the set of roots of unity (of order
bounded by the denominator of r) such that

j(σ1 σ2 , Z)r = wr (σ1 , σ2 )j(σ1 , σ2 Z)r j(σ2 , Z)r .

This map obviously only depends on r modulo Z.

 3.3 Modular forms on orthogonal groups 83

Definition 3.17. Let Γ ≤ O+ (V ) be a subgroup and r ∈ Q as above. By a

multiplier system of weight r for Γ we mean a map

χ : Γ → S 1 = {t ∈ C; |t| = 1}

satisfying

χ(σ1 σ2 ) = wr (σ1 , σ2 )χ(σ1 )χ(σ2 ).

If r ∈ Z, then χ is actually a character1 of Γ . If χ is a multiplier system of

weight r for Γ , then χ(σ)j(σ, Z)r is a cocycle of Γ .
We now define modular forms for congruence subgroups of O+ (V ).
Definition 3.18. Let Γ ≤ Γ (L) be a subgroup of finite index and χ a mul-
tiplier system for Γ of weight r ∈ Q. A meromorphic function F on Hl is
called a meromorphic modular form of weight r and multiplier system χ with
respect to Γ , if
F (σZ) = χ(σ)j(σ, Z)r F (Z) (3.47)
for all σ ∈ Γ . If F is even holomorphic on Hl then it is called a holomorphic
modular form. (Since l ≥ 3, then the Koecher principle ensures that F is also
holomorphic on the Satake boundary.)
Remark 3.19. Since l ≥ 3, the Lie group O+ (V ) has no almost simple factor
of real rank 1. This implies that the factor group of Γ as in Definition 3.18
modulo its commutator subgroup is finite (see [Mar] Proposition 6.19 on
p. 333). Thus any multiplier system of Γ has finite order.
Sometimes it is convenient to consider modular forms as homogeneous
functions on Ke + . Here we temporarily have to assume that r ∈ Z. (Otherwise
one would have to work with covers of K+ .)
If F is a meromorphic modular form on Hl of weight r, then we may
define a meromorphic function F̃ on K e + by

F̃ (tZL ) = t−r F (Z) (t ∈ C − 0).

It is easily seen that:

i) F̃ is homogeneous of degree −r, i.e. F̃ (tZL ) = t−r F̃ (ZL ) for any t ∈ C−0;
ii) F̃ is invariant under Γ , i.e. F̃ (σZL ) = χ(σ)F̃ (ZL ) for any σ ∈ Γ .
Conversely, if G is any meromorphic function on K e + satisfying (i) and (ii),
then the function F (Z) = G(ZL ) is a meromorphic modular form of weight r
on Hl in the sense of Definition 3.18. Thus modular forms can also be viewed
as meromorphic functions on K e + satisfying (i) and (ii).

Let u ∈ V = L ⊗ R be an isotropic vector and v ∈ V be orthogonal to u.

The Eichler transformation E(u, v), defined by
1
Here multiplier systems and characters are always unitary.
84 3 The Fourier expansion of the theta lift

E(u, v)(a) = a − (a, u)v + (a, v)u − q(v)(a, u)u (3.48)

for a ∈ V , is an element of O+ (V ). If u, v belong to the lattice L, then E(u, v)

lies in the discriminant kernel Γ (L).
For any k ∈ K the special Eichler transformation E(z, k) ∈ Γ (L) acts on
Hl as the translation Z 7→ Z − k and the corresponding automorphy factor
equals 1. This implies that any holomorphic modular form F of weight r ∈ Q
for the group Γ (L) is periodic with period lattice K in the following sense:
There exists a vector % ∈ K ⊗ Q (which is unique modulo K 0 ) such that

F (Z + k) = e((%, k))F (Z)

for all k ∈ K. Thus F has a Fourier expansion of the form

X
F (Z) = a(λ)e((λ, Z)).
λ∈%+K 0

If F has integral weight and trivial multiplier system, then % = 0. The

Koecher principle tells us that a(λ) 6= 0, only if λ lies in the closure of
the positive cone C ⊂ K ⊗ R (see page 77), in particular q(λ) ≥ 0. A precise
version will be stated later in Proposition 4.15.
y
The following lemma generalizes the elementary formula =(M τ ) = |cτ +d|2


for τ ∈ H and M = ac db ∈ SL2 (R).

Lemma 3.20. If σ ∈ O+ (V ) and Z = X + iY ∈ Hl , then

q(Y )
q(=(σZ)) = .
|j(σ, Z)|2

Proof. We use the formula q(Y ) = 41 (ZL , Z L ) and (3.46). One has
 
1 σ(ZL ) σ(Z L )
q(=(σZ)) = ,
4 (σZL , z) (σZ L , z)
1

= 2
σ(ZL ), σ(Z L )
4|(σZL , z)|
q(Y )
= .
|j(σ, Z)|2
t
u

Lemma 3.21. (cp. [Bo2] Lemma 13.1.) Let r ∈ Q and Γ ≤ Γ (L) be sub-
group of finite index. Suppose that Ψ is a meromorphic function on Hl for
which |Ψ (Z)|q(Y )r/2 is invariant under Γ . Then there exists a multiplier sys-
tem χ of weight r for Γ such that Ψ is a meromorphic modular form of weight
r and multiplier system χ with respect to Γ .
 3.4 Borcherds products 85

Proof. Let σ ∈ Γ . By assumption we have

|Ψ (σZ)|q(=(σZ))r/2 = |Ψ (Z)|q(Y )r/2 .

By Lemma 3.20 this means that the function |Ψ|Ψ(σZ)|

(Z)| |j(σ, Z)|
−r
is constant
with value 1. Hence, according to the maximum modulus principle, there
exists a constant χ(σ) of absolute value 1 such that

Ψ (σZ)
j(σ, Z)−r = χ(σ).
Ψ (Z)

The function χ : Γ → C is a multiplier system of weight r. t

u

3.4 Borcherds products

Theorem 3.16 states that Ψβ,m (Z) is a holomorphic function on Hl with

divisor H(β, m). Observe that Ψβ,m (Z) is not necessarily automorphic; by
construction we only know that
L
|Ψβ,m (Z)|e−ξβ,m (Z)/4

is invariant under Γ (L). However, taking suitable finite products of the Ψβ,m
L
one can attain that the main parts of the ξβ,m cancel out. Thereby one finds
a new explanation of [Bo2] Theorem 13.3 and [Bo3] from a cohomological
point of view.
Let f : H → C[L0 /L] be a nearly holomorphic modular form for Mp2 (Z)
of weight k = 1 − l/2 with principal part
X X
c(γ, n)eγ (nτ ).
γ∈L0 /L n∈Z+q(γ)
n<0

Then we call the components of

[ [
Gr(K) − H(p(γ), n)
γ∈L00 /L n∈Z+q(γ)
n<0
c(γ,n)6=0

the Weyl chambers of Gr(K) with respect to f . Let W be such a Weyl cham-
ber. For every γ ∈ L00 /L, n ∈ Z + q(γ) with n < 0 and c(γ, n) 6= 0 there is a
Weyl chamber Wγ,n of Gr(K) of index (p(γ), n) (as in section 3.1) such that
W ⊂ Wγ,n . Then \ \
W = Wγ,n .
γ∈L00 /L n∈Z+q(γ)
n<0
c(γ,n)6=0
86 3 The Fourier expansion of the theta lift

We define the Weyl vector %f (W ) attached to W and f by

1 X X
%f (W ) = c(γ, n)%p(γ),n (Wγ,n ).
2
γ∈L00 /L n∈Z+q(γ)
n<0

As before, for λ ∈ K 0 we write (λ, W ) > 0, if λ has positive inner product

with all elements in the interior or W . We will also call the subset {Z =
X + iY ∈ Hl ; Y /|Y | ∈ W } ⊂ Hl a Weyl chamber with respect to f .

Theorem 3.22 (Borcherds). Let L be an even lattice of signature (2, l)

with l ≥ 3, and z ∈ L a primitive isotropic vector. Let z 0 ∈ L0 , K = L ∩ z ⊥ ∩
⊥
z 0 , and the projection p be defined as in section 2.1. Moreover, assume that
K also contains an isotropic vector.
Let f be a nearly holomorphic modular form of weight k = 1 − l/2 whose
Fourier coefficients c(γ, n) are integral for n < 0. Then
Y Y
Ψ (Z) = Ψβ,m (Z)c(β,m)/2
β∈L0 /L m∈Z+q(β)
m<0

is a meromorphic function on Hl with the following properties:

i) It is a meromorphic modular form of (rational) weight c(0, 0)/2 for the
orthogonal group Γ (L) with some multiplier system χ of finite order. If
c(0, 0) ∈ 2Z, then χ is a character.
ii) The divisor of Ψ (Z) on Hl is given by
1 X X
(Ψ ) = c(β, m)H(β, m).
2 0
β∈L /L m∈Z+q(β)
m<0

(The multiplicities of H(β, m) are 2, if 2β = 0 in L0 /L, and 1, if 2β 6= 0

in L0 /L. Note that c(β, m) = c(−β, m) and H(β, m) = H(−β, m).)
iii) Let W ⊂ Hl be a Weyl chamber with respect to f and m0 = min{n ∈
Q; c(γ, n) 6= 0}. On the set of Z ∈ Hl , which satisfy q(Y ) > |m0 |, and
which belong to the complement of the set of poles of Ψ (Z), the function
Ψ (Z) has the normally convergent Borcherds product expansion
Y Y
c(δ,q(λ))
1 − e((δ, z 0 ) + (λ, Z))


Ψ (Z) = Ce (%f (W ), Z) .
λ∈K 0 δ∈L00 /L
(λ,W )>0 p(δ)=λ+K

Here C is a constant of absolute value 1, and %f (W ) ∈ K ⊗ R denotes

the Weyl vector attached to W and f . (Usually %f (W ) can be computed
explicitly using Theorem 3.4.)
 3.4 Borcherds products 87

Proof. Throughout the proof we write bβ,m (γ, n) instead of b(γ, n, 1 − k/2)
for the (γ, n)-th Fourier coefficient of the Poincaré series Fβ,m (τ, 1 − k/2) to
emphasize the dependence on (β, m). Note that
1 X X
f (τ ) = c(β, m)Fβ,m (τ, 1 − k/2)
2 0
β∈L /L m∈Z+q(γ)
m<0

by Proposition 1.12. Thereby the Fourier coefficients c(γ, n) of f with n ≥ 0

are given by
1 X X
c(γ, n) = c(β, m)bβ,m (γ, n). (3.49)
2 0β∈L /L m∈Z+q(γ)
m<0

In particular, by Lemma 1.13 the coefficients c(γ, n) are real numbers. Define
two functions
1 X X
Φ(Z) = c(β, m)ΦLβ,m (Z),
2 0 β∈L /L m∈Z+q(γ)
m<0
1 X X
L
ξ(Z) = c(β, m)ξβ,m (Z).
2 0β∈L /L m∈Z+q(γ)
m<0

i) First, we show that the existence of the nearly holomorphic modular

form f implies that the main terms in ξ(Z) cancel. By virtue of Proposition
L
1.16 we may replace the coefficients b(δ, q(λ)) in the Fourier expansion of ξβ,m
(see Definition 3.11) by
1
− pδ,−q(λ) (β, −m),
Γ (1 − k)
where pδ,−n (β, −m) denotes the (β, −m)-th coefficient of the Poincaré series
Pδ,−n ∈ Sκ,L . Hence the Fourier expansion of ξ(Z) can be written in the form
 
1 X |Y | K
ξ(Z) = c(β, m) √ ξβ,m (Y /|Y |) − bβ,m (0, 0) log(Y 2 )
2 2
β∈L0 /L
m∈Z+q(γ)
m<0
1 X X X
−√ c(β, m)pδ,−q(λ) (β, −m)
πΓ (1 − k) 0
λ∈K δ∈L00 /L β∈L0 /L
q(λ)<0 p(δ)=λ+K m∈Z+q(γ)
m<0
X1
× e(n(δ, z 0 ) + n(λ, X))V2−k (πn|λ||Y |, πn(λ, Y )) .
n
n≥1

But Theorem 1.17 implies that the sums

88 3 The Fourier expansion of the theta lift
X
c(β, m)pδ,−q(λ) (β, −m)
β,m

vanish. Thus ξ(Z) can be simplified:

1 X |Y | X
ξ(Z) = − log(Y 2 ) c(β, m)bβ,m (0, 0) + √ K
c(β, m)ξβ,m (Y /|Y |).
2 2 2
β,m β,m

K
If we apply the same argument to the Fourier expansion of the ξβ,m , we find
K
P
that the function β,m c(β, m)ξβ,m is rational. Hence, according to Theorem
3.6, it vanishes identically. Using (3.49) we finally obtain

ξ(Z) = −c(0, 0) log(Y 2 )

and hereby
1 X X
Φ(Z) = −c(0, 0) log(Y 2 ) + L
c(β, m)ψβ,m (Z). (3.50)
2 0
β∈L /L m∈Z+q(γ)
m<0

By Theorem 3.16 there is a real constant C such that

 
 1 X X
L

|Ψ (Z)| = C exp 
− 8 c(β, m)ψβ,m (Z)
.
0 β∈L /L m∈Z+q(γ)
m<0

Inserting (3.50) we get

|Ψ (Z)|q(Y )c(0,0)/4 = C 0 e−Φ(Z)/4 .

Since Φ(Z) is invariant under Γ (L), the function |Ψ (Z)|q(Y )c(0,0)/4 is invari-
ant, too. Using Lemma 3.21 we may infer that Ψ (Z) is a modular form of
weight c(0,0)
2 with some multiplier system χ for Γ (L).
ii) This immediately follows from Theorem 3.16.
iii) It is a consequence of Theorem 3.16 and equation (3.49) that

log |Ψ (Z)| = −2π(%f (W ), Y )

X X
+ c(δ, q(λ)) log |1 − e((δ, z 0 ) + (λ, Z))|.
λ∈K 0 δ∈L00 /L
(λ,W )>0 p(δ)=λ+K

This implies the last assertion. t

u

Remark 3.23. According to Proposition 1.16 the coefficient c(0, 0) is equal to

the linear combination
 3.4 Borcherds products 89

1 X X
− c(β, m)q(β, −m)
2 0
β∈L /L m∈Z+q(γ)
m<0

of the coefficients q(γ, n) of the Eisenstein series E ∈ Mκ,L . In particular, the

assumption that the c(γ, n) are integral for n < 0 implies that c(0, 0) ∈ Q by
Proposition 1.7.
The fact that χ has finite order implies that the Weyl vector %f (W ) lies
in K ⊗ Q. This can also be proved directly for the individual Weyl vectors
%γ,n (W ) (see Definition 3.5) using Theorem 3.4 and Theorem 2.14.

3.4.1 Examples

Let H be the lattice Z2 with the quadratic form q((a, b)) = ab. Any lattice
isomorphic to H is called a hyperbolic plane.

1. Let L = H ⊕ H be the even unimodular lattice of signature (2, 2).

Strictly speaking this case is not covered by our approach to Borcherds prod-
ucts because l = 2. However, if we define all Poincaré and Eisenstein series in
section 1.2 using Hecke’s trick, then our argument still works. There are two
minor difficulties: To get all nearly holomorphic modular forms of weight
0 (Proposition 1.12) we also have to consider Eisenstein series of weight
0. Moreover, the Eisenstein series E of weight 2 (Theorem 1.6) is in gen-
eral no longer holomorphic. Some steps of the proof of Theorem 3.22 can
be simplified. For instance one could use Theorem 2.14 and the fact that
F (1, 1, 2; z) = −z −1 log(1 − z) to compute the Weyl vector %f (W ) directly.
The results of section 3.1 are not needed in this case.
!
The space M0,L of nearly holomorphic modular forms of weight 0 is equal
to the polynomial ring C[j(τ )], where

j(τ ) = q −1 + 744 + 196884q + . . .

is the usual j-function, and q = e(τ ). In particular any Fourier polynomial

n
P
n<0 c(n)q is the principal part of a nearly holomorphic modular form.
This is equivalent to the fact that there are no elliptic cusp forms of weight
2, i.e. S2,L = {0}. We consider the nearly homomorphic modular form
X
J(τ ) = j(τ ) − 744 = c(n)q n
n≥−1

!
in M0,L . Modular forms for the orthogonal group Γ (L) can be identified with
modular forms on H×H for the group SL2 (Z)×SL2 (Z). The Heegner divisors
H(m) = 21 H(0, m) are just the Hecke correspondences T (|m|) on H × H. In
particular H(−1) is given by the translates of the diagonal {(τ, τ ); τ ∈ H}.
In this case Theorem 3.22 (when extended to l = 2) tells us that there exists
a (on H × H) holomorphic modular form Ψ for SL2 (Z) × SL2 (Z) of weight 0
90 3 The Fourier expansion of the theta lift

with divisor T (1). It is easily checked that Ψ equals the function j(τ1 )−j(τ2 ).
We obtain the product expansion
Y
j(τ1 ) − j(τ2 ) = p−1 (1 − pm q n )c(mn) ,
m>0
n∈Z

where p = e(τ1 ) and q = e(τ2 ). The product converges for =(τ1 )=(τ2 ) > 1.
This is the famous denominator formula for the monster Lie algebra [Bo4].
It was proved independently in the 80’s by Koike, Norton, and Zagier.

2. This example is due to Gritsenko and Nikulin [GN]. Let D be the

negative definite lattice Z with the quadratic form q(a) = −a2 and L the
!
lattice D ⊕ H ⊕ H of signature (2, 3). In this case the space M−1/2,L of nearly
holomorphic modular forms of weight −1/2 can be identified with the space
Je0,1 of weak Jacobi forms of weight 0 and index 1 in the sense of [EZ]. The
obstruction space S5/2,L can be identified in the same way with the space of
skew holomorphic Jacobi forms of weight 3 and index 1. This space equals
{0}. Therefore every Fourier polynomial is the principal part of a weak Jacobi
form in Je0,1 . We consider the special weak Jacobi form

φ12,1 (τ, z) X
φ0,1 (τ, z) = = c(n, r)q n ζ r
∆(τ )
n≥0
r∈Z
−1
= (ζ + 10 + ζ ) + q(10ζ −2 − 64ζ −1 + 108 − 64ζ + 10ζ 2 ) + O(q 2 )

in Je0,1 . Here we have used the notation of [EZ] §9, in particular ζ = e(z).
Modular forms for the orthogonal group Γ (L) can be identified with Siegel
modular forms for the symplectic group Sp(4, Z) of genus 2. The Heegner
divisor H(∗, m) only depends on m and is equal to the Humbert surface of
discriminant |4m| in the Siegel upper half plane HS of genus 2. For Z =
( τz τz0 ) ∈ HS we write q = e(τ ), r = e(z), and s = e(τ 0 ). According to
Theorem 3.22 there exists a holomorphic modular form Ψ (Z) of weight 5 for
the group Sp(4, Z), whose divisor is the Humbert surface of discriminant 1.
It is known that the product
(2)
Y
∆0 = ϑa,b (Z)

of the ten even theta constants is a Siegel modular form of weight 5 with the
same divisor (see [Fr1] chapter 3.1). Thus Ψ has to be a constant multiple of
(2)
∆0 . It easily checked that the right constant factor is 1/64. We obtain the
(2)
following product expansion for ∆0 :
1 (2) Y
∆ = (qrs)1/2 (1 − q i rj sk )c(ik,j) .
64 0
i,j,k∈Z
i−j/2+k>0
 3.4 Borcherds products 91

The product converges on det(Y ) > 1/4, where Y = =(Z). Maass showed in
(2)
[Ma3] that ∆0 can also be constructed as a Saito-Kurokawa lifting.

