Professional Documents
Culture Documents
Polylogarithms For GL
Polylogarithms For GL
vorgelegt von
Philipp Graf aus
Regensburg
im Jahr 2015
Die Arbeit wurde angeleitet von: Prof. Dr. Guido Kings
Contents
i
Contents
Bibliography 129
ii
1 Introduction and overview
Let G be a reductive linear algebraic group over Q and A the adeles over Q. Let us
consider the double quotient space
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.
1
1 Introduction and overview
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
•
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
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
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
1 Introduction and overview
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
This class may be specialized along the zero section to obtain polylogarithmic Eisen-
stein classes
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
4
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.
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
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
The polylogarithmic Eisenstein classes may then be decomposed with respect to this
isomorphism. We get by evaluation
5
1 Introduction and overview
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.
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.
(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.
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.
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
1 Introduction and overview
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.
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 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 ),
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).
8
1.5 Outline of the thesis
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).
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).
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 Introduction and overview
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).
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
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
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.
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 .
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 )
15
2 The topological polylogarithm
∗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.
T 7→ H1 (Ts0 , Z)
16
2.1 The logarithm sheaf on families of topological tori
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.
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.
17
2 The topological polylogarithm
deg(φ) : S → Z, s 7→ | ker(φs )|
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
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
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
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.
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
2 The topological polylogarithm
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 / ···
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.
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.
20
2.2 Limits of locally constant sheaves
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
21
2 The topological polylogarithm
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
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.
Hom( B, I ⊗k F ) / Hom( A, I ⊗k F )
Hom( B ⊗k F ∗ , I) / Hom( A ⊗k F ∗ , I)
22
2.3 Purity and localization
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.
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.
23
2 The topological polylogarithm
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
functorial in F .
F [d] → DX ⊗k or ∗X ⊗k F .
24
2.3 Purity and localization
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
25
2 The topological polylogarithm
where we used purity and the fact that (i−1 Fn ⊗k or Z/X ) is again (M-L). Finally, we have
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
f −1 (lim Fn ) → lim f −1 Fn
←− ←−
is an isomorphism.
26
2.3 Purity and localization
as the limit commutes with the section functor. The inductive limit becomes constant
and therefore
This proves the first claim. The second follows easily from the first by considering the
induced maps on the stalks.
(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
+1
i∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → lim Fn → Rj∗ j−1 lim Fn →
←− ←− ←−
27
2 The topological polylogarithm
i∗ i! F (V ) = {s ∈ F (V ) : supp(s) ⊂ Z }
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
+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.
i0 / X0
Z0
f0 f
Z
i /X
+1
i∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → lim Fn → Rj∗ j−1 lim Fn →
←− ←− ←−
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
2.3 Purity and localization
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.
+1
i∗ (i−1 (lim Fn ) ⊗k or Z/X )[−c] → lim Fn → Rj∗ j−1 lim Fn →
←− ←− ←−
29
2 The topological polylogarithm
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.
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
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.
30
2.4 Construction of the topological polylogarithm
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
31
2 The topological polylogarithm
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
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,
aug tr
π| D∗ ( Log| D ) → π| D∗ k → k.
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
purity
H 0 ({ x } , k ) → H{dx} ( T, k ) → H d ( T, k ),
32
2.4 Construction of the topological polylogarithm
purity maps it to
Proof. Using the Grothendieck-Leray spectral sequence for the composition of functors
Γ(U, −) = Γ(S , −) ◦ π|U ∗ we easily see
as
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
2 The topological polylogarithm
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.
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
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
2.4 Construction of the topological polylogarithm
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
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
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.
t−1 Log ∼
= t−1 [ N ]−1 Log ∼
= ([ N ] ◦ t)−1 Log ∼
= 0−1 Log,
35
2 The topological polylogarithm
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
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
36
2.4 Construction of the topological polylogarithm
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,
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
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 .
(−v)⊗n
exp(−v) = ∑ n!
n ≥0
with
37
2 The topological polylogarithm
We set
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
R∞ → Log∞ , f 7→ f · ρcont .
38
2.4 Construction of the topological polylogarithm
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
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
f : (Z/NZ)d \ {0} × S → k,
with
aug( f ) = ∑ f (v, ) = 0 ∈ H 0 (S , k)
v∈(Z/NZ)d \{0}
whenever k is torsion-free.
39
2 The topological polylogarithm
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.
