Eisenstein Classes and Generating Series of Modular Symbols in

EISENSTEIN CLASSES AND GENERATING SERIES OF
MODULAR SYMBOLS IN SLN
ROMAIN BRANCHEREAU
arXiv:2411.08690v1 [math.NT] 13 Nov 2024
Abstract
We define a theta lift between the homology in degree N −
1 of a locally symmetric space associated to SLN (R) and
the space of modular forms of weight N . We show that
the Fourier coefficients of this lift are Poincaré duals to
certain modular symbols associated to certain maximal
parabolic subgroups. The constant term is a canonical
cohomology classes obtained by transgressing the Euler
class of a torus bundle. This Eisenstein lift realizes a
geometric theta correspondence for the pair SLN × SL2 ,
in the spirit of Kudla-Millson. When N = 2, we show
that the lift surjects on the space of weight 2 modular
forms spanned by an Eisenstein series and the eigenforms
with non-vanishing L-function.
Contents
1. Introduction 2
1.1. The work of Kudla-Millson 2
1.2. Special cycles for SLN (R) 3
1.3. The Eisenstein class 4
1.4. Theta lifts 6
1.5. Evaluation on tori 6
1.6. The case of the modular curve 8
1.7. Outline of the paper 9
1.8. Acknowledgements 9
2. Locally symmetric spaces and special cycles 9
2.1. Locally symmetric space of SLN (R) 9
2.2. Special cycles in S 10
2.3. Quotients 12
3. Thom forms and the Mathai-Quillen formalism 15
3.1. Thom isomorphism 15
3.2. Mathai-Quillen construction 17
3.3. The form ϕ 21
3.4. The form ψ 26
3.5. The transgression form 28
4. Theta lift 30
4.1. Weil representation, intertwiners and theta series 30
4.2. Computation of the partial Fourier transforms 38
4.3. Differential forms as theta series 40
4.4. The Eisenstein class 41
1
4.5. The Fourier coefficients of the Eisenstein class 43
4.6. The constant term 50
4.7. Holomorphicity of the theta lift 53
4.8. Span of special cycles 55
4.9. Periods over tori 56
5. The case N = 2 60
5.1. Hecke operators 61
5.2. The Eisenstein lift when N = 2 62
References 65
1. Introduction
1.1. The work of Kudla-Millson. In a series of influential papers [KM86;
KM87; KM90], Kudla and Millson construct (in greater generality) a geo-
metric theta lift relating the cohomology of orthogonal locally symmetric
spaces DΓ := Γ\D of signature (p, q) to holomorphic modular forms of weight
p+q
2 . The symmetric space D ≃ SO(p, q)/SO(p) × SO(q) is the Grassma-
nian of negative q-planes in a real quadratic space of signature (p, q), and
Γ ⊂ SO(p, q) is a torsion-free subgroup preserving an integral lattice in Rp+q .
The central ingredient of their construction is a suitable differential q-form
ϕKM ∈ Ωq (D)⊗S(Rp+q ) valued in the space of Schwartz function in S(Rp+q ).
The theta machinery introduced by Weil [Wei64] produces a theta kernel
ΘKM (z, τ ) ∈ Ωq (D)Γ ⊗ M p+q (Γ′ ), z ∈ D, τ ∈H
2
where M p+q (Γ′ ) is the space of modular forms of weight (p + q)/2 and a
2
certain level Γ′ ⊂ SL2 (Z) depending on the choice of the lattice. This theta
kernel can be used to define two lifts. Let S p+q (Γ′ ) be the subspace of cusps
2
forms. On the one hand, the Petersson inner product of a cusp form f with
the theta kernel is a closed form, which defines a lift
ΛKM : S p+q (Γ′ ) −→ H q (DΓ ; C).
2
On the other hand, we can integrate the theta kernel on a cycle C in

Zq (DΓ ; Z) of degree q to define
(1.1) ΘKM : Hq (DΓ ; Z) −→ M p+q (Γ′ ).
2
The main feature of this lift is that it has the Fourier expansion
Z Z X∞
ΘKM (z, τ ) = c0 + [C, Cn ]q n
C C n=1
where the special cycles
BM
Cn ∈ Zpq−q (DΓ ; Z)
are locally finite cycles of codimension q in DΓ , and [C, Cn ] is the intersection
number in DΓ between the two cycles C and Cn .
In fact, Kudla and Millson constructed a more general lift between the co-
homology of orthogonal locally symmetric spaces and Siegel modular forms,
corresponding to the dual pair SO(p, q) × Spn (R). They also provided an
Generating series of modular symbols in SLN 3
analogous lift for locally symmetric spaces attached to unitary groups. Re-
cently, Shi [Shi23] has considered generalized special cycles on Hermitian
locally symmetric manifolds, thus extending the work of Kudla and Millson
to new dual pairs. The Kudla-Millson lift has become an important tool in
arithmetic geometry, and there is an extensive literature on generalizations
and applications of these lifts by many authors.
The aim of this paper is to construct a theta lift for the pair SLN (R) ×
SL2 (R), in the style of the Kudla-Millson lift. It is also motivated by the con-
struction of a similar lift by Bergeron-Charollois-Garcia [BCG20]. Although
the construction is different, we believe that the two lifts are the same. (As
we will show, the two lifts agree on cycles attached to tori, which suggests
that they are indeed equal.) In any case, the main novelty of the construction
in this paper is that its Fourier coefficients have a geometric interpretation,
which is not present in the work of Bergeron-Charollois-Garcia.
1.2. Special cycles for SLN (R). We consider the symmetric spaces
X := GLN (R)+ /SO(N ), S := SLN (R)/SO(N ),
where GLN (R)+ is the group of invertible matrices with positive determi-
nant. The space X can be seen as the space of real positive-definite symmet-
ric invertible matrices, and S ⊂ X as the subspace of matrices of determinant
one. Throughout this paper, we will denote by
v = (v, w) ∈ RN ⊕ RN
a pair of vectors in RN . Let ρ the representation ρ : GLN (Q) −→ QN ⊕ QN
given by ρg v := (gv, g −t w). Consider the quadratic form
B : RN ⊕ RN −→ R, B(v) := v t w,
of signature (N, N ), that we can also see as a pairing on RN × RN . Let
L ⊂ RN be an integral lattice, in the sense that L ⊂ L∗ := Hom(L, Z), and
let
L := L ⊕ L ⊂ R2N .
The dual with respect to the quadratic form is L∨ = L∗ ⊕ L∗ . Let DL :=
L∨ /L be the discriminant of the lattice, and χ ∈ C[DL ] a test function. We
will also view it as a function
χ : Q2N −→ C
that is supported on L∨ and invariant under translation by L. Let us assume1
that χ(0) = 0, and let Γ ⊂ SLN (R) be a torsion-free subgroup that preserves
χ in the sense that χ(ργ v0 ) = χ(v0 ). (Note that the groups Γ, Γ′ that we
will use from now on have nothing do to with the subgroups that appear in
1.1).
We will be interested in the cohomology of the (N 2 +N −2)/2 dimensional
locally symmetric space
SΓ := Γ\S.
1This assumption is not necessary in most parts of the paper, but is required for the
expression of the constant term.
4 Romain Branchereau
For each vector v ∈ R2N there is a natural submanifold Sv ⊂ S of codimen-

sion (N − 1), and is the image of an embedding
GLN −1 (R)+ /SO(N − 1) ֒−→ S.
The projection Zv of Sv in the quotient SΓ only depends on the class of v
in Γ\L∨ . It defines a locally finite cycle
BM
Zv ∈ Z(N 2 −N )/2 (SΓ ; Z)
of codimension N − 1 in the Borel-Moore homology. We consider the finite

linear combination
X
BM
Zn (χ) := χ(v)Zv ∈ Z(N 2 −N )/2 (SΓ ; C).
[v]∈Γ\Q2N
B(v)=n
Remark 1.1. The classical modular symbols {α, β} in H are associated to two
lines α, β ∈ P1 (Q) and generate the relative homology of the modular curve
Y0 (p). These modular symbols can be generalized to S in different ways. The
modular symbols Zn (χ) considered in this paper are associated to maximal
parabolic subgroups. Similar modular symbols have been considered in pre-
vious work [KMS00; Sch93]. On the other hand, Ash and Rudolph [AR79]
consider generalized modular symbols ZQ associated to Borel subgroups and
show that these modular symbols generate the relative homology of SΓ . We
will compute the period of the Eisenstein lift on such modular symbols in
1.5.
1.3. The Eisenstein class. Using the Mathai-Quillen formalism, we con-
struct a closed and GLN (R)+ -invariant form ϕ ∈ ΩN (X) ⊗ S(R2N ). We
define the theta series
X
Θϕχ (z, τ ) := j(hτ , i)N ω(hτ )ϕ(z, v)χ(v) ∈ ΩN (X)Γ ⊗ C ∞ (H),
v∈Q2N
where hτ ∈ SL2 (R) is any matrix sending i to τ ∈ H by Möbius transforma-

tions and j is the standard automorphy factor. The theta series is Γ-invariant
in the variable z, hence defines a form on XΓ = Γ\X. In τ it transforms like
a modular form of weight N for a congruence subgroup Γ′ ⊂ SL2 (Z) (that
depends on χ).
Remark 1.2. (See 3.3.2.) This form ϕ can also be seen the restriction of the
Kudla-Millson form along the embedding X ֒−→ D induced by ρ, where D
is the orthogonal locally symmetric space of signature (N, N ). The seesaw
relating the two forms is
SO(N, N )(Q) GL2 (Q)
GLN (Q) SL2 (Q).

Moreover, we have a diffeomorphism XΓ ≃ SΓ × R>0 , so that we can view
XΓ as fiber bundle p : XΓ −→ SΓ with fiber R>0 . The integration along the
fiber gives a form

(1.2) Eϕχ (z1 , τ, s) := p∗ Θϕχ (z, τ )u−s ∈ ΩN −1 (S)Γ ⊗ C ∞ (H),
where z = (z1 , u) is the variable on S × R>0 and we integrate over u.
Theorem 1.1. The pushforward (1.2) converges absolutely for any s ∈ C.

At s = 0 it defines a closed (N − 1)-form Eϕχ (z1 , τ ), which is invariant by
Γ ⊂ SLN (R) and transforms in τ like a modular form of weight N and level
Γ′ ⊂ SL2 (Z). In cohomology, it defines an element
∞
X
Eϕχ (z1 , τ ) = z(χ) + [Zn (χ)]q n ∈ H N −1 (SΓ ; C) ⊗ MN (Γ′ )
n=1
where [Zn (χ)] denotes the Poincaré dual to Zn (χ).
The constant term z(χ) ∈ H N −1 (SΓ ) is a canonical Eisenstein class ob-

tained by transgression of the Euler class, and for which explicit De Rham
representatives have been constructed by Bergeron-Charollois-Garcia.
For Re(s) > N , the form Eϕχ (z1 , τ, s) can be written as an Eisenstein
series. Let λ be the form
1 −1
λ := g dg + (dg t )g−t ∈ Ω1 (GLN (R)+ ) ⊗ End(RN )
2
with entries λij ∈ Ω1 (GLN (R)+ ). For a function σ : {1, . . . , N } −→ {1, . . . , N }
we define
λ(σ) := λ1σ(1) ∧ · · · ∧ λN σ(N ) ∈ ΩN −1 (GLN (R)+ ),
and for µ ∈ CN we set
µσ := µd11 · · · µdNN ,
where dm := |σ −1 (m)|. The function

s X g1−1 (vτ + w) σ
Eϕχ (g1 , τ, s)σ := Λ(s)y 2 F2 (χ)(v)
kg1−1 (vτ + w)k2N +s
v∈Q2N
in the variables (g1 , τ ) ∈ SLN (R) × H converges for Re(s) > N , where
s (−1)N −1 iN
Λ(s) := Γ N + s
2 2π N + 2
and F2 (χ) is a partial finite Fourier transform in the second variable. Let
u := det(g)1/N be the determinant function and let du/u be the GLN (R)+ -
invariant form. The form ιu ∂ λ(σ) is an (N − 1)-form on S. The following
∂u
follows from Poisson summation.
Theorem 1.2. When Re(s) > N we have

X
Eϕχ (z1 , τ, s) = Eϕχ (g1 , τ, s)σ ⊗ ιu ∂ λ(σ) ∈ ΩN −1 (S)Γ ⊗ C ∞ (H).
∂u
σ
Note that we see the right-hand side as an SO(N )-invariant form on

SLN (R) rather than a form on S = SLN (R)/SO(N ).
1.4. Theta lifts. The Kudla-Millson theta correspondence has proved use-
ful for deducing results regarding the span of the homology by special cy-
cles. There are striking applications (amongst others) of these results due to
Bergeron-Millson-Moeglin [BMM16] and Bergeron-Li-Millson-Moeglin [Ber+17].
Similarly, it is natural to wonder2 what part of the homology is generated
by the modular symbols Zn (χ).
The form Eϕχ (z1 , τ ) can be used as a theta kernel to define two lifts. On
the one hand, we have a lift
Z
′
(1.3) Eϕχ : HN −1 (SΓ ; C) −→ MN (Γ ), Eϕχ (Z, τ ) := Eϕχ (z1 , τ )
Z
analogous to the Kudla-Millson lift (1.1). Note that by Theorem 1.1, we
have
Z Z ∞
N(N−1)2 X
Eϕχ (z1 , τ ) = z(χ) + (−1) 2 [Z, Zn (χ)]q n ,
Z Z n=1
where [Z, Zn (χ)] is the intersection pairing on SΓ between the two cycles Z
and Zn (χ). On the other hand, by taking the Petersson inner product of a
cusp form with Eϕχ (z1 , τ ) we also get a lift
Λϕχ : SN (Γ′ ) −→ H N −1 (SΓ ; C).
There is a pairing
BM
(1.4) HN −1 (SΓ ; C) × H(N 2 −N )/2 (SΓ ; C) −→ C.
The following theorem is analogous to [KM88, Theorem. 4.5].
Theorem 1.3. We have

ker(Eϕχ ) ≃ span{Zn (χ) | n ∈ N>0 }⊥ ⊂ HN −1 (SΓ ; C),
where the orthogonal complement is taken with respect to the pairing (1.4).
When N > 2, under Poincaré duality H N −1 (SΓ ; C) ≃ H(N
BM
2 −N )/2 (SΓ ; C), we
also have
im(Λϕχ ) ≃ span{Zn (χ) | n ∈ N>0 }.
Remark 1.3. If χ is a test function such that χ(−v) = (−1)N −1 χ(v), then
the lift Eϕχ vanishes.
1.5. Evaluation on tori. Suppose that L = pZ2N and let χ ∈ C[DL ] be
a test function on DL = p−1 Z2N /pZ2N . Let us assume that it splits as
χ(v) = χ1 (v)χ2 (w), and let Γ ⊂ SLN (Z) be a congruence subgroup that
preserves χ. Ash and Rudolph [AR79] prove that the relative homology of
the Borel-Serre compactification S Γ is spanned by modular symbols
BM
ZQ ∈ ZN −1 (SΓ , Z),
that are translates of a split torus by an invertible matrix Q ∈ MatN (Z).

They are associated to Borel subgroups, whereas the modular symbols Zn (χ)
are associated to maximal parabolic subgroups.
2See [Maz, Question (a), p. 2]
In general, the period of Eϕχ (z1 , τ, s) over an arbitrary modular symbol

ZQ can have a pole at s = 0. However, for certain modular symbols ZQ that
are good for χ (see Definition 4.23) we have the following.
Theorem 1.4. If ZQ is a modular symbol that is good for χ, then the period
over ZQ converges and we have
Z X N
Y
(−1)N −1 iN a m τ + bm
Eϕχ (z1 , τ ) = F2 (χ)(Qv0 ) E1 τ, ,
ZQ π N pN p
v0 ∈Q2N /L m=1
where v0 = (a1 , . . . , aN , b1 , . . . , bN ) and the forms E1 (τ, λ0 ) are weight one

Eisenstein series.
On the other hand, one can evaluate the lift on a torus attached to a
totally real field F of degree N . Let m ⊂ O be an integral ideal and let
L = m ⊕ m, that we view as a lattice in R2N via the embeddings of F . By
restriction of scalars we obtain a compact cycle
Zm := U (m)1,+ \(F ⊗ R)1,+ ∈ HN −1 (SΓ ; Z)
where (F ⊗ R)1,+ are the totally positive elements of norm 1 and U (m)1,+
are the units that are congruent to 1 modulo m.
Theorem 1.5. The evaluation of the lift (1.3) on Zm is the diagonal restric-
tion Eχ (τ ) of a Hilbert-Eisenstein series Eχ (τ1 , . . . , τN )) of parallel weight one
for a subgroup of SL2 (O). Hence, it has the Fourier expansion
Z ∞
N(N−1)2 X
Eχ (τ ) = z(χ) + (−1) 2 [Zm , Zn (χ)]q n .
Zm n=1
The constant term is a linear combination of values at s = 0 of partial

zeta functions, and can be interpreted as a linking number in torus bundles
(see [Ber19] for the case of a real quadratic field).
Remark 1.4. The construction of the Eisenstein class in this paper uses the
Mathai-Quillen formalism and shares some similarities with the construction
of Bergeron-Charollois-Garcia in [BCG20; BCG23]. In [BCG20], inspired by
the existence of a similar cocycle due to Charollois-Sczech [CS16] and the
construction by Garcia [Gar18] of differential forms via superconnections,
the authors construct a lift
(1.5) EBCG : HN −1 (SΓ ) −→ MN (Γ′ ),
for some Γ′ ⊂ SL2 (Z). It follows from Theorem 1.4 and [BCG20, Theo-
rem. 6] that both lifts (1.3) and (1.5) are diagonal restrictions of Hilbert-
Eisenstein series, when evaluated on cycles attached to totally real field. On
the other hand, the computations in [BCG23] show that the periods along
modular symbols are products of weight one Eisenstein series (see [BCG23,
Théorème. 2.10] for example). Hence, although it is not completely imme-
diate by comparing the differential forms, this strongly suggests that the
two lifts agree. To get a more precise result one should show that the form
(N −1,0,N,0)
Eψ,x in [BCG20, Lemma. 28] is exactly the form Eϕχ considered in
this paper, in the case where χ is the characteristic function of a non-zero
torsion point x.
1.6. The case of the modular curve. In the special case where N = 2,
the symmetric space S is the upper half-plane H. Let p be a prime and
let L = pZ4 . For a suitable test function χ we can take Γ = Γ′ = Γ0 (p)
so that SΓ = Y0 (p) = Γ0 (p)\H. Let f1 , . . . , fr be a basis of normalized
(p)
newforms of S2 (Γ0 (p)), and E2 is the normalized Eisenstein series needed
(p)
to span M2 (Γ0 (p)). We define ωE := E2 (τ )dτ and ωf± := ωf ± ωf where
ωf = f (τ )dτ .
The Eisenstein lift specializes to
(1.6) Eϕχ : H1 (Y0 (p); C) −→ M2 (Γ0 (p)).
We will show that the special cycles
Zn (χ) = Tn {0, ∞}
are the Hecke translates of the modular symbol {0, ∞} from 0 to ∞, and we
deduce the following from Theorem 1.1.
Corollary 1.5.1. For Z ∈ H1 (Y0 (p); Z), the weight 2 modular form Eϕχ (C)
admits the Fourier expansion
Z X ∞
24
Eϕχ (Z, τ ) = ωE − [Z, Tn {0, ∞}]q n .
p−1 Z n=1
Remark 1.5. The constant term is obtained by comparing the constant term
in the spectral expansion below.
In particular, if we combine it with Theorem 1.5 we recover [DPV21,

Theorem. A] that expresses the diagonal restriction of Hilbert-Eisenstein
series as generating series of intersection numbers. Furthermore, following
[DPV21] we deduce a spectral expansion of the theta lift.
Corollary 1.5.2. For Z ∈ H1 (Y0 (p); Z), the modular form Eϕχ (Z) admits
the spectral expansion
Z X r Z
24 (p) L(fi , 1) +
Eϕχ (Z) = ωE E2 − ω fi f i .
p−1 Z iπkfi k2 Z
i=1
(p)
Let M02 (Γ0 (p)) ⊂ M2 (Γ0 (p)) be the subspace spanned by the eigenforms E2 ,
and fi for which L(fi , 1) 6= 0. The lift (1.6) is a surjective map
Eϕχ : H1 (Y0 (p); C) −→ M02 (Γ0 (p)).
Moreover, the lift is Hecke equivariant in the sense that Eϕχ (Tn Z) = Tn Eϕχ (Z)
when (n, p) = 1.
1.7. Outline of the paper.