3. Let K be the even unimodular lattice of signature (1, 33) and L =

K ⊕ H. Then k = −16, κ = 18, and S18,L = Cg(τ ), where g(τ ) is the unique
normalized elliptic cusp form of weight 18, i.e.

g(τ ) = ∆(τ )E6 (τ ) = q − 528q 2 − 4284q 3 + 147712q 4 + . . .

∞
X
= a(n)q n .
n=1

Here ∆(τ ) = q n≥1 (1 − q n )24 and Er (τ ) denotes the Eisenstein series of

Q
weight r normalized such that its constant term equals 1. The nearly holo-
morphic modular form

f (τ ) = E8 (τ )/∆(τ )2 = q −2 + 528q −1 + 86184 + 4631872q + . . .

X∞
= c(n)q n
n=−2

!
belongs to M−16,L . Write H(m) = 12 H(0, m). Theorem 3.22 implies that
there is a holomorphic modular form Ψ of weight 43092 for the orthogonal
group Γ (L) with divisor H(−2) + 528H(−1) and product expansion
Y
Ψ (Z) = e((%f (W ), Z)) (1 − e((λ, Z)))c(q(λ))
λ∈K
(λ,W )>0

for a suitable choice of W and %f (W ) ∈ K ⊗ Q. The product converges for

q(Y ) > 2. Since S18,L 6= {0}, in this case not any Fourier polynomial is the
principal part of a nearly holomorphic modular form. The obstructions are
given by the coefficients a(n) of g.
A conjecture of Lehmer says that a(n) 6= 0 for all positive integers n.
(This conjecture was originally stated for the coefficients of ∆(τ ) but was
generalized later.) Freitag pointed out that Lehmer’s conjecture can be re-
stated in terms of the geometry of the modular variety XL = Hl /Γ (L). The
vanishing of a(n) implies by Theorem 3.22 that the Heegner divisor H(−n)
vanishes in the modified divisor class group fCl(XL ) ⊗ Q of XL (see chapter
5). On the other hand it is a consequence of our converse-theorem 5.12 that
the vanishing of H(−n) in f Cl(XL ) ⊗ Q implies that a(n) = 0.
4 Some Riemann geometry on O(2, l)

The aim of this chapter is to prove Theorem 4.23. To this end we have to
exploit the Riemann structure of the quotient Hl /Γ (L).

4.1 The invariant Laplacian

In this section we consider the O+ (V )-invariant Laplacian which acts on

functions on the generalized upper half plane Hl . General references are the
books [GH, Hel, Wel].
As in section 3.2 let z ∈ L be a primitive norm 0 vector and z 0 ∈ L0
with (z, z 0 ) = 1. Then z̃ = z 0 − q(z 0 )z is an isotropic vector in L ⊗ Q. Put
⊥
K = L ∩ z ⊥ ∩ z 0 and recall the definition of Hl and K+ .
First, we introduce suitable coordinates on Hl . Let d ∈ K be a primitive
⊥
norm 0 vector and d0 ∈ K 0 with (d, d0 ) = 1. Then D = K ∩ d⊥ ∩ d0 is
a negative definite lattice and K ⊗ Q = (D ⊗ Q) ⊕ Qd0 ⊕ Qd. The vector
d˜ = d0 − q(d0 )d ∈ K ⊗ Q has norm 0 and satisfies (d, ˜ d) = 1. The lattice
˜
D ⊕ Zd ⊕ Zd is commensurable with K.
Let d3 , . . . , dl be an orthogonal basis of D ⊗ R normalized such that d23 =
· · · = d2l = −2. Then d, ˜ d, d3 , . . . , dl is a basis of K ⊗ R. If Z = z1 d˜ +
z2 d + z3 d3 + · · · + zl dl ∈ K ⊗ C, we briefly write Z = (z1 , z2 , z3 , . . . , zl ) and
denote its real part by X = (x1 , x2 , x3 , . . . , xl ), resp. its imaginary part by
Y = (y1 , y2 , y3 , . . . , yl ). We obviously have

q(Y ) = y1 y2 − y32 − y42 − · · · − yl2

and

Hl = {Z = (z1 , . . . , zl ) ∈ K ⊗ C; y1 > 0, y1 y2 − y32 − y42 − · · · − yl2 > 0}.

Moreover, sometimes we just write Z = (z1 , z2 , ZD ) for Z = (z1 , . . . , zl ) ∈ Hl ,

where ZD = z3 d3 + · · · + zl dl ∈ D ⊗ C. The real part of ZD is denoted by
XD , the imaginary part by YD .
The generalized upper half plane Hl can be viewed as a realization of the
irreducible Hermitean symmetric space O+ (V )/H, where H ∼ = SO(2)× SO(l)
denotes a maximal compact subgroup. The Killing form on the Lie algebra of
94 4 Some Riemann geometry on O(2, l)

O+ (V ) induces an O+ (V )-invariant Riemann metric g on Hl . According to

[Hel] Chap. VIII §5 any O+ (V )-invariant Riemann metric on Hl is a constant
multiple of g. In this sense the geometry on Hl is determined uniquely. By
[Hel] Chap. VIII §4 the metric g is Kählerian.
We will now describe g more explicitly. Lemma 3.20 implies that the
(1, 1)-form
i
ω = − ∂ ∂¯ log(q(Y ))
2
is invariant under O+ (V ). Moreover, ω is positive, that is

iX
ω= hµν (Z)dzµ ∧ dz̄ν ,
2 µ,ν

where h(Z) = (hµν (Z)) is a positive definite Hermitean matrix for any Z ∈
Hl . In fact, since ω is O+ (V )-invariant it suffices to check this at the special
point I = i(1, 1, 0, . . . , 0) ∈ Hl . A somewhat lengthy but trivial computation
shows that1  
1
 1 
1 2

h(I) =  .
 
4 .. 
 . 
2
Furthermore, ω is obviously closed. Hence it defines an O+ (V )-invariant
Kähler metric on Hl . Its underlying Riemann metric has to be a constant
multiple of g. It is easily seen that the attached invariant volume element
equals
dX dY
,
q(Y )l
where dX = dx1 · · · dxl and dY = dy1 · · · dyl . The invariant Laplace operator
is given by
−(d ∗ d ∗ + ∗ d ∗ d).
Here ∗ denotes the Hodge star operator for the Riemann metric g. Nakajima
[Nak] showed that, up to a constant multiple, the invariant Laplace operator
is equal to
l l
!
X 1 X
Ω= yµ yν ∂µ ∂¯ν − q(Y ) ∂1 ∂¯2 + ∂¯1 ∂2 − ∂µ ∂¯µ , (4.1)
µ,ν=1
2 µ=3

where
1
Zero matrix entries are often omitted.
 4.1 The invariant Laplacian 95
 
∂ 1 ∂ ∂
∂µ = = −i ,
∂zµ 2 ∂xµ ∂yµ
 
∂ 1 ∂ ∂
∂¯µ = = +i .
∂ z̄µ 2 ∂xµ ∂yµ
(Note that there is a misprint in [Nak] in the definition of ∆1 .) The proof
relies on the fact that there is up to a constant multiple only one O+ (V )-
invariant second order differential operator on Hl . Therefore one only has to
verify the invariance of Ω.
We will now show that for σ > 1 − k/2 and Z ∈ / H(β, m) the function
ΦLβ,m (Z, s) is an eigenfunction of Ω. First, we prove some lemmas.

Lemma 4.1. If Z = X + iY ∈ Hl and s ∈ C, then

 2 
s sl
Ωq(Y )s = − q(Y )s , (4.2)
2 4
Ω log(q(Y )) = −l/4. (4.3)

Proof. This can be verified by a straightforward computation. t

u

Recall the definition (1.34) of the Laplace operator ∆k of weight k. We de-

fine the usual Maass differential operators on smooth functions H → C[L0 /L]
by
∂
Rk = 2i + ky −1 , (4.4)
∂τ
∂
Lk = 2iy 2 . (4.5)
∂ τ̄
For any smooth function f : H → C[L0 /L] and (M, φ) ∈ Mp2 (Z) we have

(Rk f ) |k+2 (M, φ) = Rk (f |k (M, φ)) ,

(Lk f ) |k−2 (M, φ) = Lk (f |k (M, φ)) .

Hence Rk raises the weight, whereas Lk lowers it. The operator ∆k can be
expressed in terms of Rk and Lk by

∆k = Lk+2 Rk − k = Rk−2 Lk .

Lemma 4.2. Let f, g : H → C[L0 /L] be smooth functions satisfying f |k

(M, φ) = f and g |k+2 (M, φ) = g for all (M, φ) ∈ Mp2 (Z). Then

Z Z Z1/2
k−2 k
hf, giy k y=u dx.
 
hf, (Lk+2 g)iy dx dy − h(Rk f ), giy dx dy =
Fu Fu −1/2

Here Fu denotes the truncated fundamental domain (2.14).

96 4 Some Riemann geometry on O(2, l)

Proof. The assumptions imply that ω = y k hf (τ ), g(τ )idτ̄ is a SL2 (Z)-

invariant 1-form on H. By Stokes’ theorem we have
Z Z
y hf (τ ), g(τ )idτ̄ = d y k hf (τ ), g(τ )idτ̄
k

∂Fu Fu
Z  
∂ k ∂
= − y hf (τ ), g(τ )i − i y k hf (τ ), g(τ ) dx ∧ dy
∂y ∂x
Fu
Z
y k−2 hf, (Lk+2 g)i − y k h(Rk f ), gi dx ∧ dy.


=
Fu

In the integral over ∂Fu on the left hand side the contributions from SL2 (Z)-
equivalent boundary pieces cancel. Thus

Z Z1/2
k
 k 
y hf (τ ), g(τ )idτ̄ = y hf (τ ), g(τ )i y=u dx.
∂Fu −1/2

This implies the assertion. t

u

Lemma 4.3. Let f, g : H → C[L0 /L] be smooth functions satisfying f |k

(M, φ) = f and g |k (M, φ) = g for all (M, φ) ∈ Mp2 (Z). Then
Z Z
h(∆k f ), giy k−2 dx dy − hf, (∆k g)iy k−2 dx dy
Fu Fu

Z1/2 Z1/2
hf, (Lk g)iy k−2 y=u dx − h(Lk f ), giy k−2 y=u dx.
   
=
−1/2 −1/2

Proof. We write ∆k = Rk−2 Lk and apply the previous lemma twice. t

u

Lemma 4.4. For v ∈ Gr(L) − H(β, m) and s = σ + it ∈ C with σ > 1 − k/2

we have the identity
Z
dx dy
lim hFβ,m (τ, s), ∆k (ΘL (τ, v)y l/2 )iy k
u→∞ y2
Fu

= (l/4 + 1/2 − s)(l/4 − 1/2 + s)Φβ,m (v, s).

Proof. Since

∆k Fβ,m (τ, s) = (l/4 + 1/2 − s)(l/4 − 1/2 + s)Fβ,m (τ, s),

Lemma 4.3 implies that

 4.1 The invariant Laplacian 97
Z
dx dy
hFβ,m (τ, s), ∆k (ΘL (τ, v)y l/2 )iy k
y2
Fu
Z
dx dy
− (s − 1/2 − l/4)(s − 1/2 + l/4) hFβ,m (τ, s), ΘL (τ, v)y l/2 iy k
y2
Fu

Z1/2 h i
= h(Lk Fβ,m (τ, s)), ΘL (τ, v)y l/2 iy k−2 dx
y=u
−1/2

Z1/2 h i
− hFβ,m (τ, s), (Lk ΘL (τ, v)y l/2 )iy k−2 dx. (4.6)
y=u
−1/2

Thus it suffices to show that the integral on the right hand side tends to 0
as u → ∞. We have
Z1/2
h(Lk Fβ,m (τ, s)), ΘL (τ, v)y l/2 iy k−2 dx
−1/2

Z1/2   
∂ ∂
=i y +i Fβ,m (τ, s), ΘL (τ, v) dx. (4.7)
∂x ∂y
−1/2

If we insert the Fourier expansions

X X
Fβ,m (τ, s) = c(γ, n; y, s)eγ (nx),
γ∈L0 /L n∈Z+q(γ)
X
ΘL (τ, v) = exp(−4πyq(λv ) + 2πyq(λ))eλ (q(λ)x),
λ∈L0

and carry out the integration, we find that (4.7) equals

 
X
−4πyq(λv )+2πyq(λ) ∂
− ye 2πq(λ)c(λ, q(λ); y, s) + c(λ, q(λ); y, s) .
0
∂y
λ∈L

Using the explicit formulas for the coefficients c(γ, n; y, s) given in Theorem
1.9 and the asymptotic behavior of the Whittaker functions we may infer
that
∂ ∂
lim (4.7) = − lim y c(0, 0; y, s) = −b(0, 0, s) lim y y 1−s−k/2 .
y→∞ y→∞ ∂y y→∞ ∂y

Since σ > 1 − k/2 this is equal to 0. In the same way it can be seen that
Z1/2
lim hFβ,m (τ, s), (Lk ΘL (τ, v)y l/2 )iy k−2 dx = 0.
y→∞
−1/2
98 4 Some Riemann geometry on O(2, l)

Taking the limit u → ∞ in (4.6), we obtain the assertion. t

u

The following proposition is crucial for our argument.

Proposition 4.5. The Siegel theta function ΘL (τ, Z), considered as a func-
tion on H × Hl , satisfies the differential equation

∆k ΘL (τ, Z)y l/2 = −2ΩΘL (τ, Z)y l/2 .

Proof. For τ ∈ H and v ∈ Gr(L) we define a function on V = L ⊗ R by

Fτ,v : V → C, λ 7→ Fτ,v (λ) = e(τ q(λv ) + τ̄ q(λv⊥ )).

If Z ∈ Hl corresponds to v via the isomorphism Hl ∼ = Gr(L), then we also

write Fτ,Z instead of Fτ,v . For any fixed λ ∈ V we may consider Fτ,Z (λ) as
a function in τ and Z. To prove the proposition it suffices to show that

∆k Fτ,Z (λ)y l/2 = −2ΩFτ,Z (λ)y l/2 .

This essentially follows from [Sn] Lemma 1.5 as we will now indicate. As far
as possible we use the same notation as in [Sn].
We choose a basis for L and identify V with Rl+2 . Denote by Q the Gram
matrix of L with respect to this basis, so that (x, y) = xt Qy. Let g 7→ r0 (g, Q)
be the Weil representation of Mp2 (R) × O(V ) in L2 (V ) as defined in [Sn].
For fixed τ ∈ H and Z ∈ Hl the function Fτ,Z is rapidly decreasing.
According to [Sn] Lemma 1.2 we have
−l
(r0 (σ, Q)Fτ,Z )(λ) = φ(τ )−2 φ(τ ) FM τ,Z (λ), σ = (M, φ) ∈ Mp2 (R),
−1
(r0 (g, Q)Fτ,Z )(λ) = Fτ,Z (g λ) = Fτ,gZ (λ), g ∈ O+ (V ).

It is easily seen that Fi,Z satisfies the condition (1.19) in [Sn] with m = 2 − l.
If we put  1/2 −1/2  
y xy
στ = , 1 ∈ Mp2 (R)
0 y −1/2
for τ = x + iy ∈ H, then r0 (στ , Q)Fi,Z = y (2+l)/4 Fτ,Z .
The map (1.1) gives a locally isomorphic embedding of SL2 (R) into
Mp2 (R). Thus for any element u of the universal enveloping algebra U of
the Lie algebra of SL2 (R) we may consider r0 (u, Q) as a differential operator
on functions V → C. Let C be the Casimir element in U and CQ = r0 (C, Q).
Then by [Sn] Lemma 1.4 we may infer

−4∆k y l/2 Fτ,Z (λ) = (1 − l2 /4)y l/2 Fτ,Z (λ) + y −k/2 (r0 (στ , Q)CQ Fi,Z ) (λ).
(4.8)
Let % denote the representation of O+ (V ) on C ∞ (Hl ) given by

(%(g)f )(Z) = f (g −1 Z), g ∈ O+ (V ).

 4.1 The invariant Laplacian 99

For λ ∈ V we obviously have

%(g)(Fτ,Z (λ)) = r0 (g −1 , Q)Fτ,Z (λ).

