Polylogarithms For GL

Polylogarithms for GL2
over totally real fields
DISSERTATION ZUR ERLANGUNG DES DOKTORGRADES

DER NATURWISSENSCHAFTEN (DR. RER. NAT.)
DER FAKULTÄT FÜR MATHEMATIK
DER UNIVERSITÄT REGENSBURG
vorgelegt von
Philipp Graf aus
Regensburg
im Jahr 2015
Die Arbeit wurde angeleitet von: Prof. Dr. Guido Kings
Contents
1 Introduction and overview 1

1.1 Eisenstein series and cohomology . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Eisenstein cohomology for Hilbert-Blumenthal varieties . . . . . . . . . . 2
1.3 The toplogical polylogarithm and associated Eisenstein classes . . . . . . 3
1.4 Decomposition of Eisenstein classes . . . . . . . . . . . . . . . . . . . . . 5
1.5 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5.1 The topological polylogarithm . . . . . . . . . . . . . . . . . . . . . 7
1.5.2 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal-varieties 9
1.5.3 Comparison with Harder’s Eisenstein classes . . . . . . . . . . . . 10
1.6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 The topological polylogarithm 15

2.1 The logarithm sheaf on families of topological tori . . . . . . . . . . . . . . 15
2.1.1 Families of topological tori . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.2 Definition of the logarithm sheaf . . . . . . . . . . . . . . . . . . . 18
2.2 Limits of locally constant sheaves . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 limits over the natural numbers . . . . . . . . . . . . . . . . . . . . 19
2.2.2 (M-L) systems of locally constant sheaves . . . . . . . . . . . . . . 21
2.3 Purity and localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 Purity for local systems . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.2 Purity for systems of local systems and naturality . . . . . . . . . . 25
2.4 Construction of the topological polylogarithm . . . . . . . . . . . . . . . . 30
2.4.1 The localization sequence for the logarithm sheaf . . . . . . . . . . 30
2.4.2 Trivializations of the logarithm sheaf . . . . . . . . . . . . . . . . . 33
2.4.3 A differential equation for the polylogarithm . . . . . . . . . . . . . 40
2.4.4 Invariant functionals and quotient manifolds . . . . . . . . . . . . . 43
i
Contents
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties 47

3.1 The topological situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.2 The spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2 Construction of adelic polylogarithmic Eisenstein classes . . . . . . . . . 53
3.2.1 Coefficient sheaves and integral structures . . . . . . . . . . . . . 53
3.2.2 Torsion sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.3 Definition of adelic Eisenstein classes . . . . . . . . . . . . . . . . 55
3.3 Pushforward of polylogarithmic Eisenstein-classes . . . . . . . . . . . . . 59
3.3.1 Explicit description of the decomposition of the cohomology . . . . 69
4 Comparison with Harder’s Eisenstein classes 73

4.1 Nori’s calculation of the polylogarithm as current . . . . . . . . . . . . . . 73
4.1.1 Integral cohomology classes . . . . . . . . . . . . . . . . . . . . . 73
4.1.2 Solving the differential equation . . . . . . . . . . . . . . . . . . . . 75
4.2 Global description of the polylogarithm in adelic coordinates . . . . . . . . 81
4.2.1 Fourier analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.2.2 The adelic polylogarithm . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.3 Analytic continuation of the polylogarithm . . . . . . . . . . . . . . 85
4.3 The decomposition Isomorphism as Mellin-Tranformation . . . . . . . . . 87
4.4 Cohomology of the boundary . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary . . . . . 99
4.6 Comparison with Harders Eisenstein classes . . . . . . . . . . . . . . . . 110
4.7 The image of the polylogarithmic Eisenstein Operator . . . . . . . . . . . 117
4.7.1 Surjectivity of the horospherical map . . . . . . . . . . . . . . . . . 118
4.7.2 The spherical functions . . . . . . . . . . . . . . . . . . . . . . . . 123
4.7.3 The long exact sequence for the cohomology of the boundary . . . 125
4.7.4 Determination of the image . . . . . . . . . . . . . . . . . . . . . . 127
Bibliography 129
ii
1 Introduction and overview
1.1 Eisenstein series and cohomology
Let G be a reductive linear algebraic group over Q and A the adeles over Q. Let us
consider the double quotient space
XG,K := K \ G (A)/G (Q) AG (R)0 .
Here AG is the maximal split torus in the center of G, AG (R)0 ⊂ AG (R) is the connected
component of the identity, K = K∞ K f with K f ⊂ G (A f ) compact open and K∞ ⊂
G (R) maximal compact. The space XG,K has the Borel-Serre compactification XG,K .
It is a manifold with boundary, which has a stratification with respect to classes { P} of
associate parabolic Q-subgroups of G.
A representation ( E, ρ) of G in a finite dimensional Q-vectorspace defines a local system

Ẽ on XG,K and one can consider the cohomology groups H • ( XG,K , Ẽ) and H • ( XG , E
e) :=
limK H • ( XG,K , E
e). One can think of the last cohomology group as the colimit of the
−→ f
group cohomologies of all arithmetic subgroups Γ ⊂ G (Q) with values in E. From this
point of view it is clear that these cohomology groups are of fundamental arithmetic
interest.
When coefficients are extended to C, Franke (1998) showed that the cohomology
classes in these groups may be represented by automorphic forms and that one has
a direct sum decomposition of the cohomology with respect to classes { P} of asso-
ciate parabolic Q-subgroups of G as G (A f )-module, whenever AG acts by a central
character on E. The summand corresponding to the parabolic G itself is denoted by
• (X , E
Hcusp G
^ ⊗ C) and is called the cuspidal cohomology. It is build up by cusp forms, in
other words, automorphic forms whose constant terms at parabolic subgroups different
from G are zero. The cuspidal cohomology does not contribute to the cohomology of
the boundary. The complement to the cuspidal cohomology in H • ( XG , E
^ ⊗ C) is build
1
up by Eisenstein series and residues of such.
As this decomposition of the cohomology is obtained by the theory of Eisenstein series

and therefore by an analytic summation process, it is not clear whether it respects the
natural Q-structure of the cohomology. The Q-rationality of the decomposition has been
proven in the case G = ResK/Q GLn with K/Q a numberfield, see Franke (1998) Theo-
rem 20 and Franke&Schwermer (1998) 4.3. Theorem, using a classification theorem of
Jacquet & Shalika (1981) of cuspidal automorphic representations of GLn (AK ).
The Q-rationality of the decomposition above is of great arithmetic interest, as it can be

used to derive rationality results for special values of L-functions, which may occur as
constant terms of Eisenstein series or as integrals of Eisenstein series over cycles, see
Harder (1991) 3.1 or Harder (1987) (4.2.2) and V.
1.2 Eisenstein cohomology for Hilbert-Blumenthal varieties
As the main goal of this thesis is the construction of so called Eisenstein cohomology
classes for G = Res F/Q GL2 with F a totally real number field, we need to recall Harder’s
definition of Eisenstein cohomology in this situation in more detail. Harder (1987) con-
0
siders the space XG,K := K \ G (A)/G (Q) Z (R)0 , where Z ⊂ G is the center and Z (R)0
e on X 0 associated to G
the connected component of the identity, and local systems E G,K
0
representations ( E, ρ). XG,K is a disjoint union of so called Hilbert-Blumenthal varieties
and also has the Borel-Serre compactification whose boundary is homotopy equivalent
0
to ∂XG,K := K \ G (A)/B(Q) Z (R)0 with B ⊂ G the standard Borel subgroup of upper
0
triangular matrices. The natural map ∂XG,K 0
→ XG,K induces by pullback a restriction
map on cohomology resK : H • ( XG,K
0 ,Ee) → H • (∂X 0 , E
G,K
e) and by passing to the colimit
a G (A f )-equivariant map res : H • ( XG
0 ,E
e) → H • (∂X 0 , E
G
e).
• (X0 , E
The subspace HEis e), such that res : H • ( X 0 , E
e) ⊂ H • ( X 0 , E e) → im(res) is an
G G Eis G
Isomorphism, is called the Eisenstein cohomology.
Harder determines the Eisenstein cohomology in two steps. First he describes the
cohomology of the boundary. It may be understood as a sum of induced modules
G (A )
Ind B(A f ) Cφ f , where φ f is the finite component of an algebraic Hecke-character φ :
f
T (A)/T (Q) → C∗ and T is the maximal torus in G, see Harder (1987) Theorem 1.
0 is represented by a B (Q)-invariant dif-
More precisely, if a cohomology class on ∂XG
2
1.3 The toplogical polylogarithm and associated Eisenstein classes
ferential form ω, then ω is in particular U (Q)- invariant, where U ⊂ B is the unipotent

radical, and therefore ω may be developed into a Fourier series with respect to the
group U (Q). The class ω is then already determined by the constant term of the differ-
ential form ω, this means its zero-Fourier-coefficient.
As the second step Harder constructs a G (A f )-equivariant operator
•
Eis : im(res) ⊗ C → HEis ( XG0 , Ee ⊗ C),
which is a section for res. If Ẽ is not isomorphic to C, then Eis may be explicitly described
as follows: If ω ∈ im(res) ⊗ C is represented by a B(Q)-invariant differential form, then
Eis(ω ) := ∑ γ∗ ω and one can conclude that res is surjective. If Ẽ ∼
γ∈ G (Q)/B(Q) = C, the
sum above has to be understood in terms of analytic continuation and it turns out that
0 , C → H 2g−1 ( ∂X 0 , C) is not surjective, when g : = [ F : Q].
res : H 2g−1 ( XG G
1.3 The toplogical polylogarithm and associated Eisenstein

classes
The connection between Eisenstein classes and special values of L-functions may be
seen as motiviation to construct Q-rational Eisenstein classes geometrically. One way
to do so is to specialize polylogarithms. Beı̆linson & Levin (1994) constructed poly-
logarithms for relative elliptic curves. In the case of the universal elliptic curve with
level N-structure E → XG,K , G = SL2 , K f = ker(SL2 (Ẑ) → SL2 (Z/NZ)), Beilinson
proved that the polylogarithm specialized along non-zero N-torsion sections actually
yields Eisenstein classes for SL2 . Wildeshaus (1997) generalized the construction of
the polylogarithm to the setting of Shimura varieties, which also included the definiton
for abelian schemes.
Following the ideas of Beilinson and Nori (1995) we can adapt this construction eas-
ily to our topological situation. Given a group G as before with a finite dimensional
representation ψ : G → Aut(V ) we may consider the space
π : TG,K := V (Ẑ) o KAG (R)0 \ V (A) o G (A)/V (Q) o G (Q) → XG,K
for K f ⊂ ker( Aut(V (Ẑ)) → Aut(V (Z/NZ))). It is a group-object over XG,K and its
fibers are topological tori, in other words, isomorphic to products of S1 . One has the
pro-local-system Log associated to the V (Q) o G (Q)-module ∏k≥0 Symk V (Q), where
3
V (Q) acts by multiplication with the exponential series and G (Q) via ψ. Furthermore,
we consider D ⊂ TG,K , which is the union of the images of non-zero N-torsion sections
with open complement U, and the relative orientation bundle µ := det ^ (V ). From the
cohomological vanishing properties of Log one easily derives short exact localization
sequences
0 → H dim(V )−1 (U, Log ⊗ π −1 µn+1 ) → H 0 ( D, π|−D1 ∏ Symk V

e ⊗ µn ) → H 0 ( D, Q) → 0
k ≥0
for all n ∈ Z. Given a section f ∈ H 0 ( D, π|−D1 ∏k≥0 Symk V

e ⊗ µn ) mapping to zero on the
right hand side we get a unique cohomology class called the polylogarithm associated
to f
pol( f ) ∈ H dim(V )−1 (U, Log ⊗ π −1 µn+1 ).
This class may be specialized along the zero section to obtain polylogarithmic Eisen-
stein classes
(Eisk ( f ))k≥0 := 0∗ pol ( f ) ∈ ∏ H dim(V )−1 (XG,K , Symk Ve ⊗ µn+1 ).

k ≥0
Typical examples for such f may be given by functions f : V (Z/NZ) → Q with f (0) =
0 and ∑v∈V (Z/NZ) f (v) = 0. By the naturality properties of the polylogarithm these
polylogarithmic Eisenstein classes glue to G (A f )-equivariant operators
Eisk : S(V (A f ), Q)0 ⊗ H 0 ( XG , µn ) → H dim(V )−1 ( XG , Symk V

e ⊗ µ n +1 ),
where S(V (A f ), Q)0 are the Q-valued Schwartz-Bruhat functions with

R
V (A f )
f (v)dv =
f (0) = 0. Levin (2000) and Nori (1995) represented the polylogarithm cohomology
classes by explicit currents in the cohomology with C-coeffiecients. The polylogarith-
mic Eisenstein classes are then represented by Eisenstein-Kronecker series. With
this explicit description Blottière (2009a) and Kings (2008) proved in the situation G =
Res F/Q SL2 with F/Q a totally real number field, that again the polylogarithmic Eisen-
stein classes are non-trivial Eisenstein classes, as their constant terms turned out to be
special values of partial L-functions associated to the field F.
Here we only addressed the topological realization of the polylogarithm. Indeed, by

Kings (1999) the polylogarithm for abelian schemes is of motivic origin and a power-
ful tool to tackle deep arithmetic problems and conjectures. See for example Kings
(2001), where the étale elliptic polylogarithm is a decisive instrument for the proof of
the Tamagawa number conjecture for CM elliptic curves.
4
1.4 Decomposition of Eisenstein classes
1.4 Decomposition of Eisenstein classes
Even though the polygarithmic Eisenstein classes are a mighty tool in arithmetic, there
is one obvious flaw: They are all stuck in cohomological degree dim(V ) − 1.
In this work we present a decomposition principle for polylogarithmic Eisenstein classes

in the case G = Res F/Q GL2 , V = Res F/K G2a and F/Q a totally real number field.
The idea is very easy. We may see XG,K as a fiber bundle
0
ϕ : XG,K → XG,K = K \ G (A)/G (Q) Z (R)0 ,
0
where again XG,K is a space as considered by Harder (1987) in his landmark paper.
The fiber is
ϕ−1 (1) = (K∞ AG (R)0 ∩ Z (R)) \ Z (R)/Z (K )
and Z (K ) := Z (Q) ∩ K f ⊂ O F× is a subgroup of finite index of the units of O F ⊂ F the

ring of integers. By Dirichlet’s unit theorem we know that the fiber has to be compact
and that we have for the cohomology
•
H • ( ϕ−1 (1), Q) = H • ( Z (K ), Q) = H • (O F× , Q) = Hom(O F× , Q).
^
Using the coordinate on G (R) coming from the determinant we see that ϕ is up to a fi-
nite covering a trivial bundle. So one has cohomology classes in H • ( XG,K , Q) restricting
to a basis of the cohomology of all fibers of ϕ and this gives a Leray-Hirsch isomorphism
H • ( ϕ−1 (1), Q) ⊗ H • ( XG,K

0 e ⊗ µn+1 ) → H • ( XG,K , Symk V
, ϕ∗ Symk V e ⊗ µ n +1 ).
The polylogarithmic Eisenstein classes may then be decomposed with respect to this
isomorphism. We get by evaluation
Eiskq : S(V (A f ), Q)0 ⊗ H 0 ( XG , µn ) ⊗ H q ( ϕ−1 (1), Q)∗ → H dim(V )−1−q ( XG , Symk V

e ⊗ µ n +1 )
for q = 0, ..., g − 1 = dim( ϕ−1 (1)).
We want to show that this decomposition of polylogarithmic Eisenstein classes is as

non-trivial as possible. In other words, we want to get as many of Harder’s Eisenstein
cohomology classes as possible. To do so we follow Levin (2000) and Nori (1995)
to represent the polylogarithm by a current and hence the polylogarithmic Eisenstein
classes by differential forms in de Rham cohomology. The decomposition isomorphism
5
will be made explicit by fiber integration on the level of de Rham cohomology. This
gives the decomposed polylogarithmic Eisenstein classes as Mellin-transformations of
theta-series as considered by Wielonsky (1985).
To determine the image of our polylogarithmic Eisenstein classes we calculate the con-
stant terms, in other words, the restriction to the cohomology of the boundary. We al-
ready have mentioned that we have thanks to Harder (1987) a complete understanding
G (A )
0
of the cohomology of the boundary of XG,K as a sum of induced modules Ind B(A f ) Cφ f .
f
The precise relation between f and the constant term of the polylogarithmic Eisenstein
class Eisk ( f ) is then controlled by the horospherical map
G (A )
ρ : S(V (A f ), Q)0 ⊗ H 0 ( XG , µn ) → Ind B(A f ) Cφ f .
M
f
The horospherical map gives also the relation to special values of L-functions.
To understand the image of our polylogarithmic Eisenstein classes we have to deter-

mine which of the induced functions above may be realized by the horospherical map.
Our main result is as follows
Theorem. The operators Eiskq factor through the Eisenstein cohomology. The image of
Eiskq is the whole Eisenstein cohomology in degree 2g − 1 − q, if k > 0. If k = 0, Eis00
generates the Eisenstein cohomology in degree 2g − 1, for q > 0 we get all Eisenstein
classes but not those associated to spherical functions.
We want to remark a few things.
(a) The spherical functions are exactly those functions which are not in the image of
the restriction map to the cohomology of the boundary in degree 2g − 1. As the
Eiskq ( f ) have the same constant term for all q, we cannot obtain the Eisenstein
classes associated to spherical functions in cohomological degrees g, ..., 2g − 1.
0
(b) The Eisenstein cohomology of XG,K is supported in cohomological degrees 0 and
g, ..., 2g − 1, so that our Eisenstein operator actually generates most of the relevant
part.
(c) Even though Harder started with more general coefficient systems than we do,
see Harder (1987) 1.4, we get all Eisenstein cohomology classes for non-trivial
representations. The reason is that his representations are direct summands in
Symk V ⊗ det(V )n , k ≥ 1, and all weights φ|T (R)0 occuring in the cohomology of
6
1.5 Outline of the thesis
the boundary have to factor through the norm character, see Harder (1987) 2.8.
(d) We even get a much finer result. The sheaf Log has an integral structure which
allows us to define our polylogarithmic Eisenstein classes over the Ring Z[ N1 ],
when we consider XG,K and K f defines the level-N-structure. From this we also
deduce integrality results for special values of partial L-functions for totally real
fields.
(e) A proof of a decomposition isomorphism
H • ( ϕ−1 (1), Q) ⊗ H • ( XG,K

0 e ⊗ µn+1 ) → H • ( XG,K , Symk V
, ϕ∗ Symk V e ⊗ µ n +1 ).
for general reductive groups G with center Z seems to be straight forward and
is work in progress. However, it seems as if one cannot expect that the poly-
logarithm always generates much of the Eisenstein cohomology. To demon-
√
strate this problem look at K = Q( −1) and G = ResK/Q GL2 . One would
naively choose V = ResK/Q G2a . dim(V ) = 4 and this gives polylogarithmic Eisen-
0 , since
stein classes in cohomological degree 3. Now we have XG,K = XG,K
AG (R)0 K∞ = R>0 U (2) = C× U (2) = Z (R)0 K∞ and the part of Z (R), which
does not split over R, already lies in the maximal compact subgroup. So there
is no decomposition and H 3 ( XG,K , E
e) = 0 by Poincaré duality. In this case the
polylogarithm gives nothing.
This thesis has three main parts. We want to discuss them here in more detail.
1.5.1 The topological polylogarithm
In the first part we define the general geometric setup. We begin with the definition
of families of topological tori over a manifold S as a proper submersion π : T → S
with connected fibers, which is a commutative group-object over S. If S is connected
one identifies the category of families of topological tori with the category of finitely
generated free abelian groups L with π1 (S)-action (Proposition 2.1.4). This is done by
associating to L the quotient-torus ( L ⊗ R/L) ×π1 (S) S.
e Here Se is the universal cover of
7
S, where π1 (S) acts on by deck-transformations, and π1 (S) acts on the left factor by the
module structure. So the torus T is determined by a representation π1 (S) → Aut( L),
which we call the representation associated to T.
This allows us to define locally constant sheaves on T by using L o π1 (S)-modules.

Let k be any noetherian ring. L o π1 (S) acts on the set L by affine transformations.
aug
This makes the group ring k [ L] into a L o π1 (S)-module. The augmentation k [ L] →
k is equivariant making the augmentation Ideal a := ker(aug) a L o π1 (S)-module.
Consequently, k [[ L]] := limn∈N k [ L]/an+1 is a L o π1 (S)-module and the associated
←− 0
locally constant sheaf on T is called the logarithm sheaf Log.
We deduce then a purity result for projective systems of local systems indexed over the
natural numbers. If i : D → T is a closed submanifold of codimension c of our torus
we get Ri! Log = i−1 Log ⊗ or D/T [−c] (Proposition 2.3.5), where or D/T is the relative
orientation bundle. This gives a natural localization triangle
+1
i∗ i−1 Log ⊗ or D/T [−c] → Log → j∗ j−1 Log →,
where j : U → T is the inclusion of the open complement of D (Proposition 2.3.8). We

calculate the well known higher right-derived images of Log
aug∼
=
R p π∗ ( Log) = 0, p 6= d, Rd π∗ ( Log) → Rd π∗ (k ),
where d is the fiber dimension of π. From this we deduce the localization sequence
aug
0 → H d−1 (U, Log ⊗ or T/S
n +1
) → H 0 ( D, i−1 Log ⊗ orT/S
n n
) → H 0 ( D, orT/S ),
when π : D → S is a finite cover (Proposition 2.4.5).
To profit from the localization sequence, this means to get non-trivial classes in H d−1 (U, Log ⊗
n +1
or T/S ), we need a good understanding of H 0 ( D, i−1 Log ⊗ orT/S
n ). Following Beilinson
et al. (2014) we show that there is a section k → i−1 Log of aug, if D is the union of
the images of torsion sections, whose order is invertible in k (Proposition 2.4.9). This
gives us the definition of pol( f ) ∈ H d−1 (U, Log ⊗ or T/S ) for locally constant functions
f : D → k with aug( f ) = 0, as we have H 0 ( D, k ) ⊂ H 0 ( D, i−1 Log) (Proposition 2.4.14).
For computational purposes we also need a trivialization of the pro-vectorbundle asso-

ciated to Log. It comes from the isomorphism
Q[[ L]] → ∏ Symk ( L ⊗ Q), l 7→ exp(l ),

k ≥0
8
which gives the nowhere vanishing C ∞ -section v 7→ exp(−v), v ∈ L ⊗ R, and the rela-
tion of Log to the symmetric powers of the representation π1 (S) → Aut( L) associated
to T: 0−1 Log = ∏k≥0 Symk ( L^⊗ Q) (Proposition 2.4.12), where 0 : S → T is the zero
section of the group-object T.
Moreover, we show that the localization sequence of Log may be calculated by resolving
the Logarithm sheaf by non-continuous functionals. This gives a characterization of the
polylogarithms by differential equations (Proposition 2.4.19).
1.5.2 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal-varieties
0 ,when we have
First we recall the geometric situation ϕ : MK := XG,K → SK := XG,K
G = Res F/Q GL2 for a totally real field F. Let us keep the notation already fixed in
Section 1.3 and Section 1.4. We construct the tori over MK using the standard repre-
sentation G → Aut(V ), V = Res F/Q G2a . This allows us to define polylogarithms and
polylogarithmic Eisenstein classes for MK . We use the naturality properties of Log and
the localization sequence to show that we can glue our Eisenstein classes to G (A f )-
equivariant operators as described in Section 1.3 (Proposition 3.2.3).
As we have the polylogarithmic Eisenstein classes on MK , we want to decompose the

cohomology. The proof is divided up into several parts.
e ⊗ µn+1 . As ϕ is a fiber bundle it

First we calculate the higher direct images of Symk V
suffices to understand the cohomology of the fibers, which we calculate using group
cohomology H • ( Z (K ), Symk V ⊗ det(V )n+1 ). As the action of Z (K ) is semi-simple we
easily calculate these groups as H 0 ( Z (K ), Symk V ⊗ det(V )n+1 ) ⊗ H • ( Z (K ), Q) (Propo-
sition 3.3.2).
This gives consequently R• ϕ∗ (Symk V e ⊗ µn+1 ⊗ R• ϕ∗ (Q) (Corol-

e ⊗ µn+1 ) = ϕ∗ Symk V
lary 3.3.3) and we trivialize R• ϕ∗ (Q) by forms coming from global classes H• ⊂ H • (MK , Q)
using the fact that ϕ is up to a finite cover a trivial fiber bundle (Proposition 3.3.9).
These global classes define then by cup-product the decomposition isomorphism (The-
orem 3.3.11)
∪
H • (SK , ϕ∗ Symk V
e ⊗ µn+1 ) ⊗ H• → H • (MK , Symk V
e ⊗ µ n +1 ),
which we finally discuss in the setting of de Rham cohomology with the theory of fiber
integration (Section 3.3.1).
9
1.5.3 Comparison with Harder’s Eisenstein classes
In the last part we represent the polylogarithmic Eisenstein classes by differential forms
and compare them with those of Harder.
Using the ideas of Nori (1995) we represent the polylogarithms explicitly by currents
describing the Eisenstein classes as differential forms. We give then the description
in adelic coordinates and calculate the decomposition of the polylogarithmic Eisen-
stein classes by fiber integration (Lemma 4.3.6). We reinterprete this as a Mellin-
transformation of a theta-series as considered by Wielonsky (1985) (Theorem 4.3.10).
As Harder’s Eisenstein classes are determined by their restriction to the cohomlogy of

the boundary, we recall Harder’s calculation of the cohomology of the boundary of SK
(Section 4.4). We restrict then the polylogarithmic Eisenstein classes to the boundary
where they are determined by their constant terms (Proposition 4.5.1). We derive ratio-
nality and integrality results for these constant terms (Proposition 4.5.5) and define the
horospherical map controlling the relation between f and the constant term of Eisk ( f ).
We then explain that the horospherical map determines the image of our Eisenstein
operators (Proposition 4.5.7).
Next we translate our cohomolgy classes to (g, K )-cohomology and compare them there
with Harder’s Eisenstein operator. We see that the polylogarihmic Eisenstein classes
are in the image of Harder’s Eisenstein operator and therefore our operators Eiskq actu-
ally factor through the Eisenstein cohomology (Proposition 4.6.7).
Finally, we determine the image of our operators Eiskq by studying the horospherical
map. The main ingredient is to show that the horospherical map is surjective, when one
allows general Schwartz-Bruhat functions, this means functions f where not necessarily
R
V (A )
f (v)dv = f (0) = 0 (Proposition 4.7.1).
f
1.6 Acknowledgements
I would like to thank my advisor Guido Kings for giving me the opportunity to work on
his beautiful Idea to decompose the polylogarithmic Eisenstein classes for Res F/Q GL2
by using the units in the center. His encouragement was decisive for the success of this
work and i want to thank him for everything I have learned during my studies.
10
1.6 Acknowledgements
Special thanks go to my colleagues and friends Sandra Eisenreich, Johannes Sprang

and Georg Tamme. Especially Johannes and Georg have never been tired to discuss
minor or big problems, let them be of mathematical nature or not.
Finally, i would like to thank Stefan Rameseder for helping me with latex and being such
a good friend.
11
To Patricia
for love and constant support
13
2 The topological polylogarithm
2.1 The logarithm sheaf on families of topological tori
We want to start with the definition and construction of the logarithm sheaf on families
of topological tori. These will be the main geometric objects. As we want to calculate
cohomology classes on them explicitly, we need practical and explicit descriptions for
locally constant sheaves on them. This achieved by the theory representations of the
fundamental group and equivariant sheaves on the universial cover.
2.1.1 Families of topological tori
For any site C we denote by Sh(C ) the category of abelian sheaves on C. For a C ∞ -
manifold S we denote by M f d/S the category of manifolds over S .
Definition 1. Let π : T → S be a proper submersion of C ∞ -manifolds. We call π :

T → S a family of topological tori or simply a torus over S , if it is a commutative group
object in M f d/S with connected fibers. Families of topological tori together with group
homomorphisms form a subcategory of M f d/S , which we denote by Tori/S .
Remark 2.1.1. As the fibers of π are compact commutative Lie-groups, they are topo-
logical tori, in other words products of S1 := {z ∈ C : |z| = 1}.
Remark 2.1.2. We will always consider M f d/S with the usual topology of open covers.
It follows that we have a Yoneda-embedding

Tori/S → Sh( M f d/S), T 7→ X 7→ T ( X ) = Hom M f d/S ( X, T )
Let us fix a basepoint s0 ∈ S and let us take t0 := 0(s0 ) ∈ T as basepoint for T ,

where 0 : S → T is the zero section. If S is not connected, we choose a basepoint for
any connected component and do all our constructions for each connected component
15
separately. Any proper submersion of C ∞ -manifolds is a fiber bundle, see Ehresmann

(1995). From this we deduce a short exact sequence
∗i π
0 → π1 (Ts0 , t0 ) → π1 (T , t0 ) →∗ π1 (S , s0 ) → 0,
where as usual Ts0 := π −1 (s0 ) denotes the fiber over s0 . This sequence is split exact,
since π∗ has the section 0∗ , and we have a natural left π1 (S , s0 )-action on π1 (Ts0 , t0 ) =
H1 (π −1 (s0 ), Z) given by conjugation, as the latter group is commutative. In other words,
there is an isomorphism
π1 (Ts0 , t0 ) o π1 (S , s0 ) ∼
= π1 (T , t0 ), (l, g) 7→ i∗ (l )0∗ ( g),
where the structure of the semi-direct product is determined by 0 and the by the action
above.
We are going to construct locally constant sheaves on T , which are induced by π1 (T , t0 )-
modules. The functor F 7→ Ft0 , which assigns to a locally constant sheaf F its stalk at
the basepoint, establishes an equivalence between the category of locally constant
sheaves on a connected topological manifold T and the category of left-π1 (T , t0 )-
modules, see Iversen (1986) IV Theorem 9.7. Note that we consider the universal
cover Te of T equipped with a right π1 (T , t0 )-action.
Example 2.1.3. If π : T → S is a torus, we have the locally constant sheaf H :=

HomZ ( R1 π∗ (Z), Z) and Hs0 = H1 (Ts0 , Z) by Poincare-duality.
Definition 2. A lattice L is a free Z-module of finite rank. Let k → k0 be a ring homo-

morphism and M a k-module. We set Mk0 := M ⊗k k0 .
Proposition 2.1.4. Let S be a connected C ∞ -manifold, say with basepoint s0 ∈ S . Then

we have an eqivalence of categories
Tori/S → {lattices L with left-π1 (S , s0 )-action and equivariant maps}
T 7→ H1 (Ts0 , Z)
Proof. Given a π1 (S , s0 )-module L we have the torus T := LR /L. As π1 (S , s0 ) acts on

L, we get an induced action on T. We denote by S̃ the universal cover of S and set
T L := S̃ ×π1 (S ,s0 ) T, where S̃ ×π1 (S ,s0 ) T := S̃ × T/π1 (S , s0 ) and γ ∈ π1 (S , s0 ) acts on
(s, t) ∈ S̃ × T by (s, t)γ = (sγ, γ−1 t). We have a structure map π : T L → S coming
from the first projection. The action of π1 (S , s0 ) on S̃ is properly discontinuous and
16
2.1 The logarithm sheaf on families of topological tori
fixpoint free. So S̃ may be covered by Ũ ⊂ S̃ open, with Ũγ ∩ Ũ 6= ∅ implies γ = id. If

we denote by U the image of Ũ in S under the canonical map, we get π −1 (U ) = U × T.
So our quotient space is a fiber bundle with typical fiber T and π is proper. Since the
action is by diffeomorphisms, T L is even a manifold.
We endow pr1 : U × T → U with the constant group structure as a group object in
manifolds over U. By construction all transition maps of our fiber bundle come from
the π1 (S , s0 )-action on T. So all transition maps are group homomorphisms and this
means that we may glue all group structures pr1 : U × T → U to obtain a global
group structure on T L . Obviously, equivariant maps f : L → L0 induce maps on the
corresponding spaces. This gives the desired quasi-inverse functor
{lattices L with left-π1 (S , s0 )-action and equivariant maps} → Tori/S
L 7→ S̃ ×π1 (S ,s0 ) LR /L =: T L .
Remark 2.1.5. We deduce from the theorem above that our family of topological tori
T has the universal cover S̃ × H1 (Ts0 , R) where the fundamental group π1 (t0 , T ) =
H1 (Ts0 , Z) o π1 (S , s0 ) acts by ( x, v) · (l, γ) = ( xγ, γ−1 (v + l )) and the action of π1 (S , s0 )
on H1 (Ts0 , Z) is the monodromy action coming from the locally constant sheaf H.
Definition 3. A group homomorphism φ : T 0 → T of tori over S is called an isogeny, if

φ is surjective and ker(φ) is a finite covering of S . Here
ker(φ)( X ) := ker(φ : T 0 ( X ) → T ( X )), X ∈ Ob( M f d/S)
Remark 2.1.6. The kernel of a homomorphism φ : T 0 → T of tori is always repre-

sentable as a basechange with the zero section. More precisely, Proposition 2.1.4
shows that ker(φ) = S̃ ×π1 (S ,s0 ) ker(φs0 ) where φs0 : Ts00 → Ts0 is the induced homomor-
phism on the fiber over s0 .
Remark 2.1.7. Given a torus T over S we have for any N ∈ Z the multiplication by N
isogeny [ N ] : T → T . On points X ∈ Ob( M f d/S) it is determined by the functorial
group-homomorphism
[ N ] : T ( X ) → T ( X ), v 7→ N · v.
We call ker([ N ]) =: T [ N ] the N-torsion subgroup, it is a finite covering of S .
17
Definition 4. Let φ : T 0 → T be an isogeny over S . The function
deg(φ) : S → Z, s 7→ | ker(φs )|
is locally constant and is called the degree of φ.
Remark 2.1.8. We have deg([ N ]) = N dim(T /S) where
dim(T /S) : S → Z, s 7→ dim(Ts ).
Definition 5. A family of topological tori π : T → S has a level N structure, if the

sheaf of sections associated to T [ N ] in Sh( M f d/S) is constant. If the fiber dimension
dim(T /S) is constant, this just means that there is an isomorphism
T [N] ∼
= S × (Z/NZ)dim(T /S) .
Remark 2.1.9. For any ring R we set GL( M ) := Aut R− Mod ( M ), if M is an finitely gener-
ated free R-module. If π : T → S has a level-N structure, the associated representation
ρ : π1 (S , s0 ) → GL( H1 (Ts0 , Z)) factors as ρ : π1 (S , s0 ) → GL( H1 (Ts0 , Z))( N ) where
1 → GL( H1 (Ts0 , Z))( N ) → GL( H1 (Ts0 , Z)) → GL( H1 (Ts0 , Z/NZ)) → 1
is exact with the natural projection on the right hand side.
2.1.2 Definition of the logarithm sheaf
Consider the group algebra k [π1 (Ts0 , t0 )] for any commutative ring k. We denote basis
elements by l, l ∈ π1 (Ts0 , t0 ), and write elements as ∑l f (l )l, where f (l ) ∈ k and
f (l ) = 0 for almost all l. The group π1 (T , t0 ) = π1 (Ts0 , t0 ) o π1 (S , s0 ) acts naturally
on the set π1 (Ts0 , t0 ) by (l, g) ∗ v = l + g · v, l, v ∈ π1 (Ts0 , t0 ) and g ∈ π1 (S , s0 ). We
can define this action directly in π1 (T , t0 ) by x ∗ v := xv0∗ π∗ ( x −1 ), x ∈ π1 (T , t0 ) and
v ∈ π1 (Ts0 , t0 ). This action induces a natural π1 (T , t0 )-action on the group k [π1 (Ts0 , t0 )].
The group algebra comes with the natural augmentation map
aug : k [π1 (Ts0 , t0 )] → k, ∑ f ( l ) l 7 → ∑ f ( l ).

l l
The augmentation map is π1 (T , t0 )-equivariant and the kernel a = ker(aug) is again a

π1 (T , t0 )-module. In this manner we get for n ∈ N0 the π1 (T , t0 )-modules k [π1 (Ts0 , t0 )]/an+1
and k [[π1 (Ts0 , t0 )]] := lim k [π1 (Ts0 , t0 )]/an+1 .
←−
18
2.2 Limits of locally constant sheaves
Definition 6. Let T be a torus over S . We define the locally constant sheaf LogTn /S on
T as the sheaf associated to the π1 (T , t0 )-module k[π1 (Ts0 , t0 )]/an+1 , n ∈ N0 , and call
it the n-th Logarithm sheaf. lim LogTn /S =: LogT /S is called the Logarithm sheaf.
←−
Remark 2.1.10. LogT /S is also locally constant and is associated to the π1 (T , t0 )-
module k [[π1 (Ts0 , t0 )]].
Lemma 2.1.11. LogTn /S and LogT /S are natural in families of topological tori π : T →
S . In other words, given a commutative square of (pointed) families of topological tori
p
T0 /T
π0 π
q
S0 /S
with p ◦ 00 = 0 ◦ q, we have natural morphisms LogTn 0 /S 0 → p−1 LogTn /S and LogT 0 /S 0 →

p−1 LogT /S . If the diagram above is cartesian, these maps are isomorphsims.
Proof. We choose s00 ∈ S 0 with q(s00 ) = s0 and set t00 = 00 (s00 ).