• In Section 2 we define the special cycles Zn (χ) on the locally symmet-
ric space associated to SLN (R). They are finite linear combinations of
(quotients) of submanifolds Sv of codimension (N − 1) in S, which are
projections of submanifolds Xv of codimension N in X. The submanifold
Xv is the zero locus of a section sv of a bundle of rank N over X.
• After recalling the Mathai-Quillen formalism in 3.2, we use it in 3.3 and 3.4
to construct differential forms ϕ and ψ on the symmetric space X, valued
in S(R2N ). Evaluated at v, the differential form ϕ(v) is the pullback of
the Thom form by the section sv , so that it is a Poincaré dual to the
submanifold Xv . The projection formula (Proposition 3.8) relates the
integral of a form along the fiber p : X → S, with the projection X → S.
It shows that p∗ ϕ(v) is a Poincaré dual of Sv .
• In 4.1 we prove convergence results for certain theta series. This proves
the first part of Theorem 1.1. Using the differential forms and the theta
machinery we then define in 4.3 a differential form Θϕχ that transforms
like a modular form. The form Eϕχ on S is the pushforward of the theta
series Θϕχ along the fiber R>0 of the bundle p : X → S. In 4.4, we use
Poisson summation to show that the form Eϕχ is the Eisenstein form
given in Theorem 1.2. In 4.5 we show that the positive Fourier coefficients
are Poincaré duals to the special cycles, and that the negative Fourier
coefficients vanish in cohomology. This completes the proof of Theorem
1.1. In Section 4.7 we prove Theorem 1.3 (Theorem 4.21 in the text). In
4.9 we evaluate the lift on tori, yielding Theorems 1.4 and 1.5.
• In section 5 we specialize to the case N = 2 and prove Corollaries 1.5.1
and 1.5.2.
1.8. Acknowledgements. I thank Pierre Charollois, Henri Darmon and

Luis Garcia for helpful discussions around this paper and comments on earlier
versions.
2. Locally symmetric spaces and special cycles

2.1. Locally symmetric space of SLN (R). Let X be the space of real,
non-degenerate and positive-definite quadratic forms on RN . We view an
element z ∈ X as a positive definite symmetric matrix, on which the group
GLN (R)+ acts transitively by sending z to gzg t . The stabilizer of z = 1N is
K = SO(N ) so that we can identify
X ≃ GLN (R)+ /SO(N ).
2
It is a Riemannian manifold of dimension N 2+N . We also have the space
S ⊂ X of symmetric matrices of determinant one. Similarly, we have
S ≃ SLN (R)/SO(N ),
2
and the dimension of S is N +N 2
−2
. The action of GLN (R)+ on X restricts
to an action of SLN (R) on S.
Let ρ denote the action of GLN (Q) on Q2N defined by
v = (v, w) 7−→ ρg v := (gv, g −t w).
It preserves the pairing

B : QN × QN −→ Q, B(v) := v t w,
and gives a representation ρ : GLN (R)+ −→ SO(N, N ).
2.2. Special cycles in S. For v = (v, w) ∈ RN ⊕ RN we define the sub-
manifold
Xv := { z ∈ X| v = zw } ⊂ X.
The submanifold is of codimension N in X. Note that Xv is empty when
B(v) ≤ 0, since v = zw implies 0 < wt zw = wt v = B(v). The submanifold
Xv is invariant by scalar multiplication i.e Xλv = Xv . Moreover, for any
g ∈ GLN (R)+ we have
(2.1) gXv = Xρg v
since v = zw if and only if gv = (gzg t )g−t w. We can identify X ≃ S × R>0
by the diffeomorphism
1

X −→ S × R>0 , z 7−→ p(z), det(z)− N ,
1
where the projection p : X −→ S onto S is defined by p(z) = det(z)− N z.
The inverse of this map is
S × R>0 −→ X, (z1 , u) 7−→ z = z1 u.
Hence, if we identify R with R>0 via the exponential map, then we can view
X as a real vector bundle of rank one over S, which is GLN (R)+ -equivariant
in the sense that
(2.2) p(gz) = gp(z).
The image of Xv by the projection gives a submanifold
Sv := p(Xv ) ⊂ S
in S. Combining (2.1) and (2.2) we see that we have
gSv = Sρg v
for g ∈ SLN (R).
Remark 2.1. The cycle Sv only depends on the lines spanned by v and w.
We identify X with the bundle S × R>0 over S. A pair (z1 , u) ∈ S × R>0
lies in Xv exactly when v = uz1 w. By taking the length on both sides we
find that u is uniquely determined by z1 . In other words, if we restrict the
bundle S × R>0 to Sv , we can view Xv ⊂ Sv × R>0 as the image of the
section

kvk
Sv −→ Sv × R>0 , z1 7−→ z1 , .
kz1 wk
We deduce the following.
Proposition 2.1. The restriction of the projection p to Xv is a diffeomor-

phism onto Sv . Hence, the image Sv is a closed (embedded) submanifold of
S.
We can also view the submanifold Sv as an embedded copy of a smaller

symmetric space XN −1 := GLN −1 (R)+ /SO(N − 1) inside X. The setting is
similar to [KMS00; Sch93].
First, when v = (e1 , e1 ) with e1 = (1, 0, . . . , 0)t the standard basis vector,
the submanifold Xv consists of matrices of the form

1 0
z=
0 z′
where z ′ ∈ XN −1 is a symmetric positive-definite invertible matrix of size
N − 1. Hence, the submanifold Xv is the image of the embedding

1 0
j : XN −1 ֒−→ X, hSO(N − 1) 7−→ SO(N ).
0 h

v1 N −1 w1
For general v = (v, w), let us write v = ∈ R×R and w = .
v w
First, we suppose that w1 > 0. Let
√
v1 /√ n −wt /w1
mv := ∈ GLN (R)+
v/ n 1N −1
where n = B(v), and define the embedding jv : XN −1 ֒−→ X by

1 0
(2.3) jv (h) := mv .
0 h
Remark 2.2. Note that the inverse of mv is
√ √
−1 w1 / n wt/ n
(2.4) mv = ,
−vw1 /n 1N −1 − vw t /n
√ √
so that mv e1 = v/ n and m−t v e1 = w/ n. Hence, the matrix mv maps the
line spanned by e1 to the line spanned by v, and whose dual maps to the
line spanned by w.
Proposition 2.2. When w1 is positive, the submanifold Sv is the image of

p ◦ jv : XN −1 ֒−→ S,
where p is the projection from X to S.
Proof. Recall that for any matrix g ∈ GLN (R)+ we have gX(v,w) = X(gv,g−t w) .
Let

1 −wt /w1
g :=
0 1N −1
be the matrix such that g −t w1 e1 = w. If z ∈ X(v′ ,w1 e1 ) , then gz ∈ X(gv′ ,w) =
X(v,w) where

′ −1 n/w1
v := g v= ∈ R × RN −1 .
v
For v′ = (v ′ , w1 e1 ), with v ′ as above, the submanifold Xv′ consists of
positive definite invertible matrices of the form

n/w12 v t /w1
z=
v/w1 z′
for some (N −1)×(N −1)-matrix z ′ . In fact, it is the image of the embedding

jv′ : XN −1 ֒−→ X, hSO(N − 1) 7−→ jv′ (h)SO(N )
where jv′ is the embedding (2.3) and for v′ = (v ′ , w1 e1) the matrix mv′ is
√
n/w 1 0
mv ′ = √ ∈ GLN (R)+ .
v/ n 1N −1
Indeed, we see that

t 1 0 t n/w12 v t /w1
j (h)j (h) = m
v′ v′ v′ (mv ) =
′ ,
0 hht v/w1 vv t /n + hht
where vv t /n + hht is positive-definite and invertible.
So Xv = gXv′ is obtained by translating by g and is the image of the
embedding

′ 1 0
jv : XN −1 ֒−→ X, h 7−→ gjv (h) = mv
0 h
where the matrix mv is
√
v1 /√ n −wt /w1
mv = gm′v = ∈ GLN (R)+ .
v/ n 1N −1
Hence, the submanifold Sv is the image of
p ◦ jv : XN −1 ֒−→ S.

√
Note that det(mv ) = det(mv′ ) = n/|w1 | so that
1
|w1 | N 1 0
p ◦ jv (h) = √ mv .
n det(h) 0 h
Remark 2.3. If w1 is nonpositive, then one can compose the embedding with
a permutation matrix that will bring one of the positive wi ’s in the first
position. Not that at least one entry of w must be nonzero (otherwise Xv
is empty), and if all entries are negative we can replace Xv by X−v = Xv .
The submanifold Xv is then the translate by a matrix mv of the embedding
of the form
 
a 0 b
a b
j : XN −1 ֒−→ X, SO(N − 1) 7−→ 0 1 0 SO(N ).
c d
c 0 d
2.3. Quotients. Let L ⊂ RN be an integral lattice i.e. B(L, L) ⊂ Z and
let L = L ⊕ L ⊂ R2N . Let L∨ be the dual with respect to the bilinear form
Q(v1 , v2 ) = B(v1 , w2 ) + B(v1 , w2 )
on Q2N . As in the introduction we fix a function χ : DL −→ C that is
preserved by a torsion-free subgroup Γ. Hence, the quotient
XΓ := Γ\X
is a manifold. The group Γ only acts on S so that we have XΓ ≃ R>0 × SΓ
where
SΓ := Γ\S.
Example. (1) Let p be an integer and let L = pZ2N . If χ = 1v0 +L is the

characteristic function of vector v0 ∈ L r pZ2N , then χ is preserved by
the principal congruence Γ = Γ(p, N ) of matrices in SLN (Z) that are
congruent to the identity modulo p.
(2) If χ is the characteristic function of L × ZN where

t N
L = v = (v1 , v2 , · · · , vN ) ∈ Z (v1 , p) = 1, vi ∈ pZ for 2 ≤ i ≤ N
then χ is preserved by the subgroup Γ0 (p, N ) of upper triangular matrices

modulo p.
Proposition 2.3. For any positive integer n the set

∨
v ∈ Γ\L B(v) = n
is finite.
Proof. Let L∗ = Zα1 ⊕· · ·⊕ZαN be a Z-basis of L∗ and let g0 = [α1 , . . . , αN ] ∈

GLN (R)+ such that g0 ei = αi and g0−1 L∗ = ZN . Since L is an integral lat-
tice, we have B(αi , αj ) = B(g0t αi , ej ) ∈ Z, so that L′ := g0t L∗ ⊂ ZN . We
have a bijection
Γ\L∨ −→ g0−1 Γg0 \ZN ⊕ L′ , (v, w) 7−→ (g0−1 v, g0t w)
where g0−1 Γg0 ⊂ SLN (Z) is of finite index. Hence, the proposition follows
from the finiteness of

2N
v ∈ SLN (Z)\Z B(v) = n .
By the Hermite normal form, we can find γ ∈ SLN (Z) such that

D
γv = ∈ De1 ,
0N −1
where D = gcd(v1 , v2 , . . . , vN ). Thus, every orbit can be reduced to
′
D w1
SLN (Z)(v, w) = SLN (Z) , .
0N −1 w′
We have Dw1′ = v t w = n, so w1′ must be one of the divisors of n.
Reducing w2′ modulo w1′ we can write w2′ = qw1′ + r2 for some r2 ∈ Z/w1′ Z.
Then, for
 
1 q 0
γ = 0 1 0  ∈ SLN (Z)
0 0 1N −2
we have
   ′     ′ 
D w1 D w1
  ′    
ργ  0  , w2  =  0  ,  r2  .
0N −2 .
.. 0N −2 .
..
Recall that ργ acts by γ −t in the second factor. Reducing the remaining

entries w3′ , . . . , wN′ modulo w ′ , we see that every double coset can be reduced
1
to
′
D w1
SLN (Z)(v, w) = SLN (Z) ,
0N −1 r
where w1′ is a divisor of B(v) = n and r ∈ (Z/w1′ Z)N −1 .
2.3.1. Special cycles in the quotient. The stabilizer of v under the action ρ
is the subgroup

Γv := γ ∈ Γ | γv = v, wt γ = wt .
√ √
Recall from (2.4) that mv satisfies mv e1 = v/ n and m−t v e1 = w/ n. If
γ ∈ Γv then (m−1 t −1 t
v γmv )e1 = e1 and e1 (mv γmv ) = e1 . Thus, we have

−1 1 0
mv γmv = = j(γ ′ )
0 γ′
ev ⊂ SLN −1 (R) be the subgroup such that
for some γ ′ ∈ SLN −1 (R). Let Γ
e v )m−1 .
Γv := mv j(Γ v
The embedding p ◦ jv passes to the embedding of the quotient

e v \XN −1 ֒−→ Γv \S,
(2.5) p ◦ jv : Γ e v z 7−→ Γv jv (z) = Γv mv j(z)
Γ
with image Γv \Sv .
√
Remark 2.4. Although the coefficients of mv contain n, we have
√ √ √
−1 v1 /√ n −wt /w1 1 0 w1 / n wt / n
mv j(γ)mv =
v/ n 1N −1 0 γ −vw1 /n 1N −1 − vw t /n
t
1 0 1 w (γv − v) (w t − wt γ)v1
= + .
0 γ n (v − γv)w1 (v − γv)w t
e v is a congruence
Hence, if Γ is contained in SLN (Z) then the subgroup Γ
subgroup of SLN (Z).
Let us denote by Zv the image of the map
e v \XN −1 ֒−→ Γv \S −→ SΓ
jv : Γ
obtained by composing the map p ◦ jv in (2.5) with the projection of Γv \S
onto Γ\SΓ . This is the map of [KMS00, p. 116] and is proper. It defines a
locally finite cycle
Zv ∈ Z BM
N 2 −N (SΓ ; Z)
2
N 2 −N
of dimension 2 (and codimension N − 1) that depends only the equiva-
lence class v ∈ Γ\L /Z× . We then define for n ∈ N>0
∨
X
Zn (χ) := χ(v)Zv ∈ Z BMN 2 −N (SΓ ; C) .
2
v∈Γ\Q2N
B(v)=n
The sum is finite since χ is supported on L∨ .

Example. Let L = pZ2N and suppose that Γ = Γ(p2 ). For v = (e1 , e1 )

we have v1 = w1 = 1 and v = w = 0 and the congruence group Γ e v is
2
the principal congruence subgroup Γ(p , N − 1) ⊂ SLN −1 (Z) consisting of
matrices in SLN −1 (Z) that are congruent to the identity modulp p2 . More
generally, if v1 and w1 are both coprime to p2 n, then the congruence group
e v is the principal congruence subgroup Γ(p2 n, N − 1) ⊂ SLN −1 (Z). In
Γ
particular, we see that for N = 3 the special cycle Zv is the image of R>0 ×
Y (p2 n) where Y (p2 n) = Γ(p2 n)\H is a modular curve.
2.3.2. Orientability. The submanifold Xv is the zero locus of the function
fv : X −→ RN , z 7−→ zw − v.
The differential is dz fv (A) = Aw and the function is regular when B(v) > 0
(since it implies w 6= 0). At a point z ∈ Xv the tangent space splits as
Tz X ≃ Tz Xv ⊕ Nz Xv
where N Xv is the normal bundle. The kernel of the differential is Tz Xv so

that it induces an isomorphism Nz Xv ≃ T0 RN ≃ RN . Hence, an orientation
o(RN ) of RN determines an orientation o(Nz Xv ) of the normal bundle. So
after fixing an orientation o(Tz X) of X this determines an orientation of Xv
by the rule
o(Tz X) = o(Tz Xv ) ∧ o(Nz Xv )
The orientation of Sv then comes from the diffeomorphism Xv ≃ Sv . Note

that
o(Tz X−v ) = (−1)N o(Tz Xv ).
3. Thom forms and the Mathai-Quillen formalism

Let K = SO(N ) be the maximal compact subgroup of GLN (R)+ . Let
g = p ⊕ k be a Cartan decomposition of the Lie algebra g ≃ MatN (R)
of GLN (R)+ , where k = so(RN ) is the space of skew-symmetric matrices,
and p is the space of symmetric matrices. We consider the bundle E =
GLN (R)+ ×K RN . It consists of pairs [g, v] with the equivalence relation
[g, v] = [gk, k −1 v] and the projection map E → X is [g, v] 7→ gK. We view
it as a metric bundle with the metric B(v, v) = v t v = kvk2 . The bundle is
GLN (R)+ -equivariant, where GLN (R)+ acts on E by g[g ′ , v] = [gg ′ , v]. If
Γ ⊂ SLN (R) is a discrete subgroup as previously, then we have a real vector
bundle EΓ := Γ\E over XΓ , of rank N .
3.1. Thom isomorphism. Let Ωicv (EΓ ) be the space of i-forms with com-
pact vertical support i.e., with support contained in the disk bundle DEΓ ⊂
EΓ . The cohomology of this complex is the relative cohomology H i (EΓ , EΓ r
DEΓ ). Note that the relative cohomology H i (EΓ , EΓ r DEΓ ) is also the
cohomology of the complex Ωi (EΓ ) ⊕ Ωi−1 (DEΓ ) with boundary operator
δ(ω, α) = (dω, α − dω). Combining with the long exact sequence in relative
cohomology, we get the long exact sequence
H i−1 (EΓ r DEΓ ) H i (EΓ , EΓ r DEΓ ) H i (EΓ )
H i (EΓ r DEΓ ) H i+1 (EΓ , EΓ r DEΓ ) H i+1 (EΓ ).
The integration along the fibers of the bundle EΓ induces the Thom isomor-
phism
H i (EΓ , EΓ r DEΓ ) −→ H i−N (XΓ ).
The preimage of 1 ∈ H 0 (XΓ ) ≃ Z is a class
Th(EΓ ) ∈ H N (EΓ , EΓ r DEΓ )
called the Thom class. It can also be seen as a Poincaré dual of the zero
section (EΓ )0 in EΓ . It is represented by a closed differential form U ∈
ΩNrd (EΓ ) on EΓ , that has integral 1 along the fibers. The Mathai-Quillen
formalism is a canonical way to produce such a representative, that will
depend on the choice of a connection on the bundle that is compatible with
the metric.
If s : XΓ −→ EΓ is a smooth section such that s(XΓ ) is transversal to
the zero section, then by [GP74, p. 28] its zero locus s−1 (0) is a smooth
submanifold of XΓ of codimension N . The pullback s∗ Th(EΓ ) is the Poincaré
dual to the class [s−1 (0)].
3.1.1. Rapidly decreasing forms. A function φ ∈ C 0 (RN ) is said to be rapidly

decreasing if for every tuple m = (a1 , . . . , aN ) ∈ NN , we have
sup |v m φ(v)| < ∞
v∈RN
aN
where v m := v1a1 · · · vN .
Lemma 3.1. For a function φ ∈ C 0 (RN ) the following are equivalent