(4.9)

Let D be the Casimir element of the universal enveloping algebra of the Lie
algebra of O+ (V ). Then DQ = r0 (D, Q) is a differential operator on functions
V → C, and %(D) a differential operator on functions Hl → C. According
to [Hel] p. 451 ex. 5 the operator %(D) equals (up to the sign) the Laplace
operator on Hl . By (4.9) we have %(D)(Fτ,Z (λ)) = (−DQ Fτ,Z )(λ). Hence
there exists a non-zero constant cl , only depending on l, such that

cl Ω(Fτ,Z (λ)) = (DQ Fτ,Z )(λ) (4.10)

for all τ ∈ H and λ ∈ V . According to [Sn] Lemma 1.5 for any smooth rapidly
decreasing function F on V the identity

CQ F = (DQ + l2 /4 − 1)F (4.11)

holds. Combining (4.8), (4.10), and (4.11), we find

−4∆k y l/2 Fτ,Z (λ) = cl Ωy l/2 Fτ,Z (λ) (4.12)

for any λ ∈ V .
To determine the constant cl we consider the above identity (4.12) for
λ = d ∈ K. In this case we have
x2 + y12
 
y l/2 Fτ,Z (d) = y l/2 exp −πy 1 ,
q(Y )

and also (4.12) becomes much easier. (Here we have to be careful with our
notation: y = =(τ ) and Y = (y1 , . . . , yl ) = =(Z).) A straightforward compu-
tation shows that cl = 8. This completes the proof. t
u

Theorem 4.6. For Z ∈ Hl − H(β, m) and σ > 1 − k/2 the function

Φβ,m (Z, s) is an eigenfunction of the invariant Laplacian Ω on Hl . More
precisely we have
1
ΩΦβ,m (Z, s) = (s − 1/2 − l/4)(s − 1/2 + l/4)Φβ,m (Z, s).
2
Proof. Arguing as in section 2.2 we find that all iterated partial derivatives
with respect to Z of
Z
dx dy
hFβ,m (τ, s), ΘL (τ, Z)y l/2 iy k
y2
Fu

converge locally uniformly in Z as u → ∞. Therefore we have

100 4 Some Riemann geometry on O(2, l)
Z
dx dy
ΩΦβ,m (Z, s) = lim hFβ,m (τ, s), ΩΘL (τ, Z)y l/2 iy k .
u→∞ y2
Fu

By Proposition 4.5 this equals

Z
1 dx dy
− lim hFβ,m (τ, s), ∆k (ΘL (τ, Z)y l/2 )iy k .
2 u→∞ y2
Fu

Using Lemma 4.4, we obtain the assertion. t

u
Theorem 4.7. For Z ∈ Hl − H(β, m) the regularized theta integral Φβ,m (Z)
satisfies
l
ΩΦβ,m (Z) = bβ,m (0, 0).
4
Here bβ,m (0, 0) = bβ,m (0, 0, 1 − k/2) denotes the (0, 0)-th Fourier coefficient
of the Poincaré series Fβ,m (τ, 1 − k/2).
Proof. Define a function
2
R(Y, s) = √ (πq(Y ))1/2+l/4−s Γ (s − l/4)bβ,m (0, 0, s)ζ(2s − l/2).
π
In the proof of Theorem 3.9 we saw that
Φ̃β,m (Z, s) := Φβ,m (Z, s) − R(Y, s)
is holomorphic in s near 1 − k/2. Using Theorem 2.15, it can be shown in the
same way that Φ̃β,m (Z, s) is twice continuously differentiable as a function
in (Z, s) for Z ∈ Hl − H(β, m) and s varying in a neighborhood of 1 − k/2.
This implies that
ΩΦβ,m (Z) = ΩCs=1−k/2 [Φβ,m (Z, s)]
h i
= Cs=1−k/2 ΩΦ̃β,m (Z, s) + ΩCs=1−k/2 [R(Y, s)] . (4.13)

By Theorem 4.6 and Lemma 4.1 we have

ΩΦ̃β,m (Z, s) = ΩΦβ,m (Z, s) − ΩR(Y, s)
1
= (s − 1/2 − l/4)(s − 1/2 + l/4)Φ̃β,m (Z, s)
2
and therefore h i
Cs=1−k/2 ΩΦ̃β,m (Z, s) = 0. (4.14)
In the proof of Theorem 3.9 we found that
Cs=1−k/2 [R(Y, s)] = −bβ,m (0, 0) log(q(Y )) + const.
Thus, using Lemma 4.1, we get
l
ΩCs=1−k/2 [R(Y, s)] = bβ,m (0, 0). (4.15)
4
If we insert (4.14) and (4.15) into (4.13), we obtain the assertion. t
u
 4.2 Reduction theory and Lp -estimates 101

Remark 4.8. Theorem 4.6 and Theorem 4.7, together with a regularity result
for elliptic differential operators on analytic Riemann manifolds, also imply
that the functions Φβ,m (Z, s), Φβ,m (Z), and ξβ,m (Z) are real analytic (see
[Rh] §34).

4.2 Reduction theory and Lp -estimates

Recall that the invariant volume element on the upper half plane Hl is
given by dX dY
q(Y )l
. Let Γ ≤ Γ (L) be a subgroup of finite index. As usual we
write Lp (Hl /Γ ) for the Banach space of equivalence classes of measurable
Γ -invariant functions f : Hl → C with
Z
dX dY
|f (Z)|p < ∞. (4.16)
q(Y )l
Hl /Γ

To decide whether a function satisfies this integrability condition we need a

fundamental domain for the action of Γ on Hl .
We fix a coordinate system on Hl as in section 4.1 page 93. (This depends
on the choice of the vectors z, z 0 , d, d0 and the orthogonal basis d3 , . . . , dl of
D ⊗ R.) We use the notation of section 4.1.
Let
O+ +
Q (L) = {g ∈ O (V ); g(L ⊗ Q) = L ⊗ Q} (4.17)
be the rational orthogonal group of the lattice L.

Definition 4.9. Let t > 0. We define the Siegel domain St in Hl to be the

subset of Z = X + iY ∈ Hl satisfying

x21 + x22 + |q(XD )| < t2 , (4.18a)

1/t < y1 , (4.18b)
y12 < t2 q(Y ), (4.18c)
|q(YD )| < t2 y12 . (4.18d)

Proposition 4.10. Let Γ ≤ Γ (L) be a subgroup of finite index.

i) For any t > 0 and any g ∈ O+
Q (L) the set

{σ ∈ Γ ; σgSt ∩ St 6= ∅}

is finite.
ii) There exist a t > 0 and finitely many rational transformations
g1 , . . . , ga ∈ O+
Q (L) such that

S = g1 St ∪ · · · ∪ ga St

is a fundamental domain for the action of Γ on Hl , i.e. Γ S = Hl .

102 4 Some Riemann geometry on O(2, l)

Proof. This can be deduced from the general reduction theory for algebraic
groups (cf. [Bl1], [Bl2]). We briefly indicate the argument. See [Me] for a more
detailed treatment.
For any algebraic group A we denote by A+ the component of the identity.
There is a basis e3 , . . . , el of D ⊗ Q such that the restriction of q to D ⊗ Q
has the diagonal Gram matrix C = diag(−c3 , . . . , −cl ) with positive rational
˜ z̃ of L ⊗ Q, the
numbers cj . Then, with respect to the basis z, d, e3 , . . . , el , d,
quadratic form q has the Gram matrix
 
0 0 J
F = 0 C 0 ,
J 0 0

where J = ( 01 10 ). The real orthogonal group

O(V ) = {M ∈ SLl+2 (R); M tF M = F }

of V is defined over Q. The group G = OQ (L) of rational points is simply the

subgroup of matrices with rational entries, and Γ is an arithmetic subgroup
of G.
Let IL denote the special point z + id + id˜ + z̃ ∈ V (C) and [IL ] the
corresponding Element of K+ ⊂ P (V (C)). The upper half plane Hl is iden-
tified with K+ via Z 7→ [−q(Z)z + Z + z̃] as in section 3.2. Here the point
I = (i, i, 0) ∈ Hl is mapped to [IL ]. The group O+ (V ) = G+
R acts transitively
on K+ and the stabilizer K of [IL ] is a maximal compact subgroup2 . The
assignment Kg 7→ g −1 [IL ] identifies K\ O+ (V ) with K+ .
A minimal parabolic Q-subgroup of G is given by
 
 H 0 H1 H 2 
P =  0 B H3  ,
0 0 H4
 

where H0 , H4 are upper triangular and B is an orthogonal transformation in

OQ (D), with additional relations that ensure P ⊂ G. The unipotent radical
U of P is the set of matrices H in P , where H0 , H4 are unipotent, B = E,
and

H4 = H0−1 , (4.19)
CH3 + H1t JH4 = 0, (4.20)
H4t JH2 + H2t JH4 + H3t CH3 = 0. (4.21)

The inverse of a matrix H ∈ U equals

2
Only in this proof K denotes a maximal compact subgroup. Usually it denotes
a sublattice of L.
 4.2 Reduction theory and Lp -estimates 103

H4 JH3t C JH2t J
 

H −1 = 0 E C −1 H1t J  .
0 0 H0
If we write
     
1 h12 h13 . . . h1l h1,l+1 h1,l+2
H0 = , H1 = , H2 = ,
0 1 h23 . . . h2l h2,l+1 h2,l+2
then H is completely determined by h12 , . . . , h1,l+1 , h23 , . . . , h2l , and these
may be freely chosen. Thus UR is homeomorphic to R2l−2 . Equality (4.21)
implies in particular that
1 −1 2
h2,l+1 = (c h + · · · + c−1 2
l h2l ).
2 3 23
For any t > 0 the set
n Pl o
1 −1 2
ωt = H ∈ U ; h212 + h21,l+1 + 2
2
j=3 cj h1j < t and h2,l+1 < t
2

is a relatively compact open neighborhood of the identity in S

UR . This can be
checked using the above relations. Moreover, we have UR = t>0 ωt .
A maximal Q-split torus S in P is given by the set of diagonal matrices
 
λ1
 λ2 
 
 E .
−1
 
 λ2 
−1
λ1

The set of simple roots of G with respect to S and P is equal to ∆ = {α1 , α2 },

where
α1 (T ) = λ1 /λ2 , α2 (T ) = λ2
for T ∈ S as above (cf. [Bl1] p. 80, [Bl2] p. 16). Let t > 0. We denote by At
the subset
At = {T ∈ SR+ ; α1 (T ) < t, α2 (T ) < t}
of SR+ . Then the open subset

St = K · At · ωt

of GR is a Siegel domain (with respect to K, P, S) in the sense of definition

12.3 in [Bl1]. (Note the remark at the end of §12.3 in [Bl1] and the fact that
S is also a maximal split torus over R.) Now, by a straightforward calculation
it can be verified that St precisely corresponds to St under the identification
K\ O+ (V ) → Hl , Kg 7→ g −1 ◦ I. (Caution: St is defined with respect to
˜ z̃ of L ⊗ R, whereas At and ωt are defined with
the basis z, d, d3 , . . . , dl , d,
˜ z̃.) Therefore assertion (i) follows from
respect to the basis z, d, e3 , . . . , el , d,
[Bl1] Theorem 15.4 and (ii) from [Bl1] Theorem 13.1. t
u
104 4 Some Riemann geometry on O(2, l)

Let f : Hl → C be a measurable Γ -invariant function. Proposition 4.10

immediately implies that f ∈ Lp (Hl /Γ ) if and only if S |f (Z)|p dX dY
R
q(Y )l
< ∞.
Let t > 0. In the next few lemmas we derive some useful estimates that
hold on St .

Lemma 4.11. Suppose that p < l/2. Then

Z
dX dY
q(Y )p < ∞.
q(Y )l
St

Proof. Since the set of X defined by (4.18a) is compact, it suffices to prove

that Z
q(Y )p−l dY < ∞,
Rt

where Rt denotes the set of Y defined by (4.18b)–(4.18d). It is easily seen

that
y1 y 2
< q(Y ) < y1 y2 (4.22)
1 + t4
for Y ∈ Rt . Thus it suffices to show that
Z
I= (y1 y2 )p−l dy1 dy2 dYD
y1 >1/t
t2 q(Y )>y12
|q(YD )|<t2 y12

is finite. We have
Z
I t (y1 y2 )p−l dy1 dy2 dYD
y1 >1/t
y2 /y1 >1/t2
|q(YD )|<t2 y12
Z
t (y1 y2 )p−l y1l−2 dy1 dy2
y1 >1/t
y2 /y1 >1/t2
Z
t y12p−l−1 up−l dy1 du.
y1 >1/t
u>1/t2

Since l ≥ 2, the latter integral is finite for p < l/2. t

u

In particular any Siegel domain has finite volume.

Lemma 4.12. Let p < l − 1. Then

 4.2 Reduction theory and Lp -estimates 105
Z
dX dY
|(ν, Y )|p < ∞.
q(Y )l
St

for any ν ∈ K ⊗ R.

Proof. Using Minkowski’s inequality and Lemma 4.11, we find that it suffices
to prove that St y2p q(Y )−l dX dY < ∞. This can be done as in Lemma 4.11.
R

t
u

Lemma 4.13. There exists an ε > 0 such that

2(ν, Y )2 − Y 2 ν 2 > ε y22 ν12 + y12 ν22 + q(Y )|νD

2


|

for any ν = (ν1 , ν2 , νD ) ∈ K 0 and any Y ∈ Rt . Here Rt denotes the set of Y

defined by (4.18b)–(4.18d).

Proof. Let h(ν, Y ) denote the function

h(ν, Y ) = 2(ν, Y )2 − Y 2 ν 2

(ν ∈ K ⊗ R and iY ∈ Hl ). For any real orthogonal transformation g ∈

O+ (K ⊗ R) we have h(gν, gY ) = h(ν, Y ). Moreover, if Y is of the form
(y1 , y2 , 0), then
h(ν, Y ) = 2y22 ν12 + 2y12 ν22 + Y 2 |νD
2
|. (4.23)
Recall that for any isotropic vector u ∈ K ⊗ R and any v ∈ K ⊗ R with
(u, v) = 0 one has the Eichler transformation E(u, v) ∈ O+ (K ⊗ R), defined
by
E(u, v)(a) = a − (a, u)v + (a, v)u − q(v)(a, u)u.
For a given Y = (y1 , y2 , YD ) ∈ Rt we consider the special Eichler transfor-
mations

E = E(d, YD /y1 ),
˜ YD /y2 ).
Ẽ = E(d,

It is easily checked that

E(Y ) = (y1 , y2 + q(YD )/y1 , 0) ,

 
E(ν) = ν1 , ∗, νD − yν11 YD ,
Ẽ(Y ) = (y1 + q(YD )/y2 , y2 , 0) ,
 
Ẽ(ν) = ∗, ν2 , νD − yν22 YD .

Using 4.23 we obtain the following estimate:

106 4 Some Riemann geometry on O(2, l)

h(ν, Y ) = 21 h(Eν, EY ) + 12 h(Ẽν, ẼY )

≥ (y2 + q(YD )/y1 )2 ν12 + (y1 + q(YD )/y2 )2 ν22
ν1 2 ν2 2
+ q(Y ) (νD − y1 YD ) + q(Y ) (νD − y2 YD ) .

Let ε ∈ (0, 1). Then

ν1 2 ν2 2
q(Y ) (νD − y1 YD ) + q(Y ) (νD − y2 YD )
 
ν1 2 ν2 2
≥ −εq(Y ) (νD − y1 YD ) + (νD − y2 YD )
 2 !
2 ν1 y 2 + ν2 y 1 ν1 ν2 2
= −εq(Y ) νD + νD − YD −2 Y
y1 y2 y1 y2 D
2
≥ εq(Y )|νD | − 2ε|ν1 ν2 ||YD2 |.

Hence we find that

h(ν, Y ) ≥ (y2 + q(YD )/y1 )2 ν12 + (y1 + q(YD )/y2 )2 ν22

2
+ εq(Y )|νD | − 2ε|ν1 ν2 ||YD2 |. (4.24)

t4
It is easily seen that |q(YD )| < t4 +1 y1 y2 for Y ∈ Rt . This implies

y2 + q(YD )/y1 ≥ y2 /(t4 + 1),

y1 + q(YD )/y2 ≥ y1 /(t4 + 1).

If we insert this into (4.24), we get

1
h(ν, Y ) ≥ (y 2 ν 2 + y12 ν22 ) + εq(Y )|νD 2
|
2(t4 + 1)2 2 1
1
y 2 ν 2 − 8ε(t4 + 1)2 y1 y2 |ν1 ν2 | + y12 ν22 .


+
2(t4 + 1)2 2 1
1
Thus for ε < 4(t4 +1)2 we finally obtain

h(ν, Y ) ≥ ε(y22 ν12 + y12 ν22 + q(Y )|νD

2
|).

t
u

Corollary 4.14. There exists an ε > 0 with the following property: For any
ν ∈ K 0 with q(ν) ≥ 0 and any Y ∈ Rt the inequality

|(ν, Y )| ≥ ε(y2 |ν1 | + y1 |ν2 |)

holds. Here Rt denotes the set of Y defined by (4.18b)–(4.18d).

 4.2 Reduction theory and Lp -estimates 107

Proof. For q(ν) ≥ 0 one has 2(ν, Y )2 ≥ 2(ν, Y )2 − ν 2 Y 2 . Hence, according to

Lemma 4.13 there is an ε > 0 such that

(ν, Y )2 ≥ ε(y22 ν12 + y12 ν22 + q(Y )|νD

2
|) ≥ ε(y22 ν12 + y12 ν22 )

for all ν ∈ K 0 with q(ν) ≥ 0 and all Y ∈ Rt . This implies the assertion. t
u

Proposition 4.15 (Koecher principle). Let f : Hl → C be a holomorphic

function that satisfies
i) f (Z + k) = f (Z), for k ∈ K,
ii) f (σZ) = f (Z), for σ ∈ Γ (L) ∩ O(K), where O(K) is considered as a
subgroup of O(L).
Then f has a Fourier expansion of the form
X
f (Z) = a(ν)e((ν, Z)).
ν=(ν1 ,ν2 ,νD )∈K 0
ν1 ≥0; q(ν)≥0

In particular, f is bounded on the Siegel domain St .

Proof. The first assertion can be proved in the same way as the Koecher
principle for Siegel modular forms (cf. [Fr1] chapter I Hilfssatz 3.5). Note
that the assumption l ≥ 3 is crucial. The group Γ (L) ∩ O(K) plays the role
of SL(n, Z) in [Fr1]. Then the second assertion can be deduced using Lemma
4.13 or Corollary 4.14. t
u

Let t > 0 and g ∈ O+Q (L). Let f : Hl → C be a Γ -invariant measurable

function. We now consider the integral
Z
dX dY
I= |f (Z)|p
q(Y )l
gSt

over the translated Siegel domain gSt . It is convenient to use the identification
Hl → K+ , Z 7→ [Z + z̃ − q(Z)z], and to rewrite I as the integral
Z
|F (W )|p dω,
g S̃t

where S̃t = {W = [Z + z̃ − q(Z)z]; Z ∈ St }, F denotes the function on K+

corresponding to f , and dω denotes the O+ (V )-invariant measure on K+ . We
will work with different coordinate systems on K+ .
A 4-tuple Λ = (λ1 , . . . , λ4 ) of vectors in L ⊗ Q is called an admissible
index-tuple, if
i) q(λj ) = 0 for j = 1, . . . , 4,
ii) (λ1 , λ2 ) = 1 and (λ3 , λ4 ) = 1,
108 4 Some Riemann geometry on O(2, l)

iii) (λi , λj ) = 0 for i = 1, 2 and j = 3, 4,

iv) =((W, λ3 )/(W, λ1 )) > 0 for W ∈ K+ .
Then Qλ1 + Qλ2 and Qλ3 + Qλ4 are orthogonal hyperbolic planes in L ⊗ Q.
The intersection of (Qλ1 + Qλ2 )⊥ and L is a Lorentzian sublattice of L,
which will be denoted by K(Λ). The intersection of (Qλ1 + · · · + Qλ4 )⊥ and
L is a negative definite sublattice, denoted by D(Λ). If X ∈ L ⊗ R, then we
write XK(Λ) resp. XD(Λ) for the orthogonal projection of X to K(Λ) ⊗ R
resp. D(Λ) ⊗ R.
In the same way as in section 3.2 it can be seen that K+ is biholomorphic
equivalent to

Hl (Λ) = {X + iY ∈ K(Λ) ⊗ C; X, Y ∈ K(Λ) ⊗ R, (Y, λ3 ) > 0, q(Y ) > 0},

the isomorphism being given by

Hl (Λ) → K+ , Z 7→ [Z + λ2 − q(Z)λ1 ].

Let t > 0 and Λ = (λ1 , . . . , λ4 ) be an admissible index-tuple. We denote by

St (Λ) the subset of Z = X + iY ∈ Hl (Λ) satisfying

(X, λ3 )2 + (X, λ4 )2 − q(XD(Λ) ) < t2 , (4.25a)

1/t < (Y, λ3 ), (4.25b)
(Y, λ3 )2 < t2 q(Y ), (4.25c)
2 2
−q(YD(Λ) ) < t (Y, λ3 ) . (4.25d)

We define the Siegel domain S̃t (Λ) attached to Λ and t by

S̃t (Λ) = {[Z + λ2 − q(Z)λ1 ]; Z ∈ St (Λ)} ⊂ K+ .