Given a locally constant sheaf F on T associated to a π1 (T , t0 )-module M we recog-
nize p−1 F as the locally constant sheaf associated to the π1 (T 0 , t00 )-module M where
the action is induced by the map p∗ : π1 (T 0 , t00 ) → π1 (T , t0 ). The morphism p∗ :
π1 (Ts00 , t00 ) → π1 (Ts0 , t0 ) is a map of π1 (T 0 , t00 )-sets. To see this choose x ∈ π1 (T 0 , t00 )
0
and v ∈ π1 (Ts00 , t00 ) arbitrarily. We have

0
p∗ ( x ∗ v) = p∗ ( xv00∗ π∗0 ( x −1 )) = p∗ ( x ) p∗ (v) p∗ 00∗ π∗0 ( x −1 ) =
p ∗ ( x ) p ∗ ( v )0∗ π ∗ p ∗ ( x −1 ) = p ∗ ( x ) ∗ p ∗ ( v )
So p∗ : k [π1 (Ts00 , t00 )] → k [π1 (Ts0 , t0 )] is π1 (T 0 , t00 )-equivariant. This induces the de-
0
sired maps on the level of sheaves. If our diagram is cartesian the morphism p∗ :
π1 (Ts00 , t00 ) → π1 (Ts0 , t0 ) is an isomorphism inducing the desired isomorphisms on our
0
sheaves.
2.2.1 limits over the natural numbers
In the case of sheaves there can be non-trivial R p lim for p ≥ 2 and there may be no
←−
such thing as a Mittag-Leffler condition on the system ensuring the vanishing of higher
19
limits. We want to prove now that locally constant sheaves have, as expected, the same
behaviour with limits as abelian groups. Let A be an abelian category. We denote by
AN the category of all inverse systems in A indexed by the natural numbers. Objects
of AN are families ( An , an ) with objects An in A and maps an : An+1 → An . Morphisms
between ( An , an ) and ( Bn , bn ) in AN are just commutative diagrams in A of the form
··· A n +1 / An / A n −1 / ···
f n +1 fn f n −1

··· Bn+1 / Bn / Bn−1 / ···
AN is an abelian category with kernels and cokernels defined componentwise.
Proposition 2.2.1. AN has enough injective objects, if and only if A has. We can
characterize Injectives in AN as follows:
I = (In , dn ) is injective, if and only if each In is injective in A and all the dn are split
epimorphisms.
Proof. Jannsen (1988) (1.1) proposition.
From now on we want to assume that A has enough injectives and that inverse limits
indexed over N exist in A. The functor lim : AN → A sending an inverse system to its
←−
limit is additive and left exact as right-adjoint to the exact diagonal-functor ∆ sending A
to the system
id id id
··· → A → A → A → ···
In particular lim preserves injective objects. Since AN has enough injectives and lim
←− ←−
is left-exact, we form the higher right-derived functors R p lim. We will often omit the
←−
transition maps of our system and simlpy write ( An ) for objects in AN . Let us consider
A = Ab the category of abelian groups first.
Definition 7. ( An , dn ) in AbN satisfies the Mittag-Leffler condition (M-L), if for any n ∈

N the decreasing filtration of An Fm := im( An+m → An ) induced by the dk , k ≥ n,
becomes stationary.
Remark 2.2.2. If all dn are surjective, then ( An , dn ) obviously satisfies (M-L).
Proposition 2.2.3. Let ( An ) be in AbN . Then R p lim( An ) = 0 for p ≥ 2. If ( An )

←−
satisfies (M-L), then R1 lim( An ) = 0.
←−
Proof. Weibel (1994) Corollary 3.5.4. and Proposition 3.5.7.
20
2.2.2 (M-L) systems of locally constant sheaves
Let X be a topological manifold. The category Sh( X ) of sheaves on X has enough

injectives and all limits over the natural numbers exist. For (Fn ) ∈ Sh( X )N the formula
(lim Fn )(U ) = lim Fn (U ), U ⊂ X open

←− ←−
can be used as definition.
Definition 8. We say (Fn ) ∈ Sh( X )N is a system of locally constant sheaves on X, if

each Fn is a locally constant sheaf. We say that the system of locally constant sheaves
(Fn ) satisfies the Mittag-Leffler condition (M-L), if for any x ∈ X the induced system of
abelian groups (Fn,x ) does so.
Proposition 2.2.4. Let (Fn ) be a system of locally constant sheaves on a topological

manifold X. Then R p lim(Fn ) = 0 for p ≥ 2 and if (Fn ) satisfies (M-L) R1 lim(Fn ) = 0.
←− ←−
Proof. Consider an injective resolution (Fn ) ,→ (In )• in Sh( X )N . We have by definition

R p lim(Fn ) = H p (lim In• ). H p (lim In• ) is the sheaf associated to the pre-sheaf
←− ←− ←−
U 7→ H p ((lim In )(U )• ) = H p (lim In (U )• ).

←− ←−
So it suffices to show that H p (lim In (U )• ) vanishes for any U ⊂ X open and con-
←−
tractible for p ≥ 2 or p ≥ 1, if (Fn ) is (M-L).
The complexes In (U )• are injective resolutions of Mn := Fn (U ).
To see this recall that In•|U is an injective resolution of the constant sheaf Mn = Fn|U .
Here we use that locally constant sheaves are constant on simply connected manifolds
(Iversen (1986) IV.9). So we may compute
H p (U, Mn ) = H p (U, Fn|U ) = H p (In•|U (U )) = H p (In• (U ))
Now U is contractible and by homotopy-invariance of sheaf-cohomology with constant

coefficients (Iversen (1986) IV.1) we get H p (U, Mn ) = H p ({ pt.} , Mn ) which is zero
for p ≥ 1 and Mn for p = 0. So In (U )• is a resolution of Mn and it is an injective
resolution, because the section functor Γ(U, ) has the constant sheaf construction as
p
exact right-adjoint. Since (In ) is injective in Sh( X )N , all its transition-functions are split
p
epimorphic. In particular, all transition functions of (In (U )) are split epimorphic. This
21
p
implies that (In (U )) is injective in AbN . Therefore ( Mn ) ,→ (In (U ))• is an injective
resolution in AbN implying R p lim( Mn ) = H p (lim In (U )• ). Now we can use the results
←− ←−
about limits in Ab to conclude that R p lim(Fn ) = 0 for p ≥ 2 and if (Fn ) satisfies (M-L)
←−
R1 lim(Fn ) = 0.
←−
Remark 2.2.5. If (Fn ) is a (M-L) system of locally constant sheaves and (Fn ) ,→ (In )•
is an injective resolution in Sh( X )N , then lim Fn ,→ lim In• is an injective resolution.
←− ←−
This follows easily from
H p (lim In• ) = R p lim(Fn ) = 0, if p ≥ 1.

←− ←−
2.3 Purity and localization
In this section we want to prove a purity result for (M-L) systems of local systems on
topological manifolds. This gives rise to localization sequences on cohomology, which
will finally be used to derive a very easy localization sequence for the Logarithm sheaf
and to define the polylogarithm cohomology classes.
2.3.1 Purity for local systems
Let k be a noetherian ring and X a topological space.
Definition 9. A sheaf F on X is called a local k-system, or simply a local system, if it is

locally constant and if for all x ∈ X the stalks F x are projective k-modules of finite rank.
The sheaf F ∗ := Homk (F , k) is also a local k-system and is called the dual of F .
Lemma 2.3.1. If I is an injective k-sheaf on a topological space X, then F ⊗k I is

injective for any local system F .
Proof. Given any inclusion A ,→ B of k-sheaves we have a commutative diagram where

the vertical arrows are isomorphisms:
Hom( B, I ⊗k F ) / Hom( A, I ⊗k F )

Hom( B ⊗k F ∗ , I) / Hom( A ⊗k F ∗ , I)
22
To see this one uses the universal property of the tensorproduct, (F ∗ )∗ = F and
F ∗ ⊗k I = Homk (F , I). Since F ∗ is a flat k-sheaf A ⊗k F ∗ ,→ B ⊗k F ∗ is injective.
Therefore the lower horizontal arrow is surjective, because I is injective. So F ⊗k I is
injective.
Let Z ⊂ X be a closed subspace. The pushforward i∗ : Sh( Z, k ) → Sh( X, k ) on the

categories of sheaves of k-modules has a right adjoint i! : Sh( X, k ) → Sh( Z, k ) (Iversen
(1986) II. Proposition 6.6.).
Lemma 2.3.2. Let F be a local k-system on a topological space X and let G be any
sheaf on X. There is a functorial isomorphism
i! (F ⊗k G) = i−1 F ⊗k i! G .
Proof. We use the Yoneda-principle: Let H be any k-sheaf on Z. By adjunction and the
properties of ∗ and ⊗k we have isomorphisms
= Hom(H ⊗k i−1 F ∗ , i! G) ∼
Hom(H, i! G ⊗k i−1 F ) ∼ = Hom(i∗ (H ⊗k i−1 F ∗ ), G)
Moreover,
i∗ (H ⊗k i−1 F ∗ ) ∼
= i ∗ H ⊗ k i ∗ i −1 F ∗ ∼
= i∗ H ⊗k F ∗
where the last arrow comes from the adjunction morphism F → i∗ i−1 F ∗ . Therefore we
get natural in H
Hom(H, i! G ⊗k i−1 F ) ∼
= Hom(i∗ H ⊗k F ∗ , G) ∼
= Hom(i∗ H, G ⊗k F ) ∼
=
Hom(H, i! (G ⊗k F ))
As i∗ is exact, i! is left exact. Purity is concerned with the calculation of the higher right
derived functors of i! . To do so we need some notation.
Let X be a topological manifold of dimension d and k a noetherian ring.
• D + ( X, k ) denotes the derived category of bounded below complexes of sheaves

of k-modules. We identify this category with the homotopy category of bounded
below complexes of injective sheaves of k-modules.
23
• D + (k ) is the derived category of bounded below compexes of k-modules. We

identify this category with the homotopy category of bounded below complexes of
injective k-modules.
Definition 10. Let X be a topological manifold of dimension d and k a noetherian ring.

The presheaf
U ⊂ X open 7→ Homk ( Hcd (U, k ), k ) =: or X (U )
is a sheaf on X. It is even a local k-system of rank 1 (Iversen (1986) III 8.14.) and we
call it the orientation bundle of X
Definition 11. Let X be a topological manifold of dimension d. The dualizing complex

DX on X is a complex of injective sheaves on X characterized up to homotopy by
the following property (Iversen (1986)VI.2): There is an isomorphism natural in A ∈
D + ( X, k):
Hom D+ (X,k) ( A, DX ) = Hom D+ (k) (Γc ( X, A), k )
Moreover, one has a quasi-isomorphism or X [d] → DX (Iversen (1986) VI.3.2).
Lemma 2.3.3. Let X be a topological manifold of dimension d and Z ⊂ X a closed

regular submanifold of codimension c. Then one has i! DX = DZ .
Proof. Iversen (1986)VIII proposition 1.7.
Definition 12. Let X be a topological manifold of dimension d and Z ⊂ X a closed

regular submanifold of codimension c. We define the orientation of Z relative X as the
local system of rank 1 or Z/X := or Z ⊗ i−1 or ∗X .
Proposition 2.3.4. Let X be a topological manifold of dimension d, Z ⊂ X a closed

regular submanifold of codimension c and F a local system on X. Then we have in
D ( Z, k )
Ri! F = or Z/X ⊗k i−1 F [−c]
functorial in F .
Proof. We start with the quasi-isomorphism or X [d] → DX . or ∗X ⊗k F is flat and if we

tensor the last quasi-isomorphism with it, we get a quasi-isomorphism
F [d] → DX ⊗k or ∗X ⊗k F .
24
The right hand side is a complex of injectives by Lemma 2.3.1. We get
Ri! F [d] = i! ( DX ⊗k or ∗X ⊗k F ) = i! DX ⊗k i−1 or ∗X ⊗k i−1 F
by Lemma 2.3.2 and this equals by Lemma 2.3.3
DZ ⊗k i−1 or ∗X ⊗k i−1 F = or Z [d − c] ⊗k i−1 or ∗X ⊗k i−1 F = or Z/X ⊗k i−1 F [d − c]
2.3.2 Purity for systems of local systems and naturality
Proposition 2.3.5. Let X be a topological manifold of dimension d and Z ⊂ X a closed

regular submanifold of codimension c. Let (Fn ) be a system of local k-systems satisfy-
ing (M-L) on X. We have in D + ( X, k )
Ri! (lim Fn ) = lim(i−1 Fn ⊗k or Z/X )[−c] = i−1 (lim Fn ) ⊗k or Z/X [−c]

←− ←− ←−
functorial in (Fn ).
Proof. Take an injective resolution a : (Fn ) ,→ (In )• in Sh( X, k)N . We already had
the resolution b : (Fn ) ,→ ( DX ⊗k Fn )n in Sh( X, k )N functorial in systems (Fn ). Since
(In )• is a complex of injectives in Sh( X, k)N , we get a map of complexes c unique up
to homotopy making the following diagram commute
(Fn )
a / (In )•
8
b c
%
( DX ⊗k Fn )
(Iversen (1986) I Theorem 6.2). Let us consider the functor i!N : Sh( X, k ) → Sh( Z, k ),
(Gn ) 7→ (i! Gn ). We have now
i!N c p∼
=
Ri!N (Fn ) = (i! In )• ← (i! DX ⊗k Fn ) → (i−1 Fn ⊗k or Z/X )[−c],
where the map on the right-hand side is the functorial purity quasi-isomorphisms p from
proposition 2.3.4. i!N c is also a quasi-isomorphism, as
H p (i! In )• = ( H p i! In• ) = ( R p i! Fn ) = H p (i! DX ⊗k Fn ).
25
This gives a purity isomorphism
Ri!N (Fn ) = (i−1 Fn ⊗k or Z/X )[−c] ∈ D + (Sh( X, k )N )
in the derived category of bounded below complexes of systems of sheaves of k-

modules, which is functorial in systems of local systems (Fn ). As (Fn ) is (M-L), we
have that lim Fn ,→ lim In• is also an injective resolution by Remark 2.2.5. In other
←− ←−
words, R lim(Fn ) = lim Fn ∈ D + ( X, k ) and one gets in D + ( X, k )
←− ←−
Ri! lim Fn = Ri! R lim(Fn ) = R(i! ◦ lim)(Fn ) = R(lim ◦i!N )(Fn ),

←− ←− ←− ←−
since i! commutes as right-adjoint with the limit. We go on with
R lim Ri!N (Fn ) = R lim(i−1 Fn ⊗k or Z/X )[−c] = lim(i−1 Fn ⊗k or Z/X )[−c],

←− ←− ←−
where we used purity and the fact that (i−1 Fn ⊗k or Z/X ) is again (M-L). Finally, we have
lim(i−1 Fn ⊗k or Z/X ) = i−1 (lim Fn ) ⊗k or Z/X ,

←− ←−
which will follow from the following two lemmas. This will complete the proof.
Lemma 2.3.6. Let X be a topological manifold and let (Fn ) be a system of locally
constant sheaves on X. For any x ∈ X the canonical map
(lim Fn ) x → lim( Fn,x )

←− ←−
is an isomorphism. More generally, if Y is another topological manifold and f : Y → X

a continuous map, the canonical map
f −1 (lim Fn ) → lim f −1 Fn
←− ←−
is an isomorphism.
Proof. Let G be any sheaf on X. As X is a topological manifold any open neighborhood

U of X contains a contractible open neighborhood B of X. It follows that the family Bx
of contractible open neighborhoods of x is a cofinal inductive system in the family Ux of
all open neighborhoods of x. Therefore
G x = lim G(U ) = lim G( B).

−→ −→
U ∈U x B ∈B x
26
If F is locally constant, F| B is constant for B ∈ Bx , as contractible spaces are simply

connected, and we have F x = limB∈B F ( B) = F ( B) for any B ∈ Bx . We calculate
−→ x
(lim Fn ) x = lim (lim Fn )( B) = lim (lim Fn ( B))

←− −→ ←− −→ ←−
B ∈B x B ∈B x
as the limit commutes with the section functor. The inductive limit becomes constant
and therefore
lim (lim Fn ( B)) = lim Fn ( B) = lim Fn,x .

−→ ←− ←− ←−
B ∈B x
This proves the first claim. The second follows easily from the first by considering the
induced maps on the stalks.
Lemma 2.3.7. Let (Fn ) be a system of locally constant k-sheaves on a topological

manifold X and F a local k-system on X. Then the natural map
(lim Fn ) ⊗k F → lim(Fn ⊗k F )
←− ←−
is an isomorphism.
Proof. By the preceeding lemma the problem is local so it suffices to proof the corre-
sponding statement for a system of k-modules ( Fn ) and a finitely generated projective
module F. For any k-module M one has the functorial isomorphisms
Hom( M, lim Fn ⊗k F ) = lim Hom( M, Fn ⊗k F ) = lim Hom( M ⊗k F ∗ , Fn ) =

←− ←− ←−
Hom( M ⊗k F ∗ , lim Fn ) = Hom( M, (lim Fn ) ⊗k F )

←− ←−
from which the claim follows by the Yoneda-lemma.
Now we come to our important application of the purity result
Proposition 2.3.8. Let (Fn ) be a system of local systems on a topological manifold

X satisfying (M-L). Let i : Z ,→ X be the inclusion of a closed regular submanifold of
codimension c and j : U := X \ Z → X the inclusion of the open complement. One has
an exact triangle in D + ( X, k )
+1
i∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → lim Fn → Rj∗ j−1 lim Fn →
←− ←− ←−
27
Proof. Let F be any sheaf on X. Let us recall
i∗ i! F (V ) = {s ∈ F (V ) : supp(s) ⊂ Z }
(Iversen (1986) II.6.6). Therefore, we have an exact sequence
0 → i∗ i! F → F → j∗ j−1 F ,
where the arrows are the adjunction morphisms. If F is flabby, the last sequence is
even right exact. This means, if we replace F by an injective resolution I , we get an
exact sequence
0 → i∗ i! I → I → j∗ j−1 I → 0
giving rise to an exact triangle in the derived category D + ( X, k)
+1
i∗ Ri! F → F → Rj∗ j−1 F → .
If we apply now Proposition 2.3.5 to substitute i∗ Ri! F , we get the exact triangle as
claimed.
Remark 2.3.9. Proposition 2.3.8 is functorial with respect to pairs ( X, Z ), ( X 0 , Z 0 ). By

this we mean given a cartesian square of topological spaces
i0 / X0
Z0
f0 f

Z
i /X
with a topological manifold X, a closed regular submanifold Z of codimension c and

topological submersion f (see Kashiwara & Schapira (1990) Definition 3.3.1) we get by
pullback a morphism of exact triangles (apply id → R f ∗ f −1 ) from
+1
←− ←− ←−
to
+1
R f ∗0 i∗0 (i0−1 (lim f −1 Fn ) ⊗k or Z0 /X 0 )[−c] → R f ∗ lim f −1 Fn → R f ∗ Rj∗0 j0−1 lim f −1 Fn →
←− ←− ←−
To see this we recall that X 0 and Z 0 have the structure of topological manifolds and
that Z 0 has codimension c in X 0 . After exploiting the commutation rules of the functors
involved the only thing that remains to be shown is that there is a natural pullback
28
isomorphism f 0−1 or Z/X → or Z0 /X 0 . As our diagram is cartesian, we have a natural

pullback morphism i0−1 f ! k → f 0! i−1 k by Kashiwara & Schapira (1990) proposition 3.1.9
0
(iii). If we apply H dim(X )−dim(X ) , we get the natural morphsim i0−1 or X 0 /X → or Z0 /Z .
Following the proof of Kashiwara & Schapira (1990) proposition 3.3.2 we may see that it
is actually an isomorphism. By the formalism provided by Kashiwara & Schapira (1990)
remark 3.3.5 we may conclude that we also have a natural isomorphism f 0−1 or Z/X →
or Z0 /X 0 .
Remark 2.3.10. If ( X, Z ), ( X, Z 0 ) are two pairs of codimension c as above with Z 0 ⊂ Z,

we have another compatibility. Z 0 ⊂ Z is also open, as both manifolds have the same
dimension. From this we derive a natural extension by zero morphism
i∗0 (i0−1 lim Fn ⊗k or Z0 /X ) → i∗ (i−1 lim Fn ⊗k or Z/X ),

←− ←−
with the inclusions i : Z → X and i0 : Z 0 → X. If we denote by j : X \ Z → X and

j0 : X \ Z 0 → X the inclusions of the open complements, the morphism above fits into a
morphism of triangles
i∗0 (i0−1 (lim Fn ) ⊗k or Z0 /X )[−c] / lim Fn / Rj0 j0−1 lim Fn +/ 1

←− ←− ∗ ←−

i∗ (i−1 (lim F n ) ⊗k or Z/X )[−c] / lim Fn / Rj∗ j−1 lim Fn +/ 1
←− ←− ←−
The right vertical arrow is just the restriction map.
Corollary 2.3.11. Let (Fn ) be a system of local systems on a topological manifold X

satisfying (M-L). Let i : Z ,→ X be the inclusion of a closed regular submanifold of
codimension c with open complement U := X \ Z and f : X → Y a continuous map of
topological spaces. There is a long exact cohomology sequence
→ R p−1 f |U ∗ (lim Fn |U ) → R p−c f |Z∗ (lim i−1 Fn ) → R p f ∗ (lim Fn )

←− ←− ←−
→ R p f |U ∗ (lim Fn |U ) → R p+1−c f |Z∗ (lim i−1 Fn ) →

←− ←−
called the localization sequence for the pair ( X, Z ). The sequence is natural in pairs
( X, Z ) as described in Remark 2.3.9 and Remark 2.3.10.
Proof. We take the exact triangle
+1
←− ←− ←−
29
from Proposition 2.3.8 and apply R f ∗ to obtain an exact trianlge

+1
R f ∗ i∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → R f ∗ lim Fn → R f ∗ Rj∗ j−1 lim Fn →,
←− ←− ←−
which equals
+1
R f |Z∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → R f ∗ lim Fn → R f |U ∗ j−1 lim Fn →
←− ←− ←−
by composition rules of derived functors. If we apply the functor H 0 , we obtain the
statement.
2.4 Construction of the topological polylogarithm
2.4.1 The localization sequence for the logarithm sheaf
In this section we prove the important vanishing result for the cohomology of the Log-
arithm sheaf. We will then derive a simple localization sequence for Logarithm sheaf,
which enables us to define the polylogarithm cohomology classes.
From now on we will often drop the subscript k at ⊗, when the coefficient ring has been
fixed.
Proposition 2.4.1. Let π : T → S be torus of constant fiber dimension d and LogT /S =

Log the associated Logarithm sheaf over the noetherian ring k. Then
R p π∗ ( Log) = 0, p 6= d, and aug : Rd π∗ ( Log) → Rd π∗ (k)
is an isomorphism.
Proof. As the problems are local on the base, we may check the corresponding state-
ments for the stalks. Because π is proper, we have for any sheaf F on T and s ∈ S
R p π∗ (F )s = H p (π −1 (s), F|π −1 (s) ) = R p π∗ (F )s .
Now LogT /S ,s = Logπ −1 (s)/{s} by Lemma 2.1.11 and we use Beilinson et al. (2014)
Theorem 1.6.1 to conclude.
Remark 2.4.2. If π : T → S is a topological submersion, there is a relative orientation

bundle orT /S on T , see Kashiwara & Schapira (1990) Definition 3.3.3. If π : T → S is a
torus, we have a fiber bundle with orientable fibers and therefore orT /S = π −1 Rd π∗ (k )∗ .
30
Now let D ⊂ T closed such that π| D : D → S is a (finite) covering. We set U := T \ D.

One has or D = π|−D1 orS .
Proposition 2.4.3. Let F be a local k-system on S , π : T → S a family of topological

tori of constant fiberdimension d and D ⊂ T closed such that π| D is a covering. The
localization sequence for the twisted Logarithm sheaf π −1 F ⊗ Log reduces to a short
exact sequence
aug
0 → Rd−1 π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) → π| D∗ (π −1 F ⊗ Log| D ) → F → 0
Proof. Because π| D : D → S is a covering, one has for any locally constant sheaf
G on D R p π| D∗ (G) = 0, p 6= 0. To see this it suffices to show H p (π|−D1∗ (U ), G) = 0,
p 6= 0, for all small U ⊂ S . Take U contractible and uniformly covered, in other words
π|−D1∗ (U ) = äi∈I Ui with π| D|Ui : Ui → U a homeomorphism for all i ∈ I . G|Ui is
constant, as locally constant on something simply connected, and we calculate
H p (π|−D1∗ (U ), G) = H p (ä Ui , G) = ∏ H p (Ui , G) = 0, p 6= 0,
i ∈I i ∈I
by homotopy-invariance of sheaf cohomology with constant coefficients (Iversen (1986)

IV.1).
The localization sequence for (π −1 F ⊗ Logn )n
→ R p−1 π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) → R p−d π| D∗ (π|−D1 F ⊗ Log| D )
→ R p π∗ (π −1 F ⊗ Log ⊗ orT /S ) → R p π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) →
provided by Corollary 2.3.11 gives immediately isomorphisms
R p π∗ (π −1 F ⊗ Log ⊗ orT /S ) → R p π|U ∗ (π −1 F ⊗ Log ⊗ orT /S )
for p < d − 1 and p > d. We extract the exact sequence
Rd−1 π∗ (π −1 F ⊗ Log ⊗ orT /S ) → Rd−1 π|U ∗ (π −1 F ⊗ Log ⊗ orT /S )
→ π| D∗ (π −1 F ⊗ Log| D ) → Rd π∗ (π −1 F ⊗ Log ⊗ orT /S )
from our localization sequence. We have
R p π∗ (π −1 F ⊗ Log ⊗ orT /S ) = F ⊗ R p π∗ ( Log) ⊗ Rd π∗ (k )∗ = 0
31
for p 6= d and
Rd π∗ (π −1 F ⊗ Log ⊗ orT /S ) = F
by the projection formula, since π is proper and F is flat. So we have the short exact
sequence
0 → Rd−1 π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) → π| D∗ (π −1 F ⊗ Log| D ) → F
But the last arrow is obtained by applying F ⊗ and the projection formula to the epimor-
phism
aug tr
π| D∗ ( LogT /S| D ) → π| D∗ k → k,
therefore is itself an epimorphism (for tr see Iversen (1986) VII.4.). In particular
R p π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) = 0, p 6= d − 1.
Remark 2.4.4. Let us recall the augmentation morphism
aug tr
π| D∗ ( Log| D ) → π| D∗ k → k.
It is enough to do so on the fibers π −1 (s). So we have a topological torus T and D ⊂ T

is just a union of finitely many points. We have to examine the chain of maps
purity∼
= aug
H 0 ( D, LogT | D ) → HZd ( T, LogT ⊗ or T ) → H d ( T, LogT ⊗ or T ) →
∼
= ev
→ H d ( T, orT ) → H d ( T, k) ⊗ orT ( T ) → k
We have HZd ( T, Log ⊗ or T ) = H{dx} ( T, Log ⊗ or T ). We have isomorphisms

L
x∈D
purity
H 0 ({ x } , k ) → H{dx} ( T, k ) → H d ( T, k ),
as T is orientable. We choose a generator 1( x ) ∈ H 0 ({ x } , k ) and denote the corre-

sponding images with 1{ x} ∈ H{dx} ( T, k ) and ω ( x ) ∈ H d ( T, k ). We set ω ( x )∗ ∈ or T ( T )
the element dual to ω ( x ). Given
∑ f (x)1(x) ∈ H0 ( D, Log|D ), f ( x ) ∈ Logx ,

x
32
purity maps it to
∑ f (x)1{x} ⊗ ω (x)∗ ∈ HZd (T, Log ⊗ orT ).

x
This is mapped by aug to
∑ aug( f (x))ω (x) ⊗ ω (x)∗ ∈ H d (T, orT )

x
and then by ev to ∑ x aug( f ( x )) as claimed.
Proposition 2.4.5. With notations from above we have an exact sequence

res aug
0 → H d−1 (U, π −1 F ⊗ Log ⊗ orT /S ) → H 0 ( D, π −1 F ⊗ Log| D ) → H 0 (S , F )
Proof. Using the Grothendieck-Leray spectral sequence for the composition of functors
Γ(U, −) = Γ(S , −) ◦ π|U ∗ we easily see
H p (U, π −1 F ⊗ Log ⊗ orT /S ) = H p−d (S , Rd−1 π|U ∗ (π −1 F ⊗ Log ⊗ orT /S )),
as
R p π|U ∗ (π −1 F ⊗ Log ⊗ orT /S ) = 0, p 6= d − 1.
Applying H 0 (S , −) to Proposition 2.4.3 yields the result.
Definition 13. We keep the notations from above.

Given f ∈ H 0 ( D, π|−D1 F ⊗ Log| D ) with aug( f ) = 0, we define the polylogarithm associ-
ated to f as the unique cohomology class
pol( f ) ∈ H d−1 (U, π −1 F ⊗ Log ⊗ orT /S )
with res(pol( f )) = f . If 0(S) ∩ D = ∅, we define the polylogarithmic Eisenstein class

associated to f to be
Eis( f ) := 0−1 pol( f ) ∈ H d−1 (S , F ⊗ 0−1 ( Log ⊗ orT /S ))
2.4.2 Trivializations of the logarithm sheaf
Finding non trivial polylogarithm classes is by definition equivalent to finding non trivial
f ∈ H 0 ( D, π −1 F ⊗ Log| D ) with aug( f ) = 0. For which D do we have sections or a
trivialization for Log| D ?
We just write Logn = LogTn /S , when the topological situation π : T → S has been fixed.
33
Lemma 2.4.6. The augmentation map 0−1 Log → k splits. The image of 1 ∈ k is
denoted by 10 ∈ H 0 (S , 0−1 Log). 0−1 Log =: R is a sheaf of augmented algebras and
Log is an invertible π −1 R-module.
Proof. 0−1 Log is the local system associated to the π1 (S , s0 )-module

k [[π1 (Ts0 , t0 )]]. The augmentation map corresponds to the augmentation map of this
algebra. It has the obvious π1 (S , s0 )-equivariant section providing the section of the cor-
responding sheaves. Obviously π1 (S , s0 ) acts by algebra homomorphisms on k [[π1 (Ts0 , t0 )]],
therefore R is a sheaf of algebras. π −1 R is associated to the π1 (t0 , T )-module k [[π1 (Ts0 , t0 )]]
with the action
(v, γ) ∑ f (l )(l ) = ∑ f (l )(γl )

l l
for (v, γ) ∈ π1 (t0 , T ). Let us denote this module by k [[π1 (Ts0 , t0 )]]R and the module
associated to Log by k [[π1 (Ts0 , t0 )]] Log to emphasize their different π1 (t0 , T )-structures.
Then we see that product map
k [[π1 (Ts0 , t0 )]]R ⊗k k [[π1 (Ts0 , t0 )]] Log → k [[π1 (Ts0 , t0 )]] Log , f ⊗ g 7→ f · g
is π1 (t0 , T )-equivariant proving that Log is an π −1 R-module.
We set Rn := 0−1 Logn .
Lemma 2.4.7. Let φ : T 0 → T be an isogeny of tori over S . We consider the Logarithm

sheaves over a ring k and suppose that deg(φ)(s) ∈ k× for all s ∈ S , where k× always
denotes the units in a ring k.
Then the natural morphism LogT 0 /S → φ−1 LogT /S is an isomorphism.
Proof. It suffices to consider the map of underlying π1 (t00 , T 0 )- modules. It is given

φ∗
by the natural map k [[π1 (t00 , Ts00 )]] → k [[π1 (t0 , Ts0 ]]. This map is an isomorphism by
Beilinson et al. (2014) proposition 1.1.5.
Definition 14. Let X be a topological space, k → k0 a ring homomorphism and G ∈

Sh( X ) we set Gk0 := G ⊗k k0 .
Given a family of topological tori π : T → S with Logarithm sheaf Log we may consider
Log× := aug−1 (1), where aug : Log → k is the augmentation map. Log× is a sheaf
of sets on T . We may realize this sheaf as a sheaf of sections of a topological space
34
Log∗sp over S . To do so we set R := k [[π1 (t0 , Ts0 )]] and aug−1 (0) =: a the augmentation
ideal. In R we have the group of 1-units 1 + a =: R1× . The action of π1 (t0 , T ) on R
restricts to R1× making it a π1 (t0 , T )-module. Considering R1× with the discrete topology
we have
Log×sp := T̃ ×π1 (Ts ,t ) R1× .

0 0
We had T̃ = S̃ × H1 (Ts0 , R) and therefore we may write
Log×sp = S̃ ×π1 (S ,s0 ) ( H1 (Ts0 , R) ×π1 (Ts ,t ) R1× ).

0 0
The second description tells us that Log×sp may even be considered as a group object
over S with typical fiber H1 (Ts0 , R) ×π1 (Ts ,t ) R1× . We also have the sheaf R1× of 1-units
0 0
of the augmented algebra R. One has the inclusion δ : π1 (t0 , Ts0 ) ⊂ R1× giving rise to
the inclusion δ : H → R1× . If we consider the fibers of T with the discrete topology, we
may summarize this in the following pushout diagram of abelian sheaves on S
p
0 / R× / Log×sp /T /0
O1 O O
δ id
0 / HZ / HR /T /0
where we consider T and Log×sp as sheaves of sections. Let H ⊂ T be a subgroup.

By this we mean that H is a sub-local-system of T . We also may consider H as a
subspace of T covering S . Now we follow Beilinson et al. (2014) Definition 1.5.1.
Definition 15. A multiplicative trivialization of Log along H is a morphism t : H →

Log×sp , which is a section for p : Log×sp → T .
Remark 2.4.8. A multiplicative trivialization of Log gives a section 1 H ∈ Γ( H, Log| H ),

aug(1 H ) = 1, trivializing Log| H as an invertible π|−H1 R-module.
Let T (k) be the subsheaf of T consisting of all torsion sections whose order is invertible
in k and T tors the subsheaf of all torsion sections.
Proposition 2.4.9. There is a unique multiplicative trivialization ρcan : T (k) → Log×sp

for Log along T (k) . It is compatible with isogenies and for t ∈ T [ N ](S) ⊂ T k (S) it is
explicitly given by the isomorphism
t−1 Log ∼
= t−1 [ N ]−1 Log ∼
= ([ N ] ◦ t)−1 Log ∼
= 0−1 Log,
where one uses Log ∼

= [ N ]−1 Log, as deg[ N ] ∈ k× , and [ N ] ◦ t = Nt = 0, as t is
N-torsion.
35
Proof. By Beilinson et al. (2014) proposition 1.5.3. we have the statement, if S =

{ pt.}. But as Log×sp and T are fiber bundles over S , we may easily glue these local
multiplicative trivializations to a global one by using unicity.
This proposition tells us that torsion sections are a powerful tool to construct D ⊂ T for
which we can write down sections of Log| D . But before we can start calculations with
Log, let us have a closer look at the situation over k = C.
Definition 16. Let X be a topological space and R a sheaf of rings. For a projective
system (Fn ), F := lim Fn , of R-module sheaves and an R-module sheaf G on X we
←−
set
ˆ R G := lim(Fn ⊗R G).
F⊗
←−
From now on we consider the Logarithm sheaf over k = C. Let CT∞ be the sheaf of
complex valued smooth functions on T and A1T the sheaf of complex valued smooth
1-forms on T . We set
Log∞n := Logn ⊗ CT∞ and Log∞ := lim Log∞n = Log⊗

ˆ C CT∞
←−
to be the pro vector bundle associated to Log. As it comes from local systems, it is
naturally endowed with an integrable connection
id ⊗ d : Log∞ → Log∞ ⊗
ˆ C ∞ A1T ,
T
whose kernel is exactly Log. Sections of a pro vector bundles are nothing else as
compatible systems of sections.
Proposition 2.4.10. Given a lattice L one has the isomorphism of augmented algebras
l ⊗k
Q[[ L]] → ∏ Symk LQ , (l ) 7→ exp(l ) := ∑ k!
,
k ≥0 k ≥0
where the right hand side carries the Cauchy-product and the augmentation is the pro-
exp
jection onto the Sym0 LQ = Q- part. In particular, one has an isomorphism R →
exp
∏k≥0 Symk HQ of augmented Q-algebras and an isomorphism of systems (Rn ) →
∏nk=0 Symk HQ n

Proof. Beilinson et al. (2014) Corollary 1.1.10.
36
Remark 2.4.11. This isomorphism does not work integrally. Nevertheless, let k be a
torsion free ring. We have
l ⊗n
k [[ L]] → ∏ Symn LPD , (l ) 7→ exp(l ) := ∑ ,
n ≥0 n≥0 n!
into a subring of ∏k≥0 Symk LkQ . Here we define the k-submodule Symn L PD ⊂ Symk LkQ
to be generated by
[i ] [i ]
x1 1 · .. · xn n , x1 , ..., xn ∈ L,
where we define in any Q-algebra the divided powers x [n] := x

n! , x ∈ A. Symk L PD gives
rise to a local system Symk H PD , which is an integral structure for Symk HkQ . We have
the projection map pn : R → Symn H PD which is induced by
exp prn
k [[ L]] → ∏ Symn LPD → Symn LPD .
n ≥0
Carrying the action of the fundamental group from the left to the right we may under-
stand the Logarithm sheaf via the completed symmetric algebra. Set
Logsp := T̃ ×π1 (t0 ,T ) ∏ Symn H1 (π −1 (s0 ), C).

n ≥0
The sections of Log∞ may now be seen as the smooth sections s : T → Logsp . These
are π1 (t0 , T )-equivariant maps f : T̃ → ∏n≥0 Symn H1 (π −1 (s0 ), C) such that for all
n ≥ 0 the component maps f n : T̃ → Symn H1 (π −1 (s0 ), C) are smooth. In this sense it
is also clear, what we mean by smooth or continuous sections of Log×sp .
Proposition 2.4.12. There is a unique continuous multiplicative trivialization of Log. By

this we mean a continuous homomorphism ρcont : T → Log×sp , which is a section for
p : Log×sp → T . It is compatible with [ N ]-multiplication and we have ρcont|T tors = ρcan .
ρcont is explicitly given by
(−v)⊗n
exp(−v) = ∑ n!
n ≥0
with
v : T̃ = S̃ × H1 (π −1 (s0 ), R), (s, v) 7→ v ∈ ∏ Symn H1 (π −1 (s0 ), C)

n ≥0
considered as a C ∞ -section of Log on the universal cover of T .
37
Proof. π1 (t0 , T ) acts on ∏n≥0 Symn H1 (π −1 (s0 ), C) by
(l, γ) ∑ f n = exp(l ) · ∑ γ fn.

n ≥0 n ≥0
From the equivariance of exp
exp(−(v(l, γ)) = exp(−(γ−1 (v + l ))) = exp(−γ−1 l ) · γ exp(−v) = (l, γ)−1 exp(−v)
it is clear that the section above descends to a multiplicative trivialization as claimed.