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
2.4 Construction of the topological polylogarithm
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
C[0] ⊂ A X → DX [−n]
i∗ : Γc (U, i∗ A Z )∗ → Γc (U, A X )∗
H 0 ( Z, C) = H 0 ( Z, or Z ) = H 0 ( Z, DZ [c − n]) = H 0 ( Z, i! DX [c − n]) =
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
41
2 The topological polylogarithm
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
p
is injective we need to see that all transition maps are split epimorphic. Now DT is a
CT∞ -module and therefore
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 .
Proof. We set dim(T ) = n. To calculate cohmomology we can use the injective reso-
lutions
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
2.4 Construction of the topological polylogarithm
∼
=
Log| D ⊗ or − 1 ! ! −1 −1
D ⊗ i DT [−( n − d )] → i ( Log ⊗ π orS ⊗ DT [−( n − d )]).
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
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
43
2 The topological polylogarithm
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 .
(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
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
44
2.4 Construction of the topological polylogarithm
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
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
D(U 0 ) ∼
= (π∗ DX )G (U ) → D X/G (U ), T 7→ π|U 0 ∗ T = T̃.
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 .
by the transformation-formula.
45
2 The topological polylogarithm
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
(π∗ DX )G → DX/G , T 7→ T̃
(−1)deg(T ) T (π|∗U 0 dω ) = d T̃ (ω )
46
3 Polylogarithmic Eisenstein classes for
Hilbert-Blumenthal varieties
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|∞
47
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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
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
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 The topological situation
1 K ⊂ G (A) and
and its conjugates. Set K := K∞ K f ⊂ G (A), K1 := K∞ f
We define spaces
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.
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.
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
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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.
• 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.
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
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
3.1 The topological situation
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 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.
51
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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))
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.
g
H± × (R>0 \ Z (R)0 ) /G ( g f )
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
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
52
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.
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)} ,
53
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
sK : K \ G (A) →→ Q
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.
Let us apply the construction of the polylogarithm to this situation. We have the natural
short exact sequence
54
3.2 Construction of adelic polylogarithmic Eisenstein classes
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
k
in other words (EisW ( f ))k ∈ ∏k≥0 H 2g−1 (MK , Symk HK ⊗ µK ).
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 )
f ( , ) : K f \ G (A f ) × π0 ( G (R)) → S(V (A f ), k ), x 7→ f ( , x )
55
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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 (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
3.2 Construction of adelic polylogarithmic Eisenstein classes
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
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
57
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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
qW
W
MK
1
/ MK
1
µ and Log are natural in pullbacks, see Remark 2.3.9, we get the following commutative
diagram
H 2g−1 (TW \ DW
0 , Log ⊗ µ⊗n+1 ) / H 0 ( D 0 , Log 0 ⊗ µ⊗n )
1 W1 |D W1
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
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
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
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
Tor1R ( H p ( G, M ), H q ( H, N )) → 0
M
→
p + q = n +1
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
3.3 Pushforward of polylogarithmic Eisenstein-classes
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)
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).
H • ( G, NK ) = H • ( G, N ) ⊗k K
H • ( HomK[G] ( PK , NK )) = H • ( Homk[G] ( P, N )) ⊗k K = H • ( G, N ) ⊗k K
61
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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)
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
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
σ σ σ
for e ∈ Z (K ) ⊂ O × .
62
3.3 Pushforward of polylogarithmic Eisenstein-classes
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.
and we get
∑σ nσ =n σ
63
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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
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
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
3.3 Pushforward of polylogarithmic Eisenstein-classes
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)
b O ( p ) = b( p ) = ∏ pνp (b) O( p)
p ⊂ O max., p| p
!
∏ σ (b) R Gal( FHil /F )
= | N (b)| A( p) = ∏ |κ (p)|νp (b) A( p) =
σ ( p) p ⊂ O max.
[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
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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
and
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
3.3 Pushforward of polylogarithmic Eisenstein-classes
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
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.
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
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
i i
and the properties of ηi show that the last map is an isomorphism. In particular,
i i
is a quasi-isomorphism. We conclude
68
3.3 Pushforward of polylogarithmic Eisenstein-classes
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.
on cohomology.
69
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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 )
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)).
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
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]
70
3.3 Pushforward of polylogarithmic Eisenstein-classes
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 .
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
71
3 Polylogarithmic Eisenstein classes for Hilbert-Blumenthal varieties
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]
tr
Rd φ∗ (C)[−d]
= / Rd φ∗ (C)[−d]
• 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.
tr = H p ( B, tr • ) : H p ( E, f −1 V ) → H p−d ( B, V ⊗ Rd f ∗ (C))
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.
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.
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
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 ),
H p ( X, V ) = Ȟ p (U, H 0 (V )) = Ȟ p (U, V ).