˜ is an admissible index-tuple. We have Hl (Λ0 ) = Hl
Obviously Λ0 = (z, z̃, d, d)
and St (Λ0 ) = St . It is easily seen that the rational orthogonal group O+ Q (L)
+
acts transitively on admissible index-tuples. For g ∈ OQ (V ) the identity

g S̃t (λ1 , . . . , λ4 ) = S̃t (gλ1 , . . . , gλ4 ) (4.26)

holds.
In order to prove that F ∈ Lp (K+ /Γ ), it suffices to show that for any
t > 0 and any admissible index-tuple Λ the integral
Z
|F (W )|p dω (4.27)
S̃t (Λ)

is finite.
 4.2 Reduction theory and Lp -estimates 109

Lemma 4.16. Let t > 0 and Λ = (λ1 , . . . , λ4 ) be an admissible index-tuple.

For a, b ∈ Q>0 there exist t1 , t2 ∈ R with 0 < t1 < t < t2 , such that

S̃t1 (Λ) ⊂ S̃t (aλ1 , a1 λ2 , bλ3 , 1b λ4 ) ⊂ S̃t2 (Λ).

Proof. The proof is left to the reader. t

u

Lemma 4.17. Let λ1 , λ2 ∈ L ⊗ Q with q(λ1 ) = q(λ2 ) = 0 and (λ1 , λ2 ) = 1.

Let t0 > 0 and Λ0 be an admissible index-tuple of the form (λ1 , ∗, ∗, ∗). There
exist a t > 0 and an admissible index-tuple Λ of the form (λ1 , λ2 , ∗, ∗) such
that
S̃t0 (Λ0 ) ⊂ S̃t (Λ).

Proof. Write Λ0 = (λ1 , λ02 , λ03 , λ04 ) and put u = λ2 − λ02 . Then u ∈ L ⊗ Q with
(u, λ1 ) = 0. Consider the Eichler transformation E(λ1 , u) : L ⊗ Q → L ⊗ Q,

E(λ1 , u)(a) = a − (a, λ1 )u + (a, u)λ1 − q(u)(a, λ1 )λ1 .

It is easily checked that E(λ1 , u) ∈ O+ Q (L). Put Λ = E(λ1 , u)

−1 0
Λ =
0
E(λ1 , −u)Λ . Then Λ is clearly an admissible index-tuple. We have

E(λ1 , u)(λ1 ) = λ1 ,
E(λ1 , u)(λ2 ) = λ02 ,
E(λ1 , u)(k) = k − (k, λ02 )λ1 , k ∈ K(Λ).

The induced action of E(λ1 , u) on Hl (Λ) is simply the translation

E(λ1 , u)Z = Z + (λ02 )K(Λ) (Z ∈ Hl (Λ)).

Hence there is a t > 0 such that E(λ1 , u)St0 (Λ) ⊂ St (Λ). This implies

S̃t0 (Λ0 ) = S̃t0 (E(λ1 , u)Λ) = E(λ1 , u)S̃t0 (Λ) ⊂ S̃t (Λ).

t
u

Lemma 4.18. Let Λ = (λ1 , λ2 , λ3 , λ4 ) be an admissible index-tuple. More-

over, let t0 > 0 and Λ0 be an admissible index-tuple of the form (λ1 , λ2 , λ3 , ∗).
Then there exists a t > 0 such that

S̃t0 (Λ0 ) ⊂ S̃t (Λ).

Proof. This can be proved similarly as Lemma 4.17. t

u
110 4 Some Riemann geometry on O(2, l)

4.3 Modular forms whose zeros and poles lie on

Heegner divisors

Let F be a meromorphic modular form for Γ (L) of weight r. Suppose that

its divisor is a linear combination of Heegner divisors3 H(β, m). Then there
are integral coefficients c(β, m) with c(β, m) = c(−β, m) such that
1 X X
(F ) = c(β, m)H(β, m).
2 0
β∈L /L m∈Z+q(β)
m<0

The function f (Z) = log(|F (Z)|q(Y )r/2 ) is Γ (L)-invariant and has logarith-
mic singularities along Heegner divisors. In this section we show that up to
an additive constant f is equal to the regularized theta integral
1 X X
− c(β, m)Φβ,m (Z).
8 0
β∈L /L m∈Z+q(β)
m<0

In the proof we use a result due to S.-T. Yau on subharmonic functions on a

complete Riemann manifold that satisfy certain integrability conditions.
Let M be a complete connected Riemann manifold. Write ∆ for the
Laplace operator on M and (·, ·) for the inner product on square integrable
differential forms on M .

Proposition 4.19. i) Let u be a smooth function on M . Suppose that u and

∆u are square integrable. Then du is square integrable, too.
ii) The Laplace operator is symmetric in the following sense: Let u and
v be smooth square integrable functions on M . Suppose that ∆u and ∆v are
also square integrable. Then

(∆u, v) = (u, ∆v).

Proof. In the following proof we use some ideas of [Rh] §34, to simplify the
argument in [Ro].
i) Since M is complete, there exists a proper C ∞ function % : M → [0, ∞)
and a constant C > 0 such that

|d%(x)| < C

for all x ∈ M . Such a function can be obtained by regularization from the

function x 7→ d(x, o), where d(x, o) denotes the geodesic distance of x from
a fixed point o ∈ M (cf. [Rh] §34, §15). SFor any r > 0 the set Br = {x ∈
M ; %(x) ≤ r} is compact and M = r>0 Br . Let µ : [0, ∞) → [0, 1]
be a C ∞ function equal to 1 on [0, 1] and 0 on [2, ∞). Then the function
3
For a precise definition of the Heegner divisor H(β, m) see chapter 5.
 4.3 Modular forms with zeros and poles on Heegner divisors 111

σr (x) = µ(%(x)/r) has its values in [0, 1]. Moreover, it satisfies σr (x) = 1 for
x ∈ Br , σr (x) = 0 for x ∈
/ B2r , and

|dσr (x)| < D/r, (4.28)

where D > 0 is a constant which is independent of r.

Because σr has compact support, we have

(∆u, uσr2 ) = (du, σr2 du) + (σr du, 2udσr ).

1
Using the inequality |(f, g)| ≤ 2 (f, f ) + 12 (g, g), which follows from the
Schwarz inequality, we find

(σr du, σr du) ≤ (∆u, uσr2 ) + 12 (σr du, σr du) + 2(udσr , udσr )
(σr du, σr du) ≤ 2(∆u, uσr2 ) + 4(udσr , udσr )
≤ 2(∆u, uσr2 ) + 4D2 r−2 (u, u).

In the last line we have used (4.28). As r → ∞ we obtain (du, du) ≤ 2(∆u, u).
So du is square integrable.
ii) In the same way as in (i), since σr has compact support, we have

(∆u, σr v) = (du, σr dv) + (du, vdσr ).

By (4.28), as r → ∞, the second term on the right hand side tends to zero.
This implies (∆u, v) = (du, dv). t
u
Theorem 4.20. Let u be a harmonic form on M . Suppose that u ∈ Lp (M )
for some p > 1. Then u is closed and coclosed.
Proof. See [Yau] Proposition 1 on p. 663. The case p = 2 is also treated in
[Rh] Theorem 26. u
t
A continuous function u on M is called subharmonic, if it satisfies the local
harmonic maximum principle, i.e. if for any point a ∈ M and any smooth
harmonic function h defined in a neighborhood of a, the function u − h has a
local maximum at a only if it is constant in a neighborhood of a (cf. [GW]).
If u is smooth, then it is subharmonic if and only if ∆u ≥ 0 everywhere on
M . The following theorem is also due to S.-T. Yau (see [Yau] Theorem 3 on
p. 663 and Appendix (ii)).
Theorem 4.21. Let u be a non-negative continuous subharmonic function
on M . Suppose that u ∈ Lp (M ) (p > 1). Then u is constant.
Note that Theorem 4.20 for functions immediately follows from Theorem
4.21. If u is a harmonic function, then |u| is subharmonic in the above sense.
Corollary 4.22. Let M be a complete connected Riemann manifold of finite
volume. Let f ∈ Lp (M ) (p > 1) be a smooth function that satisfies ∆f = c,
where c is a real constant. Then f is constant.
112 4 Some Riemann geometry on O(2, l)

Proof. Assume that f is a solution of ∆f = c. Without any restriction we

may assume that c ≥ 0. So f is subharmonic. Let d > 0 be a large real
number. The function f +d is also subharmonic. Because M has finite volume,
f + d ∈ Lp (M ). Consider the function g = max(f + d, 0). It is continuous,
nonnegative, and belongs to Lp (M ). Moreover, it is subharmonic in the above
sense, since it is the maximum of two subharmonic functions (cf. [GW]).
According to Theorem 4.21, g is constant. This implies that f is constant on
{x ∈ M ; f (x) ≥ −d}. Since d was arbitrary, we find that f is constant on
M.
In the case p = 2 we may also argue as follows: By Proposition 4.19 we
have
c2 vol(M ) = (c, c) = (∆f, c) = (f, ∆c) = 0.
Hence f is a square integrable harmonic function. According to Theorem 4.20
it is constant. t
u
Theorem 4.23. Let F be a meromorphic modular form of weight r with
respect to Γ (L), whose divisor is a linear combination of Heegner divisors
1 X X
(F ) = c(β, m)H(β, m). (4.29)
2 0
β∈L /L m∈Z+q(β)
m<0

Define f (Z) = log(|F (Z)|q(Y )r/2 ) (with Y = =(Z)) and

1 X X
Φ(Z) = − c(β, m)Φβ,m (Z).
8 0
β∈L /L m∈Z+q(β)
m<0

Then f − Φ is constant.
Proof. Let us temporarily assume l ≥ 4. First, we will prove that f − Φ ∈
L2 (K+ /Γ (L)). It suffices to show that
Z
|f − Φ|2 dω < ∞
S̃t (Λ)

for any t > 0 and any admissible index-tuple Λ. According to Lemma 4.16–
4.18 we may assume that Λ = (z, z̃, d, d),˜ where z ∈ L is a primitive norm 0
vector, z ∈ L with (z , z) = 1, z̃ = z − q(z 0 )z, and d is a primitive norm 0
0 0 0 0
⊥
vector in K = L ∩ z ⊥ ∩ z 0 , d0 ∈ K 0 with (d0 , d) = 1, and d˜ = d0 − q(d0 )d.
This is precisely the situation we considered in the previous sections. We may
view F , f , and Φ as functions on Hl = Hl (Λ) and use the Fourier expansion
of Φ which was determined in section 3.2. We have to prove that
Z
dX dY
|f (Z) − Φ(Z)|2 < ∞.
q(Y )l
St
 4.3 Modular forms with zeros and poles on Heegner divisors 113

Define a meromorphic function Ψ on Hl by

Y Y
Ψ (Z) = Ψβ,m (Z)c(β,m)/2 .
β∈L0 /L m∈Z+q(β)
m<0

By Theorem 3.16 the quotient F/Ψ is a holomorphic function on Hl without

any zeros. Let
G = Log(F/Ψ )
be a holomorphic logarithm and g = <(G) = log |F/Ψ |. Since F is a modular
form for Γ (L) (with some multiplier system), the absolute value of F is K-
periodic, i.e. |F (Z + λ)| = |F (Z)| for any λ ∈ K. The Fourier expansions
L
(3.38) of the functions ψβ,m (Z) imply that log |Ψ | is also K-periodic. Thus
g is K-periodic, too. By the maximum modulus principle we may infer that
there is a λ0 ∈ K ⊗ R, such that

G(Z + λ) = G(Z) + 2πi(λ0 , λ)

for any λ ∈ K. Hence the function

G̃(Z) = G(Z) − 2πi(λ0 , Z)

is holomorphic on Hl and K-periodic.

Let σ ∈ Γ (L) ∩ O(K). Then σ acts linearly on Hl and j(σ, Z) = 1. This
implies |F (σZ)| = |F (Z)|. According to Theorem 3.16 we have
1X L


<(G̃(Z)) = log |F (Z)| + 2π(λ0 , Y ) + c(β, m) ψβ,m (Z) − Cβ,m .
8
β,m

L
In view of the Fourier expansion (3.38) of the ψβ,m we find

<(G̃(σZ)) = <(G̃(Z)) + `(Y ),

where `(Y ) is (a priori) a piecewise linear real function of Y . Because <(G̃(Z))

and <(G̃(σZ)) are real analytic, `(Y ) is even linear on Hl . Thus there exists
a %0 ∈ K ⊗ R with

<(G̃(σZ)) = <(G̃(Z)) + 2π(%0 , Y ).

Again, by the maximum modulus principle, there is a real constant C such

that
G̃(σZ) = G̃(Z) − 2πi(%0 , Z) + iC. (4.30)
Since G̃(Z) is K-periodic, G̃(σZ) is K-periodic, too. Using (4.30), one easily
deduces that %0 = 0. If we insert the Fourier expansion
X
G̃(Z) = a(λ)e((λ, Z)), a(λ) ∈ C,
λ∈K 0
114 4 Some Riemann geometry on O(2, l)

of G̃ into (4.30) and compare the constant terms on both sides, we find C = 0
and thereby
G̃(σZ) = G̃(Z).
But now, by the Koecher principle (Proposition 4.15), G̃ is bounded on St .
We obtain

|g(Z)| = |<(G̃(Z)) − 2π(λ0 , Y )| ≤ 2π|(λ0 , Y )| + const. (4.31)

Let us now consider the function

1 X X
L
ξ(Z) = − c(β, m)ξβ,m (Z).
8 0
β∈L /L m∈Z+q(β)
m<0

Using the asymptotic behavior (3.26) of the function Vκ , the boundedness

(2.22) of the coefficients b(γ, n) for n < 0, and Lemma 4.13, one easily obtains
that the sum over λ ∈ K 0 in the definition of ξβ,mL
(Z) (Def. 3.11) is bounded
K
on St . By means of the Fourier expansion (3.12) of ξβ,m it can be deduced
that there is a ν ∈ K ⊗ R with
K
|Y | ξβ,m (Y /|Y |) ≤ |(ν, Y )|

on St . Hence there exists a ν0 ∈ K ⊗ R such that

|ξ(Z)| ≤ |B log q(Y )| + |(ν0 , Y )| + const. (4.32)

on St , where
1 X X
B=− c(β, m)bβ,m (0, 0).
8 0
β∈L /L m∈Z+q(β)
m<0

By virtue of (4.31) and (4.32) we find

|f (Z) − Φ(Z)| = log |F (Z)/Ψ (Z)| − ξ(Z) + 2r log(q(Y ))

≤ |g(Z)| + |ξ(Z)| + | 2r log(q(Y ))|
|r|
 |(λ0 , Y )| + |(ν0 , Y )| + (|B| + 2 )| log(q(Y ))| (4.33)

on St . Using Lemma 4.11 and Lemma 4.12, we obtain that f − Φ ∈

L2 (K+ /Γ (L)).
Let us now consider the action of the invariant Laplace operator Ω on
f − Φ. According to Theorem 4.7 we have
l
ΩΦ(Z) = B.
4
Moreover, Lemma 4.1 implies that Ωf (Z) = − rl
8 . Hence
 4.3 Modular forms with zeros and poles on Heegner divisors 115

rl l
Ω(f (Z) − Φ(Z)) = − − B
8 4
is constant. Since Hl /Γ (L) is a complete Riemann manifold of finite volume,
we may apply Corollary 4.22. We find that f − Φ is constant.
In the case l = 3 the same argument shows that f − Φ ∈ Lp (K+ /Γ (L))
for p < 2. Again, by Corollary 4.22 it has to be constant. t
u
Corollary 4.24. Let F be a meromorphic modular form of weight r with
respect to Γ (L). Suppose that its divisor is a linear combination of Heegner
divisors as in (4.29). Then
1 X X
r=− c(β, m)q(β, −m),
4 0
β∈L /L m∈Z+q(β)
m<0

where the q(γ, n) denote the Fourier coefficients of the Eisenstein series E
with constant term 2e0 in Mκ,L .
Proof. We use the same notation as in the proof of Theorem 4.23. By Theo-
rem 4.23 we know that f − Φ is constant. Thus
r
<(G̃(Z)) − 2π(λ0 , Y ) − ξ(Z) + 2 log(q(Y )) (4.34)

is constant. Comparing coefficients we find that the 0-th Fourier coefficient

|Y | X X
K
−2π(λ0 , Y ) + √ c(β, m)ξβ,m (Y /|Y |) + ( 2r + B) log(q(Y ))
8 2 β∈L0 /L m∈Z+q(β)
m<0

of (4.34) is constant.
This implies in particular that the function
|Y | X X
K
−2π(λ0 , Y /|Y |) + √ c(β, m)ξβ,m (Y /|Y |)
8 2 β∈L0 /L m∈Z+q(β)
m<0

is invariant under the action of the real orthogonal group O+ (K ⊗ R). Hence
it is constant.
We may infer that there are constants C, C 0 ∈ R such that

C|Y | + ( 2r + B) log(q(Y )) = C 0 .
r
But this is only possible, if C = 0 and 2 = −B. Using Proposition 1.16, we
obtain the assertion. t
u
Remark 4.25. Let β ∈ L0 /L and m ∈ Z + q(β) with m < 0. If there is no
λ ∈ L0 with (λ, q(λ)) = (β, m), then H(β, m) = 0 in the divisor group of
Hl /Γ (L). In this case Theorem 4.23 implies that Φβ,m is constant.
5 Chern classes of Heegner divisors

Let X be a normal irreducible complex space. By a divisor D on X we mean

a formal linear combination
X
D= nY Y (nY ∈ Z)

of irreducible closed
S analytic subsets Y of codimension 1 such that the sup-
port supp(D) = nY 6=0 Y is a closed analytic subset of everywhere pure
codimension 1. Then for any compact subset K ⊂ X there are only finitely
many Y with Y ∩ K 6= ∅ and nY 6= 0. We denote the group of divisors on X
by D(X).
If Γ is a group of biholomorphic transformations of X acting properly
discontinuously, we may consider the inverse image π ∗ (D) of a divisor D
on X/Γ under the canonical projection π : X → X/Γ . For any irreducible
component Y of the inverse image of supp(D) the multiplicity of Y with
respect to π ∗ (D) equals the multiplicity of π(Y ) with respect to D. Then
π ∗ (D) is a Γ -invariant divisor on X. (So we do not take account of possibly
occurring ramification in the definition of π ∗ .)
Now let X = Hl be the generalized upper half plane as in section 3.2, and
let Γ be the orthogonal group Γ (L) or a subgroup of finite index. If β ∈ L0 /L
and m ∈ Z + q(β) with m < 0, then
X
λ⊥
λ∈β+L
q(λ)=m

is a Γ -invariant divisor on Hl with support H(β, m) (see (2.17)). It is the

inverse image under the canonical projection of an algebraic divisor on X/Γ ,
which we call the Heegner divisor of discriminant (β, m). For simplicity we
denote it by H(β, m), too. The multiplicities of all irreducible components
equal 2, if 2β = 0, and 1, if 2β 6= 0. Notice that H(β, m) = H(−β, m).
We will use the following modified divisor class group f Cl(X/Γ ): Any
meromorphic modular form f with multiplier system χ with respect to Γ
defines via its zeros and poles a Γ -invariant divisor in D(X), which is the
inverse image of an algebraic divisor (f ) in D(X/Γ ). We denote the subgroup
generated by these divisors (f ) by H(X/Γ
e ) and put
118 5 Chern classes of Heegner divisors

Cl(X/Γ ) = D(X/Γ ) H(X/Γ
f e ).