(1) The function φ is rapidly decreasing,
(2) for all M ∈ N there is a positive constant CM > 0 such that
1
|φ(v)| ≤ CM
(1 + kvk)M
for every v ∈ RN ,
′ > 0 such that for all v
(3) for all M ∈ N there is a positive constant CM
with kvk > 1 we have
′ 1
|φ(v)| ≤ CM .
kvkM
Proof. Suppose that φ is rapidly decreasing. By expanding (1 + kvk)M into

monomials we see that (1 + kvk)M |φ(v)| is bounded for every v by some
constant CM > 0. This proves that 1. implies 2., and it is immediate that 2.
implies 3. since (1 + kvk)−M ≤ kvk−M . Suppose that for all M we φ(v) =
O(kvk−M ) and let (a1 , . . . , aN ) ∈ NN . Let M := N max{a1 , . . . , aN }. Then

by the inequality of arithmetic and geometric means, we have
M kvkM
|v m | = |v1 |a1 · · · |vN |aN ≤ |v1 |2 · · · |vN |2 2N
≤ √ M.
N
For this M and v large (kvk ≥ 1) we then have
kvkM |φ(v)| C′
|v m φ(v)| ≤ √ M ≤ √ MM .
N N
It follows that
sup |v m φ(v)| < ∞.
v∈RN
kvk>1
On the other hand we have sup v∈RN |v m φ(v)| < ∞ since φ is continuous.
kvk≤1
A smooth function φ ∈ C ∞ (RN ) is in the space S(RN ) of Schwartz func-

tions if for every tuple m = (a1 , . . . , aN ) ∈ NN the derivative
∂vm φ(v) = ∂va11 · · · ∂vaNN φ(v)
is rapidly decreasing. By composition with the diffeomorphism
w
h : D −→ RN , w 7−→ p
1 − kwk2
from the unit disk D ⊂ RN onto RN we obtain a map from S(RN ) to the
space of smooth function on RN supported on D.
Let Ωird (EΓ ) ⊂ Ωi (EΓ ) be the complex of differential forms that are rapidly
decreasing in the fiber of EΓ . The pullback by h induces a map Ωird (EΓ ) −→
Ωicv (EΓ ). The Mathai-Quillen construction will produce a form in ΩN rd (EΓ )
N
whose image in Ωcv (EΓ ) will represent the Thom class.
3.2. Mathai-Quillen construction. In this section we explain how Mathai

and Quillen construct a differential form
+
U ∈ ΩN
rd (E)
GLN (R)
that is closed, GLN (R)+ -invariant and with integral 1 along the fiber. In
particular it will be Γ-invariant and descends to a form
U ∈ ΩN
rd (EΓ )
that represents the Thom class.
3.2.1. Some operations on vector bundles. The tautological bundle Etaut over
E is the pullback of E along the projection map E → X. More concretely,
we have
Etaut = GLN (R)+ ×K (RN × RN )
where the equivalence relation is [g, v, w] = [gk, k −1 v, k −1 w]. Let us consider
V V
the exterior product j Etaut = (GLN (R)+ × RN ) ×K j RN of the tauto-
logical bundle over E, where K acts by (gk, k−1 v) in the left factor and by
k−1 (v1 ∧ · · · ∧ vj ) = (k−1 v1 ) ∧ · · · ∧ (k−1 vj ) in the right factor. We also define
^ ^ j
^
+ N N
Etaut = (GLN (R) × R ) ×K R = ⊕N
j=0 Etaut .
V V
A differential i-form on E with values in j Etaut is an element in Ωi (E, j Etaut ).
It can also be seen as a basic form (K-invariant and trivial on vertical vec-
V
tors, see [BGV03, Proposition 1.9]) on GLN (R)+ ×RN with values in j RN .
Let us define
j
^ h ^j i
i,j i
Ω := Ω (E, Etaut ) ≃ Ωi (GLN (R)+ × RN ) ⊗ RN
bas
PN
and Ωi,• := j=0 Ωi,j . We define Ω•,j and Ω•,• similarly. We can equip Ω•,•
with the multiplication
Ωi,j × Ωk,l −→ Ωi+k,j+l
defined by
(ω ⊗ ν) ∧ (ω ′ ⊗ ν ′ ) := (−1)jk (ω ∧ ω ′ ) ⊗ (ν ∧ ν ′ ).
This makes Ω•,• an associative bigraded algebra over the ring of differential
forms on E. We define a positive definite bilinear form on ∧RN by
(
0 if j 6= l
B(v1 ∧ · · · ∧ vj , v1′ ∧ · · · ∧ vl′ ) = t ′
det(va vb )a,b if j = l.
With this bilinear form the splitting ∧RN = ⊕N j=0 ∧ R

j N is orthogonal.
Furthermore, since K preserves this bilinear form, it induces a bilinear form

on the bundle. The algebra structure allows us to define the exponential
map
N
X 1
exp : Ω•,• −→ Ω•,• , ω ⊗ ν 7−→ (ω ⊗ ν)k .
k!
k=0
Let e1 , . . . , eN be the standard basis of RN .

For a subset I = {i1 <
N
· · · < ik } of {1, . . . , N } let (R )I be the susbpace spanned by the vectors
ei1 , . . . , eiN . The monomials
eI := ei1 ∧ · · · ∧ eik
form an orthonormal basis of ∧RN . The Berezinian integral is the projection

onto the top dimensional component e1 ∧ · · · ∧ eN
Z B
: Ωi,• −→ ΩN (E), α = ω ⊗ ν 7−→ ω ⊗ B(ν, e1 ∧ · · · ∧ eN ).
Similarly, we have the projections

Ωi,• −→ Ωi (E), α = ω ⊗ ν 7−→ ωαI = ωB(ν, eI ).
For disjoint subsets I1 , . . . , Ik ⊂ {1, . . . , N } let define ǫ(I1 , . . . , Ik ) = ±1 by
eI1 ∧ · · · ∧ eIk = ǫ(I1 , . . . , Ik )eI1 ∪···∪Ik .
Lemma 3.2. Let α ∈ Ωi,• and β ∈ Ωk,• . We have

X
(α ∧ β)J = ǫ(I, J r I)(−1)k|I| αI ∧ βJ−I .
I⊂J
Proof. We can write

X X
α= αI1 ⊗ eI1 , β= βI2 ⊗ eI2 .
I1 ⊂{1,...,N } I2 ⊂{1,...,N }
Hence, we have
X
α∧β = (−1)k|I1 | αI1 ∧ βI2 ⊗ eI1 ∧ eI2
I1 ,I2 ⊂{1,...,N }
X
= ǫ(I1 , I2 )(−1)k|I1 | αI1 ∧ βI2 ⊗ eI1 ∪I2 .
I1 ,I2 ⊂{1,...,N }
Taking the J-th component will kill all the terms except when I1 and I2
satisfy I1 ∪ I2 = J.
3.2.2. Connection forms, covariant derivative and curvature form. Let π be

the projection of g onto k = so(RN ) defined by π(X) = 21 (X −X t ). Note that
π◦Ad(k) = Ad(k)◦π for k ∈ K, where the adjoint map is Ad(k)X = kXk−1 .
The Maurer-Cartan form is the 1-form
ϑ := g−1 dg ∈ Ω1 (GLN (R)+ ) ⊗ g,
and we define the connection form
1
θ := π(ϑ) = (ϑ − ϑt ) ∈ Ω1 (GLN (R)+ ) ⊗ k.
2
By pulling back by the projection map GLN (R)+ × RN −→ GLN (R)+ we
also get a connection form
1
θ = π(ϑ) = (ϑ − ϑt ) ∈ Ω1 (GLN (R)+ × RN ) ⊗ k.
2
The connection form defines a covariant derivative
∇ : Ωi,j −→ Ωi+1,j
by ∇(ω ⊗ ν) = (dω) ⊗ ν + (−1)i ω ⊗ (θν) where we view
θ ∈ Ω1 (GLN (R)+ × RN ) ⊗ k ⊂ Ω1 (GLN (R)+ × RN ) ⊗ so(∧RN ).
The connection is compatible with the metric induced by the bilinear form,
in the sense that for two sections s1 , s2 ∈ Ω0,j = Ω0 (E, ∧j Etaut ) we have
dB(s1 , s2 ) = B(∇s1 , s2 ) + B(s1 , ∇s2 ).
This can be seen from the fact that θ t = −θ, which implies B(θs1 , s2 ) +
B(s1 , θs2 ) = 0.
If we apply the covariant derivative twice we get ∇2 (ω ⊗ s) = ω ⊗ Rs
where
R ∈ [Ω2 (GLN (R)+ × RN ) ⊗ k]K
is the curvature form. We can identify k with ∧2 RN by the map

1X
(3.1) k −→ ∧2 RN , X 7−→ − Xij ei ∧ ej .
2
i,j
3.2.3. The Mathai-Quillen form. Let us denote by v : E −→ Etaut the tauto-

logical section defined v[g0 , v0 ] = [g0 , v0 , v0 ]. It defines an element v ∈ Ω0,1 .
Hence, we get a form
√
−2πkvk2 − 2 π∇v + R ∈ Ω0,0 ⊕ Ω1,1 ⊕ Ω2,2 .
The Mathai-Quillen form is the rapidly decreasing form defined by
Z B
N(N−1) N √ GLN (R)+
U := (−1) 2 (2π)− 2 exp −2πkvk2 − 2 π∇v + R ∈ ΩN
rd (E) .
Proposition 3.3. The form U is closed, GLN (R)+ -invariant and of integral
one along the fibers.
Proof. See Lemma 3.4 and Proposition 3.5 of [Bra23]. The fact that it is
closed come froms the fact that
Z B
√
d exp −2πkvk2 − 2 π∇v + R
Z B
√
= ∇ exp −2πkvk2 − 2 π∇v + R
Z B √ √
= (∇ + 2 πι(v)) exp −2πkvk2 − 2 π∇v + R
√ √
and that −2πkvk2 − 2 π∇v + R is annihilated by ∇ + 2 πι(v), where
ι(v) : Ωi,j −→ Ωi,j−1 is a contraction along the section v.
Proposition 3.4. Let E0 be the image of the zero section and let ω ∈
(N 2 −N )/2
Ωc (E) be a compactly supported form. Then
Z Z
ω∧U = ω.
E E0
√
Proof. Let Ut be the pullback of U by multiplication by t in the fibers, so
that U1 = U . By [MQ86] (see also Proposition 3.11) there is a transgression
form δt ∈ ΩN −1 (E r E0 ) such that
∂
dδt = Ut .
∂t
We can write
Z Z Z
ω ∧ Ut = ω ∧ Ut + ω ∧ Ut .
E E0 ErE0
Since the domain of the second integral is ErE0 we can use the transgression
form to show that the second integral is independent in t since
Z Z Z Z t1
∂
ω ∧ Ut1 − ω ∧ Ut0 = ω∧ Ut dt
ErE0 ErE0 ErE0 t0 ∂t
Z Z t1
= ω∧d δt dt
ErE0 t0
= 0.
In particular, we can take the limit as t goes to ∞. It follows from the rapid
decrease of Ut on E r E0 that
Z
lim ω ∧ Ut = 0.
t→∞ ErE
0
3.3. The form ϕ. For v = (v, w) ∈ R2N let us define the section by

g−1 v − g t w
sv : X −→ E, z 7−→ sv (z) := g, √
2
where z = gg t ∈ X. We define the form
ϕ0 (z, v) := s∗v U ∈ ΩN (X) ⊗ C ∞ (R2N ).
Since the section satisfies sv (gz) = gsρ−1
g v
(z) for any g ∈ GLN (R)+ , it is
Γv -equivariant and descends to a section
sv : Γv \X −→ Γv \E.
Hence, for fixed v ∈ R2N we have ϕ0 (z, v) ∈ ΩN (X)Γv .
(N 2 −N )/2
Proposition 3.5. Let ω ∈ Ωc (Γv \X) be a compactly supported
form. Then
Z √ Z
0
lim ω ∧ ϕ (z, tv) = ω.
t→∞ Γ \X Γv \Xv
v
√
Proof. Let Ut be the Thom form as in Proposition 3.4. It satisfies ϕ0 (z, tv) =
s∗v Ut . Let Ev ⊂ E be the image of sv . Using Proposition 3.4 we find that
Z Z
∗
ω ∧ sv Ut = s∗v (π ∗ ω ∧ Ut )
Γv \X Γv \X
Z
= π ∗ ω ∧ Ut .
Γv \Ev
We take the limit as in the proof of Proposition 3.4 and find that the latter
integral is equal to
Z Z
∗
π ω= ω.
Γv \Ev ∩E0 Γv \Xv
In the last line we used that the projection π : Ev ∩ E0 −→ Xv is a diffeo-

morphism.
Note that taking the Berezinian commutes with pullback, so that

N(N−1) Z
0 (−1) 2 −1 t 2
B √
ϕ (z, v) = N exp −πkg v − g wk exp(−2 π∇sv + R).
(2π) 2
3.3.1. Explicit formula. Let λ := 21 (ϑ + ϑt ) ∈ Ω1 (GLN (R)+ ) ⊗ End(RN ) and

let λij ∈ Ω1 (GLN (R)+ ) be its (i, j)-entry, where we identify End(RN ) ≃
MatN (R). For a subset I = {i1 < · · · < ik } ⊂ {1, . . . , N } and a function
σ : I −→ {1, . . . , N } we define the |I|-form
λ(σ) := λi1 σ(i1 ) ∧ · · · ∧ λik σ(ik ) ∈ Ω|I| (GLN (R)+ )
and the generalized Hermite polynomial Hσ ∈ C[RN ] by
N
Y
Hσ (v) := Hdm (vm )
m=1
where dm = |σ −1 (m)| and

d
d
Hd (t) := 2x − ·1
dt
denotes the single variable Hermite polynomial. This normalization of the
Hermite polynomial is sometimes called the physicist’s Hermite polynomials
and the first few Hermite polynomials are: H1 (t) = 2t, H2 (t) = 4t2 − 2,
H3 (t) = 8t3 − 12t, ...
Lemma 3.6. Let I ⊂ {1, . . . , N } be a subset. Then the I-th component is

√ |I|(|I|−1) |I| X √ −1
exp −2 π∇sv + R I = (−1) 2 2− 2 Hσ π(g v + gt w) ⊗ λ(σ),
σ
where the sum is over all functions σ : I −→ {1, . . . , N }.
Proof. For I = {1, . . . , N } this is [Bra23, Lemma. 4.4]. Note that d(g−1 )g =
−g −1 dg = −ϑ and d(gt )g −t = ϑt . Hence
ϑg−1 v + ϑt g t w
dsv = − √
2
and
g−1 v + g t w
∇sv = dsv + θsv = −λ √
2
P
where λ = 12 (ϑ + ϑt ). We write (g −1 v + gt w) = m (g−1 v + gt w)m ⊗ em to
get
XN
(g−1 v + g t w)m
∇sv = − √ λm ,
m=1
2
where
N
X
λm := λlm ⊗ el ∈ Ω1,1
l=1
and λlm = 21 (ϑlm + ϑml ). The curvature is the 2-form R ∈ [Ω2 (GLN (R)+ ×
RN ) ⊗ k]K given by R = dθ + θ 2 . Note that
1 1
θ 2 = (ϑ − ϑt )(ϑ − ϑt ) = (ϑ2 − ϑϑt − ϑt ϑ + (ϑt )2 )
4 4
and
(dg −1 )dg − (dg t )(dg−t ) ϑ2 + (ϑt )2

dθ = =− .
2 2
It follows that R = −λ2 . By the identification k ≃ ∧2 RN in (3.1), we can see

the curvature as an element R ∈ Ω2,2 = [Ω2 (GLN (R)+ × RN ) ⊗ ∧2 RN )]K .
We have
1X
R=− Rkl ek ∧ el
2
k,l
N
1 XX
=− λkm λml ⊗ ek ∧ el
2
m=1 k,l
N N
! N
!
1X X X
=− λlm ⊗ el ∧ λlm ⊗ el
2
m=1 l=1 l=1
N
X
1
=− λ2 .
2 m=1 m
For t ∈ R, we have
∞
X
√ 2 2−d/2 √
exp( 2πtλm − λm /2) = Hd ( πt)λdm .
d!
d=0
Following the computations after [Bra23, Lemma. 4.4] we write
N
Y
√ √ −1 t 1 2
exp −2 π∇sv + R = exp 2πλm (g v + g w)m − λm
2
m=1
N ∞
!
Y X 2−d/2 √
= Hd ( π(g−1 v + g t w)m )λdm
d!
m=1 d=0
N
X 2−(d1 +···+dN )/2 Y √
= Hdm ( π(g−1 v + gt w)m )λdmm .
d1 ! · · · dN !
d1 ,...,dN m=1
dm ≥0
The I-th component of λd11 ∧· · ·∧λdNN is only nonzero when d1 +· · ·+dN = |I|.
In that case, we have
 !dm 
N
^ N
X
(λd11 ∧ · · · ∧ λdNN )I =  λlm ⊗ el 
m=1 l=1
I
 
N
^ X
 
=
 dm ! (λl1 m ⊗ el1 ) ∧ · · · ∧ (λldm m ⊗ eldm )

m=1 Im ⊂I
|Im |=dm I
where the sum is over subsets Im = {l1 < · · · < ldm } ⊂ I of size dm and the
factorial comes from reordering the terms. Expanding the product shows
that
dN
X
(λd11 ∧ · · · ∧ λN )I = d1 ! · · · dN !(λI1 ⊗ eI1 ) ∧ · · · ∧ (λIN ⊗ eIN )
I1 ,...,IN
I=I1 ∪···∪IN
|I|(|I|−1) X
= (−1) 2 d1 ! · · · dN !
I1 ,...,IN
I=I1 ∪···∪IN
× (λI1 ∧ · · · ∧ λIN ) ⊗ (eI1 ∧ · · · ∧ eIN ).
To every partition of I into subsets I1 , . . . , IN corresponds a unique function
σ : I −→ {1, . . . , N } determined by σ(i) = m if and only if i ∈ Im . Note
that |σ −1 (m)| = |Im | = dm . Recall that ǫ(I1 , . . . , IN ) = ±1 is the sign that
appears when we reorder
eI1 ∧ · · · ∧ eIN = ǫ(I1 , . . . , IN )eI .
The same sign appears when we reorder
λI1 ∧ · · · ∧ λIN = ǫ(I1 , . . . , IN )λ(σ).
Hence, we conclude that
|I|(|I|−1) X
(λd11 ∧ · · · ∧ λdNN )I = (−1) 2 d1 ! · · · dN !λ(σ),
σ
|σ−1 (m)|=dm
where the sum is over all functions σ such that |σ −1 (m)| = dm for all m.
Since we sum over all nonnegative dm such that d1 + · · · + dN = |I|, we have
X X X
= .
d1 ,...,dN −1 σ σ
dm ≥0 |σ (m)|=dm

N(N−1) N
After multiplying with the exponential term and the constant (−1) 2 (2π)− 2
we get the following closed formula for ϕ0 .
Proposition 3.7. The form is explicitly

N X
ϕ0 (z, v) = 2−N π − 2 φ0σ (ρ−1 N ∞
g v) ⊗ λ(σ) ∈ Ω (X) ⊗ C (R
2N
)
σ
where z = gg t ,the sum is over functions σ : {1, . . . , N } −→ {1, . . . , N } and

√
φ0σ (v) := Hσ − π(v + w) exp −πkv − wk2 .
In the variable v ∈ R2N , the form ϕ0 is not rapidly decreasing. Hence, to

make it rapidly decreasing we define the form
ϕ ∈ ΩN (X) ⊗ S(R2N )
by setting
ϕ(z, v) := e−2πB(v,w) ϕ0 (z, v)
Z B
N(N−1) N √
= (−1) 2 (2π)− 2 exp −πkg−1 vk2 − kgt wk2 exp(−2 π∇sv + R)
N X
= 2−N π − 2 φσ (ρ−1
g v) ⊗ λ(σ)
σ
where
φσ (v, w) := φ0σ (v, w)e2πB(v)
√
= Hσ − π(v + w) exp −πkvk2 − πkwk2 .
3.3.2. Relation to the Kudla-Millson form. We refer to [Bra23] for more de-
tails on the construction of the Kudla-Millson form via the Mathai-Quillen
formalism. Let D ≃ SO(N, N )+ /SO(N ) × SO(N ) be the Grassmannian of
negative planes in the quadratic space R2N with the quadratic form B(v) of
signature (N, N ). The representation ρ induces an embedding

g 0
ρ : X ֒−→ D, g 7−→ .
0 g−t
The Kudla-Millson form is an N -form
ϕKM ∈ ΩN (D) ⊗ S(R2N ).
Let E ′ be the tautological rank N bundle E ′ = SO(N, N )+ ×SO(N )2 RN over
D, and let U ′ ∈ ΩN (E ′ ) be the Thom form. By [Bra23], there is a section
sev : D −→ E ′ such that ϕKM (v) = se∗v U ′ . Moreover, the pullback of E ′ by ρ
is precisely E and the following diagram is commutative:
E E′
sv sev
X ρ D.
Thus, we have
ϕ = ρ∗ ϕKM .
3.4. The form ψ. We have identified X ≃ S × R>0 as a bundle over S by

the map
1 1

X −→ S × R>0 , z 7−→ det(z)− N z, det(z) N
with inverse
S × R>0 −→ X, (z1 , u) 7−→ uz1 .
With respect to this splitting, a form η(z) ∈ ΩN (X) can be written

du
η(z) = ηe(z) + (−1)N −1 ιu ∂ η(z)
∂u u
where ηe(z) ∈ ΩN (X) and ιu ∂ η(z) ∈ ΩN −1 (X) are of degree 0 along R>0 .
∂u
Similarly to what we did in Section 3.1, the integration along the fibers
R>0 of the fiber bundle p : X −→ S induces the pushforward
Z ∞
N N −1 N −1 du
p∗ : Ωrd (X) −→ Ω (S), η 7−→ p∗ η := (−1) ιu ∂ η .
0 ∂u u
Note that the bundle can be seen as a vector bundle after identifying R>0 ≃
R via the exponential map. This definition of pushforward follows [BT82,
p. 61], and the next proposition is the content of [BT82, Proposition. 6.15].
dim(S)−k+1
Proposition 3.8. For η ∈ Ωkrd (X) and ω ∈ Ωc (S) we have the
projection formula
Z Z
(p∗ η) ∧ ω = η ∧ p∗ ω.
S X
Proof. The pullback p∗ ω does not depend on u. Hence, we have ιu ∂ p∗ ω = 0