The problem of unicity is local on the base S and as all spaces are fiber bundles over
S , we may reduce to the case where S is a point. This is Beilinson et al. (2014) Lemma
2.1.2.
We set
R∞n := π −1 Rn ⊗ CT∞ and R∞ = lim R∞n = R⊗C

ˆ T∞
←−
the pro vector bundle associated to π −1 R. Log∞ is now a module over this sheaf of
algebras. We also have the sheaf of differential graded algebras
ˆ T = R∞ ⊗
R⊗A ˆ C ∞ AT
T
∗ ⊂ A1 interpreting sections
with the trivial differential id ⊗ d. We have HZ ⊂ R and HC T
∗ as invariant differential forms along the fibers of π. The canonical isomorphism
of HC
End(HZ ) ∼ ∗
= HZ ⊗ HZ , id 7→ κ
ˆ T . κ is closed and multiplication with κ gives an

gives the canonical element κ ∈ R⊗A
operator on our differential graded algebra.
Corollary 2.4.13. One has the isomorphism of pro vectorbundles
R∞ → Log∞ , f 7→ f · ρcont .
If endow R∞ with the integrable connection ∇ := d − κ, this isomorphism commutes

with the connections.
Proof. As Log∞ is an invertible R∞ -module the isomorphism is clear, as aug(ρcont ) = 1

and ρcont is a global nowhere vanishing section section of Log∞ . To prove the compat-
ibility of the connections we need to calculate id ⊗ d(ρcont ). This can be done on the
universal cover T̃ . We have
(−v)⊗n id ⊗ d(−v)⊗n
id ⊗ d(exp(−v)) = id ⊗ d ∑ n!
= ∑
n!
=
n ≥0 n ≥0
38
n(−v)⊗n−1 · (id ⊗ d(−v)) (−v)⊗n

∑ n!
= ∑
n!
id ⊗ d(−v)
n ≥0 n ≥0
Write v = ∑id=1 ei ⊗ xi , for some R-basis (ei ) of H1 (Ts0 , R), where ( xi ) is the corre-
sponding dual basis of (ei ) considered as coordinate functions on H1 (Ts0 , R). Then
d d
id ⊗ d(v) = id ⊗ d ∑ ei ⊗ xi = ∑ ei ⊗ dxi = κ
i =1 i =1
We conclude id ⊗ d(ρcont ) = ρcont · (−κ ). This completes the proof.
Now we come to our case of interest.
Proposition 2.4.14. Let π : T → S be family of topological tori with level N-structure

and constant fiber dimension d. We consider
D = T [ N ] \ 0(S) ∼
= (Z/NZ)d \ {0} × S
and the Logarithmsheaf Log over a noetherian ring k, with N ∈ k× . We have then a
canonical trivialization
H 0 ( D, Log| D ) = H 0 ((Z/NZ)d \ {0} × S , R).
We had k ⊂ R and thus any locally constant function
f : (Z/NZ)d \ {0} × S → k,
with
aug( f ) = ∑ f (v, ) = 0 ∈ H 0 (S , k)
v∈(Z/NZ)d \{0}
gives rise to a section f ∈ H 0 ( D, Log| D ) of augmentation zero and therefore to a unique

polylogarithmic cohomology class
pol( f ) ∈ H d−1 (U, Log ⊗ orT /S ).
Moreover, we get the polylogarithmic Eisenstein class
0−1 pol( f ) =: Eis( f ) ∈ H d−1 (S , R ⊗ orT /S ),
which induces polylogarithmic Eisenstein classes
Eisk ( f ) ∈ H d−1 (S , Symk H PD ⊗ orT /S ),
whenever k is torsion-free.
39
Proof. k ⊂ R is Lemma 2.4.6. The trivialization is Proposition 2.4.9. Finally, we just

need Remark 2.4.4 and Definition 13 to get pol( f ). Eisk ( f ) := pk (Eisk ( f )), where
pk : R → Symk H PD as in Remark 2.4.11.
Remark 2.4.15. The restriction to torsion-free rings in the proposition above is not nec-
essary. To get the appropriate substitute for our Symn H PD one has to use the universal
divided power algebra Γk ( L), see Beilinson et al. (2014) Lemma 1.1.7.
2.4.3 A differential equation for the polylogarithm
We want explicit cohmology classes representing representing our polylogarithms over

C. To do so we need an explicit resolution of Log over C, which allows us to calculate
the localization sequence in Proposition 2.4.5.
Lemma 2.4.16. Let X be a oriented n-dimensional C ∞ -manifold and denote by A X =

(A•X , d) the de Rham complex of C-valued C ∞ -differential forms. We may represent the
dualizing complex of DX by non-continuous functionals on the de Rham complex:
p −p
DX (U ) = Γc (U, A X )∗ , U ⊂ X open
with differential d defined by

−p −p
d( T )(ω ) := (−1) p+1 T (d(ω )), ω ∈ Γc (U, A X )∗ , T ∈ Γc (U, A X )∗
Proof. The Poincare Lemma gives a quasi-isomorphism C → A X . So A X is a bounded

resolution of the constant sheaf C by soft C-sheaves. We conclude with Iversen (1986)
V.2.1.
Remark 2.4.17. DX is a complex of injective C-sheaves, which also has the structure
of C X∞ -module. One has a canonical quasi-isomorphism
or X [n] = H −n ( DX )[n] → DX
(Iversen (1986)VI.3.), which we may make explicit. Choose a volume form on X. This
p
allows us to define integration of dim( X )-forms. Given a p-form ω ∈ Γ(U, A X ) and an
η ∈ Γc (U, An− p ) we have ω ∧ η ∈ Γc (U, An ). ω ∧ η can be considered as an element
in Γc ( X, An ) by extension by zero, which can be integrated. In other words, we have a
linear map
Z
p n− p
Γ(U, A X ) → Γc (U, A X )∗ , ω 7→ ω∧ .
X
40
This map is compatible with restrictions in U ⊂ X open and we have

Z Z Z
dω ∧ η = d(ω ∧ η ) + (−1) p+1 ω ∧ dη = (−1) p+1 ω ∧ (dη )
X X X
by Stokes theorem, as our manifold has no boundary. It follows that we get a homomor-
phism of complexes
Z
A X → DX [−n], ω 7→ η 7→ ω∧η .
X
We get a chain of quasi-ismomorphisms
C[0] ⊂ A X → DX [−n]
and this just defines a trivialization C → H −n ( DX ) = or X of the orientation bundle.

Now let i : Z ,→ X be a closed regular submanifold of codimension c.
We have the pullback morphism i∗ : A X → i∗ A Z . As i is proper we also get for any
U ⊂ X open i∗ : Γc (U, A X ) → Γc (U, i∗ A Z ) and by taking duals
i∗ : Γc (U, i∗ A Z )∗ → Γc (U, A X )∗
compatible with restrictions in U. So identifying DX with the sheaf of non-continuous

functionals on A X we get i∗ DZ → DX and by adjunction i∗ : DZ → i! DX . This is exactly
the arrow inducing Lemma 2.3.3. Because i : Z → X is regularily embedded, Z carries
an orientation induced by X. We have then
H 0 ( Z, C) = H 0 ( Z, or Z ) = H 0 ( Z, DZ [c − n]) = H 0 ( Z, i! DX [c − n]) =
HZ0 ( X, DX [c − n]) = HZc ( X, or X ) = HZc ( X, C)
Here we have exactly

Z
H ( Z, C) →
0
HZc ( X, C), f 7→ ,
f
f · i∗ ω and integration is defined by the induced orientation on Z.

R R
with f
ω := Z
Proposition 2.4.18. Let π : T → S be a family of topological tori and Log = lim Logn
←−
the associated Logarithm sheaf over C and DT the dualizing complex on T . The system
of complexes
( Logn ⊗ orT−1 ⊗ DT [−dim(T )])n
is an injective resolution of ( Logn ) in Sh(T , C)N . In particular,
lim Logn ⊗ orT−1 ⊗ DT [−dim(T )]

←−
is an injective resolution of Log.
41
p
Proof. We already know that DT is injective in Sh(T , C). As Logn ⊗ orT−1 is a local
p
system Logn ⊗ orT−1 ⊗ DT is injective in Sh(T , C). In order to see that
( Logn ⊗ orT−1 ⊗ DT )n ∈ Sh(T , C)N

p
p
is injective we need to see that all transition maps are split epimorphic. Now DT is a
CT∞ -module and therefore
( Logn ⊗ orT−1 ⊗ DT )n = ( Log∞n ⊗CT∞ orT−1 ⊗ DT ).

p p
We have an isomorphism of systems ( Log∞n )n = (R∞n )n and the latter system has split
epimorphic transition maps by Proposition 2.4.10. We conclude that ( Logn ⊗ orT−1 ⊗
p
DT )n has split epimorphic transition maps and therefore is injective in Sh(T , C)N .
( Logn ⊗ orT−1 ⊗ DT [−dim(T )])n being a resolution of ( Logn )n may be checked for each
n ∈ N0 separately by using the quasi isomrphism orT [dim(T )] → DT .
Proposition 2.4.19. Let S be an oriented manifold and π : T → S a family of topolog-

ical tori with level N-structure and constant fiber dimension d. We consider the Loga-
rithm sheaf Log with C-coefficients. The polylogarithms as defined in Proposition 2.4.14
may be represented by non-continuous functionals
Z
pol( f ) ∈ Γ(T , R∞ ⊗
ˆ C ∞ π −1 orS−1 ⊗ DT ) with ∇(pol( f )) = volS ⊗
T
f
Here the differential ∇ is induced by (R∞ , ∇) and integration on S is defined with

respect to a chosen volume form volS which we also use to trivialize orS−1 .
Proof. We set dim(T ) = n. To calculate cohmomology we can use the injective reso-
lutions
lim Logn ⊗ orT /S ⊗ orT−1 ⊗ DT [−n]

←−
with differential id ⊗ d. We write these complexes as
Log∞ ⊗
ˆ C ∞ orS−1 ⊗ DT [−n]
T
using the CT∞ -module structure of DT and the rules for orientations. We have a chain of
quasi-isomorphisms
ρcan id⊗i
π −1 R → Log| D → Log| D ⊗ or − 1
D ⊗ DD [−( n − d )] →
∗
42
∼
=
Log| D ⊗ or − 1 ! ! −1 −1
D ⊗ i DT [−( n − d )] → i ( Log ⊗ π orS ⊗ DT [−( n − d )]).
The last isomorphism is Lemma 2.3.2. If we identify
i! ( Log ⊗ π −1 orS−1 ⊗ DT [−(n − d)]) = i! ( Log∞ ⊗

ˆ C ∞ π −1 orS−1 ⊗ DT [−(n − d)]),
T
R
we may give a global section of this sheaf namely ρcont ⊗ volS ⊗ f
. If we go through
the previous morphisms remebering ρcan = ρcont|T tors , we see that f maps exactly to this
section. So we just have proved
Z
ρcan c purity
H 0 ( D, R) → H 0 ( D, Log| D ) → HD (T , Log ⊗ orT /S ), f 7→ ρcont ⊗ volS ⊗ .
f
Now we use the continuous trivialization ρcont of Log to identify our injective resolutions
from above with R∞ ⊗
ˆ C ∞ orS−1 ⊗ DT [−n]. We call the differential of the last complex ∇,
T
as it is induced by (R∞ , ∇). Clearly, f corresponds now to

Z
volS ⊗ ∈ i! (R∞ ⊗
ˆ C ∞ orS−1 ⊗ DT [−n]).
T
f
The residue map res of our localization sequence is nothing but a connecting homomor-
phism of a long exact cohomology sequence. So being in the image of res just means
that there is a
Z
∞ ˆ C ∞ orS−1
pol( f ) ∈ Γ(T , R ⊗ T
⊗ DT [−n]), ∇(polD ( f )) = volS ⊗ .
f
2.4.4 Invariant functionals and quotient manifolds
To represent polylogarithms with functionals it seems promising to pullback the situation

to the universial cover of T , because the Logarithm sheaf is constant there. We would
like to write down sections invariant under the π1 (T , t0 )-action descending to sections
representing polylogarithms. This would be exactly the right thing, if we knew that poly-
logarithms may be represented by differential forms. The problem is that we cannot
pullback functionals in a naive way, as the universial covering p : T̃ → T is usually
not proper (finite). But things still work the other way round, invariant functionals on T̃
descend to functionals on T . This is what we explain now.
Let X be a C ∞ -manifold and G a group acting on X by diffeomorphisms from the right.

So for g ∈ G we have the diffeomorphism r g : X → X, x 7→ xg. We assume this
43
action properly discontinuous and fixpoint free. The quotient space with the quotient
topology has a unique C ∞ -structure making the natural covering π : X → X/G a local
diffeomorphism. We are interested in currents on X/G. For a definition of currents see
de Rham (1955) chapitre III 8. Denote the currents on a manifold V by D(V ) and the
sheaf of currents on V by DV .
Proposition 2.4.20. Let X be a G-manifold as described above. There is an isomor-

phism
(π∗ D X )G → D X/G , T 7→ T̃
Proof. Let us first construct the map. Take U ⊂ X/G open and T ∈ D(π −1 (U ))G and
a ω ∈ Ac (U ). We have to define T̃ (ω ). To do so take an open cover
S
i Ui = X such
that there are Ui0 ⊂ X open with π|Ui0 : Ui0 → Ui is a diffeomorphism. Choose a partition
of unity ∑i ei = 1 subordinated Ui . Set
T̃ (ω ) := ∑ T (π|∗U0 ei ω ) = T (∑ π|∗U0 ei ω )
i i
i i
We have to make a few remarks:

T (π|∗U 0 ei ω ) is defined, because the support of ei ω is compact and contained in Ui ∩ U,
i
therefore the the support of π|∗U 0 ei ω is compact and contained in π −1 (U ). Of course,

i
the sums are again finite and T̃ is linear.

Consider a sequence of forms (ωn ) with supp(ωn ) ⊂ K ⊂ V, where K is compact,
V a coordinate patch and ωn → 0 uniformily together with all its derivatives. By the
Leibniz-rule ei ωn converges in the same sense as ω. We have that π|−U10 (V ) is again a
i
coordinate patch on X and, since π|Ui0 is a diffeomorphism, π|∗U 0 ei ωn is a sequence of

i
forms with supp(π|∗U 0 ei ωn ) ⊂ K 0 ⊂ V 0 , where K 0 is compact, V 0 a coordinate patch and

i
π|∗U 0 ei ωn → 0 uniformily together with all its derivatives. Since T is continous, we get
i
T̃ (ωn ) = ∑ T (π|∗U0 ei ωn ) → 0
i
i
as n tends to infinity. Therefore T̃ is continuous.

Up to now T̃ seems to depend on the partition of unity (Ui , ei ) and on the choice of the
Ui0 . We will show that this is not the case.
We start with the independence of choice of the Ui0 . Given Ui0 any other set mapping
via π diffeomorphic to Ui is of the form r g (Ui0 ). We get:
T (π|∗rg (U 0 ) ei ω ) = T ((π|Ui0 ◦ r − 1 ∗ −1∗ ∗ ∗

g ) ei ω ) = T ( r g π |U 0 ei ω ) = T ( π |U 0 ei ω ) ,
i i i
44
since T is G-invariant.
Next we show the independence of the partition of unity. We assume now that we
have two partitions of unity (Ui , ei ) and (Vj , η j ). By the first step we may assume that
whenever Ui ∩ Vj 6= ∅, we have Ui0 ∩ Vj0 6= ∅. We may consider (Ui ∩ Vj , Ui0 ∩ Vj0 , ei η j ).
It is a partition of unity as the others. We get
∑ T (π|∗U0 ei ω ) = ∑ T (π|∗U0 ei ∑ ηj ω ) = ∑ T (π|∗U0 ∩V 0 ei ηj ω ),

i i i j
i i j i,j
from which the independence of the partition follows. The construction is of course
compatible with restricions in U. This means that we have the desired morphism of
sheaves
(π∗ D X )G → D X/G , T 7→ T̃.
It remains to show that it is an isomorphism. This is a local problem. Choose a co-

S•
ordinate patch U ⊂ X/G such that π −1 (U ) = g ∈ G r g (U
0 ), with all r g (U 0 ) coordinate
patches. One has the isomorphism
(π∗ D X )G (U ) = ( ∏ D(r g (U 0 )))G → D(U 0 ), ( Tg ) 7→ Tid

g∈ G
with inverse T 7→ ( g−1 T ). Now π| U 0 : U 0 → U is a diffeomorphism and we see by

pushforward π 0 : D(U 0 ) =
|U ∗
∼ D(U ) that the composite map
D(U 0 ) ∼
= (π∗ DX )G (U ) → D X/G (U ), T 7→ π|U 0 ∗ T = T̃.
is an isomorphism. This completes the proof.
Remark 2.4.21. In the following let X be an oriented manifold and let G act by oriented
diffeomorphisms. Then X/G is also oriented and we choose a volume form volX/G on
X/G. On X we take the G-invariant volume form volX := π ∗ volX/G .
• The inclusion A X ,→ D X induced by integration is G-equivariant:

Z Z Z
gω [η ] = ω ∧ r− 1∗
g η = r− 1∗ ∗
g (r g ω ∧ η ) = r ∗g ω ∧ η
X X X
by the transformation-formula.
• Differentiation of currents is a G-equivariant morphsim.
45
• One has a commutative diagram
π∗ / (π∗ A X )G
A X/G

D X/G o (π∗ D X )G
T 7→ T̃
The vertical arrows are the inclusions of forms into currents by integration. This
follows from the very definition of integration on manifolds, the choice of the atlas
π|Ui0 : Ui0 → Ui - Ui0 an atlas of X - on X/G and the formula
Z Z Z
∗ ω [η ] =
πg ∑ X
π ∗ ω ∧ π|∗U 0 ei η =
i
∑ X
π|∗U 0 ei (ω ∧ η ) =
i X/G
ω ∧ η,
i i
with ω ∈ A( X/G ), η ∈ Ac ( X/G ), ∑i ei = 1 a partition of unity subordinated (Ui ).
(π∗ DX )G → DX/G , T 7→ T̃
is a morphsim of complexes with respect to the differentiation of currents. Let

ω be a form with compact support on X/G and T a invariant current on X. As
the problem is local, we may suppose that supp(ω ) ⊂ U, with π|U 0 : U 0 → U a
diffeomorphism. Then
f (ω ) = (dT )(π ∗ 0 ω ) = (−1)deg(T ) T (dπ ∗ 0 ω ) =

dT |U |U
(−1)deg(T ) T (π|∗U 0 dω ) = d T̃ (ω )
• The whole theory of also works for non-continuous functionals.
46
3 Polylogarithmic Eisenstein classes for
Hilbert-Blumenthal varieties
We will recall the Hilbert-Blumethal varieties SK also considered by Harder (1987).

Our goal is to give a direct Q-rational construction of Harder’s Eisenstein cohomology
classes. To do so we construct a fiber bundle MK over SK . Over MK there is a fam-
ily of topological tori and we can apply our construction of polylogarithmic Eisenstein
classes. In order to get classes on SK in different cohomological degrees we have to
decompose the cohomology of MK .
3.1 The topological situation
3.1.1 Notation
Let F be a totally real number field of degree [ F : Q] = g and O ⊂ F its ring of integers.
We denote places of F by ν. If ν is an archimedian place, we write ν|∞ and call these
places also the infinite places. Non-archimedian places correspond to maximal ideals
p ⊂ O and we call these places the finite places. If we need to emphasize that ν
corresponds to the maximal ideal p, we write νp instead of ν. We have p ∩ Z = ( p) for
a prime number p ∈ N and write ν| p in this case.
Infinite places may be identified with field embeddings σ : F → R. If A is any R-algebra,
we have an isomorphism
FA := F ⊗Q A = ∏ A= ∏ A, x ⊗ a 7→ (σ ( x ) a)σ .
σ:F →R ν|∞
In this manner we may speak of the ν-component of z ∈ F ⊗Q A.

The completion of F with respect to a place ν is denoted by Fν . If ν| p, Oν is the integral
closure of Z p inside Fν and pν = (πν ) ⊂ Oν is the maximal ideal with residue field
47
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
κ (pν ) := Oν /pν = O /p =: κ (p).

For any place ν we have a normalized absolute value:
If ν|∞, | x |ν is the usual absolute value of x ∈ Fν = R. If ν is finite, we have | x |ν :=
|κ (pν )|−ν(x) = |Oν /( x )|−1 provided x ∈ Oν \ {0}. ν is considered here as a discrete
ν( x )
valuation on Fν and may also be defined by x = uπν , u ∈ Oν× .
We have the ring of adeles AF and the group of Ideles IF over F. We set A = AQ and
IQ = I. We have a decomposition into a finite and an infinite part
AF = AF,∞ × AF, f , IF = IF,∞ × IF, f .
We define Ẑ := limn∈N Z/nZ to be the profinite completion of Z and recall AF, f =

←−
Ẑ ⊗Z F. We also set Ô := Ẑ ⊗Z O .
We consider G0 := GL2 and V0 := G2a as algebraic groups over spec(O), where the
first group acts naturally on the second by matrix multiplication from the left. For any
algebraic subgroup H0 ⊂ V0 o G0 we write H := ResO /Z H0 for the restriction of scalars.
We have for example B0 ⊂ G0 the standard Borel of upper triangular matrices, U0 ⊂ B0
its unipotent radical, T0 ⊂ B0 the maximal torus and Z0 ⊂ G0 the center. For any ring R
we write H ( R) for the group of R valued points of H.
We have H (A) ⊂ ∏ν H ( Fν ), so h = (hν ) ∈ H (A) is determined by its ν-components
hν ∈ H ( Fν ). h ∈ H (A) may also be uniquely written as h = h∞ h f with h f ∈ H (A f )
and h∞ ∈ H (R). This gives for any closed subgroup K ⊂ H (A) a decomposition
K = K∞ × K f with K f = K ∩ H (A f ) and K∞ = K ∩ H (R). We consider K∞ as a Lie-
group in a natural way and for any Lie-group L we denote the connected component of
the identity by L0 . We have
G (R) = G0 ( F ⊗Z R) = ∏ GL2 (R)

ν|∞
and subgroups
K∞ : = ∏ Kν , 1
K∞ := Z0 (R)0 · ∏ Kν1 ,
ν|∞ ν|∞
where Kν := Z0 (R)SO(2) and Kν1 := SO(2). We identify the two connected components
1 0
of GL2 (R) with  , e = ±1. In this way we may view {±1} g ∼ = π0 ( G (R)) ⊂ G (R)
0 e
as a subgroup normalizing K∞ . Moreover, we will often identify Z (R) = ( F ⊗Q R)× =
∏ν|∞ R× and R× = Z0 (R) embedded diagonally.
48
3.1.2 The spaces
Take a compact open subgroup K f ⊂ G (A f ). Our main examples will be
ker G (Ẑ) → G (Z/NZ) , N ∈ N,

1 K ⊂ G (A) and
and its conjugates. Set K := K∞ K f ⊂ G (A), K1 := K∞ f
K N := ker G (Ẑ) → G (Z/NZ) K∞ .

We define spaces
SK := K \ G (A)/G (Q) and MK := K1 \ G (A)/G (Q)
by taking double quotients. These two spaces are related by the obvious projection
map φK : MK → SK .
Remark 3.1.1. Let us recall some facts which can be found mainly in Harder (1987)
1.0, 1.1.
• SK parametrizes abelian varieties A over spec(C) of dimension g with real multi-

plication by O and a level-K f -structure (Milne (1979), Theorem 1.2). SK can be
realized as the C-points of a scheme defined over Q. If g > 1, there is no uni-
versal abelian scheme above SK , as there are always non trivial automorphisms
of our abelian varieties respecting the O -structure and the level-K f -structure . To
see this consider
Z (K ) := Z (Q) ∩ K f ⊂ Z (Q) ∩ G (Ẑ) = O ×
a subgroup of finite index. If g > 1, Z (Q) ∩ K f is never trivial, because the rank of
this abelian group is positive by Dirichlet’s unit theorem.
• K f \ G (A f ) is discrete and countable.

g
• Set H± := ( F ⊗Q C)× \ ( F ⊗Q R)× . We identify 1 ⊗ i with the imaginary unit i and
get the diffeomorphism
 
a b
g
K ∞ \ G (R ) → H± , g =   7→ b + id =: τ = τ ( g)
c d a + ic
g
So G (R) acts on H± from the right by τ · g := b+dτ
a+cτ . We also have the diffeomor-
phism
g
1
K∞ \ G (R) → H± × (R>0 \ Z (R)0 ), g 7→ (i · g, | det( g)|)
49
where R>0 are of course the positive real numbers. The inverse map is
 
q 1 x
(τ = x + iy, r ) 7→ r |y|−1 ·  .
0 y
The absolute value and the square root are taken componentwise.
• The stabilizer of a connected component of K \ G (A) or K1 \ G (A) in G (Q) is

given by G ( g f ) := g−
f
1
K g
f f · G ( R ) 0 ∩ G (Q) for a g ∈ G (A ).
f f
• G (Q)/Z (K ) acts properly discontinuously on K \ G (A) and fixpointfree, if K is

small enough, say K = K N and N ≥ 3. Using the diffeomorphism above we see
that G (Q) acts properly discontinuously and without fixpoints on K1 \ G (A), if we
suppose K f to be small enough.
• Consider the ray class group Cl FK := det(K f ) ∏ν|∞ R>0 \ IF /F × . The determinant
induces maps
det : SK → Cl FK , det : M → Cl FK .
Since we have strong approximation for SL2 , the fibers of these maps are the
connected components of SK and MK . So our spaces have a finite number of
connected components, since the class number of F is finite.
• If K f is small enough, MK has a structure of a C ∞ -manifold making the canonical

projection K1 \ G (A) → MK a local diffeomorphism.
We will always assume that we have chosen K f small enough so that the quotient
spaces SK and MK have the natural structures of C ∞ -manifolds. Moreover, we consider
{±1} g = ∏ {±1} ⊂ Z (R) = ∏ R× .

ν|∞ ν|∞
Lemma 3.1.2. φK : MK → SK is a principal Z0 (R)0 {±1} g \ Z (R)/Z (K )-bundle.
Proof. Z (R) acts clearly by matrix multiplication on MK from the right, as it lies in the
center of G (A). This action factors over the group Z0 (R)0 {±1} g \ Z (R)/Z (K ), is by
diffeomorphisms and preserves the fibers of φK . What is left to be shown is that φK is
a fiber bundle and that the action of Z0 (R)0 {±1} g \ Z (R)/Z (K ) on the fibers is simply
transitive. This problem is local on the base.
 
g 1 x
H± /G ( g f ) ⊂ SK , τG ( g f ) 7→ K   g f G (Q)
0 y
50
is the inclusion of some connected components. We have
g g
φK−1 (H± /G ( g f )) = H± × (R>0 \ Z (R)0 ) /G ( g f ),

where G ( g f ) acts on the second factor by multiplication with the determinant. Here
g g
φK : H± × (R>0 \ Z (R)0 ) /G ( g f ) → H± /G ( g f )

is induced by the first projection. As G ( g f )/Z (K ) acts properly discontinuously and

g g
fixpointfree on H± , we may find for any point in H± an open neighborhood U such that
for any γ ∈ G ( g f ) Uγ ∩ U 6= ∅ implies γ ∈ Z (K ). If we denote the image of U in SK
by U 0 , we get a diffeomorphism φK−1 (U 0 ) = U 0 × (R>0 \ Z (R)0 / det( Z (K ))). So φK is a
fiber bundle. But
det : Z0 (R)0 {±1} g \ Z (R)/Z (K ) → R>0 \ Z (R)0 / det( Z (K ))
is a diffeomorphism showing that the action of Z0 (R)0 {±1} g \ Z (R)/Z (K ) on the fibers
of φK is simply transitive. This completes the proof.
Remark 3.1.3. If K = K N , N ≥ 3, we have −1 ∈

/ Z (K ) and Dirichlets unit theorem tells
us that Z (K ) acts properly discontinuously and fixpointfree on Z0 (R)0 {±1} g \ Z (R)
and that the quotient Z0 (R)0 {±1} g \ Z (R)/Z (K ) is compact.
Let us now define families of topological tori above MK . Take W f ⊂ V (A f ) o G (A f ) a

compact open subgroup such that the projection of W f onto G (A f ) gives K f . Set W :=
1 ⊂ G (A) where we always consider G (A) ⊂ V (A) o G (A) via g 7 → (0, g ). Set
W f · K∞
πW : TW := W \ V (A) o G (A)/V (Q) o G (Q) → MK , (v, g) 7→ g
Lemma 3.1.4. πW : TW → MK is a family of topological tori.
Proof. Write W f = Vf o K f , where Vf ⊂ V (A f ) is a compact open subgroup. We have

the bijection
V (A) o G (A) → V (A) × G (A), (v, g) 7→ ( g−1 v, g) =: (w, g).
In the new coordinates (w, g) the group actions look like
(w, g)(q, γ) = (γ−1 (w + q), gγ), (q, γ) ∈ V (Q) o G (Q),
(u, k)(w, g) = ((kg)−1 u + w, kg), (u, k) ∈ W.
51
When we write V (A) × G (A) we always work in the coordinates (w, g) with induced
group actions. The pullback of TW to K1 \ G (A) is W \ V (A) × G (A)/V (Q). The action
of Vf on V (A) depends on g f ∈ G (A f ). For g f ∈ G (A f ) consider K1 \ K1 g f G (R). This
is just a collection of connected components in K1 \ G (A). We may now identify
pr −1 (K1 \ K1 g f G (R)) ∼
= ( g−f 1 Vf \ V (A)/V (Q)) × (K1 \ K1 g f G (R))
W (w, h)V (Q) 7→ ( g−

f Vf + w + V ( Q ) , K h )
1 1
and the fiber over each point
g− −1
f Vf \ V (A) /V (Q) = V (R) / ( g f Vf ∩ V (Q))
1
is a topological torus. This shows that W \ V (A) × G (A)/V (Q) → K \ G (A) is a family
of topological tori. Because G (Q) acts on the latter by homomorphisms and properly
discontinuously without fixpoints on the base, TW is a family of topological tori.
Remark 3.1.5. Consider K f ⊂ G (Ẑ) and W f = V (Ẑ) o K f . Over
g
H± × (R>0 \ Z (R)0 ) /G ( g f )

our torus looks like
g
F ⊗Q C × H± × (R>0 \ Z (R)0 ) /V ( g f ) o G ( g f ),

where V ( g f ) := g−
f V (Ẑ) ∩ V (Q) and V ( g f ) o G ( g f ) acts in the following way:
1
   
l1 a b

z + l1 + τl2
(z, τ, r ) ·    ,γ =    = , τ · γ, r · det(γ) .
l2 c d a + cτ
The natural inclusion
g
F ⊗Q C × H± × (R>0 \ Z (R)0 ) /V ( g f ) o G ( g f ) → TW

is given by
   
u1 q 1 x
(z, τ, r ) 7→   , r | y | −1   = (w, g)
u2 0 y
where z = u1 + u2 τ and (w, g) f = (0, 1).
52
3.2 Construction of adelic polylogarithmic Eisenstein classes
3.2 Construction of adelic polylogarithmic Eisenstein

classes
Next we will discuss how we can associate not only to formal linear combinations of tor-
sion sections of tori but also to Schwartz-Bruhat functions f on V (A f ) polylogarithmic
Eisenstein classes Eis( f ). The naturality of the polylogarithmic construction guarantees
the G (A f )-equivariance of the operator Eis.
3.2.1 Coefficient sheaves and integral structures
We introduce some locally constant sheaves on our spaces. Given a topological space
X on which a group G acts continuously from the right we associate to any G-(left)-
module V a sheaf V on the quotient space X/G in a functorial way. We make the
constant sheaf V on X into a G-sheaf by defining for g ∈ G morphisms
g : V → g∗ V, f 7→ { x 7→ g f ( xg)} ,
if f : U → V is a locally constant map on U ⊂ X. If p : X → X/G is the canonical map,

these morphisms make p∗ V into a sheaf of G-modules and we can consider the fixed
sheaf p∗G V which is defined by p∗G V (U ) := p∗ V (U )G . This fixed sheaf is our V . We will
often just speak about the sheaf induced by a G-module or associated to a G-module.
On TW we have the logarithm sheaf LogW . It is induced by the V (Q) o G (Q)-left-module

∏k≥0 Symk V (Q), where V (Q) operates by multiplication with the exponential series and
G (Q) by its standard action, see Proposition 2.4.10.
The sheaf LogW carries an integral structure. Suppose we are on a connected compo-
nent cut out by g f ∈ G (A f ) and K f ⊂ G (Ẑ), W f = V (Ẑ) o K f ; compare Remark 3.1.5.
The fundamental group of the family of topological tori above this connected component
is G ( g f ) o V ( g f ). LogW is the locally constant sheaf associated to the G ( g f ) o V ( g f )-
module Z[[V ( g f )]] as considered in Definition 6.
Moreover, we have the locally constant sheaves Symk HK on MK associated to the
G (Q)-modules Symk V (Q) and the sheaf Symk HK0 on SK , which is associated to the
same G (Q)-module. These sheaves also have integral structures coming from Symk V ( g f ) PD .
If we consider sheaves on MK arising from G (Q)-modules, the sheaves on SK corre-
53
sponding to the same module will always be denoted by a prime.