Moreover, we write Cl(X/Γ ) for the quotient of D(X/Γ ) modulo the sub-
group of divisors coming from meromorphic modular forms of weight 0 with
trivial character. If Γ acts freely, then Cl(X/Γ ) coincides with the usual
notion of the divisor class group.

We now give an algebraic interpretation of Theorem 3.22. First we need

a rational structure on the space Sκ,L of cusp forms of weight κ = 1 + l/2.
Throughout this section let

N = min{n ∈ N; nq(γ) ∈ Z for all γ ∈ L0 }

denote the level of the lattice L. The representation %L is trivial on

Γ (N ) = {(M, ϕ) ∈ Mp2 (Z); M ≡ 1 (mod N )}

and thereby factors through the finite group


Mp2 (Z/N Z) = Mp2 (Z) Γ (N ).
P
If f = γ∈L0 /L eγ fγ ∈ Sκ,L , then each component fγ lies in the space Sκ (N )
of cusp forms of weight κ for the group Γ (N ). In other words

Sκ,L ⊂ Sκ (N ) ⊗ C[L0 /L].

Let ζN be a primitive N -th root of unity. It is easily seen that the co-
efficients of %L are contained in Q(ζN ). Indeed, it suffices to check this on
the generators T and S. For T the assertion is trivial. For S one uses the
evaluation of the Gauss sum
X p √ b+ −b−
e(q(γ)) = |L0 /L| i
γ∈L0 /L

due to Milgram (cf. [Bo2] Corollary 4.2). The following Lemma is a slight
improvement of [Bo3] Lemma 4.2.

Lemma 5.1. If κ is integral, then the space Sκ,L has a basis of cusp forms
whose Fourier coefficients all lie in Z[ζN ]. If κ is half-integral, then Sκ,L has
a basis of cusp forms whose coefficients all lie in Z[ζN 0 ], where N 0 is the least
common multiple of N and 8.

Proof. First, assume that κ is integral. We use the following well known
results on integral weight modular forms:
i) The space Sκ (N ) has a basis of cusp forms with coefficients in Z[ζN ]
(cf. [Sh] section 5 or [DI] Corollary 12.3.9).
ii) If f ∈ Sκ (N ) with coefficients in Z[1/N, ζN ] and (M, ϕ) ∈ Γ (N ),
then f |κ (M, ϕ) also lies in Sκ (N ) and has coefficients in Z[1/N, ζN ]. Here
 5 Chern classes of Heegner divisors 119

f |κ (M, ϕ) denotes the usual Petersson operator of weight κ acting on func-

tions H → C. This follows from the q-expansion principle (see [Ka] §1.6 or
[DI] section 12.3).
Consider the trace map
X
tr : Sκ (N ) ⊗ C[L0 /L] → Sκ,L , f 7→ f |∗κ (M, ϕ).
(M,ϕ)∈Mp2 (Z/N Z)

It is obviously surjective. The image of f = γ∈L0 /L eγ fγ ∈ Sκ (N ) ⊗ C[L0 /L]

P
can also be written in the form
X X
tr(f ) = %∗L (M, ϕ)−1 eγ fγ |κ (M, ϕ).
(M,ϕ)∈Mp2 (Z/N Z) γ∈L0 /L

If f has coefficients in Z[ζN ], then by (ii) tr(f ) has coefficients in Q(ζN ).

Hence by virtue of (i) we may infer that Sκ,L has a set of generators with
coefficients in Q(ζN ). After multiplying by a common denominator, we obtain
a basis with coefficients in Z[ζN ].
Now assume that κ is half integral. We claim that (i) and (ii) also hold
in the half integral weight case with N replaced by N 0 .
To prove the claim we argue similarly as in [SeSt] §5. Let θ(τ ) =
πin2 τ
P
n∈Z e be the Jacobi theta series of weight 1/2. It is well known that
θ ∈ S1/2 (8). Moreover, for any (M, ϕ) ∈ Mp2 (Z) the Fourier coefficients of
θ|1/2 (M, ϕ) lie in Z[ζ8 ]. The space Sκ (N ) can be embedded into S2κ (N 0 ) by
multiplying with θ2κ . Hence it suffices to show that the image can be defined
by linear equations with coefficients in Q(ζN 0 ). Then the claim follows from
(i), (ii) applied to the space S2κ (N 0 ).
Since θ has no zeros on H, a cusp form g belongs to the image if and
only if it vanishes with prescribed multiplicities at the cusps of Γ (N 0 ), i.e. if
certain coefficients of g in the expansions at the various cusps vanish. By (ii)
these conditions are given by linear equations over Q(ζN 0 ).
Now we can proceed as in the integral weight case. t
u

Let f ∈ Sκ,L be a cusp form with Fourier coefficients c(γ, n) and σ ∈

Gal(C/Q) an automorphism of C. Then we define the σ-conjugate of f by
X X
f σ (τ ) = cσ (γ, n)eγ (nτ ),
γ∈L0 /L n∈Z−q(γ)

where cσ (γ, n) denotes the conjugate of c(γ, n). If κ is integral (resp. half-
integral) and σ ∈ Gal(C/Q(ζN )) (resp. σ ∈ Gal(C/Q(ζN 0 ))), then Lemma
5.1 implies that f σ is also a cusp form in Sκ,L . For general σ ∈ Gal(C/Q) we
only have f σ ∈ Sκ (N ) ⊗ C[L0 /L]. In this context it would be interesting to
know, if Sκ,L always has a basis with rational Fourier coefficients.
Let
Sκ,L = Gal(C/Q) · Sκ,L
120 5 Chern classes of Heegner divisors

be the space of all Galois conjugates of the elements of Sκ,L and analogously
Mκ,L = Gal(C/Q) · Mκ,L . If R is a subring of C then we write Sκ,L (R)
resp. Sκ,L (R) for the R-module of cusp forms in Sκ,L resp. Sκ,L , whose coef-
ficients all lie in R. Observe that Lemma 5.1 implies
(
[Gal(Q(ζN )/Q) · Sκ,L (Q(ζN ))] ⊗ C, for κ ∈ Z,
Sκ,L =
[Gal(Q(ζN 0 )/Q) · Sκ,L (Q(ζN 0 ))] ⊗ C, for κ ∈ 21 Z.
In particular Sκ,L is finite dimensional.
By construction Sκ,L has a basis of cusp forms whose coefficients are
rational integers. Hence Sκ,L (Z) ⊗Z C = Sκ,L . We will also consider the dual
∗
Z-module Sκ,L (Z) of Sκ,L (Z). Special elements are the functionals
X
aβ,m : Sκ,L (Z) → Z; f= c(γ, n)eγ (nτ ) 7→ aβ,m (f ) = c(β, m)
γ,n

(β ∈ L0 /L and m ∈ Z − q(β), m > 0). We write Aκ,L P (Z) for the submodule
∗
of Sκ,L (Z) consisting of finite linear combinations β,m c(β, m)aβ,m , where
c(β, m) ∈ Z and c(β, m) = c(−β, m). It is easily seen that Aκ,L (Z) has finite
∗
index in Sκ,L (Z) and
∗
Aκ,L (Z) ⊗Z C = Sκ,L ,
∗
where Sκ,L denotes the C-dual space of Sκ,L .
1
Theorem 5.2. Let XL = Hl /Γ (L). The assignment aβ,m 7→ 2 H(β, −m)
defines a homomorphism
η : Aκ,L (Z) −→ f
Cl(XL ). (5.1)
Proof. We have to show that η is well defined. Suppose that
X X
a= c(β, m)aβ,m
β∈L0 /L m∈Z−q(β)
m>0

(c(β, m) ∈ Z and c(β, m) = c(−β, m)) is a finite linear combination which is

equal to 0 in Aκ,L (Z). Then we may also consider a as an element of the dual
space of Sκ,L . According to Theorem 1.17 there exists a nearly holomorphic
modular form f of weight k = 1 − l/2 with principal part
X X
c(β, m)eβ (−mτ ).
β∈L0 /L m∈Z−q(β)
m>0

Now Theorem 3.22 implies that there is a meromorphic modular form of

rational weight c(0, 0)/2 for Γ (L) with divisor
1 X X
c(β, m)H(β, −m),
2 0
β∈L /L m∈Z−q(β)
m>0

in other words η(a) = 0 in f

Cl(XL ). t
u
 5 Chern classes of Heegner divisors 121

As before, let X = Hl be the generalized upper half plane and Γ be

the orthogonal group Γ (L) or a subgroup of finite index. Write XL for the
quotient Hl /Γ (L).
Let us briefly recall some basic facts on Chern classes and the cohomology
of XL . For any divisor D on X/Γ one has a corresponding sheaf L(D). The
sections of L(D) over an open subset U ⊂ X/Γ are meromorphic functions
f with (f ) ≥ −D on U .
We now temporarily assume that Γ acts fixed point freely on X. Then
X/Γ is an analytic manifold and every divisor D on X/Γ a Cartier divisor,
i.e. L(D) is a line bundle. The Chern class

c(D) = c(L(D)) ∈ H 2 (X/Γ, C)

of L(D) can be constructed as follows: One chooses a meromorphic function f

on X such that (f ) equals the inverse image π ∗ (D) of D under the canonical
projection π. Then
f (γ(z))
J(γ, z) = (γ ∈ Γ )
f (z)
is an automorphy factor, i.e. a 1-cocycle of Γ in the ring of holomorphic
invertible functions on X. Hence a Hermitean metric on the bundle L(D) is
given by a positive C ∞ -function h : X → R with

h(γz) = |J(γ, z)|h(z) for all γ ∈ Γ .

Then the differential form ω = ∂ ∂¯ log(h) is Γ -invariant and closed. It defines

(via de Rham isomorphism) a cohomology class in H 2 (X/Γ, C). This is the
Chern class of D in the case that Γ acts fixed point freely on X.
In the general case one chooses a normal subgroup Γ0 ≤ Γ of finite in-
dex and obtains the Chern class c(D) by the isomorphism H 2 (X/Γ, C) ∼ =
H 2 (X/Γ0 , C)Γ/Γ0 .
The construction of the Chern class gives rise to a homomorphism

c : Cl(XL ) −→ H 2 (XL , C)

into the second cohomology. Using the results of the previous chapters, we
may determine the images of the divisors H(β, m) explicitly.
Before stating the theorem let us give an easy example. Let F : Hl → C
be a modular form of weight r with respect to Γ (L). Then |F (Z)|q(Y )r/2
is invariant under Γ (L). We may view F as a trivialization of the inverse
image L(π ∗ ((F ))) of the sheaf attached to (F ) ∈ D(XL ), and q(Y )−r/2 as
a Hermitean metric on L((F )). Hence the Chern class of the divisor (F ) is
given by
r
c((F )) = − ∂ ∂¯ log q(Y ),
2
a constant multiple of the Kähler class.
122 5 Chern classes of Heegner divisors

Theorem 5.3. The (1, 1)-form

1 ¯ L
hβ,m (Z) = ∂ ∂ξβ,m (Z)
4
is a representative of the Chern class c(H(β, m)) of the Heegner divisor
H(β, m). It has the Fourier expansion

|Y | K b(0, 0) ¯
hβ,m (Z) = ∂ ∂¯ √ ξβ,m (Y /|Y |) − ∂ ∂ log(Y 2 )
4 2 4
1 X X
+ √ b(δ, q(λ))
2 π 0 0
λ∈K δ∈L0 /L
q(λ)<0 p(δ)=λ+K
X1
× ¯ 2−k (πn|λ||Y |, πn(λ, Y )) e(n(λ, X)).
e(n(δ, z 0 ))∂ ∂V
n
n≥1

Here b(γ, n) = b(γ, n, 1 − k/2) denote the Fourier coefficients of the Poincaré
L
series Fβ,m (τ, 1 − k/2) as in Proposition 1.10.

Proof. By Theorem 3.16 the function Ψβ,m is meromorphic on Hl . Its divisor

equals π ∗ (H(β, m)). Since ξβ,m
L
is real valued and real analytic, the function
L
eξβ,m /4 is real analytic and positive. Moreover, by construction
L L
|Ψβ,m (Z)|e−ξβ,m (Z)/4 = eCβ,m /4 e−Φβ,m (Z)/4
L
is invariant under Γ (L). Thus eξβ,m /4 defines a Hermitean metric on the sheaf
L(H(β, m)). This implies the assertion. t
u

Remark 5.4. The above theorem, together with Theorem 2.12 on the singu-
larities of Φβ,m (Z), and Theorem 3.9 on its Fourier expansion, imply that
Φβ,m (Z) is something as a Green current for the divisor H(β, m) in the sense
of Arakelov geometry (see [SABK]). More precisely this function should define
a Green object with log-log-growth in the extended arithmetic intersection
theory due to Burgos, Kramer, and Kühn [BKK]. It will be interesting to
investigate this connection in future.

5.1 A lifting into the cohomology

Denote by H̃ 2 (XL , C) the quotient

H̃ 2 (XL , C) = H 2 (XL , C) C∂ ∂¯ log q(Y )



of the second cohomology H 2 (XL , C) and the span of the Kähler class
∂ ∂¯ log q(Y ). The Chern class map Cl(XL ) → H 2 (XL , C) induces a homo-
morphism
 5.1 A lifting into the cohomology 123

Cl(XL ) −→ H̃ 2 (XL , C).

c:f
If we combine c with the map η constructed in Theorem 5.2, we obtain a
homomorphism
Aκ,L (Z) −→ H̃ 2 (XL , C).
After tensoring with C one gets a linear map
∗
Sκ,L −→ H̃ 2 (XL , C), (5.2)

which is characterized by aβ,m 7→ 12 c(H(β, −m)). The image is the subspace

spanned by the classes of Heegner divisors. Define a function Ω(τ ) by
1 X X
Ω(τ ) = c(H(β, −m))eβ (mτ ). (5.3)
2 0
β∈L /L m∈Z−q(β)
m>0

Then Ω(τ ) is a priori a formal power series with coefficients in the subspace
of H̃ 2 (XL , C) spanned by the Heegner divisors. It is a consequence of the
existence of the Borcherds lift (Theorem 3.22), Serre duality (as in Theorem
1.17), and Lemma 5.1 that Ω(τ ) is in fact a modular form in Mκ,L with
values in H̃ 2 (XL , C). This argument is due to Borcherds (see [Bo3]). Hence
Ω(τ ) can be viewed as a kernel function for the map (5.2). The image of a
∗
functional a ∈ Sκ,L is given by

a(Ω(τ )) ∈ H̃ 2 (XL , C).

The purpose of this section is to refine the map (5.2) and to describe it
more precisely. First we will replace the image space H̃ 2 (XL , C) by a certain
space of harmonic (1, 1)-forms.
Let fClH (XL ) denote the subgroup of f Cl(XL ) generated by the classes of
the Heegner divisors H(β, m). Write

H1,1 (XL ) (5.4)

for the vector space of square integrable harmonic (1, 1)-forms on XL . Ac-
cording to [Bl3] (Theorem 6.2 and §7) H1,1 (XL ) is a finite dimensional space
of automorphic forms in the sense of [Ha]. In particular its elements are Z-
finite, where Z denotes the center of the universal enveloping algebra of the
Lie algebra of O(V ). By Theorem 4.20 any square integrable harmonic form
on a complete Riemann manifold is closed and thereby defines a cohomology
class via de Rham isomorphism.
Recall the basic facts on the Riemann geometry of Hl summarized in
section 4.1. We saw that the l × l matrix
∂2
 
1
h(Z) = h(Y ) = − log(q(Y ))
4 ∂yi ∂yj i,j
124 5 Chern classes of Heegner divisors

defines the (up to a constant multiple unique) O+ (V )-invariant Hermitean

metric
on Hl and thereby a Hermitean metric on XL . Let h−1 (Y ) =
hij (Y ) i,j be the inverse of h(Y ). A computation shows that
 
y1 y1 . . . . . . y1 yl  
 .. ..  0 −2q(Y )
 . .  −2q(Y ) 0 
   
h−1 (Y ) = 4  .. . .
..  + 2  q(Y ) .
   

 .
  .. 
. .
 .. .. 
  
q(Y )
y l y1 . . . . . . y l y l

The natural scalar product for (1, 1)-forms on XL is given by

Z
dX dY
(f (Z) dzi ∧ dz̄j , g(Z) dzm ∧ dz̄n ) = 4 f (Z)g(Z)him (Y )hjn (Y ) .
q(Y )l
XL

On the Siegel domain St in Hl (see Definition 4.9) the components of h−1 (Y )

satisfy the estimate

hij (Y )  q(Y ), if (i, j) 6= (2, 2).

The Kähler form ∂ ∂¯ log q(Y ) is harmonic by (4.3). We leave it to the reader
to show that it is square integrable. We put
e 1,1 (XL ) = H1,1 (XL ) C∂ ∂¯ log q(Y ).

H (5.5)

Notice that the natural map H e 1,1 (XL ) → H̃ 2 (XL , C) is in general not injec-
tive.
The following strengthening of Theorem 5.3 is crucial for our argument.