∂u
and p∗ (η ∧ p∗ ω) = (p∗ η) ∧ ω. This implies that
Z Z Z
∗ ∗
η∧p ω = p∗ (η ∧ p ω) = (p∗ η) ∧ ω.
X S S
More generally, we can consider the pushforward of any form that is inte-
grable along the fiber.
Proposition 3.9. Let d and d1 be the exterior derivatives on X and S

respectively. For a form η ∈ ΩN (X) that is integrable along the fiber we have
p∗ (dη) = d1 p∗ η + lim ηe − lim ηe ∈ ΩN −1 (S).
u→∞ u→0
Proof. For rapidly decreasing forms it is the content of Proposition 6.14.1

in [BT82]. Let us write η = ηe + η ′ du N ′
u ∈ Ω (X), where η is the contraction
∂
along u ∂u . If η is integrable along the fiber, then we have
Z ∞
du
p∗ η = η′ .
0 u
Then

∂ du
dη = d1 ηe + ηe du + d1 η ′ .
∂u u
Hence, we have
Z ∞ Z ∞
∂ du du
p∗ dη = u ηe + d1 η ′ = d1 η′ + lim ηe − lim ηe.
0 ∂u u 0 u u→∞ u→0

Let ψ(z, v) be the form
ψ(z, v) := (−1)N −1 ιu ∂ ϕ(z, v) ∈ ΩN −1 (X) ⊗ S(R2N )
∂u
so that the form ϕ(z, v) splits as

du
ϕ(z, v) = ϕ(z,
e v) + ψ(z, v) .
u
Moreover, since the form ϕ is closed, by taking the derivative on both sides
we find the transgression formula
∂
(3.2) e v) ∈ ΩN (X) ⊗ S(R2N )
d1 ψ(z, v) = u ϕ(z,
∂u
where d1 is the derivative on S.
3.4.1. Explicit formula of ψ 0 . To compute this form, note that we can write
the connection ϑ = g−1 dg in the coordinates (g1 , u) of SLN (R) × R>0 where
g = ug1 . We have
du
g−1 dg−1 = u−1 g1−1 d(g1 u) = ϑ1 +
u
−1 1

(1)
where ϑ1 = g1 dg1 . If we set λ := 2 ϑ1 + ϑ1 ∈ Ω (SLN (R)) ⊗ End(RN )
t 1
then
du
λ = λ(1) + .
u
By taking the product we can then write
du
λ(σ) = λ(1) (σ) + (−1)N −1 ιu ∂ λ(σ)
∂u u
where ιu ∂ λ(σ) ∈ Ω N −1 (S).
∂u
Proposition 3.10. The form ψ 0 (z, v) ∈ ΩN −1 (X) ⊗ S(R2N ) is equal to

N X

ψ 0 (z, v) = (−1)N −1 2−N π − 2 φ0σ (ρ−1
g v) ⊗ ιu ∂ λ(σ) .
∂u
σ
(1)
Remark 3.1. Note that λij = λij + δij du
u . Hence, we have
N
X
ιu du λ(σ) = bmσ(m) · · · λN σ(N ) ,
(−1)m λ1σ(1) · · · λ
u
m=1
σ(m)=m
where the symbol b means that we remove it from the product.

3.5. The transgression form. We have seen that the form ψ already sat-
isfies the transgression formula (3.2) with respect to R>0 acting on v by ρ.
We will need a transgression form with respect to the action of R>0 given
by rescaling the vector.
As in Proposition 3.4 we define
√ √
Ωt := −2πtkvk2 − 2 π t∇v + R
for t > 0, and the Thom form
Z B
N(N−1) N +
Ut = (−1) 2 (2π)− 2 exp(Ωt ) ∈ ΩN
rd (E)
GLN (R)
√
obtained by rescaling by et be
t in the fiber of E. Note that U1 = U . Let U
the form defined by
Z B
N(N−1)
et := −(−1) −N − N−1 N −1 +
U 2 22 π 2 v ∧ exp(Ωt ) ∈ Ωrd (E)GLN (R)
where v : E → Etaut is the tautological section as in section 3.2.3.
Proposition 3.11. We have

∂Ut 1 e
= √ dUt.
∂t t
Proof. This is [BGV03, Proposition. 1.53] with a slightly different normal-
ization. We have

∂ ∂Ωt
exp(Ωt ) = exp(Ωt )
∂t ∂t
and
√
∂Ωt π √ √
= − √ ∇ + 2 π tι(v) v.
∂t t
√ √
As in Proposition 3.3 we have ∇ + 2 π tι(v) exp(Ωt ) = 0, hence
Z B Z B
√ √
d v ∧ exp(Ωt ) = ∇ + 2 π tι(v) (v ∧ exp(Ωt ))
√ Z B
t ∂
= −√ exp(Ωt )
π ∂t
√ Z B
t ∂
= −√ exp(Ωt ).
π ∂t
N(N−1) N
The result follows after multiplying by (−1) 2 (2π)− 2 on both sides.
Let sv : X −→ E be the section considered previously. Let U e := Ue1 and
0 ∗ e
define α (z, v) := sv U ∈ Ω N −1 ∞ 2N
(X) ⊗ C (R ). We have
Z B
N(N−1) N N−1
α0 (z, v) = −(−1) 2 2− 2 π − 2 sv (z) ∧ exp(s∗v Ω1 ).
For fixed vector v ∈ R2N we can view

α0 (z, v) ∈ ΩN −1 (X)Γv .
Proposition 3.12. For any real number t > 0 we have

√ ∂ √
(3.3) dα0 (z, tv) = t ϕ0 (z, tv).
∂t
√
Proof. Since s∗v Ωt = s∗√tv Ω1 , we have s∗v Ut = s∗√tv U = ϕ0 (z, tv). On the
other hand, we have
Z B
N(N−1)
et = −(−1) 2 2− N2 π − N−1
s∗v U 2 sv (z) ∧ exp(s∗v Ωt )
1 e
= √ s∗√tv U
t
1 √
= √ α0 (z, tv).
t
Hence, we have
√ √ √
et ) = t ∂ s∗ Ut = t ∂ ϕ0 (z, tv).
dα0 (z, tv) = td(s∗v U v
∂t ∂t

3.5.1. Explicit formula for α0 . For l ∈ {1, . . . , N } and a function

σl : {1, . . . , N } r {l} −→ {1, . . . , N }
we define a smooth function on R2N
√
φ0σl (v) := (vl − wl )Hσl ( π(v + w)) exp −πkv − wk2
N
Y √
= (vl − wl ) H|σ−1 (m)| − π(vm + wm ) exp −π|vm − wm |2 .
l
m=1
Proposition 3.13. The form α0 (z, v) ∈ ΩN −1 (X) ⊗ C ∞ (R2N ) is equal to

N X
X
0 −N − N−1
α (z, v) = −2 π 2 (−1)l−1 φ0σl (ρ−1
g v) ⊗ λ(σl ),
l=1 σl
where the inner sum is over functions σl : {1, . . . , N } r {l} −→ {1, . . . , N }.
Proof. We have
Z N Z
B
1 X B √
sv (z) ∧ exp(s∗v Ω1 ) =√ ((g−1 v − gt w)l ⊗ el ) ∧ exp −2 π∇sv + R .
2 l=1
Using3 Lemma 3.2 for α = sv (z) and β = exp(s∗v Ω1 ) we find that
Z N
B
(−1)N −1 X
sv (z) ∧ exp(s∗v Ω1 ) = √ (−1)l−1 (g −1 v − gt w)l exp(s∗v Ω1 ){1,...,N }r{l} .
2 l=1
3Note that here ǫ({l}, {1, . . . , N } r {l}) = (−1)l−1 and (−1)k|{l}| = (−1)N−1 .
By Lemma 3.6, for I = {1, . . . , N } r {l} we have
√ (N−1)(N−2) N−1 X √ −1
exp −2 π∇sv + R I = (−1) 2 2− 2 Hσl π(g v + g t w) ⊗ λ(σl ),
σl
where the sum is over all functions σl : I −→ {1, . . . , N }. Combining the two
gives
Z B N X
X
N(N−1) N
sv (z) ∧ exp(s∗v Ω1 ) = (−1) 2 2− 2 (−1)l−1 φ0σl (ρ−1
g v) ⊗ λ(σl ).
l=1 σl
N(N−1) N N−1
The result follows by multipliying by the constant (−1) 2 2− 2 π − 2
and the exponential term exp −πkv − wk2 .
4. Theta lift
4.1. Weil representation, intertwiners and theta series. We consider
the Weil representation of GLN (R)×SL2 (R) on the space S(R2N ) of Schwartz
functions in two different models. Let e : R× −→ U (1) be the character
e(t) := e2iπt .
In the first model the representation ω : GLN (R)×SL2 (R) −→ U (S(R2N ))
is defined by the formulas
ω (g, h) φ(v) = ω(h)φ(ρ−1
g v),

a 0
ω 1, φ(v) = |a|N φ(av), a ∈ R× ,
0 a−1

1 n
ω 1, φ(v) = e (nB(v)) φ(v), n ∈ R,
0 1
Z
0 −1
ω 1, φ(v) = φ(v′ )e −B(v ′ , w) − B(v, w′ ) dv′ ,
1 0 R2N
where v′ = (v ′ , w′ ) and v = (v, w). In this orthogonal-symplectic model, the
quadratic form is Q(v, v′ ) = B(v ′ , w) + B(v, w′ ).
A second model ω ′ : GLN (R) × SL2 (R) −→ U (S(R2N )) for the Weil rep-
resentation is given by by
ω ′ (g, h)φ(v) := | det(g)|−1 φ(g−1 vh)
where h acts on the columns. The two representations are intertwined by
partial Fourier transform. The partial Fourier transform in the second vari-
able
F2 : S(R2N ) −→ S(R2N )
is defined by
Z
F2 (φ)(v, w) := φ(v, w′ )e(B(w, w′ ))dw′ .
RN
Similarly, we define the Fourier transform in the first variable
Z
F1 (φ)(v, w) := φ(v ′ , w)e(B(v, v ′ ))dv ′ .
RN
Lemma 4.1. We have

F2 (ω(g, h)φ)(v) = ω ′ (g, h)F2 (φ)(v)
and
F1 (ω(g, h)φ)(v) = ω ′ (g∗ , h∗ )F1 (φ)(v),
where g∗ = g −t and h∗ is defined by
∗
a b 0 1 a b 0 1 d c
= = .
c d 1 0 c d 1 0 b a

a na−1
Proof. We check it for a matrix h = . We have
0 a−1
Z

F2 (ω(g, h)φ)(v, w) = ω(g, h)φ(v, w′ )e B(w, w′ ) dw′
RN
Z
N

= |a| φ(ag −1 v, ag t w′ )e B(nv + w, w′ ) w′
RN
Z
−1 −1 −1 nv + w
= | det(g)| φ(ag v, u)e B g ,u du
RN a

nv + w
= | det(g)|−1 F2 (φ) ag −1 v, g−1
a
′
= ω (g, h)F2 (φ) (v, w) .

Let L = L × L as previously, and let χ ∈ C[DL ] be a test function. Note
that DL ≃ DL × DL where DL := L∗ /L. We can also define a finite partial
Fourier transform
F2 : C[DL ] −→ C[DL ],
by taking a finite Fourier transform in the second variable
1 X
F2 (χ)(v0 , w0′ ) := p χ(v0 , w0 )e(B(w0 , w0′ )).
|DL | w0 ∈D
L
We define F1 : C[DL ] −→ C[DL ] analogously. We can combine the two

Fourier transforms and define
F2 : S(R2N ) ⊗ C[DL ] −→ S(R2N ) ⊗ C[DL ].
We will abuse the notation and write φχ instead of φ ⊗ χ. The Fourier
transform is
F2 (φχ) = F2 (φ)F2 (χ).
For (g, h) ∈ GLN (R)+ × SL2 (R) we define the theta series
X X X
Θφχ (g, h) := ω(g, h)φ(v)χ(v) = χ(v0 ) ω(g, h)φ(v).
v∈Q2N v0 ∈DL v∈v0 +L
Using the second model of the Weil representation, we define

X X
Θ′φχ (g, h) := χ(v0 ) ω ′ (g, h)φ(v).
v0 ∈DL v∈v0 +L
Proposition 4.2 (Poisson summation). We have
Θφχ (g, h) = Θ′F2 (φχ) (g, h) = Θ′F1 (φχ) (g∗ , h∗ ).
Proof. Let φe := ω ′ (g, h)φ. Let v0 , w0 ∈ DL be the entries of v0 . Applying

the Poisson summation in the variable shows that
X X X
e w) =
φ(v, e w0 + w)
φ(v,
(v,w)∈(v0 ,w0 )+L v∈v0 +L w∈L
1 X X
=p e(B(w0 , w0′ )) e
F2 (φ)(v, w).
|DL | w′ ∈D ′
(v,w)∈(v ,w )+L
0 L 0 0
The result follows by taking the sums over all v0 ∈ DL twisted by χ, and
the fact that F2 intertwines the two representations. The second equality is
obtained in the same way, by applying the Poisson summation formula to
the first variable.
Let us write g = g1 u where (g1 , u) ∈ SLN (R) × R>0 . We want to consider

the integral along R>0 .
Proposition 4.3. The theta series converges absolutely and uniformly for
(g, h) in a compact set A ⊂ GLN (R)+ × SL2 (R). Moreover, we have
Θφχ (g, h) − F2 (φχ)(0) = O(uM ) u→0
for every M ∈ N and
Θφχ (g, h) − F1 (φχ)(0) = O(u−M ) u→∞
for every M ∈ N.
Proof. Since A is a compact set we can uniformly bound the Schwartz func-
tion ω(g, h)φ(v). More precisely, for any positive integer M there exists a
positive constant CM,A such that
1
|ω(g, h)φ(v)| ≤ CM,A
(1 + kvk)M
for all (g, h) ∈ A. We have

 
X  X 1 
|ω(g, h)φ(v)χ(v)| ≤ CM,A max |χ| 
 1 + 
kvkM 
v∈L∨ v∈L∨
v6=0
where the sum converges for M large enough (M > 2N ). This shows the
absolute convergence of the theta series. For M > 2N and any u ∈ R>0 we
have
u−M |Θφχ (g, h) − F2 (φχ)(0)| = u−M Θ′F2 (φχ) (g, h) − F2 (φχ)(0)

X
< u−M max |F2 (χ)| |ω ′ (g, h)F2 (φ)(v)|
v∈L∨
v6=0
X u−M
≤ CM,A max |F2 (χ)|
(1 + u−1 kvk)M
v∈L∨
v6=0
X 1
≤ CM,A max |F2 (χ)|
kvkM
v∈L∨
v6=0
where the sum on the right-hand side converges since M > 2N . Similarly,
we find that
Θφχ (g, h) − F1 (φχ)(0) = Θ′F1 (φχ) (g ∗ , h∗ ) − F1 (φχ)(0)
is rapidly decreasing as u → ∞.
The archimedean Schwartz functions that we will consider later satisfy
F1 (φ)(0) = F2 (φ)(0) = 0; see Proposition 4.9. In particular, in that case the
theta series is rapidly decreasing at 0 and ∞ and we get the following.
Proposition 4.4. Suppose that F1 (φ)(0) = F2 (φ)(0) = 0. The integral

Z ∞
du
Θφχ (g, h)
0 u
converges absolutely and uniformly for (g1 , h) in a compact set A ⊂ SLN (R)×
SL2 (R). Hence, it defines a smooth function in (g1 , h).
Proof. We split the integral as

Z ∞ Z 1 Z ∞
du du du
|Θφχ (g, h)| = |Θφχ (g, h)| + |Θφχ (g, h)|
0 u 0 u 1 u
and use the bounds from the previous proposition. Since |Θφχ (g, h)| ≤
CM,A uM for (g1 , h) ∈ A and some CM,A > 0, we have
Z 1
du CM,A
|Θφχ (g, h)| ≤
0 u M
and similarly for the other integral. The smoothness of the function follows
from applying the dominated convergence theorem.
However, the integral does not converge termwise absolutely, i.e. the
integral
Z ∞ X
du
|ω(g, h)φ(v)χ(v)|
0 2N
u
v∈Q
does not converge and we cannot exchange the sum and the integral. In the
next two subsections we do two things: first one can regularize the integral
by adding a term u−s and apply Poisson summation. For Re(s) large enough
the integral is termwise absolutely convergent and we can change the order
of summation and integration, yielding the Eisenstein series. On the other

hand, we can also split the theta series into singular and regular terms. For
the regular terms the series is convergent and we can exchange the sum and
the integral. This will give the Fourier expansion of the Eisenstein series.
4.1.1. Regularized integral. Note that by Poisson summation we have

Z ∞ Z ∞
du du
Θφχ (g, h) = Θ′F2 (φχ) (g, h) .
0 u 0 u
By the following Proposition, when F1 (φχ)(0) = F2 (φχ)(0) = 0 and Re(s) >
N we can write
Z ∞ X Z ∞
du du
Θφχ (g, h)u−s = ω ′ (g, h)F2 (φχ)u−s .
0 u ∨ 0
u
v∈L
Since the left-hand side converges for every s ∈ C, it provides an analytic

continuation of the right-hand side.
Proposition 4.5. Suppose that F1 (φχ)(0) = F2 (φχ)(0) = 0. Then for

Re(s) > N the integral
X Z ∞ du
|ω ′ (g, h)F2 (φχ)(v)u−s |
2N 0 u
v∈Q
converges absolutely and uniformly for (g1 , h) in a compact set A ⊂ SLN (R)×
SL2 (R).
Proof. Let ν = Re(s). We show the convergence of

X Z ∞ du X
|ω ′ (g, h)F2 (φχ)|u−ν = |F2 (χ)(v)|
0 u
v∈Q2N v∈Q2N
Z ∞
du
(4.1) × |ω ′ (g, h)F2 (φχ)(v)|u−ν .
0 u
For every M there is constant CM > 0 such that
1
|F2 (φ)(v)| ≤ CM .
(1 + kvk)M
Let us choose M such that M > N + ν where ν = Re(s). For (g1 , h) in a
compact set A ⊂ SLN (R) × SL2 (R) we can uniformly bound
1
|ω ′ (g1 , h)F2 (φ)(v)| ≤ CM,A
(1 + kvk)M
where CM,A is a positive constant. Hence we can bound
Z ∞ Z ∞
du du
|ω ′ (g, h)F2 (φ)(v)|u−ν = |ω ′ (g1 , h)F2 (φ)(u−1 v)|u−N −ν
0 u 0 u
Z ∞
1 du
≤ CM,A −1 M
u−N −ν
0 (1 + u kvk) u
1
(4.2) = CM,A β(N + ν, M − N − ν)
kvkN +ν
where the Beta function is4

Z ∞
αx1 dα
β(x1 , x2 ) := x +x
, x1 , x2 ∈ R
0 (1 + α) 1 2 α
and converges for x1 , x2 > 0. By our assumption M > N + ν, the integral
(4.2) converges when ν > −N .
Recall that χ and F2 (χ) are supported on L∨ . Since F2 (φχ) = 0, the
vector 0 is removed from the summation and Equation (4.1) can be bounded
by
X Z ∞
′ −ν du
|F2 (χ)| |ω (g, h)F2 (φ)(v)|u
0 u
v∈L∨
X 1
≤ max |F2 (χ)|CM,A β(N + ν, M − N − ν)
∨
kvkN +ν
v∈L r{0}
which converges for ν > N . We deduce that the sum converges absolutely
and uniformly on compact sets for Re(s) > N . The same holds for the
derivative of F2 (φ) in g1 and h so that the fuction is smooth.
4.1.2. Regular and singular terms. In order to compute the Fourier expan-
sion we will need to split the sum into regular and singular terms. We will
say that a vector v = (v, w) is regular if v and w are both nonzero, otherwise
we will say that v is singular. We will suppose in this section that φχ(0) = 0,
so that the vector 0 is removed from the summation. We write
L∨ r {0} = L∨ ∨ ∨
sing,1 ⊔ Lsing,2 ⊔ Lreg
where

L∨ ∨ ∨ ∨
sing,2 := v ∈ L | v = 0, w 6= 0 , Lsing,1 := v ∈ L | v 6= 0, w = 0 ,

L∨ ∨
reg := v ∈ L | v 6= 0, w 6= 0 .
Proposition 4.6. Suppose5 that φχ(0) = F1 (φχ)(0) = F2 (φχ)(0) = 0.