The sheaf µK on MK is induced by the G (Q)-module Q, where G (Q) operates by mul-
tiplication with the norm of the determinant N ◦ det. As N ◦ det( Z (K )) = 1, there is no
need to distinguish between µK on MK and µ0K on SK , as µK would be the pullback of
µ0K . We have a trivialization of µK given by the global section
sK : K \ G (A) →→ Q
g 7→ det g f f · sgn( N (det g∞ ))−1 ,

where kk f := ∏ν.∞ | |ν is the product of the finite local norms of F. Clearly, orTW /MK =
V2g
µK = HK .
As we vary the level on MK or TW , we see that our sheaves LogW are obtained by
0
pullback; for example, given W f0 ⊂ W f we have the canonical projection pW
W : TW 0 → TW
W 0 ∗ Log
and pW W = LogW 0 by Lemma 2.1.11. The same is true for Symk HK and µK . We
will drop the subscript K, when it does not cause any confusion.
3.2.2 Torsion sections
We are looking for a good class of compact open subgroups W f ⊂ V (A f ) o G (A f )

such that we can give explicit sections for πW .
From now on we will always deal with W f of the kind: W f = VK o K f , where K f ⊂

G (A f ), VK ⊂ V (A f ) are compact open subgroups and VK is an Ô -module on which K f
acts as usual by matrix-multiplication. Moreover, we suppose that we have a compact
open Ô -module VK0 ⊃ VK such that K f acts trivially on VK0 /VK , |VK0 /VK | > 1. Our main
example will be K = K N 2 , VK = NV (Ẑ) and VK0 = N −1 V (Ẑ). Given such a W we
have a natural injective arrow VK0 /VK × MK → TW , which is induced by the inclusion
VK0 × G (A) → V (A) o G (A).
We define DW as the image of this map without the image of the zero section. DW is a
closed submanifold of the torus TW , which is a finite covering of MK . We set UW to be
the open complement of DW in TW .
Let us apply the construction of the polylogarithm to this situation. We have the natural
short exact sequence
0 → H 2g−1 (UW , LogW ⊗Q πW

−1
µK ) → H 0 ( DW , LogW | DW ) → H 0 (MK , Q)
54
induced by the localization sequence for the pair ( DW , TW ). We have the inclusion
H 0 ( DW , Q) ⊂ H 0 ( DW , LogW | DW ) and on this submodule the last arrow in the short
exact sequence is just given by the trace map H 0 ( DW , Q) → H 0 (MK , Q). Exactness
of the sequence ensures that given a section f ∈ H 0 ( DW , Q) of trace 0 we have a
unique element
polW ( f ) ∈ H 2g−1 (UW , LogW ⊗Q πW

−1
µK )
mapped to f . As DW ∩ im(0) = ∅, we may specialize polW ( f ) along the zero section

of TW to get a class
EisW ( f ) ∈ H 2g−1 (MK , ∏ Symk HK ⊗ µK ) = H 2g−1 (MK , 0−1 LogW ⊗ µK ),

k ≥0
k
in other words (EisW ( f ))k ∈ ∏k≥0 H 2g−1 (MK , Symk HK ⊗ µK ).
3.2.3 Definition of adelic Eisenstein classes
We want to extend the idea of polylogarithmic Eisenstein classes to more general func-
tions f . Let us denote the space of Schwartz-Bruhat functions on V (A f ) with values
in a subfield k ⊂ C by S(V (A f ), k ), compare for example Gel0 fand et al. (1990) p.267.
Set
( Z
)
S(V (A f ), k)0 = f ∈ S(V (A f ), k ) : f (0) = 0, f dv f = 0
V (A f )
where dv f is any Haar measure on V (A f ). S(V (A f ), k )0 is a right G (A f )-module,

when we set as usual f · g f (v) = f ( g f v). This follows from the integral transformation-
formula
Z −1 Z
f ( g f v)dv f (v) = det( g f ) f
f (v)dv f (v).
V (A f ) V (A f )
We will need a more general space of functions. For any n ∈ Z denote by

S(V (A f ), µ⊗n ⊗ k) the space of functions f : V (A f ) o G (A f ) × π0 ( G (R)) → k such
that
∀ x ∈ G (A f ) × π0 ( G (R)) : f ( , x ) ∈ S(V (A f ), k),
there is some compact open subgroup K f ⊂ G (A f ) such that f factors as
f ( , ) : K f \ G (A f ) × π0 ( G (R)) → S(V (A f ), k ), x 7→ f ( , x )
55
and for all v ∈ V (A f ), x ∈ G (A f ) × π0 ( G (R)) and γ ∈ G (Q)
f (v, xγ) = N (det(γ))−n f (v, x ).
We define S(V (A f ), µ⊗n ⊗ k )0 as the subspace of S(V (A f ), µ⊗n ⊗ k ) consisting of

those functions f with f ( , x ) ∈ S(V (A f ), k)0 for all x ∈ G (A f ) × π0 ( G (R)). Again
S(V (A f ), µ⊗n ⊗ k)0 is a right G (A f ) × π0 ( G (R))-module, when we take the action in-
duced by the canonical left G (A f ) × π0 ( G (R)) action on V (A f ) o G (A f ) × π0 ( G (R)).
If k = Q we simply set S(V (A f ), µ⊗n ⊗ k ) =: S(V (A f ), µ⊗n ) and S(V (A f ), µ⊗n ⊗
k )0 =: S(V (A f ), µ⊗n )0 .
Lemma 3.2.1. S(V (A f ), µ⊗n ⊗ C)0 is generated by ϕ of the form

n
ϕ(v, g) = f (v) · η (det( g)) kdet( g)k f sgn( N (det( g))) ,
for f ∈ S(V (A f ), C)0 and η : ∏ν|∞ R>0 \ IF /F × → C× a continuous character.
Proof. Consider ϕ ∈ S(V (A f ), µ⊗n ⊗ C)0 arbitrarily. Write

n
ϕ(v, g) = F (v, g) kdet( g)k f sgn( N (det( g))) ,
for F ∈ S(V (A f ), µ⊗0 ⊗ C)0 . F factors in the second argument over some K f ⊂ G (A f )
compact open, in other words
F : K f \ G (A f ) × π0 ( G (R))/G (Q) → S(V (A f ), C)0 .
We have strong approximation for SL2,F and therefore
det : K f \ G (A f ) × π0 ( G (R))/G (Q) → det(K f ) ∏ R>0 \ IF /F × = Cl FK

ν|∞
is an isomorphism. In this manner we may interpret F in the second argument as a

function on Cl FK . Cl FK is a finite group of cardinality hK and we can describe functions
Cl FK → C as finite linear combinations of characters by the Fourier inversion theorem.
We get explicitly
F (v, g) = h−
K
1
∑ F̂ (v, η ) · η (det( g)), F̂ (v, η ) = ∑ F (v, x )η ( x ).
dK
η ∈Cl x ∈Cl FK
F
56
Finally, let us set
lim H • (MK , Symk HK ⊗ µ⊗ n • k

K ) = : H (M, Sym H ⊗ µ
⊗n
).
−→
Kf
It is a G (A f ) × π0 ( G (R))-module as explained in Harder (1987) 1.2.
Remark 3.2.2. S(V (A f ), µ⊗n )0 = S(V (A f ), Q)0 ⊗Q H 0 (M, µ⊗n ).
Proposition 3.2.3. We have a G (A f ) × π0 ( G (R))-equivariant operator
Eisk : S(V (A f ), µ⊗n )0 → H 2g−1 (M, Symk H ⊗ µ⊗n+1 )
k
which is induced by the operators EisW .
Proof. Given a φ ∈ S(V (A f ), Q)0 there is an M ∈ N such that supp(φ) ⊂ M−1 V (Ẑ)
and an M0 ∈ N such that φ is well-defined on V (A f )/M0 V (Ẑ). So considering N =
MM0 φ can be identified with a function on N −1 V (Ẑ)/NV (Ẑ). Let us investigate f ∈
S(V (A f ), µ⊗n )0 . As f factors over some compact open subgroup K f of G (A f ) and
K f \ G (A f ) × π0 ( G (R))/G (Q) is finite, we may find an N ∈ N such that f is well-
defined modulo K N 2 in the second argument and f behaves like φ above in the first
argument. In other words, f induces a function
( N −1 V (Ẑ)/NV (Ẑ) \ {0}) × K N 2 \ G (A f ) × π0 ( G (R)) → Q
which is also denoted by f . We are now in the situation of our groups W of Section 3.2.2,
namely K f = K N 2 , NV (Ẑ) = VK , VK0 = N −1 V (Ẑ) and W f = NV (Ẑ) o K N 2 . f may be
identified with a section f ∈ H 0 ( DW , µ⊗ n
K ) which actually has trace zero: We have for
all x ∈ G (A f ) × π0 ( G (R)):
∑ f (v, x ) = ∑ f (v, x ) = 0
v∈ N −1 V (Ẑ)/NV (Ẑ)\0 v∈ N −1 V (Ẑ)/NV (Ẑ)
if and only if
Z
0= ∑ f (v, x )
NV (Ẑ)
dv f =
v∈ N −1 V (Ẑ)/NV (Ẑ)
Z Z
= ∑ NV (Ẑ)
f (v + u, x )dv f (u) =
V (A f )
f (v, x )dv f .
v∈ N −1 V (Ẑ)/NV (Ẑ)
k
Consequently, we get EisW ( f ) ∈ H 2g−1 (MK , Symk HK ⊗ µ⊗
K
n +1
). We want to define
Eisk ( f ) = image of EisW

k
( f ) in H 2g−1 (M, Symk H ⊗ µ⊗n+1 ).
57
We have to show that this construction is independent of the choices made. Consider
again f ∈ S(V (A f ), µ⊗n )0 . Suppose that we have W1 and W2 both fulfilling the proper-
ties above. In other words, we have for i = 1, 2 compact open subgroups K f ,i ⊂ G (A f )
acting on the compact open Ô -modules VK0 i ⊃ VKi and acting trivially on VK0 i /VKi , such
that f is well-defined modulo Ki in the second argument, f is well-defined modulo VKi
in the first argument and for all x ∈ G (A f ) × π0 ( G (R)) supp( f ( , x )) ⊂ VK0 i .
If we set
K = K1 ∩ K2 , VK = VK1 ∩ VK1 , VK0 = VK0 1 ∩ VK0 1 , and W = W1 ∩ W2 ,

k k k
all the properties are also fulfilled by W so that we have EisW ( f ), EisW 1
( f ) and EisW 2
( f ).
To see that Eisk is well-defined we have to show that EisW
k
( f ) is obtained by pullback
k
along the canonical maps qW
Wi : MK → MKi from EisWi ( f ) for i = 1, 2. It is enough to
consider i = 1. We have a commutative square
pW
W
TW
1
/ TW
1
qW
W
MK
1
/ MK
1
0 : = pW −1 ( D ) ⊃ D , it is a finite cover of M . As the localization-sequence,

Set DW1 W1 W1 W K
µ and Log are natural in pullbacks, see Remark 2.3.9, we get the following commutative
diagram
H 2g−1 (TW1 \ DW1 , Log ⊗ µ⊗n+1 ) / H 0 ( DW , Log ⊗n

1 | DW ⊗ µ ) 1

H 2g−1 (TW \ DW
0 , Log ⊗ µ⊗n+1 ) / H 0 ( D 0 , Log 0 ⊗ µ⊗n )
1 W1 |D W1
The diagram shows that pW ∗ W ∗ 0 0 ⊗ n ).

W1 polW1 ( f ) corresponds to the section pW1 f ∈ H ( DW1 , µ
0 , see Re-
By naturality of the localization sequence for the inclusion i : DW ⊂ DW 1
mark 2.3.10, we also get the commutative diagram
H 2g−1 (TW \ DW , Log ⊗ µ⊗n+1 ) / H 0 ( DW , Log ⊗n

| DW ⊗ µ )
1
res i∗

H 2g−1 (TW \ DW
0 , Log ⊗ µ⊗n+1 ) / H 0 ( D 0 , Log 0 ⊗ µ⊗n )
1 W1 |D W1
where i∗ : H 0 ( DW , µ⊗n ) → H 0 ( DW
0 , µ⊗n ) is given by extension by zero. i f and pW ∗ f ∈
1
∗ W1
0 , µ⊗n ) coincide and consequently
H 0 ( DW 1
∗
res(polW ( f )) = pW
W1 polW1 ( f ).
58
3.3 Pushforward of polylogarithmic Eisenstein-classes
0 , we have
As 0 : MK → T meets neither DW nor DW 1
EisW ( f ) = 0∗ polW ( f ) = 0∗ res(polW ) = 0∗ pW ∗ W ∗ ∗ W ∗

W1 polW1 ( f ) = qW1 0 polW1 ( f ) = qW1 EisW1 ( f )
It follows that our operator is well-defined.

G (A f ) × π0 ( G (R))-equivariance follows easily from our construction. Given a section f
with corresponding groups K, VK , VK0 and W as above and given g ∈ G (A f ) × π0 ( G (R))
the corresponding groups for f · g can be chosen g−1 Kg, g−1 VK , g−1 VK0 and g−1 Wg. We
have the morphism
T g−1 Wg → TW , (v, h) 7→ ( gv, gh) = g(v, h)
and use Remark 2.3.9 to get the commutative diagram
H 2g−1 (TW \ DW , Log ⊗ µ⊗n+1 ) / H 0 ( DW , Log ⊗n

| DW ⊗ µ )
1

H 2g−1 (T g−1 Wg \ Dg−1 Wg , Log ⊗ µ⊗n+1 ) / H 0 ( D −1 , Log ⊗ µ⊗n )
g Wg |D g−1 Wg
As the pullback g∗ f is exactly the section corresponding to f · g, we have g∗ polW ( f ) =

polg−1 Wg ( f · g). It follows
Eis( f · g) = 0∗ polg−1 Wg ( f · g) = 0∗ g∗ polW ( f ) = g∗ 0∗ polW ( f ) = g∗ Eis( f )
and we are done.
We are going to decompose the cohomology of MK by a Leray-Hirsch theorem into a

product of the cohomology of SK and the cohomology of the fiber of φK . This will yield a
decomposition of the polylogarithmic Eisenstein classes on MK and a way to integrate
these classes to gain classes on SK .
Lemma 3.3.1. Let G and H be groups and R a Dedekind domain such that R has a
resolution P → R by finitely generated free R[ G ]-modules (for example G ∼
= Zr ). Let M
be an R[ G ]-module finitely generated free as R-module and let N be any R[ H ]-module.
Then we have the Künneth-sequence
H p ( G, M) ⊗ R H q ( H, N ) → H n ( G × H, M ⊗ R N ) →
M
0→
p+q=n
59
Tor1R ( H p ( G, M ), H q ( H, N )) → 0
M
→
p + q = n +1
Proof. We choose a resolution P• → R by finitely generated free R[ G ]-modules and a

resolution Q• → R by free R[ H ]-modules. Since R is a Dedekind domain the torsion-
free R-modules are exactly the flat R-modules and therefore any submodule of a flat R-
module is flat. With this at hand we may use Weibel (1994) Theorem 3.6.3 to conclude
that P• ⊗ R Q• → R is a resolution by free R[ G × H ]-modules. So H • ( G × H, M ⊗ R N ) =
H • ( Hom R[G× H ] ( P• ⊗ R Q• , M ⊗ R N )). Let us consider the canonical map
µ : Hom R[G] ( P, M ) ⊗ R Hom R[ H ] ( Q, N ) → Hom R[G× H ] ( P ⊗ R Q, M ⊗ R N )
µ( f ⊗ g)( p ⊗ q) = f ( p) ⊗ g(q) for p ∈ P and q ∈ Q for a finitely generated free R[ G ]-

module P and a free R[ H ]-module Q. We have
Hom R[G× H ] ( P ⊗ R Q, M ⊗ R N ) = Hom R[G× H ] ( R[ G ]n ⊗ R Q, M ⊗ R N ) = Hom R[ H ] ( Q, M ⊗ R N )n
Now M = Rm is free, so we continue by
Hom R[ H ] ( Q, N m )n = Hom R[ H ] ( Q, N )mn = Hom R[ H ] ( Q, N ) ⊗ R Mn =
= Hom R[ H ] ( Q, N ) ⊗ R Hom R[G] ( R[ G ]n , M) = Hom R[ H ] ( Q, N ) ⊗ R Hom R[G] ( P, M)
So µ is an isomorphism. Moreover, Mn = Hom R[G] ( P, M ) is a flat R-module and we can

apply Weibel (1994) Theorem 3.6.3 to the complex Hom R[G] ( P, M ) ⊗ R Hom R[ H ] ( Q, N )
to get the exact sequence
H p ( Hom R[G] ( P, M )) ⊗ R H q ( Hom R[ H ] ( Q, N )) →

M
0→
p+q=n
→ H n ( Hom R[G] ( P, M) ⊗ A Hom R[ H ] ( Q, N )) →
Tor1R ( H p ( Hom R[G] ( P, M)), H q ( Hom R[ H ] ( Q, N )) → 0

M
→
p + q = n +1
Using µ we compute this sequence as
H p ( G, M) ⊗ R H q ( H, N ) → H n ( G × H, M ⊗ R N ) →
M
0→
p+q=n
Tor1R ( H p ( G, M ), H q ( H, N )) → 0
M
→
p + q = n +1
60
Proposition 3.3.2. Let G be a finitely generated free abelian group, k a field and M an
k [ G ]-module finitely generated as k-module. We suppose that we have a field extension
k ⊂ K such that there is a decomposition
K ( χ )m(χ) ,
M
MK =
χ
with χ : G → K × running through all characters and K (χ) the K [ G ]-module K with
action by χ. Then cup-product induces an isomorphism
H 0 ( G, M) ⊗ A H • ( G, k ) = H • ( G, M)
Proof. We begin with H • ( G, K (χ)) = 0, if χ 6= 1. As χ 6= 1, we write G = Z × G 0 with a

subgroup G 0 and χ|Z 6= 1. One has K (χ) = K (χ|Z ) ⊗K K (χ|G0 ). H • (Z, K (χ|Z )) = 0 by
Weibel (1994) Example 6.1.4 and we use the Künneth-sequence
H p (Z, K (χ|Z )) ⊗ A H q ( G 0 , K (χ|G0 )) → H n ( G, K (χ)) →

M
0→
p+q=n
Tor1K ( H p (Z, K (χ|Z )), H q ( G 0 , K (χ|G0 ))) → 0

M
→
p + q = n +1
to conclude. Now we may calculate
H • ( G, MK ) = H • ( G, K ( χ )m(χ) ) = H • ( G, K (χ))m(χ) =
M M
χ χ
H • ( G, K (1))m(1) = H 0 ( G, MK ) ⊗K H • ( G, K )
and this isomorphism is induced by cup-product. We want to see that this map comes
from extension of scalars from the cup-product map
H 0 ( G, M) ⊗k H • ( G, k ) → H • ( G, M).
This is the case, because for any G-module N
H • ( G, NK ) = H • ( G, N ) ⊗k K
compatible with cup-product. To see this take a resolution P → k by finitely generated

free k [ G ]-modules. Then H • ( G, NK ) = H • ( HomK[G] ( PK , NK )). As k → K is flat and P
consists of finitely generated free k[ G ]-modules one identifies
H • ( HomK[G] ( PK , NK )) = H • ( Homk[G] ( P, N )) ⊗k K = H • ( G, N ) ⊗k K
61
proving the claim. Finally, the extension of scalars from k to K of the morphism
∪ : H 0 ( G, M) ⊗ A H • ( G, A) = H • ( G, M)
is an isomorphism, therefore ∪ is already an isomorphism, since k → K is faithfully

flat.
Corollary 3.3.3. We consider φK : MK → SK and Symn H. Then cup-product induces

an isomorphism
∪ : φK∗ Symn H ⊗Q R p φK∗ (Q) → R p φK∗ (Symn H)
Proof. It is clear that we have a global cup-product map and being an isomorphism is
a local problem on SK . As φK is a fiber bundle, it suffices to show that for all s ∈ SK the
induced maps on the fibers
∪ : H 0 (φK−∗1 (s), Symn H) ⊗Q H p (φK−∗1 (s), Q) → H p (φK−∗1 (s), Symn H)
are isomorphisms. Following Grothendieck (1957) Chapitre V we may prove this by

showing that the corresponding map on the level of group cohomology
∪ : H 0 ( Z (K ), Symn V (Q)) ⊗Q H p ( Z (K ), Q) → H p ( Z (K ), Symn V (Q))
is an isomorphism. This is the case, as our situation fulfills the requirements of Propo-
sition 3.3.2. Z (K ) is a finitely generated free abelian group by Dirichlets unit theorem
and we have the field extension Q → Q such that Symn V (Q) decomposes: We have
the isomorphism
O ⊗Z Q → ∏ Q, x ⊗ r 7→ (σ( x )r )σ .
σ:O→Q
and obtain
nσ 2
n
V (Q) ⊗Q Q = Q .
M O
SymQ SymQ
∑σ nσ =n σ
2
If we take the standard basis e1 =: X, e2 =: Y of Q , we get the Q-basis
∏ Xσk Yσn −k
σ σ σ
, k σ = 0, ..., nσ ,
σ
n V (Q) ⊗ Q, which also yields the decomposition into one-dimensional Z ( K )-
for SymQ Q
representations, as
e · ∏ Xσkσ Yσnσ −kσ = ∏ σ(e)n ∏ Xσk Yσn −k

σ σ σ σ
σ σ σ
for e ∈ Z (K ) ⊂ O × .
62
Remark 3.3.4. We set from now on A := Z[ dF1N ] where d F is the discriminant of F. We

have the Hilbert-class-field FHil of F with ring of integers O Hil . Let us set R := O Hil [ Nd1 F ].
A → R is a faithfully flat ring extension. From now on we will always consider the sheaf
Symk H PD over the ring A. The natural adjunction morphism
φK−1 φK∗ Symk H PD → Symk H PD
Z (K )
is a split monomorphism. Indeed, the corresponding map of G ( g f )- modules Symk V ( g f ) PD →
Symk V ( g f ) PD is a split mono. To see this note first that
Z (K )
Z (K )
Symk V ( g f ) PD ⊗ A R = Symk V ( g f ) PD ⊗ A R ,
as R is a free A-module. By faithful flatness it suffices now to find a retraction for the
map
Z (K )
Symk V ( g f ) PD ⊗ A R → Symk V ( g f ) PD ⊗ A R.
There are two fractional ideals a and b of O such that V ( g f ) = a ⊕ b. As we have

d F ∈ R× , there are again isomorphisms
a ⊗A R → ∏ σ(a) R, x ⊗ r 7→ (σ( x )r )σ , b ⊗ A R → ∏ σ(b) R, x ⊗ r 7→ (σ( x )r )σ

σ:O→Q σ:O→Q
and we get
Symk V ( g f ) PD ⊗ A R = SymnRσ (σ (a) R ⊕ σ(b) R) PD .

M O
∑σ nσ =n σ
In the Hilbert-class-field any Ideal of O becomes principal. Therefore we get ασ , β σ ∈

FHil with σ(a) R = ασ R and σ(b) R = β σ R. If again e1 =: X, e2 =: Y is the standard basis
of R2 , we may split Symk V ( g f ) PD ⊗ A R as in Corollary 3.3.3 into one dimensional Z (K )-
Z (K )
representations R ∏σ (ασ Xσ )[kσ ] ( β σ Yσ )[nσ −kσ ] . In Symk V ( g f ) PD ⊗ A R those one
dimensional representations occur, on which Z (K ) acts trivially. From this description
it is clear that the inclusion of the invariants is a split mono. But we can say more.
The one dimensional subrepresentations occurring in the invariants induce characters
Res F/Q Gm (Q) → Gm (Q) which have to be trivial on Z (K ). By Serre (1968) II.3.3 we
know that these have to factor through the norm character, as our field F is totally
real. In other words, R ∏σ (ασ Xσ )[kσ ] ( β σ Yσ )[nσ −kσ ] occurs in the invariants, if and only
if there is an m ∈ N0 such that for all σ : F → Q m = nσ . It follows that the module
Z (K )
Symk V ( g f ) PD ⊗ A R does not depend on the level K, as soon as Z (K ) ⊂ O × does
63
Z (K )
not contain an element of negative norm. By faithful flatness Symk V ( g f ) PD does not
depend on the level K in this case. As we have usually K f = K N , N ≥ 3, we do not
have elements of negative norm. Finally, we want to remark that we have canonically
φK∗ Symk H PD = Symk H0PD .
Corollary 3.3.5. Symk H0PD = 0, if g - k. If k = mg and Z (K ) contains no element of

negative norm then Symk H0PD 6= 0.
Remark 3.3.6. We want to describe the image

Z (K )
Z (K ) Z (K )
SymkA V ( g f ) PD → SymkA V ( g f ) PD ⊗ A R = SymkR (V ( g f ) ⊗ A R) PD
Hil /Q)
explicitly. Gal( FHil /Q) acts on R and we get RGal( F = A by Galois-theory. This
implies
Gal( FHil /Q)
Gal ( FHil /Q)

SymkA V ( g f ) PD = SymkA V ( g f ) PD ⊗ A R = SymkR (V ( g f ) ⊗ A R) PD ,
as SymkA V ( g f ) PD is a free A-module. We identify

Hil
Z (K ) Gal( F /Q)

Z (K )
SymkA V ( g f ) PD = SymkR (V ( g f ) ⊗ A R) PD ,
since the Galois-action and the Z (K )-action commute. Let us suppose k = mg, other-
wise SymkA V ( g f ) Z(K) = 0 by the considerations above. Write V ( g f ) = a ⊕ b with a as in
Z (K )
Remark 3.3.4. In SymkR (V ( g f ) ⊗ A R) PD we have the direct summand R ∏σ ( β σ Yσ )[m] .
[m]
Gal( FHil /Q) acts trivially on ∏σ Yσ , because τ · Yσ = Yτ −1 ◦σ for τ ∈ Gal( FHil /Q).
Therefore
∼
∏ σ(b)m R → R ∏( βσ Yσ )[m] = R ∏ βmσ Yσ , x 7→ x · ∏ Yσ
= [m] [m]
σ σ σ σ
as Galois-modules, where on the left-hand side we have the product of ideals.
Lemma 3.3.7.
!Gal( FHil /Q)
∏ σ (b) R = N(b) A
σ
ν (b)
where N(b) := ∏ν finite place of F |κ (pν )|ν(b) and ν(b) is defined by bOν = (πν ).
Proof. We may assume that b is integral, otherwise we multiply the ideal with a number
z ∈ Z \ {0}. A is as localization of a Dedekind ring itself a Dedekind ring. So we
64
have unique prime factorization of ideals in A and therefore our problem is local at each
maximal ideal ( p) ⊂ A. First we have
 !Gal( FHil /Q)  ! ! !
 ∏ σ (b) R  = ∏ σ (b) R ∩Q = ∏ σ (b) R ( p ) ∩Q
σ σ ( p) σ
( p)
by Galois-theory. As ( p) ⊂ Q and σ is Q-linear, we identify the last expression with

∏σ σ(b( p) ) R( p) ∩ Q. Now O( p) is a Dedekind ring with only finitely many primes and
hence a principal ideal domain. So b( p) = bO( p) for b ∈ O and we identify our ideal with
!
∏ σ ( b ) R( p) ∩ Q = N (b) R( p) ∩ Q.
σ
R( p) is the integral closure of A( p) inside FHil and as A( p) is integrally closed in Q we

get N (b) R( p) ∩ Q = N (b) A( p) . We calculate
N (b) A ( p ) = ∏ |κ (pν )|ν(b) A( p) = ∏ |κ (pν )|ν(b) A( p) .

ν finite place of F ν| p
In O( p) we have the unique prime factorization
b O ( p ) = b( p ) = ∏ pνp (b) O( p)
p ⊂ O max., p| p
and we see νp (b) = νp (b) for p| p. We get
!
∏ σ (b) R Gal( FHil /F )
= | N (b)| A( p) = ∏ |κ (p)|νp (b) A( p) =
σ ( p) p ⊂ O max.
∏ |κ (p)|νp (b) A( p) = ∏ |κ (p)|νp (b) A( p) = N (b) A( p)

p ⊂ O max., p| p p ⊂ O max., p| p
and we are done.
Proposition 3.3.8. We consider A := Z[ Nd1 F ]. V ( g f ) = g−

f V (Ẑ) ∩ V (Q) = a ⊕ b. Then
1
[m] Z (K )
N (b)m ∏σ Yσ is part of an A-basis of SymkA V ( g f ) PD and N (b) = kt2 k f , if we write
 
t1 t1 u
g f = kb = k   , k ∈ G (Ẑ).
0 t2
Proof. The first part of the proposition is clear by the preceding lemma. We have
V ( g f ) = V (b), since k−1 V (Ẑ) = V (Ẑ). From this description it is clear that b =
t2−1 Ô ∩ F and therefore N (b) = kt2 k f .
65
Proposition 3.3.9. Suppose det(K f ) = K f ∩ Z (A f ), what we always will assume from

now on. We have cohomology classes in H • (MK , Z[ 21 ]) restricting for all s ∈ SK to a
basis of H • (φK−1 (s), Z[ 21 ]). In particular, the local system R• φK∗ (Z[ 12 ]) is trivial.
Proof. Suppose that we had the cohomology classes with the desired properties. Let
H • ⊂ H • (MK , Z[ 12 ]) be the Z[ 12 ]-submodule generated by the elements restricting
to bases of the cohomology of the fibers. The natural morphism H • (MK , Z[ 12 ]) →
H 0 (SK , R• φK∗ (Z[ 21 ])) induces H • → H 0 (SK , R• φK∗ (Z[ 12 ])) and therefore an isomor-
phism of sheaves H • → R• φK∗ (Z[ 12 ]), which would be our trivialization. So let us find
these cohomology classes. Remember the determinant maps SK → Cl FK and MK →
det(K f )R>0 \ IF /F × . We consider the fiber product SK ×Cl K det(K f )R>0 \ IF /F × ,

F
where on the right hand side the structure map is the natural projection. We have the
natural maps
p : MK → SK ×Cl K det(K f )R>0 \ IF /F × , K1 gG (Q) 7→ (KgG (Q), det( g))

F
and
pr2 : SK ×Cl K det(K f )R>0 \ IF /F × → det(K f )R>0 \ IF /F ×

F
MK and SK have the same number of connected components and the cohomology
groups decompose into a direct sum of these components, so we can treat connected
components separately. We consider G ( g f ) for some g f ∈ G (A f ). We get
g g
φK : H± × (R>0 \ Z (R)0 ) /G ( g f ) → H± /G ( g f ), (τ, r ) 7→ τ

Furthermore
g g
p : H± × (R>0 \ Z (R)0 ) /G ( g f ) → H± /G ( g f ) × R>0 \ Z (R)0 / det( G ( g f ))

g
p is a map of fibrations over H± /G ( g f ). We have the projection
g
pr2 : H± /G ( g f ) × R>0 \ Z (R)0 / det( G ( g f )) → R>0 \ Z (R)0 / det( G ( g f )).
Given a basis
1
(ξ i ) ⊂ H • (R>0 \ Z (R)0 / det( G ( g f )), Z[ ])
2
the classes
1
( pr2∗ ξ i ) ⊂ H • (Hg /G ( g f ) × R>0 \ Z (R)0 / det( G ( g f )), Z[ ])
2
66
restrict back to a basis of the fibers. p respects the fibers and therefore the same is true
for ( p∗ pr2∗ ξ i ) = (ξ i ). To see this restrict p to the fiber
p : R>0 \ Z (R)0 /Z (K ) → R>0 \ Z (R)0 / det( G ( g f )), x · Z (K ) 7→ x · det( G ( g f ))
We may suppose Z (K ) = det( G ( g f )). On the left hand side e ∈ Z (K ) acts by multi-
plication with e2 , whereas e ∈ det( G ( g f )) acts on the right hand side by multiplication
with e. Therefore the map above is aN isogeny of real tori of degree ( Z (K ) : Z (K )2 ),
which is a power of 2. As 2 ∈ Z[ 21 ]× the map induces an isomorphism on cohomology.
Let us now denote the set of connected components of MK with J. We have con-
j
structed cohomology classes (ξ i ) for each j ∈ J restricting to basis of the cohomology
j
of the fibers. The family of cohomology classes (∑ j∈ J ξ i ) on MK has the desired prop-
erty: If s ∈ SK is given, then φK−1 (s) will be containend in a connected component
corresponding to some j0 ∈ J. We get
j j
( ∑ ξ i | φ −1 ( s ) ) = ( ξ i 0| φ −1 ( s ) )
j∈ J
and the right-hand side is a basis of the cohomology of the fiber φK−1 (s) by the construc-
tion above.
Remark 3.3.10. The cohomology classes described above are pullback by
det : MK → det(K f )R>0 \ IF /F ×
More precisely, H • (det(K f )R>0 \ IF /F × , Z) is torsion free, so we may understand the

latter group as a subgroup of the cohomology with real coefficients, which may be
calculated by differential forms. Consider
R>0 \ Z (R)0 / det( G ( g f )) ,→ det(K f )R>0 \ IF /F ×
the inclusion of the connected component of the identity. We calculate H • (R>0 \

Z (R)0 / det( G ( g f )), Z) by invariant differential forms. These give rise to invariant forms
on det(K f )R>0 \ IF /F × and span a subspace
1
H•K ⊂ H • (det(K f )R>0 \ IF /F × , Z[ ])
2
The pullback of these forms by det have the desired property of the proposition above
and we denote the corresponding subspace also by H•K ⊂ H • (MK , Z[ 12 ]). We have
67
the invariant 1-forms drν

rν on FR× , which we may also interprete as invariant 1-forms on
det(K f )R>0 \ IF /F × . The invariant forms H•K are R-linear combinations of wedge-
products these forms. Moreover, we set H• := H•K ⊗Z[ 1 ] Q, as this space does not
2
depend on the level K.
Theorem 3.3.11. We have an isomorphism
H • (SK , Symk H0 ) ⊗Q H• → H • (MK , Symk H), ω ⊗ η 7→ ω ∪ η
Proof. Let us choose a resolution I • of Q on MK by injective sheaves of Q-modules.

We take a basis of η1 , ..., ηl of H• in other words cohomology classes restricting to a
basis of the cohomology of the fibres of φK . Let us denote the cohomological degree
of the class ηi by di ∈ N0 (i ≥ j, then di ≥ d j ). We may interpret ηi as a morphism
ηi : Q[−di ] → I • . These morphisms induce
id ⊗ ηi = ηi : Symk H[−di ] → Symk H ⊗Q I •
By Lemma 2.3.1 we may use Symk H ⊗Q I • as an injective resolution of Symk H. Apply-

ing φK∗ and summing over all i yields
∑ ηi : φK∗ Symk H[−di ] → φK∗ (Symk H ⊗Q I • ).

M
i i
Now apply H p and get map
∑ φK∗ Symk H[−di ] → R p φK∗ (Symk H).

M
ηi :
i with di = p i with di = p
R p φK∗ (Symk H) = φK∗ Symk H ⊗ A R p φK∗ (Q)
and the properties of ηi show that the last map is an isomorphism. In particular,
∑ ηi : φK∗ Symk H[−di ] → φK∗ Symk H ⊗Q I •

M
i i
is a quasi-isomorphism. We conclude
RΓ(SK , φK∗ Symk H) ⊗Q H• = RΓ(SK , φK∗ Symk H)[−di ] =

M
= RΓ(SK , ) RφK ∗ Symk H = RΓ(SK , Symk H)
Moreover, we have Symk H0 = φK∗ Symk H completing the proof.
68
Let us consider H• as trivial G (A f ) × π0 ( G (R))-module. Set
lim H • (SK , Symk H0 ) := H • (S , Symk H0 ).

−→
Corollary 3.3.12. We have an isomorphism of G (A f ) × π0 ( G (R))-modules
H • (S , Symk H0 ) ⊗Q H• → H • (M, Symk H)
Proof. The isomorphism is Theorem 3.3.11. G (A f ) × π0 ( G (R))-equivariance follows

from the fact that H• consists of invariant forms.
Corollary 3.3.13. We have an G (A f ) × π0 ( G (R))-equivariant operator
Eiskq : S(V (A f ), µ⊗n )0 ⊗Q Hq∗ → H 2g−1−q (S , Symk H ⊗ µ⊗n+1 )
f ⊗ η ∗ 7→ Eisk ( f )(η ∗ ) for q = 0, ..., g − 1.
Proof. This is just Proposition 3.2.3 and Corollary 3.3.12
3.3.1 Explicit description of the decomposition of the cohomology
We want to calculate our polylogarithmic Eisenstein classes by differential forms in order

to compare them with those of Harder. Therefore, we need to make the decomposition
of the cohomology of MK explicit. This can be done by the theory of fiber integration
on de Rham cohomology .
Let us work over the coefficient ring Q first. We want to give the inverse map of
the natural decomposition isomorphism provided by Theorem 3.3.11. In other words,
given a basis η1 , ..., ηl of H• and ω ∈ H • (MK , Symk H) we want to know explicitly
ω1 , ..., ωl ∈ H • (SK , Symk H0 ) such that ω = ∑il=1 ωi ∪ ηi holds.
We choose the retraction Symk H → φK−1 Symk H0 of the canonical injection φK−1 Symk H0 →
Symk H provided by Remark 3.3.4.
Lemma 3.3.14. The retraction induces an isomorphism
H • (MK , Symk H) → H • (MK , φK−1 Symk H0 ).
on cohomology.
69
Proof. We know by Corollary 3.3.3 and the projection formula
R p φK∗ (φK−1 Symk H0 ) = Symk H0 ⊗ A R p φK∗ (Q) = R p φK∗ (Symk H)
so that our projection induces an isomorphism
RφK∗ (Symk H0 ) = RφK∗ (Symk H) ∈ D + (SK , Q)
in the derived category of bounded below complexes of sheaves of Q-modules. Apply-

ing R p Γ(S , ) yields the isomorphism in question.
Remark 3.3.15. We even may modify the proof of Theorem 3.3.11 to obtain an isomor-
phism
∪
H • (SK , Symk H0 ) ⊗Q H• → H • (MK , φK−1 Symk H0 )
compatible with the isomorphism of Theorem 3.3.11.
Definition 17. Let B and F be a topological manifold and f : E → B a fiber bundle with
typical fiber F. We assume F to be compact of dimension d. So f is proper. Moreover,
for a ring k we consider a k-local system V on B. There is the natural trace morphism
tr : H n ( E, f −1 V ) → H n−d ( B, V ⊗k Rd f ∗ (k)).
Let us recall how it is constructed. R p f ∗ ( f −1 V ) = 0, p > d, since
R p f ∗ ( f −1 V )b = H p ( f −1 (b), f −1 V ) = 0, p > d,
by Iversen (1986)VII 1.4, III 9.6 and 9.10. Therefore the inclusion of the truncation
induces a quasi-isomorphism τ≤ g−1 R f ∗ ( f −1 V ) → R f ∗ ( f −1 V ). We have the canonical
projection
τ≤d R f ∗ ( f −1 V ) → Rd f ∗ ( f −1 V )[−d] = V ⊗k Rd f ∗ (k )[−d],
where we used the projection formula on the right hand side (Iversen (1986) VII 2.4).
Both maps together give us the morphism
tr : R f ∗ ( f −1 V ) → V ⊗k Rd f ∗ (k )[−d]
in D + ( B, k ). Applying H n ( B, ) yields the trace morphism.
70
Remark 3.3.16. Since φK : MK → SK is a fiber bundle with compact fibers, we get
tr : H p (MK , φK−1 Symk H0 ) → H p−( g−1) (SK , Symk H0 ) ⊗Q Hg−1 .
Going through the proof of Theorem 3.3.11 we see that the composition of
∪
H p−q (SK , Symk H0 ) ⊗Q Hq → H p (MK , φK−1 Symk H0 )
M
with the trace morphism is just the projection onto H p−( g−1) (SK , Symk H0 ) ⊗Q Hg−1 .
= H g−1−• (φK−1 (s), Z) induces a Poincaré-duality H• ∼

Poincaré-duality H • (φK−1 (s), Z) ∼ =
Hg−1−• . In particular, we have the isomorphism
Z
: Hg−1 → Q = H0
and the morphism

Z
(id ⊗ ) ◦ tr : H p (MK , φK−1 Symk H0 ) → H p−( g−1) (SK , Symk H0 ),
to H• by setting it zero on H p , p 6= g − 1. By
R R
which we also denote by . We extend
Poincaré-duality we get a perfect pairing
Z
H• × H• → Q, (ω, η ) 7→ ω ∪ η.
Given our basis η1 , ..., ηl we choose a dual basis η1∗ , ..., ηl∗ with ηi ∪ η j∗ = δij .
R
Lemma 3.3.17. We consider H • (MK , φK−1 Symk H0 ) as right H• -module via cup-product.
If ω ∈ H p (MK , φK−1 Symk H0 ) is given, then we have
l Z
ω= ∑( ω ∪ ηi∗ ) ∪ ηi .
i =1
Proof. To proof this we may suppose by Theorem 3.3.11 ω = ∑il=1 ωi ∪ ηi , with ωi ∈

H p (SK , Symk H0 ). But we see with Remark 3.3.16
Z l Z l Z l Z
ω ∪ η j∗ = ∑ (ωi ∪ ηi ) ∪ η j∗ = ∑ ωi ∪ (ηi ∪ η j∗ ) = ∑ ωi ∪ (ηi ∪ η j∗ ) = ω j
i =1 i =1 i =1
and the claim follows.
As we really want to calculate cohomology classes, we finally have to make tr explicit

in de Rham cohomology. So let us consider complex coefficients and the situation
71
f : E → B from above, where we assume now B, E, F to be C ∞ -manifolds and V a

C-local system on B. We have R f ∗ ( f −1 V ) = V ⊗ f ∗ (A•E ) and the quasi-ismorphism
V ⊗ Rd f ∗ (C) → V ⊗ Rd f ∗ (C) ⊗ A•B . We need to find a chain map
tr • : V ⊗ f ∗ (A•E ) → V ⊗ Rd f ∗ (C) ⊗ A•B [−d]
such that
tr • / V ⊗ Rd f ∗ (C) ⊗ A• [−d]
V ⊗ f ∗ (A•E )
O B
tr

V ⊗ Rd f ∗ (C)[−d]
= / V ⊗ Rd f ∗ (C)[−d]
commutes. It suffices to find tr • such that

tr • / Rd f ∗ (C) ⊗ A• [−d]
f ∗ (A•E )
O B
tr

Rd φ∗ (C)[−d]
= / Rd φ∗ (C)[−d]
commutes. This is the theory of fiber integration. We assume the fiberbundle f : E → B

to be orientated in the sense of Halperin & Greub (1972). This means that we have a
d-form η on E restricting to a volume form on each fiber of f . We demand additionally
η to be closed. Therefore η defines a trivialization of Rd f ∗ (C).
Definition 18. We define the fiber integration operator
tr • : f ∗ A•E → Rd f ∗ (C) ⊗ A•B [−d]

Z
ω 7→ (−1)d(•−d) η ⊗ vol( F )−1 ω,
F
R
where F ω is defined as in Halperin & Greub (1972) with respect to η and vol( F ) :=
R
F
η.
Remark 3.3.18. • vol( F ) is a well-defined locally constant function, as vol( F )(b) =