Theorem 5.5. i) The (1, 1)-form hβ,m (Z) = 14 ∂ ∂ξ ¯ L (Z) is a harmonic

β,m
square integrable representative of the Chern class of H(β, m).
ii) The assignment
H(β, m) 7→ hβ,m (Z)
e 1,1 (XL ) such that the diagram
ClH (XL ) → H
defines a homomorphism f

ClH (XL ) /He 1,1 (XL )

LLL
f
LLL
c LLL
L& 
H̃ 2 (XL , C)

commutes.
 5.1 A lifting into the cohomology 125

¯ L (Z) = ∂ ∂Φ
Proof. i) Since ∂ ∂ξ ¯ L (Z), the (1, 1)-form ∂ ∂ξ ¯ L (Z) is in-
β,m β,m β,m
L
variant under the action of Γ (L). By Theorem 4.7 we know that Ωξβ,m is
constant. Because XL is Kählerian, the Laplace operator commutes with ∂
¯ Thus ∂ ∂ξ
and ∂. ¯ L is harmonic.
β,m
We now prove that ∂ ∂ξ¯ L is square integrable. In the same way as in the
β,m
proof of Theorem 4.23 it suffices to show that ∂ ∂ξ¯ L is square integrable on
β,m
any Siegel domain St (Λ) (where Λ is an admissible index tuple and t > 0).
By Lemmas 4.16–4.18 we may assume that Λ = (z, z̃, d, d), ˜ where z ∈ L is
a primitive norm 0 vector, z ∈ L with (z , z) = 1, z̃ = z 0 − q(z 0 )z, and d
0 0 0
⊥
is a primitive norm 0 vector in K = L ∩ z ⊥ ∩ z 0 , d0 ∈ K 0 with (d0 , d) = 1,
⊥
and d˜ = d0 − q(d0 )d. We put D = K ∩ d⊥ ∩ d0 and use the coordinates on
Hl = Hl (Λ) introduced in section 4.1 on page 93.
We consider the Fourier expansion of ∂ ∂ξ ¯ L (see Theorem 5.3). The
β,m
K
constant term involves the function ξβ,m (Y /|Y |) (see Definition 3.3). Here the
contribution ΦDβ,m can be evaluated by Theorem 2.14. If we put everything
¯ L (Z) is equal to
together we find that ∂ ∂ξ β,m

2 X X X1
√ b(δ, q(λ)) e(n(δ, z 0 ))
π 0
n
λ∈K δ∈L00 /L n≥1
q(λ)<0 p(δ)=λ+K
¯ κ (πn|λ||Y |, πn(λ, Y )) e(n(λ, X))
× ∂ ∂V
λ2 YD2
 
¯ 2
X
¯ 1 2
− b(0, 0)∂ ∂ log(Y ) + 4 π∂ ∂ (λ, YD ) −
0
y1 l−2
λ∈D
λ+D=p2 (β)
q(λ)=m
 κ−1
4 X X X 1
¯ πn|λ||Y |
+ b(δ, q(λ)) ∂ ∂y1
π n2 y1
λ∈D 0 −0 δ∈K00 /K n≥1
p(δ)=λ+D
   
2πn|λ||Y | (λ, YD )
× Kκ−1 e n + n(δ, d0 ) . (5.6)
y1 y1

(Since the computation is similar to the argument in the proof of Theorem

5.8 we omit the details.)
According to (2.22) the coefficients b(δ, n) satisfy

b(δ, n) = O(1), n → −∞. (5.7)

Moreover, we will use the following estimates. Let ε > 0 and a, b ∈ N0 . Then
 a  b √
∂ ∂ A2 +B 2
Vκ (A, B) ε,a,b e−2(1−ε) (5.8)
∂A ∂B
√
uniformly on A2 + B 2 > ε (compare (3.26)), and
126 5 Chern classes of Heegner divisors
 a
∂
Kκ−1 (y) ε,a e−y (5.9)
∂y
uniformly on y > ε (see [AbSt] chapter 9).
By means of (5.7), (5.8), and Lemma 4.13 one can show that the first sum
over λ ∈ K 0 in (5.6) is rapidly decreasing on St . Hence it is square integrable
over St . The term b(0, 0)∂ ∂¯ log(Y 2 ) is a multiple of the Kähler form. It is also
square integrable.
We now prove that the third term in (5.6) is square integrable over St . For
l = 3 it vanishes identically. Hence we may temporarily assume that l ≥ 4. It
suffices to show that
∂2 1 λ2 YD2
 
2
(λ, YD ) − dzi ∧ dz̄j
∂yi ∂yj y1 l−2
is square integrable for all i, j = 1, . . . , l. On St the estimate
(
∂2 1 2 2
 
λ Y 1/y1 , if i 6= 2 and j 6= 2,
(λ, YD )2 − D

∂yi ∂yj y1 l−2 0, if i = 2 or j = 2,

holds. Thus it suffices to show that the integral

Z
dX dY
I = y1−2 hii (Y )hjj (Y )
q(Y )l
St

is finite for all i 6= 2 and j 6= 2. We have

Z Z
−2 2 dX dY dX dY
I  y1 q(Y )  y22 .
q(Y )l q(Y )l
St St

The latter integral is finite by Lemma 4.12.

Let us now consider the last term in (5.6). We prove that for any i, j =
1, . . . , l the (1, 1)-form
X 1 ∂2  κ−1
X X πn|λ||Y |
fij dzi ∧ dz̄j = b(δ, q(λ)) y 1
0 0
n2 ∂yi ∂yj y1
λ∈D −0 δ∈K0 /K n≥1
p(δ)=λ+D
   
2πn|λ||Y | (λ, YD ) 0
× Kκ−1 e n + n(δ, d ) dzi ∧ dz̄j (5.10)
y1 y1
is square integrable over St . There is a δ > 0 such that

|λ||Y |/y1 > δ (5.11)

for all Z ∈ St and all λ ∈ D0 − 0. Let b ∈ Z. It is a consequence of (5.9) and

(5.11) that there exists a polynomial P (Y, λ) of degree r in Y and degree r0
in λ such that
 5.1 A lifting into the cohomology 127
κ−1
∂2
  
|λ||Y | 2π|λ||Y |
y1 Kκ−1 e((λ, YD )/y1 )
∂yi ∂yj y1 y1
b
y12

|P (Y, λ)|
 e−π|λ||Y |/y1
y1r+1 |λ2 |Y 2

for all Z ∈ St and all λ ∈ D0 − 0. Using (4.22) we find that the right hand
side satisfies the estimate
b
y12

|P (Y, λ)|
e−π|λ||Y |/y1
y1r+1 |λ2 |Y 2
 b−r  b−r0 /2
1 |P (Y, λ)| y1 1
 e−π|λ||Y |/y1
y1 y2r |λ|r0 y2 |λ2 |
 b−r  b−r0 /2
1 y1 1
 e−π|λ||Y |/y1 .
y 1 y2 |λ2 |

Hence the series in (5.10) has (up to a multiplicative constant) the majorant
0
X X 1 1  y1 b−r  1 b−r /2
2 y 2 |λ2 |
e−πn|λ||Y |/y1
0
n 1 y 2 n
λ∈D −0 n≥1

on St . If we choose b large enough a priori, then this is bounded by y1 /y22 on

St . Thus we find
Z Z 2
dX dY y1 ii dX dY
|fij (Z)|2 hii (Y )hjj (Y )  h (Y )hjj (Y )
q(Y )l y24 q(Y )l
St St
Z
dX dY
 q(Y ) .
q(Y )l
St

The latter integral is finite for l ≥ 3 by Lemma 4.11. This concludes the proof
of the first assertion.
ii) In order to prove that the above homomorphism is well defined it
suffices to show: For any meromorphic modular form F whose divisor
1 X X
(F ) = c(β, m)H(β, m)
2 0
β∈L /L m∈Z+q(β)
m<0

(c(β, m) ∈ Z with c(β, m) = c(−β, m)) is a linear combination of Heegner

divisors, the corresponding linear combination
X
¯ L (Z)
c(β, m)∂ ∂ξ β,m
β,m
128 5 Chern classes of Heegner divisors

e 1,1 (XL ).
equals zero in H
By Theorem 4.23 we have
  1X
log |F (Z)|q(Y )r/2 = − c(β, m)ΦL
β,m (Z),
8
β,m

where r denotes the weight of F . This implies that

X X
¯ L (Z) =
c(β, m)∂ ∂ξ ¯ L (Z)
c(β, m)∂ ∂Φ
β,m β,m
β,m β,m
 
= −8∂ ∂¯ log |F (Z)|q(Y )r/2
= −4r∂ ∂¯ log (q(Y ))

is a multiple of the Kähler class. t

u

We now consider the generating series

1 ¯ 1 X X
Ω L (Z, τ ) = ∂ ∂ log(Y 2 )e0 + hβ,−m (Z)eβ (mτ ) (5.12)
4 2 0
β∈L /L m∈Z−q(β)
m>0

of the special representatives hβ,−m (Z) of the Chern classes of the Heegner
divisors. (The presence of the constant term will become clear in the proof of
Theorem 5.8.) By a similar argument as in the proof of Theorem 5.5 it can
be shown that Ω L (Z, τ ) converges normally and is square integrable in Z.
Hence we may consider it as an element of H1,1 (XL ) ⊗ C[L0 /L] for fixed τ .

In the rest of this section we shall show that for fixed Z the components
of Ω L (Z, τ ) are contained in the space Mκ,L . Thus Ω L (Z, τ ) can be viewed
as the kernel function of a lifting

ϑ : Sκ,L −→ H1,1 (XL ); f 7→ f (τ ), Ω L (Z, τ ) τ

from elliptic cusp forms to H1,1 (XL ). The idea of the proof is the same as
in [Za]: We write Ω L (Z, τ ) as an explicit linear combination of Poincaré and
Eisenstein series in Mκ,L . This description is crucial for our further argument.
It can be used to evaluate the scalar product of Ω L (Z, τ ) and a cusp form
f ∈ Sκ,L explicitly. Thereby one obtains the Fourier expansion of ϑ(f ) in
terms of the Fourier coefficients of f . The map ϑ is one possible generalization
of the Doi-Naganuma map [DN, Na, Za]. In the O(2, 3)-case of Siegel modular
forms of genus 2 such a generalization was given by Piateskii-Shapiro using
representation theoretic methods ([PS1, PS2], see also [Wei]).

For technical reasons we first introduce two linear operators between the
spaces Sκ,L and Sκ,K where K is a sublattice of L as in section 2.1.
 5.1 A lifting into the cohomology 129
P
Lemma 5.6. Let F = γ∈K 0 /K eγ fγ be a modular form in Mκ,K . Define a
0
function F |↑L
K : H → C[L /L] by
X
F |↑L
K= eβ fp(β) .
β∈L00 /L

Then F |↑L
K ∈ Mκ,L .

0
Proof. Denote the components of F |↑L K by gβ (β ∈ L /L). It suffices to check
the transformation behavior for the generators T , S of Mp2 (Z). For T this is
immediately verified. Regarding S we have to show that
√ 2κ √ l−2 X
τ i
gβ (−1/τ ) = p e((β, δ))gδ (τ ).
0
|L /L| δ∈L0 /L

The right hand side obviously equals

√ 2κ √ l−2
τ i X
p e((β, δ))fp(δ) (τ ). (5.13)
0
|L /L| δ∈L00 /L

A set of representatives for L00 /L is given by λ = γ − (γ, ζ)z/N + bz/N , where

γ runs through a set of representatives for K 0 /K and b runs modulo N . We
find
√ 2κ √ l−2 X
τ i X
(5.13) = p e ((β, γ − (γ, ζ)z/N )) fγ (τ ) e((β, z)b/N ).
|L0 /L| γ∈K 0 /K b (N )

/ L00 /L, and N if β ∈ L00 /L. In the second case we

The latter sum is 0 if β ∈
obtain
√ 2κ √ l−2 X
τ i
(5.13) = p e ((p(β), γ)) fγ (τ )
0
|K /K| γ∈K 0 /K
= fp(β) (−1/τ ) = gβ (−1/τ ).

t
u
P
Lemma 5.7. Let F = β∈L0 /L eβ fβ be a modular form in Mκ,L . Define a
0
function F |↓L
K : H → C[K /K] by
X X
F |↓L
K= eγ fδ .
γ∈K 0 /K δ∈L00 /L
p(δ)=γ

Then F |↓L
K ∈ Mκ,K .
130 5 Chern classes of Heegner divisors

Proof. This can be proved in the same way as Theorem 2.6. t

u

It is easily seen that the operators ↑L L

K and ↓K are adjoint with respect to
the Petersson scalar product. Moreover, they take cusp forms to cusp forms.

For the rest of this section we fix a decomposition of the lattice L. Let
z ∈ L be a primitive norm 0 vector and define z 0 , N , L00 , p as in section
2.1. Then the lattice K = L ∩ z 0⊥ ∩ z ⊥ has signature (1, l − 1) and L ⊗ R =
(K ⊗ R) ⊕ Rz 0 ⊕ Rz. We assume that there also exists a primitive norm 0
vector d ∈ K and define d0 , M , K00 , p analogous to z 0 , N , L00 , p. Then the
lattice D = K ∩ d0⊥ ∩ d⊥ is negative definite and K ⊗ R = (D ⊗ R) ⊕ Rd0 ⊕ Rd.
If λ ∈ L ⊗ R then we write λK resp. λD for the orthogonal projection of λ to
K ⊗ R resp. D ⊗ R. Recall that k = 1 − l/2 and κ = 1 + l/2.

Theorem 5.8. The function Ω L (Z, τ ) defined in (5.12) can be written in

the form
1 ¯
Ω L (Z, τ ) = Ω K (Y, τ ) |↑L 2 L
K + ∂ ∂ log(Y )E (τ )
8
1 X X X1
− √ e(n(δ, z 0 ))
4 πΓ (κ − 1) 0 0
n
λ∈K δ∈L0 /L n≥1
q(λ)<0 p(δ)=λ+K

¯ κ (πn|λ||Y |, πn(λ, Y )) e(n(λ, X))P L

× ∂ ∂V δ,−q(λ) (τ ), (5.14)

where
1 X X
¯ |ξ K (Y /|Y |)eβ (mτ ).
Ω K (Y, τ ) = √ ∂ ∂|Y β,−m (5.15)
8 2 β∈K 0 /K m∈Z−q(β)
m>0

The function Ω K (Y, τ ) is equal to

¯ Y )−1 ΘD (YD , τ ) |↑K − 1 ¯ Y)

X X X 1
π∂ ∂(d, D ∂ ∂(d,
2πΓ (κ − 1) n2
0 λ∈D −0 δ∈K00 /K n≥1
p(δ)=λ+D
 κ−1    
πn|λ||Y | 2πn|λ||Y | (λ, Y ) 0 K
× Kκ−1 e n + n(δ, d ) Pδ,−q(λ) (τ ).
(d, Y ) (d, Y ) (d, Y )
(5.16)

Here ΘD (YD , τ ) ∈ Sκ,D denotes the theta series

X  λ2 YD2

D 1 X 2
Θ (YD , τ ) = (λ, YD ) − eβ (−q(λ)τ )
2 0
l−2
β∈D /D λ∈β+D

attached to the negative definite lattice D and the harmonic polynomial

λ2 Y 2
(λ, YD )2 − l−2D . Furthermore, Pδ,n
K L
resp. Pδ,n denote the Poincaré series
 5.1 A lifting into the cohomology 131

in Sκ,K resp. Sκ,L , and E L the Eisenstein series with constant term 2e0 in
Mκ,L (cf. section 1.2).
In particular, for fixed Z the components of Ω K (Y, τ ) lie in Sκ,K . Simi-
larly Ω L (Z, τ ) can be viewed as an element of H1,1 (XL ) ⊗ Mκ,L .

Proof. Throughout the proof we write bL β,m (γ, n) instead of b(γ, n, 1 − k/2)
L
for the (γ, n)-th Fourier coefficient of the Poincaré series Fβ,m (τ, 1 − k/2) to
emphasize the dependence on L and (β, m).
To prove (5.14) we insert the Fourier expansions of the differential forms
hβ,−m (Z) into (5.12) and exchange the order of summation. We find

1 X X
¯ |ξ K
Ω L (Z, τ ) = √ ∂ ∂|Y p(β),−m (Y /|Y |)eβ (mτ )
8 2 β∈L0 /L m∈Z−q(β)
0
m>0
1 1 X X
+ ∂ ∂¯ log(Y 2 )e0 − ∂ ∂¯ log(Y 2 ) bL
β,−m (0, 0)eβ (mτ )
4 8 0
β∈L /L m∈Z−q(β)
m>0
1 X X X1
¯ κ (πn|λ||Y |, πn(λ, Y ))
+ √ e(n(δ, z 0 ))∂ ∂V
4 π 0
n
λ∈K δ∈L00 /L n≥1
q(λ)<0 p(δ)=λ+K
X X
× e(n(λ, X)) bL
β,−m (δ, q(λ))eβ (mτ ). (5.17)
β∈L0 /L m∈Z−q(β)
m>0

According to Proposition 1.16 one has

X X
2e0 − bL L
β,−m (0, 0)eβ (mτ ) = E (τ ),
β∈L0 /L m∈Z−q(β)
m>0
X X 1
bL
β,−m (δ, q(λ))eβ (mτ ) = − PL (τ ).
Γ (κ − 1) δ,−q(λ)
β∈L0 /L m∈Z−q(β)
m>0

If we put this into (5.17), we obtain (5.14).

Using (3.12) it can be seen in the same way that Ω K (Y, τ ) is equal to
132 5 Chern classes of Heegner divisors
"
Y2

1 ¯ X X
0
∂∂ − 2(d , Y ) ΦD
p(β),−m
16 (d, Y )
β∈K00 /K m∈Z−q(β)
m>0
#
8π X
2
+ (λ, Y ) eβ (mτ )
(Y, d) 0
λ∈D
λ+D=p(β)
q(λ)=−m

1 X X X (d, Y )  πn|λ||Y | κ−1

− ∂ ∂¯
2πΓ (κ − 1) n2 (d, Y )
λ∈D −0 δ∈K00 /K n≥1
0

p(δ)=λ+D
   
2πn|λ||Y | (λ, Y ) 0 K
× Kκ−1 e n + n(δ, d ) Pδ,−q(λ) (τ ). (5.18)
(d, Y ) (d, Y )
¯
Since the ∂ ∂-operator annihilates all terms which are linear in Y , the first
summand in (5.18) is equal to
" #
1 ¯ −1
X X
2 D
X
2
∂ ∂(d, Y ) YD Φp(β),−m + 8π (λ, YD ) eβ (mτ ).
16 0 0
β∈K0 /K m∈Z−q(β) λ∈D
m>0 λ+D=p(β)
q(λ)=−m
(5.19)
According to Theorem 2.14 we have
16πm
ΦD
p(β),−m = # {λ ∈ p(β) + D; q(λ) = −m} .
l−2
Thus (5.19) can be rewritten as

λ2 YD2
 
π ¯ X X X
∂ ∂(d, Y )−1 (λ, YD )2 − eβ (mτ )
2 l−2
β∈K00 /K m∈Z−q(β) λ∈D 0
m>0 λ+D=p(β)
q(λ)=−m

λ2 YD2
 
π ¯ X X
= ∂ ∂(d, Y )−1 (λ, YD )2 − eβ (−q(λ)τ )
2 l−2
β∈K00 /K λ∈p(β)+D
¯ Y )−1 ΘD (YD , τ ) |↑K .
= π∂ ∂(d, D

Putting this into (5.18) we get (5.16). The fact that ΘD (YD , τ ) is a cusp form
in Sκ,D follows from the usual theta transformation formula ([Bo2] Theorem
4.1). t
u
Theorem 5.9. The lifting

ϑ : Sκ,L −→ H1,1 (XL ); f 7→ f (τ ), Ω L (Z, τ ) τ

has the following properties:

 5.1 A lifting into the cohomology 133
P
i) Let f = γ,n c(γ, n)eγ (nτ ) be a cusp form in Sκ,L . Then the image of
f has the Fourier expansion
X X X
ϑ(f )(Z) = ϑ0 (f )(Y )−2−κ π 1/2−κ |λ|2−2κ n−1 e(n(δ, z 0 ))
λ∈K 0 n≥1 δ∈L00 /L
q(λ)<0 p(δ)=λ+K
¯ κ (πn|λ||Y |, πn(λ, Y )) e(n(λ, X)),
× c(δ, −q(λ))∂ ∂V

where the 0-th Fourier coefficient ϑ0 (f ) is given by

ϑ0 (f )(Y ) = f (τ ) |↓L K


K , Ω (Y, τ ) τ .

ii) The diagram

∗
Sκ,L / Sκ,L ϑ / H1,1 (XL ) (5.20)
O


∗
Sκ,L
η
/f
ClH ⊗C /H
e 1,1 (XL )

∗ ∗
commutes. Here Sκ,L → Sκ,L is given by the restriction of functionals, and
∗
Sκ,L is identified with Sκ,L by means of the Petersson scalar product (·, f ) 7→
f . The right vertical arrow denotes the canonical projection.