Then, we have
Z ∞ Z ∞ X
du du
Θφχ (g, h) = ω(g, h)φχ(v)
0 u 0 v∈L∨ u
sing,1
Z ∞ X
du
+ ω ′ (g, h)F2 (φχ)(v)
0 v∈L∨ u
sing,2
X Z ∞ du
+ ω(g, h)φχ(v) .
∨ 0 u
v∈Lreg
4If we set t = α/(1 + α) then we have the usual expression β(x , x ) = R 1 tx1 −1 (1 −
1 2 0
t)x2 −1 dt.
5By Proposition 4.9, the Schwartz functions that we will consider all satisfy F (φ)(0) =
1
F2 (φ)(0) = 0, and we have assumed throughout this paper than χ(0) = 0.
Moreover, for Re(s) > N we have

Z ∞ X Z ∞
du du
Θφχ (g, h)u−s = ω(g, h)φχ(v)u−s
0 u ∨ 0 u
v∈Lsing,1
X Z ∞
du
+ ω ′ (g, h)F2 (φχ)(v)u−s
0 u
v∈L∨
sing,2
X Z ∞
du
+ ω(g, h)φχ(v)u−s .
0 u
v∈L∨
reg
Proof. Since 0 is removed from the summation and since the theta series is
absolutely convergent, we can split the theta series as
X X X
Θφχ (g, h) = ω(g, h)φχ(v) + ω(g, h)φχ(v) + ω(g, h)φχ(v).
v∈L∨
sing,1 v∈L∨
sing,2
v∈L∨
reg
Since F2 is the Fourier transform in the variable w, we can apply Poisson

summation to the second singular sum:
X X
ω(g, h)φχ(v) = ω ′ (g, h)F2 (φχ)(v).
v∈L∨ v∈L∨
v=0 v=0
For the first part, the convergence of the two singular integral follows from
Proposition 4.3. The change of the order of summation and integral for
the regular term is justified by Lemma 4.7. For the second part of the
proposition, the order of the integral and sum in the singular terms can be
changed for Re(s) > N by Lemma 4.8.
Lemma 4.7. For any s ∈ C, the integral

Z ∞ X
du
ω(g, h)φχ(v)u−s
0 ∨
u
v∈Lreg
converges uniformly for (g1 , h) in a compact subset of SLN (R) × SL2 (R).
Proof. We want to show the convergence of

Z ∞ X
du
|ω(g, h)φχ(v)| ,
0 ∨
u
v∈Lreg
which is equivalent to show the convergence of

X Z ∞ du
|ω(g, h)φχ(v)| .
∨ 0 u
v∈Lreg
Let A ⊂ SLN (R) × SL2 (R) be a compact set. For any integer M we can
uniformly bound for any (g1 , h) ∈ A
1
|ω(g1 , h)φ(v)| ≤ CM,A
(1 + kvk)M
where CM,A > 0 is a positive constant depending on M and A. Hence, we

have
1
|ω(g, h)φ(v)| ≤ CM,A
(1 + kρ−1
u vk)
M
1
≤ CM,A p
(1 + u kvk + u2 kwk|)M
−2
1
≤ CM,A −2
(u kvk + u2 kwk2 )M/2
2
where v = (v, w). Let ν = Re(s) and let M be large enough so that M ≥ |ν|.
We can bound the integral by
Z ∞ Z ∞
du u−ν du
ω(g, h)φ(v)u−s ≤ CM,A −2 kvk2 + u2 kwk2 )M/2 u
0 u 0 (u
Z ∞
CM,A 1 α−ν/2 dα
≤
2 kwk(M −ν)/2 kvk(M +ν)/2 0 (α−1 + α)M/2 α

CM,A 1 M −ν M +ν
≤ β , ,
4 kwk(M −ν)/2 kvk(M +ν)/2 4 4
where we used the substitution u2 = αkvk/kwk, and the right-hand side
converges since M − ν and M + ν are both positive.
Recall that the Weierstrass M -test [Rud76, Theorem. 7.10] asserts that
if a family of function Fd (g1 ,P
h) is uniformly bounded P by |Fd (g1 , h)| ≤ Md
by a sequence Md such that d Md converges, then d Fd (g1 , h) converges
uniformly (and absolutely). When (g1 , h) is in compact set A we have
X Z ∞ du
Fd (g1 , h) := ω(g, h)φ(v)u−s
u
v∈Nd 0

CM,A M −ν M +ν |Nd |
≤ β , M −|ν|
4 4 4 d 2
where

Nd := v ∈ L∨ reg d ≤ kvkkwk ≤ d + 1 .
The cardinality |Nd | is finite and grows polynomially in d. Hence, the sum
X Z ∞ du
∞
X
|ω(g, h)φχ(v)| ≤ max |χ| Fd (g1 , h)
∨ 0 u
v∈Lreg d=1
converges absolutely and uniformly on compact sets.
Lemma 4.8. The integrals

Z ∞ X
du
|ω(g, h)φχ(v)u−s |
0 u
v∈L∨
sing,i
converge uniformly for (g1 , h) in a compact sets A ⊂ SLN (R) × SL2 (R) for
Re(s) > N when i = 1 (resp. Re(s) < −N when i = 2).
Proof. Let v = (v, 0) be a singular vector with v 6= 0, and suppose that

ν = Re(s) > N . For every integer M > ν we have a positive constant CM,A
such that
1
|ω(g, h)φ(v)| ≤ CM,A .
(1 + u−1 kvk)M
Thus, we can bound
Z ∞ Z ∞
du u−ν du CM,A
|ω(g, h)φ(v)| u−ν ≤ CM,A = β(ν, M − ν).
0 u 0 (1 + u−1 kvk)M u kvkν
The β-function β(ν, M − ν) converges, since M − ν > 0 and ν > 0. Hence,
the sum
X Z ∞ du X 1
|ω(g, h)φχ(v)| u−ν ≤ CM,A β (ν, M − ν) max |χ|
∨ 0 u ∗
kvkν
v∈Lsing,1 w∈L
v6=0
converges since ν > N . Similarly, for a singular vector v = (0, w) with w 6= 0

we get
X Z ∞ X
du 1
|ω(g, h)φχ(v)| u−ν ≤ CM,A β (−ν, M + ν) max |χ| ,
0 u ∗
kvk−ν
v∈L∨
sing,2
w∈L
v6=0
where the right-hand side converges for ν < −N .
4.2. Computation of the partial Fourier transforms. We will need to

compute the Fourier transforms of the functions φσ and φσl that appear
in the explicit expressions of the differential forms ϕ and α. For a vector
µ ∈ CN and a function σ partitioning {1, . . . , N } let us write
µσ := µd11 · · · µdNN
where dm = |σ −1 (m)|.
Proposition 4.9. The partial Fourier transforms of φσ and φσl are

N
F2 (φσ )(v) = iN 2N π 2 (vi + w)σ exp −πkvk2 − πkwk2
and
" #
N N −1 N−1 2π|vl i + wl |2 + |σl−1 (l)|
F2 (φσl )(v) = −i 2 π 2
2π(vl i + wl )

× (vi + w)σl exp −πkvk2 − πkwk2 .
In particular, we have F1 (ϕ)(0) = F2 (ϕ)(0) = F1 (α)(0) = F2 (α)(0) = 0 so

that the theta series Θϕχ (z, h) and Θαχ (z, h) are rapidly decreasing at 0 and
∞.
Proof. The proposition follows from the computation of the partial Fourier
transform of
√
φσ (v, w) = Hσ π(v + w) exp −πkvk2 − πkwk2
N
Y √ 2 2

= Hdm π(vm + wm ) exp −πvm − πwm .
m=1
For x, y ∈ R we have the generating series
∞
√ X √ 2 2t
d
exp −πx2 − πy 2 + 2 π(x + y)t − t2 = Hd ( π(x + y))e−πx −πy .
d!
d=0
A direct computation shows that
Z √
−πx2 −πy 2 +2 π(x+y)t−t2 ′
e e2iπyy dy
R
√
Z √ √ 2
=e −π(x2 +(y ′ )2 )+2 π(x+iy ′ )t
e−( πy−t−iy ′ π )
dy
R
∞
X
−πx2 −π(y ′ )2 ) √ d td
=e 2d π (x + iy ′ )d
d!
d=0
where the integral is a Gaussian integral equal to 1. Hence, the partial
√ 2 2
Fourier transform of Hd ( π(x + y))e−πx −πy is
√ d 2 ′ 2 √ d d 2 ′ 2
2d π (x + iy ′ )d e−πx −π(y ) = id 2d π (ix + y ′ ) e−πx −π(y ) ,
and the partial Fourier transform of φσ (v) is
N
N Y dm
F2 (φσ )(v) = iN 2N π 2 exp −πkvk2 − πkwk2 (ivm + wm ) ,
m=1
where N = d1 + · · · + dN .
Recall that
√
φσl (v) = (vl − wl )Hσl ( π(v + w)) exp −πkvk2 − πkwk2
N
Y √ 2 2

= (vl − wl ) H|σ−1 (m)| ( π(vm + wm )) exp −πvm − πwm ).
l
m=1
The partial Fourier transform is computed separately for each factor of the
product. For the factors m 6= l, we use the previous computations for φσ .
For the factor m = l, we need to compute the partial Fourier transform of
√ 2 2
(x − y)Hd ( π(x + y))e−πx −πy . In general, the transform of function of the
form (x − y)f (y) is

′ 1 ∂
F2 ((x − y)f )(y ) = x + F2 (f )(y ′ ).
2πi ∂y ′
√
One can check that the partial Fourier transform of (x − y)Hd ( π(x +
2 2
y))e−πx −πy is
" #
√ 2π|ix + y ′ |2 + d
d d 2 ′ 2
−id+1 2d π (ix + y ′ ) e−πx −π(y ) .
2π(ix + y ′ )
For the last statement, note that

φσ (v, w) = φσ (w, v), φσl (v, w) = −φσl (w, v)
We deduce that
F1 (φσ )(v, w) = F2 (φσ )(w, v), F1 (φσl )(v, w) = −F2 (φσl )(w, v).
In particular, it follows that F1 (ϕ)(0) = F2 (ϕ)(0) and F1 (α)(0) = −F2 (α)(0).
The vanishing of F2 (φσ )(0) is clear. The vanishing of F2 (φσl )(0) is also clear
when N > 1 since
N
Y −1
(vi + w)σl = (vl i + wl ) (vm i + wm )|σl (m)|
m=1
m6=l
vanishes at v = 0. When N = 1 then l = 1 and σl is a function on an empty

set. Hence, the form is

F2 (φσl )(v) = −2iπ(vi + w) exp −πkvk2 − πkwk2 ,
which also vanishes at v = 0.
4.3. Differential forms as theta series. Let us now consider the theta
series for the Schwartz function constructed in Section 3. As previously we
let χ ∈ C[DL ] be a test function preserved by a torsion-free congruence
subgroup Γ.
Using the Weil representation, we can write φσ (ρ−1
g v) = ω(g, 1)φσ (v) so
that
N X
ϕ(z, v) = 2−N π − 2 ω(g, 1)φσ (v) ⊗ λ(σ) ∈ ΩN (X) ⊗ S(R2N )
σ
and similarly
N X
X
N−1
α(z, v) = −2−N π − 2 (−1)l−1 ω(g, 1)φσl (v) ⊗ λ(σl ).
l=1 σl
Lemma 4.10. The functions φσ and φσl satisfy
ω(1, k)φσ (v) = j(k, i)−N φσ (v), ω(1, k)φσl (v) = j(k, i)−(N −2) φσl (v)
for k ∈ SO(2).
Proof. Recall the partial Fourier transform intertwines the two representa-
tions ω and ω ′ . Thus, it is equivalent to show that φσ satisfies ω ′ (1, k)F2 (φσ )(v) =
j(k, i)−N F2 (φσ )(v). Let

c −s cos(θ) − sin(θ)
k= := ∈ SO(2).
s c sin(θ) cos(θ)
We have
(cv + sw)i + (−sv + cw) = (c + is)(vi + w) = j(k, i)(vi + w).
Since (vi + w)σ is a polynomial of degree N in the entries of the vector vi+w,
we see that
N
ω ′ (1, k)F2 (φσ )(v, w) = F2 (φσ )(cv + sw, −sv + cw) = j(k, i) F2 (φσ )(v, w).
The result for φσ follows from the fact that j(k, i) = c − is = j(k, i)−1 . The
computation is analogous for the function φσl , noting that (vi + w)σl is of
degree (N − 1) and that |j(k, i)|2 = 1.
Following the construction of theta series seen in 4.1, we consider for any
h ∈ SL2 (R) the form
N X
ω(1, h)ϕ(z, v) = 2−N π − 2 ω(g, h)φσ (v) ⊗ λ(σ)
σ
where z = gg t and σ is a function on {1, . . . , N }. For τ = x + iy ∈ H we

now put
X
Θϕχ (z, τ ) := j(hτ , i)N χ(v)ω(1, hτ )ϕ(z, v) ∈ ΩN (X)Γ ⊗ C ∞ (H)
v∈Q2N
where hτ ∈ SL2 (R) is any matrix sending i to τ . We can take

√ √
y x y−1
hτ = √ −1 ,
0 y
√
for which j(hτ , i) = y −1 . Since we have multiplied by j(hτ , i)N , Lemma
4.10 implies that the function is independent of the choice of hτ . For γ ∈ Γ′
we have hγτ = γhτ k for some k ∈ SO(2). Thus
Θϕχ (z, γτ ) = j(γ, τ )N Θϕχ (z, τ ),

so the form transforms like a modular form of weight N for Γ′ ⊂ SL2 (Z).
Note that the form is not holomorphic but holomorphic in cohomology, as
we will see in 4.7
Similarly, we define the theta series
√ X
Θψχ (z, τ ) := y −N χ(v)ω(1, hτ )ψ(z, v) ∈ ΩN (X)Γ ⊗ C ∞ (H)
v∈Q2N
which also transforms like a modular form of weight N for Γ′ . On the other
hand, the transgressed theta series
√ X
(4.3) Θαχ (z, τ ) := y−(N −2) χ(v)ω(1, hτ )α(z, v)
v∈Q2N
transforms like a modular form of weight (N − 2) for Γ′ .
4.4. The Eisenstein class. With respect to the splitting X ≃ S × R>0 we

can write
du
Θϕχ (z, τ ) = Θϕχ
e (z, τ ) + Θψχ (z, τ ) .
u
By Proposition 4.9 we have F1 (ϕ)(0) = F2 (ϕ)(0) = 0. It follows from

Proposition 4.3 that Θϕχ (z, τ ) and Θψχ (z, τ ) are rapidly decreasing as u
goes to 0 and ∞. Thus, the integral
Z ∞
du
Eϕχ (z1 , τ ) := p∗ (Θϕχ (z, τ )) = Θψχ (z, τ )
0 u
converges absolutely and uniformly on compact sets, by Proposition 4.4.
Proposition 4.11. The form Eϕχ (z1 , τ ) is closed.
Proof. Since Θϕχ (z, τ ) is closed and is rapidly decreasing at 0 and ∞, Propo-
sition 3.9 implies that
d1 Eϕχ (z1 , τ ) = d1 p∗ (Θϕχ (z, τ )) = p∗ (dΘϕχ (z, τ )) = 0.

We can view Eϕχ (z1 , τ ) as the value at s = 0 of the integral
Z ∞ du
Eϕχ (z1 , τ, s) := p∗ Θ′F2 (ϕχ) (z, τ )u−s = Θ′F2 (ψχ) (z, τ )u−s .
0 u
It converges absolutely for every s ∈ C, and as in Proposition 4.5 it converges
termwise absolutely for Re(s) > N .
Proposition 4.12. For Re(s) > N we can write

X
Eϕχ (z1 , τ, s) = Eϕχ (g1 , τ, s)σ ⊗ ιu ∂ λ(σ) ,
∂u
σ
where

s X g1−1 (vτ + w) σ
Eϕχ (g1 , τ, s)σ := Λ(s)y 2 F2 (χ)(v)
kg1−1 (vτ + w)k2N +s
v∈Q2N
v6=0
and
s (−1)N −1 iN
Λ(s) := Γ N + s .
2 2π N + 2
Proof. Since the integral converges absolutely for Re(s) > N , we can use
Poisson summation to write
Z ∞
du
Eϕχ (z1 , τ, s) = Θ′F2 (ψχ) (z, τ )u−s
0 u
X Z ∞
N du
= j(hτ , i) F2 (χ)(v) ω ′ (1, hτ )F2 (ψ)(z, v)u−s .
2N 0 u
v∈Q
Using the explicit formula for ψ in Proposition 3.10 and for F2 (φσ ) in
Proposition 4.9, we find that
j(hτ , i)N ω ′ (1, hτ )F2 (ψ)(z, v)

(−1)N −1 iN X g1 (vτ + w) σ − yuπ2 kg1−1 (vτ +w)k2
−1
= e ⊗ ι ∂ λ(σ)
u ∂u
yN σ
u2N
so that for v 6= 0 we have

Z ∞
N du
j(hτ , i) ω ′ (1, hτ )F2 (ψ)(z, v)u−s
0 u
s
X y 2 g1−1 (vτ + w)
σ
= Λ(s) −1 2N +s
⊗ ι ∂ λ(σ) .
u ∂u
σ kg1 (vτ + w)k
Example. When N = 1. We have X ≃ R>0 and S is a point. Let u be the

coordinate on X. We have
du √ v v 2
ϕ0 (u, v) = ψ 0 (u, v) = y + uw e−yπ( u −uw) .
u u
Then
√
j(hτ , i)ω(1, hτ )ϕ(u, v) = ϕ0 (u, yv)q B(v)
v v 2
= + uw e−π( u −uw) q B(v) .
u
It follows from Proposition 4.12 that
s i X y2
s
Eϕχ (τ, s) = Γ 1 + s F2 (χ)(v) .