H p ( X, V ⊗ A R) = Ȟ p (U, V ⊗ A R) = Ȟ p (U, V ) ⊗ A R = H p ( X, V ) ⊗ A R,
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
4.1 Nori’s calculation of the polylogarithm as current
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.
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
75
4 Comparison with Harder’s Eisenstein classes
−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
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
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
4.1 Nori’s calculation of the polylogarithm as current
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
For this see Schwartz (1966) VII 1. This gives us the Fourier expansion
77
4 Comparison with Harder’s Eisenstein classes
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
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
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
78
4.1 Nori’s calculation of the polylogarithm as current
( 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
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
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
since
79
4 Comparison with Harder’s Eisenstein classes
So we should have
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
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
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
81
4 Comparison with Harder’s Eisenstein classes
can.
λ p : Q p → Q p /Z p ⊂ Q/Z ⊂ R/Z,
λ∞ : R → R/Z, x 7→ − x mod Z.
V (A) → V
\ (A), x → ψ0 h x, i := ψ0 (h x, i).
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)
82
4.2 Global description of the polylogarithm in adelic coordinates
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
−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
ϕ= ∑ ϕ(u)χu ,
u∈ N −1 V (Ẑ)/NV (Ẑ)
83
4 Comparison with Harder’s Eisenstein classes
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
dw1σ ∧ dw2σ
V
σ volV
volT /M := −
= .
kdet gk f sgn(N(det g)) 1 kdet gk f sgn(N(det g))−1
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
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
84
4.2 Global description of the polylogarithm in adelic coordinates
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 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,
85
4 Comparison with Harder’s Eisenstein classes
√
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
ϕ f ⊗ ϕ ∞ ( v ) : = ϕ f ( v f ) ϕ ∞ ( v ∞ ), v ∈ V (A).
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}
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.
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
4 Comparison with Harder’s Eisenstein classes
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.
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
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
4.3 The decomposition Isomorphism as Mellin-Tranformation
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
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
• 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
4 Comparison with Harder’s Eisenstein classes
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)!
.
for k = gm ≥ 0.
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
Set
√
ut−1 ĝ− 1 !
√ f (1, τ )
ψ( ut−1 ) := ∑ φ( fˆ) p
| Im(τ )|
l, g .
l ∈V (Q)\{0}
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
4.3 The decomposition Isomorphism as Mellin-Tranformation
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 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
91
4 Comparison with Harder’s Eisenstein classes
√
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.
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
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
γ∈ G (Q)/B(Q)) t∈ Z (A)
92
4.3 The decomposition Isomorphism as Mellin-Tranformation
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
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
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
p+q=2g−1
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
4 Comparison with Harder’s Eisenstein classes
√
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
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
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)
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
95
4 Comparison with Harder’s Eisenstein classes
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))
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)
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).
96
4.4 Cohomology of the boundary
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.
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
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
97
4 Comparison with Harder’s Eisenstein classes
t7→(s,t)
1 → Z (R)0 /Z (K )+ → SKT → π0 (SKT ) = K T T (R)0 \ T (A)/T (Q) → 1.
Harder considers algebraic Hecke characters of type γm,n (Harder (1987) (2.5.2)), in
other words, continuous homomorphisms
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).
Q · φ̃ f ⊗ H p ( T/Z ) ∼
= H p− g (S B , Symk H0 ⊗ µ⊗n ⊗ Q)
M
∗n
ψ · N(ω )m ⊗ volV dx1 ∧ ... ∧ dx g
ψ ⊗ η 7→ √ ∪η
dF
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 =
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.
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
p+q=2g−1
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
99
4 Comparison with Harder’s Eisenstein classes
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))
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 .
and this Tate-integral exists by Wielonsky (1985) III Proposition 5. To treat the second
summand we consider the bijection
100
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
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
theexterior 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
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
4 Comparison with Harder’s Eisenstein classes
∏ 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
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
102
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
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
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
4 Comparison with Harder’s Eisenstein classes
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
104
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
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 .
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 ×
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
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
4 Comparison with Harder’s Eisenstein classes
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
Of course, we can refine this statement, if we use the integral structures on cohomology,
which we have chosen.
f (0) = 0, ∑ f (v) = 0.
v∈V ( Z/NZ)
106
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
∗
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×
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
[ 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
107
4 Comparison with Harder’s Eisenstein classes
and
is part of an A-basis of
by
m +2 ×
fˆ(tg f e1 , g)χ(det( g f )t)
det( g f )t
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
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
φ: type(φ)=γm,n+1 ,φ̃ f |Z(R)=1
im(resS ◦ Eiskq ) ∼
= im(ρ0m,n ) ⊗ H( T/Z ) g−1−q .