Proof. i) This is an immediate consequence of Theorem 5.8 and Proposition

1.5.
ii) By Theorem 5.2 and Theorem 5.5 the image of the functional aβ,m ∈
∗
Sκ,L under the maps in the lower line is given by 12 hβ,−m (Z). On the other
hand its image in Sκ,L equals

(4πm)κ−1
Pβ,m .
2Γ (κ − 1)

According to Theorem 5.8, for fixed Z the function Ω(Z, τ )− 18 E(τ )∂ ∂¯ log(Y 2 )
is a cusp form. Using Proposition 1.5 we find that the image of aβ,m in
H1,1 (XL ) is equal to
1 1
hβ,−m (Z) + bβ,−m (0, 0)∂ ∂¯ log(Y 2 ).
2 8
This implies the assertion. t
u

5.1.1 Comparison with the classical theta lift

We now compare the map ϑ constructed above with the theta lifting from el-
liptic cusp forms to holomorphic cusp forms on the orthogonal group O+ (V )
due to Oda [Od] and Rallis-Schiffmann [RS]. We use the description of
134 5 Chern classes of Heegner divisors

Borcherds [Bo2], because it fits most easily into our setting. Let us first recall
how holomorphic cusp forms of weight l contribute to the cohomology of XL .
It is easily verified that the determinant of the Jacobi matrix of the trans-
formation Z 7→ σ(Z) (σ ∈ O+ (V )) at Z ∈ Hl is equal to

j(σ, Z)−l

(see [Fr1] chapter I Hilfssatz 1.6 for the case of Siegel modular forms). Thus,
if f is a holomorphic modular form of weight l for Γ (L), then

ωf = f dz1 ∧ · · · ∧ dzl

is a holomorphic differential form of type (l, 0), which is invariant under Γ (L).
If f is a cusp form, then ωf is square integrable.
If r ∈ Q, we denote by Sr (Γ (L)) the space of holomorphic cusp forms of
weight r for the group Γ (L). Moreover, we write Hl,0 (XL ) for the space of
square integrable holomorphic l-forms on the quotient XL .

Lemma 5.10. The assignment f 7→ ωf defines an isomorphism

Sl (Γ (L))
∼ / Hl,0 (XL ).

Proof. This can be proved in the same way as Satz 2.6 in chapter III of [Fr1].
t
u

Note that the natural map Hl,0 (XL ) → H l (XL , C) to the middle coho-
mology of XL is injective.

We write −L for the lattice L as a Z-module, but equipped with the

quadratic form −q(·). Then the representation %−L is the dual representation
of %L . This follows immediately from the definition.
The classical theta lift of Oda and Rallis-Schiffmann in weight κ = 1 + l/2
is a map from Sκ,−L to Sl (Γ (L)). It can be viewed as a lifting

ι : Sκ,−L −→ Hl,0 (XL )

or as a lifting to the middle cohomology of XL . According to Theorem 14.3

in [Bo2], if f ∈ Sκ,−L with Fourier coefficients c(γ, n) (γ ∈ L0 /L and n ∈ Z +
q(γ)), then there is a square integrable holomorphic l-form ι(f ) ∈ Hl,0 (XL )
with Fourier expansion
X X X
ι(f )(Z) = nl−1 e(n(δ, z 0 ))c(δ, q(λ)/n2 )e((λ, Z)) dz1 . . . dzl .
λ∈K 0 n|λ δ∈L00 /L
q(λ)>0 p(δ)=λ/n+K
(λ,d)>0

runs over all n ∈ N with λ/n ∈ K 0 . The map

P
Here the second sum n|λ
 5.2 Modular forms with zeros and poles on Heegner divisors II 135

ϑ : Sκ,L −→ H1,1 (XL )

(cf. Theorem 5.9) can be regarded as a lifting to the second cohomology of

XL . If f ∈ Sκ,L with Fourier coefficients c(γ, n) (γ ∈ L0 /L and n ∈ Z − q(γ)),
then there exists a square integrable harmonic (1, 1)-form ϑ(f ) with Fourier
expansion
X X X
ϑ(f )(Z) = ϑ0 (f )(Y )−2−κ π 1/2−κ |λ|−l nl−1 e(n(δ, z 0 ))
λ∈K 0 n|λ δ∈L00 /L
q(λ)<0 p(δ)=λ/n+K
¯ κ (π|λ||Y |, π(λ, Y )) e((λ, X)).
× c(δ, −q(λ)/n2 )∂ ∂V

The similarity of the Fourier expansions of the two liftings is striking. In

both cases the arithmetic part of the coefficient with index λ (and q(λ) 6= 0)
is given by X X
nl−1 e(n(δ, z 0 ))c(δ, |q(λ)|/n2 ).
n|λ δ∈L00 /L
p(δ)=λ/n+K

We see that in some cases an elliptic modular form f ∈ Sκ,L can contribute
to both, the middle cohomology and the second cohomology of XL . For in-
stance, if the level of L is 1 or 2, then %L = %−L and therefore Sκ,L = Sκ,−L .

In a series of papers Kudla and Millson constructed liftings from the

cohomology with compact support of locally symmetric spaces associated to
orthogonal (or unitary) groups to classical holomorphic Siegel modular forms
on congruence subgroups [KM1, KM2, KM3]. Serious difficulties arise, if one
tries to generalize their construction to obtain liftings from the cohomol-
ogy with arbitrary support to classical modular forms. In particular the lifts
are no longer holomorphic and the geometric meaning of the Fourier coef-
ficients of the non-holomorphic part remains unclear. A first step towards
such a generalization was made by J. Funke in his thesis [Fu]. In our case the
(generalized) lifting of Kudla and Millson would go from H 2 (XL , C) to (non-
holomorphic) elliptic modular forms of weight κ. One should expect that it
is in some sense adjoint to our map ϑ. It would be interesting to understand
this connection in detail.

5.2 Modular forms whose zeros and poles lie on

Heegner divisors II

We pick up again our study of meromorphic modular forms whose divisor is

a linear combination of Heegner divisors. In section 4.3 we saw that such a
modular form F is (up to an additive constant) equal to a regularized theta
lift Φ of a linear combination of Maass-Poincaré series Fβ,m .
136 5 Chern classes of Heegner divisors

In this section we show, how Theorem 5.9 can be used to answer the
question, whether F is already the Borcherds lift of a nearly holomorphic
modular form as in Theorem 3.22.
Recall that a hyperbolic plane is a lattice which is isomorphic to the lattice
Z2 with the quadratic form q((a, b)) = ab.
We define two subspaces of Sκ,L by
X
−
Sκ,L = {f = c(γ, n)eγ (nτ ) ∈ Sκ,L ; c(λ, −q(λ)) = 0 for all λ ∈ L0 },
γ,n
(5.21)
 ⊥
+ −
Sκ,L = Sκ,L . (5.22)

− +
If the lattice L splits a hyperbolic plane over Z, then Sκ,L = 0 and Sκ,L =
Sκ,L . It is an immediate consequence of Theorem 5.9 (i) that the lifting ϑ
−
vanishes identically on Sκ,L .

Theorem 5.11. Assume that the restricted map

+ e 1,1 (XL )
ϑ : Sκ,L −→ H

is injective. Then every meromorphic modular form F with respect to Γ (L),

whose divisor is a linear combination of Heegner divisors
1 X
(F ) = c(λ)H(λ, q(λ)) (5.23)
2 0
λ∈L
q(λ)<0

(c(λ) ∈ Z with c(λ) = c(−λ) and c(λ) = 0 for all but finitely many λ),
is a Borcherds product; i.e. there exists a nearly holomorphic modular form
!
f ∈ Mk,L with principal part
X
c(λ)eλ (q(λ)τ ) (5.24)
0
λ∈L
q(λ)<0

such that F is the Borcherds lift of f as in Theorem 3.22.

Proof. Let F be a meromorphic modular form for the group Γ (L) with divisor
(5.23). Then
1 X
c(λ)c(H(λ, q(λ))) = 0
2 0 λ∈L

e 1,1 (XL ). According to Theorem 5.9 this implies that

in H
!
X
κ−1
ϑ c(λ)(−q(λ)) Pλ,−q(λ) = 0
λ∈L0
 5.2 Modular forms with zeros and poles on Heegner divisors II 137

e 1,1 (XL ). Using the assumption on ϑ we find that

in H
X
c(λ)(−q(λ))κ−1 Pλ,−q(λ)
λ∈L0

−
is contained in Sκ,L . On the other hand, by Proposition 1.5 this linear com-
−
bination of Poincaré series is orthogonal to Sκ,L . Therefore it vanishes iden-
tically. But this is equivalent to saying
X
c(λ)aλ,−q(λ) = 0
λ∈L0

∗
in Sκ,L . By Theorem 1.17 there exists a nearly holomorphic modular form
!
f ∈ Mk,L with principal part (5.24). Let B(f ) be the Borcherds lift of f
as in Theorem 3.22. Then B(f )/F is a holomorphic modular form (with
a multiplier system of finite order) without any zeros on Hl . Hence it is
constant. t
u

Theorem 5.12. Let L be an even lattice of signature (2, l), that splits two
orthogonal hyperbolic planes over Z. Then every meromorphic modular form
for the group Γ (L), whose divisor is a linear combination of Heegner divisors,
is a Borcherds product as in Theorem 3.22.

Proof. We write L as an orthogonal sum L = D ⊕ H1 ⊕ H2 , where D is a

negative definite lattice and H1 , H2 are hyperbolic planes. Put K = D ⊕ H1 .
+
According to Theorem 5.11 it suffices to prove that the map ϑ : Sκ,L →
1,1
H (XL ) is injective.
e
+
Let f ∈ Sκ,L with ϑ(f ) = 0. Denote the Fourier coefficients of f by c(γ, n).
−
We have to show that f ∈ Sκ,L .
For a suitable choice of the vectors z and z 0 the Fourier expansion of ϑ(f )
is given by
X X
ϑ(f )(Z) = ϑ0 (f )(Y ) − 2−κ π 1/2−κ |λ|2−2κ n2κ−3 c(λ/n, −q(λ)/n2 )
λ∈K 0 n|λ
q(λ)<0
¯ κ (π|λ||Y |, π(λ, Y )) e((λ, X)).
× ∂ ∂V

The assumption ϑ(f ) = 0 implies that

X
|λ|2−2κ n2κ−3 c(λ/n, −q(λ)/n2 ) = 0
n|λ

for all λ ∈ K 0 with q(λ) < 0. By an easy inductive argument we find that
c(λ, −q(λ)) = 0 for all λ ∈ K 0 with q(λ) < 0.
Recall that there is an isomorphism between H1 ⊕ H2 and the lattice
M2 (Z) of integral 2 × 2 matrices such that the quadratic form q corresponds
138 5 Chern classes of Heegner divisors

to the determinant on M2 (Z). The group SL2 (Z) acts on M2 (Z) by multipli-
cation from both sides. This gives rise to a homomorphism

SL2 (Z) × SL2 (Z) → O+ (H1 ⊕ H2 )

(cf. [FH] Lemma 4.4). Hence, by the theorem of elementary divisors for
SL2 (Z), for any λ ∈ L0 there is a σ ∈ O+ (H1 ⊕ H2 ) ⊂ Γ (L) such that
σ(λ) ∈ K 0 .
Thereby we obtain c(λ, −q(λ)) = 0 for all λ ∈ L0 with q(λ) < 0. t
u

It seems likely that Theorem 5.11 can also be used to prove a more general
version of Theorem 5.12. However, this seems to be a non-trivial problem,
which might require additional arguments like some newform theory or Hecke
theory for the space Sκ,L and the lifting ϑ.

Example 5.13. Fix a positive integer t. Let L be the lattice L = D ⊕H1 ⊕H2 ,
where H1 , H2 are hyperbolic planes and D denotes the lattice Z with the
negative definite quadratic form q(a) = −ta2 . Obviously L has signature
(2, 3), k = −1/2, and κ = 5/2. The discriminant group L0 /L ∼ = D0 /D has
order 2t. In the same way as in [EZ] §5 it can be seen that the space S5/2,L
cusp
is isomorphic to the space J3,t of skew-holomorphic Jacobi cusp forms of
weight 3 and index t (see [Sk], [Ko]).
The group Γ (L) is isomorphic to the paramodular group of level t (cf. [GN]
Lemma 1.9 and [GrHu]). Moreover, the quotient XL = Hl /Γ (L) is the moduli
space of Abelian surfaces with a (1, t)-polarization. The Heegner divisors
H(β, m) are known as Humbert surfaces. Since L splits two hyperbolic planes
over Z, any H(β, m) is a prime divisor in D(XL ) (cf. [FH] Lemma 4.4).
By Theorem 5.9 we have an injective map
cusp
ϑ : J3,t −→ H1,1 (XL )
cusp
from J3,t into the space of square integrable harmonic (1, 1)-forms on XL .
cusp
Observe that the 0-th Fourier coefficient ϑ0 (f )(Y ) of the lift of f ∈ J3,t
can be identified with the Shimura lift of f . According to Theorem 5.12 any
meromorphic modular form with respect to Γ (L), whose only zeros and poles
lie on Humbert surfaces, is a Borcherds product in the sense of Theorem 3.22.
In this case the classical theta lift is essentially the Maass lift. It gives rise
to an injective map
cusp
ι : J3,t −→ H3,0 (XL )
from holomorphic Jacobi cusp forms of weight 3 and index t into H3,0 (XL ).
We obtain a partial answer to a question raised by W. Kohnen ([Ko] Problem
(ii) in §2). Holomorphic Jacobi forms of weight 3 contribute via Maass lift
to the middle cohomology of XL , whereas skew-holomorphic Jacobi forms
contribute to the second cohomology. Here it is natural to ask, if there is a
similar connection in higher weights.
References

[AbSt] M. Abramowitz and I. Stegun, Pocketbook of Mathematical Functions,

Verlag Harri Deutsch (1984).
[Bo1] R. E. Borcherds, Automorphic forms on Os+2,2 (R) and infinite products,
Invent. Math. 120 (1995), 161–213.
[Bo2] R. E. Borcherds, Automorphic forms with singularities on Grassmannians,
Invent. Math. 132 (1998), 491–562.
[Bo3] R. E. Borcherds, The Gross-Kohnen-Zagier theorem in higher dimensions,
Duke Math. J. 97 (1999), 219–233.
[Bo4] R. E. Borcherds, Sporadic groups and string theory, First E. C. M., Vol. I
(Paris 1992), 411-421, Progr. Math. 119 (1994), Birkhäuser, Basel.
[Bl1] A. Borel, Introduction aux groupes arithmétiques, Hermann (1969).
[Bl2] A. Borel, Linear Algebraic Groups. In: Algebraic groups and discontinuous
subgroups, Proceedings of Symposia in Pure Mathematics 9, American
Mathematical Society (1966), 3–19.
[Bl3] A. Borel, Cohomology of arithmetic groups, Proc. Int. Congr. of Mathe-
matics, Vancouver 1974, Vol. 1 (1975), 435–442.
[Br1] J. H. Bruinier, Borcherds products and Chern classes of Hirzebruch-Zagier
divisors, Invent. math. 138 (1999), 51–83.
[Br2] J. H. Bruinier, Borcherdsprodukte und Chernsche Klassen von Hirze-
bruch-Zagier-Zykeln, Dissertation, Universität Heidelberg (1998).
[BK] J. H. Bruinier and M. Kuss, Eisenstein series attached to lattices and mod-
ular forms on orthogonal groups, preprint (2000), to appear in Manuscr.
Math.
[BKK] J. I. Burgos, J. Kramer, and U. Kühn , manuscript, in preparation (2001).
[DI] F. Diamond and J. Im, Modular forms and modular curves, Canadian
Mathematical Society Conference Proceedings 17 (1995), 39–133.
[DN] K. Doi and H. Naganuma, On the functional equation of certain Dirichlet
series, Invent. Math. 9 (1969), 1–14.
[E1] A. Erdélyi, W. Magnus, F. Oberhettinger and F. G. Tricomi, Higher Tran-
scendental Functions, McGraw-Hill (1953).
[E2] A. Erdélyi, W. Magnus, F. Oberhettinger and F. G. Tricomi, Tables of
Integral Transforms, vol. I, McGraw-Hill (1954).
[EZ] M. Eichler and D. Zagier, The Theory of Jacobi Forms, Progress in Math.
55 (1985), Birkhäuser.
[Fa] J. Fay, Fourier coefficients of the resolvent for a Fuchsian group, J. reine
angew. Math. 294 (1977), 143–203.
[Fr1] E. Freitag, Siegelsche Modulfunktionen, Springer-Verlag (1983).
[Fr2] E. Freitag, Singular Modular Forms and Theta Relations, Lecture Notes
in Mathematics 1487, Springer-Verlag (1991).
140 References

[FH] E. Freitag and C. F. Hermann, Some modular varieties of low dimension,

Adv. Math. 152 (2000), 203–287.
[Fu] J. Funke, Rational quadratic divisors and automorphic forms, PhD-thesis,
University of Maryland (1999).
[GH] P. Griffiths and J. Harris, Principles of algebraic geometry, John Wiley
(1978).
[GrHu] V. Gritsenko and K. Hulek, Minimal Siegel modular threefolds, Math.
Proc. Camb. Phil. Soc. 123 (1998), 461–485.
[GN] V. Gritsenko and V. Nikulin, Automorphic forms and Lorentzian Kac-
Moody algebras. Part II, Intern. J. of Math. 9 (1998), 201–275.
[GR] R. C. Gunning and H. Rossi, Analytic functions of several complex vari-
ables, Prentice-Hall (1965).
[GW] R. E. Greene and H. Wu, C ∞ approximations of convex, subharmonic,
and plurisubharmonic functions, Ann. scient. Éc. Norm. Sup. 12 (1979),
47–84.
[Ha] Harish-Chandra, Automorphic Forms on Semisimple Lie Groups, Lecture
Notes in Mathematics 62, Springer-Verlag (1968).
[He] D. A. Hejhal, The Selberg Trace Formula for PSL(2, R), Lecture Notes in
Mathematics 1001, Springer-Verlag (1983).
[Hel] S. Helgason, Differential geometry and symmetric spaces, Academic Press
(1962).
[HM] J. Harvey and G. Moore, Algebras, BPS states, and strings, Nuclear Phys.
B 463 (1996), no. 2-3, 315–368.
[HZ] F. Hirzebruch und D. Zagier, Intersection Numbers of Curves on Hilbert
Modular Surfaces and Modular Forms of Nebentypus, Invent. Math. 36
(1976), 57–113.
[Ho] R. Howe, θ-series and invariant theory, Proceedings of Symposia in Pure
Mathematics 33, part 1, American Mathematical Society (1979), 275–285.
[Ka] N. Katz, p-adic properties of modular schemes and modular forms. In:
Modular Functions of One Variable III, Lecture Notes in Math. 350,
Springer-Verlag (1973), 69–190.
[Ko] W. Kohnen, Jacobi forms and Siegel modular forms: Recent results and
problems, L’Ens. Math. 39 (1993), 121–136.
[Kon] M. Kontsevich, Product formulas for modular forms on O(2, n), Séminaire
Bourbaki 821 (1996).
[KM1] S. Kudla and J. Millson, The theta correspondence and harmonic forms
I, Math. Ann. 274, (1986), 353–378.
[KM2] S. Kudla and J. Millson, The theta correspondence and harmonic forms
II, Math. Ann. 277, (1987), 267–314.
[KM3] S. Kudla and J. Millson, Intersection numbers of cycles on locally sym-
metric spaces and Fourier coefficients of holomorphic modular forms in
several complex variables, IHES Publi. Math. 71 (1990), 121–172.
[Ma1] H. Maass, Modular functions of one complex variable, Spinger-Verlag
(1983).
[Ma2] H. Maass, Automorphe Funktionen in mehreren Veränderlichen und
Dirichletsche Reihen, Abh. Math. Sem. Univ. Hamburg 16 (1949), 72–
100.
[Ma3] H. Maass, Über ein Analogon zur Vermutung von Saito-Kurokawa, Invent.
Math. 60 (1980), 85–104.
 References 141