2 2π 1+ 2 2
(vτ + w)|vτ + w|s
v∈Q
v6=0
For suitable test functions we obtain the coherent or incoherent weight one
Eisenstein series [KRY99], and more generally in [Kud97].
4.5. The Fourier coefficients of the Eisenstein class. We will need to

split the theta series in regular and singular terms, as in Section 4.1.2. Recall
that the Weil representation acts by
√ √
ω(1, hτ )ϕ(z, v) = y −N ϕ0 (z, yv)q B(v) , q = e2iπτ ,
so that
X √ ′
Θϕχ (z, τ ) = χ(v)ϕ0 (z, yv)q B(v) ∈ ΩN (X)Γ ⊗ C ∞ (H)Γ
v∈Q2N
and
X √ ′
Θψχ (z, τ ) = χ(v)ψ 0 (z, yv)q B(v) ∈ ΩN (X)Γ ⊗ C ∞ (H)Γ .
v∈Q2N
For any class v ∈ Γ\Q2N we set

Z ∞ X
√ du
Ev (z1 , y) := ψ 0 (z, yv) ∈ ΩN −1 (SΓ ) ⊗ C ∞ (R>0 ).
0 ′
u
v ∈Γv
We unfold the integral and get

Z ∞ Z ∞X X X
du √ du
Θψχ (z, τ ) = χ(v)ψ 0 (z, yv)q B(v)
0 u 0 u
n∈Z v∈Γ\Q2N v′ ∈Γv
B(v)=n
 
X
 X


=  χ(v)Ev (z1 , y) q n .
 
n∈Z 2N v∈Γ\Q
B(v)=n
Note that by Lemma 4.6, if v is regular, then we have

X Z ∞ √ du
Ev (z1 , y) := ψ 0 (z, yv) ,
′ 0 u
v ∈Γv
whereas for the singular vectors one needs to add u−s to change the order of
integration and summation. Thus, the n-th Fourier coefficient of Eϕχ (z1 , τ )
is given by
X
an (Eϕχ (z1 , τ )) = χ(v)Ev (z1 , y).
v∈Γ\Q2N
B(v)=n
For the constant term, we will consider the regular and singular vectors
separately and write
X X
a0 (Eϕχ (z1 , τ )) = χ(v)Ev (z1 , y) + χ(v)Ev (z1 , y)
v∈Γ\L∨
sing,1 v∈Γ\L∨
sing,2
X
(4.4) + χ(v)Ev (z1 , y).
v∈Γ\L∨
reg
B(v)=0
In the next section, we will show that the regular isotropic terms, and the
regular negative terms Ev (z1 , y) with B(v) < 0 are trivial in cohomology.
4.5.1. Pairings. We have the following pairing

2 −N )/2
Hc(N (SΓ ; C) × H N −1 (SΓ ; C) −→ C,
defined by
Z
′
[ω, ω ] = ω ∧ ω′.
SΓ
(N 2 −N )/2
On the other hand, by Poincaré duality we have Hc (SΓ ; C) ≃ HN −1 (SΓ ; C).
Hence, the pairing is
HN −1 (SΓ ; C) × H N −1 (SΓ ; C) −→ C,
where
Z
N(N−1)2
′
[Z, ω ] = (−1) 2 ω′.
Z
(N 2 −N )/2
The Poincaré dual ωZ in Hc (SΓ ; C) of a class Z in HN −1 (SΓ ; C) is
such that
Z Z
N(N−1)2
′ ′
[Z, ω ] = (−1) 2 ω = ωZ ∧ ω ′ = [ωZ , ω ′ ].
Z SΓ
ω′
for any ∈ H N −1 (S Γ ; C). Similarly, by Poincaré duality we have H
N −1 (S ; C) ≃
Γ
BM
H(N 2 −N )/2 (SΓ ; C), where the Borel-Moore homology is the homology of lo-
cally finite chains. Thus, the pairing becomes

2 −N )/2
Hc(N BM
(SΓ ; C) × H(N 2 −N )/2 (SΓ ; C) −→ C,
where
Z
[ω, Z ′ ] = ω.
Z′
Finally, combining the two gives the intersection pairing in homology
BM
HN −1 (SΓ ; C) × H(N 2 −N )/2 (SΓ ; C) −→ C,
where
Z
N(N−1)2
′
[Z, Z ] = (−1) 2 ωZ ′
Z
Z Z
= ωZ = ωZ ∧ ωZ ′ = [ωZ , ωZ ′ ].
Z′ SΓ
N(N−1)2
Note that we have [Z, Z ′ ] = (−1) 2 [Z ′ , Z].
4.5.2. The negative and regular isotropic orbits. Recall from (3.3) that α0 (z, v)
is an (N − 1)-form on X satisfying
√ ∂ √
dα0 (z, tv) = t ϕ0 (z, tv).
∂t
The form α0 is not rapidly decreasing on R2N , since
α0 (z, v) = pz (v) exp(−πkg−1 v − gt wk2 )
where z = ggt and pz (v) is a differential form that is polynomial in v. For
z ∈ X r Xv , the integral
Z ∞ √ dt
ξ 0 (z, y, v) := α0 (z, tv) ∈ ΩN −1 (X r Xv )
y t
√
converges and satisfies dξ 0 (z, y, v) = ϕ0 (z, yv).
Lemma 4.13. Let v be a regular vector in R2N . The sum

X Z ∞ √ dt
G(z, y, v) := α0 (z, tv′ ) , u, y ∈ R>0 ,
′ y t
v ∈Γv
converges absolutely and uniformly for z in a compact set A ⊂ X r Xv .

Moreover, for every m ∈ Z and u ∈ R>0 we have that |um G(z, y, v)| is
bounded i.e. the function is rapidly decreasing as u goes to 0 and ∞.
Proof. Let (v ′ , w′ ) ∈ RN × RN be the entries of v′ . First, note that

α0 (z, v′ ) = pz (v′ ) exp(−πkg−1 v ′ − gt w′ k2 )
= pz (v′ ) exp(−πkρ−1 ′ 2
g v k ) exp(2πn)
where n = B(v′ ) = B(v) and z = ggt . By abuse of notation we can write

for M ≥ 0 and z1 = g1 g1t in a compact set A
1
|α0 (z, v′ )| ≤ CM (z1 ) .
kρ−1 ′ M
1+ g vk
For M ≥ 1 we have
Z ∞ Z ∞
−1 ′ M 0
√ ′ dt kρ−1
g vk
′ M
dt
kρg v k |α (z, tv )| ≤ CM (z1 ) √ −1
y t y (1 + tkρg v k) t
′ M
Z ∞
1
≤ CM (z1 ) M dt
1+
y t 2
2CM (z1 )
= √ =: DM (z1 , y).
M yM
Since v′ 6= 0, the integral is bounded by DM (z1 , y)/kρ−1 ′ M and the sum
g vk
converges.
By the binomial formula and the sub-multiplicativity of the operator
norm6, we have for M, M ′ ∈ N the inequality
′
u−2(M −M ) ′ 2M ′ 2M ′ ′
2
kv k kw k ≤ kg−1 v ′ k2M kgt w′ k2M
K
′ ′ 2(M +M ′ )
≤ (kg−1 v ′ k2 + kg t w′ k2 )M +M = kρ−1
g vk
′
where K := supg1 ∈A kg−1 kM kgt kM . Thus, for any M, M ′ and v′ ∈ Γv we
get the bound
Z ∞ ′
√ dt DM +M ′ (z1 , y) u(M −M )
α0 (z, tv′ ) ≤ .
y t K2 kv ′ kM kw′ kM ′
Note that v ′ and w′ are both nonzero since v is regular. Hence, for u ∈ R>0
and any m ∈ Z we can take M large and set M ′ = M + m so that
X Z ∞ √ dt DM +M ′ (z1 , y) X 1
m
u α0 (z, tv′ ) ≤ 2
′ y t K ′
kv k kw′ kM +m
′ M
v ∈Γv v ∈Γv
D2M +m (z1 , y) X 1
≤ 2 ′ ′ min(M,M +m)
K ′
(kv kkw k)
v ∈Γv
X∞
D2M +m (z1 , y) |Nd |
=
K2 dmin(M,M +m)
d=1
where Nd := {v′
∈ Γv | d ≤ kv ′ kkw′ k
≤ d + 1} is finite and polynomial in d.
The sum on the right hand-side converges for M large enough.
6Here we take the norm kgk = sup kgvk

.
v∈RN kvk
It follows that the sum over a regular orbit defines a form

X
ξ 0 (z, y, v) ∈ ΩN −1 (XΓ r Xv ).
v′ ∈Γv
Moreover, we have
X X √
d ξ 0 (z, y, v′ ) = ϕ0 (z, yv′ ).
v′ ∈Γv v′ ∈Γv
Proposition 4.14. If v is a regular vector then the form Ev (z1 , y) ∈ ΩN −1 (SΓ )

is exact on SΓ r Zv . In particular, if v is a regular vector with B(v) ≤ 0,
then
Z
ω ∧ Ev (z1 , y) = 0.
SΓ
P
Proof. The previous lemma shows that v′ ∈Γv ξ 0 (z, y, v′ ) is rapidly decreas-
ing at 0 and ∞ when v is a regular vector. Thus, it is integrable in u and
we have
! !
X X
d1 p∗ ξ 0 (z, y, v′ ) = p∗ d ξ 0 (z, y, v′ )
v′ ∈Γv v′ ∈Γv
!
X √
0 ′
= p∗ ϕ (z, yv ) = Ev (z1 , y).
v′ ∈Γv
If v is a vector such that B(v) ≤ 0, then Zv is empty. This implies the

second part of the proposition.
4.5.3. The positive Fourier coefficients of the Eisenstein class. The special
cycles Zv are not embedded submanifolds (they are only images of proper
immersions). If ω is a compactly supported form, then the pairing is given
Z
[ω, Zv ] = p∗v ω
Γv \Sv
where pv the projection Γv \S −→ SΓ . The integral is convergent since the

composition
Γv \Sv ֒−→ Γv \S −→ SΓ
is proper.
The following proposition implies that
X
an (Eϕχ (z1 , τ )) = χ(v)Ev (z1 , y).
v∈Γ\Q2N
B(v)=n
represents the Poincaré dual to Zn (χ).
Proposition 4.15. If v is a regular vector in Q2N with B(v) = n > 0, then

the form Ev (z1 , y) represents the Poincaré dual to Zv in H N −1 (SΓ ). Hence,
(N 2 −N )/2
for every compactly supported form ω ∈ Ωc (SΓ ) we have
Z
[ω, Ev (z1 , y)] = [ω, Zv ] = ω.
Zv
Proof. Let us rewrite the sum Ev (z1 , y) as

X Z ∞ √ du
Ev (z1 , y) = ψ 0 (z, yρ−1
γ v)
u
γ∈Γv \Γ 0
X √
= p∗ ϕ0 (z, yρ−1
γ v) ∈ Ω
N −1
(SΓ )
γ∈Γv \Γ
where Γv denotes the stabilizer of v. We can assume without loss of gener-

ality that y = 1, since S√yv = Sv . Let us drop this variable z temporarily
and write ϕ0 (v). We want to show that
 
Z X Z
0 −1 
ω∧  p∗ ϕ (ργ v) = p∗v ω,
SΓ γ∈Γv \Γ Γv \Sv
where pv : Γv \S −→ SΓ is the projection map. By unfolding, this is equiva-

lent to
 
Z X Z
0 −1 

ω∧  p∗ ϕ (ργ v) = (p∗v ω) ∧ p∗ ϕ0 (v) .
SΓ γ∈Γv \Γ Γv \S
2
We have to show that for a form η ∈ Ω(N −N )/2 (Γv \SΓ ) that has compact
support in Γ\S, we have
Z Z
0

η ∧ p∗ ϕ (v) = η.
Γv \S Γv \Sv
This is similar to [Gar18, Proposition. 4.8] and we follow the proof. Let us
denote by I(t) the integral
Z √
I(t) := η ∧ p∗ ϕ0 ( tv) , t > 0.
Γv \S
By the transgression formula (3.3), we have
Z t2 √ dt
0
√ 0
√
ϕ ( t2 v) − ϕ ( t1 v) = d α0 ( tv) .
t1 t
Since v is regular
√ and the exponential term of α0 (v) is exp(−πku−1 v−uwk2 ),
the form α0 ( tv) is rapidly decreasing as u goes to ∞ and we have
Z t2
0
√ 0
√ 0
√ dt
p∗ ϕ ( t2 v) − p∗ ϕ ( t1 v) = d1 p∗ α ( tv) .
t1 t
It follows that the integral I(t) is independent of t and that
I(1) = lim I(t).
t→∞
Suppose that the support of η in SΓ is contained in a contractible relatively
compact open set U ⊂ SΓ . Otherwise, we may cover supp(η) by a finite
union of such open subsets. Let us denote by ̟ : S −→ SΓ the projection
map and let K ⊂ S be an open subset that is diffeomorphic to U under
(N 2 −N )/2
̟. Let ηK ∈ Ωc (S) be the form supported on K that is obtained
P by
restricting the pullback of η to K so that the pullback of η to S is γ∈Γ ηγK .
Then
 
Z X √
I(t) = ηK ∧  p∗ ϕ0 (ρ−1
γ tv)
S γ∈Γv \Γ
X Z √
= ηK ∧ p∗ ϕ0 (ρ−1
γ tv).
γ∈Γv \Γ S
Note that given a compact K ⊂ S, there are only finitely many v′ ∈ Γv such
that Sv′ intersects K. Indeed, suppose that z1 ∈ Sv′ ∩ K and let v′ = (v ′ , w′ ).
Then z1 w′ = v ′ implies that (w′ )t z1 w′ = B(v ′ , w′ ) = n. There are only
finitely many w′ that satisfies this condition for a fixed z1 , and the same
holds if z1 is in compact set. Similarly, we have (v ′ )t z1−1 v ′ = B(v ′ , w′ ) = n
and only finitely many v ′ can satisfy this. We can split I(t) as a sum over
the two sets
n o n o
S1 := γ ∈ Γv \Γ | Sρ−1
γ v
∩ K 6
= ∅ , S2 := γ ∈ Γv \Γ | Sρ−1
γ v
∩ K = ∅ ,
where S1 is finite. For the sum over S2 , since the terms are rapidly decreasing
(since z1 6∈ Sρ−1
γ v
) and we exchange the sum and we can apply the dominated
convergence theorem to change the order of the limit and the sum. It follows
that the integral
XZ √
lim ηK ∧ p∗ ϕ0 (ρ−1
γ tv) = 0
t→∞ S
γ∈S2
vanishes. On the other hand, by Proposition 3.5, the sum over S1 is

XZ √ XZ
0 −1
lim ηK ∧ p∗ ϕ (ργ tv) = ηK .
t→∞ S Sρ−1 v
γ∈S1 γ∈S1 γ
For γ ∈ S2 the submanifold Sρ−1

γ v
does not intersect the support of η, thus
we can add these terms to the latter sum and get
X Z √ X Z
0 −1
lim ηK ∧ p∗ ϕ (ργ tv) = ηK
t →∞
γ∈Γv \Γ S γ∈Γv \Γ Sρ−1
γ v
Z X
= ηγK
Sv γ∈Γ \Γ
v
Z X
= ηγK
γv \Sv γ∈Γ
Z
= η.
Γv \Sv

4.6. The constant term. In this section, we show that the constant term
is given by the Eisenstein classes constructed by Bergeron-Charollois-Garcia.
Let L be a lattice in RN as previously. The quotient
T := Γ\(S × T ) −→ Γ\S
is a torus bundle with fibers T := RN /L, where Γ acts by γ(z, v + L) =
(γz, γv + L).
4.6.1. Topological Eisenstein classes. We recall the construction of the canon-

ical Eisenstein classes following Bergeron-Charollois-Garcia [BCG20]. Simi-
lar classes have been previously considered by Bismut-Cheeger [BC92], and
Faltings [Fal05] for Siegel spaces. Note that the space X (resp. S + ) in loc.
cit. is the space S (resp. X) in this paper. Let m be an integer coprime
to |DL | and let T [m] = Γ\(S × T [m]) be the m-torsion points on T , and
T0 = Γ\(S × {0}) the image of the zero section. The Thom isomorphism
induces an isomorphism
H 0 (T [m]) −→ H N (T , T r T [m]).
Sullivan shows that the image of the class
[T [m] − mN T0 ] = 0 ∈ H N (T , T r T [m])
is trivial in cohomology. The long exact sequence
H N −2 (T r T [m]) H N −1 (T , T r T [m]) H N −1 (T )
H N −1 (T r T [m]) H N (T , T r T [m]) H N (T )
shows that it has a preimage in H N −1 (T r T [m]), only determined up to an

element in H N −1 (T ).
To find a canonical representatives, let a be an integer coprime to m. It
induces a finite covering map T −→ T and a pullback
a∗ : H N (T , T r T [m]) −→ H N (T , T r T [am]).
It also defines a pushforward
X
a∗ : ΩN (S × T ) −→ ΩN (S × T ), a∗ (ω)(z1 , v) = ω(z, w)
w∈T
aw=v
by summing over the fiber. The map is Γ-equivariant, and

a∗ ΩN −1 (S × (T r T [am]) ⊂ ΩN −1 (S × (T r T [m]).
Thus, it induces a pushforward map
a∗ : H N (T , T r T [am]) −→ H N (T , T r T [m]).
Finally, the pullback by the inclusion T [m] ⊂ T [am] is
i∗ : H N (T r T [m]) −→ H N (T r T [am]).
The following theorem is [BCG20, Section. 3].
Theorem 4.16. There is a unique class

zm ∈ H N −1 (T r T [m]; Q)
such that a∗ ι∗ zm = zm for all integers a coprime to m.
Let v0 ∈ DL be nonzero i.e. represented by a vector v0 6∈ L∗ not in L.

Since the order is coprime to m, it defines a section
v0 : SΓ −→ T r T [m].
The canonical Eisenstein class is the class
1
z(v0 ) := N v ∗ zm ∈ H N −1 (SΓ , Q),
m −1 0
where the normalization by mN − 1 is chosen so that the class does not
depend on m.
4.6.2. De Rham representative. We have an isomorphism
(4.5) X × RN −→ E, (z, v) −→ [g, g −1 v]
where z = ggt . Let Φ ∈ ΩN (X × RN ) be the pullback of the Thom form U
by this map. Let us denote by tv the translation tv : X × V −→ V defined
by tv (z, v ′ ) = (z, v + v ′ ). The form
X
ΘΦ (z, v ′ ) := t∗v Φ(z, v ′ ) ∈ ΩN (X × T )Γ
v∈L
is closed and Γ-invariant. By [BCG20, Proposition. 16], it is a Thom form
of the torus bundle. We write
e v ′ ) + Ψ(z, v ′ ) du
Φ(z, v ′ ) = Φ(z,
u
N
where Φ ∈ Ω (X × T ) and Ψ ∈ Ω N −1 (X × T ) are of degree 0 along R>0 .
Similarly, we define
X
ΘΨ (z, v ′ ) := t∗v Ψ(z, v ′ ) ∈ ΩN −1 (X × T )Γ .
v∈L
Since the exponential factor of Ψ is of the form exp(−u−1 kv ′ +vk2 ), the form
ΘΨ (z, v ′ ) is rapidly decreasing as u → 0 if v ′ is not in L. Hence, the integral
Z ∞
′ du
EΨ (z1 , v , s) := ΘΨ (z, v ′ )u−s ∈ ΩN −1 (T − T0 )
0 u
converges for Re(s) > 0 and away from the zero section. It has a meromor-
phic continuation, with at most a simple pole at s = 1. This is the form
defined by Bergeron-Charollois-Garcia in [BCG20, section. 8.5].
Let v0 be a vector in L∗ and suppose that Γ ⊂ SLN (Z) preserves the class
of v0 ∈ DL . Let us also denote by v0 the corresponding section
SΓ −→ T r T0 , z 7−→ (z, v0 ).
The pullback defines a closed form
EΨ,v0 (z1 , s) := v0∗ EΨ (z1 , v ′ , s) ∈ ΩN −1 (SΓ ).
The relation between the topological construction of the Eisenstein class and
the forms is the following theorem [BCG20, Theorem. 23].
Theorem 4.17. The class −z(v0 ) in H N −1 (SΓ , Q) is represented by the

form EΨ,v0 .
For v0 ∈ DL let us define the test functions χ1 (v0 ) := χ(v0 , 0) and

χ2 (w0 ) := χ(0, w0 ). Let
X
z(χi ) := χi (v0 )z(v0 ).
v0 ∈DL
Let χ
b2 be the finite Fourier transform of χ2 and set
z(χ) := −z(χ1 ) − z(b
χ2 ).
Proposition 4.18. We have

a0 (Eϕχ (z1 , τ )) = z(χ) in H N −1 (SΓ , Q).
Proof. We have seen in (4.4) that the constant term is

X X
a0 (Eϕχ (z1 , τ )) = χ(v)Ev (z1 , y) + χ(v)Ev (z1 , y)
v∈Γ\L∨
sing,1 v∈Γ\L∨
sing,2
X
+ χ(v)Ev (z1 , y),
v∈Γ\L∨
reg
B(v)=0
where, by Proposition 4.14, the regular term is exact.