R
f −1 ( b )
η is finite and η is closed.
• The factor of −1 is just for the proper sign conventions, as d[ p] = (−1) p d for the
differential of a shifted complex. So tr • is a chain map.
• From tr • η = η ⊗ 1 it follows that tr • has the desired property, in other words
tr = H p ( B, tr • ) : H p ( E, f −1 V ) → H p−d ( B, V ⊗ Rd f ∗ (C))
is the map from Definition 17.
72
4 Comparison with Harder’s Eisenstein
classes
Kings (2008) and Blottière (2009b) have shown that polylogarithmic Eisenstein classes
may restrict non trivially to cohomology classes on pointed neighborhoods of the cusps
of Hilbert-Blumenthal varieties. Restricted to pointed neighborhoods of the cusps poly-
logarithmic Eisenstein classes may be evaluated giving rise to special values of partial
L-functions of the totally real field F.
Harder on the other hand started with cohomology classes near the cusps and con-
structed an operator from the cohomology near the cusps into the cohomology of the
Hilbert-Blumenthal variety, which is a section for the restriction map. The image of this
operator is called the Eisenstein cohomology.
Now we want to represent the polylogarithmic Eisenstein classes by differential forms
so that we finally can compare them with Harder’s Eisenstein cohomology classes.
4.1 Nori’s calculation of the polylogarithm as current
One of the big advantages of the polylogarithm is that it can actually be calculated.
For example, this has been done by Nori (1995) and A. Levin in Levin (2000) with C-
coefficients by using currents. We will follow Nori’s approach. We recall a well known
lemma which justifies the extension of coefficients to C.
4.1.1 Integral cohomology classes
Lemma 4.1.1. Let X be a topological manifold which has a finite good cover U =
(Ui )i=1,...,m , in other words a cover by finitely many open sets each homeomorphic to Rn
such that all finite intersections Ui0 ∩ ... ∩ Ui p are again homeomorphic to Rn . Moreover,
73
4 Comparison with Harder’s Eisenstein classes
we consider V a locally constant sheaf of A-modules on X and A → R a flat ring

extension. Then the natural morphism
H • ( X, V ) ⊗ A R → H • ( X, V ⊗ A R)
is an isomorphism.
Proof. Let U = (Ui ) be a finite good cover of X. We have the the Čech to derived
functor spectral sequence Godement (1973) Théorème 5.4.1.
p,q
E2 = Ȟ p (U, H q (V )) ⇒ H p+q ( X, V ),
with the presheaf H q (V )(V ) = H q (V, V ), V ⊂ X open. Consider Ui0 , ..., Ui p ∈ U. As U

is good, we get that Ui0 ∩ ... ∩ Ui p is homeomorphic to Rn and in particular contractible.
It follows that V|Ui ∩...∩Ui p is constant and therefore
0
H q (V )(Ui0 ∩ ... ∩ Ui p ) = H q (Ui0 ∩ ... ∩ Ui p , V ) = 0, q > 0,
by homotopy invariance of sheaf cohomology with constant coefficients Iversen (1986)

p,q
IV. Theorem 1.1. We conclude C p (U, H q (V )) = 0, if q > 0, and therefore E2 = 0, q 6= 0.
The spectral sequence degenerates and the edge morphisms yield isomorphisms
H p ( X, V ) = Ȟ p (U, H 0 (V )) = Ȟ p (U, V ).
If A → R is a flat ring extension, we see
H p ( X, V ⊗ A R) = Ȟ p (U, V ⊗ A R) = Ȟ p (U, V ) ⊗ A R = H p ( X, V ) ⊗ A R,
as the covering U is finite.
Remark 4.1.2. Let X a topological manifold with a finite good cover and V an A-
local system. By induction on the cardinality of good covers one easily shows us-
ing the Mayer-Vietoris sequence that H • ( X, V ) is a finitely generated A-module. Let
now A ⊂ C be a principal ideal domain. By Lemma 4.1.1 we may embed our co-
homology groups of local A-systems - up to torsion - in the cohomology groups with
C-coefficients. More precisely, H • ( X, V ) decomposes into a direct sum of a free A-
module H • ( X, V )free and a torsion A-module H • ( X, V )tor . We have a natural injection
H • ( X, V )free ,→ H • ( X, V ⊗ A C), isomorphisms
H • ( X, V )free ⊗ A C ∼
= H • ( X, V ) ⊗ A C ∼
= H • ( X, V ⊗ A C)
74
can
and H • ( X, V )free coincides with im( H • ( X, V ) → H • ( X, V ⊗ A C)) inside H • ( X, V ⊗ A C).
We call im( H • ( X, V ) → H • ( X, V ⊗ A C)) the A-integral or simply the integral classes
inside H • ( X, V ⊗ A C). If X is a manifold, integral classes may then be calculated by
differential forms or currents with values in these local systems. In particular, we will
represent the polylogarithm by a current as proposed in Proposition 2.4.19.
4.1.2 Solving the differential equation
Let us consider πW : TW → MK . Let us suppose K = K N and W = V (Ẑ) o K N . To

find a representative for the polylogarithm we may treat each connected component of
MK N separately. So our topological situation is of the form
π : V (R)/V ( g f ) × M /G ( g f ) → M/G ( g f ),

where
∼
∏ SO(2) \ G(R) → M :=
= g
(τ, r ) ∈ H± × Z (R) : im(rτ ) ∈ Z (R)0 , g 7→ (i · g, det( g)),

ν|∞
 
a b
and γ =   ∈ G (R) acts on M by (τ, r )γ := b+dτ
· det(γ) . We write as
a+cτ , r
c d
before V ( g f ) = a ⊕ b as a sum of two fractional ideals. Set L := V ( g f ), Γ := G ( g f ) and
π̃ : T̃ := V (R)/L × M → M the projection. Moreover, we consider R>0 ⊂ Z (R) acting
on M.
By Proposition 2.4.20 we have
!Γ
∞
R ⊗CT̃/Γ
∞ ⊗D
T̃/Γ = q∗ ∏ Sym V (C) ⊗ DT̃
k
,
k ≥0
where q : T̃ → T̃/Γ is the canonical projection. We will construct a Γ-invariant 2g − 1-

current on T̃, which descends to the right current pol( f ) on T̃/Γ. Let us fix coordinates
∼
=
first. We have FR → ∏σ R, x ⊗ r → (σ( x )r )σ and after fixing an ordering of the set of
embeddings σ : F → R we get an isomorphism ∏σ R → Rg . Since V (R) = FR2 , we
g
take the coordinates w = (w1 , w2 ), where wi = (wiσ )σ , i = 1, 2. H± is considered as an
open subspace of the complex manifold FC and we take the induced coordinates τ =
(τσ )σ . The corresponding real coordinates are denoted by ( x, y), i.e. τ = x + iy. The
(FC -valued) differential forms dw1 , dw2 are globally defined on T̃ making the algebra of
75
global differentials on T̃ into a free C ∞ ( T̃ )-module of finite rank. We trivialize or −

M/Γ ⊗ C
1
by choosing the volume form

^ dτ ∧ dτ ^ dr
vol M/Γ := ∧
σ
τ−τ σ σ
r σ
−1
and orR
>0 \ M/Γ
⊗ C by
g
^ dτ ∧ dτ (−1)i−1 dr dr dr
∧∑
c
volM := volR>0 \ M/Γ := ∧ .. .. ∧ ,
σ
τ−τ σ i =1
g r1 ri r g
dr
where b means that we omit r i . We also need an orientation of the fibers of our family
Here we take θ := vol1(T ) σ dw1σ ∧ dw2σ , where the volume is calculated with
V
of tori.
to volV := σ dw1σ ∧ dw2σ . θ defines a Z-orientation on the torus T.
V
respect
We can identify distributions φ on T with 0-currents, when we use our volume form θ:
Given a form η with compact support of top degree on T we may write it uniquely as f θ
with a C ∞ -function f on T. Then we define φ(η ) =: φ( f ). We have, for example, the
distribution δ f = ∑t∈T f (t)δt , where f : T → C is a map of finite support and δt is the
Delta Distribution in t.
Currents φ on T can be considered as linear functionals φ : Ac ( T̃ ) → Ac ( M); compare
de Rham (1955) 2, Theoreme 9. In this manner we can consider δ f θ as a 2g-current on
T̃ with values in forms on M. Given a ω ∈ Γc ( T̃, C T̃∞ ⊗π̃ −1 C M∞ π̃ −1 A M ) we have explicitly
δ f θ (ω ) = δ f (w)(θ ∧ ω ) = ∑ f (t)ω (t, −)

t∈ T
where δ f (w) means that δ f just acts on the coordinates of T. Moreover, we have
Z
1 ⊗ δf θ = ∑ f (t)
{t}× M
,
t∈ T
where 1 ⊗ δ f θ means the tensor product of currents on M and T and integration is

defined with respect to our fixed orientations. In general, all currents on T can be
identified with currents on T̃ with values in forms on M and these can be identified via
integration over M with currents on T̃. So we may identify 1 ⊗ δ f θ with δ f θ.
With this overview at hand we can say in which space we want to look for pol( f ). We
consider the space generated by forms φ ∧ ω, where φ is a 0-current on T and ω a
smooth form on T̃. Then φ ∧ ω (η ) = φ(w)(ω ∧ η ) as currents on T̃ with values in forms
on M. Consider the the pullback of Log∞ to T̃. We identify it with the trivial pro vector
bundle ∏k≥0 Symk V (C) ⊗ C T̃∞ . We have the connection ∇ = d − κ and are interested in
76
solving the equation ∇(φ) = δ f θ of currents on T̃ with values in forms on M, where the
support of f consists of points of order N on the torus T and aug( f ) = ∑t∈T f (t) = 0.
Our arithmetic group Γ guarantees that t : (τ, r ) 7→ (t, τ, r ) defines a torsion section of
π
T̃/Γ → M/Γ so that δ f θ descends to a current on T̃/Γ, which can be identified with the
functional
Z Z
∑ f (t) M/Γ
∗
t =
f
. (4.1)
t∈ T
Currents φ on T have a Fourier expansion. Since each current φ on T can be uniquely

written as differential form with distributional coefficients in the basis dw1 , dw2 , we
may define Fourier expansion coefficientwise. Therefore, it suffices to settle the case,
when φ itself is a distribution. But then we have φ = ∑λ∈ L∗ φλ exp(2πiλ(w)), where
Hom( L, Z) =: L∗ and φλ = φ(exp(−2πiλ(w))). Now we may evaluate φ as
φ( g) = ∑∗ φλ g−λ , with g = ∑∗ gλ exp(2πiλ(w)) ∈ C ∞ (T ).

λ∈ L λ∈ L
For this see Schwartz (1966) VII 1. This gives us the Fourier expansion
δf θ = ∑ ∑ f (t) exp(2πiλ(w − t))θ = ∑ ∗ ∑ f (t) exp(2πiλ(w − t))θ,

λ∈ L∗ t∈ T 06 = λ ∈ L t ∈ T
as aug( f ) = 0. In this spirit let us write φ = ∑λ∈ L∗ φλ exp(2πiλ(w)) as Fourier series

where φλ are smooth forms on M with values in the completed symmetric algebra
tensored with closed invariant forms on T. Applying ∇ we get the differential equation
d(φλ ) + Aλ ∧ φλ = ∑ f (t) exp(2πiλ(−t))θ, λ ∈ L∗ , Aλ := 2πidλ − κ

t∈ T
for each Fourier coefficient. We try to solve this equation in

•
∏ Symk V (C) ⊗ A( M) ⊗C V (C) ∗ ⊂ ∏ Symk V (C) ⊗ AT̃ (T̃ )
^
R=
k k
R p,q is bigraded: R p,q := ∏k Symk V (C) ⊗ A p ( M ) ⊗C V (C)∗ . Let us first con-

L Vq
R=
sider the case λ = 0. We get the equation d(φ0 ) − κ ∧ φ0 = 0, as δ f has 0 zero-Fourier-
coefficient. So we simply set φ0 = 0. We endow V (R) with a symplectic structure:
h., .i : V (R) ∧ V (R) → R, w1 ∧ w2 → Tr (det(w1 , w2 ))
with Tr ( x ⊗ y) := Tr F/Q ( x ⊗ y) = ∑σ σ( x )y for x ⊗ y ∈ FR . Next we fix global (FC -

valued) vector fields on T̃
τe1 − e2 τe1 − e2
u := , u := − ,
τ−τ τ−τ
77
where the tangent bundle of T is trivialized by e1 7→ ∂w1 , e2 7→ ∂w2 . These vector fields
∞
reflect the fact that we have a decomposition into types HZ ⊗ C M/Γ = H0,−1 ⊕ H−1,0
coming from the Hodge theory of the fibers. The two subspaces are lagrangian with
1
respect to h., .iC and we have det(u, u) = τ −τ . Whenever ξ is a local section of HZ ⊗
∞
C M/Γ denote by ξ 0,−1 , ξ −1,0 the decomposition into types. In this spirit u defines a frame
for H−1,0 and u for H0,−1 and since we have
w = w1 e1 + w2 e2 = (w1 + τw2 )u + (w1 + τw2 )u
for w ∈ V (R), we see w−1,0 = (w1 + w2 τ )u. Moreover, the decomposition into types is
Γ-equivariant and we get
 
a b
(γw)−1,0 = γw−1,0 , γu = ( a + cτ )u, γ =  .
c d
Since h., .i is non-degenerate, we may express L∗ by it. If we set d ⊂ O to be the

different-ideal of O , we have an isomorphism
b−1 d−1 ⊕ a−1 d−1 → L∗ , l 7→ hl, .i := λ
of O -modules. We will stick to the vector fields

l −1,0
, 0 6 = l = ( l1 , l2 ) ∈ L ∗ ⊂ V ( R ) .
r
We have
−1,0
l −1,0 det γ−1 γl
γ =
r r
compatible with the natural Γ action on L∗
D E D E
γλ(w) := λ(γ−1 w) = l, γ−1 w = det γ−1 γl, w ,
l −1,0
if λ = hl, .i. Now we follow Noris construction of pol( f ).The vector field r on T̃ defines
a linear operator on the differential algebra R onto itself by contraction. Let us denote
this operator by il . Furthermore, define the operator Cl := d + Al ∧ on R, where Al is the
Aλ from above. Then Cl ◦ Cl = 0, il ◦ il = 0 and Cl ◦ il + il ◦ Cl is an isomorphsim. The
first assertion is immediately clear. For the second write Fl = (dil + il d) + ( Al il + il Al ).
l −1,0
The first summand is the Lie derivative L l −1,0 with respect to the vectorfield r . It is a
r
2g
sum of maps R p,q → R p+1,q−1 and therefore L l −1,0 = 0 and L l −1,0 is nilpotent on R.
r r
( Al il + il Al )(ω ) = Al (il ω ) + il ( Al ∧ ω ) = Al (il ω ) + il ( Al ) ∧ ω − Al ∧ il (ω )
78
by the rules of interior multiplication and therefore
( Al il + il Al )(ω ) = il ( Al ) ∧ ω = il ( Al )ω
is just multiplication by
l −1,0 l −1,0
−1,0
l l −1,0
il ( Al ) = 2πidλ( ) − κ( ) = 2πi l, −
r r r r
a function with values in the completed symmetric algebra of V (C). We calculate

−1,0 −1,0
l 0,−1 l
l, = l , ,
r r
since the Hodge decomposition is langrangian, and endow FC with the natural hermitian
metric induced by Tr to get
l + l τ 2

l −1,0

2
2πi l 0,−1 , = π p1 = Fl (τ, r ).

r rIm(τ )
Note γFl (τ, r ) = Fl (τγ, rdetγ) = Fdet γ−1 γl (τ, r ). Since l 6= 0, we have Fl (τ, r ) ∈ C×
∞ , because the leading
and therefore il ( Al ) is certainly invertible in ∏k Symk V (C) ⊗ C M
coefficient is invertible. It follows that Al il + il Al is invertible.
l −1,0
Now L l −1,0 and il ( Al ) commute, since the vector field r takes its values in the tangent
r
∞ -linear. We conclude that the sum
space of T and therefore its Lie derivative is C M
L l−1,0 + il ( Al ) is also invertible. Explicitly,
r
(il Cl + Cl il )−1 = (L l−1,0 + il ( Al ))−1 =

r
∑ (−1)k il ( Al )−(k+1) Lk− l 1,0
r
.
k ≥0
This is a finite sum. Set ωt,l := exp(−2πiλ(t))θ. Then il (il Cl + Cl il )−1 (ωt,l ) is a solution
for d(φλ ) + Aλ ∧ φλ = Cl (φλ ) = exp(2πiλ(−t))θ. To prove this note that Cl and il Cl +
Cl il commute. Therefore Cl and (il Cl + Cl il )−1 commute and
ωt,l = (il Cl + Cl il )(il Cl + Cl il )−1 (ωt,l ) = Cl il (il Cl + Cl il )−1 (ωt,l ),
since
il Cl (il Cl + Cl il )−1 (ωt,l ) = il (il Cl + Cl il )−1 Cl (ωt,l ) = 0,
as Cl ωt,l = 0. Because il L l −1,0 = il dil we get

r
φl = φλ = ∑ f (t) ∑ (−1)k il ( Al )−(k+1) il (dil )k ωt,l

t∈ T k ≥0
79
So we should have
∑t∈T f (t) exp(2πi hl, −ti)il (dil )k θ

φ= ∑ ∑ (−1)k ! k +1 exp(2πi hl, wi)
l +l τ 2 l −1,0

k ≥0 06 = l ∈ L ∗
√rIm(τ ) − r
1 2
π
The sum over k only has nontrivial summands between zero and 2g − 1. We rewrite
 −(k+1)
l + l τ 2 l −1,0

1 2
π −

p
rIm(τ ) r
as binomial series
√ −2( k + n +1) ⊗n
(k + n)! π ( l + l τ ) l −1,0
∑ k!n! prIm(τ )
1 2
.

n ≥0 r
So in total φ should be given by

⊗n
l −1,0
(k + n)!(−1)k r ∑t∈T f (t) exp(2πi hl, w − ti)il (dil )k θ
∑ ∑ π ( l1 + l2 τ ) 2 ( k + n + 1 )
√
n,k ≥0 06=l ∈ L∗
k!n! √

rIm(τ )
We want to see that this expression defines a current with values in the completed
symmetric algebra. To do so we have to control the coefficients of the Fourier series.
The first part is to control
1
.
π ( l1 + l2 τ ) 2 ( k + n + 1 )
√
√
rIm(τ )
√
π (l +l τ )
Consider √ 1 2 as a continuous map S : V (R) × M → FC , which is FR -linear in
rIm(τ )
the first argument. Take any compact set K in M and K 0 the one-sphere in V (R). The
product is again compact and kSk has a minimal value on K 0 × K. Since S(., τ, r ) is
injective for each (τ, r ), this minimum has to be bigger than zero. Therefore, there is a
constant C > 0 such that kS(v, τ, r )k > C kvk for all 0 6= v ∈ V (R) and (τ, r ) ∈ K. It
follows that
1 1
2( k + n +1) ≤
C k l k ( k + n +1)
2
√
π ( l1 + l2 τ )
√
rIm(τ )
for all l and for any (τ, r ) in a compact set K with C only depending on K. Next we
−1,0 ⊗n
have to consider l r il (dil )k θ. We have to estimate each coefficient in front of an
80
4.2 Global description of the polylogarithm in adelic coordinates
C ∞ ( M)-basis of Symk V (C)⊗A V (C)∗ . Each coefficient can be controlled

Vq
ˆ p ( M ) ⊗C
by C 0 kl kn+k+1 for (τ, r ) in a compact set.
Now we want to see that each coefficient of φ, which is given as a formal Fourier series,
actually converges to a distribution on T̃. But we have just shown that the coefficients
can be controlled by C kl kn+k+1 for all (τ, r ) ∈ K compact. Schwartz (1966) VII.1 tells
us that φ is a current, since our estimates are locally uniform in (τ, r ). The next step is
to prove that the current φ is Γ-invariant. First we have γ∗ φl = φdet γ−1 γl , as
γ∗ il (dil )k θ = idet γ−1 γl d(idet γ−1 γl )k θ,
since γ∗ ◦ il = idet γ−1 γl ◦ γ∗ and θ is Γ-invariant. Therefore

D E
γ∗ ∑ φl exp(2πi hl, wi) = ∑(γ∗ φl ) exp(2πi l, γ−1 w ) =
l l
D E
∑ φdet γ− γl exp(2πi
1 det γ−1 γl, w ).
l
Evaluating on a test form and using locally uniform convergence we conclude that this
equals ∑l φl exp(2πi hl, wi) again. Finally, we have thanks to Nori (1995).
Theorem 4.1.3.
⊗n
l −1,0
(k + n)!(−1)k r ∑t∈T f (t) exp(2πi hl, w − ti)il (dil )k θ
φ= ∑ ∑ π ( l1 + l2 τ ) 2 ( k + n + 1 )
√ ⊗ volM
n,k ≥0 06=l ∈ L∗
√rIm(τ )
k!n!
defines a R>0 × Γ-invariant current with values in the completed symmetric algebra on
T̃ that satisfies ∇(φ) = δ f θ. In particular, φ induces a current on R>0 \ T̃/Γ with values
in Log∞ ⊗ π −1 orR
−1
, which equals pol( f ) as cohomology class.
>0 \ M/Γ
Proof. See Proposition 2.4.19, Proposition 2.4.20 and Equation (4.1).
4.2 Global description of the polylogarithm in adelic

coordinates
We already have an explicit description of the polylogarithm as a current on each con-

nected component of TW . However, the choice of a connected component of MK al-
ways means a choice of a specific lattice L inside V (R), which represents this con-
nected component. We want a description of the polylogarithm, which does not depend
81
on a particular choice of a lattice. To achieve this we use adelic coordinates. We start

by recalling Fourier analysis on V (A).
4.2.1 Fourier analysis
Let us take the skew-symmetric non-degenerate bilinear Form
V (A) × V (A) → A, h x, yi := Tr F/Q (det( x, y))
Let us fix a character ψ0 : A → C× trivial on Q. We define ψ0 = exp(2πiλ) where we

take λ : A → R/Z, λ = ∑ν place of Q λν , with
can.
λ p : Q p → Q p /Z p ⊂ Q/Z ⊂ R/Z,
λ∞ : R → R/Z, x 7→ − x mod Z.
We identify V (A) with its Pontryagin dual by
V (A) → V
\ (A), x → ψ0 h x, i := ψ0 (h x, i).
We need to fix a Haar measure dv on V (A) = A2F . To do so we fix a Haar measure

dx on AF and take dv to be the product measure. We take Haar measures dxν on Fν
for each place ν. If ν|∞, we have Fν = R and we take the Lebesgue measure. If ν| p
1
we fix dxν by dxν (Oν ) = N (dν )− 2 , where dν ⊂ Oν is the different of Oν /Z p . We take
then dx = ∏ν place of F dxν in the sense of Tate, see Tate (1967) XV 3.3. In a similar
manner we get the Haar measure dv on V (A f ) and if we restrict ψ0 to V (A f ), we get
the topological isomorphism
V (A f ) → V\
(A f ), x → ψ0 h x, i .
For any integrable function ϕ : V (A) → C and any g ∈ G (A) we have the transforma-
tion formula
Z Z
−1
ϕ( gv)dv = kdet( g)k ϕ(v)dv,
v ∈V (A) v ∈V (A)
with k k := ∏ν place of F | |ν the Tate-character and similarly for integrable ϕ : V (A f ) → C

Z −1 Z
ϕ( g f v)dv = det( g f ) ϕ
ϕ(v)dv.
v ∈V (A f ) v ∈V (A f )
82
We define the Fourier transform of a Schwartz-Bruhat function ϕ : V (A) → C by

Z
ϕ̂( x ) := ϕ(v)ψ0 h x, vidv.
V (A)
The measure dv is self dual, in other words, the Fourier inversion formula holds: ϕ̂ˆ = ϕ.
There is no twist by −1∗ as in the one dimensional formula. To see this consider
V (A) × V (A) → A, ( x, y) := Tr F/Q ( x1 y1 + x2 y2 ),
use Fubini, the one dimensional Fourier inversion formula,

 
0 −1
h x, yi = ( Jx, y) ,   = J, J 2 = −1, t J = J −1 ,
1 0
−1
ϕ · g)( x ) = kdet( g)k ϕ̂( ĝ−1 x ),
and the transformation formula. We have the formula (\
for ϕ ∈ S(V (A)) and g ∈ G (A). Here ĝ denotes the adjoint operator of g with re-
spect to h , i. It is explicitly given by ĝ = det g · g−1 . Similarly, we get (\
ϕ · g f )( x ) =
−1 −
det( g f ) ϕ̂( ĝ x ), for ϕ ∈ S(V (A f )) and g f ∈ G (A f ).
1

f f
We want to examine ϕ̂ of a Schwartz-Bruhat function ϕ : V (A f ) → C. Let us assume

that supp( ϕ) ⊂ N −1 V (Ẑ) and ϕ is well-defined modulo NV (Ẑ). We may write
ϕ= ∑ ϕ(u)χu ,
u∈ N −1 V (Ẑ)/NV (Ẑ)
where u = u + NV (Ẑ) and χ M is the characteristic function of a measurable set M ⊂

V (A f ). We calculate
Z
ϕ̂( x ) = ∑ ϕ(u)
V (A f )
χu (v)ψ0 h x, vidv
u∈ N −1 V (Ẑ)/NV (Ẑ)
The latter integral is

Z Z Z
χu (v)ψ0 h x, vidv = χu (v + u)ψ0 h x, v + uidv = χu−u (v)ψ0 h x, v + uidv
V (A f ) V (A f ) V (A f )
Z Z
= χ NV (Ẑ) (v)ψ0 h x, v + uidv = ψ0 h x, ui ψ0 h x, vidv.
V (A f ) NV (Ẑ)
R
The last integral is now easy to compute. NV (Ẑ)
ψ0 h x, vidv is zero, whenever ψ0 h x, i
is not the trivial character on NV (Ẑ). This is exactly the case, when x ∈ / N −1 d−1 V (Ẑ).
Otherwise the integral is just NV (Ẑ) dv = ( N 2g d F )−1 . This gives the final formula
R
ϕ̂( x ) = ( N 2g d F )−1 ∑ ϕ(u)ψ0 h x, uiχ N −1 d−1 V (Ẑ) ( x ).

u∈ N −1 V (Ẑ)/NV (Ẑ)
83
4.2.2 The adelic polylogarithm
Let us give the description of the polylogarithm current on TW in global adelic coordi-
nates. First we trivialize the orientation bundle by the section
2g
vol∗T /M −1 0 0
^ ^
:= kdet gk f sgn(N(det g)) Xσ ∧ Yσ ∈ H (M, µ) = H (M, H)
σ
which is the element dual to the form
dw1σ ∧ dw2σ
V
σ volV
volT /M := −
= .
kdet gk f sgn(N(det g)) 1 kdet gk f sgn(N(det g))−1
Here we extended the Norm character N and k k f to the whole of IF by setting N = 1

on the finite Ideles and k.k f = 1 on FR× .
As the level W f varies, this form may not give each fiber volume one. Nevertheless,
if K f ⊂ G (Ẑ) and W f = V (Ẑ) o K f the form volT /M gives each fiber the volume d F .
Suppose we are given a function in S(V (A f ), µ⊗n )0 . Of course, any such function may
be written as
n
(v, g) 7→ kdet( g)k f sgn(N(det( g)))−1 f (v, g) = ϕ(v, g),
with f ∈ S(V (A f ), µ⊗0 )0 such that f factors in the second argument over K f = K N 2 ,
supp( f ( , g)) ⊂ N −1 V (Ẑ) and f ( , g) is well-defined modulo NV (Ẑ) for all g ∈ G (A).
Let us recall the main steps in the calculation of pol( f ) = polW ( f ), W f = NV (Ẑ) o K N 2 .
First of all we did our calculations in the coordinates (w, g), see Remark 3.1.5. This
means that we associated to f on each fiber
V (R)/V (Q) ∩ ( g−1 NV (Ẑ) + V (R)) ∼

= g−1 NV (Ẑ) \ V (A)/V (Q)
the delta distribution
∑ f ( gw, g)δw = ∑ f (v, g)δg−1 v .

w∈ g−1 N −1 V (Ẑ)/g−1 NV (Ẑ) v∈ N −1 V (Ẑ)/NV (Ẑ)
We fixed the lattice L given by V (Q) ∩ ( Ng−1 V (Ẑ) + V (R)) and the dual lattice L∗ =
V (Q) ∩ ( N −1 d−1 ĝV (Ẑ) + V (R)) in V (Q) with respect to h , i. The canonical map
N −2 L/L = g−1 N −1 V (Ẑ) ∩ V (Q)/g−1 NV (Ẑ) ∩ V (Q) → g−1 N −1 V (Ẑ)/g−1 NV (Ẑ)
84
is an isomorphism. We constructed the polylogarithm pol( f ) with respect to

∑w∈ N −2 L/L f ( gw, g)δw as the current
−1,0 ⊗ j
vol
( j + k)!(−1)k f˜(l, g) l r il (dil )k d TN/2gM
∑ ∑∗
F
√ 2( j + k +1) exp(2πi hl, w∞ i) ⊗ volM ,
j,k ≥0 06=l ∈ L ( l + l )
√
π 1 2 τ
j!k!
rIm(τ )
where f˜(l, g) was actually given by

D E
∑ f ( gw, g) exp(2πi hl, −wi) = ∑ f (v, g)ψ0 l, − g−1 v , l ∈ L∗ .
w∈ N −2 L/L v∈ N −1 V (Ẑ)/NV (Ẑ)
On the left-hand side of the last equation we interpret w ∈ N −2 L ⊂ V (R) in the infinite
component, whereas on the right-hand side we have v ∈ N −1 V (Ẑ) ⊂ V (A f ). The
factor ( N 2g d F )−1 appears, because ( N 2g d F )−1 volT /M is the normalized volume form
of the fibers in the situation W f = NV (Ẑ) o K N 2 , which we used for the construction of
pol( f ). If we have Fourier analysis for Schwartz-Bruhat functions on V (A f ) in mind, we
see ( N 2g d F )−1 f˜(l, g) = fˆ( ĝ−1 l, g). Here fˆ means that we do Fourier tranformation in
the first argument. More precisely, ĝ−1 ∗ χ N −1 d−1 V (Ẑ) = χ N −1 d−1 ĝV (Ẑ) so that for l ∈ V (Q)
ĝ−1 ∗ χ N −1 d−1 V (Ẑ) (l ) 6= 0 if and only if l ∈ L∗ . With this notation at hand we may write
pol( f ) as
−1,0 ⊗ j
( j + k)!(−1)k fˆ( ĝ−1 l, g) l r il (dil )k volT /M
∑ ∑ π ( l1 + l2 τ ) 2 ( j + k + 1 )
√ ψ0 hl, w∞ i ⊗ volM .
j,k≥0 l ∈V (Q)\{0}
j!k! √

rIm(τ )
If we multiply with our trivialization volT∗n/M , we may write pol( ϕ) as

−1,0 ⊗ j
( j + k)!(−1)k fˆ( ĝ−1 l, g) l r il (dil )k volT /M ⊗ vol∗Tn/M
∑ ∑ π ( l1 + l2 τ ) 2 ( j + k + 1 )
√ ψ0 hl, w∞ i ⊗ volM .
j,k ≥0 l ∈V (Q)\{0}
j!k! √

rIm(τ )
4.2.3 Analytic continuation of the polylogarithm
If we want to specialize the polylogarithm along the zero section, we need to know,
whether this current may represented by a smooth differential form on U. This is what
Levin proved in Blottière (2009b). Let us quickly recall how things work. We have
identifications O ⊗ R2 ∼
= O ⊗ C, e1 7→ 1, e2 7→ i,
∏ SO(2) \ G(R) → M, g∞ 7→ (ig∞ , det g∞ ) = (τ, r ),

ν|∞
85
√
π (l + l τ ) Γ(s) du
Z
1 2 −1
= π ĝ∞ l , and s = exp(−ux )us , x ∈ R>0 .

p
rIm(τ ) x R>0 u
This yields
√
π ( l + l τ ) −2( j + k +1) Z

−1 2 u j+k+1 du

1 2
= exp(−uπ ĝ∞ l ) .

p
( j + k)! u

rIm(τ ) R>0
Given ϕ f ∈ S(V (A f )) and ϕ∞ ∈ S(V (R)) we define ϕ f ⊗ ϕ∞ ∈ S(V (A)) by
ϕ f ⊗ ϕ ∞ ( v ) : = ϕ f ( v f ) ϕ ∞ ( v ∞ ), v ∈ V (A).
w 7→ exp(−π kwk2 ) is fixed by Fourier transformation, so we have a (V (A f ) o G (A f )) ×

π0 ( G∞ )-equivariant Theta-embedding
φ : S(V (A f )) → S(V (A)), f 7→ f ⊗ exp(−π k x k2 ),

[
which is compatible with Fourier transformation, in other words, φ ( f ) = φ( fˆ). We can
write the polylogarithm as
( j + k)!
∑ j!k!
(−1)k φ( j,k) ⊗ vol∗Tn/M ⊗ volM ,
j,k ≥0
ˆ
√ −1 −1,0 ⊗ j
Z φ( f )( u ĝ l, g) l r u j+k+1 du k
u il ( dil ) volT /M
φ ( j,k)
:= ∑ ( j + k)!
ψ0 hl, w∞ i
l ∈V (Q)\{0} u∈R>0
The first observation is that for k + j + 1 > 2g we already have honest differential forms,
because the sum over l converges locally uniformly. For these big k + j + 1 we may
also put the integral over u ∈ R>0 in front of all sums by uniform convergence. Let us
consider the differential form
√ −1,0 ⊗ j
Z φ( fˆ)( u ĝ−1 l, g) l r us du k
u il ( dil ) volT /M
∑
( j,k)
φs := · ψ0 hl, w∞ i
u∈R>0 l ∈V (Q)\{0} Γ(s)
( j,k)
for complex s ∈ C, Re(s) > 2g. Levin proves by using Poisson-summation that φs has
an analytic continuation as a holomorphic function in s ∈ C on U. This analytic contin-
uation yields the representation of the polylogarithmic current as a differential form on
( j,k)
U, see also Beilinson et al. (2014) Theorem 2.3.6.: pol( ϕ) = ∑ j,k≥0 φj+k+1 ⊗ vol∗T /M .
The additional vol∗T /M and the missing volM are due to the fact that differential forms
resolve the constant sheaf C, whereas the currents resolve the orientation bundle of
TW . Now we may specialize our polylogarithm to get the representation of Eisk ( ϕ) as
√ ⊗k
−φ( fˆ)( u ĝ−1 l, g) l −1,0

du
Z
lim
s→2g+k u∈R>0
∑ k! ( 2g − 1 ) ! r
us il (dil )2g−1 volT /M ⊗ volT∗n/+M
u
1
l ∈V (Q)\{0}
for k ≥ 0 and the limit exists by analytic continuation.
86
4.3 The decomposition Isomorphism as Mellin-Tranformation
Next we are going apply Section 3.3.1 to Eisk ( ϕ). The fiber integral of Eisk ( ϕ) turns
out to be a Mellin-Transformation of a Theta-series. Given this description of Eisk ( ϕ)
and having Wielonsky (1985) in mind the connection to Eisenstein series and Harder’s
Eisenstein classes is immediately apparent. But before we can apply the fiber integral,
dri
we need to express Eisk ( ϕ) by HC
• which is generated by
ri , i = 1, ..., g.
Set h gi := {1, ..., g} considered as an ordered set. If A is any R-algebra (non-commutative),

g
I ⊂ h gi an ordered subset and x ∈ FA = ∏i=1 A, set x I := ∏i∈ I xi and N( x ) := xh gi . If
i ∈ h gi we also write I \ i for I \ {i }, I ∪ i for I ∪ {i } and I c = h gi \ I.
Lemma 4.3.1.
!
g
il (dil )2g−1 volV

l1 + τl2 l1 + τl2 l1 + τl2
(2g − 1)!
=N
r (τ − τ ) ∑∏ d
r (τ − τ )
∧ dτ j ∧
r (τ − τ ) i
dτ i
i =1 i 6 = j j
Proof. We follow the computation of Blottière (2009a) 4.7. Set dw1 + τdw2 =: η 1 ,
η1
τ −τ =: η 2 . We have
g
η1 − η1
η 1 (u) = 1, η 1 (u) = 0, dη 1 = dτ ∧ dw2 = dτ ∧
τ−τ
, volV = ∏ ηi1 ∧ ηi2 .
i =1
2g−1
il (dil )2g−1 volV = il L l −1,0 volV
r
and we calculate
g
∏ ηi1 ∧ ηi2 =
2g−1 2g−1
L l−1,0 volV = L l−1,0
r r i =1
g
(2g − 1)! ∑ ∏ L l−1,0 η j1 ∧ L l−1,0 η j2 ∧ ηi1 ∧ L l−1,0 ηi2 + L l−1,0 ηi1 ∧ ηi2 ,
r r r r
i =1 j 6 = i
as L2l −1,0 ηi1 = L2l −1,0 ηi2 = 0 for i = 1, ..., g and L l −1,0 satisfies the Leibniz rule. Moreover,
r r r
l1 + τl2 l + τl2
L l−1,0 η 1 = (dil + il d)(η 1 ) = d( )+ 1 dτ,
r r r (τ − τ )
l1 + τl2 dτ
L l−1,0 η 2 = (dil + il d)(η 2 ) = il d(η 2 ) =
r r (τ − τ ) τ − τ
and il L l −1,0 ηi1 = il L l −1,0 ηi2 = il ηi2 = 0, i = 1, ..., g. Now we get
r r
g
il (dil )2g−1 volV
= ∑ ∏ L l−1,0 η j1 ∧ L l−1,0 η j2 ∧ il ηi1 ∧ L l−1,0 ηi2 =
(2g − 1)! i =1 j 6 = i r r r
87
g 2
l1 + τl2 l + τl2 l1 + τl2 dτ l1 + τl2
∑∏ d(
r
)+ 1
r (τ − τ )
dτ ∧
r (τ − τ ) τ − τ
∧
r (τ − τ )
dτ i =
i =1 j 6 = i j i
! !
g
d( l1 +rτl2 )
2
l + τl2 dτ l1 + τl2 l1 + τl2
∑∏ τ−τ
+ 1
r (τ − τ ) τ − τ
∧
r (τ − τ )
dτ ∧
r (τ − τ )
dτ i =
i =1 j 6 = i j i
g
l1 + τl2 2

l1 + τl2 l1 + τl2
∑ ∏ d r(τ − τ ) ∧ r(τ − τ ) dτ ∧ r(τ − τ ) dτi
i =1 j 6 = i j i
Lemma 4.3.2.
l1 + τl2 −1 il (dil )2g−1 volV