[Mar] G. A. Margulis, Discrete Subgroups of Semisimple Lie Groups, Springer-

Verlag (1991).
[Me] O. Meyer, Über die Satakekompaktifizierung arithmetischer Quotienten
von beschränkten hermiteschen Bereichen des Types IVn , Diplomarbeit,
Universität Heidelberg (1998).
[Na] H. Naganuma, On the coincidence of two Dirichlet series associated with
cusp forms of Hecke’s “Neben”-type and Hilbert modular forms over a real
quadratic field, J. Math. Soc. Japan 25 (1973), 547–555.
[Nak] S. Nakajima, On Invariant Differential Operators on Bounded Symmetric
Domains of Type IV, Proc. Japan Acad. 58 Ser. A (1982), 235–238.
[Ni] D. Niebur, A class of nonanalytic automorphic functions, Nagoya Math.
J. 52 (1973), 133–145.
[No] A. Nobs, Die irreduziblen Darstellungen der Gruppen SL2 (Zp ), insbeson-
dere SL2 (Z2 ). I. Teil, Comment Math. Helvetici 51 (1976), 465–489.
[Od] T. Oda, On Modular Forms Associated with Indefinite Quadratic Forms
of Signature (2, n − 2), Math. Ann. 231 (1977), 97–144.
[PS1] I. Piateskii-Shapiro, On the Saito-Kurokawa lifting, Invent. Math. 71
(1983), 309–338.
[PS2] I. Piateskii-Shapiro, Cuspidal automorphic representations associated to
parabolic subgroups and Ramanujan conjecture. In: Number theory re-
lated to Fermat’s last theorem, Prog. Math. 26, Birkhäuser (1982), 143–
151.
[RS] S. Rallis and G. Schiffmann, On a relation between SL2 cusp forms and
cusp forms on tube domains associated to orthogonal groups, Trans. AMS
263 (1981), 1–58.
[Rh] G. de Rham, Differentiable Manifolds, Springer-Verlag (1984).
[Ro] W. Roelcke, Über den Laplace-Operator auf Riemannschen Manigfaltig-
keiten mit diskontinuierlichen Gruppen, Math. Nachr. 21 (1960), 131–149.
[Sc] B. Schoeneberg, Elliptic Modular Functions, Springer-Verlag (1974).
[SeSt] J.-P. Serre and H. M. Stark, Modular forms of weight 1/2. In: Modular
Functions of One Variable VI, Lecture Notes in Math. 627, Springer-
Verlag (1977), 27–67.
[Sh] G. Shimura, Introduction to the Arithmetic Theory of Automorphic Func-
tions, Princeton University Press (1971).
[Sk] N.-P. Skoruppa, Developments in the theory of Jacobi forms. In: Proceed-
ings of the conference on automorphic funtions and their applications,
Chabarovsk (eds.: N. Kuznetsov and V. Bykovsky), The USSR Academy
of Science (1990), 167–185. (see also MPI-preprint 89-40, Bonn (1989).)
[Sn] T. Shintani, On construction of holomorphic cusp forms of half integral
weight, Nagoya Math. J. 58 (1975), 83–126.
[SABK] C. Soulé, D. Abramovich, J.-F. Burnol, and J. Kramer, Lectures on
Arakelov Geometry, Cambridge Studies in Advanced Mathematics 33,
Cambridge University Press (1992).
[Wei] R. Weissauer, The Picard group of Siegel modular threefolds, J. reine
angew. Math. 430 (1992), 179–211.
[Wel] R. O. Wells, Differential analysis on complex manifolds, Springer-Verlag
(1980).
[Yau] S.-T. Yau, Some function-theoretic properties of complete Riemannian
Manifolds and their applications to geometry, Indiana Univ. Math. J. 25
(1976), 659–670.
142 References

[Za] D. Zagier, Modular Forms Associated to Real Quadratic Fields, Invent.

Math. 30 (1975), 1–46.
Notation

√
· The principal branch of the square root
f (·)  g(·) |f (·)| ≤ C|g(·)|
f (·) ε g(·) |f (·)| ≤ C(ε)|g(·)|
f ≈g See p. 51
|κ The Petersson slash operator of weight κ with representation %L (p. 17)
|∗κ The Petersson slash operator of weight κ with representation %∗L (p. 17)
ha, bi The standard scalar product on C[L0 /L] for a, b ∈ C[L0 /L]
(a, b) The bilinear form on L for a, b ∈ L
(f, g) The Petersson scalar product of f, g ∈ Sκ,L (p. 22)
↑LK See p. 129
↓LK See p. 129
λ2 = (λ, λ) for λ ∈ V
|λ| = |(λ, λ)|1/2 for λ ∈ V
λ⊥ The orthogonal complement of λ
λv The orthogonal projection of λ ∈ V onto v

∗ ∗
aγ,n A certain functional in Sκ,L (p. 36) or Sκ,L (Z) (p. 120)
Aκ,L (Z) See p. 120
b(γ, n, s) The Fourier coefficients of Fβ,m (τ, s) (p. 30)
bβ,m (γ, n) = b(γ, n) = b(γ, n, 1 − k/2)
Br (x) The r-th Bernoulli polynomial (p. 61)
Br (x) See p. 62
c(D) The Chern class of the divisor D in the second cohomology
c(γ, n; y, s) The Fourier coefficients of Fβ,m (τ, s) (p. 57)
C The positive cone in K ⊗ R (p. 77)
Cβ,m A certain constant (p. 72)
Cl(X/Γ ) The divisor class group of X/Γ (p. 117)
Cl(X/Γ )
f The modified divisor class group of X/Γ (p. 117)
ClH (XL )
f The subgroup of f Cl(XL ) generated by the Heegner divisors H(β, m)
C The complex numbers
C[L0 /L] The group algebra of the discriminant group L0 /L
Cs=a [f (s)] The constant term of the Laurent expansion of f at s = a (p. 50)
χ A multiplier system or character
d Often a primitive isotropic vector in K
144 Notation

d0 An element of K 0 with (d, d0 ) = 1

d˜ Usually d0 − q(d0 )d
⊥
D = K ∩ d⊥ ∩ d0 , or a divisor
δ∗,∗ The Kronecker-delta
∆k The Laplace operator of weight k (p. 28)
e(z) = e2πiz for z ∈ C
eγ An element of the standard basis (eγ )γ∈L0 /L of C[L0 /L]
eγ (τ ) = eγ e(τ )
EβL (τ ) An Eisenstein series in Mκ,L (p. 23)
E(τ ) = E0 (τ ) = E0L (τ )
E(u, v) An Eichler transformation (p. 105)
η See p. 120
L
Fβ,m (τ, s) A non-holomorphic Poincaré series of weight k (p. 29)
Feβ,m (τ, s) See p. 30
F (a, b, c; z) The Gauss hypergeometric function as in [AbSt]
K
Fβ,m (τ, s; r, t) See p. 45
F The standard fundamental domain for SL2 (Z)
Fu The truncated fundamental domain (p. 47)
Gal(E/F ) The Galois group of a field extension E/F
Gr(L) The Grassmannian of L
Γ Usually a subgroup of finite index of Γ (L)
Γ (N ) The principal congruence subgroup of level N
Γ (y) The Gamma function
Γ (a, x) The incomplete Gamma function as in [AbSt] (p. 28)
Γ1 = SL2 (Z)
Γ∞ = {( 10 n1 ) ; n ∈ Z} ≤ SL2 (Z)
Γ̃∞ = {(( 10 n1 ) , 1); n ∈ Z} ≤ Mp2 (Z)
Γ (L) = O+ (V ) ∩ Od (L)
H(β, m) The Heegner divisor of index (β, m) (p. 47 and p. 117)
hβ,m (Z) ¯ L (Z) of the Chern class of H(β, m)
The representative 41 ∂ ∂ξ β,m
Hc∗ (β, m, γ, n) A generalized Kloosterman sum (p. 19)
Hc (β, m, γ, n) A generalized Kloosterman sum (p. 30)
H 2 (XL , C) The second cohomology of XL with coefficients in C
2
H̃ (XL , C) See p. 122
H1,1 (XL ) The space of square integrable harmonic (1, 1)-forms on XL (p. 123)
He 1,1 (XL ) See p. 124
H The complex upper half plane {τ ∈ C; =(τ ) > 0}
Hl The generalized upper half plane (p. 77)
Hl (Λ) The generalized upper half plane attached to Λ (p. 108)
H The upper half space model of hyperbolic space (p. 65)
Iν (y) The I-Bessel function as in [AbSt] or [E1]
=(·) The imaginary part
Jν (y) The J-Bessel function as in [AbSt] or [E1]
j(σ, Z) The automorphic factor (σ(ZL ), z) (p. 82)
 Notation 145

Jk,t The space of holomorphic Jacobi forms of weight k and index t

Jk,t The space of skew-holomorphic Jacobi forms of weight k and index t
Kν (y) The K-Bessel function as in [AbSt] or [E1]
k Usually 1 − l/2
κ Usually 1 + l/2
⊥
K = L ∩ z⊥ ∩ z0
K See p. 76
K+ A component of K
Ke+ The cone over K+ (p. 82)
l An integer ≥ 3
L An even lattice of signature (b+ , b− )
L0 The dual lattice of L
L00 = {λ ∈ L0 ; (λ, z) ≡ 0 (mod N )}
Lk A Maass differential operator (p. 95)
Lp (Hl /Γ ) See p. 101
Λ Usually√an admissible index-tuple

(p. 107)
M̃ = (M, cτ + d) for M = ac db ∈ SL2 (R)
Mν, µ (y) The M -Whittaker function as in [AbSt] (p. 27)
Ms (y) = y −k/2 M−k/2, s−1/2 (y)
Mκ,L The space of modular forms of weight κ with respect to %∗L and Mp2 (Z)
!
Mk,L The space of nearly holomorphic modular forms of weight k (p. 34)
Mp2 (R) The metaplectic cover of SL2 (R) (p. 15)
Mp2 (Z) The integral metaplectic group (p. 16)
Mκ,L = Gal(C/Q) · Mκ,L (p. 120)
µ Usually a certain vector in V ∩ z ⊥ (p. 42)
N The unique positive integer with (z, L) = N Z, or the level of L
N The zero-quadric in P (V (C)) (p. 76)
N = {1, 2, 3, . . . }
N0 = N ∪ {0}
O(V ) The (special) orthogonal group of V (p. 40)
O+ (V ) The connected component of the identity of O(V )
O(L) = {g ∈ O(V ); gL = L}
Od (L) The discriminant kernel of O(L), see p. 40
O+Q (L) The rational orthogonal group of L (p. 101)
Ω The invariant Laplace operator on Hl (p. 94)
Ω(τ ) A kernel function, see p. 123
Ω L (Z, τ ) The kernel function of ϑ, see p. 128
Ω K (Y, τ ) A kernel function, see p. 130
p Usually a certain projection L00 → K 0 (p. 41)
L
Pβ,m (τ ) The Poincaré series of index (β, m) in Sκ,L (p. 19)
L
pβ,m (γ, n) The Fourier coefficients of Pβ,m
P (V (C)) The projective space of V (C)
ΦLβ,m (v, s) The regularized theta lift of Fβ,m (τ, s) (p. 47)
Φβ,m (v) = Cs=1−k/2 [Φβ,m (v, s)], the regularized theta lift of Fβ,m (τ, 1 − k/2)
146 Notation

L
ψβ,m See p. 66 and p. 79
Ψβ,m (Z) See p. 80
q(a) = 21 (a, a), the quadratic form on L
qβ (γ, n) The Fourier coefficients of EβL (τ )
q(γ, n) The Fourier coefficients of E(τ )
Q The rational numbers
Rk A Maass differential operator (p. 95)
R The real numbers
Rt The set of Y defined by (4.18b)–(4.18d)
<(·) The real part
%L The Weil representation attached to (L0 /L, q) (p. 16)
%γδ (M, φ) A coefficient of the representation %L (p. 17)
%∗L The dual representation of %L
%β,m (W ) See p. 67
%f (W ) See p. 86
s A complex
variable
√

S = 01 −1 0 , τ ∈ Mp2 (Z)
SL2 (R) The group of real 2 × 2-matrices of determinant 1
SL2 (Z) The group of integral 2 × 2-matrices of determinant 1
Sκ,L The subspace of cusp forms in Mκ,L
∗
Sκ,L The dual space of Sκ,L
−
Sκ,L A certain subspace of Sκ,L (p. 136)
+
Sκ,L A certain subspace of Sκ,L (p. 136)
supp(D) The support of the divisor D
St A Siegel domain in Hl (p. 101)
S̃t (Λ) A Siegel domain in K+ (p. 108)
S(β, m, U ) See p. 47 and p. 81
Sκ,L = Gal(C/Q) · Sκ,L (p. 119)
Sκ,L (Z) The module of cusp forms in Sκ,L whose coefficients all lie in Z
∗
Sκ,L (Z) The dual module of Sκ,L (Z)
σ Often the real part of s
T = (( 10 11 ) , 1) ∈ Mp2 (Z)
τ A variable in H
ΘL (τ, v; r, t) The Siegel theta function attached to L (p. 39)
θγ (τ, v; r, t) A component of ΘL (τ, v; r, t)
ϑ A lifting, see p. 128
ϑ0 (f )(Y ) The 0-th Fourier coefficient of ϑ(f ) (p. 133)
ΘD (YD , τ ) A theta series, see p. 130
v Usually an element of Gr(L)
V L⊗R
V (C) =V ⊗C
Vκ (A, B) A special function, see p. 71
w The orthogonal complement of zv in v
W A Weyl chamber (p. 61), or a complex variable
 Notation 147

Wν, µ (y) The W -Whittaker function as in [AbSt] (p. 27)

Ws (y) = |y|−k/2 Wk/2 sgn(y), s−1/2 (|y|)
x The real part of τ
X The real part of Z ∈ Hl , or a normal irreducible complex space
XL = Hl /Γ (L)
L
ξβ,m See p. 66 and p. 79
y The imaginary part of τ
Y Usually the imaginary part of Z ∈ Hl
z Usually a primitive isotropic vector in L
z0 An element of L0 with (z, z 0 ) = 1
z̃ Usually z 0 −

q(z
0

)z
−1 0
Z = 0 −1 , i ∈ Mp2 (Z), or a variable in Hl
ZL Usually (Z, 1, −q(Z) − q(z 0 )) ∈ Ke + for Z ∈ Hl
Z The integers
ζ A vector in L with (ζ, z) = N
ζ(s) The Riemann zeta function
ζN A primitive N -th root of unity
Index

admissible index-tuple, 110 fundamental domain, 22

arrow-operators, 131 – truncated, 47
automorphic form, 85
automorphy factor, 85, 123 Gauss hypergeometric function, 54
Gram determinant, 41
Bernoulli Grassmannian, 39
– number, 26 Green current, 124
– polynomial, 63 Green object, 124
Bessel function, 19, 30, 64
Borcherds product, 87, 88, 139 Hankel integral, 25
– generalized, 63, 82 Heegner divisor, 80, 119
Hermitean metric, 123, 126
Cartier divisor, 123 Hermitean symmetric space, 73, 95
Casimir element, 101 Humbert surface, 93
character, 85 Humbert surfaces, 140
Chern class hyperbolic plane, 91, 138
– of a divisor, 123 hyperbolic space, 63
– of a Heegner divisor, 124, 126 – hyperboloid model, 66, 79
cocycle relation, 85 – upper half space model, 67
complex structure, 78
cusp form, 18 incomplete Gamma function, 28

discriminant group, 16 Jacobi form, 18

discriminant kernel, 40 – skew-holomorphic, 18, 140
divisor, 119 – weak, 92
– class group, 119
– group, 119 Kähler class, 124
Doi-Naganuma lifting, 131 Kähler metric, 96
dual lattice, 16 kernel function, 125, 130
dual representation, 17 Kloosterman sum, 19, 30
Koecher principle, 85, 86, 109
Eichler transformation, 86, 107, 111
Eisenstein series, 23 Laplace operator
error function, 54 – for O(2, l), 96
Euler-Mascheroni constant, 75 – hyperbolic, 72
– of weight k, 28
Fourier coefficients, 18, 30 – on a Riemann manifold, 112
Fourier expansion, 18, 29, 86 lattice, 16
– of the theta lift, 63 – Lorentzian, 63
150 Index

Laurant expansion, 47 – on (1, 1)-forms, 126

Lehmer conjecture, 94 – Petersson, 22
level, 120 Serre duality, 36, 125
Shimura lift, 140
Maass operators, 97 Siegel domain, 103, 110
metaplectic group, 15 Siegel theta function, 39
modular form, 18 singularity, 52
– elliptic , 25 skew-holomorphic Jacobi form, 18, 140
– nearly holomorphic, 34 subharmonic, 113
– on O(2, l), 85
multiplier system, 85 theta function, 39
theta integral, 46
orthogonal group, 40 theta lift, 39, 135
– regularized, 47
paramodular group, 140
theta transformation formula, 40
Petersson coefficient formula, 22
tube domain, 79
Petersson slash operator, 17
pluriharmonic, 82
unfolding trick, 23, 54, 55, 57
Poincaré series, 19
upper half plane, 15
– non-holomorphic, 29
– generalized, 79
principal part, 34
upper half space model, 67
projection, 41, 42

quadratic form, 16 weak Jacobi form, 92

Weil representation, 16, 100
Ramanujan sum, 25 Weyl chamber, 63, 80, 88
Ramanujan-Petersson conjecture, 20 Weyl vector, 69, 88
reduction theory, 103, 104 Whittaker differential equation, 27
Riemann manifold, 112 Whittaker function, 27
Riemann metric, 95 – asymptotic properties, 50
– invariant, 96 – generalized, 73

scalar product, 17 zero-quadric, 78