Note that the following diagram commutes
X × RN G ×K RN
v
s(v,0)
X
where the top arrow is the map (4.5) and sv (z) = [gz , gz−1 v − gzt w] is the
section used to construct ϕ0 (v) = s∗v U . Hence, we have v ∗ Φ(z, v ′ ) =
ϕ0 (z, (v, 0)). It follows that
X
v0∗ ΘΦ (z, v ′ ) = v0∗ t∗v Φ(z, v ′ )
v∈L
X
= (v0 + v)∗ Φ(z, v ′ )
v∈L
X
= ϕ0 (z, (v, 0))
v∈v0 +L
X X
= ϕ0 (z, (v ′ , 0)).
v∈Γ\(v0 +L) v′ ∈Γv
du
Hence, taking u components gives
X X
v0∗ ΘΨ (z, v ′ ) = ψ 0 (z, (v ′ , 0))
v∈Γ\(v0 +L) v′ ∈Γv
and the integral over R>0 is

Z ∞
du
EΨ,v0 (z1 , s) = v0∗ ΘΨ (z, v ′ )u−s
0 u
Z ∞ X X du
= ψ 0 (z, (v ′ , 0))u−s
0 u
v∈Γ\(v0 +L) v′ ∈Γv
X
= E(v,0) (z1 , 1).
v∈Γ\(v0 +L)
Finally, note that the function E(v,0) (z1 , y) satisfies the homogeneity property
E(v,0) (z1 , y, s) = y s E(v,0) (z1 , 1, s),
which can be seen from the change of variable t = uy, and shows that
X
E(v,0) (z1 , y) = EΨ,v0 .
[v]∈Γ\(v0 +L)
It follows that
X X
χ(v)Ev (z1 , y) = χ(v, 0)E(v,0) (z1 , y)
v∈Γ\L∨
sing,1 v∈Γ\L∗
X X
= χ1 (v0 ) E(v,0) (z1 , y)
v0 ∈DL v∈Γ\(v0 +L)
X
= χ1 (v0 )EΨ,v0
v0 ∈DL
represents −z(χ1 ), and the second singular term follows from Poisson sum-
mation.
4.7. Holomorphicity of the theta lift. The form Eϕχ (z1 , τ ) transforms
like a modular of weight N . To obtain a lift to holomorphic modular forms,
we haveto show that
the forms vanishes (in cohomology) under the operator
∂ 1 ∂ ∂
∂τ = 2 ∂x + i ∂y .
Recall that we defined in (4.3) the theta series
X √
Θαχ (z, τ ) = y χ(v)α0 (z, yv)q B(v) .
v∈Q2N
Since Θαχ is rapidly decreasing at 0 and ∞, the integral

Eαχ (z1 , τ ) := p∗ (Θαχ (z, τ )) ∈ ΩN −2 (SΓ ) ⊗ C ∞ (H)
along the fiber of X −→ S converges.
Theorem 4.19. The form Eαχ (z1 , τ ) transforms like a modular form of
weight N − 2 and
∂
2iy 2 Eϕχ (z, τ ) = dEαχ (z1 , τ ).
∂τ
In particular, the lift Eϕχ (C, τ ) is holomorphic.
Proof. The form α0 satisfies

√ ∂ 0 √ ∂ √
dα0 (z, yv) = y ϕ (z, yv) = 2iy ϕ0 (z, yv).
∂y ∂ τ̄
Hence, we have
X √
dΘαχ (z, τ ) = y χ(v)dα0 (z, yv)q B(v)
v∈L∨
X ∂ 0 √
= 2iy 2 χ(v) ϕ (z, yv)q B(v)
∂ τ̄
v∈L∨
∂
= 2iy 2 Θϕχ (z, τ ).
∂ τ̄
Since Θαχ (z, τ ) is rapidly decreasing at 0 and ∞, we deduce from Proposition
3.9 that
∂
2iy Eϕχ (z1 , τ ) = p∗ (dΘαχ (z, τ )) = d1 p∗ Θαχ (z, τ ).
∂τ

(N 2 −N )/2
For a compactly supported form ω ∈ Hc (SΓ ), we define the pairing
Z
Eϕχ (ω, τ ) := [ω, Eϕχ (z1 , τ )] = ω ∧ Eϕχ (z1 , τ ).
SΓ
(N 2 −N )/2
By Poincaré duality, we have Hc (SΓ ; C) ≃ HN −1 (SΓ ; C). If ω = ωC
(N 2 −N )/2
is a form in Ωc (SΓ ; C) representing the Poincaré dual of a cycle C,
then we write
Z
N(N−1)2
Eϕχ (C, τ ) := (−1) 2 Eϕχ (ωC , τ ) = Eϕχ (z1 , τ ).
C
Let MN (Γ′ ) be the space of holomorphic modular forms of weight N and

level Γ′ ⊂ SL2 (Z). We combine sections 4.5.3 and 4.5.2 to deduce the fol-
lowing theorem.
Theorem 4.20. The form Eϕχ (z1 , τ ) defines a lift
Eϕχ : HN −1 (SΓ ; Z) −→ MN (Γ′ ), Z 7−→ Eϕχ (Z),
where for a cycle Z ∈ ZN −1 (SΓ , Z) we have

Z ∞
X
N(N−1)2
Eϕχ (Z, τ ) = z(χ) + (−1) 2 [Z, Zn (χ)] q n .
Z n=1
Remark 4.1. Recall that as oriented submanifolds we have S−v = (−1)N Sv .

Hence, if χ is a test function such that χ(−v) = (−1)N −1 χ(v), then Eϕχ (Z, τ )
vanishes for every Z.
4.8. Span of special cycles. We also get a dual lift

Λϕχ : SN (Γ′ ) −→ H N −1 (SΓ ; C), f 7−→ Λϕχ (f )
where for a weight N cusp form f ∈ SN (Γ′ ) we define
Λϕχ (z1 , f ) := hEϕχ (z1 , τ ), f ipet
Z
= Eϕχ (z1 , τ )f (τ )y N −2 dxdy, z1 ∈ SΓ .
Γ′ \H
Since the integrals are absolutely convergent, it follows from Fubinis theorem
that
[ω, Λϕχ (f )] = hEϕχ (ω), f ipet
Recall that we have a pairing
BM
[−, −] : HN −1 (SΓ ; C) × H(N 2 −N )/2 (SΓ ; C) −→ C.
BM
For a subspace W ⊂ H(N 2 −N )/2 (SΓ ; C) we denote its orthogonal complement
by

W ⊥ := Z ∈ HN −1 (SΓ ; C) | [Z, Z ′ ] = 0 for all Z ′ ∈ W .
The following theorem is analogous to [KM88, Theorem. 4.5]. Note that here
we extend the coefficients to C to get a lift
Eϕχ : HN −1 (SΓ ; C) −→ MN (Γ′ ).
Theorem 4.21. We have

ker(Eϕχ ) ≃ span{Zn (χ) | n ∈ N>0 }⊥ ⊂ HN −1 (SΓ ; C).
Moreover, when N > 2, under Poincaré duality H N −1 (SΓ ; C) ≃ H(N
BM
2 −N )/2 (SΓ ; C)
we also have
im(Λϕχ ) ≃ span{Zn (χ) | n ∈ N>0 }.
Proof. The first follows from the definition of the theta lift. The lift Eϕχ (Z)
vanishes if and only if all the Fourier coefficients [Z, Zn (χ)] vanishes, i.e.
when
Z ∈ {Zn (χ) | n ∈ N>0 }⊥ .
When N > 2, let us show that we have im(Λϕχ ) ≃ ker(Eϕχ )⊥ , or equiva-
lently
im(Λϕχ )⊥ ≃ ker(Eϕχ ).
For any N we have ker(Eϕχ ) ⊂ im(Λϕχ )⊥ . Indeed, if ω is an element of
ker(Eϕχ ), then for η = Λϕχ (f ) in the image of Λϕχ we have
[ω, η] = [ω, Λϕχ (f )] = [Eϕχ (ω), f ] = 0.
For the other inclusion, let Γ′∞ ⊂ Γ′ be the stabilizer of ∞ in Γ′ . For n ≥ 1
we define the Poincaré series Pn ∈ SN (Γ′ ) by
X exp(2iπγτ )
Pn (τ ) := c ,
′ ′
j(γ, τ )N
γ∈Γ∞ \Γ
where the sum converges for N > 2 and the constant c is chosen such that
for a modular form f ∈ MN (Γ′ ) we have
hPn (τ ), f ipet = an (f ).
If ω ∈ im(Λϕχ )⊥ , then for any n ≥ 1 we have
0 = [ω, Λϕχ (Pn )] = [Eϕχ (ω), Pn ] = [ω, Zn (χ)],
since [ω, Zn (χ)] is the n-th Fourier coefficient of Eϕχ (ω). It follows that
Eϕχ (ω) = 0.
Remark 4.2. Equivalently, we have ker(Λϕχ ) = im(Eϕχ ) ∩ SN (Γ′ ).
4.9. Periods over tori. Let T (Q) ⊂ GLN (Q) be a torus. We first consider
the split torus
  
 T1
 

 .  ×
T0 (Q) =  ..  , Tm ∈ Q ≃ (Q× )N .

 

TN
tN−1
After the change of variables (T1 , . . . , TN ) = ( tu1 , . . . , 1
u , ut1 ···tN ) we see
(1) (1)
that we have a diffeomorphism T0 (R) ≃ T0 (R) × R>0 where T0 (Q) ⊂
SLN (Q) is the torus
  

 t1 


 .  

(1)  . .  ×
T0 (Q) =   Tm ∈ Q ≃ (Q× )N −1 .

  tN −1  


 

(t1 · · · tN −1 )−1
The intersection T0 (R) ∩ SO(N ) is the maximal compact {±1}N −1 ⊂ T0 (R).
Hence the quotient
(1) (1) N −1
T0 (R)/T0 (R) ∩ SO(N ) ≃ R>0
can be embedded in S. Let ϑT0 ∈ Ω1 (T0 (R)) ⊗ End(RN ) be the pullback of
the Maurer-Cartan ϑ to T0 . Then
 
dT1 /T1
 .. 
 . 
ϑT0 =  
 dTN −1 /TN −1 
dTN /TN ,
and λT0 = 12 (ϑT0 + ϑtT0 ) = ϑT0 . Since (λT0 )ij = 0 if i 6= j, it follows that
(4.6) λT0 (σ) = (λT0 )1σ(1) ∧ · · · ∧ (λT0 )N σ(N )
can only be nonzero if σ is the identity function σ(i) = i. For this function,
we see that (4.6) is equal to
dT1 dTN
λT0 (σ) = ∧ ··· ∧ .
T1 TN
∂
The contraction along u ∂u is

dT1 dTN dt1 dtN −1
ιu ∂ ∧ ··· ∧ =N ∧ ··· ∧ .
∂u T1 TN t1 tN −1
N −1
Hence, the restriction of Eϕχ (z1 , τ, s) to R>0 is
s X n(vτ + w) dt1 dtN −1

Eϕχ (t, τ, s) = Λ(s)y 2 F2 (χ)(v) N ∧ ··· ∧
kt−1 (vτ + w)k2N +s t1 tN −1
v∈Q2N
v6=0
where t = (t1 , . . . , tN −1 ) and
N −1
!N +s/2
X |vm τ + wm |2
−1 2N +s
kt (vτ + w)k = + (t1 · · · tN −1 )2 |vN τ + wN |2 .
t2m
m=1
Note that for a vector µ ∈ CN we denote its norm by n(µ) = µ1 · · · µN .

N −1
To compute the integral over R>0 it will be more convenient to change
the variables and compute
Z Z Z ∞
Eϕχ (z1 , τ, s) = Θϕχ (z, τ )u−s
(1) (1) (1) (1)
T0 (R)/T0 (R)∩SO(N ) T0 (R)/T0 (R)∩SO(N ) 0
Z
= Θϕχ (z, τ )u−s .
T0 (R)/T0 (R)∩SO(N )
Using that
Z
1 s −N + s ∞
2 du
Γ N+ α 2 = e−αu u2N +s
2 2 0 u
(for α = kt−1 (vτ +w)k2N +s ) and the change of variable (T1 , . . . , TN ) =
( tu1 , . . . , tuN , ut1 · · · tN ) we find that
Z
1 s 1 dt1 dt
Γ N+ −1 2N +s
N ∧ ··· ∧
2 2 RN−1 >0
kt (vτ + w)k t1 tN −1
YN Z ∞
2 2 2+s/N dTm
= e−Tm |vm τ +wm | Tm
m=1 0
Tm
1 s N Y
N
1
= Γ 1 + .
2N N |v m τ + w m |2+s/N
m=1
Recall that we have

dT1 dTN dt1 dtN −1
ιu ∂ ∧ ··· ∧ =N ∧ ··· ∧ .
∂u T1 TN t1 tN −1
We deduce that
Z X s
y2
Eϕχ (z1 , τ, s) = Λ′ (s) F2 (χ)(v)
RN−1 n(vτ + w)|n(vτ + w)|s/N
>0 v∈Q2N
where

s N
′ (−1)N −1 iN Γ 1 + N
Λ (s) = s .
πN + 2
4.9.1. Integral over modular symbols. Let L = pZN and L = L × L. Let

v0 = (v0 , w0 ) where v0 = (a1 , · · · , aN )t and w0 = (b1 , · · · , bN )t are two
vectors in L∗ = p−1 ZN . We will say that v0 is good if it satisfies that for
each m the entry am is not in pZ. In other words, the entry am is of the
form pr2 + pZ with 1 ≤ r ≤ p2 − 1.
Let χ ∈ C[DL ] be a test function and assume througout this section that
it splits as χ(v) = χ1 (v)χ2 (w), with two test functions χi ∈ C[DL ].
Definition 4.22. We say that χ ∈ C[DL ] is good if χ1 (0) = 0. Hence, a

good test function is only supported on good vectors.
Remark 4.3. Since χ splits, we have F2 (χ) = χ1 χb2 where χ b2 is the Fourier
transform of χ2 in C[DL ]. Hence, such a test function χ is good if and only
F2 (χ) is good.
Ash and Rudolph [AR79] proved that the relative homology of S Γ is gener-
ated by modular symbols ZQ attached to an invertible matrix Q ∈ MatN (Z)
(if det(Q) = 0, then the modular symbol is trivial in homology). First, let
Z∞ ∈ ZN BM (S ; Z) be the image of T (1) (R)T (1) (R)t ≃ RN −1 in S . The
−1 Γ 0 0 >0 Γ
7 t
modular symbol ZQ is the image of SQ := QS∞ Q in SΓ and defines a class
BM
ZQ ∈ H N −1 (SΓ , Z).
Suppose that F2 (χ) is good. Then by the previous computations 8

Z X s
′ y2
Eϕχ (z1 , τ, s) = Λ (s) F2 (χ)(v)
Z∞ 2N
n(vτ + w)|n(vτ + w)|s/N
v∈Q
s
X X y2
′
= Λ (s) F2 (χ)(v0 )
v0 ∈DL
v∈v0 +pZN
w∈w0 +pZN
X N
Y
′ −N am τ + bm s
= Λ (s)p F2 (χ)(v0 ) E1 τ, , ,
p N
v0 ∈DL m=1
where E1 (τ, λ0 , s) is the weight one Eisenstein series

X 1
E1 (τ, λ0 , s) :=
λ∈Z+τ Z
(λ0 + λ)|λ0 + λ|s/N
for λ0 ∈ C. It admits an analytic continuation to s = 0 if λ0 6∈ Zτ + Z. If

we suppose that F2 (χ) is good, then whenever F2 (χ)(v0 ) is nonzero we have
that
a m τ + bm
6∈ Zτ + Z.
p
The previous computation shows that given a good test function, the integral
on the modular symbol Z∞ converges.
7It is exactly the set D in [AR79, p. 245].

8Note that we can add 0 ∈ L∨ to the summation since F (χ)(0) = 0.
2
Definition 4.23. Let χ ∈ C[DL ] be a test function. We say that the matrix
Q is good for χ if Q∗ (χ1 )(v) := χ1 (Qv) is good.
Remark 4.4. Note that since Q has integral entries, we have QL∨ ⊂ L∨ and
QL ⊂ L. Thus, the map v 7→ χ(Qv) can be viewed as a test function in
C[DL ], supported on QL∨ ⊂ L∨ and L-invariant.
Using the equivariance of the form ϕ, we find that
Q∗ F2 (ϕχ)(v) = F2 (ϕ)(Q−1 v)F2 (χ)(v).
when det(Q) 6= 0. If follows that along the modular symbol ZQ with
det(Q) 6= 0, we have
Z X s
y2
Eϕχ (z1 , τ, s) = Λ′ (s) F2 (χ)(Qv)
ZQ 2N
v∈Q
X N
Y
′ N a m τ + bm s
= Λ (s)p F2 (χ)(Qv0 ) E1 τ, , .
m=1
p N
v0 ∈Q2N /L
If Q is good for χ, then Q∗ F2 (χ)(v0 ) = χ1 (Qv0 )b

χ2 (Qw0 ) is supported on
the cosets v0 + L with good v0 = (v0 , w0 ). Thus, the Eisenstein series on
the right-hand side admits an analytic continuation to s = 0.
Theorem 4.24. Let χ ∈ C[DL ] be a test function that splits as χ(v) =

χ1 (v)χ2 (w). If Q is good for χ, then the period over the modular symbol ZQ
converges and we have
Z X N
Y
(−1)N −1 iN a m τ + bm
Eϕχ (z1 , τ ) = F2 (χ)(Qv0 ) E1 τ, .
ZQ π N pN m=1
p
v0 ∈Q2N /L
Remark 4.5. For an N -tuple γ = (γ1 , . . . , γN ), one can consider similar pe-
riods over ZQ(γ) . Similar computation show that obtains the homogeneous
cocycle valued in meromorphic functions H × CN considered in Zhang’s the-
sis [Zha20, Section. 4.1.1] and by Bergeron-Charollois-Garcia in [BCG23,
Theorem. 2.10].
4.9.2. Tori attached to totally real fields. Let F be a totally real field of
degree N with ring of integers O. Let m ⊂ O be an integral ideal, that
we view as a lattice L = m in RN using the N real embeddings of F . Let
L = m ⊕ m. Note that with this embedding, we have
B(α, β) = trF |Q (αβ).
Hence, we have m∗ = m−1 d−1 and L∨ = m−1 d−1 ⊕ m−1 d−1 , where d−1 is
the inverse different ideal. Let χ be a test function in C[DL ]. The totally
positive elements F 1,+ of norm in F can be diagonally embedded
(F ⊗ R)1,+ ֒−→ SLN (R)
and the image is the split torus T0 (R). Let

U (m)1,+ := x ∈ O 1,+ | x − 1 ∈ m
be the totally positive units O 1,+ that are conguent to 1 modulo m. Since
U (m)1,+ preserves the discriminant module, it preserves χ and is contained
in Γ, the stabilizer of χ. The quotient
Zm := U (m)1,+ \(F ⊗ R)1,+ ∈ HN −1 (SΓ ; Z)
is compact and defines a homology class in SΓ . It follows from the above
computations that
Z
Eϕ (z1 , τ, s) = E (τ, s)
Zm
is the diagonal restriction of a Hilbert-Eisenstein series of parallel weight one
N
iN Γ 1 + Ns X
Eχ (τ , s) = s F2 (χ)(v0 )E (τ , v0 τ + w0 , s)
πN + 2 v0 ∈D L
where for λ0 ∈ CN we set

s
X n(Im(τ )) 2N
E (τ , λ0 , s) = .
n(λ0 + vτ + w)|n(λ0 + vτ + w)|s/N
v∈(m⊕m)/U (m)1,+
v6=(0,0)
Here vτ + w denotes the vector in CN that has entries vm τm + wm , where

v = (v, w) = (v1 , . . . , vN , w1 , . . . , wN ). The Hilbert-Eisenstein series is of
parallel weight one and level Γ′ where Γ(m) ⊂ Γ′ ⊂ SL2 (O). We deduce the
following.
Theorem 4.25. The diagonal restriction of the Hilbert-Eisenstein series has

the Fourier expansion
Z ∞
N(N−1)2 X
Eχ (τ ) = z(χ) + (−1) 2 [Zm , Zn (χ)]q n .
Zm n=1
5. The case N = 2
Let X be the space of positive definite symmetric 2 by 2 matrices and
S ⊂ X the matrices of determinant 1. We have
H ≃ SLN (R)/SO(2) ≃ S.
Let z1 = a + ib be the coordinates on H. The first map sends z1 ∈ H to
g1 SO(2) where
√ √
b a/√b
g1 = .
0 1/ b
The second map sends it to the positive definite quadratic form

t 1 a2 + b2 a
g1 g1 = ∈ S.
b a 1
We extend it to

u a 2 + b2 a
H × R>0 −→ X, (z1 , u) 7−→ z1 u = .
b a 1
Suppose that χ is the characteristic function of L ⊕ Z2 defined in Example