N =
r (τ − τ ) (2g − 1)!
g
l1 + τl2 l1 + τl2 dτ ∧ dτ dr ∧ dτ
∑ ∑ (−1)| I |
r ( τ − τ ) (i ∪ I )c r ( τ − τ ) I ∪i τ − τ (i ∪ I )c
∧
r I
∧ dτ i
i =1 I ⊂h gi\i
Proof. We have

l1 + τl2 l1 + τl2 dr l1 + τl2 −1
d ∧ dτ = − +d r ∧ dτ
r (τ − τ ) r (τ − τ ) r τ−τ
and
l1 + τl2 − τl2 + τl2

l1 + τl2
d ∧ dτ = d ∧ dτ =
τ−τ τ−τ
l + τl2 dτ ∧ dτ

l1 + τl2 l + τl2
d − l2 ∧ dτ = d 1 ∧ dτ = 1
τ−τ τ−τ (τ − τ ) τ − τ
If we plug these formulas in Lemma 4.3.1, we get the desired formula.
dr
Corollary 4.3.3. The coefficients in front of the invariant forms r I, I ⊂ h gi, of
N(r )2 i l ( dil )2g−1 vol V do not depend on r.
⊗k
g 1 l −1,0
Lemma 4.3.4. We set ω := (τ − τ )u and N(ω ) := ∏i=1 ωi . The projection of k! r
onto the Symk V (C) Z(K) -part is
l1 + τl2 m N(ω )m

N , k = gm,
r (τ − τ ) (m!) g
and zero otherwise.
88
xk
Proof. We set x [k] := k! for x in any Q-Algebra R. We have ( x + y)[k] = ∑kl=0 x [l ] y[k−l ] .
Doing induction on n ∈ N we may easily proof the formula
![k]
n n
∑ xi = ∑ ∏ x [k ] i
i =1 (k i )∈N0n :∑ k i =k i =1
l −1,0 g l −1,0
Write r = ∑ i =1 r i and get
⊗k ⊗[k] !⊗[k] ⊗[ki ]

g g
l −1,0 l −1,0 l −1,0 l −1,0

1
k! r
=
r
= ∑ r i =
g
∑ ∏ r
=
i =1 (k i )∈N0 :∑ k i =k i =1 i
g ⊗ki g ⊗ki
l −1,0

1 1 l1 + τl2
g
∑ ∏ kσ ! r
= ∑g
∏ kσ ! r (τ − τ )
ω
(k i )∈N0 :∑ k i =k i =1 i (k i )∈N0 :∑ k i =k i =1 i
The projection of this element onto the Symk V (C) Z(K) -part are those summands where
the Z (K ) action factors through the norm character. Thus the only summand living in
Symk V (C) Z(K) is the one corresponding to k i = m for i = 1, ..., g, see Remark 3.3.4.
Before we can go on, we need to fix measures again. Let us consider the Ideles IF . If
|κ (pν )| dxν
ν - ∞, we take d× xν := |κ (pν )|−1 | xν |ν
, as Haar measure of the multiplicative group Fν× ,
where dxν is the Haar measure on the additive group Fν , which we already have fixed.
dxν
If ν|∞, we have Fν = R and we take d× xν := | xν |
, where on the right-hand side dxν
is the usual Lebesgue measure on R. These local measures induce global measures
d× x on IF and IF, f and FR× as in Section 4.2.1.
Remark 4.3.5. • Let G be a locally compact group and H ⊂ G a normal and closed
subgroup. Suppose we have fixed (left) Haar measures dg on G and dh on H.
Then there is a uniquely determined (left) Haar measure dgH on G/H such that
Z Z Z
f ( g)dg = f ( gh)dh dgH
G G/H H
holds for all integrable functions f on G.
• If H ⊂ G is an open subgroup and we have fixed a Haar measure dg on G, we

always take on H the restriction of dg as Haar measure on H.
• Let us set RK := d× x (R>0 \ FR∗,0 / det( Z (K ))), where the measure d× x is induced
by the measure dt
t on R>0 , d× x on FR∗,0 and the counting measure on Z (K ).
• In the following calculations the choice of the Haar measures will be clear from
the context.
89
Set
m
l1 + τl2
Θl := N N(ω )m N(r )2 il (dil )2g−1 volT /M ⊗ volT∗n/+M
1
(4.2)
τ−τ
−1
a form constant in r and C (k ) := (m!) g (2g−1)!
.
Lemma 4.3.6. We have

C ( k )2 g φ( fˆ)(t ĝ−1 l, g)| N(t)|2(m+2+ξ ) ×
Z
k
Eis ( ϕ) = lim
ξ →0 RK t∈{±1} g \ Z (R)/Z (K ) l ∈V (Q)\{0}
∑ N(det( g))m+2
d t · Θl
for k = gm ≥ 0.
Proof. The fibers R>0 \ Z (R)0 / det( Z (K )) ∼

= φK−1 (KgG (Q)) have been oriented with
dr
respect to the forms r i where r = det( g∞ ), compare Remark 3.3.10. We have to write
everything in base and fiber coordinates
  √ −1
1 x √ −1 u ĝ (1, τ )
  =: (1, τ ) and u ĝ = p f , r ∈ FR∗,0 .
0 y r | Im(τ )|
Now we use Section 3.3.1 and Corollary 4.3.3 to identify the cohomology class Eisk ( ϕ)
with
√ −1
u ĝ (1,τ )
φ( fˆ) √ f l, g us
C (k) r | Im(τ )| du ×
Z Z
lim
s→2g+k RK r ∈R>0 \ FR∗,0 / det( Z (K ))
∑
u∈R>0 l ∈V (Q)\{0} N(r )m+2 sgn N((det( g)))m+2 u
d r · Θl
R
Here we have already interchanged the limit and r ∈R>0 \ FR∗,0 / det( Z (K ))
. This is possible by
the dominated convergence theorem, as the limit in s gives a continuous function in r,
which is integrable over the compact set R>0 \ FR∗,0 /Z (K )2 . Remember the isomorphism
ϕ : R>0 {±1} g \ Z (R)/Z (K ) → R>0 \ FR∗,0 / det( Z (K )), t 7→ t2 = r, ϕ∗ d× r = 2g−1 d× t.
Set
√
ut−1 ĝ− 1 !
√ f (1, τ )
ψ( ut−1 ) := ∑ φ( fˆ) p
| Im(τ )|
l, g .
l ∈V (Q)\{0}
This allows us to write Eisk ( ϕ) as

√
C ( k )2 g −1 ψ( ut−1 )us du ×
Z Z
lim d t · Θl =
s→2g+k RK t∈R>0 {±1} g \ FR∗ /Z (K ) u ∈ R>0 N(t)2(m+2) sgn N(det( g))m+2 u
C ( k )2 g ψ((vt)−1 )v−2s dv ×
Z Z
lim 2 ( m + 2 ) m + 2
d t · Θl =
s→2g+k RK t∈R>0 {±1} g \ FR∗ /Z (K ) v ∈ R>0 N( t ) sgn N(det( g)) v
90
2s
C ( k )2 g ψ(vt)(v g ) g N(t)2(m+2) dv ×
Z Z
lim d t · Θl =
s→2g+k RK t∈R>0 {±1} g \ FR∗ /Z (K ) v ∈ R>0 sgn N(det( g))m+2 v
Set FR1 := {t ∈ FR∗ : | N(t)| = 1} and get isomorphisms
incl.
{±1} g \ FR1 /Z (K ) → R>0 {±1} g \ FR∗ /Z (K ) and FR1 × R>0 → FR× , (t, v) 7→ tv.
2s
C ( k )2 g ψ(vt)(v g ) g dv
Z Z
k
Eis ( ϕ) = lim dt · Θl =
s→2g+k RK t∈{±1} g \ FR1 /Z (K ) v ∈ R>0 sgn N(det( g))m+2 v
2s
C ( k )2 g ψ(t)| N(t)| g
Z
lim d× t · Θl =
s→2g+k RK t∈{±1} g \ FR∗ /Z (K ) sgn N(det( g))m+2
C ( k )2 g φ( fˆ)(t ĝ−1 l, g)| N(t)|2(m+2+ξ ) ×
Z
lim
ξ →0 RK t∈{±1} g \ FR∗ /Z (K ) l ∈V (Q)\{0}
∑ N(det( g))m+2
d t · Θl
The reparametrization of the limit is possible as our analytic continuation is a continuous

function is s.
Remark 4.3.7. From now on we assume K f is so small that Z (K ) ⊂ Z (R)0 . Then we

may write
C ( k )2 g φ( fˆ)(t ĝ−1 l, g) N(t)2(m+2+ξ ) d× t

Z
Eisk ( ϕ) =
R K ξ →0
lim ∑
t∈ Z (R)0 /Z (K ) l ∈V (Q)\{0} N(det( g∞ ))m+2
· Θl .
The next step is to interpret this integral as a global Tate-integral. To do this we decom-
pose Eisk ( ϕ) into several summands.
There is a short exact sequence
1 → Z (R)0 /Z (K )+ → K Zf \ Z (A)/Z (Q) → Cl FK → 1,
with K Z := K ∩ Z (A) and Cl FK = K Zf Z (R)0 \ Z (A)/Z (Q) (we always suppose K Zf =

det(K f )) and Z (K )+ := Z (K ) ∩ Z (R)0 . Denote by Cl
dK the Pontryagin dual of the last
F
group and let hK be its cardinality. Moreover, we define
n o
dK : χ(t) = sgn(N(t))m , t ∈ Z (R) .
dK (m) := χ ∈ Cl
Cl F F
Remark 4.3.8. Recall the Iwasawa-decomposition on G0 (R) = G0 ( Fν ) for ν|∞:

 
a b
g=  = k( g) a( g)n( g) :
c d
   
1 a −c cos(θ ) −sin(θ )
k( g) = √  =  ∈ SO(2),
a2 + c2 c a sin(θ ) cos(θ )
91
√   
a2 + c2 0 t1 0
a( g) =  =  ∈ T0 (R),
det( g)
0 √ 0 t2
a2 + c2
   
ab+dc
1 a2 + c2 
1 x
n( g) =  =  ∈ U0 (R).
0 1 0 1
We can consider k, a, n as continuous functions G ( Fν ) → C. Moreover, we set b := an.
One also has a Iwasawa-decomposition for g ∈ G0 ( Fν ) and ν a finite place:

 
t1 t1 x
g = kb with k ∈ G0 (Oν ) and b =   ∈ B0 ( Fν ).
0 t2
This decomposition is not unique, as b is just well-defined up to elements from B0 (Oν ).

Nevertheless, the Iwasawa-decomposition often suffices to define functions on G0 ( Fν ).
For example, if χ : Fν× → C× is an unramified quasi-character, say χ( x ) = | x |sν for
s ∈ C, then we have well-defined continuous functions G0 ( Fν ) → C× , g 7→ χ(t1 ) or
χ(t2 ). Finally, we get the Iwasawa-decomposition for g ∈ G (A). Of course, it is defined
place by place and is given by
   
a b t1 t1 x
  = g = kb with k ∈ G (Ẑ) · ∏ SO(2) and b =   ∈ B (A).
c d ν|∞ 0 t2
Again, this decomposition is just well-defined up to elements from B(Ẑ). For the infinite
( aν +icν )
places ν|∞ we have functions θν : G0 ( Fν ) → R/2πZ defined by eiθ
ν := e
iθν = √
2 2
.
aν +cν
For i ∈ h gi and I ⊂ h gi \ i we define the form
(−1)| I | (2g − 1)! N(e−(m+1)iθ )e− iθ iθ

I ∪i e ( I ∪i )c
Θ I,i := ·
(−2i )k+2g kdet gk f sgn(N(det g))−1 N(t2 )m+2
dτ ∧ dτ dr ∧ dτ
N( ω ) m ∧ ∧ dτ i ⊗ volT∗n/+M
1
(4.3)
τ − τ h gi\(i∪ I ) r I
We have Schwartz-Bruhat-functions
φ( fˆ)m ˆ 1 2 m +1 1
I ∪i ( v, g ) : = φ ( f )( v, g ) N( v∞ + iv∞ ) (v∞ + iv2∞ ) I ∪i (v1∞ − iv2∞ )( I ∪i)c
on V (A). Write φ( fˆ)m ˆ m ˆ m ˆ m ˆ dK

I ∪i = φ ( f ) I ∪i,∞ ⊗ φ ( f ) I ∪i, f with φ ( f ) I ∪i, f = f . For χ ∈ Cl F and
s ∈ C we may define differential forms
Z
∑
s ×
EiskI,i ( ϕ, χ, s) = φ( fˆ)m
I ∪i ( tgγe , g ) χ (det( g ) t ) kdet( g ) t k d t · Θ I,i .
1
γ∈ G (Q)/B(Q)) t∈ Z (A)
92
Lemma 4.3.9. If Re(s) > 2, EiskI,i ( ϕ, χ, s) is an honest differential form and a holomor-
phic function in s. EiskI,i ( ϕ, χ, s) has an analytic continuation as meromorphic function in
s to the whole complex plane and does not have a pole at s = 2. In particular, the limit
lim EiskI,i ( ϕ, χ, m + 2 + 2ξ ) =: EiskI,i ( ϕ, χ)

ξ →0
exists and gives a well-defined differential form on MK .
Proof. The first statement is Wielonsky (1985) III Proposition 6. Wielonsky considers
functions in the variable det(t), whereas we have the variable t. Therefore Wielonsky
has the statement for σ > 1 and we have our statement for Re(s) > 2. The second
( fˆ)m (0) = 0.
statement is Wielonsky (1985) III Proposition 9, as we have φ\ I ∪i
Let us set κ F := d× t(R>0 \ Z (A)/Z (Q)).
Theorem 4.3.10. Suppose we are given a function in ϕ ∈ S(V (A f ), µ⊗n )0 . Write

n
(v, g) 7→ kdet( g)k f sgn(N(det( g)))−1 f (v, g) = ϕ(v, g),
with f ∈ S(V (A f ), µ⊗0 )0 such that f factors in the second argument over K f with
Z (K ) ⊂ Z (R)0 . The polylogarithmic Eisenstein class associated to ϕ
Eisk ( ϕ) ∈ H 2g−1 (M, Symk H ⊗ µ⊗n+1 ) = H p (S , Symk H0 ⊗ µ⊗n+1 ) ⊗ Hq

M
p+q=2g−1
may be represented by the differential form

g
2C (k)
κ F i∑ ∑ ∑ EiskI,i ( ϕ, χ)
=1 I ⊂h gi\i dK (m)
χ∈Cl F
Proof. Define sgn(r ) := (sgn(r )ν )ν|∞ := (sgn(rν ))ν|∞ , r ∈ FR× and S : V (R) → FC ,
v = (v1 , v2 ) 7→ v1 + iv2 . For g ∈ G (R) we have S(k ( g)v) = eiθ ( g) S(v) and
 
− 1 − 1 1 1 x
sgn(det( g))k ( ĝ−1 ) = kd ( g) , b( ĝ−1 ) = sgn(det( g))bd
( g) = √  .
ry 0 y
Therefore
sgn(N(r ))m+2 ˆ m
N(e−(m+1)iθ )e− iθ iθ
I ∪i e ( I ∪i )c √ φ( f ) I ∪i ( ĝ−1 v, g) =
N( ry)m+2
m +1
v1∞ + τv2∞ v1∞ + τv2∞ v1∞ + τv2∞

φ( fˆ)( ĝ−1 v, g) N .
ry ry I ∪i ry ( I ∪i )c
93
√
We see sgn(r ) ry = sgn(r )|t2 | = t2 and with Lemma 4.3.2, the definition of Θl (Equa-
tion (4.2)) and Θ I,i (Equation (4.3)) we conclude for t ∈ Z (R)0
g
∑ ∑ φ( fˆ)m −1
I ∪i ( t ĝ l, g )| N( t )|
m+2+2ξ
Θ I,i = φ( fˆ)(t ĝ−1 l, g) N(t)2(m+2+ξ ) N(r )−(m+2) Θl .
i =1 I ⊂h gi\i
We express the characteristic function of Z (R)0 /Z (K )+ ⊂ K Zf \ Z (A)/Z (Q) by h− 1

K ∑χ∈Cl
dK χ
F
by considering a character χ as a function on Z (A). We get
d× t(K Zf ) RK hK
Eisk ( ϕ) =
2g C (k)
g Z
∑ ∑ ∑ lim
ξ →0
∑
t∈ Z (A)/Z (Q) l ∈V (Q)\{0}
φ( fˆ)m −1
I ∪i ( t ĝ l, g ) χ ( t ) k t k
m+2+2ξ ×
d t · Θ I,i .
i =1 I ⊂h gi\i χ∈Cl
dK
F
Using short exact sequences one calculates 2−( g−1) d× t(K Zf ) RK hK = κ F . For Re(ξ ) > 0
the inner sum converges absolutely and locally uniformly. Therefore we may rearrange
summation ∑l ∈V (Q)\{0} = ∑l ∈V (Q)\{0}/Z(Q) ∑z∈Z(Q) , use the bijection G (Q)/B(Q) →
V (Q) \ {0} /Z (Q), γ 7→ γe1 and interchange Z(A)/Z(Q) and ∑γ∈G(Q)/B(Q) . We get in
R
total
Z
lim
ξ →0 t∈ Z (A)/Z (Q) l ∈V (Q)\{0}
∑ φ( fˆ)m −1
I ∪i ( t ĝ l, g ) χ ( t ) k t k
m+2+2ξ ×
d t · Θ I,i =
Z
lim ∑ φ( fˆ)m −1 1
I ∪i ( t ĝ γe , g ) χ ( t ) k t k
m+2+2ξ ×
d t · Θ I,i =
γ∈ G (Q)/B(Q) t∈ Z (A)
ξ →0
Z
∑
m+2+2ξ ×
lim φ( fˆ)m 1
I ∪i ( tgγe , g ) χ (det( g ) t ) kdet( g ) t k d t · Θ I,i .
γ∈ G (Q)/B(Q) t∈ Z (A)
ξ →0
For this see also Wielonsky (1985) III Proposition 7. So we already have
g
2C (k )
κ F i∑ ∑ ∑
k
Eis ( ϕ) = EiskI,i ( ϕ, χ).
=1 I ⊂h gi\i dK
χ∈Cl F
To prove the theorem it suffices to show that EiskI,i ( ϕ, χ) = 0, whenever χ ∈ dK (m). So

/ Cl F
g
/ Cl (m) and an e ∈ {±1} ⊂ Z (R) with χ(e) sgn(N(e)) 6= 1. Then
take a χ ∈ d K
F
m
Z
×
φ( fˆ)m 1 ξ
I ∪i ( tgγe , g ) χ (det( g ) t ) kdet( g ) t k d t =
t ∈ Z (A)
Z
×
φ( fˆ)m 1 ξ
I ∪i ( etgγe , g ) χ ( e det( g ) t ) k e det( g ) t k d t =
t ∈ Z (A)
Z
ξ ×
χ(e) sgn(N(e))m φ( fˆ)m 1
I ∪i ( tgγe , g ) χ (det( g ) t ) kdet( g ) t k d t,
t ∈ Z (A)
94
4.4 Cohomology of the boundary
since
φ( fˆ)m 1 m ˆ m 1
I ∪i ( etgγe , g ) = sgn(N( e )) φ ( f ) I ∪i ( tgγe , g ).
So we conclude
Z
ξ ×
φ( fˆ)m 1
I ∪i ( tgγe , g ) χ (det( g ) t ) kdet( g ) t k d t = 0
t ∈ Z (A)
and therefore EiskI,i ( ϕ, χ) = 0.
As Harder’s Eisenstein classes are by definition determined by their restriction to the

cohomology of the boundary of the Borel-Serre compactification of SK , we need a thor-
ough understanding of the cohomology of the boundary to compare the polylogarith-
mic Eisenstein classes with those of Harder. In this section we quickly recall Harder’s
description of the cohomology of the boundary as a module induced from B(A f ) to
G (A f ).
From now on we consider Q-coefficients, but we will emphasize, when things can be
defined integrally. We introduce the spaces ∂SK := K \ G (A)/B(Q) and ∂MK :=
K1 \ G (A)/B(Q). SK has the Borel-Serre compactification SK . It is a manifold with
boundary and we have that the boundary SK \ SK is homotopy equivalent to ∂SK , see
Harder (1987) 2.1. Therefore, we refer to H • (∂SK , Symk H0 ⊗ µ⊗n ) as the cohomology
of the boundary.
The natural map jSK : ∂SK → SK induces by pullback a restriction map on cohomology
resSK : H • (SK , Symk H0 ⊗ µ⊗n ) → H • (∂SK , Symk H0 ⊗ µ⊗n )
and performing the colimits over K these maps glue to G (A f ) × π0 ( G∞ )-equivariant

maps resS : H • (S , Symk H0 ⊗ µ⊗n ) → H • (∂S , Symk H0 ⊗ µ⊗n ) of the colimits of the
groups above .
Analogously we define resM : H • (M, Symk H0 ⊗ µ⊗n ) → H • (∂M, Symk H0 ⊗ µ⊗n ). The
cohomology groups of the boundary are induced G (A f ) × π0 ( G∞ )-modules. More
precisely, set K B := K ∩ B(A), K1,B := K1 ∩ B(A) and get the natural inclusions
SKB := K B \ B(A)/B(Q) → ∂SK and MKB := K1,B \ B(A)/B(Q) → ∂MK . SKB ⊂ ∂SK is
95
an inclusion of several connected components of ∂SK . As usual we set
lim H • (SKB , Symk H0 ⊗ µ⊗n ) =: H • (S B , Symk H0 ⊗ µ⊗n ).

−→
K
These groups are B(A f ) × π0 ( B(R))-modules and we have
G (A )×π0 ( G∞ )
Ind B(A f )×π H • (S B , Symk H0 ⊗ µ⊗n ) = H • (∂S , Symk H0 ⊗ µ⊗n )
f 0 ( B (R))
as G (A f ) × π0 ( G∞ )-modules (Harder (1982), p. 117.).
Next consider W0 = Ga ⊂ V0 = G2a embedded in the first component. If g f ∈ G (A f )

is given, we define W ( g f ) := W (Q) ∩ V ( g f ). W (Q) is a B(Q) invariant submodule of
V (Q) and W ( g f ) is a B( g f ) := G ( g f ) ∩ B(Q) invariant submodule. Using multiplication
we get the B( g f )-invariant submodule
Symk V ( g f ) PD ∩ Symk−1 V (Q) · W (Q) ⊂ Symk V ( g f ) PD
and the B( g f )-module
Symk V ( g f ) PD /(Symk V ( g f ) PD ∩ Symk−1 V (Q) · W (Q))
We set again A := Z[ Nd1 F ] and R := O FHil [ Nd1 F ]. Using the faithfully flat ring extension
A → R and the techniques provided by Remark 3.3.4 and Remark 3.3.6 we see that
Z (K )
SymkA V ( g f ) PD /(SymkA V ( g f ) PD ∩ Symk−1 V (Q) · W (Q)) Z(K)
is a free A-module of rank one generated by the residue class of

 
t1 t1 u
kt2 k f ∏ Yσ , g f = k 
[m]  , k ∈ G (Ẑ),
σ 0 t2
if k = gm and zero otherwise. In particular, the B( g f )-action factors over T ( g f ) the

image of B( g f ) in T (Q).
Let us consider the locally constant sheaf ω k on ∂SK , which is associated to the T (Q)-
module Symk V (Q) Z(K) /(Symk−1 V (Q) · W (Q)) Z(K) . It has an integral structure ω kPD
described above. We have the natural projection map Symk H0 → ω k of sheaves on SKB .
Set K T := image of K B in T (A).
The natural map B → T induces a fiber bundle
p : SKB → K T \ T (A)/T (Q) =: SKT
96
with fiber p−1 (1) = K ∩ U (A) \ U (A)/U (Q), which is a topological torus of dimension
g. As described in Section 3.3.1 we have the trace morphism
tr
Rp∗ (Symk H0 ⊗ µ⊗n ) → Rp∗ (ω k ⊗ µ⊗n ) → R g p∗ (Q) ⊗ ω k ⊗ µ⊗n [− g].
Here we used the projection formula and that ω k is pullback by p. Of course, the trace
morphism is defined integrally. Now we have the following result due to Harder.
Theorem 4.4.1. The trace morphism
H p (SKB , Symk H0 ⊗ µ⊗n ) → H p− g (SKT , R g p∗ (Q) ⊗ ω k ⊗ µ⊗n ), p ≥ g,
is an isomorphism.
Proof. By faithfully flatness and Lemma 4.1.1 we may extend coefficients from Q to Q.
In this situation see Harder (1987) II. Important is Harder (1987) 2.8 for the occurring
weights.
Harder gives a complete description of the cohomology group on the right-hand side,
what we will recall now. First we trivialize the sheaf R g p∗ (Q). We fix an ordering
Σ = σ1 , ..., σg . We write elements

 
t1 t1 x
b=  ∈ B(R) = B0 ( F ⊗Q R)
0 t2
and x = ( x1 , ..., x g ) ∈ F ⊗Q R = Rg . Set

dx1 ∧ ... ∧ dx g
volS B /S T (t) := √ .
d F t2 t1 sgn(N(t2−1 t1 ))−1
−1
f
The form is T (Q) invariant, closed and gives each fiber p−1 ( p(b)) a rational volume,
therefore we may interprete volS B /S T ∈ H 0 (SKT , R g p∗ (Q)). If we have K = K N , then
N − g volS B /S T gives each fiber exactly volume one and defines an integral trivialization.
For k = mg we have
N( ω ) m
kt2 kmf sgn(t2 )−m := kt2 kmf sgn(t2 )−m ∏(−Yσ )[m] ∈ H 0 (SKT , ω kPD )
(m!) g σ
in the cohomology with A-coefficients. It defines an A-integral trivialization of ω kPD .

Finally,
kdet( g)k f sgn(N(det( g)))−1 volV∗ = vol∗T /M , volV∗ :=

^
Xσ ∧ Yσ
σ
97
is a trivialization of µ. If we have K = K N , then it even defines a trivialization over the

ring A, as one sees by considering the volumes of the fibers of TW → MK N , W f =
V (Ẑ) o K N .
Suppose K T = Z (A) ∩ K × Z (A) ∩ K and set Z (K )0 = Z (K ) ∩ Z (R)+ . We have a
short exact sequence of groups
t7→(s,t)
1 → Z (R)0 /Z (K )+ → SKT → π0 (SKT ) = K T T (R)0 \ T (A)/T (Q) → 1.
The cohomology of a connected component of SKT is therefore isomorphic to the coho-

mology of Z (R)0 /Z (K )+ . Because Z (K )+ acts properly discontinuously and fixpoint
free on Z (R)0 , we get H • ( Z (R)0 /Z (K )+ , Z) = H • ( Z (K )+ , Z). We may interprete
these classes as invariant forms on SKT as described in Remark 3.3.10 and as such
they span a free Z-submodule H• ( T/Z )K ⊂ H • (SKT , Z) isomorphic to H • ( Z (K )+ , Z).
We set H• ( T/Z ) := H• ( T/Z )K ⊗Z Q as this space does not depend on the level K.
For m ∈ N0 and n ∈ Z we fix characters
γm,n : T (R) → R× , (t1 , t2 ) 7→ N(t2 )−(m+n+1) N(t1 )−(n−1) .
Harder considers algebraic Hecke characters of type γm,n (Harder (1987) (2.5.2)), in
other words, continuous homomorphisms
φ : T (A)/T (Q) → C× , φ|T (R)0 = γm,n

−1
| T (R)0
,
to decompose the cohomology of the boundary. We have φ|T (R) = γm,n sgn(φ) for a
certain character sgn(φ) : T (R)0 \ T (R) → {±1}. We set φ̃ f := φ f · sgn(φ), where
as usual φ f = φ|T (A f ) . Then φ̃ f ( xa) = φ̃ f ( x )γm,n ( a) for all a ∈ T (Q) and x ∈ T (A).
Moreover, φ̃ f takes its values in Q and therefore Gal(Q/Q) acts on the set of φ̃ f by
σ · φ̃ f ( x ) := σ (φ̃ f ( x )), σ ∈ Gal(Q/Q), x ∈ T (A).
Theorem 4.4.2. If k = mg and p ≥ g, we have an isomorphism of Gal(Q/Q)-modules
Q · φ̃ f ⊗ H p ( T/Z ) ∼
= H p− g (S B , Symk H0 ⊗ µ⊗n ⊗ Q)
M
φ: type(φ)=γm,n ,φ̃ f |Z(R)=1
∗n
ψ · N(ω )m ⊗ volV dx1 ∧ ... ∧ dx g
ψ ⊗ η 7→ √ ∪η
dF
Proof. Harder (1987) Theorem 1.
98
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
Remark 4.4.3. Harders Q-structure on cohomology, compare Harder (1987) 1.3, 2.4
and (2.7.1), agrees with our natural Q-structure on H • (SK , Symk H0 ⊗ µ⊗n ). The point is
that Harders Q-structure on cohomology is H • (SK , Symk H0 ⊗ µ⊗n ⊗ Q)Gal(Q/Q) , where
the Galois action comes from functoriality, compare Remark 3.3.6 and Harder (1987)
1.3, 1.4. We have by Lemma 4.1.1
Gal(Q/Q)
H • (SK , Symk H0 ⊗ µ⊗n ⊗ Q)Gal(Q/Q) = H • (SK , Symk H0 ⊗ µ⊗n ) ⊗ Q =
H • (SK , Symk H0 ⊗ µ⊗n ).
Remark 4.4.4. Even though Harder started with more general local systems than Symk H0 ⊗
µ⊗n , see Harder (1987) 1.4, he does not obtain more cohomology groups in degree
p = g, ..., 2g − 1 on the boundary than we do. Harder (1987) (2.8.2) tells us that we get
all weights occuring in cohomological degrees p = g, ..., 2g − 1.
4.5 Restriction of polylogarithmic Eisenstein classes to the

boundary
The next step is to consider the restriction of our polylogarithmic Eisenstein classes to
the boundary. We have the isomorphism of groups
 
1 x
A F → U (A), x 7 →  =u
0 1
inducing on U (A) a Haar measure du corresponding to our Haar measure dx on AF .
Proposition 4.5.1. The cohomology class
resS (Eisk ( ϕ)) ∈ H p (∂S , Symk H0 ⊗ µ⊗n+1 ) ⊗ Hq

M
p+q=2g−1
for ϕ ∈ S(V (A f ), µ⊗n )0 may be represented by the differential form

g
2C (k)
Z
κ F i∑ ∑ ∑
m +2 ×
φ( fˆ)m 1
I ∪i ( tge , g ) χ (det( g ) t ) kdet( g ) t k d t · Θ I,i
=1 I ⊂h gi\i dK (m) t ∈ Z (A)
χ∈Cl F
Proof. Let us set for a moment

Z
m+2+2ξ ×
FI ∪i (χ, ξ, g) := φ( fˆ)m 1
I ∪i ( tge , g ) χ (det( g ) t ) kdet( g ) t k d t.
t ∈ Z (A)
99
By Theorem 4.4.2 and Theorem 4.4.1 we know that we may perform fiber integration,
see Section 3.3.1, to simplify our cohomology classes. The connected components of
∂SK are fiber bundles and the fiber passing through g ∈ G (A) is
U∞ /g f K f g−
f ∩ U (Q) = gKg
1 −1
∩ U (A) \ U (A)/U (Q) ,→ K \ G (A)/B(Q) = ∂SK ,
u 7→ gu. As we have
g
2C (k )
κ F i∑ ∑ ∑
Eisk ( ϕ) = EiskI,i ( ϕ, χ),
=1 I ⊂h gi\i dK (m)
χ∈Cl F
we have to do fiber integration for each EiskI,i ( ϕ, χ) and for those we have
Z
du( g f K f g− 1 −1
f )
u∈U∞ /g f K f g−
lim ∑ FI ∪i (χ, ξ, tg f k ∞ a∞ u∞ γ)du · Θ I,i ,
f ∩U ( Q )
1 ξ →0
γ∈ G (Q)/B(Q))
where we have used the Iwasawa-decomposition of g∞ = k ∞ a∞ n∞ ∈ G (R) and that

the forms Θ I,i are constant in u direction. If we use translation invariance of du, we may
write the last integral as
Z
lim
u∈U (A)/U (Q) ξ →0 γ∈ G (Q)/B(Q))
∑ FI ∪i (χ, ξ, tguγ)du · Θ I,i .
By dominated convergence we may interchange the limit and integration, as the limit is a
continuous function in u and therefore integrable over the compact space U (A)/U (Q).
We obtain
Z
lim
ξ →0
∑
u∈U (A)/U (Q) γ∈ G (Q)/B(Q))
FI ∪i (χ, ξ, guγ)du · Θ I,i .
We write our integral as a sum

Z
lim FI ∪i (χ, ξ, gu)du · Θ I,i +
ξ →0 u∈U (A)/U (Q)
Z
lim
ξ →0
∑
u∈U (A)/U (Q) 16=γ∈ G (Q)/B(Q))
FI ∪i (χ, ξ, guγ)du · Θ I,i .
The first summand does not depend on u, as ue1 = e1 , so it equals

Z
m +2 ×
φ( fˆ)m 1
I ∪i ( tge , g ) χ (det( g ) t ) kdet( g ) t k d t · Θ I,i
t ∈ Z (A)
and this Tate-integral exists by Wielonsky (1985) III Proposition 5. To treat the second
summand we consider the bijection
U (Q) → {γB(Q) ∈ G (Q)/B(Q) : γ ∈

/ B(Q)}
100
     
1 α α −1 0 −1
u=  7→ uJ =  , J =  
0 1 1 0 1 0
and obtain the integral
Z Z
m+2+2ξ ×
lim φ( fˆ)m 1
I ∪i ( tguJe , g ) χ (det( g ) t ) kdet( g ) t k d tdu · Θ I,i .
ξ →0 u ∈U ( A ) t ∈ Z (A)
This integral is zero by the following lemma and our proposition is proved.
Lemma 4.5.2.
Z Z
m+2+2ξ ×
lim φ( fˆ)m 1
I ∪i ( tguJe , g ) χ (det( g ) t ) kdet( g ) t k d tdu = 0
ξ →0 u ∈U ( A ) t ∈ Z (A)
Proof. By Fubinis theorem we may switch the order of integration and obtain the integral
Z Z
m+2+2ξ
lim φ( fˆ)m 1
I ∪i ( tguJe , g ) χ (det( g ) t ) kdet( g ) t k dud× t =
ξ →0 t ∈ Z (A) u ∈U ( A )
Z Z
m+2+2ξ
lim φ( fˆ)m 1
I ∪i | g ( tuJe , g ) χ (det( g ) t ) kdet( g ) t k dud× t,
ξ →0 t ∈ Z (A) u ∈U ( A )
where as usual ψ| g (v) = ψ( gv) for a Schwartz-Bruhat-function ψ ∈ S(V (A)). The ad-
vantage of this integral is that we may interpret the inner integral as a Fourier transfor-
mation, whereas
 theexterior integral may be examined with the theory of Tate-integrals.
1 x
We had u =   and get
0 1
Z Z
m+2+2ξ
lim φ( fˆ)m
I ∪i | g (( tx, t ), g ) χ (det( g ) t ) kdet( g ) t k dxd× t =
ξ →0 t ∈ Z (A) x ∈A F
Z Z
lim k t k −1 φ( fˆ)m
I ∪i | g (( x, t ), g ) χ (det( g ) t ) kdet( g ) t k
m+2+2ξ
dxd× t =
ξ →0 t ∈I F x ∈A F
Z Z
χ(det( g)) kdet( g)km+2 lim φ( fˆ)m
I ∪i | g (( x, t ), g ) χ ( t ) k t k
m+1+2ξ
dxd× t
ξ →0 t ∈I F x ∈A F
To examine the integral

Z Z
m+1+2ξ
lim φ( fˆ)m
I ∪i | g (( x, t ), g ) χ ( t ) k t k dxd× t
ξ →0 t ∈I F x ∈A F
we treat the finite and the infinite places separately. Let us start with the finite part. It is
of the form
Z Z
fˆ| g (( x, t), g)dx ktk1+m+2ξ χ(t)d× t.
t∈IF, f x ∈AF, f
101
We define the Schwartz-Bruhat function

Z
ϕ : AF, f → C, ϕ(u) := fˆ| g (( x, u), g)dx.
x ∈AF, f
We denote the Fourier transform of Schwartz-Bruhat functions ϕ : AF → C by

Z
Pϕ(v) := φ( x )ψ0 ( Tr F/Q ( xv))dx,
x ∈A F
where ψ0 : A/Q → C× is our fixed non-trivial character. We see

Z Z Z
Pϕ(0) = fˆ| g ((u, x ), g)dudx = fˆ| g (v, g)dv =
x ∈AF, f u∈AF, f v ∈V (A f )
kdet( g)k−f 1 fˆˆ(0, g) = kdet( g)k−f 1 f (0, g) = 0.

p(m)
We add an appropriate Euler-factor at infinity. For a place ν|∞ we set ϕν (u) :=
p(m)
exp(−πu2 ), if 2|m, and ϕν (u) := exp(−πu2 )u, if 2 - m. We define
∏ R → C, (uν ) 7→ ∏ ϕν
p(m) p(m)
ϕ∞ : F ⊗Q R = ( u ν ),
ν|∞ ν|∞
p(m)
ϕ p(m) : AF → C, u = (u f , u∞ ) 7→ ϕ∞ (u∞ ) ϕ(u f ),
and consider the Tate-integral lim t∈IF ϕ p(m) (t)χ(t) ktkm+1+2ξ d× t. The integral exists
R
ξ →0
in any case, since Pϕ p(m) (0) = 0 and we do not have a pole at m + 1 + 2ξ = 1, see
Tate (1967) Main Theorem 4.4.1. Moreover, the integral over the infinite places exists
and is non-zero as computed in Tate (1967) 2.5. We conclude that the finite integral
lim t∈I ϕ(t)χ(t) ktkm+1+2ξ d× t exists.
R
ξ →0 F, f
So it suffices to to show that