108
4.5 Restriction of polylogarithmic Eisenstein classes to the boundary
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))
φ: type(φ)=γm,n+1 ,φ̃ f |Z(R)=1
is surjective. Clearly,
H• ( T/Z ) ⊗ HC
•
= H • ( T (R)0 /T (Q) ∩ (K f · T (R)0 , C)
c0 dy dr
2 g −1
· ∑ sgn( I )
y I
∧
r h g−1i\ I
,
I ⊂h g−1i
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 Comparison with Harder’s 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.
which is a section for resS . We denote the image of the operator Eis Harder by
•
HEis (S , Symk H0 ⊗ µ⊗n+1 )
Theorem 4.6.2.
•
im(Eiskq ) ⊂ HEis (S , Symk H0 ⊗ µ⊗n+1 )
110
4.6 Comparison with Harders Eisenstein classes
follows.
From now on we suppose that ϕ is of the form above. For I $ h gi we define Θ̃ I as the
form
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
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
∗ 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
4 Comparison with Harder’s Eisenstein classes
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
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
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
112
4.6 Comparison with Harders Eisenstein classes
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
by
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
4 Comparison with Harder’s Eisenstein classes
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
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
∗ 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 =
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
4.6 Comparison with Harders Eisenstein classes
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
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
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
4 Comparison with Harder’s Eisenstein classes
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 )
G ( g f ) := ( g−
f K f g f · G (R) ) ∩ G (Q), B ( g f ) = : G ( g f ) ∩ B (A).
1 0
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)
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 ·
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)
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
lim
ξ →0
∑ FI,mf (χ, 0, g f s)
s∈ G ( g f )\ G (Q)/B(Q)
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)
( 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)
117
4 Comparison with Harder’s Eisenstein classes
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 )
it even suffices to understand the image of ρ0m,0 . But first we want to understand a more
general map.
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 ×
fˆ(tg f e1 , g)χ(det( g f )t)
det( g f )t
ρm ( f )( g) := d t,
dK (m) (−2πi )k+2g t ∈ Z (A f )
χ∈Cl F
if f is K f -invariant.
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
4.7 The image of the polylogarithmic Eisenstein Operator
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
K N \ G (A f ) × π0 ( G (R))/P(A f ) × π0 ( P∞ ) → C.
ψ : K N \ G (Ẑ)/P(Ẑ) = G (Z/NZ)/P(Z/NZ ) → C.
119
4 Comparison with Harder’s Eisenstein classes
where the local functions sψν : Fν → C are defined by sψν := χOν2 , if ν - N, and
(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
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 )
x ∈(O /N O)×
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
4.7 The image of the polylogarithmic Eisenstein Operator
−(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)
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 )
121
4 Comparison with Harder’s Eisenstein classes
For any set of representatives {u} the function ϕ{u} ∈ S(V (A f ), µ⊗0 ⊗ C) defined by
its Fourier transform
is a preimage of ψ under ρm .
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
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,
122
4.7 The image of the polylogarithmic Eisenstein Operator
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
G (A )×π0 ( G (R))
Ind B(A f )×π C · φ̃ f , φ|T1 (A) = k · k2 ,
f 0 ( B (R))
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
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)
123
4 Comparison with Harder’s Eisenstein classes
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)
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)
124
4.7 The image of the polylogarithmic Eisenstein Operator
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 .
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)
and we conclude that the vectorspace ker(Ψm,n ) has a natural Q structure which we
denote by ker(Ψm,n|Q ).
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.
0 → j! j−1 F → F → i∗ i−1 F → 0.
125
4 Comparison with Harder’s Eisenstein classes
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) → · · ·
(N)
and im(resSK N ) has codimension |Cl FK N | = |Cl F | in H 2g−1 (∂SK N , Q).
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 |.
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 The image of the polylogarithmic Eisenstein Operator
Now we can completely determine the image of our polylogarithmic Eisenstein operator.
K N
C · S(φ)
M
φ: type(φ)=γ0,0 ,φ̃ f |Z(R)=1 ,φ|T1 (A) =k · k2
[ 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 )
Therefore
KN
ker(Ψ0,0 ) = im(ρ00,−1 )K N = im(resSKN )
127
4 Comparison with Harder’s Eisenstein classes
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.
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.
Franke, J. (1998). Harmonic analysis in weighted L2 -spaces. Ann. Sci. École Norm.
Sup. (4), 31(2), 181–279.
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.
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.
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.
130
Bibliography
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.
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.
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.
131
Bibliography
132