2.3, where L ⊂ Z2 consists of pairs (v1 , v2 ) where (v1 , p) = 1 and v2 ∈ pZ.
The function χ is preserved by Γ = Γ0 (p) and the locally symmetric space
is the modular curve
SΓ = Y0 (p) = Γ0 (p)\H.
5.1. Hecke operators. For two points α, β ∈ P1 (Q) let {α, β} be the geo-
desic joining the two points on the boundary of H. For n > 0 we define

(n) a b
∆0 (p) = M = ∈ Mat2 (Z) det(M ) = n > 0, (a, p) = 1 .
pc d
The congruence subgroup Γ0 (p) ⊂ SLN (Z) acts on the left and the right
on ∆0 (p)(n) and the double coset
R0 (p)(n) = Γ0 (p)\∆0 (p)(n)
is finite. We will also denote by {α, β} the image of {α, β} in the modular
curve Γ0 (p)\H. The modular symbol {α, β} represents a 1-cycle relative to
the cusps of Γ0 (p)\H. An explicit set of representatives for R0 (p)(n) is given
by
G dG
′ −1
(n) d b
∆0 (p) = Γ0 (p) .
′
0 d′
dd =n b=0
d>0
(d,p)=1
The action of the coset representatives induces the Hecke operator

′ −1
X dX dα + b dβ + b
Tn {α, β} := , .
′
d′ d′
dd =n b=0
d>0
(d,p)=1
In particular, we have
−1
X dX
′
b
Tn {0, ∞} = ,∞ .
d′
dd′ =n b=0
d>0
(d,p)=1
The Hecke operators act on the homology H1 (Y0 (p); Z) in a similar way.
Proposition 5.1. We have Zn (χ) = Tn {0, ∞}.
Proof. Let v = (v, w) where v = (v1 , v2 )t and w = (w1 , w2 )t . Let us compute

Xv . It is the set of z = (z1 , u) such that v = z1 uw i.e.

v1 u a2 + b2 a w1 u w1 (a2 + b2 ) + w2 a
= = .
v2 b a 1 w2 y w1 a + w2
Hence the projection Sv is the set of τ such that
(5.1) v2 (w1 (a2 + b2 ) + w2 a) = v1 (w1 a + w2 )
(= bu−1 for some u ∈ R>0 ).
If w1 = 0, then v2 6= 0 and w2 6= 0 since B(v) > 0. The equation is v2 a = v1

and describes the geodesic from vv21 to ∞ in H. If v2 = 0, then v1 6= 0 and
w1 6= 0. The equation is 0 = w1 a + w2 and describes the geodesic from − w 2
w1
to ∞. If w1 v2 6= 0, then we can rewrite equation (5.1) as

v2 w2 − v1 w1 2 v2 w2 + v1 w1 2
a+ + b2 = ,
2w1 v2 2w1 v2
which describes the geodesic with endpoints w v1
w1 and − v2 . We have showed
2
that the submanifold is the geodesic

w2 v1 w1 v
Xv = ,− if v = , 1 .
w1 v2 w2 v2
By the proof of Proposition 2.3, we find that a set of representatives for
Γ0 (p)\(L × Z2 ) is
′
d d
Rn = v = , n = dd′ , (d, p) = 1, 0 < d, 0 ≤ b ≤ d′ − 1 .
0 b
Hence, we have
−1
X dX
′
b
Zn (χ) = ,∞ = Tn {0, ∞}.
d′
dd′ =n b=0
d>0
(d,p)=1

The Hecke operators also act on differentials forms as follows. If ω ∈
Ω1 (H)Γ0 (p) is a Γ0 (p)-invariant form on H, then
X
Tn ω := X ∗ω
X∈R0 (p)(n)
is a Γ0 (p)-invariant form on H. It induces an action of Hecke operators on

the cohomology group H 1 (Y0 (p)). If ω{α,β} ∈ Ω1 (Y0 (p)) is a Poincaré dual
to {α, β}, then Tn ω{α,β} = ωTn {α,β} . Finally, the Hecke operators acts on a
modular form f ∈ M2 (Γ0 (p)) by
X d′ −1
1 X dz + b
Tn f (z) := f .
′
(d′ )2 d′
dd =n b=0
d>0
(d,p)=1
If ωf = f (z)dz, then Tn (ωf ) = ωTn f .

5.2. The Eisenstein lift when N = 2. With the setup as above, the
Eisenstein lift is
Eϕχ : H1 (Y0 (p); C) ≃ Hc1 (Y0 (p); C) −→ M2 (Γ0 (p)).
(p)
Let {E2 , f1 , . . . , fr } be a basis of M2 (Γ0 (p)), where fi ∈ S2 (Γ0 (p)) is a
(p)
basis of normalized newforms and E2 is the Eisenstein series
(p) p − 1 X (p)
E2 = + σ1 (n)q n .
24
n≥1
(p)
We define ωE := E2 (τ )dτ and ωf± := (ωf ± ωf ) where ωf = f (τ )dτ .
Corollary 5.1.1. For ω ∈ Hc1 (Y0 (p); C) we have the Fourier expansion
∞
X
Eϕχ (ω, τ ) = [ω, ωE ] − [ω, Tn {0, ∞}]q n ∈ M2 (Γ0 (p)).
n=1
and the spectral expansion

X L(fi , 1) r
24 (p)
Eϕχ (ω) = [ω, ωE ]E2 − [ω, ωf+i ]fi .
p−1 iπkfi k2
i=1
In particular, we deduce that the image of this lift is precisely the space
(p)
M02 (Γ0 (p)) spanned by E2 and the forms fi for which L(fi , 1) 6= 0. More-
over, the lift is Hecke equivariant i.e., satisfies Eϕχ (Tn ω) = Tn Eϕχ (ω) when
(p, n) = 1.
Proof. The first part follows from Proposition 5.1 and the Fourier expansion
of Eϕχ . The winding element Gw ∈ H 1 (Y0 (p)) is such that
Z ∞
[ω, Gw ] = ω.
0
We follow the proof of [DPV21, Lemma. 3.4] with sligthly different normal-
izations. We write the winding element as
r
X
Gw = αE ωE + (α+ + − −
fi ωfi + αfi ωfi ).
i=1

1 1
If ω0 denotes the homology class represented by , then we deduce
0 1
(p)
from [ω0 , ωE ] = a0 (E2 ) and [ω0 , Gw ] = 1 that
24
αE = .
p−1
Using the following properties
[ωf+i , ωf+j ] = [ωf−i , ωf−j ] = 0,
[ωf+i , ωf−j ] = −[ωf−j , ωf+i ] = −2hfi , fj i
L(f, 1)
(5.2) [ωf− , Gw ] = − ,
iπ
[ωf+ , Gw ] = 0,
we find that
L(fi , 1)
α−
fi = 0, α+
fi = − ,
iπkfi k2
and
X L(fi , 1) r
24
Gw = ωE − ω+ .
p−1 iπkfi k2 fi
i=1
Since fi is a newform and has real Fourier coefficients, the n-th Fourier
coefficient of Eϕχ (ω) is
X L(fi , 1)r
24 (p)
[ω, Tn Gw ] = [ω, ωE ]an (E2 ) − [ω, ωf+i ]an (fi ).
p−1 iπkfi k2
i=1
Thus, we have
X L(fi , 1) r
24 (p)
Eϕχ (ω) = [ω, ωE ]E2 − [ω, ωf+i ]fi .
p−1 iπkfi k2
i=1
The surjectivity follows from the fact that
(p)
Eϕχ (ωE ) = E2
2L(fi , 1)
Eϕχ (ωf−i ) = fi
iπ
Eϕχ (ωf+i ) = 0.
Finally, since the Hecke operator Tn is self-adjoint when (n, p) = 1 [Miy89,
Theorem. 4.5.4], we have
(p) (p)
[Tn ω, ωE ] = [ω, Tn ωE ]E2 = an (E2 )[ω, ωE ].
Similarly, we have [Tn ω, ωf−i ] = an (f )[ω, ωf−i ] and this proves the Hecke equiv-
ariance.
Remark 5.1. It follows from the proof of the the previous corollary that the
kernel of Eϕχ is spanned by all the forms ωf+i , and the forms ωf−i for which
L(fi , 1) 6= 0. Moreover, by Theorem 4.21 we have
ker(Eϕχ ) ≃ span{Tn {0, ∞} | n ∈ N>0 }⊥ ⊂ Hc1 (Y0 (p); C).
It follows that
span{Tn {0, ∞} | n ∈ N>0 }⊥ ={ωf+i }⊕{ωf−i such that L(fi , 1) 6= 0}.

This equality can be seen directly. Since the modular symbols generate
the relative homology H1BM (Y0 (p); C) ≃ H1 (X0 (p), ∂X0 (p); C), the Poincaré-
Lefschetz pairing is
Z β
1
Hc (Y0 (p); C) × H1 (X0 (p), ∂X0 (p); C) −→ C, [ω, {α, β}] = ω.
α
Since Hc1 (Y0 (p), C)
is spanned by the newforms ωf+i , ωf−i ,
it follows from the
h i
properties (5.2) that ωf+i , Tn {0, ∞} = 0 and
h i h i
ωf−i , Tn {0, ∞} = Tn∗ ωf−i , {0, ∞}
h i
= ωa−n (fi )fi , {0, ∞}
an (fi )L(fi , 1)
=− .
iπ
References
[AR79] Avner Ash and Lee Rudolph. “The modular symbol and contin-
ued fractions in higher dimensions”. In: Inventiones mathematicae
55.3 (1979), pp. 241–250.
[Ber19] Nicolas Bergeron. “Enlacement dans les fibrés en tores et fonc-
tions L de Hecke”. In: Congrès de la Société mathématique de
France Sémin. Congr. 33 (2019), pp. 173–186.
[BCG20] Nicolas Bergeron, Pierre Charollois, and Luis E. Garcia. “Trans-
gressions of the Euler class and Eisenstein cohomology of GLN (Z)”.
In: Japanese Journal of Mathematics 15.2 (2020), pp. 311–379.
[BCG23] Nicolas Bergeron, Pierre Charollois, and Luis E. Garcia. Cocycles
de groupe pour GLn et arrangements d’hyperplans. Vol. 39. CRM
Monograph Series. American Mathematical Society, Providence,
RI, 2023, p. 127.
[BMM16] Nicolas Bergeron, John Millson, and Colette Moeglin. “The Hodge
conjecture and arithmetic quotients of complex balls”. In: Acta
Mathematica 216.1 (2016), pp. 1 –125.
[Ber+17] Nicolas Bergeron et al. “The Noether-Lefschetz conjecture and
generalizations”. In: Inventiones mathematicae 208.2 (2017), pp. 501–
552.
[BGV03] N. Berline, E. Getzler, and M. Vergne. Heat Kernels and Dirac
Operators. Grundlehren Text Editions. Springer Berlin Heidel-
berg, 2003.
[BC92] Jean-Michel Bismut and Jeff Cheeger. “Transgressed Euler classes
of SL(2n, Z) vector bundles, adiabatic limits of eta invariants and
special values of L-functions”. In: Annales scientifiques de l’École
Normale Supérieure 25.4 (1992), pp. 335–391.
[BT82] Raoul Bott and Loring W Tu. Differential forms in algebraic
topology. 1st ed. Vol. 82. Springer-Verlag New York, 1982.
[Bra23] Romain Branchereau. “The Kudla-Millson form via the Mathai-
Quillen formalism”. In: Canadian Journal of Mathematics (2023),
pp. 1–26.
[CS16] Pierre Charollois and Robert Sczech. “Elliptic functions accord-
ing to Eisenstein and Kronecker: an update”. In: Eur. Math. Soc.
Newsl. 101 (2016), pp. 8–14.
[DPV21] Henri Darmon, Alice Pozzi, and Jan Vonk. “Diagonal restrictions
of p-adic Eisenstein families”. In: Mathematische Annalen 379.1
(2021), pp. 503–548.
[Fal05] Gerd Faltings. “Arithmetic Eisenstein Classes on the Siegel Space:
Some Computations”. In: Number Fields and Function Fields—
Two Parallel Worlds. Ed. by Gerard van der Geer, Ben Moonen,
and René Schoof. Birkhäuser Boston, 2005.
[Gar18] Luis E. Garcia. “Superconnections, theta series, and period do-
mains”. In: Advances in Mathematics 329 (2018), pp. 555 –589.
[GP74] Victor Guillemin and Alan Pollack. Differential topology. Prentice-
Hall, Inc., Englewood Cliffs, NJ, 1974, pp. xvi+222.
[KMS00] D. Kazhdan, B. Mazur, and C.-G. Schmidt. “Relative modular

symbols and Rankin-Selberg convolutions”. In: J. Reine Angew.
Math. 519 (2000), pp. 97–141.
[Kud97] Stephen S. Kudla. “Central Derivatives of Eisenstein Series and
Height Pairings”. In: Annals of Mathematics 146.3 (1997), pp. 545–
646.
[KM86] Stephen S. Kudla and John J. Millson. “The theta correspon-
dence and harmonic forms. I”. In: Mathematische Annalen 274.3
(1986), pp. 353–378.
[KM87] Stephen S. Kudla and John J. Millson. “The theta correspon-
dence and harmonic forms. II”. In: Mathematische Annalen 277.2
(1987), pp. 267–314.
[KM88] Stephen S. Kudla and John J. Millson. “Tubes, Cohomology with
Growth Conditions and an Application to the Theta Correspon-
dence”. In: vol. 40. 1. Cambridge University Press, 1988, pp. 1–
37.
[KM90] Stephen S. Kudla and John J. Millson. “Intersection numbers
of cycles on locally symmetric spaces and Fourier coefficients of
holomorphic modular forms in several complex variables”. en. In:
Publications Mathématiques de l’IHÉS 71 (1990), pp. 121–172.
[KRY99] Stephen S. Kudla, Michael Rapoport, and Tonghai Yang. “On the
derivative of an eisenstein series of weight one”. In: International
Mathematics Research Notices 1999.7 (Jan. 1999), pp. 347–385.
[MQ86] Varghese Mathai and Daniel Quillen. “Superconnections, thom
classes, and equivariant differential forms”. In: Topology 25.1 (1986),
pp. 85–110.
[Maz] Barry Mazur. “Arithmetic in the geometry of symmetric spaces”.
Unpublished letter to John Millson, available on Barry Mazur’s
webpage. url: https://bpb-us-e1.wpmucdn.com/sites.harvard.edu/dist/a/189/file
[Miy89] Toshitsune Miyake. “Modular Groups and Modular Forms”. In:
Modular Forms. Berlin, Heidelberg: Springer Berlin Heidelberg,
1989, pp. 96–194.
[Rud76] Walter Rudin. Principles of Mathematical Analysis, third edition.
McGraw-Hill, Inc., 1976.
[Sch93] Claus-G. Schmidt. “Relative modular symbols and p-adic Rankin-
Selberg convolutions”. In: Invent. Math. 112.1 (1993), pp. 31–76.
[Shi23] Yousheng Shi. “Theta series and generalized special cycles on
Hermitian locally symmetric manifolds”. In: Mathematische An-
nalen 386.3 (2023), pp. 1463–1553.
[Wei64] André Weil. “Sur certains groupes d’opérateurs unitaires”. In:
Acta Mathematica 111 (1964), pp. 143 –211.
[Zha20] Hao Zhang. “Elliptic Cocycle for GLN (Z) and Hecke Operators”.
Sorbonne Université. PhD thesis. 2020. url: https://www.imj-prg.fr/theses/pdf/hao_

Eisenstein Classes and Generating Series of Modular Symbols in

Uploaded by

Copyright:

Available Formats

Eisenstein Classes and Generating Series of Modular Symbols in

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Eisenstein Classes and Generating Series of Modular Symbols in

Uploaded by

Copyright:

Available Formats

EISENSTEIN CLASSES AND GENERATING SERIES OF

MODULAR SYMBOLS IN SLN

On the other hand, we can integrate the theta kernel on a cycle C in

For each vector v ∈ R2N there is a natural submanifold Sv ⊂ S of codimen-

of codimension N − 1 in the Borel-Moore homology. We consider the finite

where hτ ∈ SL2 (R) is any matrix sending i to τ ∈ H by Möbius transforma-

GLN (Q) SL2 (Q).

where z = (z1 , u) is the variable on S × R>0 and we integrate over u.

Theorem 1.1. The pushforward (1.2) converges absolutely for any s ∈ C.

where [Zn (χ)] denotes the Poincaré dual to Zn (χ).

The constant term z(χ) ∈ H N −1 (SΓ ) is a canonical Eisenstein class ob-

and for µ ∈ CN we set

where dm := |σ −1 (m)|. The function

Theorem 1.2. When Re(s) > N we have

Note that we see the right-hand side as an SO(N )-invariant form on

The following theorem is analogous to [KM88, Theorem. 4.5].

Theorem 1.3. We have

that are translates of a split torus by an invertible matrix Q ∈ MatN (Z).

In general, the period of Eϕχ (z1 , τ, s) over an arbitrary modular symbol

where v0 = (a1 , . . . , aN , b1 , . . . , bN ) and the forms E1 (τ, λ0 ) are weight one

The constant term is a linear combination of values at s = 0 of partial

(1.6) Eϕχ : H1 (Y0 (p); C) −→ M2 (Γ0 (p)).

We will show that the special cycles

In particular, if we combine it with Theorem 1.5 we recover [DPV21,

Eϕχ : H1 (Y0 (p); C) −→ M02 (Γ0 (p)).

1.7. Outline of the paper.

1.8. Acknowledgements. I thank Pierre Charollois, Henri Darmon and

2. Locally symmetric spaces and special cycles

It preserves the pairing

Proposition 2.1. The restriction of the projection p to Xv is a diffeomor-

We can also view the submanifold Sv as an embedded copy of a smaller

Proposition 2.2. When w1 is positive, the submanifold Sv is the image of

for some (N −1)×(N −1)-matrix z ′ . In fact, it is the image of the embedding

Example. (1) Let p be an integer and let L = pZ2N . If χ = 1v0 +L is the

then χ is preserved by the subgroup Γ0 (p, N ) of upper triangular matrices

Proposition 2.3. For any positive integer n the set

Proof. Let L∗ = Zα1 ⊕· · ·⊕ZαN be a Z-basis of L∗ and let g0 = [α1 , . . . , αN ] ∈

Recall that ργ acts by γ −t in the second factor. Reducing the remaining

The embedding p ◦ jv passes to the embedding of the quotient

The sum is finite since χ is supported on L∨ .

Example. Let L = pZ2N and suppose that Γ = Γ(p2 ). For v = (e1 , e1 )

2.3.2. Orientability. The submanifold Xv is the zero locus of the function

where N Xv is the normal bundle. The kernel of the differential is Tz Xv so

o(Tz X) = o(Tz Xv ) ∧ o(Nz Xv )

The orientation of Sv then comes from the diffeomorphism Xv ≃ Sv . Note

o(Tz X−v ) = (−1)N o(Tz Xv ).

3. Thom forms and the Mathai-Quillen formalism

cohomology, we get the long exact sequence

H i−1 (EΓ r DEΓ ) H i (EΓ , EΓ r DEΓ ) H i (EΓ )

H i (EΓ r DEΓ ) H i+1 (EΓ , EΓ r DEΓ ) H i+1 (EΓ ).

3.1.1. Rapidly decreasing forms. A function φ ∈ C 0 (RN ) is said to be rapidly

Lemma 3.1. For a function φ ∈ C 0 (RN ) the following are equivalent

Proof. Suppose that φ is rapidly decreasing. By expanding (1 + kvk)M into

O(kvk−M ) and let (a1 , . . . , aN ) ∈ NN . Let M := N max{a1 , . . . , aN }. Then

A smooth function φ ∈ C ∞ (RN ) is in the space S(RN ) of Schwartz func-

3.2. Mathai-Quillen construction. In this section we explain how Mathai

that represents the Thom class.

k−1 (v1 ∧ · · · ∧ vj ) = (k−1 v1 ) ∧ · · · ∧ (k−1 vj ) in the right factor. We also define

With this bilinear form the splitting ∧RN = ⊕N j=0 ∧ R