Z Z
m+1+2ξ
lim φ( fˆ)m
I ∪i | g,∞ (( x, t ), g ) χ ( t ) k t k dxd× t = 0.
ξ →0 t∈ FR× x ∈ FR
We transform the integral back and and have to show the vanishing of
Z Z
m+2+2ξ
lim φ( fˆ)m
I ∪i | g,∞ (( tx, t ), g ) χ ( t ) k t k dxd× t =
ξ →0 t∈ FR× x ∈ FR
Z Z
m+2+2ξ ×
2g lim φ( fˆ)m
I ∪i | g,∞ (( tx, t ), g ) k t k d tdx.
ξ →0 x ∈ FR t∈ FR∗,0
dK (m). We use again the Iwasawa-

The last equality holds, since we have χ ∈ Cl F
decomposition for g ∈ G (R) and calculate the last integral as
−iθ
2g N(e(m+1)iθ )eiθ
I ∪i e ( I ∪i )c
Z Z
m+2+2ξ ×
lim φ( fˆ)m
I ∪i,∞ ( t ( x, y ), g ) k t k d t · dx,
N ( t 1 ) m +2 ξ →0 x ∈ FR t∈ FR∗,0
102
t2
with y = t1 . To end the proof we show
Z Z
m+2+2ξ ×
lim φ( fˆ)m
I ∪i,∞ ( t ( x, y ), g ) k t k d t · dx = 0.
ξ →0 x ∈ FR t∈ FR∗,0
We have
2
φ( fˆ)m
I ∪i,∞ ( t ( x, y ), g ) = exp(− π k t ( x, y )k ) N( t ( x + iy ))
m +1
(t( x + iy)) I ∪i (t( x − iy))( I ∪i)c ,
so
Γ(m + 2 + ξ ) g N( x + iy)m+1 ( x + iy) I ∪i ( x − iy)( I ∪i)c
Z
ˆ m m+2+2ξ ×
φ( f ) I ∪i,∞ (t( x, y), g) ktk d t = g g ( m +2+ ξ )
t∈ FR∗,0 2 π | N( x + iy)|2(m+2+ξ )
and the integral in question is therefore
Γ ( m + 2) g dx
Z
lim .
2 g π g ( m +2) ξ →0 x ∈ FR N( x − iy)(m+1) ( x − iy) I ∪i ( x + iy)( I ∪i)c | N( x + iy)|2ξ
This can be calculated as
Γ ( m + 2) g dx
Z
π g ( m +2) x ∈ FR N( x − iy)(m+1) ( x − iy) I ∪i ( x + iy)( I ∪i)c

by dominated convergence, because the integral is dominated by
dx
Z
c , c ∈ R>0 ,
x ∈ FR | N( x + iy)|(m+2)
which exists by Fubini, as the corresponding one dimensional integrals x∈R | x+iydx|(m+2)
R
exist for y ∈ R× . This is seen as follows. The integral equals x∈R 2 dx

R
m
2 ( +1)
. We split
( x +y ) 2
it up as
Z 1 Z ∞
dx dx dx
Z
m =2 m +2 m .
x ∈R ( x 2 + y 2 ) ( 2 +1) 0 ( x2 + y 2 ) ( 2 +1) 1 ( x2 + y 2 ) ( 2 +1)
1
Because m is as function in x continuous on [0, 1], there is a constant C >
( x 2 + y 2 ) ( 2 +1)
0 such that 2 21 ( m +1) < C, x ∈ [0, 1]. Therefore we may estimate our integral by
R ∞ (x +dxy ) 2
2C + 2 1 Moreover, we have 2 21 ( m +1) ≤ xm1+2 , x ∈ [1, ∞], and our
m .
( x 2 + y 2 ) ( 2 +1) ( x +y ) 2
R∞
integral may be estimated by 2C + 2 1 xmdx+2 , which is finite as m ≥ 0. Let us come
back to
Γ ( m + 2) g dx
Z
,
2 g π g ( m +2) x ∈ FR N( x − iy)(m+1) ( x − iy) I ∪i ( x + iy)( I ∪i)c
which is a product of integrals. If we want to show that it vanishes, it suffices to show
the vanishing of a single factor, for example the ith-factor. This factor is essentially the
integral
∞
1 1
Z
dx = = 0.
x ∈R ( x − iy)(m+2) −(m + 1)( x − iy)m+1 −∞
103
Consider the Iwasawa-decomposition

 
t1 t1 x
g = kb ∈ G (A) with b =   and k ∈ G (Ẑ) · ∏ SO(2).
0 t2 ν|∞
Let us define the differential form on K1 \ G (A)

g
(−2) g dt2,j

volP := ∑
κF ∏
∧ dx j ∧ dxi .
i =1 j 6 =i
t2,j
Lemma 4.5.3. We have

g
κ volP −2(−1)| I | dτ ∧ dτ dr ∧ dτ

−1
F = ∑ ∑
τ − τ (i ∪ I )c
∧
r
∧ dτ i
t2 t1 sgn(N(t2−1 t1 )) −1 −1
i =1 I ⊂h gi\i t2 t1 sgn(N( t2 t1 )) I

f f
as cohomology classes in H 2g−1 (∂MK , C) = H p (∂SK , C) ⊗ Hq . More pre-

L
p+q=2g−1
cisely, if A = Z[ Nd1 F ],K = K N and N ≥ 3
d× t(K N,
Z )h
f K N vol P
√
(−2) g d F t2−1 t1 sgn(N(t2−1 t1 ))

f
is in the image of the cohomology with A-coefficients and part of an A-basis of the
A-integral classes.
dτ ∧dτ dx ∧dy
Proof. We have τ −τ = y and we see that the top x-component of
g
dτ ∧ dτ dr ∧ dτ
∑ ∑ τ − τ (i ∪ I )c
∧
r I
∧ dτ i
i =1 I ⊂h gi\i
equals
g
dx ∧ dy dr ∧ dx
∑ ∑ y (i ∪ I )c
∧
r I
∧ dxi .
i =1 I ⊂h gi\i
By Theorem 4.4.1 we already know that these two classes have to be cohomologous.
We have the formulas
1 dx ∧ dy dr

1 1 dt2
d ∧ dx = − ∧ dx , ryd ∧ dx = (−2) ∧ dx
ry ry y r ry t2
from which the claimed equality of forms follows. What remains to be shown is the
integrality statement. Consider components {±1} g R>0 \ B(R)/B( g f ) ⊂ ∂MK N . Set
P0 = g ∈ G0 : ge1 = e1 and get P = ResO /Z P0 . Define

P(R)1 := x ∈ P(R) : x ∈ P(R)0 , N(det( x )) = 1 .

104
This gives the submanifold
i : P(R)1 /P( g f ) → {±1} g R>0 \ B(R)/B( g f ) with P( g f ) := B( g f ) ∩ P(R)0 .
The form
d× t(K N,
Z )h
KN
√ f volP
(−2) g d F t2 t1 sgn(N(t2−1 t1 ))
− 1

f
is a volume form of the compact space on the left-hand side giving it volume N g ∈ A× .
So the form is (up to a factor of N g ) dual to the fundamental class under Poincaré duality
and hence defines an A-integral class on P(R)1 /P( g f ). As we also have a right-inverse
   
t1 t1 x 1 x
p : {±1} g R>0 \ B(R)/B( g f ) → P(R)1 /P( g f ),   7→  | t2 t1−1
|

0 t2 0 √
n
| N(t2 t1−1 )|
for the map i, we may conclude that our form is part of an A-basis of the cohomology
of ∂MK N .
Proposition 4.5.4. For ϕ ∈ S(V (A f ), µ⊗n )0 K f -invariant the cohomology class
resS (Eisk ( ϕ)) ∈ H p (∂S , Symk H0 ⊗ µ⊗n+1 ) ⊗ Hq

M
p+q=2g−1
equals
Γ ( m + 2) g
Z
∑
m +2 ×
fˆ(tg f e1 , g)χ(det( g f )t) det( g f )t

d t·
dK (m) (−2πi )k+2g t ∈ Z (A f )
χ∈Cl F
∗ n +1
N(ω )m volP ⊗ volV
−n
(m!) g kdet( g)k f sgn(N(det( g)))−1
In particular,
Γ ( m + 2) g
Z
∑
m +2 ×

d t
dK (m) (−2πi )k+2g t ∈ Z (A f )
χ∈Cl F
is a rational number.
Proof. We calculate
Z
m +2 ×
φ( fˆ)m 1
I ∪i,∞ ( tg∞ e , g∞ ) χ (det( g∞ ) t ) kdet( g∞ ) t k d t=
t ∈ Z (R)
105
N ( t 2 ) m +2 Γ ( m + 2 ) g
−iθ
N(e(m+1)iθ )eiθ
I ∪i e ( I ∪i )c
π g ( m +2)
Remembering the definition of the Θ I,i (Equation (4.3)) the formula for resS (Eisk ( ϕ))
follows from Proposition 4.5.1 and Lemma 4.5.3. The rationality statement follows,
since
kt2 kmf +2 sgn(N(t2 ))m+2 kdet( g)knf sgn(N(det g))n N(ω )m volP ⊗ volV∗n+1 =
kt2 kmf +2 sgn(N(t2 ))m+2 kdet( g)knf sgn(N(det g))n ∏(−Yσ )m volP ⊗ volV∗n+1
σ
as cohomology classes in H • (∂S , ω k ⊗ µ⊗n+1 ) ⊗ H• . The latter defines a rational class,

compare Lemma 4.5.3 and Section 4.4 and recall d× t(K Zf ) ∈ Q× √1d , since K Zf ⊂ Ô ×
F
is a subgroup of finite index and d× t(Ô × ) = √1 .

dF
Of course, we can refine this statement, if we use the integral structures on cohomology,
which we have chosen.
Proposition 4.5.5. Let N ≥ 3 be an integer, A := Z[ dF1N ] and f ∈ map(V (Z/NZ), A)

with
f (0) = 0, ∑ f (v) = 0.
v∈V ( Z/NZ)
We denote by O × ( N )+ ⊂ O × the group of totally positive units which are congruent to

one modulo N. We write
 
t1 t1 x
g = k  ∈ G (A f ) × π0 ( G (R)), k ∈ G (Ẑ),
0 t2
and get a well-defined function in map(K N \ G (A f ) × π0 ( G (R))/B(Q), A)

√
(−1) g d F Γ(m + 2) g fˆ( ĝ f le1 )
g 7→
(−2πi )k+2g sgn(N(t2 ))m+2 kt2 kmf +2 ×
∑× N ( l ) m +2
∈A
l ∈ F /O ( N )+
Proof. We apply the construction of polylogarithmic Eisensteinclasses to f and obtain

by Proposition 2.4.14
Eisk ( f ) ∈ H 2g−1 (MK N , Symk H PD ⊗ µ)
with A-coefficients. Since Symk H0PD ⊂ Symk H PD is a direct summand, we obtain by

restricting Eisk ( f ) to the boundary ∂MK N
resM (Eisk ( f )) ∈ H 2g−1 (∂MK N , ω kPD ⊗ µ)
106
with A-coefficients. We have calculated resM (Eisk ( f )) with C-coefficients as

Γ ( m + 2) g
Z
∑
m +2 ×
fˆ(tg f e1 )χ(det( g f )t) det( g f )t

d t·
(−2πi )k+2g t ∈ Z (A f )
χ∈Cl
[
KZ
(m)
N
∗
N(ω )m volP ⊗ volV
.
(m!) g
Consider ϕ∞ ∈ S( FR ) defined by ϕ∞ (v) := exp(− kvk2 ) N(v)m+2 , v ∈ FR . We have for
dK (m)
χ ∈ Cl F
Z
ϕ∞ (t)χ∞ (t)| N(t)|m+2 d× t = Γ(m + 2) g .
t∈ FR×
Define ϕ ∈ S(AF ) by ϕ(v) := ϕ∞ (v∞ ) fˆ( g f v f e1 ) for v ∈ AF . We calculate the integral
Γ ( m + 2) g
Z
∑
m +2 ×
fˆ(tg f e1 )χ(det( g f )t) det( g f )t

d t=
[ K
(−2πi )k+2g t ∈ Z (A f )
χ∈Cl F N (m)
1
Z
∑
m +2 ×
ϕ(t)χ(det( g f )t) det( g f )t d t=
[ K
(−2πi )k+2g t ∈ Z (A)
χ∈Cl F N (m)
1
Z
∑
m +2 ×
ϕ(t)χ(det( g f )t) det( g f )t d t=
[ K
(−2πi )k+2g t ∈ Z (A)
χ∈Cl F N
d× t(K N,
Z ) Z
∑ ∑× ϕ(tl )χ(det( g f )t) det( g f )t

f m +2
d× t =

[ K
(−2πi )k+2g Z \ Z (A) /F ×
t∈K N, f l∈F
χ∈Cl F N
d× t(K N,
Z )h
KN
Z
∑× ϕ(det( g f )−1 tl ) ktkm+2 d× t =
f
(−2πi )k+2g t ∈ Z (R)0 / O × ( N ) + l∈F
d× t(K N,
Z )h
KN
Z
∑×
f
ϕ(det( g f )−1 tl ) ktkm+2 d× t =
(−2πi )k+2g ×
l ∈ F /O ( N )+ t ∈ Z (R)0
Z ) h Γ ( m + 2) g
d× t(K N, KN fˆ( ĝ f le1 )
∑×
f
.
2g (−2πi )k+2g ×
l ∈ F /O ( N )+
N ( l ) m +2
The form
d× t(K N, Z )h
KN
√ f volP
d F (−2) g tt21 sgn(N( tt12 ))

f
is part of an A-basis of the cohomology of ∂MK N as seen in Lemma 4.5.3. We already

have mentioned in Section 4.4 that
kt2 kmf sgn(N(t2 ))m N(ω )m
kt2 kmf sgn(N(t2 )) m
∏(−Yσ ) [m]
=
(m!) g
∈ H 0 (∂MK N , ω kPD )
σ
107
and
kdet( g)k f sgn(N(det( g))) volV∗ ∈ H 0 (∂SK N , µ)
define A-trivializations and therefore

m ∗
d× t(K N,
Z )h
f
m m
K N k t2 k f sgn(N( t2 )) kdet( g )k f sgn(N(det( g ))) N( ω ) vol P ⊗ volV
√
d F (m!) g (−2) g tt12 sgn(N( tt21 ))

f
is part of an A-basis of
im( H 2g−1 (∂MK N , ω kPD ⊗ µ) → H 2g−1 (∂MK N , ω kPD ⊗ µ ⊗ C)).
Now we may conclude that

√
(−1) g d F Γ(m + 2) g fˆ( ĝ f le1 )
(−2πi )k+2g sgn(N(t2 ))m+2 kt2 kmf +2 ×
∑× N ( l ) m +2
∈ A,
l ∈ F /O ( N )+
as resM (Eisk ( f )) is A-integral.
Definition 19. Let us define the horospherical map

G (A )×π0 ( G (R))
ρ0m,n : S(V (A), µ⊗n ⊗ Q)0 → Ind B(A f )×π Q · φ,
M
f 0 ( B (R))
φ: type(φ)=γm,n+1 ,φ̃ f |Z(R)=1
by
m +2 ×
R
Γ ( m + 2) g t ∈ Z (A f )
d t
ρ0m,n ( ϕ)( g) := ∑ (−2πi )k+2g
−n ,
dK (m)
χ∈Cl F
kdet( g)k f sgn(N(det( g)))−1
where we have written

n
ϕ(v, g) = f (v, g) kdet( g)k f sgn(N(det( g)))−1
with f ∈ S(V (A f ), µ⊗0 ⊗ Q)0 K f -invariant.
Remark 4.5.6. ρ0m,n is well-defined, this means independent of the choice of K f and
Q-rational so that
 Gal(Q/Q)
G (A )×π0 ( G (R))
ρ0m,n : S(V (A), µ⊗n )0 →  Ind B(A f )×π ( B(R)) Q · φ
M
.
f 0
Proposition 4.5.7. We have
im(resS ◦ Eiskq ) ∼
= im(ρ0m,n ) ⊗ H( T/Z ) g−1−q .
108
Proof. We prove the proposition with C-coefficients and all of the following construc-
tions will respect the chosen Q-structures. By Proposition 4.5.4
N( ω ) m ∗ n +1
resS ◦ Eisk ( ϕ) = ρ0m,n ( ϕ) · volP ⊗ volV ∈ H • (∂S , Symk H0 ⊗ µ⊗n+1 ) ⊗ HC
•
.
(m!) g
Set
g ( g −1) √ g
(−1) 2 (−2) g d F dt2 dx1 ∧ ... ∧ dxn
volT2 =
κF ∑ (−1)i−1 t2 h gi\i
, volU = √
dF
i =1
N( ω ) m ∗ n +1
and get volP = volT2 ∧ volU . Dividing resS ◦ Eisk ( ϕ) by (m!) g
volU ⊗ volV we get
G (A )×π0 ( G (R))
ρ0m,n ( ϕ) · volT2 ∈ Ind B(A f )×π C · φ · H• ( T/Z ) ⊗ HC
•
M
f 0 ( B (R))
So what remains to be shown is that

•,∗
HC → H• ( T/Z )C , v 7→ volT2 (v)
is surjective. Clearly,
H• ( T/Z ) ⊗ HC
•
= H • ( T (R)0 /T (Q) ∩ (K f · T (R)0 , C)
and we may interpret volT2 as rational volume form in t2 -direction. Therefore

dt2
volT2 = c0 ·
t 2 h g −1i
dy
for a c0 ∈ C× . We take dr
i = 1, ..., g − 1, as basis for H1C and y , i = 1, ..., g − 1, as
r i, i
dt

dy
√
basis for H1 ( T/Z )C . We have t2,i2,i = 12 y + drr i , as t2,i = yi ri , and therefore
i
!
1 dy dr 1 dy dr
volT2 = c0 · + ∧ ... ∧ + =
2 y1 r1 2 y g −1 r g −1
c0 dy dr
2 g −1
· ∑ sgn( I )
y I
∧
r h g−1i\ I
,
I ⊂h g−1i
where sgn( I ) is the sign of the permutation which describes h g − 1i = I ∪ (h g − 1i \ I ).

∗
If we take the dual basis V (r )i ∈ H1,
C of
dr
r i, i = 1, ..., g − 1, we may see
c0 dy
volT2 (V (r ) J ) = · sgn(h g − 1i \ J ) , J ⊂ h g − 1i .
2 g −1 y h g−1i\ J
•,∗
In particular, HC → H• ( T/Z )C , v 7→ volT2 (v) is surjective.
Remark 4.5.8. The last proposition shows that for any ϕ ∈ S(V (A f ), µ⊗n ⊗ C)0
dτ − dτ
ρ0m,n ( ϕ) ∧ dτ I , I ( h gi
τ − τ Ic
is in the image of ∑q resS ◦ Eiskq with C-coefficients.
109
4.6 Comparison with Harders Eisenstein classes
The reason why we talk about polylogarithmic Eisenstein classes is that these classes
are actually Eisenstein cohomology classes in the sense of Harder. This is what we
prove now.
Theorem 4.6.1. Define H̃ • (S , Symk H0 ⊗ µ⊗n+1 ) :=

ker resS : H • (S , Symk H0 ⊗ µ⊗n+1 ) → H • (∂S , Symk H0 ⊗ µ⊗n+1 ) .
One has a G (A f ) × π0 ( G (R))-equivariant operator
Eis Harder : im(resS ) → H • (S , Symk H0 ⊗ µ⊗n+1 ),
which is a section for resS . We denote the image of the operator Eis Harder by
•
HEis (S , Symk H0 ⊗ µ⊗n+1 )
and get a decomposition of G (A f ) × π0 ( G (R))-modules
H • (S , Symk H0 ⊗ µ⊗n+1 ) = H̃ • (S , Symk H0 ⊗ µ⊗n+1 ) ⊕ HEis

•
(S , Symk H0 ⊗ µ⊗n+1 )
Proof. Harder (1987) Theorem 2.
The aim of this section is to prove the following
Theorem 4.6.2.
•
im(Eiskq ) ⊂ HEis (S , Symk H0 ⊗ µ⊗n+1 )
• ⊂ H • (M , C) by the invariant differential forms

We generated HC dr
i ∈ h gi \ g. Let
K r i,
dr
V (r )i be the invariant vectorfields dual to r i and set

V (r ) I := V (r )i1 ∧ ... ∧ V (r )iq , I = i1 , ..., iq ⊂ h g − 1i , q = 2g − 1 − p.
Lemma 4.6.3. To prove Theorem 4.6.2 it suffices to show

•
Eiskq ( ϕ)(V (r ) I ) ∈ HEis (S , Symk H0 ⊗ µ⊗n+1 ⊗ C),
for all I ⊂ h g − 1i and

n
ϕ(v, g) = f (v) · η (det( g)) kdet( g)k f sgn(N(det( g))) ,
for f ∈ S(V (A f ))0 and η : ∏ν|∞ R>0 \ IF /F × → C a continuous character.
110
Proof. To prove Theorem 4.6.2 we may extend coefficients to C. The ϕ described

above generate S(V (A f ), µ⊗n ⊗ C)0 . As (V (r ) I ) I ⊂h g−1i is a basis of HC
•∗ , the lemma
follows.
From now on we suppose that ϕ is of the form above. For I $ h gi we define Θ̃ I as the
form
(−2i )−(k+2g) N(e−(m+1)iθ )m+1 e− iθ iθ

I eIc m dτ ∧ dτ
− n N( ω ) ∧ dτ I ⊗ volV∗n+1
−
τ τ I c
N( t2 ) m + 2 det g f sgn(N(det g∞ )) − 1
f
k
f I ( ϕ, χ) :=
and set Eis
Z
∑
m+2+2ξ ×
lim φ( fˆ)m 1
I ( tgγe , g ) η (det( g )) χ (det( g ) t ) kdet( g ) t k d t · Θ̃ I .
γ∈ G (Q)/B(Q)) t∈ Z (A)
ξ →0
Lemma 4.6.4. To prove Theorem 4.6.2 it suffices to show that the differential form
k • k 0 ⊗n+1 ⊗ C) for all I $ h g i.
∑χ∈Cl
dK (m) Eis I ( ϕ, χ ) defines a class in HEis (S , Sym H ⊗ µ
f
F
Proof. Eiskq ( ϕ)(V (r ) I ) are C-linear combinations of ∑χ∈Cl fk

dK (m) Eis I ( ϕ, χ ).
F
To prove this last step we have to rewrite our classes in (g, K )-cohomology, as Harder
constructed his operator Eis Harder in this setting.
Remark 4.6.5. The local system Symk H0 ⊗ µ⊗n+1 corresponds to the G (Q)-representation
k V (Q) ⊗ vol∗n+1 on which G (Q) acts by
SymQ V
∗ n +1 ∗ n +1
ρ( g)v1 ...vk volV := N(det( g))n+1 gv1 ...gvk volV ,
for v1 , ..., vn ∈ V (Q), g ∈ G (Q) and gvi the standard action by matrix multiplication. Set
g := Lie( G (R)) = T1 G (R), k := Lie(K∞ ) = T1 K∞ .
Lie(R>0 · SO(2)). Moreover, g

L L
Of course, we have g = ν|∞ Lie( G0 ( Fν )) and k = ν|∞
and k are FR -modules. We get the isomorphism Ψ
Γ(SK , ASK ⊗Q Symn H0 ⊗ µ⊗n+1 ) =
∗ n +1
H 0 ( G (Q), Γ(K \ G (A), AK\G(A) ⊗Q Symn V (Q) ⊗ volV )) →
•
g/k, C ∞ (K f \ G (A)/G (Q)) ⊗ Symn V (Q) ⊗ volV
∗ n +1
^
HomK∞ ( )
111
as described in Borel & Wallach (2000) chapter VII, proposition 2.5. Given a differtential
V•
form η on the left hand side and X ∈ g/k this means explicitly
Ψ(η )( X )( g) := ρ( g∞ )η ( p( g))(dp g dr g,1 X ) where
r g : K f \ G (A) → K f \ G (A), x 7→ xg, and p : K f \ G (A) → K \ G (A)
is the canonical map. Unlike Borel & Wallach (2000) we consider right translation in-
stead of left translation due to the fact that G (Q) acts from the right. We can make
everything explicit by giving a basis for g/k ⊗ C. We take
     
1 1 ± i 1 0 0 1
P± :=  = 1 ± 1  ⊗i
2 ± i −1 2 0 −1 2 1 0
as a basis for Lie( G0 (R))/Lie(R>0 SO(2))C , compare Harder (1993) p. 130. We have
for the adjoint representation
 
a −c
Ad(k ) P± = kP± k−1 = ( a ∓ ic)2 P± , k =   ∈ SO(2).
c a
We have H± := C \ R and the isomorphism

 
b + id w2 a b
G0 (R) → H± , g 7→ z = ig = = , for g =   , w1 := a + ic, w2 := b + id,
a + ic w1 c d
and we get
∂ ∂ ∂ ∂
dr g,1 P+ = (w1 + w2 )|(w1 ,w2 )=(1,i) g , dr g,1 P− = (w1 + w2 ) ,
∂w1 ∂w2 ∂w1 ∂w2 |(w1 ,w2 )=(1,i) g
det( g) t2
dz( p( g))(dr g,1 P− ) = 0, dz( p( g))(dr g,1 P+ ) = 2i 2
= 2i ei2θ and
( a − ic) t1
det( g) t2
dz( p( g))(dr g,1 P+ ) = 0, dz( p( g))(dr g,1 P− ) = −2i = −2i e−i2θ
( a + ic)2 t1
where we used the Iwasawa-decomposition of g ∈ G0 (R), see Remark 4.3.8. Finally,
consider ω0 := (e1 z − e2 ) ∈ C ∞ (H± , V0 (R)) and rewrite it using the G0 (R)-action on
V0 (R) as
−idet( g) 1
ω0 = g −1 ( (e − ie2 )) = g−1 (−it2 eiθ (e1 − ie2 )).
a − ic
Remember that we had

n
ϕ(v, g) = f (v) · η (det( g)) kdet( g)k f sgn(N(det( g))) ,
112
for f ∈ S(V (A f ))0 and η : ∏ν|∞ R>0 \ IF /F × → C× a continuous character. For

dK (m) we consider the algebraic Hecke-character
χ ∈ Cl F
φ : T (A)/T (Q) → C× , (t1 , t1 ) 7→ η (t1 t2 )χ(t2 ) kt2 km+n+2 kt1 kn .
φ is of type γm,n+1 and φ̃ f |Z(R) = 1, therefore φ is a character contributing to the coho-

mology of the boundary. Write
 
t1 t1 x
g = kb, k ∈ SL2 (Ô) · K1 , b =   ∈ B (A).
0 t2
m ( χ, ξ, g ) : =
Given η, χ and f as above we have well-defined functions FI,∞
t2 t2
| N( )|ξ η∞ (det( g)) N(−it2 eiθ )m N(2i )(ei2θ ) I N(det( g))n+1 sgn(N(det( g)))n
t1 t1
on G (R), FI,mf (χ, ξ, g f ) :=
ξ 2+ m
Γ(m + 2) g tt21 η f (det( g)) t∈Z(A fˆ(tg f e1 )χ f (det( g f )t) det( g f )t f d× t
R
f f)
−n
(−2πi )k+2g det( g f ) f
on G (A f ) and FIm (χ, ξ, ) := FI,mf (χ, ξ, ) ⊗ FI,∞

m ( χ, ξ, ) on G (A). In other words,
FIm (χ, ξ, g) := FI,mf (χ, ξ, g f ) · FI,∞

m
(χ, ξ, g∞ ).
Lemma 4.6.6. We define

•
∗ n +1
ωm g/k, V ∗ t k
V (C) ⊗ volV
^
I ( ϕ, χ, ξ ) ∈ Hom K∞ ( ξ ⊗ SymC )
φ t2

1
by
( P− ∧ P+ ) I c ∧ P+ I 7→ FIm (χ, ξ, ) N(e1 − e2 i )m volV∗n+1
and all elements not collinear with ( P− ∧ P+ ) I c ∧ P+ I mapping to zero.
Proof. For the definition of V ∗ t see Harder (1987) 3.3 and p. 79. We have to show
ξ
φ t2

1  
t t x
that ω m  1 1  ∈ B(A) and
I ( ϕ, χ, ξ ) actually defines a form as claimed. For b =
0 t2
g ∈ G (A) we have
ξ
t2
FIm (χ, ξ, gb) = m
t φ(t1 , t2 ) FI (χ, ξ, g).

1
113
K f and K∞ -finiteness of FIm (χ, ξ, ) is clear and therefore FIm (χ, ξ, ) ∈ V ∗ t ξ . So what
φ t2

1
remains to be shown is that
( P− ∧ P+ ) I c ∧ P+ I 7→ FIm (χ, ξ, ) N(e1 − e2 i )m volV∗n+1
is K∞ -linear. Consider
     
x −y x − y cos θ ν − sin θ ν
k = z  ∈ K∞ with z ∈ Z (R)0 and   = .
y x y x sin θν cos θν
ν
We have
  −1
x −y
Ad(k−1 ) ( P− ∧ P+ ) I c ∧ P+ I = Ad   ( P− ∧ P+ ) I c ∧ P+ I =
y x
e2iθ
I ( P− ∧ P+ ) I c ∧ P+ I
and on the other hand
∗ n +1
k −1 ω m −1 m m
I ( ϕ, χ, ξ )(( P− ∧ P+ ) I c ∧ P+ I )( kg ) = k FI ( χ, ξ, kg ) N( e1 − e2 i ) volV =
∗ n +1
N(z−1 )2(n+1) N(z−1 e−iθ )m FIm (χ, ξ, kg) N(e1 − e2 i )m volV =
N(z−1 )2(n+1) N(z−1 e−iθ )m FI,mf (χ, ξ, g f ) FI,∞

m
(ξ, kg∞ ) N(e1 − e2 i )m volV∗n+1 =
m m ∗ n +1
e2iθ
I FI ( χ, ξ, g ) N( e1 − e2 i ) volV
m ( χ, ξ, kg ) and it follows that ω m ( ϕ, χ, ξ ) is (right)-K -equivariant.

by the definition of FI,∞ I ∞
Harder defines
•
Eis∗ (ω m g/k, C ∞ (K f \ G (A)/G (Q)) ⊗ SymC
k ∗ n +1
V (C) ⊗ volV
^
I ( ϕ, χ, ξ )) ∈ Hom K∞ ( )
by
( P− ∧ P+ ) I c ∧ P+ I 7→ ∑ ∗ n +1
FIm (χ, ξ, gγ) N(e1 − e2 i )m volV
γ∈ G (Q)/B(Q)
and all elements not collinear with ( P− ∧ P+ ) I c ∧ P+ I mapping to zero, see Harder (1987)
(4.2.2).
114
Proposition 4.6.7.
dτ − dτ
∑ lim Eis∗ (ω m
ξ →0
0
I ( ϕ, χ, ξ )) = Eis Harder ( ρm,n ( ϕ ) N( ω )
m
τ − τ Ic
∧ dτ I ⊗ volV∗n+1 ),
dK (m)
χ∈Cl F
k
lim Eis∗ (ω m
I ( ϕ, χ, ξ )) = Ψ (Eis I ( ϕ, χ ))
f
ξ →0
and Theorem 4.6.2 is proved.
Proof. We have
dτ − dτ
ρ0m,n ( ϕ) N(ω )m ∧ dτ I ⊗ volV∗n+1 =
τ−τ I c
dτ ∧ dτ
∑ FI,mf (χ, 0, g)η∞ (det( g)) sgn(N(det( g)))n N(ω )m
τ − τ Ic
∧ dτ I ⊗ volV∗n+1 .
dK (m)
χ∈Cl F
Following Remark 4.6.5 and Harder (1987) p.79 and p.80 we see that
dτ − dτ
∑ lim Eis∗ (ω m
ξ →0
0
I ( ϕ, χ, ξ )) = Eis Harder ( ρm,n ( ϕ ) N( ω )
m
τ − τ Ic
∧ dτ I ⊗ volV∗n+1 )
dK (m)
χ∈Cl F
by the very definition of Harder. The right hand side of the equation is defined, as we
know by Remark 4.5.8 that
dτ − dτ
ρ0m,n ( ϕ) N(ω )m ∧ dτ I ⊗ volV∗n+1 ∈ im(resS ).
τ − τ Ic
k
So what remains to be shown is lim Eis∗ (ω m
I ( ϕ, χ, ξ )) = Ψ (Eis I ( ϕ, χ )). We start with
f
ξ →0
the calculation of f kI ( ϕ, χ)).

Ψ(Eis Using Remark 4.6.5 we see that it is determined by
Z
∑
m+2+2ξ ×
lim φ( fˆ)m 1
I ( tgγe ) η (det( g )) χ (det( g ) t ) kdet( g ) t k d t=
γ∈ G (Q)/B(Q)) t∈ Z (A)
ξ →0
Z
∑
m+2+2ξ ×
lim φ( fˆ)m 1
I ( tgγe ) η (det( gγ )) χ (det( gγ ) t ) kdet( gγ ) t k d t=
γ∈ G (Q)/B(Q)) t∈ Z (A)
ξ →0
Z
∑
m+2+2ξ ×
fˆ(tg f γe1 )η f (det( g f γ))χ f (det( g f γ)t) det( g f γ)t f

lim d t·
γ∈ G (Q)/B(Q)) t∈ Z (A f )
ξ →0
Z
m+2+2ξ ×
φ( fˆ)m 1
I,∞ ( tg∞ γe ) η∞ (det( g∞ γ )) χ∞ (det( g∞ γ ) t )| N(det( g∞ γ ) t )| d t
t ∈ Z (R)
Let us set
Z
m+2+2ξ ×
F∞ (ξ, g) := φ( fˆ)m 1
I,∞ ( tg∞ e ) η∞ (det( g∞ )) χ∞ (det( g∞ ) t )| N(det( g∞ ) t )| d t,
t ∈ Z (R)
115
Z m+2+2ξ ×
fˆ(tg f e1 )η f (det( g f ))χ f (det( g f )t) det( g f )t f

Ff (ξ, g) := d t
t ∈ Z (A f )
for the moment. Consider
G ( g f ) := ( g−
f K f g f · G (R) ) ∩ G (Q), B ( g f ) = : G ( g f ) ∩ B (A).
1 0
G ( g f ) acts on G (Q)/B(Q) by left translation and we decompose the latter into G ( g f )

orbits. Now we may write our sum above as
lim
ξ →0
∑ ∑ Ff (ξ, gsγ) F∞ (ξ, gsγ)
s∈ G ( g f )\ G (Q)/B(Q) γ∈ G ( g f s)/B( g f s)
G ( g f ) \ G (Q)/B(Q) is finite, as G (Z) \ G (Q)/B(Q) ∼

= Cl F is the class group of F.
Moreover, G ( g f ) does not affect Ff (ξ, ): By construction of Eisk we have chosen the
level K f for f and η such that f (kv) = f (v) and η (det(k )) = 1 for all k ∈ K f and
v ∈ V (A f ), see Proposition 3.2.3. Therefore we may write our sum as
∑ Ff (0, gs) lim

ξ →0
∑ F∞ (ξ, gsγ)
s∈ G ( g f )\ G (Q)/B(Q) γ∈ G ( g f s)/B( g f s)
Let us take care of F∞ . We write

   
1 x a b
g∞ = kb = kt1   and sγ =  .
0 y c d
We get
−iθ
F∞ (ξ, gsγ) = N(eiθ )m+1 eiθ
I e I c η∞ (det( g∞ sγ )) N( t2 )
m +2
| N(t2 )|2ξ ·
N(det(sγ))m+2 | N(det(sγ))|2ξ N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c Γ(m + 2 + ξ ) g

.
π k+2g+ gξ | N( a + cτ )|2(m+2+ξ )
Using | N(det(γ))| = 1 we calculate
Γ ( m + 2) g
lim
ξ →0
∑ F∞ (ξ, gsγ) =
π k+2g
N(eiθ )m+1 eiθ −iθ
I e I c N( t2 )
m +2
·
γ∈ G ( g f s)/B( g f s)
η∞ (det( gsγ)) N(det(sγ))m+2 N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c

lim
ξ →0
∑ | N( a + cτ )|2(m+2+ξ )
=
γ∈ G ( g f s)/B( g f s)
η∞ (det(sγ))| N(det(sγ))|ξ N(det(sγ))m+2 N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c

lim
ξ →0
∑ | N( a + cτ )|2(m+2+ξ )
·
γ∈ G ( g f s)/B( g f s)
Γ ( m + 2) g −iθ t2
k + 2g
η∞ (det( g)) N(eiθ )m+1 eiθ
I e I c N( t2 )
m +2
| N( )|ξ
π t1
116
4.7 The image of the polylogarithmic Eisenstein Operator
Looking carefully at the formulas in Remark 4.6.5 and recalling
det( g f s) −n | N(det(sγ))|−n = det( g f ) −n

f f
f kI ( ϕ, χ)) is the form that maps ( P− ∧ P+ ) c ∧ P+ I to

we may conclude that Ψ(Eis I
lim
ξ →0
∑ FI,mf (χ, 0, g f s)
s∈ G ( g f )\ G (Q)/B(Q)
η∞ (det(sγ))| N(det(sγ))|ξ N(det(sγ))m+2 N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c

∑ | N(det(sγ))|−n | N( a + cτ )|2(m+2+ξ )
·
γ∈ G ( g f s)/B( g f s)
t2 ξ t2
sgn(N(det( g)))n η∞ (det( g))| N( )| N(−it2 eiθ (e1 − ie2 ))m N(2i )(ei2θ ) I N(det( g))n+1 volV∗n+1
t1 t1
and all the elements not collinear with ( P− ∧ P+ ) I c ∧ P+ I to zero. The last sum is nothing
else but
lim
ξ →0
∑ FI,mf (χ, ξ, g f γ) FI,∞
m
(χ, ξ, g) N(e1 − e2 i )m volV∗n+1 ·
γ∈ G (Q)/B(Q)
η∞ (det(sγ)) N(det(sγ))m+2 | N(det(sγ))|ξ N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c

| N(det(sγ))|−n | N( a + cτ )|2(m+2+ξ )
Finally,
det(sγ) a + cτ iθ ( g)
t2 ( gsγ) = t2 ( g) and eiθ ( gsγ) = e
| a + cτ | | a + cτ |
m ( χ, ξ, gsγ ) equals
and therefore FI,∞
m η∞ (det(sγ)) N(det(sγ))m+2 | N(det(sγ))|ξ N( a + cτ )m+1 ( a + cτ ) I ( a + cτ ) I c

FI,∞ (χ, ξ, g) ·
| N(det(sγ))|−n | N( a + cτ )|2(m+2+ξ )
k
Consequently Ψ(Eis
f I ( ϕ, χ)) is the form that maps
( P− ∧ P+ ) I c ∧ P+ I 7→ lim
ξ →0
∑ ∗ n +1
FIm (χ, ξ, gγ) N(e1 − e2 i )m volV .
γ∈ G (Q)/B(Q)
This form is exactly lim Eis∗ (ω m

I ( ϕ, χ, ξ )).
ξ →0
The polylogarithmic Eisenstein classes are built up by Eisenstein series associated

to Schwartz-Bruhat functions on V (A), while Harder’s Eisenstein series are associ-
ated to functions on G (A), which are induced from algebraic Hecke-characters on
117
T (A)/T (Q). Eisenstein series associated to such induced functions on G (A) can be
represented as finite sums of Eisenstein series coming from Schwartz-Bruhat functions
on V (A), see Jacquet & Zagier (1987) Lemma. So we have a good reason to believe
that the image of the polylogarithmic Eisenstein operator is quite big. In this section we
determine im(Eiskq ) ⊂ HEis
• (S , Symk H 0 ⊗ µ⊗n+1 ) completely. As
•
resS : HEis (S , Symk H0 ⊗ µ⊗n+1 ) → H • (∂S , Symk H0 ⊗ µ⊗n+1 )
is an isomorphism of G (A f ) × π0 ( G (R))-modules onto its image, it suffices to examine

the image of resS ◦ Eiskq . This last operator has already been calculated in Proposi-
tion 4.5.4 and we know by Proposition 4.5.7 that it suffices to understand the image of
the horospherical map ρ0m,n . As we may multiply our functions by
kdet( g)k−f n sgn(N(det( g)))−n , n ∈ Z,
it even suffices to understand the image of ρ0m,0 . But first we want to understand a more
general map.
Definition 20. Define the horospherical map for m ≥ 0
G (A )×π0 ( G (R))
ρm : S(V (A f ), µ⊗0 ⊗ C) → Ind B(A f )×π C · φ̃ f
M
f 0 ( B (R))
φ: type(φ)=γm,1 ,φ̃ f |Z(R)=1
by
Γ ( m + 2) g
Z
∑
m +2 ×

ρm ( f )( g) := d t,
dK (m) (−2πi )k+2g t ∈ Z (A f )
χ∈Cl F
if f is K f -invariant.
4.7.1 Surjectivity of the horospherical map
Proposition 4.7.1. The horospherical map ρm is surjective.
Proof. The idea how to prove this proposition is essentially already in Jacquet & Zagier
(1987) Lemma. Nevertheless, we give the proof for the sake of completeness. Set for
N∈N
:= UN ∏ R>0 \ IF /F × .
(N)
UN := ker(Ô × → (O /N O)× ), Cl F
ν|∞
118
Let φ be an algebraic Hecke-character as above. We may write
φ ( t 1 , t 2 ) = η ( t 1 t 2 ) χ 0 ( t 2 ) k t 2 k m +2 , ( t 1 , t 2 ) ∈ T (A ),
[ (N)
for characters η, χ0 ∈ Cl F for some N ∈ N. Then
φ̃ f (t1 , t2 ) = η (t1 t2 )χ0 (t2 ) kt2 kmf +2 sgn(N(t2 ))m+2 , (t1 , t2 ) ∈ T (A).
(N)
φ̃ f |Z(R) = 1 means χ0 (t) = sgn(N(t))m+2 , t ∈ FR× , in other words χ ∈ Cl F (m) :=
Cl FK N . Now take
G (A )×π0 ( G (R))
ψ ∈ Ind B(A f )×π C · φ̃ f .
f 0 ( B (R))
By choosing N big enough we may also assume that ψ(kx ) = ψ( x ) holds for all x ∈
G (A f ) × π0 ( G (R)) and k ∈ K N . Define ψ by
ψ( x ) = η (det( x ))ψ(det( x )−1 · x ), x ∈ G (A f ) × π0 ( G (R)).
As det(K N ) = UN , we still have ψ(kx ) = ψ( x ) for all x ∈ G (A f ) × π0 ( G (R)) and

 
t1 t1 u
k ∈ K N , but now we have for b =   ∈ B(A f ) × π0 ( B(R))
0 t2
m +2
−(m+2)
ψ( xb) = ψ( x )χ0 (t1−1 ) t1−1 sgn(N(t1−1 ))m+2 = ψ( x )χ0f (t1 )−1 kt1 k f .

f
This means that ψ is a function
K N \ G (A f ) × π0 ( G (R))/P(A f ) × π0 ( P∞ ) → C.
We restrict ψ to G (Ẑ) and obtain
ψ : K N \ G (Ẑ)/P(Ẑ) = G (Z/NZ)/P(Z/NZ ) → C.
We consider the embedding i : G (A f )/P(A f ) ,→ V (A f ), x 7→ xe1 , and make ψ to

a function on V (A f ) by extending it by zero outside G (A f )/P(A f ). When we restrict
ψ to V (Ẑ), it even induces a function on V (Z/NZ) and may also be interpreted as a
Schwartz-Bruhat function sψ ∈ S(V (A f ), C). sψ can be explicitly described as follows:
Take a complete set of representatives x ∈ G (Ẑ) of G (Z/NZ)/P(Z/NZ). Then
sψ := ∑ ψ( x )χ xe1 + NV (Ẑ) = ⊗ν-∞ sψν

xmodN ∈ G (Z/NZ)/P(Z/NZ)
119
where the local functions sψν : Fν → C are defined by sψν := χOν2 , if ν - N, and
sψν := ∑ ψ( xν )χ xν e1 + N Oν2 = ψ · χG0 (Oν )/P0 (Oν ) , if ν| N.

xν ∈ G0 (Oν /N Oν )/P0 (Oν /N Oν )
(N)
Take a full set of representatives u ∈ IF of Cl F . We define ϕ ∈ S(V (A f ), µ⊗0 ⊗ C) by
its Fourier transform with respect to v
ϕ̂(v, g) := ∑ χ0f (u) kukmf +2 η (det( g))sψ(uv).

u
Let us calculate ρm ( ϕ). We have by definition
Γ ( m + 2) g
Z
∑
m +2 ×
ϕ̂(tg f e1 , g)χ(det( g f )t) det( g f )t

ρm ( ϕ)( g) = d t=
[ (N)
(−2πi )k+2g t ∈ Z (A f )
χ∈Cl F (m)
Γ ( m + 2) g
∑ ∑ χ0f (u) kukmf +2 η (det( g))
(−2πi )k+2g
·
u [ (N)
χ∈Cl F (m)
Z m +2 ×
sψ(utg f e1 )χ(det( g f )t) det( g f )t

d t.
t ∈ Z (A f )
So we need to understand the last integral. Analogous to sψ we define the Schwartz-

Bruhat function sχ0−1 ∈ S(AF, f , C). Extend χ0−1 by zero to a function on AF, f and
choose again a full set of representatives x ∈ Ô × of (O /N O)× = UN \ Ô × . Then
sχ0−1 := ∑ χ0−1 ( x )χ x+ N Ô = ⊗ν|∞ sχ0−

ν
1
x ∈(O /N O)×
and the local functions on Fν are defined by
sχν0−1 := χOν , if ν - N, and sχν0−1 := ∑ χ0−1 ( xν )χ xν + N Oν = χ0−1 · χOν× , if ν| N.

xν ∈(Oν /N Oν )×
Write g f = xb for x ∈ G (Ẑ) and b ∈ B(A f ) and see
sψ(tug f e1 ) = sψ( xtut1 e1 ) = ∏ χO ((xtut1 e1 )ν ) ∏ ψ((xtut1 )ν )χG (O )/P (O ) ((xtut1 e1 )ν ).

2
ν 0 ν 0 ν
ν- N ν| N
As x ∈ G (Ẑ), we may write this as
∏ χO ((tut1 )ν ) ∏ ψ((xtut1 )ν )χO× ((tut1 )ν ).

ν ν
ν- N ν| N
Now
−(m+2)
ψ(( xtut1 )ν ) = k(tut1 )ν k f χ0−1 ((tut1 )ν )ψ( xν ) = χ0−1 ((tut1 )ν )ψ( xν ) for (tut1 )ν ∈ Oν×
120
and we get using ψ|K N = 1
∏ χO ((tut1 )ν ) ∏ ψ(xν )χ0−1 ((tut1 )ν )χO× ((tut1 )ν ) =

ν ν
ν- N ν| N
ψ( x ) ∏ χOν ((tut1 )ν ) ∏ χ0−1 ((tut1 )ν )χOν× ((tut1 )ν ) = ψ( x )sχ0−1 (tut1 )

ν- N ν| N
Now we may write our integral as

Z m +2 ×
sψ(utg f e1 )χ(det( g f )t) det( g f )t

d t=
t ∈ Z (A f )
Z m +2 ×
sχ0−1 (utt1 )χ(det( g f )t) det( g f )t

ψ( x ) d t=
t ∈ Z (A f )
−(m+2) m +2 Z
ψ( x )χ f (ut1 )−1 kut1 k f sχ0−1 (t)χ(t) ktkm+2 d× t

χ f (det( g f )) det( g f ) f
t ∈ Z (A f )
Remembering
−(m+2)
ψ( g) = ψ( g det( g)−1 ) = ψ( g)η (det( g))−2 χ0f (det( g)−1 ) kdet( g)k f
(−2πi )k+2g
we see now that ρ ( ϕ)( g)
Γ ( m +2) g m
equals
Z
ψ( g) ∑ ∑ χ0f · χ−f 1 (ut1 det( g)−1 ) · sχ0−1 (t)χ(t) ktkm+2 d× t =
u t ∈ Z (A f )
[ (N)
χ∈Cl F (m)
Z
ψ( g) ∑ ∑ χ0 · χ−1 (ut1 det( g)−1 ) · sχ0−1 (t)χ(t) ktkm+2 d× t.
u t ∈ Z (A f )
[ (N)
χ∈Cl F (m)
Of course, ∑u χ0 · χ−1 (ut1 det( g)−1 ) = 0, whenever χ 6= χ0 , and therefore

(N)
|Cl F |Γ(m + 2) g
Z
ρm ( ϕ) = sχ0−1 (t)χ0 (t) ktkm+2 d× t · ψ.
(−2πi )k+2g t ∈ Z (A f )
To complete our proof we need to show that

Z
sχ0−1 (t)χ0 (t) ktkm+2 d× t 6= 0.
t ∈ Z (A f )
But following Tate (1967) p.342 f.f. and using sχ0−1 = ⊗ν-∞ sχ0−
ν
1 we calculate
1 1
Z
sχ0−1 (t)χ0 (t) ktkm+2 d× t = √ ∏ 1 − χ0 (π −(m+2)
t ∈ Z (A f ) dF ν- N∞ ν ) N (pν )
an convergent Euler-product and therefore non-zero.
121
Remark 4.7.2. We keep the notations from above. Let us set

(N)
|Cl f | Γ ( m + 2) g Z
Λ N (χ, m + 2) = sχ0−1 (t)χ0 (t) ktkm+2 d× t
(−2πi )k+2g t ∈ Z (A f )
For any set of representatives {u} the function ϕ{u} ∈ S(V (A f ), µ⊗0 ⊗ C) defined by
its Fourier transform
ϕ̂{u} (v, g) := Λ N (χ, m + 2)−1 ∑ χ0f (u) kukmf +2 η (det( g))sψ(uv)

u
is a preimage of ψ under ρm .
Let us set T 1 := ker(det : T → ResO /Z Gm ).
Lemma 4.7.3.
G (A )×π0 ( G (R))
C · φ̃ f
M
Ind B(A f )×π
f 0 ( B (R))
φ: type(φ)=γm,1 ,φ̃ f |Z(R)=1
is in the image of ρ0m,0 , if φ|T1 (A) 6= k · k2 .
Proof. Write φ(t1 , t2 ) = η (t1 t2 )χ(t2 ) kt2 km+2 . φ|T1 (A) 6= k · k2 tells us that m ≥ 1 or χ 6=
1 and if m = 0 we know χ| F× = 1. So there is an u0 ∈ IF such that χ f (u0 ) ku0 kmf 6= 1.
R
We calculate
Z
χ f (u0 ) ku0 kmf ϕ̂{u} (v, g)dv =
v ∈V (A f )
Z
Λ N (χ, m + 2)−1 ∑ χ f (u0 u) kuu0 kmf kuk2f η (det( g)) sψ(uv)dv
u v ∈V (A f )
Z
Λ N (χ, m + 2)−1 ∑ χ f (u0 u) ku0 ukmf kuk2f η (det( g)) sψ(u0 u(u0−1 v))dv
u v ∈V (A f )
Z
Λ N (χ, m + 2) −1
∑ χ f (u0 u) ku0 ukmf kuu0 k2f η (det( g))
v ∈V (A f )
sψ(u0 uv)dv =
u
Z
ϕ̂{u0 u} (v, g)dv.
v ∈V (A f )
In other words,
χ f (u0 )−1 ku0 k−f m ϕ̂{u0 u} − ϕ̂{u} ∈ S(V (A f ), µ⊗0 ⊗ C)0
and by Fourier inversion
χ f (u0 )−1 ku0 k−f m ϕ{u0 u} − ϕ{u} ∈ S(V (A f ), µ⊗0 ⊗ C)0 .
122
Now we see
χ f (u0 )−1 ku0 k−f m ϕ{u0 u} − ϕ{u} χ f (u0 )−1 ku0 k−f m ϕ{u0 u} − ϕ{u}
! !
ρ0m,0 = ρm =
χ f (u0 )−1 ku0 k−f m − 1 χ f (u0 )−1 ku0 k−f m − 1
χ f (u0 )−1 ku0 k−f m ρm ( ϕ{u0 u} ) − ρm ( ϕ{u} ) χ f (u0 )−1 ku0 k−f m ψ − ψ
= =ψ
χ f (u0 )−1 ku0 k−f m − 1 χ f (u0 )−1 ku0 k−f m − 1
4.7.2 The spherical functions
So what is left to be considered are
G (A )×π0 ( G (R))
Ind B(A f )×π C · φ̃ f , φ|T1 (A) = k · k2 ,
f 0 ( B (R))
in particular m = 0. We choose the unique right-invariant Haar measure ds on SL2 (AF, f )

with ds(SL2 (Ô)) = 1. The group SL2 (AF, f ) is unimodular.
Definition 21. Let φ : T (A)/T (Q) → C× be an algebraic Hecke-character of type γ0,n

with φ|T1 (A) = k · k2 . Then we have the spherical function
G (A )×π0 ( G (R))
S(φ) ∈ Ind B(A f )×π C · φ̃ f
f 0 ( B (R))
defined by
 
t1 t1 u
S(φ)( g) := φ̃ f (b) := φ̃ f (t1 , t2 ), if g = xb, x ∈ SL2 (Ô), b =   ∈ B (A)
0 t2
Lemma 4.7.4. Let φ : T (A) → C× be an algebraic Hecke-character of type γ0,n with

φ|T1 (A) = k · k2 . Then we have an G (A f ) × π0 ( G (R)) equivariant projection operator
G (A )×π0 ( G (R))
Ψφ : Ind B(A f )×π C · φ̃ f → C · S(φ)
f 0 ( B (R))
defined by
Z
Ψφ (ψ)( g) := ψ(sg)ds
s∈SL2 (Ô)
Proof. Write g = xb, x ∈ SL2 (Ô), as above. Then

Z Z
Ψφ (ψ)( g) := ψ(sg)ds = ψ(sxb)ds =
s∈SL2 (Ô) s∈SL2 (Ô)
123
Z Z
φ̃ f (b) ψ(sx )ds = S(φ)( g) ψ(s)ds = S(φ)( g)Ψφ (ψ)(1).
s∈SL2 (Ô) s∈SL2 (Ô)
Since SL2 (AF, f ) is unimodular, the operator Ψφ is also G (A f ) × π0 ( G (R)) equivariant.
Remark 4.7.5. Ψφ (S(φ)) = S(φ), so Ψφ is a split epimorphism, and ψ ∈ ker(Ψφ ) if and

only if
Z
ψ(s)ds = Ψφ (ψ)(1) = 0.
s∈SL2 (Ô)
Lemma 4.7.6. Let φ : T (A) → C× be an algebraic Hecke-character of type γ0,1 with

φ̃ f |Z(R) = 1 and φ|T1 (A) = k · k2 . Then ker(Ψφ ) is in the image of ρ00,0 .
Proof. Let ψ ∈ ker(Ψφ ) be given. We suppose ψ is left K N -invariant and φ is UN invari-

(N)
ant. For any set of representatives {u} ⊂ IF of Cl F the function ϕ{u} ∈ S(V (A f ), µ⊗0 ⊗
C) defined by its Fourier transform
ϕ̂{u} (v, g) := Λ N (χ, m + 2)−1 ∑ χ0f (u) kukmf +2 η (det( g))sψ(uv)

u
is a preimage of ψ under ρm . To prove the lemma we show that ϕ{u} ∈ S(V (A f ), µ⊗0 ⊗
C)0 . By Fourier inversion we have to show that ϕ̂{u} ∈ S(V (A f ), µ⊗0 ⊗ C)0 and there-
fore it suffices to prove
Z
sψ(uv)dv = 0.
v ∈V (A f )
Recall how we defined sψ in the course of the proof of proposition 4.7.1 and calculate
Z Z
v ∈V (A f )
sψ(uv)dv = kuk−f 2
v ∈V (A f )
sψ(v)dv = kuk−f 2 dv( NV (Ẑ)) ∑ sψ(v) =
v∈V (Z/NZ)
kuk−f 2 dv( NV (Ẑ)) ∑ ψ( x ) =

x ∈ G (Z/NZ)/P(Z/NZ)
kuk−f 2 dv( NV (Ẑ))

kuk−f 2 dv( NV (Ẑ)) ∑ ψ(s) =
|U0 (O /N O)| ∑ ψ ( s ).
s∈SL2 (O /N O)/U0 (O /N O) s∈SL2 (O /N O)
(1)
If we set K N ∩ SL2 (Ô) := K N we may write this as
kuk−f 2 dv( NV (Ẑ))

|U0 (O /N O)| ∑ ψ(s) =
(1)
s∈K N \SL2 (Ô)
kuk−f 2 dv( NV (Ẑ)) Z kuk−f 2 dv( NV (Ẑ)) Z

ψ(s)ds = ψ(s)ds = 0,
|U0 (O /N O)|ds(K N )
(1) s∈SL2 (Ô) |U0 (O /N O)|ds(K1N ) s∈SL2 (Ô)
as ψ|SL2 (Ô) = ψ|SL2 (Ô) .
124
Now we may define the operator Ψm,n

G (A )×π0 ( G (R))
C · φ̃ f → C · S ( φ ),
M M
Ind B(A f )×π
f 0 ( B (R))
φ: type(φ)=γm,n ,φ̃ f |Z(R)=1 2
φ: type(φ)=γm,n ,φ̃ f |Z(R)=1 ,φ|T1 (A) =k · k
where Ψm,n is zero on those summands with φ|T1 (A) 6= k · k2 and Ψm,n equals Ψφ de-
fined above on those summands with φ|T1 (A) = k · k2 .
Remark 4.7.7. Ψm,n is defined over Q. To see this consider a function

G (A )×π0 ( G (R))
Z
ψ∈ Ind B(A f )×π ( B(R)) Q · φ̃ f , Ψφ (ψ) = ψ(s)ds · S(φ).
f 0
s∈SL2 (Ô)
There is some compact open subgroup K f ⊂ SL2 (Ô) such that

Z
s∈SL2 (Ô)
ψ(s)ds = ∑ ψ(s)ds(K f ).
s∈K f \SL2 (Ô)
As the index of K f ⊂ SL2 (Ô) is finite and ds(SL2 (Ô)) = 1, we have ds(K f ) ∈ Q× , and
therefore
Z
×
ψ(s)ds and S(φ)( g) ∈ Q for g ∈ G (A).
s∈SL2 (Ô)
Moreover, Ψm,n is Galois-equivariant. This follows from

Z Z
σ · ψ(s)ds =
s∈SL2 (Ô)
∑ σ · ψ(s)ds(K f ) = σ s∈SL2 (Ô)
ψ(s)ds , σ ∈ Gal(Q/Q),
s∈K f \SL2 (Ô)
and we conclude that the vectorspace ker(Ψm,n ) has a natural Q structure which we
denote by ker(Ψm,n|Q ).
Corollary 4.7.8. ker(Ψm,n+1|Q ) is in the image of ρ0m,n .
4.7.3 The long exact sequence for the cohomology of the boundary
The next step is to show that ker(Ψm,n+1|Q ) is exactly the image of ρ0m,n . To do so we use
a long exact cohomology sequence relating the cohomology with compact supports, the
usual cohomology of SK and the cohomology of the boundary.
We denote again by SK the Borel-Serre compactification of SK . Consider j : SK → SK

the open inclusion and i : SK \ SK → SK the closed inclusion of the boundary. For any
abelian sheaf F on SK we have the exact sequence
0 → j! j−1 F → F → i∗ i−1 F → 0.
125
p
If we apply Hc (SK , ) = H p (SK , ) to this sequence, we obtain a long exact cohomology
sequence
p p +1
· · · → Hc (SK , j! Q) → H p (SK , Q) → H p (SK \ SK , Q) → Hc (SK , j! Q) → · · ·
p p
Hc (SK , j! ) = Hc (SK , ), SK \ SK is homotopy equivalent to ∂SK and SK is homotopy
equivalent to SK by Harder (1987) 2.1, so we get the long exact cohomology sequence
p p +1
· · · → Hc (SK , Q) → H p (SK , Q) → H p (∂SK , Q) → Hc (SK , Q) → · · ·
Lemma 4.7.9. We have an exact sequence

resSK
2g
H 2g−1 (SK N , Q) −→N H 2g−1 (∂SK N , Q) → Hc (SK N , Q) → 0
(N)
and im(resSK N ) has codimension |Cl FK N | = |Cl F | in H 2g−1 (∂SK N , Q).
Proof. We start with the exact sequence

resSK
2g
H 2g−1 (SK N , Q) −→N H 2g−1 (∂SK N , F ) → Hc (SK N , Q) → H 2g (SK N , Q)
We use Poincaré duality to see
2g
Hc (SK N , C) = H 0 (SK N , C), H 2g (SK N , C) = Hc0 (SK N , C) = 0
The latter is zero, because SK N is not compact. The first group is non-zero, we even
conclude
2g (N)
dimQ Hc (SK N , Q) = dimC H 0 (SK N , C) = |Cl FK N | = |Cl F |.
So we have the exact sequence

resSK
2g
H 2g−1 (SK N , Q) −→N H 2g−1 (∂SK N , Q) → Hc (SK N , Q) → 0
and the codimension of im(resSK N ) in H 2g−1 (∂SK N , Q) equals
2g (N)
dimQ Hc (SK N , Q) = |Cl F |.
This lemma tells us that im(Eis00 ) cannot give more than a subspace of codimension
(N)
|Cl F | in the cohomology of the boundary in cohomological degree 2g − 1. In particular
ρ00,−1 cannot be surjective.
126
4.7.4 Determination of the image
Now we can completely determine the image of our polylogarithmic Eisenstein operator.
Lemma 4.7.10. im(ρ0m,n ) = ker(Ψm,n+1|Q )
Proof. It suffices to prove the case m = 0 and n = −1 with complex coefficients.

Consider the epimorphism (the map is split)
 K N
KN G (A )×π ( G (R))
Ψ0,0 = Ψ0,0 :  Ind B(A f )×π 0( B(R)) C · φ̃ f 
M
0
→
f
φ: type(φ)=γ0,0 ,φ̃ f |Z(R)=1
 K N
C · S(φ)
 M 

φ: type(φ)=γ0,0 ,φ̃ f |Z(R)=1 ,φ|T1 (A) =k · k2
Any φ on the right hand side is of the form

t2
φ ( t1 , t2 ) = η ( t1 t2 )
t

1
and η has to be trivial on UN . This follows from
S(φ)(kb) = φ̃ f (kb), k ∈ K N ∩ B(A f ), b ∈ B(A f ).
[ KN
Therefore η ∈ Cl F and the number of different φ on the right hand side is exactly
|Cl FK N |. So the codimension of ker(ΨK N ) inside
 K N
 IndG(A f )×π0 (G(R)) C · φ̃ f 
M
B(A f )×π0 ( B(R))
φ: type(φ)=γ0,0 ,φ̃ f |Z(R)=1
KN
is |Cl FK N |. By Corollary 4.7.8 we know that ker(Ψ0,0 ) ⊂ im(ρ00,−1 )K N , but
im(ρ00,−1 )K N ⊂ im(resSK N )
and im(resSK N ) has already codimension |Cl FK N | inside

 K N
H 2g−1 (∂SK N , C) =  IndG(A f )×π0 (G(R)) C · φ̃ f 

M
B(A f )×π0 ( B(R))
.
φ: type(φ)=γ0,0 ,φ̃ f |Z(R)=1
Therefore
KN
ker(Ψ0,0 ) = im(ρ00,−1 )K N = im(resSKN )
and consequently im(ρ00,−1 ) = ker(Ψ0,0 ).
127
Theorem 4.7.11. Let
Eiskq : S(V (A f ), µ⊗n ) ⊗ Hq∗ → H 2g−1−q (S , Symk H0 ⊗ µ⊗n+1 )
be the polylogarithmic Eisenstein operator defined in Corollary 3.3.13. The image of

this operator is isomorphic to ker(Ψm,n+1|Q ) ⊗ H( T/Z ) g−1−q .
Proof. By Theorem 4.6.2 and Theorem 4.6.1 we know that resS : im Eiskq → im(resS ◦ Eiskq )
is an isomorphism. By Proposition 4.5.7 we have the isomorphism
im(resS ◦ Eiskq ) ∼
= im(ρ0m,n ) ⊗ H( T/Z ) g−1−q .
Lemma 4.7.10 tells us ker(Ψm,n+1|Q ) = im(ρ0m,n ), from which the theorem follows.
2g−1−q
Corollary 4.7.12. If k > 0, then im(Eiskq ) = HEis (S , Symk H0 ⊗ µ⊗n+1 ). Moreover,
2g−1 2g−1−q
im(Eis00 ) = HEis (S , µ⊗n+1 ) and im(Eis0q )K N ⊂ HEis (SK N , µ⊗n+1 ) is a subspace of
codimension |Cl FK N |, if q > 0.
Proof. This is just Theorem 4.7.11 together with Harder (1987) Theorem 2. For the
calculation of the codimension see also the proof of Lemma 4.7.10.
128
Bibliography
Beilinson, A., Kings, G., & Levin, A. (2014). Topological polylogarithms and p-adic
interpolation of l-values of totally real fields. arXiv:1410.4741.
Beı̆linson, A. & Levin, A. (1994). The elliptic polylogarithm. In Motives (Seattle, WA,
1991), volume 55 of Proc. Sympos. Pure Math. (pp. 123–190). Amer. Math. Soc.,
Providence, RI.
Blottière, D. (2009a). Les classes d’Eisenstein des variétés de Hilbert-Blumenthal. Int.

Math. Res. Not. IMRN, (pp. 3236–3263).
Blottière, D. (2009b). Réalisation de Hodge du polylogarithme d’un schéma abélien. J.

Inst. Math. Jussieu, 8(1), 1–38.
Borel, A. & Wallach, N. (2000). Continuous cohomology, discrete subgroups, and repre-
sentations of reductive groups, volume 67 of Mathematical Surveys and Monographs.
American Mathematical Society, Providence, RI, second edition.
de Rham, G. (1955). Variétés différentiables. Formes, courants, formes harmoniques.

Actualités Sci. Ind., no. 1222 = Publ. Inst. Math. Univ. Nancago III. Hermann et Cie,
Paris.
Ehresmann, C. (1995). Les connexions infinitésimales dans un espace fibré différen-

tiable. In Séminaire Bourbaki, Vol. 1 (pp. Exp. No. 24, 153–168). Soc. Math. France,
Paris.
Franke, J. (1998). Harmonic analysis in weighted L2 -spaces. Ann. Sci. École Norm.
Sup. (4), 31(2), 181–279.
Franke, J. & Schwermer, J. (1998). A decomposition of spaces of automorphic forms,

and the Eisenstein cohomology of arithmetic groups. Math. Ann., 311(4), 765–790.
129
Bibliography
Gel0 fand, I. M., Graev, M. I., & Pyatetskii-Shapiro, I. I. (1990). Representation theory
and automorphic functions, volume 6 of Generalized Functions. Academic Press,
Inc., Boston, MA. Translated from the Russian by K. A. Hirsch, Reprint of the 1969
edition.
Godement, R. (1973). Topologie algébrique et théorie des faisceaux. Hermann, Paris.

Troisième édition revue et corrigée, Publications de l’Institut de Mathématique de
l’Université de Strasbourg, XIII, Actualités Scientifiques et Industrielles, No. 1252.
Grothendieck, A. (1957). Sur quelques points d’algèbre homologique. Tôhoku Math. J.

(2), 9, 119–221.
Halperin, S. & Greub, W. H. (1972). Fibre integration and some of its applications.
Collect. Math., 23, 195–212.
Harder, G. (1982). Period integrals of Eisenstein cohomology classes and special val-
ues of some L-functions. In Number theory related to Fermat’s last theorem (Cam-
bridge, Mass., 1981), volume 26 of Progr. Math. (pp. 103–142). Birkhäuser Boston,
Boston, MA.
Harder, G. (1987). Eisenstein cohomology of arithmetic groups. The case GL2 . Invent.
Math., 89(1), 37–118.
Harder, G. (1991). Eisenstein cohomology of arithmetic groups and its applications to

number theory. In Proceedings of the International Congress of Mathematicians, Vol.
I, II (Kyoto, 1990) (pp. 779–790).: Math. Soc. Japan, Tokyo.
Harder, G. (1993). Eisensteinkohomologie und die Konstruktion gemischter Motive,

volume 1562 of Lecture Notes in Mathematics. Springer-Verlag, Berlin.
Iversen, B. (1986). Cohomology of sheaves. Universitext. Springer-Verlag, Berlin.
Jacquet, H. & Shalika, J. A. (1981). On Euler products and the classification of auto-
morphic forms. II. Amer. J. Math., 103(4), 777–815.
Jacquet, H. & Zagier, D. (1987). Eisenstein series and the Selberg trace formula. II.
Trans. Amer. Math. Soc., 300(1), 1–48.
Jannsen, U. (1988). Continuous étale cohomology. Math. Ann., 280(2), 207–245.
130
Bibliography
Kashiwara, M. & Schapira, P. (1990). Sheaves on manifolds, volume 292 of

Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Math-
ematical Sciences]. Springer-Verlag, Berlin. With a chapter in French by Christian
Houzel.
Kings, G. (1999). K-theory elements for the polylogarithm of abelian schemes. J. Reine
Angew. Math., 517, 103–116.
Kings, G. (2001). The Tamagawa number conjecture for CM elliptic curves. Invent.
Math., 143(3), 571–627.
Kings, G. (2008). Degeneration of polylogarithms and special values of L-functions for

totally real fields. Doc. Math., 13, 131–159.
Levin, A. (2000). Polylogarithmic currents on abelian varieties. In Regulators in analysis,

geometry and number theory, volume 171 of Progr. Math. (pp. 207–229). Birkhäuser
Boston, Boston, MA.
Milne, J. S. (1979). Points on Shimura varieties mod p. In Automorphic forms, repre-

sentations and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis,
Ore., 1977), Part 2, Proc. Sympos. Pure Math., XXXIII (pp. 165–184). Amer. Math.
Soc., Providence, R.I.
Nori, M. V. (1995). Some Eisenstein cohomology classes for the integral unimodular
group. In Proceedings of the International Congress of Mathematicians, Vol. 1, 2
(Zürich, 1994) (pp. 690–696).: Birkhäuser, Basel.
Schwartz, L. (1966). Théorie des distributions. Publications de l’Institut de Mathéma-

tique de l’Université de Strasbourg, No. IX-X. Nouvelle édition, entiérement corrigée,
refondue et augmentée. Hermann, Paris.
Serre, J.-P. (1968). Abelian l-adic representations and elliptic curves. McGill University
lecture notes written with the collaboration of Willem Kuyk and John Labute. W. A.
Benjamin, Inc., New York-Amsterdam.
Tate, J. T. (1967). Fourier analysis in number fields, and Hecke’s zeta-functions. In

Algebraic Number Theory (Proc. Instructional Conf., Brighton, 1965) (pp. 305–347).
Thompson, Washington, D.C.
131
Bibliography
Weibel, C. A. (1994). An introduction to homological algebra, volume 38 of Cambridge

Studies in Advanced Mathematics. Cambridge University Press, Cambridge.
Wielonsky, F. (1985). Séries d’Eisenstein, intégrales toroïdales et une formule de

Hecke. Enseign. Math. (2), 31(1-2), 93–135.
Wildeshaus, J. (1997). Realizations of polylogarithms, volume 1650 of Lecture Notes

in Mathematics. Springer-Verlag, Berlin.
132

Polylogarithms For GL

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

Polylogarithms For GL

Uploaded by

Copyright:

Available Formats

Polylogarithms for GL2

over totally real fields

DISSERTATION ZUR ERLANGUNG DES DOKTORGRADES

1 Introduction and overview 1

2 The topological polylogarithm 15

3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties 47

4 Comparison with Harder’s Eisenstein classes 73

1.1 Eisenstein series and cohomology

XG,K := K \ G (A)/G (Q) AG (R)0 .

A representation ( E, ρ) of G in a finite dimensional Q-vectorspace defines a local system

up by Eisenstein series and residues of such.

As this decomposition of the cohomology is obtained by the theory of Eisenstein series

The Q-rationality of the decomposition above is of great arithmetic interest, as it can be

1.2 Eisenstein cohomology for Hilbert-Blumenthal varieties

ferential form ω, then ω is in particular U (Q)- invariant, where U ⊂ B is the unipotent

As the second step Harder constructs a G (A f )-equivariant operator

1.3 The toplogical polylogarithm and associated Eisenstein

π : TG,K := V (Ẑ) o KAG (R)0 \ V (A) o G (A)/V (Q) o G (Q) → XG,K

0 → H dim(V )−1 (U, Log ⊗ π −1 µn+1 ) → H 0 ( D, π|−D1 ∏ Symk V

for all n ∈ Z. Given a section f ∈ H 0 ( D, π|−D1 ∏k≥0 Symk V

pol( f ) ∈ H dim(V )−1 (U, Log ⊗ π −1 µn+1 ).

(Eisk ( f ))k≥0 := 0∗ pol ( f ) ∈ ∏ H dim(V )−1 (XG,K , Symk Ve ⊗ µn+1 ).

Eisk : S(V (A f ), Q)0 ⊗ H 0 ( XG , µn ) → H dim(V )−1 ( XG , Symk V

where S(V (A f ), Q)0 are the Q-valued Schwartz-Bruhat functions with

Here we only addressed the topological realization of the polylogarithm. Indeed, by

1.4 Decomposition of Eisenstein classes

In this work we present a decomposition principle for polylogarithmic Eisenstein classes

ϕ−1 (1) = (K∞ AG (R)0 ∩ Z (R)) \ Z (R)/Z (K )

and Z (K ) := Z (Q) ∩ K f ⊂ O F× is a subgroup of finite index of the units of O F ⊂ F the

H • ( ϕ−1 (1), Q) ⊗ H • ( XG,K

Eiskq : S(V (A f ), Q)0 ⊗ H 0 ( XG , µn ) ⊗ H q ( ϕ−1 (1), Q)∗ → H dim(V )−1−q ( XG , Symk V

for q = 0, ..., g − 1 = dim( ϕ−1 (1)).

We want to show that this decomposition of polylogarithmic Eisenstein classes is as

To understand the image of our polylogarithmic Eisenstein classes we have to deter-

Our main result is as follows

We want to remark a few things.

(e) A proof of a decomposition isomorphism

H • ( ϕ−1 (1), Q) ⊗ H • ( XG,K

1.5 Outline of the thesis

1.5.1 The topological polylogarithm

This allows us to define locally constant sheaves on T by using L o π1 (S)-modules.

locally constant sheaf on T is called the logarithm sheaf Log.

where j : U → T is the inclusion of the open complement of D (Proposition 2.3.8). We

when π : D → S is a finite cover (Proposition 2.4.5).

For computational purposes we also need a trivialization of the pro-vectorbundle asso-

Q[[ L]] → ∏ Symk ( L ⊗ Q), l 7→ exp(l ),

1.5.2 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal-varieties

As we have the polylogarithmic Eisenstein classes on MK , we want to decompose the

e ⊗ µn+1 . As ϕ is a fiber bundle it

This gives consequently R• ϕ∗ (Symk V e ⊗ µn+1 ⊗ R• ϕ∗ (Q) (Corol-

1.5.3 Comparison with Harder’s Eisenstein classes

As Harder’s Eisenstein classes are determined by their restriction to the cohomlogy of

Special thanks go to my colleagues and friends Sandra Eisenreich, Johannes Sprang

2.1 The logarithm sheaf on families of topological tori

2.1.1 Families of topological tori

Definition 1. Let π : T → S be a proper submersion of C ∞ -manifolds. We call π :

Let us fix a basepoint s0 ∈ S and let us take t0 := 0(s0 ) ∈ T as basepoint for T ,

separately. Any proper submersion of C ∞ -manifolds is a fiber bundle, see Ehresmann

Example 2.1.3. If π : T → S is a torus, we have the locally constant sheaf H :=

Definition 2. A lattice L is a free Z-module of finite rank. Let k → k0 be a ring homo-

Proposition 2.1.4. Let S be a connected C ∞ -manifold, say with basepoint s0 ∈ S . Then

Tori/S → {lattices L with left-π1 (S , s0 )-action and equivariant maps}

Proof. Given a π1 (S , s0 )-module L we have the torus T := LR /L. As π1 (S , s0 ) acts on